From 7bb481fda9ecb134804b49c2ce77ca28f7eea583 Mon Sep 17 00:00:00 2001 From: Hans-Christoph Steiner Date: Fri, 30 Mar 2012 20:42:12 -0400 Subject: Imported Upstream version 2.0.3 --- ext/fts3/README.content | 178 ++ ext/fts3/README.syntax | 209 ++ ext/fts3/README.tokenizers | 133 ++ ext/fts3/README.txt | 4 + ext/fts3/fts3.c | 5092 ++++++++++++++++++++++++++++++++++++++++++++ ext/fts3/fts3.h | 26 + ext/fts3/fts3Int.h | 523 +++++ ext/fts3/fts3_aux.c | 474 +++++ ext/fts3/fts3_expr.c | 984 +++++++++ ext/fts3/fts3_hash.c | 383 ++++ ext/fts3/fts3_hash.h | 112 + ext/fts3/fts3_icu.c | 258 +++ ext/fts3/fts3_porter.c | 645 ++++++ ext/fts3/fts3_snippet.c | 1500 +++++++++++++ ext/fts3/fts3_term.c | 369 ++++ ext/fts3/fts3_test.c | 324 +++ ext/fts3/fts3_tokenizer.c | 489 +++++ ext/fts3/fts3_tokenizer.h | 152 ++ ext/fts3/fts3_tokenizer1.c | 233 ++ ext/fts3/fts3_write.c | 3397 +++++++++++++++++++++++++++++ ext/fts3/fts3speed.tcl | 122 ++ ext/fts3/mkfts3amal.tcl | 115 + 22 files changed, 15722 insertions(+) create mode 100644 ext/fts3/README.content create mode 100644 ext/fts3/README.syntax create mode 100644 ext/fts3/README.tokenizers create mode 100644 ext/fts3/README.txt create mode 100644 ext/fts3/fts3.c create mode 100644 ext/fts3/fts3.h create mode 100644 ext/fts3/fts3Int.h create mode 100644 ext/fts3/fts3_aux.c create mode 100644 ext/fts3/fts3_expr.c create mode 100644 ext/fts3/fts3_hash.c create mode 100644 ext/fts3/fts3_hash.h create mode 100644 ext/fts3/fts3_icu.c create mode 100644 ext/fts3/fts3_porter.c create mode 100644 ext/fts3/fts3_snippet.c create mode 100644 ext/fts3/fts3_term.c create mode 100644 ext/fts3/fts3_test.c create mode 100644 ext/fts3/fts3_tokenizer.c create mode 100644 ext/fts3/fts3_tokenizer.h create mode 100644 ext/fts3/fts3_tokenizer1.c create mode 100644 ext/fts3/fts3_write.c create mode 100644 ext/fts3/fts3speed.tcl create mode 100644 ext/fts3/mkfts3amal.tcl (limited to 'ext/fts3') diff --git a/ext/fts3/README.content b/ext/fts3/README.content new file mode 100644 index 0000000..ab98675 --- /dev/null +++ b/ext/fts3/README.content @@ -0,0 +1,178 @@ + +FTS4 CONTENT OPTION + + Normally, in order to create a full-text index on a dataset, the FTS4 + module stores a copy of all indexed documents in a specially created + database table. + + As of SQLite version 3.7.9, FTS4 supports a new option - "content" - + designed to extend FTS4 to support the creation of full-text indexes where: + + * The indexed documents are not stored within the SQLite database + at all (a "contentless" FTS4 table), or + + * The indexed documents are stored in a database table created and + managed by the user (an "external content" FTS4 table). + + Because the indexed documents themselves are usually much larger than + the full-text index, the content option can sometimes be used to achieve + significant space savings. + +CONTENTLESS FTS4 TABLES + + In order to create an FTS4 table that does not store a copy of the indexed + documents at all, the content option should be set to an empty string. + For example, the following SQL creates such an FTS4 table with three + columns - "a", "b", and "c": + + CREATE VIRTUAL TABLE t1 USING fts4(content="", a, b, c); + + Data can be inserted into such an FTS4 table using an INSERT statements. + However, unlike ordinary FTS4 tables, the user must supply an explicit + integer docid value. For example: + + -- This statement is Ok: + INSERT INTO t1(docid, a, b, c) VALUES(1, 'a b c', 'd e f', 'g h i'); + + -- This statement causes an error, as no docid value has been provided: + INSERT INTO t1(a, b, c) VALUES('j k l', 'm n o', 'p q r'); + + It is not possible to UPDATE or DELETE a row stored in a contentless FTS4 + table. Attempting to do so is an error. + + Contentless FTS4 tables also support SELECT statements. However, it is + an error to attempt to retrieve the value of any table column other than + the docid column. The auxiliary function matchinfo() may be used, but + snippet() and offsets() may not. For example: + + -- The following statements are Ok: + SELECT docid FROM t1 WHERE t1 MATCH 'xxx'; + SELECT docid FROM t1 WHERE a MATCH 'xxx'; + SELECT matchinfo(t1) FROM t1 WHERE t1 MATCH 'xxx'; + + -- The following statements all cause errors, as the value of columns + -- other than docid are required to evaluate them. + SELECT * FROM t1; + SELECT a, b FROM t1 WHERE t1 MATCH 'xxx'; + SELECT docid FROM t1 WHERE a LIKE 'xxx%'; + SELECT snippet(t1) FROM t1 WHERE t1 MATCH 'xxx'; + + Errors related to attempting to retrieve column values other than docid + are runtime errors that occur within sqlite3_step(). In some cases, for + example if the MATCH expression in a SELECT query matches zero rows, there + may be no error at all even if a statement does refer to column values + other than docid. + +EXTERNAL CONTENT FTS4 TABLES + + An "external content" FTS4 table is similar to a contentless table, except + that if evaluation of a query requires the value of a column other than + docid, FTS4 attempts to retrieve that value from a table (or view, or + virtual table) nominated by the user (hereafter referred to as the "content + table"). The FTS4 module never writes to the content table, and writing + to the content table does not affect the full-text index. It is the + responsibility of the user to ensure that the content table and the + full-text index are consistent. + + An external content FTS4 table is created by setting the content option + to the name of a table (or view, or virtual table) that may be queried by + FTS4 to retrieve column values when required. If the nominated table does + not exist, then an external content table behaves in the same way as + a contentless table. For example: + + CREATE TABLE t2(id INTEGER PRIMARY KEY, a, b, c); + CREATE VIRTUAL TABLE t3 USING fts4(content="t2", a, c); + + Assuming the nominated table does exist, then its columns must be the same + as or a superset of those defined for the FTS table. + + When a users query on the FTS table requires a column value other than + docid, FTS attempts to read this value from the corresponding column of + the row in the content table with a rowid value equal to the current FTS + docid. Or, if such a row cannot be found in the content table, a NULL + value is used instead. For example: + + CREATE TABLE t2(id INTEGER PRIMARY KEY, a, b, c, d); + CREATE VIRTUAL TABLE t3 USING fts4(content="t2", b, c); + + INSERT INTO t2 VALUES(2, 'a b', 'c d', 'e f'); + INSERT INTO t2 VALUES(3, 'g h', 'i j', 'k l'); + INSERT INTO t3(docid, b, c) SELECT id, b, c FROM t2; + + -- The following query returns a single row with two columns containing + -- the text values "i j" and "k l". + -- + -- The query uses the full-text index to discover that the MATCH + -- term matches the row with docid=3. It then retrieves the values + -- of columns b and c from the row with rowid=3 in the content table + -- to return. + -- + SELECT * FROM t3 WHERE t3 MATCH 'k'; + + -- Following the UPDATE, the query still returns a single row, this + -- time containing the text values "xxx" and "yyy". This is because the + -- full-text index still indicates that the row with docid=3 matches + -- the FTS4 query 'k', even though the documents stored in the content + -- table have been modified. + -- + UPDATE t2 SET b = 'xxx', c = 'yyy' WHERE rowid = 3; + SELECT * FROM t3 WHERE t3 MATCH 'k'; + + -- Following the DELETE below, the query returns one row containing two + -- NULL values. NULL values are returned because FTS is unable to find + -- a row with rowid=3 within the content table. + -- + DELETE FROM t2; + SELECT * FROM t3 WHERE t3 MATCH 'k'; + + When a row is deleted from an external content FTS4 table, FTS4 needs to + retrieve the column values of the row being deleted from the content table. + This is so that FTS4 can update the full-text index entries for each token + that occurs within the deleted row to indicate that that row has been + deleted. If the content table row cannot be found, or if it contains values + inconsistent with the contents of the FTS index, the results can be difficult + to predict. The FTS index may be left containing entries corresponding to the + deleted row, which can lead to seemingly nonsensical results being returned + by subsequent SELECT queries. The same applies when a row is updated, as + internally an UPDATE is the same as a DELETE followed by an INSERT. + + Instead of writing separately to the full-text index and the content table, + some users may wish to use database triggers to keep the full-text index + up to date with respect to the set of documents stored in the content table. + For example, using the tables from earlier examples: + + CREATE TRIGGER t2_bu BEFORE UPDATE ON t2 BEGIN + DELETE FROM t3 WHERE docid=old.rowid; + END; + CREATE TRIGGER t2_bd BEFORE DELETE ON t2 BEGIN + DELETE FROM t3 WHERE docid=old.rowid; + END; + + CREATE TRIGGER t2_bu AFTER UPDATE ON t2 BEGIN + INSERT INTO t3(docid, b, c) VALUES(new.rowid, new.b, new.c); + END; + CREATE TRIGGER t2_bd AFTER INSERT ON t2 BEGIN + INSERT INTO t3(docid, b, c) VALUES(new.rowid, new.b, new.c); + END; + + The DELETE trigger must be fired before the actual delete takes place + on the content table. This is so that FTS4 can still retrieve the original + values in order to update the full-text index. And the INSERT trigger must + be fired after the new row is inserted, so as to handle the case where the + rowid is assigned automatically within the system. The UPDATE trigger must + be split into two parts, one fired before and one after the update of the + content table, for the same reasons. + + FTS4 features a special command similar to the 'optimize' command that + deletes the entire full-text index and rebuilds it based on the current + set of documents in the content table. Assuming again that "t3" is the + name of the external content FTS4 table, the command is: + + INSERT INTO t3(t3) VALUES('rebuild'); + + This command may also be used with ordinary FTS4 tables, although it may + only be useful if the full-text index has somehow become corrupt. It is an + error to attempt to rebuild the full-text index maintained by a contentless + FTS4 table. + + diff --git a/ext/fts3/README.syntax b/ext/fts3/README.syntax new file mode 100644 index 0000000..01bc80c --- /dev/null +++ b/ext/fts3/README.syntax @@ -0,0 +1,209 @@ + +1. OVERVIEW + + This README file describes the syntax of the arguments that may be passed to + the FTS3 MATCH operator used for full-text queries. For example, if table + "t1" is an Fts3 virtual table, the following SQL query: + + SELECT * FROM t1 WHERE MATCH + + may be used to retrieve all rows that match a specified for full-text query. + The text "" should be replaced by either the name of the fts3 table + (in this case "t1"), or by the name of one of the columns of the fts3 + table. should be replaced by an SQL expression that + computes to a string containing an Fts3 query. + + If the left-hand-side of the MATCH operator is set to the name of the + fts3 table, then by default the query may be matched against any column + of the table. If it is set to a column name, then by default the query + may only match the specified column. In both cases this may be overriden + as part of the query text (see sections 2 and 3 below). + + As of SQLite version 3.6.8, Fts3 supports two slightly different query + formats; the standard syntax, which is used by default, and the enhanced + query syntax which can be selected by compiling with the pre-processor + symbol SQLITE_ENABLE_FTS3_PARENTHESIS defined. + + -DSQLITE_ENABLE_FTS3_PARENTHESIS + +2. STANDARD QUERY SYNTAX + + When using the standard Fts3 query syntax, a query usually consists of a + list of terms (words) separated by white-space characters. To match a + query, a row (or column) of an Fts3 table must contain each of the specified + terms. For example, the following query: + + MATCH 'hello world' + + matches rows (or columns, if is the name of a column name) that + contain at least one instance of the token "hello", and at least one + instance of the token "world". Tokens may be grouped into phrases using + quotation marks. In this case, a matching row or column must contain each + of the tokens in the phrase in the order specified, with no intervening + tokens. For example, the query: + + MATCH '"hello world" joe" + + matches the first of the following two documents, but not the second or + third: + + "'Hello world', said Joe." + "One should always greet the world with a cheery hello, thought Joe." + "How many hello world programs could their be?" + + As well as grouping tokens together by phrase, the binary NEAR operator + may be used to search for rows that contain two or more specified tokens + or phrases within a specified proximity of each other. The NEAR operator + must always be specified in upper case. The word "near" in lower or mixed + case is treated as an ordinary token. For example, the following query: + + MATCH 'engineering NEAR consultancy' + + matches rows that contain both the "engineering" and "consultancy" tokens + in the same column with not more than 10 other words between them. It does + not matter which of the two terms occurs first in the document, only that + they be seperated by only 10 tokens or less. The user may also specify + a different required proximity by adding "/N" immediately after the NEAR + operator, where N is an integer. For example: + + MATCH 'engineering NEAR/5 consultancy' + + searches for a row containing an instance of each specified token seperated + by not more than 5 other tokens. More than one NEAR operator can be used + in as sequence. For example this query: + + MATCH 'reliable NEAR/2 engineering NEAR/5 consultancy' + + searches for a row that contains an instance of the token "reliable" + seperated by not more than two tokens from an instance of "engineering", + which is in turn separated by not more than 5 other tokens from an + instance of the term "consultancy". Phrases enclosed in quotes may + also be used as arguments to the NEAR operator. + + Similar to the NEAR operator, one or more tokens or phrases may be + separated by OR operators. In this case, only one of the specified tokens + or phrases must appear in the document. For example, the query: + + MATCH 'hello OR world' + + matches rows that contain either the term "hello", or the term "world", + or both. Note that unlike in many programming languages, the OR operator + has a higher precedence than the AND operators implied between white-space + separated tokens. The following query matches documents that contain the + term 'sqlite' and at least one of the terms 'fantastic' or 'impressive', + not those that contain both 'sqlite' and 'fantastic' or 'impressive': + + MATCH 'sqlite fantastic OR impressive' + + Any token that is part of an Fts3 query expression, whether or not it is + part of a phrase enclosed in quotes, may have a '*' character appended to + it. In this case, the token matches all terms that begin with the characters + of the token, not just those that exactly match it. For example, the + following query: + + MATCH 'sql*' + + matches all rows that contain the term "SQLite", as well as those that + contain "SQL". + + A token that is not part of a quoted phrase may be preceded by a '-' + character, which indicates that matching rows must not contain the + specified term. For example, the following: + + MATCH '"database engine" -sqlite' + + matches rows that contain the phrase "database engine" but do not contain + the term "sqlite". If the '-' character occurs inside a quoted phrase, + it is ignored. It is possible to use both the '-' prefix and the '*' postfix + on a single term. At this time, all Fts3 queries must contain at least + one term or phrase that is not preceded by the '-' prefix. + + Regardless of whether or not a table name or column name is used on the + left hand side of the MATCH operator, a specific column of the fts3 table + may be associated with each token in a query by preceding a token with + a column name followed by a ':' character. For example, regardless of what + is specified for , the following query requires that column "col1" + of the table contains the term "hello", and that column "col2" of the + table contains the term "world". If the table does not contain columns + named "col1" and "col2", then an error is returned and the query is + not run. + + MATCH 'col1:hello col2:world' + + It is not possible to associate a specific table column with a quoted + phrase or a term preceded by a '-' operator. A '*' character may be + appended to a term associated with a specific column for prefix matching. + +3. ENHANCED QUERY SYNTAX + + The enhanced query syntax is quite similar to the standard query syntax, + with the following four differences: + + 1) Parenthesis are supported. When using the enhanced query syntax, + parenthesis may be used to overcome the built-in precedence of the + supplied binary operators. For example, the following query: + + MATCH '(hello world) OR (simple example)' + + matches documents that contain both "hello" and "world", and documents + that contain both "simple" and "example". It is not possible to forumlate + such a query using the standard syntax. + + 2) Instead of separating tokens and phrases by whitespace, an AND operator + may be explicitly specified. This does not change query processing at + all, but may be used to improve readability. For example, the following + query is handled identically to the one above: + + MATCH '(hello AND world) OR (simple AND example)' + + As with the OR and NEAR operators, the AND operator must be specified + in upper case. The word "and" specified in lower or mixed case is + handled as a regular token. + + 3) The '-' token prefix is not supported. Instead, a new binary operator, + NOT, is included. The NOT operator requires that the query specified + as its left-hand operator matches, but that the query specified as the + right-hand operator does not. For example, to query for all rows that + contain the term "example" but not the term "simple", the following + query could be used: + + MATCH 'example NOT simple' + + As for all other operators, the NOT operator must be specified in + upper case. Otherwise it will be treated as a regular token. + + 4) Unlike in the standard syntax, where the OR operator has a higher + precedence than the implicit AND operator, when using the enhanced + syntax implicit and explict AND operators have a higher precedence + than OR operators. Using the enhanced syntax, the following two + queries are equivalent: + + MATCH 'sqlite fantastic OR impressive' + MATCH '(sqlite AND fantastic) OR impressive' + + however, when using the standard syntax, the query: + + MATCH 'sqlite fantastic OR impressive' + + is equivalent to the enhanced syntax query: + + MATCH 'sqlite AND (fantastic OR impressive)' + + The precedence of all enhanced syntax operators, in order from highest + to lowest, is: + + NEAR (highest precedence, tightest grouping) + NOT + AND + OR (lowest precedence, loosest grouping) + + Using the advanced syntax, it is possible to specify expressions enclosed + in parenthesis as operands to the NOT, AND and OR operators. However both + the left and right hand side operands of NEAR operators must be either + tokens or phrases. Attempting the following query will return an error: + + MATCH 'sqlite NEAR (fantastic OR impressive)' + + Queries of this form must be re-written as: + + MATCH 'sqlite NEAR fantastic OR sqlite NEAR impressive' diff --git a/ext/fts3/README.tokenizers b/ext/fts3/README.tokenizers new file mode 100644 index 0000000..e06803a --- /dev/null +++ b/ext/fts3/README.tokenizers @@ -0,0 +1,133 @@ + +1. FTS3 Tokenizers + + When creating a new full-text table, FTS3 allows the user to select + the text tokenizer implementation to be used when indexing text + by specifying a "tokenize" clause as part of the CREATE VIRTUAL TABLE + statement: + + CREATE VIRTUAL TABLE USING fts3( + [, tokenize []] + ); + + The built-in tokenizers (valid values to pass as ) are + "simple" and "porter". + + should consist of zero or more white-space separated + arguments to pass to the selected tokenizer implementation. The + interpretation of the arguments, if any, depends on the individual + tokenizer. + +2. Custom Tokenizers + + FTS3 allows users to provide custom tokenizer implementations. The + interface used to create a new tokenizer is defined and described in + the fts3_tokenizer.h source file. + + Registering a new FTS3 tokenizer is similar to registering a new + virtual table module with SQLite. The user passes a pointer to a + structure containing pointers to various callback functions that + make up the implementation of the new tokenizer type. For tokenizers, + the structure (defined in fts3_tokenizer.h) is called + "sqlite3_tokenizer_module". + + FTS3 does not expose a C-function that users call to register new + tokenizer types with a database handle. Instead, the pointer must + be encoded as an SQL blob value and passed to FTS3 through the SQL + engine by evaluating a special scalar function, "fts3_tokenizer()". + The fts3_tokenizer() function may be called with one or two arguments, + as follows: + + SELECT fts3_tokenizer(); + SELECT fts3_tokenizer(, ); + + Where is a string identifying the tokenizer and + is a pointer to an sqlite3_tokenizer_module + structure encoded as an SQL blob. If the second argument is present, + it is registered as tokenizer and a copy of it + returned. If only one argument is passed, a pointer to the tokenizer + implementation currently registered as is returned, + encoded as a blob. Or, if no such tokenizer exists, an SQL exception + (error) is raised. + + SECURITY: If the fts3 extension is used in an environment where potentially + malicious users may execute arbitrary SQL (i.e. gears), they should be + prevented from invoking the fts3_tokenizer() function, possibly using the + authorisation callback. + + See "Sample code" below for an example of calling the fts3_tokenizer() + function from C code. + +3. ICU Library Tokenizers + + If this extension is compiled with the SQLITE_ENABLE_ICU pre-processor + symbol defined, then there exists a built-in tokenizer named "icu" + implemented using the ICU library. The first argument passed to the + xCreate() method (see fts3_tokenizer.h) of this tokenizer may be + an ICU locale identifier. For example "tr_TR" for Turkish as used + in Turkey, or "en_AU" for English as used in Australia. For example: + + "CREATE VIRTUAL TABLE thai_text USING fts3(text, tokenizer icu th_TH)" + + The ICU tokenizer implementation is very simple. It splits the input + text according to the ICU rules for finding word boundaries and discards + any tokens that consist entirely of white-space. This may be suitable + for some applications in some locales, but not all. If more complex + processing is required, for example to implement stemming or + discard punctuation, this can be done by creating a tokenizer + implementation that uses the ICU tokenizer as part of its implementation. + + When using the ICU tokenizer this way, it is safe to overwrite the + contents of the strings returned by the xNext() method (see + fts3_tokenizer.h). + +4. Sample code. + + The following two code samples illustrate the way C code should invoke + the fts3_tokenizer() scalar function: + + int registerTokenizer( + sqlite3 *db, + char *zName, + const sqlite3_tokenizer_module *p + ){ + int rc; + sqlite3_stmt *pStmt; + const char zSql[] = "SELECT fts3_tokenizer(?, ?)"; + + rc = sqlite3_prepare_v2(db, zSql, -1, &pStmt, 0); + if( rc!=SQLITE_OK ){ + return rc; + } + + sqlite3_bind_text(pStmt, 1, zName, -1, SQLITE_STATIC); + sqlite3_bind_blob(pStmt, 2, &p, sizeof(p), SQLITE_STATIC); + sqlite3_step(pStmt); + + return sqlite3_finalize(pStmt); + } + + int queryTokenizer( + sqlite3 *db, + char *zName, + const sqlite3_tokenizer_module **pp + ){ + int rc; + sqlite3_stmt *pStmt; + const char zSql[] = "SELECT fts3_tokenizer(?)"; + + *pp = 0; + rc = sqlite3_prepare_v2(db, zSql, -1, &pStmt, 0); + if( rc!=SQLITE_OK ){ + return rc; + } + + sqlite3_bind_text(pStmt, 1, zName, -1, SQLITE_STATIC); + if( SQLITE_ROW==sqlite3_step(pStmt) ){ + if( sqlite3_column_type(pStmt, 0)==SQLITE_BLOB ){ + memcpy(pp, sqlite3_column_blob(pStmt, 0), sizeof(*pp)); + } + } + + return sqlite3_finalize(pStmt); + } diff --git a/ext/fts3/README.txt b/ext/fts3/README.txt new file mode 100644 index 0000000..517a2a0 --- /dev/null +++ b/ext/fts3/README.txt @@ -0,0 +1,4 @@ +This folder contains source code to the second full-text search +extension for SQLite. While the API is the same, this version uses a +substantially different storage schema from fts1, so tables will need +to be rebuilt. diff --git a/ext/fts3/fts3.c b/ext/fts3/fts3.c new file mode 100644 index 0000000..12013f2 --- /dev/null +++ b/ext/fts3/fts3.c @@ -0,0 +1,5092 @@ +/* +** 2006 Oct 10 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This is an SQLite module implementing full-text search. +*/ + +/* +** The code in this file is only compiled if: +** +** * The FTS3 module is being built as an extension +** (in which case SQLITE_CORE is not defined), or +** +** * The FTS3 module is being built into the core of +** SQLite (in which case SQLITE_ENABLE_FTS3 is defined). +*/ + +/* The full-text index is stored in a series of b+tree (-like) +** structures called segments which map terms to doclists. The +** structures are like b+trees in layout, but are constructed from the +** bottom up in optimal fashion and are not updatable. Since trees +** are built from the bottom up, things will be described from the +** bottom up. +** +** +**** Varints **** +** The basic unit of encoding is a variable-length integer called a +** varint. We encode variable-length integers in little-endian order +** using seven bits * per byte as follows: +** +** KEY: +** A = 0xxxxxxx 7 bits of data and one flag bit +** B = 1xxxxxxx 7 bits of data and one flag bit +** +** 7 bits - A +** 14 bits - BA +** 21 bits - BBA +** and so on. +** +** This is similar in concept to how sqlite encodes "varints" but +** the encoding is not the same. SQLite varints are big-endian +** are are limited to 9 bytes in length whereas FTS3 varints are +** little-endian and can be up to 10 bytes in length (in theory). +** +** Example encodings: +** +** 1: 0x01 +** 127: 0x7f +** 128: 0x81 0x00 +** +** +**** Document lists **** +** A doclist (document list) holds a docid-sorted list of hits for a +** given term. Doclists hold docids and associated token positions. +** A docid is the unique integer identifier for a single document. +** A position is the index of a word within the document. The first +** word of the document has a position of 0. +** +** FTS3 used to optionally store character offsets using a compile-time +** option. But that functionality is no longer supported. +** +** A doclist is stored like this: +** +** array { +** varint docid; +** array { (position list for column 0) +** varint position; (2 more than the delta from previous position) +** } +** array { +** varint POS_COLUMN; (marks start of position list for new column) +** varint column; (index of new column) +** array { +** varint position; (2 more than the delta from previous position) +** } +** } +** varint POS_END; (marks end of positions for this document. +** } +** +** Here, array { X } means zero or more occurrences of X, adjacent in +** memory. A "position" is an index of a token in the token stream +** generated by the tokenizer. Note that POS_END and POS_COLUMN occur +** in the same logical place as the position element, and act as sentinals +** ending a position list array. POS_END is 0. POS_COLUMN is 1. +** The positions numbers are not stored literally but rather as two more +** than the difference from the prior position, or the just the position plus +** 2 for the first position. Example: +** +** label: A B C D E F G H I J K +** value: 123 5 9 1 1 14 35 0 234 72 0 +** +** The 123 value is the first docid. For column zero in this document +** there are two matches at positions 3 and 10 (5-2 and 9-2+3). The 1 +** at D signals the start of a new column; the 1 at E indicates that the +** new column is column number 1. There are two positions at 12 and 45 +** (14-2 and 35-2+12). The 0 at H indicate the end-of-document. The +** 234 at I is the next docid. It has one position 72 (72-2) and then +** terminates with the 0 at K. +** +** A "position-list" is the list of positions for multiple columns for +** a single docid. A "column-list" is the set of positions for a single +** column. Hence, a position-list consists of one or more column-lists, +** a document record consists of a docid followed by a position-list and +** a doclist consists of one or more document records. +** +** A bare doclist omits the position information, becoming an +** array of varint-encoded docids. +** +**** Segment leaf nodes **** +** Segment leaf nodes store terms and doclists, ordered by term. Leaf +** nodes are written using LeafWriter, and read using LeafReader (to +** iterate through a single leaf node's data) and LeavesReader (to +** iterate through a segment's entire leaf layer). Leaf nodes have +** the format: +** +** varint iHeight; (height from leaf level, always 0) +** varint nTerm; (length of first term) +** char pTerm[nTerm]; (content of first term) +** varint nDoclist; (length of term's associated doclist) +** char pDoclist[nDoclist]; (content of doclist) +** array { +** (further terms are delta-encoded) +** varint nPrefix; (length of prefix shared with previous term) +** varint nSuffix; (length of unshared suffix) +** char pTermSuffix[nSuffix];(unshared suffix of next term) +** varint nDoclist; (length of term's associated doclist) +** char pDoclist[nDoclist]; (content of doclist) +** } +** +** Here, array { X } means zero or more occurrences of X, adjacent in +** memory. +** +** Leaf nodes are broken into blocks which are stored contiguously in +** the %_segments table in sorted order. This means that when the end +** of a node is reached, the next term is in the node with the next +** greater node id. +** +** New data is spilled to a new leaf node when the current node +** exceeds LEAF_MAX bytes (default 2048). New data which itself is +** larger than STANDALONE_MIN (default 1024) is placed in a standalone +** node (a leaf node with a single term and doclist). The goal of +** these settings is to pack together groups of small doclists while +** making it efficient to directly access large doclists. The +** assumption is that large doclists represent terms which are more +** likely to be query targets. +** +** TODO(shess) It may be useful for blocking decisions to be more +** dynamic. For instance, it may make more sense to have a 2.5k leaf +** node rather than splitting into 2k and .5k nodes. My intuition is +** that this might extend through 2x or 4x the pagesize. +** +** +**** Segment interior nodes **** +** Segment interior nodes store blockids for subtree nodes and terms +** to describe what data is stored by the each subtree. Interior +** nodes are written using InteriorWriter, and read using +** InteriorReader. InteriorWriters are created as needed when +** SegmentWriter creates new leaf nodes, or when an interior node +** itself grows too big and must be split. The format of interior +** nodes: +** +** varint iHeight; (height from leaf level, always >0) +** varint iBlockid; (block id of node's leftmost subtree) +** optional { +** varint nTerm; (length of first term) +** char pTerm[nTerm]; (content of first term) +** array { +** (further terms are delta-encoded) +** varint nPrefix; (length of shared prefix with previous term) +** varint nSuffix; (length of unshared suffix) +** char pTermSuffix[nSuffix]; (unshared suffix of next term) +** } +** } +** +** Here, optional { X } means an optional element, while array { X } +** means zero or more occurrences of X, adjacent in memory. +** +** An interior node encodes n terms separating n+1 subtrees. The +** subtree blocks are contiguous, so only the first subtree's blockid +** is encoded. The subtree at iBlockid will contain all terms less +** than the first term encoded (or all terms if no term is encoded). +** Otherwise, for terms greater than or equal to pTerm[i] but less +** than pTerm[i+1], the subtree for that term will be rooted at +** iBlockid+i. Interior nodes only store enough term data to +** distinguish adjacent children (if the rightmost term of the left +** child is "something", and the leftmost term of the right child is +** "wicked", only "w" is stored). +** +** New data is spilled to a new interior node at the same height when +** the current node exceeds INTERIOR_MAX bytes (default 2048). +** INTERIOR_MIN_TERMS (default 7) keeps large terms from monopolizing +** interior nodes and making the tree too skinny. The interior nodes +** at a given height are naturally tracked by interior nodes at +** height+1, and so on. +** +** +**** Segment directory **** +** The segment directory in table %_segdir stores meta-information for +** merging and deleting segments, and also the root node of the +** segment's tree. +** +** The root node is the top node of the segment's tree after encoding +** the entire segment, restricted to ROOT_MAX bytes (default 1024). +** This could be either a leaf node or an interior node. If the top +** node requires more than ROOT_MAX bytes, it is flushed to %_segments +** and a new root interior node is generated (which should always fit +** within ROOT_MAX because it only needs space for 2 varints, the +** height and the blockid of the previous root). +** +** The meta-information in the segment directory is: +** level - segment level (see below) +** idx - index within level +** - (level,idx uniquely identify a segment) +** start_block - first leaf node +** leaves_end_block - last leaf node +** end_block - last block (including interior nodes) +** root - contents of root node +** +** If the root node is a leaf node, then start_block, +** leaves_end_block, and end_block are all 0. +** +** +**** Segment merging **** +** To amortize update costs, segments are grouped into levels and +** merged in batches. Each increase in level represents exponentially +** more documents. +** +** New documents (actually, document updates) are tokenized and +** written individually (using LeafWriter) to a level 0 segment, with +** incrementing idx. When idx reaches MERGE_COUNT (default 16), all +** level 0 segments are merged into a single level 1 segment. Level 1 +** is populated like level 0, and eventually MERGE_COUNT level 1 +** segments are merged to a single level 2 segment (representing +** MERGE_COUNT^2 updates), and so on. +** +** A segment merge traverses all segments at a given level in +** parallel, performing a straightforward sorted merge. Since segment +** leaf nodes are written in to the %_segments table in order, this +** merge traverses the underlying sqlite disk structures efficiently. +** After the merge, all segment blocks from the merged level are +** deleted. +** +** MERGE_COUNT controls how often we merge segments. 16 seems to be +** somewhat of a sweet spot for insertion performance. 32 and 64 show +** very similar performance numbers to 16 on insertion, though they're +** a tiny bit slower (perhaps due to more overhead in merge-time +** sorting). 8 is about 20% slower than 16, 4 about 50% slower than +** 16, 2 about 66% slower than 16. +** +** At query time, high MERGE_COUNT increases the number of segments +** which need to be scanned and merged. For instance, with 100k docs +** inserted: +** +** MERGE_COUNT segments +** 16 25 +** 8 12 +** 4 10 +** 2 6 +** +** This appears to have only a moderate impact on queries for very +** frequent terms (which are somewhat dominated by segment merge +** costs), and infrequent and non-existent terms still seem to be fast +** even with many segments. +** +** TODO(shess) That said, it would be nice to have a better query-side +** argument for MERGE_COUNT of 16. Also, it is possible/likely that +** optimizations to things like doclist merging will swing the sweet +** spot around. +** +** +** +**** Handling of deletions and updates **** +** Since we're using a segmented structure, with no docid-oriented +** index into the term index, we clearly cannot simply update the term +** index when a document is deleted or updated. For deletions, we +** write an empty doclist (varint(docid) varint(POS_END)), for updates +** we simply write the new doclist. Segment merges overwrite older +** data for a particular docid with newer data, so deletes or updates +** will eventually overtake the earlier data and knock it out. The +** query logic likewise merges doclists so that newer data knocks out +** older data. +** +** TODO(shess) Provide a VACUUM type operation to clear out all +** deletions and duplications. This would basically be a forced merge +** into a single segment. +*/ + +#include "fts3Int.h" +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) + +#if defined(SQLITE_ENABLE_FTS3) && !defined(SQLITE_CORE) +# define SQLITE_CORE 1 +#endif + +#include +#include +#include +#include +#include +#include + +#include "fts3.h" +#ifndef SQLITE_CORE +# include "sqlite3ext.h" + SQLITE_EXTENSION_INIT1 +#endif + +static int fts3EvalNext(Fts3Cursor *pCsr); +static int fts3EvalStart(Fts3Cursor *pCsr); +static int fts3TermSegReaderCursor( + Fts3Cursor *, const char *, int, int, Fts3MultiSegReader **); + +/* +** Write a 64-bit variable-length integer to memory starting at p[0]. +** The length of data written will be between 1 and FTS3_VARINT_MAX bytes. +** The number of bytes written is returned. +*/ +int sqlite3Fts3PutVarint(char *p, sqlite_int64 v){ + unsigned char *q = (unsigned char *) p; + sqlite_uint64 vu = v; + do{ + *q++ = (unsigned char) ((vu & 0x7f) | 0x80); + vu >>= 7; + }while( vu!=0 ); + q[-1] &= 0x7f; /* turn off high bit in final byte */ + assert( q - (unsigned char *)p <= FTS3_VARINT_MAX ); + return (int) (q - (unsigned char *)p); +} + +/* +** Read a 64-bit variable-length integer from memory starting at p[0]. +** Return the number of bytes read, or 0 on error. +** The value is stored in *v. +*/ +int sqlite3Fts3GetVarint(const char *p, sqlite_int64 *v){ + const unsigned char *q = (const unsigned char *) p; + sqlite_uint64 x = 0, y = 1; + while( (*q&0x80)==0x80 && q-(unsigned char *)p>= 7; + }while( v!=0 ); + return i; +} + +/* +** Convert an SQL-style quoted string into a normal string by removing +** the quote characters. The conversion is done in-place. If the +** input does not begin with a quote character, then this routine +** is a no-op. +** +** Examples: +** +** "abc" becomes abc +** 'xyz' becomes xyz +** [pqr] becomes pqr +** `mno` becomes mno +** +*/ +void sqlite3Fts3Dequote(char *z){ + char quote; /* Quote character (if any ) */ + + quote = z[0]; + if( quote=='[' || quote=='\'' || quote=='"' || quote=='`' ){ + int iIn = 1; /* Index of next byte to read from input */ + int iOut = 0; /* Index of next byte to write to output */ + + /* If the first byte was a '[', then the close-quote character is a ']' */ + if( quote=='[' ) quote = ']'; + + while( ALWAYS(z[iIn]) ){ + if( z[iIn]==quote ){ + if( z[iIn+1]!=quote ) break; + z[iOut++] = quote; + iIn += 2; + }else{ + z[iOut++] = z[iIn++]; + } + } + z[iOut] = '\0'; + } +} + +/* +** Read a single varint from the doclist at *pp and advance *pp to point +** to the first byte past the end of the varint. Add the value of the varint +** to *pVal. +*/ +static void fts3GetDeltaVarint(char **pp, sqlite3_int64 *pVal){ + sqlite3_int64 iVal; + *pp += sqlite3Fts3GetVarint(*pp, &iVal); + *pVal += iVal; +} + +/* +** When this function is called, *pp points to the first byte following a +** varint that is part of a doclist (or position-list, or any other list +** of varints). This function moves *pp to point to the start of that varint, +** and sets *pVal by the varint value. +** +** Argument pStart points to the first byte of the doclist that the +** varint is part of. +*/ +static void fts3GetReverseVarint( + char **pp, + char *pStart, + sqlite3_int64 *pVal +){ + sqlite3_int64 iVal; + char *p; + + /* Pointer p now points at the first byte past the varint we are + ** interested in. So, unless the doclist is corrupt, the 0x80 bit is + ** clear on character p[-1]. */ + for(p = (*pp)-2; p>=pStart && *p&0x80; p--); + p++; + *pp = p; + + sqlite3Fts3GetVarint(p, &iVal); + *pVal = iVal; +} + +/* +** The xDisconnect() virtual table method. +*/ +static int fts3DisconnectMethod(sqlite3_vtab *pVtab){ + Fts3Table *p = (Fts3Table *)pVtab; + int i; + + assert( p->nPendingData==0 ); + assert( p->pSegments==0 ); + + /* Free any prepared statements held */ + for(i=0; iaStmt); i++){ + sqlite3_finalize(p->aStmt[i]); + } + sqlite3_free(p->zSegmentsTbl); + sqlite3_free(p->zReadExprlist); + sqlite3_free(p->zWriteExprlist); + sqlite3_free(p->zContentTbl); + + /* Invoke the tokenizer destructor to free the tokenizer. */ + p->pTokenizer->pModule->xDestroy(p->pTokenizer); + + sqlite3_free(p); + return SQLITE_OK; +} + +/* +** Construct one or more SQL statements from the format string given +** and then evaluate those statements. The success code is written +** into *pRc. +** +** If *pRc is initially non-zero then this routine is a no-op. +*/ +static void fts3DbExec( + int *pRc, /* Success code */ + sqlite3 *db, /* Database in which to run SQL */ + const char *zFormat, /* Format string for SQL */ + ... /* Arguments to the format string */ +){ + va_list ap; + char *zSql; + if( *pRc ) return; + va_start(ap, zFormat); + zSql = sqlite3_vmprintf(zFormat, ap); + va_end(ap); + if( zSql==0 ){ + *pRc = SQLITE_NOMEM; + }else{ + *pRc = sqlite3_exec(db, zSql, 0, 0, 0); + sqlite3_free(zSql); + } +} + +/* +** The xDestroy() virtual table method. +*/ +static int fts3DestroyMethod(sqlite3_vtab *pVtab){ + Fts3Table *p = (Fts3Table *)pVtab; + int rc = SQLITE_OK; /* Return code */ + const char *zDb = p->zDb; /* Name of database (e.g. "main", "temp") */ + sqlite3 *db = p->db; /* Database handle */ + + /* Drop the shadow tables */ + if( p->zContentTbl==0 ){ + fts3DbExec(&rc, db, "DROP TABLE IF EXISTS %Q.'%q_content'", zDb, p->zName); + } + fts3DbExec(&rc, db, "DROP TABLE IF EXISTS %Q.'%q_segments'", zDb,p->zName); + fts3DbExec(&rc, db, "DROP TABLE IF EXISTS %Q.'%q_segdir'", zDb, p->zName); + fts3DbExec(&rc, db, "DROP TABLE IF EXISTS %Q.'%q_docsize'", zDb, p->zName); + fts3DbExec(&rc, db, "DROP TABLE IF EXISTS %Q.'%q_stat'", zDb, p->zName); + + /* If everything has worked, invoke fts3DisconnectMethod() to free the + ** memory associated with the Fts3Table structure and return SQLITE_OK. + ** Otherwise, return an SQLite error code. + */ + return (rc==SQLITE_OK ? fts3DisconnectMethod(pVtab) : rc); +} + + +/* +** Invoke sqlite3_declare_vtab() to declare the schema for the FTS3 table +** passed as the first argument. This is done as part of the xConnect() +** and xCreate() methods. +** +** If *pRc is non-zero when this function is called, it is a no-op. +** Otherwise, if an error occurs, an SQLite error code is stored in *pRc +** before returning. +*/ +static void fts3DeclareVtab(int *pRc, Fts3Table *p){ + if( *pRc==SQLITE_OK ){ + int i; /* Iterator variable */ + int rc; /* Return code */ + char *zSql; /* SQL statement passed to declare_vtab() */ + char *zCols; /* List of user defined columns */ + + sqlite3_vtab_config(p->db, SQLITE_VTAB_CONSTRAINT_SUPPORT, 1); + + /* Create a list of user columns for the virtual table */ + zCols = sqlite3_mprintf("%Q, ", p->azColumn[0]); + for(i=1; zCols && inColumn; i++){ + zCols = sqlite3_mprintf("%z%Q, ", zCols, p->azColumn[i]); + } + + /* Create the whole "CREATE TABLE" statement to pass to SQLite */ + zSql = sqlite3_mprintf( + "CREATE TABLE x(%s %Q HIDDEN, docid HIDDEN)", zCols, p->zName + ); + if( !zCols || !zSql ){ + rc = SQLITE_NOMEM; + }else{ + rc = sqlite3_declare_vtab(p->db, zSql); + } + + sqlite3_free(zSql); + sqlite3_free(zCols); + *pRc = rc; + } +} + +/* +** Create the backing store tables (%_content, %_segments and %_segdir) +** required by the FTS3 table passed as the only argument. This is done +** as part of the vtab xCreate() method. +** +** If the p->bHasDocsize boolean is true (indicating that this is an +** FTS4 table, not an FTS3 table) then also create the %_docsize and +** %_stat tables required by FTS4. +*/ +static int fts3CreateTables(Fts3Table *p){ + int rc = SQLITE_OK; /* Return code */ + int i; /* Iterator variable */ + sqlite3 *db = p->db; /* The database connection */ + + if( p->zContentTbl==0 ){ + char *zContentCols; /* Columns of %_content table */ + + /* Create a list of user columns for the content table */ + zContentCols = sqlite3_mprintf("docid INTEGER PRIMARY KEY"); + for(i=0; zContentCols && inColumn; i++){ + char *z = p->azColumn[i]; + zContentCols = sqlite3_mprintf("%z, 'c%d%q'", zContentCols, i, z); + } + if( zContentCols==0 ) rc = SQLITE_NOMEM; + + /* Create the content table */ + fts3DbExec(&rc, db, + "CREATE TABLE %Q.'%q_content'(%s)", + p->zDb, p->zName, zContentCols + ); + sqlite3_free(zContentCols); + } + + /* Create other tables */ + fts3DbExec(&rc, db, + "CREATE TABLE %Q.'%q_segments'(blockid INTEGER PRIMARY KEY, block BLOB);", + p->zDb, p->zName + ); + fts3DbExec(&rc, db, + "CREATE TABLE %Q.'%q_segdir'(" + "level INTEGER," + "idx INTEGER," + "start_block INTEGER," + "leaves_end_block INTEGER," + "end_block INTEGER," + "root BLOB," + "PRIMARY KEY(level, idx)" + ");", + p->zDb, p->zName + ); + if( p->bHasDocsize ){ + fts3DbExec(&rc, db, + "CREATE TABLE %Q.'%q_docsize'(docid INTEGER PRIMARY KEY, size BLOB);", + p->zDb, p->zName + ); + } + if( p->bHasStat ){ + fts3DbExec(&rc, db, + "CREATE TABLE %Q.'%q_stat'(id INTEGER PRIMARY KEY, value BLOB);", + p->zDb, p->zName + ); + } + return rc; +} + +/* +** Store the current database page-size in bytes in p->nPgsz. +** +** If *pRc is non-zero when this function is called, it is a no-op. +** Otherwise, if an error occurs, an SQLite error code is stored in *pRc +** before returning. +*/ +static void fts3DatabasePageSize(int *pRc, Fts3Table *p){ + if( *pRc==SQLITE_OK ){ + int rc; /* Return code */ + char *zSql; /* SQL text "PRAGMA %Q.page_size" */ + sqlite3_stmt *pStmt; /* Compiled "PRAGMA %Q.page_size" statement */ + + zSql = sqlite3_mprintf("PRAGMA %Q.page_size", p->zDb); + if( !zSql ){ + rc = SQLITE_NOMEM; + }else{ + rc = sqlite3_prepare(p->db, zSql, -1, &pStmt, 0); + if( rc==SQLITE_OK ){ + sqlite3_step(pStmt); + p->nPgsz = sqlite3_column_int(pStmt, 0); + rc = sqlite3_finalize(pStmt); + }else if( rc==SQLITE_AUTH ){ + p->nPgsz = 1024; + rc = SQLITE_OK; + } + } + assert( p->nPgsz>0 || rc!=SQLITE_OK ); + sqlite3_free(zSql); + *pRc = rc; + } +} + +/* +** "Special" FTS4 arguments are column specifications of the following form: +** +** = +** +** There may not be whitespace surrounding the "=" character. The +** term may be quoted, but the may not. +*/ +static int fts3IsSpecialColumn( + const char *z, + int *pnKey, + char **pzValue +){ + char *zValue; + const char *zCsr = z; + + while( *zCsr!='=' ){ + if( *zCsr=='\0' ) return 0; + zCsr++; + } + + *pnKey = (int)(zCsr-z); + zValue = sqlite3_mprintf("%s", &zCsr[1]); + if( zValue ){ + sqlite3Fts3Dequote(zValue); + } + *pzValue = zValue; + return 1; +} + +/* +** Append the output of a printf() style formatting to an existing string. +*/ +static void fts3Appendf( + int *pRc, /* IN/OUT: Error code */ + char **pz, /* IN/OUT: Pointer to string buffer */ + const char *zFormat, /* Printf format string to append */ + ... /* Arguments for printf format string */ +){ + if( *pRc==SQLITE_OK ){ + va_list ap; + char *z; + va_start(ap, zFormat); + z = sqlite3_vmprintf(zFormat, ap); + if( z && *pz ){ + char *z2 = sqlite3_mprintf("%s%s", *pz, z); + sqlite3_free(z); + z = z2; + } + if( z==0 ) *pRc = SQLITE_NOMEM; + sqlite3_free(*pz); + *pz = z; + } +} + +/* +** Return a copy of input string zInput enclosed in double-quotes (") and +** with all double quote characters escaped. For example: +** +** fts3QuoteId("un \"zip\"") -> "un \"\"zip\"\"" +** +** The pointer returned points to memory obtained from sqlite3_malloc(). It +** is the callers responsibility to call sqlite3_free() to release this +** memory. +*/ +static char *fts3QuoteId(char const *zInput){ + int nRet; + char *zRet; + nRet = 2 + strlen(zInput)*2 + 1; + zRet = sqlite3_malloc(nRet); + if( zRet ){ + int i; + char *z = zRet; + *(z++) = '"'; + for(i=0; zInput[i]; i++){ + if( zInput[i]=='"' ) *(z++) = '"'; + *(z++) = zInput[i]; + } + *(z++) = '"'; + *(z++) = '\0'; + } + return zRet; +} + +/* +** Return a list of comma separated SQL expressions and a FROM clause that +** could be used in a SELECT statement such as the following: +** +** SELECT FROM %_content AS x ... +** +** to return the docid, followed by each column of text data in order +** from left to write. If parameter zFunc is not NULL, then instead of +** being returned directly each column of text data is passed to an SQL +** function named zFunc first. For example, if zFunc is "unzip" and the +** table has the three user-defined columns "a", "b", and "c", the following +** string is returned: +** +** "docid, unzip(x.'a'), unzip(x.'b'), unzip(x.'c') FROM %_content AS x" +** +** The pointer returned points to a buffer allocated by sqlite3_malloc(). It +** is the responsibility of the caller to eventually free it. +** +** If *pRc is not SQLITE_OK when this function is called, it is a no-op (and +** a NULL pointer is returned). Otherwise, if an OOM error is encountered +** by this function, NULL is returned and *pRc is set to SQLITE_NOMEM. If +** no error occurs, *pRc is left unmodified. +*/ +static char *fts3ReadExprList(Fts3Table *p, const char *zFunc, int *pRc){ + char *zRet = 0; + char *zFree = 0; + char *zFunction; + int i; + + if( p->zContentTbl==0 ){ + if( !zFunc ){ + zFunction = ""; + }else{ + zFree = zFunction = fts3QuoteId(zFunc); + } + fts3Appendf(pRc, &zRet, "docid"); + for(i=0; inColumn; i++){ + fts3Appendf(pRc, &zRet, ",%s(x.'c%d%q')", zFunction, i, p->azColumn[i]); + } + sqlite3_free(zFree); + }else{ + fts3Appendf(pRc, &zRet, "rowid"); + for(i=0; inColumn; i++){ + fts3Appendf(pRc, &zRet, ", x.'%q'", p->azColumn[i]); + } + } + fts3Appendf(pRc, &zRet, "FROM '%q'.'%q%s' AS x", + p->zDb, + (p->zContentTbl ? p->zContentTbl : p->zName), + (p->zContentTbl ? "" : "_content") + ); + return zRet; +} + +/* +** Return a list of N comma separated question marks, where N is the number +** of columns in the %_content table (one for the docid plus one for each +** user-defined text column). +** +** If argument zFunc is not NULL, then all but the first question mark +** is preceded by zFunc and an open bracket, and followed by a closed +** bracket. For example, if zFunc is "zip" and the FTS3 table has three +** user-defined text columns, the following string is returned: +** +** "?, zip(?), zip(?), zip(?)" +** +** The pointer returned points to a buffer allocated by sqlite3_malloc(). It +** is the responsibility of the caller to eventually free it. +** +** If *pRc is not SQLITE_OK when this function is called, it is a no-op (and +** a NULL pointer is returned). Otherwise, if an OOM error is encountered +** by this function, NULL is returned and *pRc is set to SQLITE_NOMEM. If +** no error occurs, *pRc is left unmodified. +*/ +static char *fts3WriteExprList(Fts3Table *p, const char *zFunc, int *pRc){ + char *zRet = 0; + char *zFree = 0; + char *zFunction; + int i; + + if( !zFunc ){ + zFunction = ""; + }else{ + zFree = zFunction = fts3QuoteId(zFunc); + } + fts3Appendf(pRc, &zRet, "?"); + for(i=0; inColumn; i++){ + fts3Appendf(pRc, &zRet, ",%s(?)", zFunction); + } + sqlite3_free(zFree); + return zRet; +} + +/* +** This function interprets the string at (*pp) as a non-negative integer +** value. It reads the integer and sets *pnOut to the value read, then +** sets *pp to point to the byte immediately following the last byte of +** the integer value. +** +** Only decimal digits ('0'..'9') may be part of an integer value. +** +** If *pp does not being with a decimal digit SQLITE_ERROR is returned and +** the output value undefined. Otherwise SQLITE_OK is returned. +** +** This function is used when parsing the "prefix=" FTS4 parameter. +*/ +static int fts3GobbleInt(const char **pp, int *pnOut){ + const char *p; /* Iterator pointer */ + int nInt = 0; /* Output value */ + + for(p=*pp; p[0]>='0' && p[0]<='9'; p++){ + nInt = nInt * 10 + (p[0] - '0'); + } + if( p==*pp ) return SQLITE_ERROR; + *pnOut = nInt; + *pp = p; + return SQLITE_OK; +} + +/* +** This function is called to allocate an array of Fts3Index structures +** representing the indexes maintained by the current FTS table. FTS tables +** always maintain the main "terms" index, but may also maintain one or +** more "prefix" indexes, depending on the value of the "prefix=" parameter +** (if any) specified as part of the CREATE VIRTUAL TABLE statement. +** +** Argument zParam is passed the value of the "prefix=" option if one was +** specified, or NULL otherwise. +** +** If no error occurs, SQLITE_OK is returned and *apIndex set to point to +** the allocated array. *pnIndex is set to the number of elements in the +** array. If an error does occur, an SQLite error code is returned. +** +** Regardless of whether or not an error is returned, it is the responsibility +** of the caller to call sqlite3_free() on the output array to free it. +*/ +static int fts3PrefixParameter( + const char *zParam, /* ABC in prefix=ABC parameter to parse */ + int *pnIndex, /* OUT: size of *apIndex[] array */ + struct Fts3Index **apIndex /* OUT: Array of indexes for this table */ +){ + struct Fts3Index *aIndex; /* Allocated array */ + int nIndex = 1; /* Number of entries in array */ + + if( zParam && zParam[0] ){ + const char *p; + nIndex++; + for(p=zParam; *p; p++){ + if( *p==',' ) nIndex++; + } + } + + aIndex = sqlite3_malloc(sizeof(struct Fts3Index) * nIndex); + *apIndex = aIndex; + *pnIndex = nIndex; + if( !aIndex ){ + return SQLITE_NOMEM; + } + + memset(aIndex, 0, sizeof(struct Fts3Index) * nIndex); + if( zParam ){ + const char *p = zParam; + int i; + for(i=1; i module name ("fts3" or "fts4") +** argv[1] -> database name +** argv[2] -> table name +** argv[...] -> "column name" and other module argument fields. +*/ +static int fts3InitVtab( + int isCreate, /* True for xCreate, false for xConnect */ + sqlite3 *db, /* The SQLite database connection */ + void *pAux, /* Hash table containing tokenizers */ + int argc, /* Number of elements in argv array */ + const char * const *argv, /* xCreate/xConnect argument array */ + sqlite3_vtab **ppVTab, /* Write the resulting vtab structure here */ + char **pzErr /* Write any error message here */ +){ + Fts3Hash *pHash = (Fts3Hash *)pAux; + Fts3Table *p = 0; /* Pointer to allocated vtab */ + int rc = SQLITE_OK; /* Return code */ + int i; /* Iterator variable */ + int nByte; /* Size of allocation used for *p */ + int iCol; /* Column index */ + int nString = 0; /* Bytes required to hold all column names */ + int nCol = 0; /* Number of columns in the FTS table */ + char *zCsr; /* Space for holding column names */ + int nDb; /* Bytes required to hold database name */ + int nName; /* Bytes required to hold table name */ + int isFts4 = (argv[0][3]=='4'); /* True for FTS4, false for FTS3 */ + const char **aCol; /* Array of column names */ + sqlite3_tokenizer *pTokenizer = 0; /* Tokenizer for this table */ + + int nIndex; /* Size of aIndex[] array */ + struct Fts3Index *aIndex = 0; /* Array of indexes for this table */ + + /* The results of parsing supported FTS4 key=value options: */ + int bNoDocsize = 0; /* True to omit %_docsize table */ + int bDescIdx = 0; /* True to store descending indexes */ + char *zPrefix = 0; /* Prefix parameter value (or NULL) */ + char *zCompress = 0; /* compress=? parameter (or NULL) */ + char *zUncompress = 0; /* uncompress=? parameter (or NULL) */ + char *zContent = 0; /* content=? parameter (or NULL) */ + + assert( strlen(argv[0])==4 ); + assert( (sqlite3_strnicmp(argv[0], "fts4", 4)==0 && isFts4) + || (sqlite3_strnicmp(argv[0], "fts3", 4)==0 && !isFts4) + ); + + nDb = (int)strlen(argv[1]) + 1; + nName = (int)strlen(argv[2]) + 1; + + aCol = (const char **)sqlite3_malloc(sizeof(const char *) * (argc-2) ); + if( !aCol ) return SQLITE_NOMEM; + memset((void *)aCol, 0, sizeof(const char *) * (argc-2)); + + /* Loop through all of the arguments passed by the user to the FTS3/4 + ** module (i.e. all the column names and special arguments). This loop + ** does the following: + ** + ** + Figures out the number of columns the FTSX table will have, and + ** the number of bytes of space that must be allocated to store copies + ** of the column names. + ** + ** + If there is a tokenizer specification included in the arguments, + ** initializes the tokenizer pTokenizer. + */ + for(i=3; rc==SQLITE_OK && i8 + && 0==sqlite3_strnicmp(z, "tokenize", 8) + && 0==sqlite3Fts3IsIdChar(z[8]) + ){ + rc = sqlite3Fts3InitTokenizer(pHash, &z[9], &pTokenizer, pzErr); + } + + /* Check if it is an FTS4 special argument. */ + else if( isFts4 && fts3IsSpecialColumn(z, &nKey, &zVal) ){ + struct Fts4Option { + const char *zOpt; + int nOpt; + } aFts4Opt[] = { + { "matchinfo", 9 }, /* 0 -> MATCHINFO */ + { "prefix", 6 }, /* 1 -> PREFIX */ + { "compress", 8 }, /* 2 -> COMPRESS */ + { "uncompress", 10 }, /* 3 -> UNCOMPRESS */ + { "order", 5 }, /* 4 -> ORDER */ + { "content", 7 } /* 5 -> CONTENT */ + }; + + int iOpt; + if( !zVal ){ + rc = SQLITE_NOMEM; + }else{ + for(iOpt=0; iOptnOpt && !sqlite3_strnicmp(z, pOp->zOpt, pOp->nOpt) ){ + break; + } + } + if( iOpt==SizeofArray(aFts4Opt) ){ + *pzErr = sqlite3_mprintf("unrecognized parameter: %s", z); + rc = SQLITE_ERROR; + }else{ + switch( iOpt ){ + case 0: /* MATCHINFO */ + if( strlen(zVal)!=4 || sqlite3_strnicmp(zVal, "fts3", 4) ){ + *pzErr = sqlite3_mprintf("unrecognized matchinfo: %s", zVal); + rc = SQLITE_ERROR; + } + bNoDocsize = 1; + break; + + case 1: /* PREFIX */ + sqlite3_free(zPrefix); + zPrefix = zVal; + zVal = 0; + break; + + case 2: /* COMPRESS */ + sqlite3_free(zCompress); + zCompress = zVal; + zVal = 0; + break; + + case 3: /* UNCOMPRESS */ + sqlite3_free(zUncompress); + zUncompress = zVal; + zVal = 0; + break; + + case 4: /* ORDER */ + if( (strlen(zVal)!=3 || sqlite3_strnicmp(zVal, "asc", 3)) + && (strlen(zVal)!=4 || sqlite3_strnicmp(zVal, "desc", 4)) + ){ + *pzErr = sqlite3_mprintf("unrecognized order: %s", zVal); + rc = SQLITE_ERROR; + } + bDescIdx = (zVal[0]=='d' || zVal[0]=='D'); + break; + + default: /* CONTENT */ + assert( iOpt==5 ); + sqlite3_free(zUncompress); + zContent = zVal; + zVal = 0; + break; + } + } + sqlite3_free(zVal); + } + } + + /* Otherwise, the argument is a column name. */ + else { + nString += (int)(strlen(z) + 1); + aCol[nCol++] = z; + } + } + + /* If a content=xxx option was specified, the following: + ** + ** 1. Ignore any compress= and uncompress= options. + ** + ** 2. If no column names were specified as part of the CREATE VIRTUAL + ** TABLE statement, use all columns from the content table. + */ + if( rc==SQLITE_OK && zContent ){ + sqlite3_free(zCompress); + sqlite3_free(zUncompress); + zCompress = 0; + zUncompress = 0; + if( nCol==0 ){ + sqlite3_free((void*)aCol); + aCol = 0; + rc = fts3ContentColumns(db, argv[1], zContent, &aCol, &nCol, &nString); + } + assert( rc!=SQLITE_OK || nCol>0 ); + } + if( rc!=SQLITE_OK ) goto fts3_init_out; + + if( nCol==0 ){ + assert( nString==0 ); + aCol[0] = "content"; + nString = 8; + nCol = 1; + } + + if( pTokenizer==0 ){ + rc = sqlite3Fts3InitTokenizer(pHash, "simple", &pTokenizer, pzErr); + if( rc!=SQLITE_OK ) goto fts3_init_out; + } + assert( pTokenizer ); + + rc = fts3PrefixParameter(zPrefix, &nIndex, &aIndex); + if( rc==SQLITE_ERROR ){ + assert( zPrefix ); + *pzErr = sqlite3_mprintf("error parsing prefix parameter: %s", zPrefix); + } + if( rc!=SQLITE_OK ) goto fts3_init_out; + + /* Allocate and populate the Fts3Table structure. */ + nByte = sizeof(Fts3Table) + /* Fts3Table */ + nCol * sizeof(char *) + /* azColumn */ + nIndex * sizeof(struct Fts3Index) + /* aIndex */ + nName + /* zName */ + nDb + /* zDb */ + nString; /* Space for azColumn strings */ + p = (Fts3Table*)sqlite3_malloc(nByte); + if( p==0 ){ + rc = SQLITE_NOMEM; + goto fts3_init_out; + } + memset(p, 0, nByte); + p->db = db; + p->nColumn = nCol; + p->nPendingData = 0; + p->azColumn = (char **)&p[1]; + p->pTokenizer = pTokenizer; + p->nMaxPendingData = FTS3_MAX_PENDING_DATA; + p->bHasDocsize = (isFts4 && bNoDocsize==0); + p->bHasStat = isFts4; + p->bDescIdx = bDescIdx; + p->zContentTbl = zContent; + zContent = 0; + TESTONLY( p->inTransaction = -1 ); + TESTONLY( p->mxSavepoint = -1 ); + + p->aIndex = (struct Fts3Index *)&p->azColumn[nCol]; + memcpy(p->aIndex, aIndex, sizeof(struct Fts3Index) * nIndex); + p->nIndex = nIndex; + for(i=0; iaIndex[i].hPending, FTS3_HASH_STRING, 1); + } + + /* Fill in the zName and zDb fields of the vtab structure. */ + zCsr = (char *)&p->aIndex[nIndex]; + p->zName = zCsr; + memcpy(zCsr, argv[2], nName); + zCsr += nName; + p->zDb = zCsr; + memcpy(zCsr, argv[1], nDb); + zCsr += nDb; + + /* Fill in the azColumn array */ + for(iCol=0; iColazColumn[iCol] = zCsr; + zCsr += n+1; + assert( zCsr <= &((char *)p)[nByte] ); + } + + if( (zCompress==0)!=(zUncompress==0) ){ + char const *zMiss = (zCompress==0 ? "compress" : "uncompress"); + rc = SQLITE_ERROR; + *pzErr = sqlite3_mprintf("missing %s parameter in fts4 constructor", zMiss); + } + p->zReadExprlist = fts3ReadExprList(p, zUncompress, &rc); + p->zWriteExprlist = fts3WriteExprList(p, zCompress, &rc); + if( rc!=SQLITE_OK ) goto fts3_init_out; + + /* If this is an xCreate call, create the underlying tables in the + ** database. TODO: For xConnect(), it could verify that said tables exist. + */ + if( isCreate ){ + rc = fts3CreateTables(p); + } + + /* Figure out the page-size for the database. This is required in order to + ** estimate the cost of loading large doclists from the database. */ + fts3DatabasePageSize(&rc, p); + p->nNodeSize = p->nPgsz-35; + + /* Declare the table schema to SQLite. */ + fts3DeclareVtab(&rc, p); + +fts3_init_out: + sqlite3_free(zPrefix); + sqlite3_free(aIndex); + sqlite3_free(zCompress); + sqlite3_free(zUncompress); + sqlite3_free(zContent); + sqlite3_free((void *)aCol); + if( rc!=SQLITE_OK ){ + if( p ){ + fts3DisconnectMethod((sqlite3_vtab *)p); + }else if( pTokenizer ){ + pTokenizer->pModule->xDestroy(pTokenizer); + } + }else{ + assert( p->pSegments==0 ); + *ppVTab = &p->base; + } + return rc; +} + +/* +** The xConnect() and xCreate() methods for the virtual table. All the +** work is done in function fts3InitVtab(). +*/ +static int fts3ConnectMethod( + sqlite3 *db, /* Database connection */ + void *pAux, /* Pointer to tokenizer hash table */ + int argc, /* Number of elements in argv array */ + const char * const *argv, /* xCreate/xConnect argument array */ + sqlite3_vtab **ppVtab, /* OUT: New sqlite3_vtab object */ + char **pzErr /* OUT: sqlite3_malloc'd error message */ +){ + return fts3InitVtab(0, db, pAux, argc, argv, ppVtab, pzErr); +} +static int fts3CreateMethod( + sqlite3 *db, /* Database connection */ + void *pAux, /* Pointer to tokenizer hash table */ + int argc, /* Number of elements in argv array */ + const char * const *argv, /* xCreate/xConnect argument array */ + sqlite3_vtab **ppVtab, /* OUT: New sqlite3_vtab object */ + char **pzErr /* OUT: sqlite3_malloc'd error message */ +){ + return fts3InitVtab(1, db, pAux, argc, argv, ppVtab, pzErr); +} + +/* +** Implementation of the xBestIndex method for FTS3 tables. There +** are three possible strategies, in order of preference: +** +** 1. Direct lookup by rowid or docid. +** 2. Full-text search using a MATCH operator on a non-docid column. +** 3. Linear scan of %_content table. +*/ +static int fts3BestIndexMethod(sqlite3_vtab *pVTab, sqlite3_index_info *pInfo){ + Fts3Table *p = (Fts3Table *)pVTab; + int i; /* Iterator variable */ + int iCons = -1; /* Index of constraint to use */ + + /* By default use a full table scan. This is an expensive option, + ** so search through the constraints to see if a more efficient + ** strategy is possible. + */ + pInfo->idxNum = FTS3_FULLSCAN_SEARCH; + pInfo->estimatedCost = 500000; + for(i=0; inConstraint; i++){ + struct sqlite3_index_constraint *pCons = &pInfo->aConstraint[i]; + if( pCons->usable==0 ) continue; + + /* A direct lookup on the rowid or docid column. Assign a cost of 1.0. */ + if( pCons->op==SQLITE_INDEX_CONSTRAINT_EQ + && (pCons->iColumn<0 || pCons->iColumn==p->nColumn+1 ) + ){ + pInfo->idxNum = FTS3_DOCID_SEARCH; + pInfo->estimatedCost = 1.0; + iCons = i; + } + + /* A MATCH constraint. Use a full-text search. + ** + ** If there is more than one MATCH constraint available, use the first + ** one encountered. If there is both a MATCH constraint and a direct + ** rowid/docid lookup, prefer the MATCH strategy. This is done even + ** though the rowid/docid lookup is faster than a MATCH query, selecting + ** it would lead to an "unable to use function MATCH in the requested + ** context" error. + */ + if( pCons->op==SQLITE_INDEX_CONSTRAINT_MATCH + && pCons->iColumn>=0 && pCons->iColumn<=p->nColumn + ){ + pInfo->idxNum = FTS3_FULLTEXT_SEARCH + pCons->iColumn; + pInfo->estimatedCost = 2.0; + iCons = i; + break; + } + } + + if( iCons>=0 ){ + pInfo->aConstraintUsage[iCons].argvIndex = 1; + pInfo->aConstraintUsage[iCons].omit = 1; + } + + /* Regardless of the strategy selected, FTS can deliver rows in rowid (or + ** docid) order. Both ascending and descending are possible. + */ + if( pInfo->nOrderBy==1 ){ + struct sqlite3_index_orderby *pOrder = &pInfo->aOrderBy[0]; + if( pOrder->iColumn<0 || pOrder->iColumn==p->nColumn+1 ){ + if( pOrder->desc ){ + pInfo->idxStr = "DESC"; + }else{ + pInfo->idxStr = "ASC"; + } + pInfo->orderByConsumed = 1; + } + } + + assert( p->pSegments==0 ); + return SQLITE_OK; +} + +/* +** Implementation of xOpen method. +*/ +static int fts3OpenMethod(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCsr){ + sqlite3_vtab_cursor *pCsr; /* Allocated cursor */ + + UNUSED_PARAMETER(pVTab); + + /* Allocate a buffer large enough for an Fts3Cursor structure. If the + ** allocation succeeds, zero it and return SQLITE_OK. Otherwise, + ** if the allocation fails, return SQLITE_NOMEM. + */ + *ppCsr = pCsr = (sqlite3_vtab_cursor *)sqlite3_malloc(sizeof(Fts3Cursor)); + if( !pCsr ){ + return SQLITE_NOMEM; + } + memset(pCsr, 0, sizeof(Fts3Cursor)); + return SQLITE_OK; +} + +/* +** Close the cursor. For additional information see the documentation +** on the xClose method of the virtual table interface. +*/ +static int fts3CloseMethod(sqlite3_vtab_cursor *pCursor){ + Fts3Cursor *pCsr = (Fts3Cursor *)pCursor; + assert( ((Fts3Table *)pCsr->base.pVtab)->pSegments==0 ); + sqlite3_finalize(pCsr->pStmt); + sqlite3Fts3ExprFree(pCsr->pExpr); + sqlite3Fts3FreeDeferredTokens(pCsr); + sqlite3_free(pCsr->aDoclist); + sqlite3_free(pCsr->aMatchinfo); + assert( ((Fts3Table *)pCsr->base.pVtab)->pSegments==0 ); + sqlite3_free(pCsr); + return SQLITE_OK; +} + +/* +** If pCsr->pStmt has not been prepared (i.e. if pCsr->pStmt==0), then +** compose and prepare an SQL statement of the form: +** +** "SELECT FROM %_content WHERE rowid = ?" +** +** (or the equivalent for a content=xxx table) and set pCsr->pStmt to +** it. If an error occurs, return an SQLite error code. +** +** Otherwise, set *ppStmt to point to pCsr->pStmt and return SQLITE_OK. +*/ +static int fts3CursorSeekStmt(Fts3Cursor *pCsr, sqlite3_stmt **ppStmt){ + int rc = SQLITE_OK; + if( pCsr->pStmt==0 ){ + Fts3Table *p = (Fts3Table *)pCsr->base.pVtab; + char *zSql; + zSql = sqlite3_mprintf("SELECT %s WHERE rowid = ?", p->zReadExprlist); + if( !zSql ) return SQLITE_NOMEM; + rc = sqlite3_prepare_v2(p->db, zSql, -1, &pCsr->pStmt, 0); + sqlite3_free(zSql); + } + *ppStmt = pCsr->pStmt; + return rc; +} + +/* +** Position the pCsr->pStmt statement so that it is on the row +** of the %_content table that contains the last match. Return +** SQLITE_OK on success. +*/ +static int fts3CursorSeek(sqlite3_context *pContext, Fts3Cursor *pCsr){ + int rc = SQLITE_OK; + if( pCsr->isRequireSeek ){ + sqlite3_stmt *pStmt = 0; + + rc = fts3CursorSeekStmt(pCsr, &pStmt); + if( rc==SQLITE_OK ){ + sqlite3_bind_int64(pCsr->pStmt, 1, pCsr->iPrevId); + pCsr->isRequireSeek = 0; + if( SQLITE_ROW==sqlite3_step(pCsr->pStmt) ){ + return SQLITE_OK; + }else{ + rc = sqlite3_reset(pCsr->pStmt); + if( rc==SQLITE_OK && ((Fts3Table *)pCsr->base.pVtab)->zContentTbl==0 ){ + /* If no row was found and no error has occured, then the %_content + ** table is missing a row that is present in the full-text index. + ** The data structures are corrupt. */ + rc = FTS_CORRUPT_VTAB; + pCsr->isEof = 1; + } + } + } + } + + if( rc!=SQLITE_OK && pContext ){ + sqlite3_result_error_code(pContext, rc); + } + return rc; +} + +/* +** This function is used to process a single interior node when searching +** a b-tree for a term or term prefix. The node data is passed to this +** function via the zNode/nNode parameters. The term to search for is +** passed in zTerm/nTerm. +** +** If piFirst is not NULL, then this function sets *piFirst to the blockid +** of the child node that heads the sub-tree that may contain the term. +** +** If piLast is not NULL, then *piLast is set to the right-most child node +** that heads a sub-tree that may contain a term for which zTerm/nTerm is +** a prefix. +** +** If an OOM error occurs, SQLITE_NOMEM is returned. Otherwise, SQLITE_OK. +*/ +static int fts3ScanInteriorNode( + const char *zTerm, /* Term to select leaves for */ + int nTerm, /* Size of term zTerm in bytes */ + const char *zNode, /* Buffer containing segment interior node */ + int nNode, /* Size of buffer at zNode */ + sqlite3_int64 *piFirst, /* OUT: Selected child node */ + sqlite3_int64 *piLast /* OUT: Selected child node */ +){ + int rc = SQLITE_OK; /* Return code */ + const char *zCsr = zNode; /* Cursor to iterate through node */ + const char *zEnd = &zCsr[nNode];/* End of interior node buffer */ + char *zBuffer = 0; /* Buffer to load terms into */ + int nAlloc = 0; /* Size of allocated buffer */ + int isFirstTerm = 1; /* True when processing first term on page */ + sqlite3_int64 iChild; /* Block id of child node to descend to */ + + /* Skip over the 'height' varint that occurs at the start of every + ** interior node. Then load the blockid of the left-child of the b-tree + ** node into variable iChild. + ** + ** Even if the data structure on disk is corrupted, this (reading two + ** varints from the buffer) does not risk an overread. If zNode is a + ** root node, then the buffer comes from a SELECT statement. SQLite does + ** not make this guarantee explicitly, but in practice there are always + ** either more than 20 bytes of allocated space following the nNode bytes of + ** contents, or two zero bytes. Or, if the node is read from the %_segments + ** table, then there are always 20 bytes of zeroed padding following the + ** nNode bytes of content (see sqlite3Fts3ReadBlock() for details). + */ + zCsr += sqlite3Fts3GetVarint(zCsr, &iChild); + zCsr += sqlite3Fts3GetVarint(zCsr, &iChild); + if( zCsr>zEnd ){ + return FTS_CORRUPT_VTAB; + } + + while( zCsrzEnd ){ + rc = FTS_CORRUPT_VTAB; + goto finish_scan; + } + if( nPrefix+nSuffix>nAlloc ){ + char *zNew; + nAlloc = (nPrefix+nSuffix) * 2; + zNew = (char *)sqlite3_realloc(zBuffer, nAlloc); + if( !zNew ){ + rc = SQLITE_NOMEM; + goto finish_scan; + } + zBuffer = zNew; + } + assert( zBuffer ); + memcpy(&zBuffer[nPrefix], zCsr, nSuffix); + nBuffer = nPrefix + nSuffix; + zCsr += nSuffix; + + /* Compare the term we are searching for with the term just loaded from + ** the interior node. If the specified term is greater than or equal + ** to the term from the interior node, then all terms on the sub-tree + ** headed by node iChild are smaller than zTerm. No need to search + ** iChild. + ** + ** If the interior node term is larger than the specified term, then + ** the tree headed by iChild may contain the specified term. + */ + cmp = memcmp(zTerm, zBuffer, (nBuffer>nTerm ? nTerm : nBuffer)); + if( piFirst && (cmp<0 || (cmp==0 && nBuffer>nTerm)) ){ + *piFirst = iChild; + piFirst = 0; + } + + if( piLast && cmp<0 ){ + *piLast = iChild; + piLast = 0; + } + + iChild++; + }; + + if( piFirst ) *piFirst = iChild; + if( piLast ) *piLast = iChild; + + finish_scan: + sqlite3_free(zBuffer); + return rc; +} + + +/* +** The buffer pointed to by argument zNode (size nNode bytes) contains an +** interior node of a b-tree segment. The zTerm buffer (size nTerm bytes) +** contains a term. This function searches the sub-tree headed by the zNode +** node for the range of leaf nodes that may contain the specified term +** or terms for which the specified term is a prefix. +** +** If piLeaf is not NULL, then *piLeaf is set to the blockid of the +** left-most leaf node in the tree that may contain the specified term. +** If piLeaf2 is not NULL, then *piLeaf2 is set to the blockid of the +** right-most leaf node that may contain a term for which the specified +** term is a prefix. +** +** It is possible that the range of returned leaf nodes does not contain +** the specified term or any terms for which it is a prefix. However, if the +** segment does contain any such terms, they are stored within the identified +** range. Because this function only inspects interior segment nodes (and +** never loads leaf nodes into memory), it is not possible to be sure. +** +** If an error occurs, an error code other than SQLITE_OK is returned. +*/ +static int fts3SelectLeaf( + Fts3Table *p, /* Virtual table handle */ + const char *zTerm, /* Term to select leaves for */ + int nTerm, /* Size of term zTerm in bytes */ + const char *zNode, /* Buffer containing segment interior node */ + int nNode, /* Size of buffer at zNode */ + sqlite3_int64 *piLeaf, /* Selected leaf node */ + sqlite3_int64 *piLeaf2 /* Selected leaf node */ +){ + int rc; /* Return code */ + int iHeight; /* Height of this node in tree */ + + assert( piLeaf || piLeaf2 ); + + sqlite3Fts3GetVarint32(zNode, &iHeight); + rc = fts3ScanInteriorNode(zTerm, nTerm, zNode, nNode, piLeaf, piLeaf2); + assert( !piLeaf2 || !piLeaf || rc!=SQLITE_OK || (*piLeaf<=*piLeaf2) ); + + if( rc==SQLITE_OK && iHeight>1 ){ + char *zBlob = 0; /* Blob read from %_segments table */ + int nBlob; /* Size of zBlob in bytes */ + + if( piLeaf && piLeaf2 && (*piLeaf!=*piLeaf2) ){ + rc = sqlite3Fts3ReadBlock(p, *piLeaf, &zBlob, &nBlob, 0); + if( rc==SQLITE_OK ){ + rc = fts3SelectLeaf(p, zTerm, nTerm, zBlob, nBlob, piLeaf, 0); + } + sqlite3_free(zBlob); + piLeaf = 0; + zBlob = 0; + } + + if( rc==SQLITE_OK ){ + rc = sqlite3Fts3ReadBlock(p, piLeaf?*piLeaf:*piLeaf2, &zBlob, &nBlob, 0); + } + if( rc==SQLITE_OK ){ + rc = fts3SelectLeaf(p, zTerm, nTerm, zBlob, nBlob, piLeaf, piLeaf2); + } + sqlite3_free(zBlob); + } + + return rc; +} + +/* +** This function is used to create delta-encoded serialized lists of FTS3 +** varints. Each call to this function appends a single varint to a list. +*/ +static void fts3PutDeltaVarint( + char **pp, /* IN/OUT: Output pointer */ + sqlite3_int64 *piPrev, /* IN/OUT: Previous value written to list */ + sqlite3_int64 iVal /* Write this value to the list */ +){ + assert( iVal-*piPrev > 0 || (*piPrev==0 && iVal==0) ); + *pp += sqlite3Fts3PutVarint(*pp, iVal-*piPrev); + *piPrev = iVal; +} + +/* +** When this function is called, *ppPoslist is assumed to point to the +** start of a position-list. After it returns, *ppPoslist points to the +** first byte after the position-list. +** +** A position list is list of positions (delta encoded) and columns for +** a single document record of a doclist. So, in other words, this +** routine advances *ppPoslist so that it points to the next docid in +** the doclist, or to the first byte past the end of the doclist. +** +** If pp is not NULL, then the contents of the position list are copied +** to *pp. *pp is set to point to the first byte past the last byte copied +** before this function returns. +*/ +static void fts3PoslistCopy(char **pp, char **ppPoslist){ + char *pEnd = *ppPoslist; + char c = 0; + + /* The end of a position list is marked by a zero encoded as an FTS3 + ** varint. A single POS_END (0) byte. Except, if the 0 byte is preceded by + ** a byte with the 0x80 bit set, then it is not a varint 0, but the tail + ** of some other, multi-byte, value. + ** + ** The following while-loop moves pEnd to point to the first byte that is not + ** immediately preceded by a byte with the 0x80 bit set. Then increments + ** pEnd once more so that it points to the byte immediately following the + ** last byte in the position-list. + */ + while( *pEnd | c ){ + c = *pEnd++ & 0x80; + testcase( c!=0 && (*pEnd)==0 ); + } + pEnd++; /* Advance past the POS_END terminator byte */ + + if( pp ){ + int n = (int)(pEnd - *ppPoslist); + char *p = *pp; + memcpy(p, *ppPoslist, n); + p += n; + *pp = p; + } + *ppPoslist = pEnd; +} + +/* +** When this function is called, *ppPoslist is assumed to point to the +** start of a column-list. After it returns, *ppPoslist points to the +** to the terminator (POS_COLUMN or POS_END) byte of the column-list. +** +** A column-list is list of delta-encoded positions for a single column +** within a single document within a doclist. +** +** The column-list is terminated either by a POS_COLUMN varint (1) or +** a POS_END varint (0). This routine leaves *ppPoslist pointing to +** the POS_COLUMN or POS_END that terminates the column-list. +** +** If pp is not NULL, then the contents of the column-list are copied +** to *pp. *pp is set to point to the first byte past the last byte copied +** before this function returns. The POS_COLUMN or POS_END terminator +** is not copied into *pp. +*/ +static void fts3ColumnlistCopy(char **pp, char **ppPoslist){ + char *pEnd = *ppPoslist; + char c = 0; + + /* A column-list is terminated by either a 0x01 or 0x00 byte that is + ** not part of a multi-byte varint. + */ + while( 0xFE & (*pEnd | c) ){ + c = *pEnd++ & 0x80; + testcase( c!=0 && ((*pEnd)&0xfe)==0 ); + } + if( pp ){ + int n = (int)(pEnd - *ppPoslist); + char *p = *pp; + memcpy(p, *ppPoslist, n); + p += n; + *pp = p; + } + *ppPoslist = pEnd; +} + +/* +** Value used to signify the end of an position-list. This is safe because +** it is not possible to have a document with 2^31 terms. +*/ +#define POSITION_LIST_END 0x7fffffff + +/* +** This function is used to help parse position-lists. When this function is +** called, *pp may point to the start of the next varint in the position-list +** being parsed, or it may point to 1 byte past the end of the position-list +** (in which case **pp will be a terminator bytes POS_END (0) or +** (1)). +** +** If *pp points past the end of the current position-list, set *pi to +** POSITION_LIST_END and return. Otherwise, read the next varint from *pp, +** increment the current value of *pi by the value read, and set *pp to +** point to the next value before returning. +** +** Before calling this routine *pi must be initialized to the value of +** the previous position, or zero if we are reading the first position +** in the position-list. Because positions are delta-encoded, the value +** of the previous position is needed in order to compute the value of +** the next position. +*/ +static void fts3ReadNextPos( + char **pp, /* IN/OUT: Pointer into position-list buffer */ + sqlite3_int64 *pi /* IN/OUT: Value read from position-list */ +){ + if( (**pp)&0xFE ){ + fts3GetDeltaVarint(pp, pi); + *pi -= 2; + }else{ + *pi = POSITION_LIST_END; + } +} + +/* +** If parameter iCol is not 0, write an POS_COLUMN (1) byte followed by +** the value of iCol encoded as a varint to *pp. This will start a new +** column list. +** +** Set *pp to point to the byte just after the last byte written before +** returning (do not modify it if iCol==0). Return the total number of bytes +** written (0 if iCol==0). +*/ +static int fts3PutColNumber(char **pp, int iCol){ + int n = 0; /* Number of bytes written */ + if( iCol ){ + char *p = *pp; /* Output pointer */ + n = 1 + sqlite3Fts3PutVarint(&p[1], iCol); + *p = 0x01; + *pp = &p[n]; + } + return n; +} + +/* +** Compute the union of two position lists. The output written +** into *pp contains all positions of both *pp1 and *pp2 in sorted +** order and with any duplicates removed. All pointers are +** updated appropriately. The caller is responsible for insuring +** that there is enough space in *pp to hold the complete output. +*/ +static void fts3PoslistMerge( + char **pp, /* Output buffer */ + char **pp1, /* Left input list */ + char **pp2 /* Right input list */ +){ + char *p = *pp; + char *p1 = *pp1; + char *p2 = *pp2; + + while( *p1 || *p2 ){ + int iCol1; /* The current column index in pp1 */ + int iCol2; /* The current column index in pp2 */ + + if( *p1==POS_COLUMN ) sqlite3Fts3GetVarint32(&p1[1], &iCol1); + else if( *p1==POS_END ) iCol1 = POSITION_LIST_END; + else iCol1 = 0; + + if( *p2==POS_COLUMN ) sqlite3Fts3GetVarint32(&p2[1], &iCol2); + else if( *p2==POS_END ) iCol2 = POSITION_LIST_END; + else iCol2 = 0; + + if( iCol1==iCol2 ){ + sqlite3_int64 i1 = 0; /* Last position from pp1 */ + sqlite3_int64 i2 = 0; /* Last position from pp2 */ + sqlite3_int64 iPrev = 0; + int n = fts3PutColNumber(&p, iCol1); + p1 += n; + p2 += n; + + /* At this point, both p1 and p2 point to the start of column-lists + ** for the same column (the column with index iCol1 and iCol2). + ** A column-list is a list of non-negative delta-encoded varints, each + ** incremented by 2 before being stored. Each list is terminated by a + ** POS_END (0) or POS_COLUMN (1). The following block merges the two lists + ** and writes the results to buffer p. p is left pointing to the byte + ** after the list written. No terminator (POS_END or POS_COLUMN) is + ** written to the output. + */ + fts3GetDeltaVarint(&p1, &i1); + fts3GetDeltaVarint(&p2, &i2); + do { + fts3PutDeltaVarint(&p, &iPrev, (i1pos(*pp1) && pos(*pp2)-pos(*pp1)<=nToken). i.e. +** when the *pp1 token appears before the *pp2 token, but not more than nToken +** slots before it. +** +** e.g. nToken==1 searches for adjacent positions. +*/ +static int fts3PoslistPhraseMerge( + char **pp, /* IN/OUT: Preallocated output buffer */ + int nToken, /* Maximum difference in token positions */ + int isSaveLeft, /* Save the left position */ + int isExact, /* If *pp1 is exactly nTokens before *pp2 */ + char **pp1, /* IN/OUT: Left input list */ + char **pp2 /* IN/OUT: Right input list */ +){ + char *p = *pp; + char *p1 = *pp1; + char *p2 = *pp2; + int iCol1 = 0; + int iCol2 = 0; + + /* Never set both isSaveLeft and isExact for the same invocation. */ + assert( isSaveLeft==0 || isExact==0 ); + + assert( p!=0 && *p1!=0 && *p2!=0 ); + if( *p1==POS_COLUMN ){ + p1++; + p1 += sqlite3Fts3GetVarint32(p1, &iCol1); + } + if( *p2==POS_COLUMN ){ + p2++; + p2 += sqlite3Fts3GetVarint32(p2, &iCol2); + } + + while( 1 ){ + if( iCol1==iCol2 ){ + char *pSave = p; + sqlite3_int64 iPrev = 0; + sqlite3_int64 iPos1 = 0; + sqlite3_int64 iPos2 = 0; + + if( iCol1 ){ + *p++ = POS_COLUMN; + p += sqlite3Fts3PutVarint(p, iCol1); + } + + assert( *p1!=POS_END && *p1!=POS_COLUMN ); + assert( *p2!=POS_END && *p2!=POS_COLUMN ); + fts3GetDeltaVarint(&p1, &iPos1); iPos1 -= 2; + fts3GetDeltaVarint(&p2, &iPos2); iPos2 -= 2; + + while( 1 ){ + if( iPos2==iPos1+nToken + || (isExact==0 && iPos2>iPos1 && iPos2<=iPos1+nToken) + ){ + sqlite3_int64 iSave; + iSave = isSaveLeft ? iPos1 : iPos2; + fts3PutDeltaVarint(&p, &iPrev, iSave+2); iPrev -= 2; + pSave = 0; + assert( p ); + } + if( (!isSaveLeft && iPos2<=(iPos1+nToken)) || iPos2<=iPos1 ){ + if( (*p2&0xFE)==0 ) break; + fts3GetDeltaVarint(&p2, &iPos2); iPos2 -= 2; + }else{ + if( (*p1&0xFE)==0 ) break; + fts3GetDeltaVarint(&p1, &iPos1); iPos1 -= 2; + } + } + + if( pSave ){ + assert( pp && p ); + p = pSave; + } + + fts3ColumnlistCopy(0, &p1); + fts3ColumnlistCopy(0, &p2); + assert( (*p1&0xFE)==0 && (*p2&0xFE)==0 ); + if( 0==*p1 || 0==*p2 ) break; + + p1++; + p1 += sqlite3Fts3GetVarint32(p1, &iCol1); + p2++; + p2 += sqlite3Fts3GetVarint32(p2, &iCol2); + } + + /* Advance pointer p1 or p2 (whichever corresponds to the smaller of + ** iCol1 and iCol2) so that it points to either the 0x00 that marks the + ** end of the position list, or the 0x01 that precedes the next + ** column-number in the position list. + */ + else if( iCol1=pEnd ){ + *pp = 0; + }else{ + sqlite3_int64 iVal; + *pp += sqlite3Fts3GetVarint(*pp, &iVal); + if( bDescIdx ){ + *pVal -= iVal; + }else{ + *pVal += iVal; + } + } +} + +/* +** This function is used to write a single varint to a buffer. The varint +** is written to *pp. Before returning, *pp is set to point 1 byte past the +** end of the value written. +** +** If *pbFirst is zero when this function is called, the value written to +** the buffer is that of parameter iVal. +** +** If *pbFirst is non-zero when this function is called, then the value +** written is either (iVal-*piPrev) (if bDescIdx is zero) or (*piPrev-iVal) +** (if bDescIdx is non-zero). +** +** Before returning, this function always sets *pbFirst to 1 and *piPrev +** to the value of parameter iVal. +*/ +static void fts3PutDeltaVarint3( + char **pp, /* IN/OUT: Output pointer */ + int bDescIdx, /* True for descending docids */ + sqlite3_int64 *piPrev, /* IN/OUT: Previous value written to list */ + int *pbFirst, /* IN/OUT: True after first int written */ + sqlite3_int64 iVal /* Write this value to the list */ +){ + sqlite3_int64 iWrite; + if( bDescIdx==0 || *pbFirst==0 ){ + iWrite = iVal - *piPrev; + }else{ + iWrite = *piPrev - iVal; + } + assert( *pbFirst || *piPrev==0 ); + assert( *pbFirst==0 || iWrite>0 ); + *pp += sqlite3Fts3PutVarint(*pp, iWrite); + *piPrev = iVal; + *pbFirst = 1; +} + + +/* +** This macro is used by various functions that merge doclists. The two +** arguments are 64-bit docid values. If the value of the stack variable +** bDescDoclist is 0 when this macro is invoked, then it returns (i1-i2). +** Otherwise, (i2-i1). +** +** Using this makes it easier to write code that can merge doclists that are +** sorted in either ascending or descending order. +*/ +#define DOCID_CMP(i1, i2) ((bDescDoclist?-1:1) * (i1-i2)) + +/* +** This function does an "OR" merge of two doclists (output contains all +** positions contained in either argument doclist). If the docids in the +** input doclists are sorted in ascending order, parameter bDescDoclist +** should be false. If they are sorted in ascending order, it should be +** passed a non-zero value. +** +** If no error occurs, *paOut is set to point at an sqlite3_malloc'd buffer +** containing the output doclist and SQLITE_OK is returned. In this case +** *pnOut is set to the number of bytes in the output doclist. +** +** If an error occurs, an SQLite error code is returned. The output values +** are undefined in this case. +*/ +static int fts3DoclistOrMerge( + int bDescDoclist, /* True if arguments are desc */ + char *a1, int n1, /* First doclist */ + char *a2, int n2, /* Second doclist */ + char **paOut, int *pnOut /* OUT: Malloc'd doclist */ +){ + sqlite3_int64 i1 = 0; + sqlite3_int64 i2 = 0; + sqlite3_int64 iPrev = 0; + char *pEnd1 = &a1[n1]; + char *pEnd2 = &a2[n2]; + char *p1 = a1; + char *p2 = a2; + char *p; + char *aOut; + int bFirstOut = 0; + + *paOut = 0; + *pnOut = 0; + + /* Allocate space for the output. Both the input and output doclists + ** are delta encoded. If they are in ascending order (bDescDoclist==0), + ** then the first docid in each list is simply encoded as a varint. For + ** each subsequent docid, the varint stored is the difference between the + ** current and previous docid (a positive number - since the list is in + ** ascending order). + ** + ** The first docid written to the output is therefore encoded using the + ** same number of bytes as it is in whichever of the input lists it is + ** read from. And each subsequent docid read from the same input list + ** consumes either the same or less bytes as it did in the input (since + ** the difference between it and the previous value in the output must + ** be a positive value less than or equal to the delta value read from + ** the input list). The same argument applies to all but the first docid + ** read from the 'other' list. And to the contents of all position lists + ** that will be copied and merged from the input to the output. + ** + ** However, if the first docid copied to the output is a negative number, + ** then the encoding of the first docid from the 'other' input list may + ** be larger in the output than it was in the input (since the delta value + ** may be a larger positive integer than the actual docid). + ** + ** The space required to store the output is therefore the sum of the + ** sizes of the two inputs, plus enough space for exactly one of the input + ** docids to grow. + ** + ** A symetric argument may be made if the doclists are in descending + ** order. + */ + aOut = sqlite3_malloc(n1+n2+FTS3_VARINT_MAX-1); + if( !aOut ) return SQLITE_NOMEM; + + p = aOut; + fts3GetDeltaVarint3(&p1, pEnd1, 0, &i1); + fts3GetDeltaVarint3(&p2, pEnd2, 0, &i2); + while( p1 || p2 ){ + sqlite3_int64 iDiff = DOCID_CMP(i1, i2); + + if( p2 && p1 && iDiff==0 ){ + fts3PutDeltaVarint3(&p, bDescDoclist, &iPrev, &bFirstOut, i1); + fts3PoslistMerge(&p, &p1, &p2); + fts3GetDeltaVarint3(&p1, pEnd1, bDescDoclist, &i1); + fts3GetDeltaVarint3(&p2, pEnd2, bDescDoclist, &i2); + }else if( !p2 || (p1 && iDiff<0) ){ + fts3PutDeltaVarint3(&p, bDescDoclist, &iPrev, &bFirstOut, i1); + fts3PoslistCopy(&p, &p1); + fts3GetDeltaVarint3(&p1, pEnd1, bDescDoclist, &i1); + }else{ + fts3PutDeltaVarint3(&p, bDescDoclist, &iPrev, &bFirstOut, i2); + fts3PoslistCopy(&p, &p2); + fts3GetDeltaVarint3(&p2, pEnd2, bDescDoclist, &i2); + } + } + + *paOut = aOut; + *pnOut = (p-aOut); + assert( *pnOut<=n1+n2+FTS3_VARINT_MAX-1 ); + return SQLITE_OK; +} + +/* +** This function does a "phrase" merge of two doclists. In a phrase merge, +** the output contains a copy of each position from the right-hand input +** doclist for which there is a position in the left-hand input doclist +** exactly nDist tokens before it. +** +** If the docids in the input doclists are sorted in ascending order, +** parameter bDescDoclist should be false. If they are sorted in ascending +** order, it should be passed a non-zero value. +** +** The right-hand input doclist is overwritten by this function. +*/ +static void fts3DoclistPhraseMerge( + int bDescDoclist, /* True if arguments are desc */ + int nDist, /* Distance from left to right (1=adjacent) */ + char *aLeft, int nLeft, /* Left doclist */ + char *aRight, int *pnRight /* IN/OUT: Right/output doclist */ +){ + sqlite3_int64 i1 = 0; + sqlite3_int64 i2 = 0; + sqlite3_int64 iPrev = 0; + char *pEnd1 = &aLeft[nLeft]; + char *pEnd2 = &aRight[*pnRight]; + char *p1 = aLeft; + char *p2 = aRight; + char *p; + int bFirstOut = 0; + char *aOut = aRight; + + assert( nDist>0 ); + + p = aOut; + fts3GetDeltaVarint3(&p1, pEnd1, 0, &i1); + fts3GetDeltaVarint3(&p2, pEnd2, 0, &i2); + + while( p1 && p2 ){ + sqlite3_int64 iDiff = DOCID_CMP(i1, i2); + if( iDiff==0 ){ + char *pSave = p; + sqlite3_int64 iPrevSave = iPrev; + int bFirstOutSave = bFirstOut; + + fts3PutDeltaVarint3(&p, bDescDoclist, &iPrev, &bFirstOut, i1); + if( 0==fts3PoslistPhraseMerge(&p, nDist, 0, 1, &p1, &p2) ){ + p = pSave; + iPrev = iPrevSave; + bFirstOut = bFirstOutSave; + } + fts3GetDeltaVarint3(&p1, pEnd1, bDescDoclist, &i1); + fts3GetDeltaVarint3(&p2, pEnd2, bDescDoclist, &i2); + }else if( iDiff<0 ){ + fts3PoslistCopy(0, &p1); + fts3GetDeltaVarint3(&p1, pEnd1, bDescDoclist, &i1); + }else{ + fts3PoslistCopy(0, &p2); + fts3GetDeltaVarint3(&p2, pEnd2, bDescDoclist, &i2); + } + } + + *pnRight = p - aOut; +} + +/* +** Argument pList points to a position list nList bytes in size. This +** function checks to see if the position list contains any entries for +** a token in position 0 (of any column). If so, it writes argument iDelta +** to the output buffer pOut, followed by a position list consisting only +** of the entries from pList at position 0, and terminated by an 0x00 byte. +** The value returned is the number of bytes written to pOut (if any). +*/ +int sqlite3Fts3FirstFilter( + sqlite3_int64 iDelta, /* Varint that may be written to pOut */ + char *pList, /* Position list (no 0x00 term) */ + int nList, /* Size of pList in bytes */ + char *pOut /* Write output here */ +){ + int nOut = 0; + int bWritten = 0; /* True once iDelta has been written */ + char *p = pList; + char *pEnd = &pList[nList]; + + if( *p!=0x01 ){ + if( *p==0x02 ){ + nOut += sqlite3Fts3PutVarint(&pOut[nOut], iDelta); + pOut[nOut++] = 0x02; + bWritten = 1; + } + fts3ColumnlistCopy(0, &p); + } + + while( paaOutput); i++){ + if( pTS->aaOutput[i] ){ + if( !aOut ){ + aOut = pTS->aaOutput[i]; + nOut = pTS->anOutput[i]; + pTS->aaOutput[i] = 0; + }else{ + int nNew; + char *aNew; + + int rc = fts3DoclistOrMerge(p->bDescIdx, + pTS->aaOutput[i], pTS->anOutput[i], aOut, nOut, &aNew, &nNew + ); + if( rc!=SQLITE_OK ){ + sqlite3_free(aOut); + return rc; + } + + sqlite3_free(pTS->aaOutput[i]); + sqlite3_free(aOut); + pTS->aaOutput[i] = 0; + aOut = aNew; + nOut = nNew; + } + } + } + + pTS->aaOutput[0] = aOut; + pTS->anOutput[0] = nOut; + return SQLITE_OK; +} + +/* +** Merge the doclist aDoclist/nDoclist into the TermSelect object passed +** as the first argument. The merge is an "OR" merge (see function +** fts3DoclistOrMerge() for details). +** +** This function is called with the doclist for each term that matches +** a queried prefix. It merges all these doclists into one, the doclist +** for the specified prefix. Since there can be a very large number of +** doclists to merge, the merging is done pair-wise using the TermSelect +** object. +** +** This function returns SQLITE_OK if the merge is successful, or an +** SQLite error code (SQLITE_NOMEM) if an error occurs. +*/ +static int fts3TermSelectMerge( + Fts3Table *p, /* FTS table handle */ + TermSelect *pTS, /* TermSelect object to merge into */ + char *aDoclist, /* Pointer to doclist */ + int nDoclist /* Size of aDoclist in bytes */ +){ + if( pTS->aaOutput[0]==0 ){ + /* If this is the first term selected, copy the doclist to the output + ** buffer using memcpy(). */ + pTS->aaOutput[0] = sqlite3_malloc(nDoclist); + pTS->anOutput[0] = nDoclist; + if( pTS->aaOutput[0] ){ + memcpy(pTS->aaOutput[0], aDoclist, nDoclist); + }else{ + return SQLITE_NOMEM; + } + }else{ + char *aMerge = aDoclist; + int nMerge = nDoclist; + int iOut; + + for(iOut=0; iOutaaOutput); iOut++){ + if( pTS->aaOutput[iOut]==0 ){ + assert( iOut>0 ); + pTS->aaOutput[iOut] = aMerge; + pTS->anOutput[iOut] = nMerge; + break; + }else{ + char *aNew; + int nNew; + + int rc = fts3DoclistOrMerge(p->bDescIdx, aMerge, nMerge, + pTS->aaOutput[iOut], pTS->anOutput[iOut], &aNew, &nNew + ); + if( rc!=SQLITE_OK ){ + if( aMerge!=aDoclist ) sqlite3_free(aMerge); + return rc; + } + + if( aMerge!=aDoclist ) sqlite3_free(aMerge); + sqlite3_free(pTS->aaOutput[iOut]); + pTS->aaOutput[iOut] = 0; + + aMerge = aNew; + nMerge = nNew; + if( (iOut+1)==SizeofArray(pTS->aaOutput) ){ + pTS->aaOutput[iOut] = aMerge; + pTS->anOutput[iOut] = nMerge; + } + } + } + } + return SQLITE_OK; +} + +/* +** Append SegReader object pNew to the end of the pCsr->apSegment[] array. +*/ +static int fts3SegReaderCursorAppend( + Fts3MultiSegReader *pCsr, + Fts3SegReader *pNew +){ + if( (pCsr->nSegment%16)==0 ){ + Fts3SegReader **apNew; + int nByte = (pCsr->nSegment + 16)*sizeof(Fts3SegReader*); + apNew = (Fts3SegReader **)sqlite3_realloc(pCsr->apSegment, nByte); + if( !apNew ){ + sqlite3Fts3SegReaderFree(pNew); + return SQLITE_NOMEM; + } + pCsr->apSegment = apNew; + } + pCsr->apSegment[pCsr->nSegment++] = pNew; + return SQLITE_OK; +} + +/* +** Add seg-reader objects to the Fts3MultiSegReader object passed as the +** 8th argument. +** +** This function returns SQLITE_OK if successful, or an SQLite error code +** otherwise. +*/ +static int fts3SegReaderCursor( + Fts3Table *p, /* FTS3 table handle */ + int iIndex, /* Index to search (from 0 to p->nIndex-1) */ + int iLevel, /* Level of segments to scan */ + const char *zTerm, /* Term to query for */ + int nTerm, /* Size of zTerm in bytes */ + int isPrefix, /* True for a prefix search */ + int isScan, /* True to scan from zTerm to EOF */ + Fts3MultiSegReader *pCsr /* Cursor object to populate */ +){ + int rc = SQLITE_OK; /* Error code */ + sqlite3_stmt *pStmt = 0; /* Statement to iterate through segments */ + int rc2; /* Result of sqlite3_reset() */ + + /* If iLevel is less than 0 and this is not a scan, include a seg-reader + ** for the pending-terms. If this is a scan, then this call must be being + ** made by an fts4aux module, not an FTS table. In this case calling + ** Fts3SegReaderPending might segfault, as the data structures used by + ** fts4aux are not completely populated. So it's easiest to filter these + ** calls out here. */ + if( iLevel<0 && p->aIndex ){ + Fts3SegReader *pSeg = 0; + rc = sqlite3Fts3SegReaderPending(p, iIndex, zTerm, nTerm, isPrefix, &pSeg); + if( rc==SQLITE_OK && pSeg ){ + rc = fts3SegReaderCursorAppend(pCsr, pSeg); + } + } + + if( iLevel!=FTS3_SEGCURSOR_PENDING ){ + if( rc==SQLITE_OK ){ + rc = sqlite3Fts3AllSegdirs(p, iIndex, iLevel, &pStmt); + } + + while( rc==SQLITE_OK && SQLITE_ROW==(rc = sqlite3_step(pStmt)) ){ + Fts3SegReader *pSeg = 0; + + /* Read the values returned by the SELECT into local variables. */ + sqlite3_int64 iStartBlock = sqlite3_column_int64(pStmt, 1); + sqlite3_int64 iLeavesEndBlock = sqlite3_column_int64(pStmt, 2); + sqlite3_int64 iEndBlock = sqlite3_column_int64(pStmt, 3); + int nRoot = sqlite3_column_bytes(pStmt, 4); + char const *zRoot = sqlite3_column_blob(pStmt, 4); + + /* If zTerm is not NULL, and this segment is not stored entirely on its + ** root node, the range of leaves scanned can be reduced. Do this. */ + if( iStartBlock && zTerm ){ + sqlite3_int64 *pi = (isPrefix ? &iLeavesEndBlock : 0); + rc = fts3SelectLeaf(p, zTerm, nTerm, zRoot, nRoot, &iStartBlock, pi); + if( rc!=SQLITE_OK ) goto finished; + if( isPrefix==0 && isScan==0 ) iLeavesEndBlock = iStartBlock; + } + + rc = sqlite3Fts3SegReaderNew(pCsr->nSegment+1, + iStartBlock, iLeavesEndBlock, iEndBlock, zRoot, nRoot, &pSeg + ); + if( rc!=SQLITE_OK ) goto finished; + rc = fts3SegReaderCursorAppend(pCsr, pSeg); + } + } + + finished: + rc2 = sqlite3_reset(pStmt); + if( rc==SQLITE_DONE ) rc = rc2; + + return rc; +} + +/* +** Set up a cursor object for iterating through a full-text index or a +** single level therein. +*/ +int sqlite3Fts3SegReaderCursor( + Fts3Table *p, /* FTS3 table handle */ + int iIndex, /* Index to search (from 0 to p->nIndex-1) */ + int iLevel, /* Level of segments to scan */ + const char *zTerm, /* Term to query for */ + int nTerm, /* Size of zTerm in bytes */ + int isPrefix, /* True for a prefix search */ + int isScan, /* True to scan from zTerm to EOF */ + Fts3MultiSegReader *pCsr /* Cursor object to populate */ +){ + assert( iIndex>=0 && iIndexnIndex ); + assert( iLevel==FTS3_SEGCURSOR_ALL + || iLevel==FTS3_SEGCURSOR_PENDING + || iLevel>=0 + ); + assert( iLevelaIndex==0 ); + + memset(pCsr, 0, sizeof(Fts3MultiSegReader)); + + return fts3SegReaderCursor( + p, iIndex, iLevel, zTerm, nTerm, isPrefix, isScan, pCsr + ); +} + +/* +** In addition to its current configuration, have the Fts3MultiSegReader +** passed as the 4th argument also scan the doclist for term zTerm/nTerm. +** +** SQLITE_OK is returned if no error occurs, otherwise an SQLite error code. +*/ +static int fts3SegReaderCursorAddZero( + Fts3Table *p, /* FTS virtual table handle */ + const char *zTerm, /* Term to scan doclist of */ + int nTerm, /* Number of bytes in zTerm */ + Fts3MultiSegReader *pCsr /* Fts3MultiSegReader to modify */ +){ + return fts3SegReaderCursor(p, 0, FTS3_SEGCURSOR_ALL, zTerm, nTerm, 0, 0,pCsr); +} + +/* +** Open an Fts3MultiSegReader to scan the doclist for term zTerm/nTerm. Or, +** if isPrefix is true, to scan the doclist for all terms for which +** zTerm/nTerm is a prefix. If successful, return SQLITE_OK and write +** a pointer to the new Fts3MultiSegReader to *ppSegcsr. Otherwise, return +** an SQLite error code. +** +** It is the responsibility of the caller to free this object by eventually +** passing it to fts3SegReaderCursorFree() +** +** SQLITE_OK is returned if no error occurs, otherwise an SQLite error code. +** Output parameter *ppSegcsr is set to 0 if an error occurs. +*/ +static int fts3TermSegReaderCursor( + Fts3Cursor *pCsr, /* Virtual table cursor handle */ + const char *zTerm, /* Term to query for */ + int nTerm, /* Size of zTerm in bytes */ + int isPrefix, /* True for a prefix search */ + Fts3MultiSegReader **ppSegcsr /* OUT: Allocated seg-reader cursor */ +){ + Fts3MultiSegReader *pSegcsr; /* Object to allocate and return */ + int rc = SQLITE_NOMEM; /* Return code */ + + pSegcsr = sqlite3_malloc(sizeof(Fts3MultiSegReader)); + if( pSegcsr ){ + int i; + int bFound = 0; /* True once an index has been found */ + Fts3Table *p = (Fts3Table *)pCsr->base.pVtab; + + if( isPrefix ){ + for(i=1; bFound==0 && inIndex; i++){ + if( p->aIndex[i].nPrefix==nTerm ){ + bFound = 1; + rc = sqlite3Fts3SegReaderCursor( + p, i, FTS3_SEGCURSOR_ALL, zTerm, nTerm, 0, 0, pSegcsr); + pSegcsr->bLookup = 1; + } + } + + for(i=1; bFound==0 && inIndex; i++){ + if( p->aIndex[i].nPrefix==nTerm+1 ){ + bFound = 1; + rc = sqlite3Fts3SegReaderCursor( + p, i, FTS3_SEGCURSOR_ALL, zTerm, nTerm, 1, 0, pSegcsr + ); + if( rc==SQLITE_OK ){ + rc = fts3SegReaderCursorAddZero(p, zTerm, nTerm, pSegcsr); + } + } + } + } + + if( bFound==0 ){ + rc = sqlite3Fts3SegReaderCursor( + p, 0, FTS3_SEGCURSOR_ALL, zTerm, nTerm, isPrefix, 0, pSegcsr + ); + pSegcsr->bLookup = !isPrefix; + } + } + + *ppSegcsr = pSegcsr; + return rc; +} + +/* +** Free an Fts3MultiSegReader allocated by fts3TermSegReaderCursor(). +*/ +static void fts3SegReaderCursorFree(Fts3MultiSegReader *pSegcsr){ + sqlite3Fts3SegReaderFinish(pSegcsr); + sqlite3_free(pSegcsr); +} + +/* +** This function retreives the doclist for the specified term (or term +** prefix) from the database. +*/ +static int fts3TermSelect( + Fts3Table *p, /* Virtual table handle */ + Fts3PhraseToken *pTok, /* Token to query for */ + int iColumn, /* Column to query (or -ve for all columns) */ + int *pnOut, /* OUT: Size of buffer at *ppOut */ + char **ppOut /* OUT: Malloced result buffer */ +){ + int rc; /* Return code */ + Fts3MultiSegReader *pSegcsr; /* Seg-reader cursor for this term */ + TermSelect tsc; /* Object for pair-wise doclist merging */ + Fts3SegFilter filter; /* Segment term filter configuration */ + + pSegcsr = pTok->pSegcsr; + memset(&tsc, 0, sizeof(TermSelect)); + + filter.flags = FTS3_SEGMENT_IGNORE_EMPTY | FTS3_SEGMENT_REQUIRE_POS + | (pTok->isPrefix ? FTS3_SEGMENT_PREFIX : 0) + | (pTok->bFirst ? FTS3_SEGMENT_FIRST : 0) + | (iColumnnColumn ? FTS3_SEGMENT_COLUMN_FILTER : 0); + filter.iCol = iColumn; + filter.zTerm = pTok->z; + filter.nTerm = pTok->n; + + rc = sqlite3Fts3SegReaderStart(p, pSegcsr, &filter); + while( SQLITE_OK==rc + && SQLITE_ROW==(rc = sqlite3Fts3SegReaderStep(p, pSegcsr)) + ){ + rc = fts3TermSelectMerge(p, &tsc, pSegcsr->aDoclist, pSegcsr->nDoclist); + } + + if( rc==SQLITE_OK ){ + rc = fts3TermSelectFinishMerge(p, &tsc); + } + if( rc==SQLITE_OK ){ + *ppOut = tsc.aaOutput[0]; + *pnOut = tsc.anOutput[0]; + }else{ + int i; + for(i=0; ipSegcsr = 0; + return rc; +} + +/* +** This function counts the total number of docids in the doclist stored +** in buffer aList[], size nList bytes. +** +** If the isPoslist argument is true, then it is assumed that the doclist +** contains a position-list following each docid. Otherwise, it is assumed +** that the doclist is simply a list of docids stored as delta encoded +** varints. +*/ +static int fts3DoclistCountDocids(char *aList, int nList){ + int nDoc = 0; /* Return value */ + if( aList ){ + char *aEnd = &aList[nList]; /* Pointer to one byte after EOF */ + char *p = aList; /* Cursor */ + while( peSearch==FTS3_DOCID_SEARCH || pCsr->eSearch==FTS3_FULLSCAN_SEARCH ){ + if( SQLITE_ROW!=sqlite3_step(pCsr->pStmt) ){ + pCsr->isEof = 1; + rc = sqlite3_reset(pCsr->pStmt); + }else{ + pCsr->iPrevId = sqlite3_column_int64(pCsr->pStmt, 0); + rc = SQLITE_OK; + } + }else{ + rc = fts3EvalNext((Fts3Cursor *)pCursor); + } + assert( ((Fts3Table *)pCsr->base.pVtab)->pSegments==0 ); + return rc; +} + +/* +** This is the xFilter interface for the virtual table. See +** the virtual table xFilter method documentation for additional +** information. +** +** If idxNum==FTS3_FULLSCAN_SEARCH then do a full table scan against +** the %_content table. +** +** If idxNum==FTS3_DOCID_SEARCH then do a docid lookup for a single entry +** in the %_content table. +** +** If idxNum>=FTS3_FULLTEXT_SEARCH then use the full text index. The +** column on the left-hand side of the MATCH operator is column +** number idxNum-FTS3_FULLTEXT_SEARCH, 0 indexed. argv[0] is the right-hand +** side of the MATCH operator. +*/ +static int fts3FilterMethod( + sqlite3_vtab_cursor *pCursor, /* The cursor used for this query */ + int idxNum, /* Strategy index */ + const char *idxStr, /* Unused */ + int nVal, /* Number of elements in apVal */ + sqlite3_value **apVal /* Arguments for the indexing scheme */ +){ + int rc; + char *zSql; /* SQL statement used to access %_content */ + Fts3Table *p = (Fts3Table *)pCursor->pVtab; + Fts3Cursor *pCsr = (Fts3Cursor *)pCursor; + + UNUSED_PARAMETER(idxStr); + UNUSED_PARAMETER(nVal); + + assert( idxNum>=0 && idxNum<=(FTS3_FULLTEXT_SEARCH+p->nColumn) ); + assert( nVal==0 || nVal==1 ); + assert( (nVal==0)==(idxNum==FTS3_FULLSCAN_SEARCH) ); + assert( p->pSegments==0 ); + + /* In case the cursor has been used before, clear it now. */ + sqlite3_finalize(pCsr->pStmt); + sqlite3_free(pCsr->aDoclist); + sqlite3Fts3ExprFree(pCsr->pExpr); + memset(&pCursor[1], 0, sizeof(Fts3Cursor)-sizeof(sqlite3_vtab_cursor)); + + if( idxStr ){ + pCsr->bDesc = (idxStr[0]=='D'); + }else{ + pCsr->bDesc = p->bDescIdx; + } + pCsr->eSearch = (i16)idxNum; + + if( idxNum!=FTS3_DOCID_SEARCH && idxNum!=FTS3_FULLSCAN_SEARCH ){ + int iCol = idxNum-FTS3_FULLTEXT_SEARCH; + const char *zQuery = (const char *)sqlite3_value_text(apVal[0]); + + if( zQuery==0 && sqlite3_value_type(apVal[0])!=SQLITE_NULL ){ + return SQLITE_NOMEM; + } + + rc = sqlite3Fts3ExprParse(p->pTokenizer, p->azColumn, p->bHasStat, + p->nColumn, iCol, zQuery, -1, &pCsr->pExpr + ); + if( rc!=SQLITE_OK ){ + if( rc==SQLITE_ERROR ){ + static const char *zErr = "malformed MATCH expression: [%s]"; + p->base.zErrMsg = sqlite3_mprintf(zErr, zQuery); + } + return rc; + } + + rc = sqlite3Fts3ReadLock(p); + if( rc!=SQLITE_OK ) return rc; + + rc = fts3EvalStart(pCsr); + + sqlite3Fts3SegmentsClose(p); + if( rc!=SQLITE_OK ) return rc; + pCsr->pNextId = pCsr->aDoclist; + pCsr->iPrevId = 0; + } + + /* Compile a SELECT statement for this cursor. For a full-table-scan, the + ** statement loops through all rows of the %_content table. For a + ** full-text query or docid lookup, the statement retrieves a single + ** row by docid. + */ + if( idxNum==FTS3_FULLSCAN_SEARCH ){ + zSql = sqlite3_mprintf( + "SELECT %s ORDER BY rowid %s", + p->zReadExprlist, (pCsr->bDesc ? "DESC" : "ASC") + ); + if( zSql ){ + rc = sqlite3_prepare_v2(p->db, zSql, -1, &pCsr->pStmt, 0); + sqlite3_free(zSql); + }else{ + rc = SQLITE_NOMEM; + } + }else if( idxNum==FTS3_DOCID_SEARCH ){ + rc = fts3CursorSeekStmt(pCsr, &pCsr->pStmt); + if( rc==SQLITE_OK ){ + rc = sqlite3_bind_value(pCsr->pStmt, 1, apVal[0]); + } + } + if( rc!=SQLITE_OK ) return rc; + + return fts3NextMethod(pCursor); +} + +/* +** This is the xEof method of the virtual table. SQLite calls this +** routine to find out if it has reached the end of a result set. +*/ +static int fts3EofMethod(sqlite3_vtab_cursor *pCursor){ + return ((Fts3Cursor *)pCursor)->isEof; +} + +/* +** This is the xRowid method. The SQLite core calls this routine to +** retrieve the rowid for the current row of the result set. fts3 +** exposes %_content.docid as the rowid for the virtual table. The +** rowid should be written to *pRowid. +*/ +static int fts3RowidMethod(sqlite3_vtab_cursor *pCursor, sqlite_int64 *pRowid){ + Fts3Cursor *pCsr = (Fts3Cursor *) pCursor; + *pRowid = pCsr->iPrevId; + return SQLITE_OK; +} + +/* +** This is the xColumn method, called by SQLite to request a value from +** the row that the supplied cursor currently points to. +*/ +static int fts3ColumnMethod( + sqlite3_vtab_cursor *pCursor, /* Cursor to retrieve value from */ + sqlite3_context *pContext, /* Context for sqlite3_result_xxx() calls */ + int iCol /* Index of column to read value from */ +){ + int rc = SQLITE_OK; /* Return Code */ + Fts3Cursor *pCsr = (Fts3Cursor *) pCursor; + Fts3Table *p = (Fts3Table *)pCursor->pVtab; + + /* The column value supplied by SQLite must be in range. */ + assert( iCol>=0 && iCol<=p->nColumn+1 ); + + if( iCol==p->nColumn+1 ){ + /* This call is a request for the "docid" column. Since "docid" is an + ** alias for "rowid", use the xRowid() method to obtain the value. + */ + sqlite3_result_int64(pContext, pCsr->iPrevId); + }else if( iCol==p->nColumn ){ + /* The extra column whose name is the same as the table. + ** Return a blob which is a pointer to the cursor. + */ + sqlite3_result_blob(pContext, &pCsr, sizeof(pCsr), SQLITE_TRANSIENT); + }else{ + rc = fts3CursorSeek(0, pCsr); + if( rc==SQLITE_OK && sqlite3_data_count(pCsr->pStmt)>(iCol+1) ){ + sqlite3_result_value(pContext, sqlite3_column_value(pCsr->pStmt, iCol+1)); + } + } + + assert( ((Fts3Table *)pCsr->base.pVtab)->pSegments==0 ); + return rc; +} + +/* +** This function is the implementation of the xUpdate callback used by +** FTS3 virtual tables. It is invoked by SQLite each time a row is to be +** inserted, updated or deleted. +*/ +static int fts3UpdateMethod( + sqlite3_vtab *pVtab, /* Virtual table handle */ + int nArg, /* Size of argument array */ + sqlite3_value **apVal, /* Array of arguments */ + sqlite_int64 *pRowid /* OUT: The affected (or effected) rowid */ +){ + return sqlite3Fts3UpdateMethod(pVtab, nArg, apVal, pRowid); +} + +/* +** Implementation of xSync() method. Flush the contents of the pending-terms +** hash-table to the database. +*/ +static int fts3SyncMethod(sqlite3_vtab *pVtab){ + int rc = sqlite3Fts3PendingTermsFlush((Fts3Table *)pVtab); + sqlite3Fts3SegmentsClose((Fts3Table *)pVtab); + return rc; +} + +/* +** Implementation of xBegin() method. This is a no-op. +*/ +static int fts3BeginMethod(sqlite3_vtab *pVtab){ + TESTONLY( Fts3Table *p = (Fts3Table*)pVtab ); + UNUSED_PARAMETER(pVtab); + assert( p->pSegments==0 ); + assert( p->nPendingData==0 ); + assert( p->inTransaction!=1 ); + TESTONLY( p->inTransaction = 1 ); + TESTONLY( p->mxSavepoint = -1; ); + return SQLITE_OK; +} + +/* +** Implementation of xCommit() method. This is a no-op. The contents of +** the pending-terms hash-table have already been flushed into the database +** by fts3SyncMethod(). +*/ +static int fts3CommitMethod(sqlite3_vtab *pVtab){ + TESTONLY( Fts3Table *p = (Fts3Table*)pVtab ); + UNUSED_PARAMETER(pVtab); + assert( p->nPendingData==0 ); + assert( p->inTransaction!=0 ); + assert( p->pSegments==0 ); + TESTONLY( p->inTransaction = 0 ); + TESTONLY( p->mxSavepoint = -1; ); + return SQLITE_OK; +} + +/* +** Implementation of xRollback(). Discard the contents of the pending-terms +** hash-table. Any changes made to the database are reverted by SQLite. +*/ +static int fts3RollbackMethod(sqlite3_vtab *pVtab){ + Fts3Table *p = (Fts3Table*)pVtab; + sqlite3Fts3PendingTermsClear(p); + assert( p->inTransaction!=0 ); + TESTONLY( p->inTransaction = 0 ); + TESTONLY( p->mxSavepoint = -1; ); + return SQLITE_OK; +} + +/* +** When called, *ppPoslist must point to the byte immediately following the +** end of a position-list. i.e. ( (*ppPoslist)[-1]==POS_END ). This function +** moves *ppPoslist so that it instead points to the first byte of the +** same position list. +*/ +static void fts3ReversePoslist(char *pStart, char **ppPoslist){ + char *p = &(*ppPoslist)[-2]; + char c = 0; + + while( p>pStart && (c=*p--)==0 ); + while( p>pStart && (*p & 0x80) | c ){ + c = *p--; + } + if( p>pStart ){ p = &p[2]; } + while( *p++&0x80 ); + *ppPoslist = p; +} + +/* +** Helper function used by the implementation of the overloaded snippet(), +** offsets() and optimize() SQL functions. +** +** If the value passed as the third argument is a blob of size +** sizeof(Fts3Cursor*), then the blob contents are copied to the +** output variable *ppCsr and SQLITE_OK is returned. Otherwise, an error +** message is written to context pContext and SQLITE_ERROR returned. The +** string passed via zFunc is used as part of the error message. +*/ +static int fts3FunctionArg( + sqlite3_context *pContext, /* SQL function call context */ + const char *zFunc, /* Function name */ + sqlite3_value *pVal, /* argv[0] passed to function */ + Fts3Cursor **ppCsr /* OUT: Store cursor handle here */ +){ + Fts3Cursor *pRet; + if( sqlite3_value_type(pVal)!=SQLITE_BLOB + || sqlite3_value_bytes(pVal)!=sizeof(Fts3Cursor *) + ){ + char *zErr = sqlite3_mprintf("illegal first argument to %s", zFunc); + sqlite3_result_error(pContext, zErr, -1); + sqlite3_free(zErr); + return SQLITE_ERROR; + } + memcpy(&pRet, sqlite3_value_blob(pVal), sizeof(Fts3Cursor *)); + *ppCsr = pRet; + return SQLITE_OK; +} + +/* +** Implementation of the snippet() function for FTS3 +*/ +static void fts3SnippetFunc( + sqlite3_context *pContext, /* SQLite function call context */ + int nVal, /* Size of apVal[] array */ + sqlite3_value **apVal /* Array of arguments */ +){ + Fts3Cursor *pCsr; /* Cursor handle passed through apVal[0] */ + const char *zStart = ""; + const char *zEnd = ""; + const char *zEllipsis = "..."; + int iCol = -1; + int nToken = 15; /* Default number of tokens in snippet */ + + /* There must be at least one argument passed to this function (otherwise + ** the non-overloaded version would have been called instead of this one). + */ + assert( nVal>=1 ); + + if( nVal>6 ){ + sqlite3_result_error(pContext, + "wrong number of arguments to function snippet()", -1); + return; + } + if( fts3FunctionArg(pContext, "snippet", apVal[0], &pCsr) ) return; + + switch( nVal ){ + case 6: nToken = sqlite3_value_int(apVal[5]); + case 5: iCol = sqlite3_value_int(apVal[4]); + case 4: zEllipsis = (const char*)sqlite3_value_text(apVal[3]); + case 3: zEnd = (const char*)sqlite3_value_text(apVal[2]); + case 2: zStart = (const char*)sqlite3_value_text(apVal[1]); + } + if( !zEllipsis || !zEnd || !zStart ){ + sqlite3_result_error_nomem(pContext); + }else if( SQLITE_OK==fts3CursorSeek(pContext, pCsr) ){ + sqlite3Fts3Snippet(pContext, pCsr, zStart, zEnd, zEllipsis, iCol, nToken); + } +} + +/* +** Implementation of the offsets() function for FTS3 +*/ +static void fts3OffsetsFunc( + sqlite3_context *pContext, /* SQLite function call context */ + int nVal, /* Size of argument array */ + sqlite3_value **apVal /* Array of arguments */ +){ + Fts3Cursor *pCsr; /* Cursor handle passed through apVal[0] */ + + UNUSED_PARAMETER(nVal); + + assert( nVal==1 ); + if( fts3FunctionArg(pContext, "offsets", apVal[0], &pCsr) ) return; + assert( pCsr ); + if( SQLITE_OK==fts3CursorSeek(pContext, pCsr) ){ + sqlite3Fts3Offsets(pContext, pCsr); + } +} + +/* +** Implementation of the special optimize() function for FTS3. This +** function merges all segments in the database to a single segment. +** Example usage is: +** +** SELECT optimize(t) FROM t LIMIT 1; +** +** where 't' is the name of an FTS3 table. +*/ +static void fts3OptimizeFunc( + sqlite3_context *pContext, /* SQLite function call context */ + int nVal, /* Size of argument array */ + sqlite3_value **apVal /* Array of arguments */ +){ + int rc; /* Return code */ + Fts3Table *p; /* Virtual table handle */ + Fts3Cursor *pCursor; /* Cursor handle passed through apVal[0] */ + + UNUSED_PARAMETER(nVal); + + assert( nVal==1 ); + if( fts3FunctionArg(pContext, "optimize", apVal[0], &pCursor) ) return; + p = (Fts3Table *)pCursor->base.pVtab; + assert( p ); + + rc = sqlite3Fts3Optimize(p); + + switch( rc ){ + case SQLITE_OK: + sqlite3_result_text(pContext, "Index optimized", -1, SQLITE_STATIC); + break; + case SQLITE_DONE: + sqlite3_result_text(pContext, "Index already optimal", -1, SQLITE_STATIC); + break; + default: + sqlite3_result_error_code(pContext, rc); + break; + } +} + +/* +** Implementation of the matchinfo() function for FTS3 +*/ +static void fts3MatchinfoFunc( + sqlite3_context *pContext, /* SQLite function call context */ + int nVal, /* Size of argument array */ + sqlite3_value **apVal /* Array of arguments */ +){ + Fts3Cursor *pCsr; /* Cursor handle passed through apVal[0] */ + assert( nVal==1 || nVal==2 ); + if( SQLITE_OK==fts3FunctionArg(pContext, "matchinfo", apVal[0], &pCsr) ){ + const char *zArg = 0; + if( nVal>1 ){ + zArg = (const char *)sqlite3_value_text(apVal[1]); + } + sqlite3Fts3Matchinfo(pContext, pCsr, zArg); + } +} + +/* +** This routine implements the xFindFunction method for the FTS3 +** virtual table. +*/ +static int fts3FindFunctionMethod( + sqlite3_vtab *pVtab, /* Virtual table handle */ + int nArg, /* Number of SQL function arguments */ + const char *zName, /* Name of SQL function */ + void (**pxFunc)(sqlite3_context*,int,sqlite3_value**), /* OUT: Result */ + void **ppArg /* Unused */ +){ + struct Overloaded { + const char *zName; + void (*xFunc)(sqlite3_context*,int,sqlite3_value**); + } aOverload[] = { + { "snippet", fts3SnippetFunc }, + { "offsets", fts3OffsetsFunc }, + { "optimize", fts3OptimizeFunc }, + { "matchinfo", fts3MatchinfoFunc }, + }; + int i; /* Iterator variable */ + + UNUSED_PARAMETER(pVtab); + UNUSED_PARAMETER(nArg); + UNUSED_PARAMETER(ppArg); + + for(i=0; idb; /* Database connection */ + int rc; /* Return Code */ + + /* As it happens, the pending terms table is always empty here. This is + ** because an "ALTER TABLE RENAME TABLE" statement inside a transaction + ** always opens a savepoint transaction. And the xSavepoint() method + ** flushes the pending terms table. But leave the (no-op) call to + ** PendingTermsFlush() in in case that changes. + */ + assert( p->nPendingData==0 ); + rc = sqlite3Fts3PendingTermsFlush(p); + + if( p->zContentTbl==0 ){ + fts3DbExec(&rc, db, + "ALTER TABLE %Q.'%q_content' RENAME TO '%q_content';", + p->zDb, p->zName, zName + ); + } + + if( p->bHasDocsize ){ + fts3DbExec(&rc, db, + "ALTER TABLE %Q.'%q_docsize' RENAME TO '%q_docsize';", + p->zDb, p->zName, zName + ); + } + if( p->bHasStat ){ + fts3DbExec(&rc, db, + "ALTER TABLE %Q.'%q_stat' RENAME TO '%q_stat';", + p->zDb, p->zName, zName + ); + } + fts3DbExec(&rc, db, + "ALTER TABLE %Q.'%q_segments' RENAME TO '%q_segments';", + p->zDb, p->zName, zName + ); + fts3DbExec(&rc, db, + "ALTER TABLE %Q.'%q_segdir' RENAME TO '%q_segdir';", + p->zDb, p->zName, zName + ); + return rc; +} + +/* +** The xSavepoint() method. +** +** Flush the contents of the pending-terms table to disk. +*/ +static int fts3SavepointMethod(sqlite3_vtab *pVtab, int iSavepoint){ + UNUSED_PARAMETER(iSavepoint); + assert( ((Fts3Table *)pVtab)->inTransaction ); + assert( ((Fts3Table *)pVtab)->mxSavepoint < iSavepoint ); + TESTONLY( ((Fts3Table *)pVtab)->mxSavepoint = iSavepoint ); + return fts3SyncMethod(pVtab); +} + +/* +** The xRelease() method. +** +** This is a no-op. +*/ +static int fts3ReleaseMethod(sqlite3_vtab *pVtab, int iSavepoint){ + TESTONLY( Fts3Table *p = (Fts3Table*)pVtab ); + UNUSED_PARAMETER(iSavepoint); + UNUSED_PARAMETER(pVtab); + assert( p->inTransaction ); + assert( p->mxSavepoint >= iSavepoint ); + TESTONLY( p->mxSavepoint = iSavepoint-1 ); + return SQLITE_OK; +} + +/* +** The xRollbackTo() method. +** +** Discard the contents of the pending terms table. +*/ +static int fts3RollbackToMethod(sqlite3_vtab *pVtab, int iSavepoint){ + Fts3Table *p = (Fts3Table*)pVtab; + UNUSED_PARAMETER(iSavepoint); + assert( p->inTransaction ); + assert( p->mxSavepoint >= iSavepoint ); + TESTONLY( p->mxSavepoint = iSavepoint ); + sqlite3Fts3PendingTermsClear(p); + return SQLITE_OK; +} + +static const sqlite3_module fts3Module = { + /* iVersion */ 2, + /* xCreate */ fts3CreateMethod, + /* xConnect */ fts3ConnectMethod, + /* xBestIndex */ fts3BestIndexMethod, + /* xDisconnect */ fts3DisconnectMethod, + /* xDestroy */ fts3DestroyMethod, + /* xOpen */ fts3OpenMethod, + /* xClose */ fts3CloseMethod, + /* xFilter */ fts3FilterMethod, + /* xNext */ fts3NextMethod, + /* xEof */ fts3EofMethod, + /* xColumn */ fts3ColumnMethod, + /* xRowid */ fts3RowidMethod, + /* xUpdate */ fts3UpdateMethod, + /* xBegin */ fts3BeginMethod, + /* xSync */ fts3SyncMethod, + /* xCommit */ fts3CommitMethod, + /* xRollback */ fts3RollbackMethod, + /* xFindFunction */ fts3FindFunctionMethod, + /* xRename */ fts3RenameMethod, + /* xSavepoint */ fts3SavepointMethod, + /* xRelease */ fts3ReleaseMethod, + /* xRollbackTo */ fts3RollbackToMethod, +}; + +/* +** This function is registered as the module destructor (called when an +** FTS3 enabled database connection is closed). It frees the memory +** allocated for the tokenizer hash table. +*/ +static void hashDestroy(void *p){ + Fts3Hash *pHash = (Fts3Hash *)p; + sqlite3Fts3HashClear(pHash); + sqlite3_free(pHash); +} + +/* +** The fts3 built-in tokenizers - "simple", "porter" and "icu"- are +** implemented in files fts3_tokenizer1.c, fts3_porter.c and fts3_icu.c +** respectively. The following three forward declarations are for functions +** declared in these files used to retrieve the respective implementations. +** +** Calling sqlite3Fts3SimpleTokenizerModule() sets the value pointed +** to by the argument to point to the "simple" tokenizer implementation. +** And so on. +*/ +void sqlite3Fts3SimpleTokenizerModule(sqlite3_tokenizer_module const**ppModule); +void sqlite3Fts3PorterTokenizerModule(sqlite3_tokenizer_module const**ppModule); +#ifdef SQLITE_ENABLE_ICU +void sqlite3Fts3IcuTokenizerModule(sqlite3_tokenizer_module const**ppModule); +#endif + +/* +** Initialise the fts3 extension. If this extension is built as part +** of the sqlite library, then this function is called directly by +** SQLite. If fts3 is built as a dynamically loadable extension, this +** function is called by the sqlite3_extension_init() entry point. +*/ +int sqlite3Fts3Init(sqlite3 *db){ + int rc = SQLITE_OK; + Fts3Hash *pHash = 0; + const sqlite3_tokenizer_module *pSimple = 0; + const sqlite3_tokenizer_module *pPorter = 0; + +#ifdef SQLITE_ENABLE_ICU + const sqlite3_tokenizer_module *pIcu = 0; + sqlite3Fts3IcuTokenizerModule(&pIcu); +#endif + +#ifdef SQLITE_TEST + rc = sqlite3Fts3InitTerm(db); + if( rc!=SQLITE_OK ) return rc; +#endif + + rc = sqlite3Fts3InitAux(db); + if( rc!=SQLITE_OK ) return rc; + + sqlite3Fts3SimpleTokenizerModule(&pSimple); + sqlite3Fts3PorterTokenizerModule(&pPorter); + + /* Allocate and initialise the hash-table used to store tokenizers. */ + pHash = sqlite3_malloc(sizeof(Fts3Hash)); + if( !pHash ){ + rc = SQLITE_NOMEM; + }else{ + sqlite3Fts3HashInit(pHash, FTS3_HASH_STRING, 1); + } + + /* Load the built-in tokenizers into the hash table */ + if( rc==SQLITE_OK ){ + if( sqlite3Fts3HashInsert(pHash, "simple", 7, (void *)pSimple) + || sqlite3Fts3HashInsert(pHash, "porter", 7, (void *)pPorter) +#ifdef SQLITE_ENABLE_ICU + || (pIcu && sqlite3Fts3HashInsert(pHash, "icu", 4, (void *)pIcu)) +#endif + ){ + rc = SQLITE_NOMEM; + } + } + +#ifdef SQLITE_TEST + if( rc==SQLITE_OK ){ + rc = sqlite3Fts3ExprInitTestInterface(db); + } +#endif + + /* Create the virtual table wrapper around the hash-table and overload + ** the two scalar functions. If this is successful, register the + ** module with sqlite. + */ + if( SQLITE_OK==rc + && SQLITE_OK==(rc = sqlite3Fts3InitHashTable(db, pHash, "fts3_tokenizer")) + && SQLITE_OK==(rc = sqlite3_overload_function(db, "snippet", -1)) + && SQLITE_OK==(rc = sqlite3_overload_function(db, "offsets", 1)) + && SQLITE_OK==(rc = sqlite3_overload_function(db, "matchinfo", 1)) + && SQLITE_OK==(rc = sqlite3_overload_function(db, "matchinfo", 2)) + && SQLITE_OK==(rc = sqlite3_overload_function(db, "optimize", 1)) + ){ + rc = sqlite3_create_module_v2( + db, "fts3", &fts3Module, (void *)pHash, hashDestroy + ); + if( rc==SQLITE_OK ){ + rc = sqlite3_create_module_v2( + db, "fts4", &fts3Module, (void *)pHash, 0 + ); + } + return rc; + } + + /* An error has occurred. Delete the hash table and return the error code. */ + assert( rc!=SQLITE_OK ); + if( pHash ){ + sqlite3Fts3HashClear(pHash); + sqlite3_free(pHash); + } + return rc; +} + +/* +** Allocate an Fts3MultiSegReader for each token in the expression headed +** by pExpr. +** +** An Fts3SegReader object is a cursor that can seek or scan a range of +** entries within a single segment b-tree. An Fts3MultiSegReader uses multiple +** Fts3SegReader objects internally to provide an interface to seek or scan +** within the union of all segments of a b-tree. Hence the name. +** +** If the allocated Fts3MultiSegReader just seeks to a single entry in a +** segment b-tree (if the term is not a prefix or it is a prefix for which +** there exists prefix b-tree of the right length) then it may be traversed +** and merged incrementally. Otherwise, it has to be merged into an in-memory +** doclist and then traversed. +*/ +static void fts3EvalAllocateReaders( + Fts3Cursor *pCsr, /* FTS cursor handle */ + Fts3Expr *pExpr, /* Allocate readers for this expression */ + int *pnToken, /* OUT: Total number of tokens in phrase. */ + int *pnOr, /* OUT: Total number of OR nodes in expr. */ + int *pRc /* IN/OUT: Error code */ +){ + if( pExpr && SQLITE_OK==*pRc ){ + if( pExpr->eType==FTSQUERY_PHRASE ){ + int i; + int nToken = pExpr->pPhrase->nToken; + *pnToken += nToken; + for(i=0; ipPhrase->aToken[i]; + int rc = fts3TermSegReaderCursor(pCsr, + pToken->z, pToken->n, pToken->isPrefix, &pToken->pSegcsr + ); + if( rc!=SQLITE_OK ){ + *pRc = rc; + return; + } + } + assert( pExpr->pPhrase->iDoclistToken==0 ); + pExpr->pPhrase->iDoclistToken = -1; + }else{ + *pnOr += (pExpr->eType==FTSQUERY_OR); + fts3EvalAllocateReaders(pCsr, pExpr->pLeft, pnToken, pnOr, pRc); + fts3EvalAllocateReaders(pCsr, pExpr->pRight, pnToken, pnOr, pRc); + } + } +} + +/* +** Arguments pList/nList contain the doclist for token iToken of phrase p. +** It is merged into the main doclist stored in p->doclist.aAll/nAll. +** +** This function assumes that pList points to a buffer allocated using +** sqlite3_malloc(). This function takes responsibility for eventually +** freeing the buffer. +*/ +static void fts3EvalPhraseMergeToken( + Fts3Table *pTab, /* FTS Table pointer */ + Fts3Phrase *p, /* Phrase to merge pList/nList into */ + int iToken, /* Token pList/nList corresponds to */ + char *pList, /* Pointer to doclist */ + int nList /* Number of bytes in pList */ +){ + assert( iToken!=p->iDoclistToken ); + + if( pList==0 ){ + sqlite3_free(p->doclist.aAll); + p->doclist.aAll = 0; + p->doclist.nAll = 0; + } + + else if( p->iDoclistToken<0 ){ + p->doclist.aAll = pList; + p->doclist.nAll = nList; + } + + else if( p->doclist.aAll==0 ){ + sqlite3_free(pList); + } + + else { + char *pLeft; + char *pRight; + int nLeft; + int nRight; + int nDiff; + + if( p->iDoclistTokendoclist.aAll; + nLeft = p->doclist.nAll; + pRight = pList; + nRight = nList; + nDiff = iToken - p->iDoclistToken; + }else{ + pRight = p->doclist.aAll; + nRight = p->doclist.nAll; + pLeft = pList; + nLeft = nList; + nDiff = p->iDoclistToken - iToken; + } + + fts3DoclistPhraseMerge(pTab->bDescIdx, nDiff, pLeft, nLeft, pRight,&nRight); + sqlite3_free(pLeft); + p->doclist.aAll = pRight; + p->doclist.nAll = nRight; + } + + if( iToken>p->iDoclistToken ) p->iDoclistToken = iToken; +} + +/* +** Load the doclist for phrase p into p->doclist.aAll/nAll. The loaded doclist +** does not take deferred tokens into account. +** +** SQLITE_OK is returned if no error occurs, otherwise an SQLite error code. +*/ +static int fts3EvalPhraseLoad( + Fts3Cursor *pCsr, /* FTS Cursor handle */ + Fts3Phrase *p /* Phrase object */ +){ + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + int iToken; + int rc = SQLITE_OK; + + for(iToken=0; rc==SQLITE_OK && iTokennToken; iToken++){ + Fts3PhraseToken *pToken = &p->aToken[iToken]; + assert( pToken->pDeferred==0 || pToken->pSegcsr==0 ); + + if( pToken->pSegcsr ){ + int nThis = 0; + char *pThis = 0; + rc = fts3TermSelect(pTab, pToken, p->iColumn, &nThis, &pThis); + if( rc==SQLITE_OK ){ + fts3EvalPhraseMergeToken(pTab, p, iToken, pThis, nThis); + } + } + assert( pToken->pSegcsr==0 ); + } + + return rc; +} + +/* +** This function is called on each phrase after the position lists for +** any deferred tokens have been loaded into memory. It updates the phrases +** current position list to include only those positions that are really +** instances of the phrase (after considering deferred tokens). If this +** means that the phrase does not appear in the current row, doclist.pList +** and doclist.nList are both zeroed. +** +** SQLITE_OK is returned if no error occurs, otherwise an SQLite error code. +*/ +static int fts3EvalDeferredPhrase(Fts3Cursor *pCsr, Fts3Phrase *pPhrase){ + int iToken; /* Used to iterate through phrase tokens */ + char *aPoslist = 0; /* Position list for deferred tokens */ + int nPoslist = 0; /* Number of bytes in aPoslist */ + int iPrev = -1; /* Token number of previous deferred token */ + + assert( pPhrase->doclist.bFreeList==0 ); + + for(iToken=0; iTokennToken; iToken++){ + Fts3PhraseToken *pToken = &pPhrase->aToken[iToken]; + Fts3DeferredToken *pDeferred = pToken->pDeferred; + + if( pDeferred ){ + char *pList; + int nList; + int rc = sqlite3Fts3DeferredTokenList(pDeferred, &pList, &nList); + if( rc!=SQLITE_OK ) return rc; + + if( pList==0 ){ + sqlite3_free(aPoslist); + pPhrase->doclist.pList = 0; + pPhrase->doclist.nList = 0; + return SQLITE_OK; + + }else if( aPoslist==0 ){ + aPoslist = pList; + nPoslist = nList; + + }else{ + char *aOut = pList; + char *p1 = aPoslist; + char *p2 = aOut; + + assert( iPrev>=0 ); + fts3PoslistPhraseMerge(&aOut, iToken-iPrev, 0, 1, &p1, &p2); + sqlite3_free(aPoslist); + aPoslist = pList; + nPoslist = aOut - aPoslist; + if( nPoslist==0 ){ + sqlite3_free(aPoslist); + pPhrase->doclist.pList = 0; + pPhrase->doclist.nList = 0; + return SQLITE_OK; + } + } + iPrev = iToken; + } + } + + if( iPrev>=0 ){ + int nMaxUndeferred = pPhrase->iDoclistToken; + if( nMaxUndeferred<0 ){ + pPhrase->doclist.pList = aPoslist; + pPhrase->doclist.nList = nPoslist; + pPhrase->doclist.iDocid = pCsr->iPrevId; + pPhrase->doclist.bFreeList = 1; + }else{ + int nDistance; + char *p1; + char *p2; + char *aOut; + + if( nMaxUndeferred>iPrev ){ + p1 = aPoslist; + p2 = pPhrase->doclist.pList; + nDistance = nMaxUndeferred - iPrev; + }else{ + p1 = pPhrase->doclist.pList; + p2 = aPoslist; + nDistance = iPrev - nMaxUndeferred; + } + + aOut = (char *)sqlite3_malloc(nPoslist+8); + if( !aOut ){ + sqlite3_free(aPoslist); + return SQLITE_NOMEM; + } + + pPhrase->doclist.pList = aOut; + if( fts3PoslistPhraseMerge(&aOut, nDistance, 0, 1, &p1, &p2) ){ + pPhrase->doclist.bFreeList = 1; + pPhrase->doclist.nList = (aOut - pPhrase->doclist.pList); + }else{ + sqlite3_free(aOut); + pPhrase->doclist.pList = 0; + pPhrase->doclist.nList = 0; + } + sqlite3_free(aPoslist); + } + } + + return SQLITE_OK; +} + +/* +** This function is called for each Fts3Phrase in a full-text query +** expression to initialize the mechanism for returning rows. Once this +** function has been called successfully on an Fts3Phrase, it may be +** used with fts3EvalPhraseNext() to iterate through the matching docids. +** +** If parameter bOptOk is true, then the phrase may (or may not) use the +** incremental loading strategy. Otherwise, the entire doclist is loaded into +** memory within this call. +** +** SQLITE_OK is returned if no error occurs, otherwise an SQLite error code. +*/ +static int fts3EvalPhraseStart(Fts3Cursor *pCsr, int bOptOk, Fts3Phrase *p){ + int rc; /* Error code */ + Fts3PhraseToken *pFirst = &p->aToken[0]; + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + + if( pCsr->bDesc==pTab->bDescIdx + && bOptOk==1 + && p->nToken==1 + && pFirst->pSegcsr + && pFirst->pSegcsr->bLookup + && pFirst->bFirst==0 + ){ + /* Use the incremental approach. */ + int iCol = (p->iColumn >= pTab->nColumn ? -1 : p->iColumn); + rc = sqlite3Fts3MsrIncrStart( + pTab, pFirst->pSegcsr, iCol, pFirst->z, pFirst->n); + p->bIncr = 1; + + }else{ + /* Load the full doclist for the phrase into memory. */ + rc = fts3EvalPhraseLoad(pCsr, p); + p->bIncr = 0; + } + + assert( rc!=SQLITE_OK || p->nToken<1 || p->aToken[0].pSegcsr==0 || p->bIncr ); + return rc; +} + +/* +** This function is used to iterate backwards (from the end to start) +** through doclists. It is used by this module to iterate through phrase +** doclists in reverse and by the fts3_write.c module to iterate through +** pending-terms lists when writing to databases with "order=desc". +** +** The doclist may be sorted in ascending (parameter bDescIdx==0) or +** descending (parameter bDescIdx==1) order of docid. Regardless, this +** function iterates from the end of the doclist to the beginning. +*/ +void sqlite3Fts3DoclistPrev( + int bDescIdx, /* True if the doclist is desc */ + char *aDoclist, /* Pointer to entire doclist */ + int nDoclist, /* Length of aDoclist in bytes */ + char **ppIter, /* IN/OUT: Iterator pointer */ + sqlite3_int64 *piDocid, /* IN/OUT: Docid pointer */ + int *pnList, /* IN/OUT: List length pointer */ + u8 *pbEof /* OUT: End-of-file flag */ +){ + char *p = *ppIter; + + assert( nDoclist>0 ); + assert( *pbEof==0 ); + assert( p || *piDocid==0 ); + assert( !p || (p>aDoclist && p<&aDoclist[nDoclist]) ); + + if( p==0 ){ + sqlite3_int64 iDocid = 0; + char *pNext = 0; + char *pDocid = aDoclist; + char *pEnd = &aDoclist[nDoclist]; + int iMul = 1; + + while( pDociddoclist; + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + + if( p->bIncr ){ + assert( p->nToken==1 ); + assert( pDL->pNextDocid==0 ); + rc = sqlite3Fts3MsrIncrNext(pTab, p->aToken[0].pSegcsr, + &pDL->iDocid, &pDL->pList, &pDL->nList + ); + if( rc==SQLITE_OK && !pDL->pList ){ + *pbEof = 1; + } + }else if( pCsr->bDesc!=pTab->bDescIdx && pDL->nAll ){ + sqlite3Fts3DoclistPrev(pTab->bDescIdx, pDL->aAll, pDL->nAll, + &pDL->pNextDocid, &pDL->iDocid, &pDL->nList, pbEof + ); + pDL->pList = pDL->pNextDocid; + }else{ + char *pIter; /* Used to iterate through aAll */ + char *pEnd = &pDL->aAll[pDL->nAll]; /* 1 byte past end of aAll */ + if( pDL->pNextDocid ){ + pIter = pDL->pNextDocid; + }else{ + pIter = pDL->aAll; + } + + if( pIter>=pEnd ){ + /* We have already reached the end of this doclist. EOF. */ + *pbEof = 1; + }else{ + sqlite3_int64 iDelta; + pIter += sqlite3Fts3GetVarint(pIter, &iDelta); + if( pTab->bDescIdx==0 || pDL->pNextDocid==0 ){ + pDL->iDocid += iDelta; + }else{ + pDL->iDocid -= iDelta; + } + pDL->pList = pIter; + fts3PoslistCopy(0, &pIter); + pDL->nList = (pIter - pDL->pList); + + /* pIter now points just past the 0x00 that terminates the position- + ** list for document pDL->iDocid. However, if this position-list was + ** edited in place by fts3EvalNearTrim(), then pIter may not actually + ** point to the start of the next docid value. The following line deals + ** with this case by advancing pIter past the zero-padding added by + ** fts3EvalNearTrim(). */ + while( pIterpNextDocid = pIter; + assert( pIter>=&pDL->aAll[pDL->nAll] || *pIter ); + *pbEof = 0; + } + } + + return rc; +} + +/* +** +** If *pRc is not SQLITE_OK when this function is called, it is a no-op. +** Otherwise, fts3EvalPhraseStart() is called on all phrases within the +** expression. Also the Fts3Expr.bDeferred variable is set to true for any +** expressions for which all descendent tokens are deferred. +** +** If parameter bOptOk is zero, then it is guaranteed that the +** Fts3Phrase.doclist.aAll/nAll variables contain the entire doclist for +** each phrase in the expression (subject to deferred token processing). +** Or, if bOptOk is non-zero, then one or more tokens within the expression +** may be loaded incrementally, meaning doclist.aAll/nAll is not available. +** +** If an error occurs within this function, *pRc is set to an SQLite error +** code before returning. +*/ +static void fts3EvalStartReaders( + Fts3Cursor *pCsr, /* FTS Cursor handle */ + Fts3Expr *pExpr, /* Expression to initialize phrases in */ + int bOptOk, /* True to enable incremental loading */ + int *pRc /* IN/OUT: Error code */ +){ + if( pExpr && SQLITE_OK==*pRc ){ + if( pExpr->eType==FTSQUERY_PHRASE ){ + int i; + int nToken = pExpr->pPhrase->nToken; + for(i=0; ipPhrase->aToken[i].pDeferred==0 ) break; + } + pExpr->bDeferred = (i==nToken); + *pRc = fts3EvalPhraseStart(pCsr, bOptOk, pExpr->pPhrase); + }else{ + fts3EvalStartReaders(pCsr, pExpr->pLeft, bOptOk, pRc); + fts3EvalStartReaders(pCsr, pExpr->pRight, bOptOk, pRc); + pExpr->bDeferred = (pExpr->pLeft->bDeferred && pExpr->pRight->bDeferred); + } + } +} + +/* +** An array of the following structures is assembled as part of the process +** of selecting tokens to defer before the query starts executing (as part +** of the xFilter() method). There is one element in the array for each +** token in the FTS expression. +** +** Tokens are divided into AND/NEAR clusters. All tokens in a cluster belong +** to phrases that are connected only by AND and NEAR operators (not OR or +** NOT). When determining tokens to defer, each AND/NEAR cluster is considered +** separately. The root of a tokens AND/NEAR cluster is stored in +** Fts3TokenAndCost.pRoot. +*/ +typedef struct Fts3TokenAndCost Fts3TokenAndCost; +struct Fts3TokenAndCost { + Fts3Phrase *pPhrase; /* The phrase the token belongs to */ + int iToken; /* Position of token in phrase */ + Fts3PhraseToken *pToken; /* The token itself */ + Fts3Expr *pRoot; /* Root of NEAR/AND cluster */ + int nOvfl; /* Number of overflow pages to load doclist */ + int iCol; /* The column the token must match */ +}; + +/* +** This function is used to populate an allocated Fts3TokenAndCost array. +** +** If *pRc is not SQLITE_OK when this function is called, it is a no-op. +** Otherwise, if an error occurs during execution, *pRc is set to an +** SQLite error code. +*/ +static void fts3EvalTokenCosts( + Fts3Cursor *pCsr, /* FTS Cursor handle */ + Fts3Expr *pRoot, /* Root of current AND/NEAR cluster */ + Fts3Expr *pExpr, /* Expression to consider */ + Fts3TokenAndCost **ppTC, /* Write new entries to *(*ppTC)++ */ + Fts3Expr ***ppOr, /* Write new OR root to *(*ppOr)++ */ + int *pRc /* IN/OUT: Error code */ +){ + if( *pRc==SQLITE_OK ){ + if( pExpr->eType==FTSQUERY_PHRASE ){ + Fts3Phrase *pPhrase = pExpr->pPhrase; + int i; + for(i=0; *pRc==SQLITE_OK && inToken; i++){ + Fts3TokenAndCost *pTC = (*ppTC)++; + pTC->pPhrase = pPhrase; + pTC->iToken = i; + pTC->pRoot = pRoot; + pTC->pToken = &pPhrase->aToken[i]; + pTC->iCol = pPhrase->iColumn; + *pRc = sqlite3Fts3MsrOvfl(pCsr, pTC->pToken->pSegcsr, &pTC->nOvfl); + } + }else if( pExpr->eType!=FTSQUERY_NOT ){ + assert( pExpr->eType==FTSQUERY_OR + || pExpr->eType==FTSQUERY_AND + || pExpr->eType==FTSQUERY_NEAR + ); + assert( pExpr->pLeft && pExpr->pRight ); + if( pExpr->eType==FTSQUERY_OR ){ + pRoot = pExpr->pLeft; + **ppOr = pRoot; + (*ppOr)++; + } + fts3EvalTokenCosts(pCsr, pRoot, pExpr->pLeft, ppTC, ppOr, pRc); + if( pExpr->eType==FTSQUERY_OR ){ + pRoot = pExpr->pRight; + **ppOr = pRoot; + (*ppOr)++; + } + fts3EvalTokenCosts(pCsr, pRoot, pExpr->pRight, ppTC, ppOr, pRc); + } + } +} + +/* +** Determine the average document (row) size in pages. If successful, +** write this value to *pnPage and return SQLITE_OK. Otherwise, return +** an SQLite error code. +** +** The average document size in pages is calculated by first calculating +** determining the average size in bytes, B. If B is less than the amount +** of data that will fit on a single leaf page of an intkey table in +** this database, then the average docsize is 1. Otherwise, it is 1 plus +** the number of overflow pages consumed by a record B bytes in size. +*/ +static int fts3EvalAverageDocsize(Fts3Cursor *pCsr, int *pnPage){ + if( pCsr->nRowAvg==0 ){ + /* The average document size, which is required to calculate the cost + ** of each doclist, has not yet been determined. Read the required + ** data from the %_stat table to calculate it. + ** + ** Entry 0 of the %_stat table is a blob containing (nCol+1) FTS3 + ** varints, where nCol is the number of columns in the FTS3 table. + ** The first varint is the number of documents currently stored in + ** the table. The following nCol varints contain the total amount of + ** data stored in all rows of each column of the table, from left + ** to right. + */ + int rc; + Fts3Table *p = (Fts3Table*)pCsr->base.pVtab; + sqlite3_stmt *pStmt; + sqlite3_int64 nDoc = 0; + sqlite3_int64 nByte = 0; + const char *pEnd; + const char *a; + + rc = sqlite3Fts3SelectDoctotal(p, &pStmt); + if( rc!=SQLITE_OK ) return rc; + a = sqlite3_column_blob(pStmt, 0); + assert( a ); + + pEnd = &a[sqlite3_column_bytes(pStmt, 0)]; + a += sqlite3Fts3GetVarint(a, &nDoc); + while( anDoc = nDoc; + pCsr->nRowAvg = (int)(((nByte / nDoc) + p->nPgsz) / p->nPgsz); + assert( pCsr->nRowAvg>0 ); + rc = sqlite3_reset(pStmt); + if( rc!=SQLITE_OK ) return rc; + } + + *pnPage = pCsr->nRowAvg; + return SQLITE_OK; +} + +/* +** This function is called to select the tokens (if any) that will be +** deferred. The array aTC[] has already been populated when this is +** called. +** +** This function is called once for each AND/NEAR cluster in the +** expression. Each invocation determines which tokens to defer within +** the cluster with root node pRoot. See comments above the definition +** of struct Fts3TokenAndCost for more details. +** +** If no error occurs, SQLITE_OK is returned and sqlite3Fts3DeferToken() +** called on each token to defer. Otherwise, an SQLite error code is +** returned. +*/ +static int fts3EvalSelectDeferred( + Fts3Cursor *pCsr, /* FTS Cursor handle */ + Fts3Expr *pRoot, /* Consider tokens with this root node */ + Fts3TokenAndCost *aTC, /* Array of expression tokens and costs */ + int nTC /* Number of entries in aTC[] */ +){ + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + int nDocSize = 0; /* Number of pages per doc loaded */ + int rc = SQLITE_OK; /* Return code */ + int ii; /* Iterator variable for various purposes */ + int nOvfl = 0; /* Total overflow pages used by doclists */ + int nToken = 0; /* Total number of tokens in cluster */ + + int nMinEst = 0; /* The minimum count for any phrase so far. */ + int nLoad4 = 1; /* (Phrases that will be loaded)^4. */ + + /* Tokens are never deferred for FTS tables created using the content=xxx + ** option. The reason being that it is not guaranteed that the content + ** table actually contains the same data as the index. To prevent this from + ** causing any problems, the deferred token optimization is completely + ** disabled for content=xxx tables. */ + if( pTab->zContentTbl ){ + return SQLITE_OK; + } + + /* Count the tokens in this AND/NEAR cluster. If none of the doclists + ** associated with the tokens spill onto overflow pages, or if there is + ** only 1 token, exit early. No tokens to defer in this case. */ + for(ii=0; ii0 ); + + + /* Iterate through all tokens in this AND/NEAR cluster, in ascending order + ** of the number of overflow pages that will be loaded by the pager layer + ** to retrieve the entire doclist for the token from the full-text index. + ** Load the doclists for tokens that are either: + ** + ** a. The cheapest token in the entire query (i.e. the one visited by the + ** first iteration of this loop), or + ** + ** b. Part of a multi-token phrase. + ** + ** After each token doclist is loaded, merge it with the others from the + ** same phrase and count the number of documents that the merged doclist + ** contains. Set variable "nMinEst" to the smallest number of documents in + ** any phrase doclist for which 1 or more token doclists have been loaded. + ** Let nOther be the number of other phrases for which it is certain that + ** one or more tokens will not be deferred. + ** + ** Then, for each token, defer it if loading the doclist would result in + ** loading N or more overflow pages into memory, where N is computed as: + ** + ** (nMinEst + 4^nOther - 1) / (4^nOther) + */ + for(ii=0; iinOvfl) + ){ + pTC = &aTC[iTC]; + } + } + assert( pTC ); + + if( ii && pTC->nOvfl>=((nMinEst+(nLoad4/4)-1)/(nLoad4/4))*nDocSize ){ + /* The number of overflow pages to load for this (and therefore all + ** subsequent) tokens is greater than the estimated number of pages + ** that will be loaded if all subsequent tokens are deferred. + */ + Fts3PhraseToken *pToken = pTC->pToken; + rc = sqlite3Fts3DeferToken(pCsr, pToken, pTC->iCol); + fts3SegReaderCursorFree(pToken->pSegcsr); + pToken->pSegcsr = 0; + }else{ + /* Set nLoad4 to the value of (4^nOther) for the next iteration of the + ** for-loop. Except, limit the value to 2^24 to prevent it from + ** overflowing the 32-bit integer it is stored in. */ + if( ii<12 ) nLoad4 = nLoad4*4; + + if( ii==0 || pTC->pPhrase->nToken>1 ){ + /* Either this is the cheapest token in the entire query, or it is + ** part of a multi-token phrase. Either way, the entire doclist will + ** (eventually) be loaded into memory. It may as well be now. */ + Fts3PhraseToken *pToken = pTC->pToken; + int nList = 0; + char *pList = 0; + rc = fts3TermSelect(pTab, pToken, pTC->iCol, &nList, &pList); + assert( rc==SQLITE_OK || pList==0 ); + if( rc==SQLITE_OK ){ + int nCount; + fts3EvalPhraseMergeToken(pTab, pTC->pPhrase, pTC->iToken,pList,nList); + nCount = fts3DoclistCountDocids( + pTC->pPhrase->doclist.aAll, pTC->pPhrase->doclist.nAll + ); + if( ii==0 || nCountpToken = 0; + } + + return rc; +} + +/* +** This function is called from within the xFilter method. It initializes +** the full-text query currently stored in pCsr->pExpr. To iterate through +** the results of a query, the caller does: +** +** fts3EvalStart(pCsr); +** while( 1 ){ +** fts3EvalNext(pCsr); +** if( pCsr->bEof ) break; +** ... return row pCsr->iPrevId to the caller ... +** } +*/ +static int fts3EvalStart(Fts3Cursor *pCsr){ + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + int rc = SQLITE_OK; + int nToken = 0; + int nOr = 0; + + /* Allocate a MultiSegReader for each token in the expression. */ + fts3EvalAllocateReaders(pCsr, pCsr->pExpr, &nToken, &nOr, &rc); + + /* Determine which, if any, tokens in the expression should be deferred. */ + if( rc==SQLITE_OK && nToken>1 && pTab->bHasStat ){ + Fts3TokenAndCost *aTC; + Fts3Expr **apOr; + aTC = (Fts3TokenAndCost *)sqlite3_malloc( + sizeof(Fts3TokenAndCost) * nToken + + sizeof(Fts3Expr *) * nOr * 2 + ); + apOr = (Fts3Expr **)&aTC[nToken]; + + if( !aTC ){ + rc = SQLITE_NOMEM; + }else{ + int ii; + Fts3TokenAndCost *pTC = aTC; + Fts3Expr **ppOr = apOr; + + fts3EvalTokenCosts(pCsr, 0, pCsr->pExpr, &pTC, &ppOr, &rc); + nToken = pTC-aTC; + nOr = ppOr-apOr; + + if( rc==SQLITE_OK ){ + rc = fts3EvalSelectDeferred(pCsr, 0, aTC, nToken); + for(ii=0; rc==SQLITE_OK && iipExpr, 1, &rc); + return rc; +} + +/* +** Invalidate the current position list for phrase pPhrase. +*/ +static void fts3EvalInvalidatePoslist(Fts3Phrase *pPhrase){ + if( pPhrase->doclist.bFreeList ){ + sqlite3_free(pPhrase->doclist.pList); + } + pPhrase->doclist.pList = 0; + pPhrase->doclist.nList = 0; + pPhrase->doclist.bFreeList = 0; +} + +/* +** This function is called to edit the position list associated with +** the phrase object passed as the fifth argument according to a NEAR +** condition. For example: +** +** abc NEAR/5 "def ghi" +** +** Parameter nNear is passed the NEAR distance of the expression (5 in +** the example above). When this function is called, *paPoslist points to +** the position list, and *pnToken is the number of phrase tokens in, the +** phrase on the other side of the NEAR operator to pPhrase. For example, +** if pPhrase refers to the "def ghi" phrase, then *paPoslist points to +** the position list associated with phrase "abc". +** +** All positions in the pPhrase position list that are not sufficiently +** close to a position in the *paPoslist position list are removed. If this +** leaves 0 positions, zero is returned. Otherwise, non-zero. +** +** Before returning, *paPoslist is set to point to the position lsit +** associated with pPhrase. And *pnToken is set to the number of tokens in +** pPhrase. +*/ +static int fts3EvalNearTrim( + int nNear, /* NEAR distance. As in "NEAR/nNear". */ + char *aTmp, /* Temporary space to use */ + char **paPoslist, /* IN/OUT: Position list */ + int *pnToken, /* IN/OUT: Tokens in phrase of *paPoslist */ + Fts3Phrase *pPhrase /* The phrase object to trim the doclist of */ +){ + int nParam1 = nNear + pPhrase->nToken; + int nParam2 = nNear + *pnToken; + int nNew; + char *p2; + char *pOut; + int res; + + assert( pPhrase->doclist.pList ); + + p2 = pOut = pPhrase->doclist.pList; + res = fts3PoslistNearMerge( + &pOut, aTmp, nParam1, nParam2, paPoslist, &p2 + ); + if( res ){ + nNew = (pOut - pPhrase->doclist.pList) - 1; + assert( pPhrase->doclist.pList[nNew]=='\0' ); + assert( nNew<=pPhrase->doclist.nList && nNew>0 ); + memset(&pPhrase->doclist.pList[nNew], 0, pPhrase->doclist.nList - nNew); + pPhrase->doclist.nList = nNew; + *paPoslist = pPhrase->doclist.pList; + *pnToken = pPhrase->nToken; + } + + return res; +} + +/* +** This function is a no-op if *pRc is other than SQLITE_OK when it is called. +** Otherwise, it advances the expression passed as the second argument to +** point to the next matching row in the database. Expressions iterate through +** matching rows in docid order. Ascending order if Fts3Cursor.bDesc is zero, +** or descending if it is non-zero. +** +** If an error occurs, *pRc is set to an SQLite error code. Otherwise, if +** successful, the following variables in pExpr are set: +** +** Fts3Expr.bEof (non-zero if EOF - there is no next row) +** Fts3Expr.iDocid (valid if bEof==0. The docid of the next row) +** +** If the expression is of type FTSQUERY_PHRASE, and the expression is not +** at EOF, then the following variables are populated with the position list +** for the phrase for the visited row: +** +** FTs3Expr.pPhrase->doclist.nList (length of pList in bytes) +** FTs3Expr.pPhrase->doclist.pList (pointer to position list) +** +** It says above that this function advances the expression to the next +** matching row. This is usually true, but there are the following exceptions: +** +** 1. Deferred tokens are not taken into account. If a phrase consists +** entirely of deferred tokens, it is assumed to match every row in +** the db. In this case the position-list is not populated at all. +** +** Or, if a phrase contains one or more deferred tokens and one or +** more non-deferred tokens, then the expression is advanced to the +** next possible match, considering only non-deferred tokens. In other +** words, if the phrase is "A B C", and "B" is deferred, the expression +** is advanced to the next row that contains an instance of "A * C", +** where "*" may match any single token. The position list in this case +** is populated as for "A * C" before returning. +** +** 2. NEAR is treated as AND. If the expression is "x NEAR y", it is +** advanced to point to the next row that matches "x AND y". +** +** See fts3EvalTestDeferredAndNear() for details on testing if a row is +** really a match, taking into account deferred tokens and NEAR operators. +*/ +static void fts3EvalNextRow( + Fts3Cursor *pCsr, /* FTS Cursor handle */ + Fts3Expr *pExpr, /* Expr. to advance to next matching row */ + int *pRc /* IN/OUT: Error code */ +){ + if( *pRc==SQLITE_OK ){ + int bDescDoclist = pCsr->bDesc; /* Used by DOCID_CMP() macro */ + assert( pExpr->bEof==0 ); + pExpr->bStart = 1; + + switch( pExpr->eType ){ + case FTSQUERY_NEAR: + case FTSQUERY_AND: { + Fts3Expr *pLeft = pExpr->pLeft; + Fts3Expr *pRight = pExpr->pRight; + assert( !pLeft->bDeferred || !pRight->bDeferred ); + + if( pLeft->bDeferred ){ + /* LHS is entirely deferred. So we assume it matches every row. + ** Advance the RHS iterator to find the next row visited. */ + fts3EvalNextRow(pCsr, pRight, pRc); + pExpr->iDocid = pRight->iDocid; + pExpr->bEof = pRight->bEof; + }else if( pRight->bDeferred ){ + /* RHS is entirely deferred. So we assume it matches every row. + ** Advance the LHS iterator to find the next row visited. */ + fts3EvalNextRow(pCsr, pLeft, pRc); + pExpr->iDocid = pLeft->iDocid; + pExpr->bEof = pLeft->bEof; + }else{ + /* Neither the RHS or LHS are deferred. */ + fts3EvalNextRow(pCsr, pLeft, pRc); + fts3EvalNextRow(pCsr, pRight, pRc); + while( !pLeft->bEof && !pRight->bEof && *pRc==SQLITE_OK ){ + sqlite3_int64 iDiff = DOCID_CMP(pLeft->iDocid, pRight->iDocid); + if( iDiff==0 ) break; + if( iDiff<0 ){ + fts3EvalNextRow(pCsr, pLeft, pRc); + }else{ + fts3EvalNextRow(pCsr, pRight, pRc); + } + } + pExpr->iDocid = pLeft->iDocid; + pExpr->bEof = (pLeft->bEof || pRight->bEof); + } + break; + } + + case FTSQUERY_OR: { + Fts3Expr *pLeft = pExpr->pLeft; + Fts3Expr *pRight = pExpr->pRight; + sqlite3_int64 iCmp = DOCID_CMP(pLeft->iDocid, pRight->iDocid); + + assert( pLeft->bStart || pLeft->iDocid==pRight->iDocid ); + assert( pRight->bStart || pLeft->iDocid==pRight->iDocid ); + + if( pRight->bEof || (pLeft->bEof==0 && iCmp<0) ){ + fts3EvalNextRow(pCsr, pLeft, pRc); + }else if( pLeft->bEof || (pRight->bEof==0 && iCmp>0) ){ + fts3EvalNextRow(pCsr, pRight, pRc); + }else{ + fts3EvalNextRow(pCsr, pLeft, pRc); + fts3EvalNextRow(pCsr, pRight, pRc); + } + + pExpr->bEof = (pLeft->bEof && pRight->bEof); + iCmp = DOCID_CMP(pLeft->iDocid, pRight->iDocid); + if( pRight->bEof || (pLeft->bEof==0 && iCmp<0) ){ + pExpr->iDocid = pLeft->iDocid; + }else{ + pExpr->iDocid = pRight->iDocid; + } + + break; + } + + case FTSQUERY_NOT: { + Fts3Expr *pLeft = pExpr->pLeft; + Fts3Expr *pRight = pExpr->pRight; + + if( pRight->bStart==0 ){ + fts3EvalNextRow(pCsr, pRight, pRc); + assert( *pRc!=SQLITE_OK || pRight->bStart ); + } + + fts3EvalNextRow(pCsr, pLeft, pRc); + if( pLeft->bEof==0 ){ + while( !*pRc + && !pRight->bEof + && DOCID_CMP(pLeft->iDocid, pRight->iDocid)>0 + ){ + fts3EvalNextRow(pCsr, pRight, pRc); + } + } + pExpr->iDocid = pLeft->iDocid; + pExpr->bEof = pLeft->bEof; + break; + } + + default: { + Fts3Phrase *pPhrase = pExpr->pPhrase; + fts3EvalInvalidatePoslist(pPhrase); + *pRc = fts3EvalPhraseNext(pCsr, pPhrase, &pExpr->bEof); + pExpr->iDocid = pPhrase->doclist.iDocid; + break; + } + } + } +} + +/* +** If *pRc is not SQLITE_OK, or if pExpr is not the root node of a NEAR +** cluster, then this function returns 1 immediately. +** +** Otherwise, it checks if the current row really does match the NEAR +** expression, using the data currently stored in the position lists +** (Fts3Expr->pPhrase.doclist.pList/nList) for each phrase in the expression. +** +** If the current row is a match, the position list associated with each +** phrase in the NEAR expression is edited in place to contain only those +** phrase instances sufficiently close to their peers to satisfy all NEAR +** constraints. In this case it returns 1. If the NEAR expression does not +** match the current row, 0 is returned. The position lists may or may not +** be edited if 0 is returned. +*/ +static int fts3EvalNearTest(Fts3Expr *pExpr, int *pRc){ + int res = 1; + + /* The following block runs if pExpr is the root of a NEAR query. + ** For example, the query: + ** + ** "w" NEAR "x" NEAR "y" NEAR "z" + ** + ** which is represented in tree form as: + ** + ** | + ** +--NEAR--+ <-- root of NEAR query + ** | | + ** +--NEAR--+ "z" + ** | | + ** +--NEAR--+ "y" + ** | | + ** "w" "x" + ** + ** The right-hand child of a NEAR node is always a phrase. The + ** left-hand child may be either a phrase or a NEAR node. There are + ** no exceptions to this - it's the way the parser in fts3_expr.c works. + */ + if( *pRc==SQLITE_OK + && pExpr->eType==FTSQUERY_NEAR + && pExpr->bEof==0 + && (pExpr->pParent==0 || pExpr->pParent->eType!=FTSQUERY_NEAR) + ){ + Fts3Expr *p; + int nTmp = 0; /* Bytes of temp space */ + char *aTmp; /* Temp space for PoslistNearMerge() */ + + /* Allocate temporary working space. */ + for(p=pExpr; p->pLeft; p=p->pLeft){ + nTmp += p->pRight->pPhrase->doclist.nList; + } + nTmp += p->pPhrase->doclist.nList; + aTmp = sqlite3_malloc(nTmp*2); + if( !aTmp ){ + *pRc = SQLITE_NOMEM; + res = 0; + }else{ + char *aPoslist = p->pPhrase->doclist.pList; + int nToken = p->pPhrase->nToken; + + for(p=p->pParent;res && p && p->eType==FTSQUERY_NEAR; p=p->pParent){ + Fts3Phrase *pPhrase = p->pRight->pPhrase; + int nNear = p->nNear; + res = fts3EvalNearTrim(nNear, aTmp, &aPoslist, &nToken, pPhrase); + } + + aPoslist = pExpr->pRight->pPhrase->doclist.pList; + nToken = pExpr->pRight->pPhrase->nToken; + for(p=pExpr->pLeft; p && res; p=p->pLeft){ + int nNear; + Fts3Phrase *pPhrase; + assert( p->pParent && p->pParent->pLeft==p ); + nNear = p->pParent->nNear; + pPhrase = ( + p->eType==FTSQUERY_NEAR ? p->pRight->pPhrase : p->pPhrase + ); + res = fts3EvalNearTrim(nNear, aTmp, &aPoslist, &nToken, pPhrase); + } + } + + sqlite3_free(aTmp); + } + + return res; +} + +/* +** This function is a helper function for fts3EvalTestDeferredAndNear(). +** Assuming no error occurs or has occurred, It returns non-zero if the +** expression passed as the second argument matches the row that pCsr +** currently points to, or zero if it does not. +** +** If *pRc is not SQLITE_OK when this function is called, it is a no-op. +** If an error occurs during execution of this function, *pRc is set to +** the appropriate SQLite error code. In this case the returned value is +** undefined. +*/ +static int fts3EvalTestExpr( + Fts3Cursor *pCsr, /* FTS cursor handle */ + Fts3Expr *pExpr, /* Expr to test. May or may not be root. */ + int *pRc /* IN/OUT: Error code */ +){ + int bHit = 1; /* Return value */ + if( *pRc==SQLITE_OK ){ + switch( pExpr->eType ){ + case FTSQUERY_NEAR: + case FTSQUERY_AND: + bHit = ( + fts3EvalTestExpr(pCsr, pExpr->pLeft, pRc) + && fts3EvalTestExpr(pCsr, pExpr->pRight, pRc) + && fts3EvalNearTest(pExpr, pRc) + ); + + /* If the NEAR expression does not match any rows, zero the doclist for + ** all phrases involved in the NEAR. This is because the snippet(), + ** offsets() and matchinfo() functions are not supposed to recognize + ** any instances of phrases that are part of unmatched NEAR queries. + ** For example if this expression: + ** + ** ... MATCH 'a OR (b NEAR c)' + ** + ** is matched against a row containing: + ** + ** 'a b d e' + ** + ** then any snippet() should ony highlight the "a" term, not the "b" + ** (as "b" is part of a non-matching NEAR clause). + */ + if( bHit==0 + && pExpr->eType==FTSQUERY_NEAR + && (pExpr->pParent==0 || pExpr->pParent->eType!=FTSQUERY_NEAR) + ){ + Fts3Expr *p; + for(p=pExpr; p->pPhrase==0; p=p->pLeft){ + if( p->pRight->iDocid==pCsr->iPrevId ){ + fts3EvalInvalidatePoslist(p->pRight->pPhrase); + } + } + if( p->iDocid==pCsr->iPrevId ){ + fts3EvalInvalidatePoslist(p->pPhrase); + } + } + + break; + + case FTSQUERY_OR: { + int bHit1 = fts3EvalTestExpr(pCsr, pExpr->pLeft, pRc); + int bHit2 = fts3EvalTestExpr(pCsr, pExpr->pRight, pRc); + bHit = bHit1 || bHit2; + break; + } + + case FTSQUERY_NOT: + bHit = ( + fts3EvalTestExpr(pCsr, pExpr->pLeft, pRc) + && !fts3EvalTestExpr(pCsr, pExpr->pRight, pRc) + ); + break; + + default: { + if( pCsr->pDeferred + && (pExpr->iDocid==pCsr->iPrevId || pExpr->bDeferred) + ){ + Fts3Phrase *pPhrase = pExpr->pPhrase; + assert( pExpr->bDeferred || pPhrase->doclist.bFreeList==0 ); + if( pExpr->bDeferred ){ + fts3EvalInvalidatePoslist(pPhrase); + } + *pRc = fts3EvalDeferredPhrase(pCsr, pPhrase); + bHit = (pPhrase->doclist.pList!=0); + pExpr->iDocid = pCsr->iPrevId; + }else{ + bHit = (pExpr->bEof==0 && pExpr->iDocid==pCsr->iPrevId); + } + break; + } + } + } + return bHit; +} + +/* +** This function is called as the second part of each xNext operation when +** iterating through the results of a full-text query. At this point the +** cursor points to a row that matches the query expression, with the +** following caveats: +** +** * Up until this point, "NEAR" operators in the expression have been +** treated as "AND". +** +** * Deferred tokens have not yet been considered. +** +** If *pRc is not SQLITE_OK when this function is called, it immediately +** returns 0. Otherwise, it tests whether or not after considering NEAR +** operators and deferred tokens the current row is still a match for the +** expression. It returns 1 if both of the following are true: +** +** 1. *pRc is SQLITE_OK when this function returns, and +** +** 2. After scanning the current FTS table row for the deferred tokens, +** it is determined that the row does *not* match the query. +** +** Or, if no error occurs and it seems the current row does match the FTS +** query, return 0. +*/ +static int fts3EvalTestDeferredAndNear(Fts3Cursor *pCsr, int *pRc){ + int rc = *pRc; + int bMiss = 0; + if( rc==SQLITE_OK ){ + + /* If there are one or more deferred tokens, load the current row into + ** memory and scan it to determine the position list for each deferred + ** token. Then, see if this row is really a match, considering deferred + ** tokens and NEAR operators (neither of which were taken into account + ** earlier, by fts3EvalNextRow()). + */ + if( pCsr->pDeferred ){ + rc = fts3CursorSeek(0, pCsr); + if( rc==SQLITE_OK ){ + rc = sqlite3Fts3CacheDeferredDoclists(pCsr); + } + } + bMiss = (0==fts3EvalTestExpr(pCsr, pCsr->pExpr, &rc)); + + /* Free the position-lists accumulated for each deferred token above. */ + sqlite3Fts3FreeDeferredDoclists(pCsr); + *pRc = rc; + } + return (rc==SQLITE_OK && bMiss); +} + +/* +** Advance to the next document that matches the FTS expression in +** Fts3Cursor.pExpr. +*/ +static int fts3EvalNext(Fts3Cursor *pCsr){ + int rc = SQLITE_OK; /* Return Code */ + Fts3Expr *pExpr = pCsr->pExpr; + assert( pCsr->isEof==0 ); + if( pExpr==0 ){ + pCsr->isEof = 1; + }else{ + do { + if( pCsr->isRequireSeek==0 ){ + sqlite3_reset(pCsr->pStmt); + } + assert( sqlite3_data_count(pCsr->pStmt)==0 ); + fts3EvalNextRow(pCsr, pExpr, &rc); + pCsr->isEof = pExpr->bEof; + pCsr->isRequireSeek = 1; + pCsr->isMatchinfoNeeded = 1; + pCsr->iPrevId = pExpr->iDocid; + }while( pCsr->isEof==0 && fts3EvalTestDeferredAndNear(pCsr, &rc) ); + } + return rc; +} + +/* +** Restart interation for expression pExpr so that the next call to +** fts3EvalNext() visits the first row. Do not allow incremental +** loading or merging of phrase doclists for this iteration. +** +** If *pRc is other than SQLITE_OK when this function is called, it is +** a no-op. If an error occurs within this function, *pRc is set to an +** SQLite error code before returning. +*/ +static void fts3EvalRestart( + Fts3Cursor *pCsr, + Fts3Expr *pExpr, + int *pRc +){ + if( pExpr && *pRc==SQLITE_OK ){ + Fts3Phrase *pPhrase = pExpr->pPhrase; + + if( pPhrase ){ + fts3EvalInvalidatePoslist(pPhrase); + if( pPhrase->bIncr ){ + assert( pPhrase->nToken==1 ); + assert( pPhrase->aToken[0].pSegcsr ); + sqlite3Fts3MsrIncrRestart(pPhrase->aToken[0].pSegcsr); + *pRc = fts3EvalPhraseStart(pCsr, 0, pPhrase); + } + + pPhrase->doclist.pNextDocid = 0; + pPhrase->doclist.iDocid = 0; + } + + pExpr->iDocid = 0; + pExpr->bEof = 0; + pExpr->bStart = 0; + + fts3EvalRestart(pCsr, pExpr->pLeft, pRc); + fts3EvalRestart(pCsr, pExpr->pRight, pRc); + } +} + +/* +** After allocating the Fts3Expr.aMI[] array for each phrase in the +** expression rooted at pExpr, the cursor iterates through all rows matched +** by pExpr, calling this function for each row. This function increments +** the values in Fts3Expr.aMI[] according to the position-list currently +** found in Fts3Expr.pPhrase->doclist.pList for each of the phrase +** expression nodes. +*/ +static void fts3EvalUpdateCounts(Fts3Expr *pExpr){ + if( pExpr ){ + Fts3Phrase *pPhrase = pExpr->pPhrase; + if( pPhrase && pPhrase->doclist.pList ){ + int iCol = 0; + char *p = pPhrase->doclist.pList; + + assert( *p ); + while( 1 ){ + u8 c = 0; + int iCnt = 0; + while( 0xFE & (*p | c) ){ + if( (c&0x80)==0 ) iCnt++; + c = *p++ & 0x80; + } + + /* aMI[iCol*3 + 1] = Number of occurrences + ** aMI[iCol*3 + 2] = Number of rows containing at least one instance + */ + pExpr->aMI[iCol*3 + 1] += iCnt; + pExpr->aMI[iCol*3 + 2] += (iCnt>0); + if( *p==0x00 ) break; + p++; + p += sqlite3Fts3GetVarint32(p, &iCol); + } + } + + fts3EvalUpdateCounts(pExpr->pLeft); + fts3EvalUpdateCounts(pExpr->pRight); + } +} + +/* +** Expression pExpr must be of type FTSQUERY_PHRASE. +** +** If it is not already allocated and populated, this function allocates and +** populates the Fts3Expr.aMI[] array for expression pExpr. If pExpr is part +** of a NEAR expression, then it also allocates and populates the same array +** for all other phrases that are part of the NEAR expression. +** +** SQLITE_OK is returned if the aMI[] array is successfully allocated and +** populated. Otherwise, if an error occurs, an SQLite error code is returned. +*/ +static int fts3EvalGatherStats( + Fts3Cursor *pCsr, /* Cursor object */ + Fts3Expr *pExpr /* FTSQUERY_PHRASE expression */ +){ + int rc = SQLITE_OK; /* Return code */ + + assert( pExpr->eType==FTSQUERY_PHRASE ); + if( pExpr->aMI==0 ){ + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + Fts3Expr *pRoot; /* Root of NEAR expression */ + Fts3Expr *p; /* Iterator used for several purposes */ + + sqlite3_int64 iPrevId = pCsr->iPrevId; + sqlite3_int64 iDocid; + u8 bEof; + + /* Find the root of the NEAR expression */ + pRoot = pExpr; + while( pRoot->pParent && pRoot->pParent->eType==FTSQUERY_NEAR ){ + pRoot = pRoot->pParent; + } + iDocid = pRoot->iDocid; + bEof = pRoot->bEof; + assert( pRoot->bStart ); + + /* Allocate space for the aMSI[] array of each FTSQUERY_PHRASE node */ + for(p=pRoot; p; p=p->pLeft){ + Fts3Expr *pE = (p->eType==FTSQUERY_PHRASE?p:p->pRight); + assert( pE->aMI==0 ); + pE->aMI = (u32 *)sqlite3_malloc(pTab->nColumn * 3 * sizeof(u32)); + if( !pE->aMI ) return SQLITE_NOMEM; + memset(pE->aMI, 0, pTab->nColumn * 3 * sizeof(u32)); + } + + fts3EvalRestart(pCsr, pRoot, &rc); + + while( pCsr->isEof==0 && rc==SQLITE_OK ){ + + do { + /* Ensure the %_content statement is reset. */ + if( pCsr->isRequireSeek==0 ) sqlite3_reset(pCsr->pStmt); + assert( sqlite3_data_count(pCsr->pStmt)==0 ); + + /* Advance to the next document */ + fts3EvalNextRow(pCsr, pRoot, &rc); + pCsr->isEof = pRoot->bEof; + pCsr->isRequireSeek = 1; + pCsr->isMatchinfoNeeded = 1; + pCsr->iPrevId = pRoot->iDocid; + }while( pCsr->isEof==0 + && pRoot->eType==FTSQUERY_NEAR + && fts3EvalTestDeferredAndNear(pCsr, &rc) + ); + + if( rc==SQLITE_OK && pCsr->isEof==0 ){ + fts3EvalUpdateCounts(pRoot); + } + } + + pCsr->isEof = 0; + pCsr->iPrevId = iPrevId; + + if( bEof ){ + pRoot->bEof = bEof; + }else{ + /* Caution: pRoot may iterate through docids in ascending or descending + ** order. For this reason, even though it seems more defensive, the + ** do loop can not be written: + ** + ** do {...} while( pRoot->iDocidbEof==0 ); + }while( pRoot->iDocid!=iDocid && rc==SQLITE_OK ); + fts3EvalTestDeferredAndNear(pCsr, &rc); + } + } + return rc; +} + +/* +** This function is used by the matchinfo() module to query a phrase +** expression node for the following information: +** +** 1. The total number of occurrences of the phrase in each column of +** the FTS table (considering all rows), and +** +** 2. For each column, the number of rows in the table for which the +** column contains at least one instance of the phrase. +** +** If no error occurs, SQLITE_OK is returned and the values for each column +** written into the array aiOut as follows: +** +** aiOut[iCol*3 + 1] = Number of occurrences +** aiOut[iCol*3 + 2] = Number of rows containing at least one instance +** +** Caveats: +** +** * If a phrase consists entirely of deferred tokens, then all output +** values are set to the number of documents in the table. In other +** words we assume that very common tokens occur exactly once in each +** column of each row of the table. +** +** * If a phrase contains some deferred tokens (and some non-deferred +** tokens), count the potential occurrence identified by considering +** the non-deferred tokens instead of actual phrase occurrences. +** +** * If the phrase is part of a NEAR expression, then only phrase instances +** that meet the NEAR constraint are included in the counts. +*/ +int sqlite3Fts3EvalPhraseStats( + Fts3Cursor *pCsr, /* FTS cursor handle */ + Fts3Expr *pExpr, /* Phrase expression */ + u32 *aiOut /* Array to write results into (see above) */ +){ + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + int rc = SQLITE_OK; + int iCol; + + if( pExpr->bDeferred && pExpr->pParent->eType!=FTSQUERY_NEAR ){ + assert( pCsr->nDoc>0 ); + for(iCol=0; iColnColumn; iCol++){ + aiOut[iCol*3 + 1] = (u32)pCsr->nDoc; + aiOut[iCol*3 + 2] = (u32)pCsr->nDoc; + } + }else{ + rc = fts3EvalGatherStats(pCsr, pExpr); + if( rc==SQLITE_OK ){ + assert( pExpr->aMI ); + for(iCol=0; iColnColumn; iCol++){ + aiOut[iCol*3 + 1] = pExpr->aMI[iCol*3 + 1]; + aiOut[iCol*3 + 2] = pExpr->aMI[iCol*3 + 2]; + } + } + } + + return rc; +} + +/* +** The expression pExpr passed as the second argument to this function +** must be of type FTSQUERY_PHRASE. +** +** The returned value is either NULL or a pointer to a buffer containing +** a position-list indicating the occurrences of the phrase in column iCol +** of the current row. +** +** More specifically, the returned buffer contains 1 varint for each +** occurence of the phrase in the column, stored using the normal (delta+2) +** compression and is terminated by either an 0x01 or 0x00 byte. For example, +** if the requested column contains "a b X c d X X" and the position-list +** for 'X' is requested, the buffer returned may contain: +** +** 0x04 0x05 0x03 0x01 or 0x04 0x05 0x03 0x00 +** +** This function works regardless of whether or not the phrase is deferred, +** incremental, or neither. +*/ +char *sqlite3Fts3EvalPhrasePoslist( + Fts3Cursor *pCsr, /* FTS3 cursor object */ + Fts3Expr *pExpr, /* Phrase to return doclist for */ + int iCol /* Column to return position list for */ +){ + Fts3Phrase *pPhrase = pExpr->pPhrase; + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + char *pIter = pPhrase->doclist.pList; + int iThis; + + assert( iCol>=0 && iColnColumn ); + if( !pIter + || pExpr->bEof + || pExpr->iDocid!=pCsr->iPrevId + || (pPhrase->iColumnnColumn && pPhrase->iColumn!=iCol) + ){ + return 0; + } + + assert( pPhrase->doclist.nList>0 ); + if( *pIter==0x01 ){ + pIter++; + pIter += sqlite3Fts3GetVarint32(pIter, &iThis); + }else{ + iThis = 0; + } + while( iThisdoclist, and +** * any Fts3MultiSegReader objects held by phrase tokens. +*/ +void sqlite3Fts3EvalPhraseCleanup(Fts3Phrase *pPhrase){ + if( pPhrase ){ + int i; + sqlite3_free(pPhrase->doclist.aAll); + fts3EvalInvalidatePoslist(pPhrase); + memset(&pPhrase->doclist, 0, sizeof(Fts3Doclist)); + for(i=0; inToken; i++){ + fts3SegReaderCursorFree(pPhrase->aToken[i].pSegcsr); + pPhrase->aToken[i].pSegcsr = 0; + } + } +} + +/* +** Return SQLITE_CORRUPT_VTAB. +*/ +#ifdef SQLITE_DEBUG +int sqlite3Fts3Corrupt(){ + return SQLITE_CORRUPT_VTAB; +} +#endif + +#if !SQLITE_CORE +/* +** Initialize API pointer table, if required. +*/ +int sqlite3_extension_init( + sqlite3 *db, + char **pzErrMsg, + const sqlite3_api_routines *pApi +){ + SQLITE_EXTENSION_INIT2(pApi) + return sqlite3Fts3Init(db); +} +#endif + +#endif diff --git a/ext/fts3/fts3.h b/ext/fts3/fts3.h new file mode 100644 index 0000000..c1aa8ca --- /dev/null +++ b/ext/fts3/fts3.h @@ -0,0 +1,26 @@ +/* +** 2006 Oct 10 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This header file is used by programs that want to link against the +** FTS3 library. All it does is declare the sqlite3Fts3Init() interface. +*/ +#include "sqlite3.h" + +#ifdef __cplusplus +extern "C" { +#endif /* __cplusplus */ + +int sqlite3Fts3Init(sqlite3 *db); + +#ifdef __cplusplus +} /* extern "C" */ +#endif /* __cplusplus */ diff --git a/ext/fts3/fts3Int.h b/ext/fts3/fts3Int.h new file mode 100644 index 0000000..78392ec --- /dev/null +++ b/ext/fts3/fts3Int.h @@ -0,0 +1,523 @@ +/* +** 2009 Nov 12 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +*/ +#ifndef _FTSINT_H +#define _FTSINT_H + +#if !defined(NDEBUG) && !defined(SQLITE_DEBUG) +# define NDEBUG 1 +#endif + +/* +** FTS4 is really an extension for FTS3. It is enabled using the +** SQLITE_ENABLE_FTS3 macro. But to avoid confusion we also all +** the SQLITE_ENABLE_FTS4 macro to serve as an alisse for SQLITE_ENABLE_FTS3. +*/ +#if defined(SQLITE_ENABLE_FTS4) && !defined(SQLITE_ENABLE_FTS3) +# define SQLITE_ENABLE_FTS3 +#endif + +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) + +/* If not building as part of the core, include sqlite3ext.h. */ +#ifndef SQLITE_CORE +# include "sqlite3ext.h" +extern const sqlite3_api_routines *sqlite3_api; +#endif + +#include "sqlite3.h" +#include "fts3_tokenizer.h" +#include "fts3_hash.h" + +/* +** This constant controls how often segments are merged. Once there are +** FTS3_MERGE_COUNT segments of level N, they are merged into a single +** segment of level N+1. +*/ +#define FTS3_MERGE_COUNT 16 + +/* +** This is the maximum amount of data (in bytes) to store in the +** Fts3Table.pendingTerms hash table. Normally, the hash table is +** populated as documents are inserted/updated/deleted in a transaction +** and used to create a new segment when the transaction is committed. +** However if this limit is reached midway through a transaction, a new +** segment is created and the hash table cleared immediately. +*/ +#define FTS3_MAX_PENDING_DATA (1*1024*1024) + +/* +** Macro to return the number of elements in an array. SQLite has a +** similar macro called ArraySize(). Use a different name to avoid +** a collision when building an amalgamation with built-in FTS3. +*/ +#define SizeofArray(X) ((int)(sizeof(X)/sizeof(X[0]))) + + +#ifndef MIN +# define MIN(x,y) ((x)<(y)?(x):(y)) +#endif + +/* +** Maximum length of a varint encoded integer. The varint format is different +** from that used by SQLite, so the maximum length is 10, not 9. +*/ +#define FTS3_VARINT_MAX 10 + +/* +** FTS4 virtual tables may maintain multiple indexes - one index of all terms +** in the document set and zero or more prefix indexes. All indexes are stored +** as one or more b+-trees in the %_segments and %_segdir tables. +** +** It is possible to determine which index a b+-tree belongs to based on the +** value stored in the "%_segdir.level" column. Given this value L, the index +** that the b+-tree belongs to is (L<<10). In other words, all b+-trees with +** level values between 0 and 1023 (inclusive) belong to index 0, all levels +** between 1024 and 2047 to index 1, and so on. +** +** It is considered impossible for an index to use more than 1024 levels. In +** theory though this may happen, but only after at least +** (FTS3_MERGE_COUNT^1024) separate flushes of the pending-terms tables. +*/ +#define FTS3_SEGDIR_MAXLEVEL 1024 +#define FTS3_SEGDIR_MAXLEVEL_STR "1024" + +/* +** The testcase() macro is only used by the amalgamation. If undefined, +** make it a no-op. +*/ +#ifndef testcase +# define testcase(X) +#endif + +/* +** Terminator values for position-lists and column-lists. +*/ +#define POS_COLUMN (1) /* Column-list terminator */ +#define POS_END (0) /* Position-list terminator */ + +/* +** This section provides definitions to allow the +** FTS3 extension to be compiled outside of the +** amalgamation. +*/ +#ifndef SQLITE_AMALGAMATION +/* +** Macros indicating that conditional expressions are always true or +** false. +*/ +#ifdef SQLITE_COVERAGE_TEST +# define ALWAYS(x) (1) +# define NEVER(X) (0) +#else +# define ALWAYS(x) (x) +# define NEVER(X) (x) +#endif + +/* +** Internal types used by SQLite. +*/ +typedef unsigned char u8; /* 1-byte (or larger) unsigned integer */ +typedef short int i16; /* 2-byte (or larger) signed integer */ +typedef unsigned int u32; /* 4-byte unsigned integer */ +typedef sqlite3_uint64 u64; /* 8-byte unsigned integer */ + +/* +** Macro used to suppress compiler warnings for unused parameters. +*/ +#define UNUSED_PARAMETER(x) (void)(x) + +/* +** Activate assert() only if SQLITE_TEST is enabled. +*/ +#if !defined(NDEBUG) && !defined(SQLITE_DEBUG) +# define NDEBUG 1 +#endif + +/* +** The TESTONLY macro is used to enclose variable declarations or +** other bits of code that are needed to support the arguments +** within testcase() and assert() macros. +*/ +#if defined(SQLITE_DEBUG) || defined(SQLITE_COVERAGE_TEST) +# define TESTONLY(X) X +#else +# define TESTONLY(X) +#endif + +#endif /* SQLITE_AMALGAMATION */ + +#ifdef SQLITE_DEBUG +int sqlite3Fts3Corrupt(void); +# define FTS_CORRUPT_VTAB sqlite3Fts3Corrupt() +#else +# define FTS_CORRUPT_VTAB SQLITE_CORRUPT_VTAB +#endif + +typedef struct Fts3Table Fts3Table; +typedef struct Fts3Cursor Fts3Cursor; +typedef struct Fts3Expr Fts3Expr; +typedef struct Fts3Phrase Fts3Phrase; +typedef struct Fts3PhraseToken Fts3PhraseToken; + +typedef struct Fts3Doclist Fts3Doclist; +typedef struct Fts3SegFilter Fts3SegFilter; +typedef struct Fts3DeferredToken Fts3DeferredToken; +typedef struct Fts3SegReader Fts3SegReader; +typedef struct Fts3MultiSegReader Fts3MultiSegReader; + +/* +** A connection to a fulltext index is an instance of the following +** structure. The xCreate and xConnect methods create an instance +** of this structure and xDestroy and xDisconnect free that instance. +** All other methods receive a pointer to the structure as one of their +** arguments. +*/ +struct Fts3Table { + sqlite3_vtab base; /* Base class used by SQLite core */ + sqlite3 *db; /* The database connection */ + const char *zDb; /* logical database name */ + const char *zName; /* virtual table name */ + int nColumn; /* number of named columns in virtual table */ + char **azColumn; /* column names. malloced */ + sqlite3_tokenizer *pTokenizer; /* tokenizer for inserts and queries */ + char *zContentTbl; /* content=xxx option, or NULL */ + + /* Precompiled statements used by the implementation. Each of these + ** statements is run and reset within a single virtual table API call. + */ + sqlite3_stmt *aStmt[27]; + + char *zReadExprlist; + char *zWriteExprlist; + + int nNodeSize; /* Soft limit for node size */ + u8 bHasStat; /* True if %_stat table exists */ + u8 bHasDocsize; /* True if %_docsize table exists */ + u8 bDescIdx; /* True if doclists are in reverse order */ + int nPgsz; /* Page size for host database */ + char *zSegmentsTbl; /* Name of %_segments table */ + sqlite3_blob *pSegments; /* Blob handle open on %_segments table */ + + /* TODO: Fix the first paragraph of this comment. + ** + ** The following hash table is used to buffer pending index updates during + ** transactions. Variable nPendingData estimates the memory size of the + ** pending data, including hash table overhead, but not malloc overhead. + ** When nPendingData exceeds nMaxPendingData, the buffer is flushed + ** automatically. Variable iPrevDocid is the docid of the most recently + ** inserted record. + ** + ** A single FTS4 table may have multiple full-text indexes. For each index + ** there is an entry in the aIndex[] array. Index 0 is an index of all the + ** terms that appear in the document set. Each subsequent index in aIndex[] + ** is an index of prefixes of a specific length. + */ + int nIndex; /* Size of aIndex[] */ + struct Fts3Index { + int nPrefix; /* Prefix length (0 for main terms index) */ + Fts3Hash hPending; /* Pending terms table for this index */ + } *aIndex; + int nMaxPendingData; /* Max pending data before flush to disk */ + int nPendingData; /* Current bytes of pending data */ + sqlite_int64 iPrevDocid; /* Docid of most recently inserted document */ + +#if defined(SQLITE_DEBUG) || defined(SQLITE_COVERAGE_TEST) + /* State variables used for validating that the transaction control + ** methods of the virtual table are called at appropriate times. These + ** values do not contribution to the FTS computation; they are used for + ** verifying the SQLite core. + */ + int inTransaction; /* True after xBegin but before xCommit/xRollback */ + int mxSavepoint; /* Largest valid xSavepoint integer */ +#endif +}; + +/* +** When the core wants to read from the virtual table, it creates a +** virtual table cursor (an instance of the following structure) using +** the xOpen method. Cursors are destroyed using the xClose method. +*/ +struct Fts3Cursor { + sqlite3_vtab_cursor base; /* Base class used by SQLite core */ + i16 eSearch; /* Search strategy (see below) */ + u8 isEof; /* True if at End Of Results */ + u8 isRequireSeek; /* True if must seek pStmt to %_content row */ + sqlite3_stmt *pStmt; /* Prepared statement in use by the cursor */ + Fts3Expr *pExpr; /* Parsed MATCH query string */ + int nPhrase; /* Number of matchable phrases in query */ + Fts3DeferredToken *pDeferred; /* Deferred search tokens, if any */ + sqlite3_int64 iPrevId; /* Previous id read from aDoclist */ + char *pNextId; /* Pointer into the body of aDoclist */ + char *aDoclist; /* List of docids for full-text queries */ + int nDoclist; /* Size of buffer at aDoclist */ + u8 bDesc; /* True to sort in descending order */ + int eEvalmode; /* An FTS3_EVAL_XX constant */ + int nRowAvg; /* Average size of database rows, in pages */ + sqlite3_int64 nDoc; /* Documents in table */ + + int isMatchinfoNeeded; /* True when aMatchinfo[] needs filling in */ + u32 *aMatchinfo; /* Information about most recent match */ + int nMatchinfo; /* Number of elements in aMatchinfo[] */ + char *zMatchinfo; /* Matchinfo specification */ +}; + +#define FTS3_EVAL_FILTER 0 +#define FTS3_EVAL_NEXT 1 +#define FTS3_EVAL_MATCHINFO 2 + +/* +** The Fts3Cursor.eSearch member is always set to one of the following. +** Actualy, Fts3Cursor.eSearch can be greater than or equal to +** FTS3_FULLTEXT_SEARCH. If so, then Fts3Cursor.eSearch - 2 is the index +** of the column to be searched. For example, in +** +** CREATE VIRTUAL TABLE ex1 USING fts3(a,b,c,d); +** SELECT docid FROM ex1 WHERE b MATCH 'one two three'; +** +** Because the LHS of the MATCH operator is 2nd column "b", +** Fts3Cursor.eSearch will be set to FTS3_FULLTEXT_SEARCH+1. (+0 for a, +** +1 for b, +2 for c, +3 for d.) If the LHS of MATCH were "ex1" +** indicating that all columns should be searched, +** then eSearch would be set to FTS3_FULLTEXT_SEARCH+4. +*/ +#define FTS3_FULLSCAN_SEARCH 0 /* Linear scan of %_content table */ +#define FTS3_DOCID_SEARCH 1 /* Lookup by rowid on %_content table */ +#define FTS3_FULLTEXT_SEARCH 2 /* Full-text index search */ + + +struct Fts3Doclist { + char *aAll; /* Array containing doclist (or NULL) */ + int nAll; /* Size of a[] in bytes */ + char *pNextDocid; /* Pointer to next docid */ + + sqlite3_int64 iDocid; /* Current docid (if pList!=0) */ + int bFreeList; /* True if pList should be sqlite3_free()d */ + char *pList; /* Pointer to position list following iDocid */ + int nList; /* Length of position list */ +}; + +/* +** A "phrase" is a sequence of one or more tokens that must match in +** sequence. A single token is the base case and the most common case. +** For a sequence of tokens contained in double-quotes (i.e. "one two three") +** nToken will be the number of tokens in the string. +*/ +struct Fts3PhraseToken { + char *z; /* Text of the token */ + int n; /* Number of bytes in buffer z */ + int isPrefix; /* True if token ends with a "*" character */ + int bFirst; /* True if token must appear at position 0 */ + + /* Variables above this point are populated when the expression is + ** parsed (by code in fts3_expr.c). Below this point the variables are + ** used when evaluating the expression. */ + Fts3DeferredToken *pDeferred; /* Deferred token object for this token */ + Fts3MultiSegReader *pSegcsr; /* Segment-reader for this token */ +}; + +struct Fts3Phrase { + /* Cache of doclist for this phrase. */ + Fts3Doclist doclist; + int bIncr; /* True if doclist is loaded incrementally */ + int iDoclistToken; + + /* Variables below this point are populated by fts3_expr.c when parsing + ** a MATCH expression. Everything above is part of the evaluation phase. + */ + int nToken; /* Number of tokens in the phrase */ + int iColumn; /* Index of column this phrase must match */ + Fts3PhraseToken aToken[1]; /* One entry for each token in the phrase */ +}; + +/* +** A tree of these objects forms the RHS of a MATCH operator. +** +** If Fts3Expr.eType is FTSQUERY_PHRASE and isLoaded is true, then aDoclist +** points to a malloced buffer, size nDoclist bytes, containing the results +** of this phrase query in FTS3 doclist format. As usual, the initial +** "Length" field found in doclists stored on disk is omitted from this +** buffer. +** +** Variable aMI is used only for FTSQUERY_NEAR nodes to store the global +** matchinfo data. If it is not NULL, it points to an array of size nCol*3, +** where nCol is the number of columns in the queried FTS table. The array +** is populated as follows: +** +** aMI[iCol*3 + 0] = Undefined +** aMI[iCol*3 + 1] = Number of occurrences +** aMI[iCol*3 + 2] = Number of rows containing at least one instance +** +** The aMI array is allocated using sqlite3_malloc(). It should be freed +** when the expression node is. +*/ +struct Fts3Expr { + int eType; /* One of the FTSQUERY_XXX values defined below */ + int nNear; /* Valid if eType==FTSQUERY_NEAR */ + Fts3Expr *pParent; /* pParent->pLeft==this or pParent->pRight==this */ + Fts3Expr *pLeft; /* Left operand */ + Fts3Expr *pRight; /* Right operand */ + Fts3Phrase *pPhrase; /* Valid if eType==FTSQUERY_PHRASE */ + + /* The following are used by the fts3_eval.c module. */ + sqlite3_int64 iDocid; /* Current docid */ + u8 bEof; /* True this expression is at EOF already */ + u8 bStart; /* True if iDocid is valid */ + u8 bDeferred; /* True if this expression is entirely deferred */ + + u32 *aMI; +}; + +/* +** Candidate values for Fts3Query.eType. Note that the order of the first +** four values is in order of precedence when parsing expressions. For +** example, the following: +** +** "a OR b AND c NOT d NEAR e" +** +** is equivalent to: +** +** "a OR (b AND (c NOT (d NEAR e)))" +*/ +#define FTSQUERY_NEAR 1 +#define FTSQUERY_NOT 2 +#define FTSQUERY_AND 3 +#define FTSQUERY_OR 4 +#define FTSQUERY_PHRASE 5 + + +/* fts3_write.c */ +int sqlite3Fts3UpdateMethod(sqlite3_vtab*,int,sqlite3_value**,sqlite3_int64*); +int sqlite3Fts3PendingTermsFlush(Fts3Table *); +void sqlite3Fts3PendingTermsClear(Fts3Table *); +int sqlite3Fts3Optimize(Fts3Table *); +int sqlite3Fts3SegReaderNew(int, sqlite3_int64, + sqlite3_int64, sqlite3_int64, const char *, int, Fts3SegReader**); +int sqlite3Fts3SegReaderPending( + Fts3Table*,int,const char*,int,int,Fts3SegReader**); +void sqlite3Fts3SegReaderFree(Fts3SegReader *); +int sqlite3Fts3AllSegdirs(Fts3Table*, int, int, sqlite3_stmt **); +int sqlite3Fts3ReadLock(Fts3Table *); +int sqlite3Fts3ReadBlock(Fts3Table*, sqlite3_int64, char **, int*, int*); + +int sqlite3Fts3SelectDoctotal(Fts3Table *, sqlite3_stmt **); +int sqlite3Fts3SelectDocsize(Fts3Table *, sqlite3_int64, sqlite3_stmt **); + +void sqlite3Fts3FreeDeferredTokens(Fts3Cursor *); +int sqlite3Fts3DeferToken(Fts3Cursor *, Fts3PhraseToken *, int); +int sqlite3Fts3CacheDeferredDoclists(Fts3Cursor *); +void sqlite3Fts3FreeDeferredDoclists(Fts3Cursor *); +void sqlite3Fts3SegmentsClose(Fts3Table *); + +/* Special values interpreted by sqlite3SegReaderCursor() */ +#define FTS3_SEGCURSOR_PENDING -1 +#define FTS3_SEGCURSOR_ALL -2 + +int sqlite3Fts3SegReaderStart(Fts3Table*, Fts3MultiSegReader*, Fts3SegFilter*); +int sqlite3Fts3SegReaderStep(Fts3Table *, Fts3MultiSegReader *); +void sqlite3Fts3SegReaderFinish(Fts3MultiSegReader *); + +int sqlite3Fts3SegReaderCursor( + Fts3Table *, int, int, const char *, int, int, int, Fts3MultiSegReader *); + +/* Flags allowed as part of the 4th argument to SegmentReaderIterate() */ +#define FTS3_SEGMENT_REQUIRE_POS 0x00000001 +#define FTS3_SEGMENT_IGNORE_EMPTY 0x00000002 +#define FTS3_SEGMENT_COLUMN_FILTER 0x00000004 +#define FTS3_SEGMENT_PREFIX 0x00000008 +#define FTS3_SEGMENT_SCAN 0x00000010 +#define FTS3_SEGMENT_FIRST 0x00000020 + +/* Type passed as 4th argument to SegmentReaderIterate() */ +struct Fts3SegFilter { + const char *zTerm; + int nTerm; + int iCol; + int flags; +}; + +struct Fts3MultiSegReader { + /* Used internally by sqlite3Fts3SegReaderXXX() calls */ + Fts3SegReader **apSegment; /* Array of Fts3SegReader objects */ + int nSegment; /* Size of apSegment array */ + int nAdvance; /* How many seg-readers to advance */ + Fts3SegFilter *pFilter; /* Pointer to filter object */ + char *aBuffer; /* Buffer to merge doclists in */ + int nBuffer; /* Allocated size of aBuffer[] in bytes */ + + int iColFilter; /* If >=0, filter for this column */ + int bRestart; + + /* Used by fts3.c only. */ + int nCost; /* Cost of running iterator */ + int bLookup; /* True if a lookup of a single entry. */ + + /* Output values. Valid only after Fts3SegReaderStep() returns SQLITE_ROW. */ + char *zTerm; /* Pointer to term buffer */ + int nTerm; /* Size of zTerm in bytes */ + char *aDoclist; /* Pointer to doclist buffer */ + int nDoclist; /* Size of aDoclist[] in bytes */ +}; + +/* fts3.c */ +int sqlite3Fts3PutVarint(char *, sqlite3_int64); +int sqlite3Fts3GetVarint(const char *, sqlite_int64 *); +int sqlite3Fts3GetVarint32(const char *, int *); +int sqlite3Fts3VarintLen(sqlite3_uint64); +void sqlite3Fts3Dequote(char *); +void sqlite3Fts3DoclistPrev(int,char*,int,char**,sqlite3_int64*,int*,u8*); +int sqlite3Fts3EvalPhraseStats(Fts3Cursor *, Fts3Expr *, u32 *); +int sqlite3Fts3FirstFilter(sqlite3_int64, char *, int, char *); + +/* fts3_tokenizer.c */ +const char *sqlite3Fts3NextToken(const char *, int *); +int sqlite3Fts3InitHashTable(sqlite3 *, Fts3Hash *, const char *); +int sqlite3Fts3InitTokenizer(Fts3Hash *pHash, const char *, + sqlite3_tokenizer **, char ** +); +int sqlite3Fts3IsIdChar(char); + +/* fts3_snippet.c */ +void sqlite3Fts3Offsets(sqlite3_context*, Fts3Cursor*); +void sqlite3Fts3Snippet(sqlite3_context *, Fts3Cursor *, const char *, + const char *, const char *, int, int +); +void sqlite3Fts3Matchinfo(sqlite3_context *, Fts3Cursor *, const char *); + +/* fts3_expr.c */ +int sqlite3Fts3ExprParse(sqlite3_tokenizer *, + char **, int, int, int, const char *, int, Fts3Expr ** +); +void sqlite3Fts3ExprFree(Fts3Expr *); +#ifdef SQLITE_TEST +int sqlite3Fts3ExprInitTestInterface(sqlite3 *db); +int sqlite3Fts3InitTerm(sqlite3 *db); +#endif + +/* fts3_aux.c */ +int sqlite3Fts3InitAux(sqlite3 *db); + +void sqlite3Fts3EvalPhraseCleanup(Fts3Phrase *); + +int sqlite3Fts3MsrIncrStart( + Fts3Table*, Fts3MultiSegReader*, int, const char*, int); +int sqlite3Fts3MsrIncrNext( + Fts3Table *, Fts3MultiSegReader *, sqlite3_int64 *, char **, int *); +char *sqlite3Fts3EvalPhrasePoslist(Fts3Cursor *, Fts3Expr *, int iCol); +int sqlite3Fts3MsrOvfl(Fts3Cursor *, Fts3MultiSegReader *, int *); +int sqlite3Fts3MsrIncrRestart(Fts3MultiSegReader *pCsr); + +int sqlite3Fts3DeferredTokenList(Fts3DeferredToken *, char **, int *); + +#endif /* !SQLITE_CORE || SQLITE_ENABLE_FTS3 */ +#endif /* _FTSINT_H */ diff --git a/ext/fts3/fts3_aux.c b/ext/fts3/fts3_aux.c new file mode 100644 index 0000000..ada85d7 --- /dev/null +++ b/ext/fts3/fts3_aux.c @@ -0,0 +1,474 @@ +/* +** 2011 Jan 27 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +*/ +#include "fts3Int.h" +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) + +#include +#include + +typedef struct Fts3auxTable Fts3auxTable; +typedef struct Fts3auxCursor Fts3auxCursor; + +struct Fts3auxTable { + sqlite3_vtab base; /* Base class used by SQLite core */ + Fts3Table *pFts3Tab; +}; + +struct Fts3auxCursor { + sqlite3_vtab_cursor base; /* Base class used by SQLite core */ + Fts3MultiSegReader csr; /* Must be right after "base" */ + Fts3SegFilter filter; + char *zStop; + int nStop; /* Byte-length of string zStop */ + int isEof; /* True if cursor is at EOF */ + sqlite3_int64 iRowid; /* Current rowid */ + + int iCol; /* Current value of 'col' column */ + int nStat; /* Size of aStat[] array */ + struct Fts3auxColstats { + sqlite3_int64 nDoc; /* 'documents' values for current csr row */ + sqlite3_int64 nOcc; /* 'occurrences' values for current csr row */ + } *aStat; +}; + +/* +** Schema of the terms table. +*/ +#define FTS3_TERMS_SCHEMA "CREATE TABLE x(term, col, documents, occurrences)" + +/* +** This function does all the work for both the xConnect and xCreate methods. +** These tables have no persistent representation of their own, so xConnect +** and xCreate are identical operations. +*/ +static int fts3auxConnectMethod( + sqlite3 *db, /* Database connection */ + void *pUnused, /* Unused */ + int argc, /* Number of elements in argv array */ + const char * const *argv, /* xCreate/xConnect argument array */ + sqlite3_vtab **ppVtab, /* OUT: New sqlite3_vtab object */ + char **pzErr /* OUT: sqlite3_malloc'd error message */ +){ + char const *zDb; /* Name of database (e.g. "main") */ + char const *zFts3; /* Name of fts3 table */ + int nDb; /* Result of strlen(zDb) */ + int nFts3; /* Result of strlen(zFts3) */ + int nByte; /* Bytes of space to allocate here */ + int rc; /* value returned by declare_vtab() */ + Fts3auxTable *p; /* Virtual table object to return */ + + UNUSED_PARAMETER(pUnused); + + /* The user should specify a single argument - the name of an fts3 table. */ + if( argc!=4 ){ + *pzErr = sqlite3_mprintf( + "wrong number of arguments to fts4aux constructor" + ); + return SQLITE_ERROR; + } + + zDb = argv[1]; + nDb = strlen(zDb); + zFts3 = argv[3]; + nFts3 = strlen(zFts3); + + rc = sqlite3_declare_vtab(db, FTS3_TERMS_SCHEMA); + if( rc!=SQLITE_OK ) return rc; + + nByte = sizeof(Fts3auxTable) + sizeof(Fts3Table) + nDb + nFts3 + 2; + p = (Fts3auxTable *)sqlite3_malloc(nByte); + if( !p ) return SQLITE_NOMEM; + memset(p, 0, nByte); + + p->pFts3Tab = (Fts3Table *)&p[1]; + p->pFts3Tab->zDb = (char *)&p->pFts3Tab[1]; + p->pFts3Tab->zName = &p->pFts3Tab->zDb[nDb+1]; + p->pFts3Tab->db = db; + p->pFts3Tab->nIndex = 1; + + memcpy((char *)p->pFts3Tab->zDb, zDb, nDb); + memcpy((char *)p->pFts3Tab->zName, zFts3, nFts3); + sqlite3Fts3Dequote((char *)p->pFts3Tab->zName); + + *ppVtab = (sqlite3_vtab *)p; + return SQLITE_OK; +} + +/* +** This function does the work for both the xDisconnect and xDestroy methods. +** These tables have no persistent representation of their own, so xDisconnect +** and xDestroy are identical operations. +*/ +static int fts3auxDisconnectMethod(sqlite3_vtab *pVtab){ + Fts3auxTable *p = (Fts3auxTable *)pVtab; + Fts3Table *pFts3 = p->pFts3Tab; + int i; + + /* Free any prepared statements held */ + for(i=0; iaStmt); i++){ + sqlite3_finalize(pFts3->aStmt[i]); + } + sqlite3_free(pFts3->zSegmentsTbl); + sqlite3_free(p); + return SQLITE_OK; +} + +#define FTS4AUX_EQ_CONSTRAINT 1 +#define FTS4AUX_GE_CONSTRAINT 2 +#define FTS4AUX_LE_CONSTRAINT 4 + +/* +** xBestIndex - Analyze a WHERE and ORDER BY clause. +*/ +static int fts3auxBestIndexMethod( + sqlite3_vtab *pVTab, + sqlite3_index_info *pInfo +){ + int i; + int iEq = -1; + int iGe = -1; + int iLe = -1; + + UNUSED_PARAMETER(pVTab); + + /* This vtab delivers always results in "ORDER BY term ASC" order. */ + if( pInfo->nOrderBy==1 + && pInfo->aOrderBy[0].iColumn==0 + && pInfo->aOrderBy[0].desc==0 + ){ + pInfo->orderByConsumed = 1; + } + + /* Search for equality and range constraints on the "term" column. */ + for(i=0; inConstraint; i++){ + if( pInfo->aConstraint[i].usable && pInfo->aConstraint[i].iColumn==0 ){ + int op = pInfo->aConstraint[i].op; + if( op==SQLITE_INDEX_CONSTRAINT_EQ ) iEq = i; + if( op==SQLITE_INDEX_CONSTRAINT_LT ) iLe = i; + if( op==SQLITE_INDEX_CONSTRAINT_LE ) iLe = i; + if( op==SQLITE_INDEX_CONSTRAINT_GT ) iGe = i; + if( op==SQLITE_INDEX_CONSTRAINT_GE ) iGe = i; + } + } + + if( iEq>=0 ){ + pInfo->idxNum = FTS4AUX_EQ_CONSTRAINT; + pInfo->aConstraintUsage[iEq].argvIndex = 1; + pInfo->estimatedCost = 5; + }else{ + pInfo->idxNum = 0; + pInfo->estimatedCost = 20000; + if( iGe>=0 ){ + pInfo->idxNum += FTS4AUX_GE_CONSTRAINT; + pInfo->aConstraintUsage[iGe].argvIndex = 1; + pInfo->estimatedCost /= 2; + } + if( iLe>=0 ){ + pInfo->idxNum += FTS4AUX_LE_CONSTRAINT; + pInfo->aConstraintUsage[iLe].argvIndex = 1 + (iGe>=0); + pInfo->estimatedCost /= 2; + } + } + + return SQLITE_OK; +} + +/* +** xOpen - Open a cursor. +*/ +static int fts3auxOpenMethod(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCsr){ + Fts3auxCursor *pCsr; /* Pointer to cursor object to return */ + + UNUSED_PARAMETER(pVTab); + + pCsr = (Fts3auxCursor *)sqlite3_malloc(sizeof(Fts3auxCursor)); + if( !pCsr ) return SQLITE_NOMEM; + memset(pCsr, 0, sizeof(Fts3auxCursor)); + + *ppCsr = (sqlite3_vtab_cursor *)pCsr; + return SQLITE_OK; +} + +/* +** xClose - Close a cursor. +*/ +static int fts3auxCloseMethod(sqlite3_vtab_cursor *pCursor){ + Fts3Table *pFts3 = ((Fts3auxTable *)pCursor->pVtab)->pFts3Tab; + Fts3auxCursor *pCsr = (Fts3auxCursor *)pCursor; + + sqlite3Fts3SegmentsClose(pFts3); + sqlite3Fts3SegReaderFinish(&pCsr->csr); + sqlite3_free((void *)pCsr->filter.zTerm); + sqlite3_free(pCsr->zStop); + sqlite3_free(pCsr->aStat); + sqlite3_free(pCsr); + return SQLITE_OK; +} + +static int fts3auxGrowStatArray(Fts3auxCursor *pCsr, int nSize){ + if( nSize>pCsr->nStat ){ + struct Fts3auxColstats *aNew; + aNew = (struct Fts3auxColstats *)sqlite3_realloc(pCsr->aStat, + sizeof(struct Fts3auxColstats) * nSize + ); + if( aNew==0 ) return SQLITE_NOMEM; + memset(&aNew[pCsr->nStat], 0, + sizeof(struct Fts3auxColstats) * (nSize - pCsr->nStat) + ); + pCsr->aStat = aNew; + pCsr->nStat = nSize; + } + return SQLITE_OK; +} + +/* +** xNext - Advance the cursor to the next row, if any. +*/ +static int fts3auxNextMethod(sqlite3_vtab_cursor *pCursor){ + Fts3auxCursor *pCsr = (Fts3auxCursor *)pCursor; + Fts3Table *pFts3 = ((Fts3auxTable *)pCursor->pVtab)->pFts3Tab; + int rc; + + /* Increment our pretend rowid value. */ + pCsr->iRowid++; + + for(pCsr->iCol++; pCsr->iColnStat; pCsr->iCol++){ + if( pCsr->aStat[pCsr->iCol].nDoc>0 ) return SQLITE_OK; + } + + rc = sqlite3Fts3SegReaderStep(pFts3, &pCsr->csr); + if( rc==SQLITE_ROW ){ + int i = 0; + int nDoclist = pCsr->csr.nDoclist; + char *aDoclist = pCsr->csr.aDoclist; + int iCol; + + int eState = 0; + + if( pCsr->zStop ){ + int n = (pCsr->nStopcsr.nTerm) ? pCsr->nStop : pCsr->csr.nTerm; + int mc = memcmp(pCsr->zStop, pCsr->csr.zTerm, n); + if( mc<0 || (mc==0 && pCsr->csr.nTerm>pCsr->nStop) ){ + pCsr->isEof = 1; + return SQLITE_OK; + } + } + + if( fts3auxGrowStatArray(pCsr, 2) ) return SQLITE_NOMEM; + memset(pCsr->aStat, 0, sizeof(struct Fts3auxColstats) * pCsr->nStat); + iCol = 0; + + while( iaStat[0].nDoc++; + eState = 1; + iCol = 0; + break; + + /* State 1. In this state we are expecting either a 1, indicating + ** that the following integer will be a column number, or the + ** start of a position list for column 0. + ** + ** The only difference between state 1 and state 2 is that if the + ** integer encountered in state 1 is not 0 or 1, then we need to + ** increment the column 0 "nDoc" count for this term. + */ + case 1: + assert( iCol==0 ); + if( v>1 ){ + pCsr->aStat[1].nDoc++; + } + eState = 2; + /* fall through */ + + case 2: + if( v==0 ){ /* 0x00. Next integer will be a docid. */ + eState = 0; + }else if( v==1 ){ /* 0x01. Next integer will be a column number. */ + eState = 3; + }else{ /* 2 or greater. A position. */ + pCsr->aStat[iCol+1].nOcc++; + pCsr->aStat[0].nOcc++; + } + break; + + /* State 3. The integer just read is a column number. */ + default: assert( eState==3 ); + iCol = (int)v; + if( fts3auxGrowStatArray(pCsr, iCol+2) ) return SQLITE_NOMEM; + pCsr->aStat[iCol+1].nDoc++; + eState = 2; + break; + } + } + + pCsr->iCol = 0; + rc = SQLITE_OK; + }else{ + pCsr->isEof = 1; + } + return rc; +} + +/* +** xFilter - Initialize a cursor to point at the start of its data. +*/ +static int fts3auxFilterMethod( + sqlite3_vtab_cursor *pCursor, /* The cursor used for this query */ + int idxNum, /* Strategy index */ + const char *idxStr, /* Unused */ + int nVal, /* Number of elements in apVal */ + sqlite3_value **apVal /* Arguments for the indexing scheme */ +){ + Fts3auxCursor *pCsr = (Fts3auxCursor *)pCursor; + Fts3Table *pFts3 = ((Fts3auxTable *)pCursor->pVtab)->pFts3Tab; + int rc; + int isScan; + + UNUSED_PARAMETER(nVal); + UNUSED_PARAMETER(idxStr); + + assert( idxStr==0 ); + assert( idxNum==FTS4AUX_EQ_CONSTRAINT || idxNum==0 + || idxNum==FTS4AUX_LE_CONSTRAINT || idxNum==FTS4AUX_GE_CONSTRAINT + || idxNum==(FTS4AUX_LE_CONSTRAINT|FTS4AUX_GE_CONSTRAINT) + ); + isScan = (idxNum!=FTS4AUX_EQ_CONSTRAINT); + + /* In case this cursor is being reused, close and zero it. */ + testcase(pCsr->filter.zTerm); + sqlite3Fts3SegReaderFinish(&pCsr->csr); + sqlite3_free((void *)pCsr->filter.zTerm); + sqlite3_free(pCsr->aStat); + memset(&pCsr->csr, 0, ((u8*)&pCsr[1]) - (u8*)&pCsr->csr); + + pCsr->filter.flags = FTS3_SEGMENT_REQUIRE_POS|FTS3_SEGMENT_IGNORE_EMPTY; + if( isScan ) pCsr->filter.flags |= FTS3_SEGMENT_SCAN; + + if( idxNum&(FTS4AUX_EQ_CONSTRAINT|FTS4AUX_GE_CONSTRAINT) ){ + const unsigned char *zStr = sqlite3_value_text(apVal[0]); + if( zStr ){ + pCsr->filter.zTerm = sqlite3_mprintf("%s", zStr); + pCsr->filter.nTerm = sqlite3_value_bytes(apVal[0]); + if( pCsr->filter.zTerm==0 ) return SQLITE_NOMEM; + } + } + if( idxNum&FTS4AUX_LE_CONSTRAINT ){ + int iIdx = (idxNum&FTS4AUX_GE_CONSTRAINT) ? 1 : 0; + pCsr->zStop = sqlite3_mprintf("%s", sqlite3_value_text(apVal[iIdx])); + pCsr->nStop = sqlite3_value_bytes(apVal[iIdx]); + if( pCsr->zStop==0 ) return SQLITE_NOMEM; + } + + rc = sqlite3Fts3SegReaderCursor(pFts3, 0, FTS3_SEGCURSOR_ALL, + pCsr->filter.zTerm, pCsr->filter.nTerm, 0, isScan, &pCsr->csr + ); + if( rc==SQLITE_OK ){ + rc = sqlite3Fts3SegReaderStart(pFts3, &pCsr->csr, &pCsr->filter); + } + + if( rc==SQLITE_OK ) rc = fts3auxNextMethod(pCursor); + return rc; +} + +/* +** xEof - Return true if the cursor is at EOF, or false otherwise. +*/ +static int fts3auxEofMethod(sqlite3_vtab_cursor *pCursor){ + Fts3auxCursor *pCsr = (Fts3auxCursor *)pCursor; + return pCsr->isEof; +} + +/* +** xColumn - Return a column value. +*/ +static int fts3auxColumnMethod( + sqlite3_vtab_cursor *pCursor, /* Cursor to retrieve value from */ + sqlite3_context *pContext, /* Context for sqlite3_result_xxx() calls */ + int iCol /* Index of column to read value from */ +){ + Fts3auxCursor *p = (Fts3auxCursor *)pCursor; + + assert( p->isEof==0 ); + if( iCol==0 ){ /* Column "term" */ + sqlite3_result_text(pContext, p->csr.zTerm, p->csr.nTerm, SQLITE_TRANSIENT); + }else if( iCol==1 ){ /* Column "col" */ + if( p->iCol ){ + sqlite3_result_int(pContext, p->iCol-1); + }else{ + sqlite3_result_text(pContext, "*", -1, SQLITE_STATIC); + } + }else if( iCol==2 ){ /* Column "documents" */ + sqlite3_result_int64(pContext, p->aStat[p->iCol].nDoc); + }else{ /* Column "occurrences" */ + sqlite3_result_int64(pContext, p->aStat[p->iCol].nOcc); + } + + return SQLITE_OK; +} + +/* +** xRowid - Return the current rowid for the cursor. +*/ +static int fts3auxRowidMethod( + sqlite3_vtab_cursor *pCursor, /* Cursor to retrieve value from */ + sqlite_int64 *pRowid /* OUT: Rowid value */ +){ + Fts3auxCursor *pCsr = (Fts3auxCursor *)pCursor; + *pRowid = pCsr->iRowid; + return SQLITE_OK; +} + +/* +** Register the fts3aux module with database connection db. Return SQLITE_OK +** if successful or an error code if sqlite3_create_module() fails. +*/ +int sqlite3Fts3InitAux(sqlite3 *db){ + static const sqlite3_module fts3aux_module = { + 0, /* iVersion */ + fts3auxConnectMethod, /* xCreate */ + fts3auxConnectMethod, /* xConnect */ + fts3auxBestIndexMethod, /* xBestIndex */ + fts3auxDisconnectMethod, /* xDisconnect */ + fts3auxDisconnectMethod, /* xDestroy */ + fts3auxOpenMethod, /* xOpen */ + fts3auxCloseMethod, /* xClose */ + fts3auxFilterMethod, /* xFilter */ + fts3auxNextMethod, /* xNext */ + fts3auxEofMethod, /* xEof */ + fts3auxColumnMethod, /* xColumn */ + fts3auxRowidMethod, /* xRowid */ + 0, /* xUpdate */ + 0, /* xBegin */ + 0, /* xSync */ + 0, /* xCommit */ + 0, /* xRollback */ + 0, /* xFindFunction */ + 0, /* xRename */ + 0, /* xSavepoint */ + 0, /* xRelease */ + 0 /* xRollbackTo */ + }; + int rc; /* Return code */ + + rc = sqlite3_create_module(db, "fts4aux", &fts3aux_module, 0); + return rc; +} + +#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) */ diff --git a/ext/fts3/fts3_expr.c b/ext/fts3/fts3_expr.c new file mode 100644 index 0000000..1c3a790 --- /dev/null +++ b/ext/fts3/fts3_expr.c @@ -0,0 +1,984 @@ +/* +** 2008 Nov 28 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This module contains code that implements a parser for fts3 query strings +** (the right-hand argument to the MATCH operator). Because the supported +** syntax is relatively simple, the whole tokenizer/parser system is +** hand-coded. +*/ +#include "fts3Int.h" +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) + +/* +** By default, this module parses the legacy syntax that has been +** traditionally used by fts3. Or, if SQLITE_ENABLE_FTS3_PARENTHESIS +** is defined, then it uses the new syntax. The differences between +** the new and the old syntaxes are: +** +** a) The new syntax supports parenthesis. The old does not. +** +** b) The new syntax supports the AND and NOT operators. The old does not. +** +** c) The old syntax supports the "-" token qualifier. This is not +** supported by the new syntax (it is replaced by the NOT operator). +** +** d) When using the old syntax, the OR operator has a greater precedence +** than an implicit AND. When using the new, both implicity and explicit +** AND operators have a higher precedence than OR. +** +** If compiled with SQLITE_TEST defined, then this module exports the +** symbol "int sqlite3_fts3_enable_parentheses". Setting this variable +** to zero causes the module to use the old syntax. If it is set to +** non-zero the new syntax is activated. This is so both syntaxes can +** be tested using a single build of testfixture. +** +** The following describes the syntax supported by the fts3 MATCH +** operator in a similar format to that used by the lemon parser +** generator. This module does not use actually lemon, it uses a +** custom parser. +** +** query ::= andexpr (OR andexpr)*. +** +** andexpr ::= notexpr (AND? notexpr)*. +** +** notexpr ::= nearexpr (NOT nearexpr|-TOKEN)*. +** notexpr ::= LP query RP. +** +** nearexpr ::= phrase (NEAR distance_opt nearexpr)*. +** +** distance_opt ::= . +** distance_opt ::= / INTEGER. +** +** phrase ::= TOKEN. +** phrase ::= COLUMN:TOKEN. +** phrase ::= "TOKEN TOKEN TOKEN...". +*/ + +#ifdef SQLITE_TEST +int sqlite3_fts3_enable_parentheses = 0; +#else +# ifdef SQLITE_ENABLE_FTS3_PARENTHESIS +# define sqlite3_fts3_enable_parentheses 1 +# else +# define sqlite3_fts3_enable_parentheses 0 +# endif +#endif + +/* +** Default span for NEAR operators. +*/ +#define SQLITE_FTS3_DEFAULT_NEAR_PARAM 10 + +#include +#include + +/* +** isNot: +** This variable is used by function getNextNode(). When getNextNode() is +** called, it sets ParseContext.isNot to true if the 'next node' is a +** FTSQUERY_PHRASE with a unary "-" attached to it. i.e. "mysql" in the +** FTS3 query "sqlite -mysql". Otherwise, ParseContext.isNot is set to +** zero. +*/ +typedef struct ParseContext ParseContext; +struct ParseContext { + sqlite3_tokenizer *pTokenizer; /* Tokenizer module */ + const char **azCol; /* Array of column names for fts3 table */ + int bFts4; /* True to allow FTS4-only syntax */ + int nCol; /* Number of entries in azCol[] */ + int iDefaultCol; /* Default column to query */ + int isNot; /* True if getNextNode() sees a unary - */ + sqlite3_context *pCtx; /* Write error message here */ + int nNest; /* Number of nested brackets */ +}; + +/* +** This function is equivalent to the standard isspace() function. +** +** The standard isspace() can be awkward to use safely, because although it +** is defined to accept an argument of type int, its behaviour when passed +** an integer that falls outside of the range of the unsigned char type +** is undefined (and sometimes, "undefined" means segfault). This wrapper +** is defined to accept an argument of type char, and always returns 0 for +** any values that fall outside of the range of the unsigned char type (i.e. +** negative values). +*/ +static int fts3isspace(char c){ + return c==' ' || c=='\t' || c=='\n' || c=='\r' || c=='\v' || c=='\f'; +} + +/* +** Allocate nByte bytes of memory using sqlite3_malloc(). If successful, +** zero the memory before returning a pointer to it. If unsuccessful, +** return NULL. +*/ +static void *fts3MallocZero(int nByte){ + void *pRet = sqlite3_malloc(nByte); + if( pRet ) memset(pRet, 0, nByte); + return pRet; +} + + +/* +** Extract the next token from buffer z (length n) using the tokenizer +** and other information (column names etc.) in pParse. Create an Fts3Expr +** structure of type FTSQUERY_PHRASE containing a phrase consisting of this +** single token and set *ppExpr to point to it. If the end of the buffer is +** reached before a token is found, set *ppExpr to zero. It is the +** responsibility of the caller to eventually deallocate the allocated +** Fts3Expr structure (if any) by passing it to sqlite3_free(). +** +** Return SQLITE_OK if successful, or SQLITE_NOMEM if a memory allocation +** fails. +*/ +static int getNextToken( + ParseContext *pParse, /* fts3 query parse context */ + int iCol, /* Value for Fts3Phrase.iColumn */ + const char *z, int n, /* Input string */ + Fts3Expr **ppExpr, /* OUT: expression */ + int *pnConsumed /* OUT: Number of bytes consumed */ +){ + sqlite3_tokenizer *pTokenizer = pParse->pTokenizer; + sqlite3_tokenizer_module const *pModule = pTokenizer->pModule; + int rc; + sqlite3_tokenizer_cursor *pCursor; + Fts3Expr *pRet = 0; + int nConsumed = 0; + + rc = pModule->xOpen(pTokenizer, z, n, &pCursor); + if( rc==SQLITE_OK ){ + const char *zToken; + int nToken, iStart, iEnd, iPosition; + int nByte; /* total space to allocate */ + + pCursor->pTokenizer = pTokenizer; + rc = pModule->xNext(pCursor, &zToken, &nToken, &iStart, &iEnd, &iPosition); + + if( rc==SQLITE_OK ){ + nByte = sizeof(Fts3Expr) + sizeof(Fts3Phrase) + nToken; + pRet = (Fts3Expr *)fts3MallocZero(nByte); + if( !pRet ){ + rc = SQLITE_NOMEM; + }else{ + pRet->eType = FTSQUERY_PHRASE; + pRet->pPhrase = (Fts3Phrase *)&pRet[1]; + pRet->pPhrase->nToken = 1; + pRet->pPhrase->iColumn = iCol; + pRet->pPhrase->aToken[0].n = nToken; + pRet->pPhrase->aToken[0].z = (char *)&pRet->pPhrase[1]; + memcpy(pRet->pPhrase->aToken[0].z, zToken, nToken); + + if( iEndpPhrase->aToken[0].isPrefix = 1; + iEnd++; + } + + while( 1 ){ + if( !sqlite3_fts3_enable_parentheses + && iStart>0 && z[iStart-1]=='-' + ){ + pParse->isNot = 1; + iStart--; + }else if( pParse->bFts4 && iStart>0 && z[iStart-1]=='^' ){ + pRet->pPhrase->aToken[0].bFirst = 1; + iStart--; + }else{ + break; + } + } + + } + nConsumed = iEnd; + } + + pModule->xClose(pCursor); + } + + *pnConsumed = nConsumed; + *ppExpr = pRet; + return rc; +} + + +/* +** Enlarge a memory allocation. If an out-of-memory allocation occurs, +** then free the old allocation. +*/ +static void *fts3ReallocOrFree(void *pOrig, int nNew){ + void *pRet = sqlite3_realloc(pOrig, nNew); + if( !pRet ){ + sqlite3_free(pOrig); + } + return pRet; +} + +/* +** Buffer zInput, length nInput, contains the contents of a quoted string +** that appeared as part of an fts3 query expression. Neither quote character +** is included in the buffer. This function attempts to tokenize the entire +** input buffer and create an Fts3Expr structure of type FTSQUERY_PHRASE +** containing the results. +** +** If successful, SQLITE_OK is returned and *ppExpr set to point at the +** allocated Fts3Expr structure. Otherwise, either SQLITE_NOMEM (out of memory +** error) or SQLITE_ERROR (tokenization error) is returned and *ppExpr set +** to 0. +*/ +static int getNextString( + ParseContext *pParse, /* fts3 query parse context */ + const char *zInput, int nInput, /* Input string */ + Fts3Expr **ppExpr /* OUT: expression */ +){ + sqlite3_tokenizer *pTokenizer = pParse->pTokenizer; + sqlite3_tokenizer_module const *pModule = pTokenizer->pModule; + int rc; + Fts3Expr *p = 0; + sqlite3_tokenizer_cursor *pCursor = 0; + char *zTemp = 0; + int nTemp = 0; + + const int nSpace = sizeof(Fts3Expr) + sizeof(Fts3Phrase); + int nToken = 0; + + /* The final Fts3Expr data structure, including the Fts3Phrase, + ** Fts3PhraseToken structures token buffers are all stored as a single + ** allocation so that the expression can be freed with a single call to + ** sqlite3_free(). Setting this up requires a two pass approach. + ** + ** The first pass, in the block below, uses a tokenizer cursor to iterate + ** through the tokens in the expression. This pass uses fts3ReallocOrFree() + ** to assemble data in two dynamic buffers: + ** + ** Buffer p: Points to the Fts3Expr structure, followed by the Fts3Phrase + ** structure, followed by the array of Fts3PhraseToken + ** structures. This pass only populates the Fts3PhraseToken array. + ** + ** Buffer zTemp: Contains copies of all tokens. + ** + ** The second pass, in the block that begins "if( rc==SQLITE_DONE )" below, + ** appends buffer zTemp to buffer p, and fills in the Fts3Expr and Fts3Phrase + ** structures. + */ + rc = pModule->xOpen(pTokenizer, zInput, nInput, &pCursor); + if( rc==SQLITE_OK ){ + int ii; + pCursor->pTokenizer = pTokenizer; + for(ii=0; rc==SQLITE_OK; ii++){ + const char *zByte; + int nByte, iBegin, iEnd, iPos; + rc = pModule->xNext(pCursor, &zByte, &nByte, &iBegin, &iEnd, &iPos); + if( rc==SQLITE_OK ){ + Fts3PhraseToken *pToken; + + p = fts3ReallocOrFree(p, nSpace + ii*sizeof(Fts3PhraseToken)); + if( !p ) goto no_mem; + + zTemp = fts3ReallocOrFree(zTemp, nTemp + nByte); + if( !zTemp ) goto no_mem; + + assert( nToken==ii ); + pToken = &((Fts3Phrase *)(&p[1]))->aToken[ii]; + memset(pToken, 0, sizeof(Fts3PhraseToken)); + + memcpy(&zTemp[nTemp], zByte, nByte); + nTemp += nByte; + + pToken->n = nByte; + pToken->isPrefix = (iEndbFirst = (iBegin>0 && zInput[iBegin-1]=='^'); + nToken = ii+1; + } + } + + pModule->xClose(pCursor); + pCursor = 0; + } + + if( rc==SQLITE_DONE ){ + int jj; + char *zBuf = 0; + + p = fts3ReallocOrFree(p, nSpace + nToken*sizeof(Fts3PhraseToken) + nTemp); + if( !p ) goto no_mem; + memset(p, 0, (char *)&(((Fts3Phrase *)&p[1])->aToken[0])-(char *)p); + p->eType = FTSQUERY_PHRASE; + p->pPhrase = (Fts3Phrase *)&p[1]; + p->pPhrase->iColumn = pParse->iDefaultCol; + p->pPhrase->nToken = nToken; + + zBuf = (char *)&p->pPhrase->aToken[nToken]; + if( zTemp ){ + memcpy(zBuf, zTemp, nTemp); + sqlite3_free(zTemp); + }else{ + assert( nTemp==0 ); + } + + for(jj=0; jjpPhrase->nToken; jj++){ + p->pPhrase->aToken[jj].z = zBuf; + zBuf += p->pPhrase->aToken[jj].n; + } + rc = SQLITE_OK; + } + + *ppExpr = p; + return rc; +no_mem: + + if( pCursor ){ + pModule->xClose(pCursor); + } + sqlite3_free(zTemp); + sqlite3_free(p); + *ppExpr = 0; + return SQLITE_NOMEM; +} + +/* +** Function getNextNode(), which is called by fts3ExprParse(), may itself +** call fts3ExprParse(). So this forward declaration is required. +*/ +static int fts3ExprParse(ParseContext *, const char *, int, Fts3Expr **, int *); + +/* +** The output variable *ppExpr is populated with an allocated Fts3Expr +** structure, or set to 0 if the end of the input buffer is reached. +** +** Returns an SQLite error code. SQLITE_OK if everything works, SQLITE_NOMEM +** if a malloc failure occurs, or SQLITE_ERROR if a parse error is encountered. +** If SQLITE_ERROR is returned, pContext is populated with an error message. +*/ +static int getNextNode( + ParseContext *pParse, /* fts3 query parse context */ + const char *z, int n, /* Input string */ + Fts3Expr **ppExpr, /* OUT: expression */ + int *pnConsumed /* OUT: Number of bytes consumed */ +){ + static const struct Fts3Keyword { + char *z; /* Keyword text */ + unsigned char n; /* Length of the keyword */ + unsigned char parenOnly; /* Only valid in paren mode */ + unsigned char eType; /* Keyword code */ + } aKeyword[] = { + { "OR" , 2, 0, FTSQUERY_OR }, + { "AND", 3, 1, FTSQUERY_AND }, + { "NOT", 3, 1, FTSQUERY_NOT }, + { "NEAR", 4, 0, FTSQUERY_NEAR } + }; + int ii; + int iCol; + int iColLen; + int rc; + Fts3Expr *pRet = 0; + + const char *zInput = z; + int nInput = n; + + pParse->isNot = 0; + + /* Skip over any whitespace before checking for a keyword, an open or + ** close bracket, or a quoted string. + */ + while( nInput>0 && fts3isspace(*zInput) ){ + nInput--; + zInput++; + } + if( nInput==0 ){ + return SQLITE_DONE; + } + + /* See if we are dealing with a keyword. */ + for(ii=0; ii<(int)(sizeof(aKeyword)/sizeof(struct Fts3Keyword)); ii++){ + const struct Fts3Keyword *pKey = &aKeyword[ii]; + + if( (pKey->parenOnly & ~sqlite3_fts3_enable_parentheses)!=0 ){ + continue; + } + + if( nInput>=pKey->n && 0==memcmp(zInput, pKey->z, pKey->n) ){ + int nNear = SQLITE_FTS3_DEFAULT_NEAR_PARAM; + int nKey = pKey->n; + char cNext; + + /* If this is a "NEAR" keyword, check for an explicit nearness. */ + if( pKey->eType==FTSQUERY_NEAR ){ + assert( nKey==4 ); + if( zInput[4]=='/' && zInput[5]>='0' && zInput[5]<='9' ){ + nNear = 0; + for(nKey=5; zInput[nKey]>='0' && zInput[nKey]<='9'; nKey++){ + nNear = nNear * 10 + (zInput[nKey] - '0'); + } + } + } + + /* At this point this is probably a keyword. But for that to be true, + ** the next byte must contain either whitespace, an open or close + ** parenthesis, a quote character, or EOF. + */ + cNext = zInput[nKey]; + if( fts3isspace(cNext) + || cNext=='"' || cNext=='(' || cNext==')' || cNext==0 + ){ + pRet = (Fts3Expr *)fts3MallocZero(sizeof(Fts3Expr)); + if( !pRet ){ + return SQLITE_NOMEM; + } + pRet->eType = pKey->eType; + pRet->nNear = nNear; + *ppExpr = pRet; + *pnConsumed = (int)((zInput - z) + nKey); + return SQLITE_OK; + } + + /* Turns out that wasn't a keyword after all. This happens if the + ** user has supplied a token such as "ORacle". Continue. + */ + } + } + + /* Check for an open bracket. */ + if( sqlite3_fts3_enable_parentheses ){ + if( *zInput=='(' ){ + int nConsumed; + pParse->nNest++; + rc = fts3ExprParse(pParse, &zInput[1], nInput-1, ppExpr, &nConsumed); + if( rc==SQLITE_OK && !*ppExpr ){ + rc = SQLITE_DONE; + } + *pnConsumed = (int)((zInput - z) + 1 + nConsumed); + return rc; + } + + /* Check for a close bracket. */ + if( *zInput==')' ){ + pParse->nNest--; + *pnConsumed = (int)((zInput - z) + 1); + return SQLITE_DONE; + } + } + + /* See if we are dealing with a quoted phrase. If this is the case, then + ** search for the closing quote and pass the whole string to getNextString() + ** for processing. This is easy to do, as fts3 has no syntax for escaping + ** a quote character embedded in a string. + */ + if( *zInput=='"' ){ + for(ii=1; iiiDefaultCol; + iColLen = 0; + for(ii=0; iinCol; ii++){ + const char *zStr = pParse->azCol[ii]; + int nStr = (int)strlen(zStr); + if( nInput>nStr && zInput[nStr]==':' + && sqlite3_strnicmp(zStr, zInput, nStr)==0 + ){ + iCol = ii; + iColLen = (int)((zInput - z) + nStr + 1); + break; + } + } + rc = getNextToken(pParse, iCol, &z[iColLen], n-iColLen, ppExpr, pnConsumed); + *pnConsumed += iColLen; + return rc; +} + +/* +** The argument is an Fts3Expr structure for a binary operator (any type +** except an FTSQUERY_PHRASE). Return an integer value representing the +** precedence of the operator. Lower values have a higher precedence (i.e. +** group more tightly). For example, in the C language, the == operator +** groups more tightly than ||, and would therefore have a higher precedence. +** +** When using the new fts3 query syntax (when SQLITE_ENABLE_FTS3_PARENTHESIS +** is defined), the order of the operators in precedence from highest to +** lowest is: +** +** NEAR +** NOT +** AND (including implicit ANDs) +** OR +** +** Note that when using the old query syntax, the OR operator has a higher +** precedence than the AND operator. +*/ +static int opPrecedence(Fts3Expr *p){ + assert( p->eType!=FTSQUERY_PHRASE ); + if( sqlite3_fts3_enable_parentheses ){ + return p->eType; + }else if( p->eType==FTSQUERY_NEAR ){ + return 1; + }else if( p->eType==FTSQUERY_OR ){ + return 2; + } + assert( p->eType==FTSQUERY_AND ); + return 3; +} + +/* +** Argument ppHead contains a pointer to the current head of a query +** expression tree being parsed. pPrev is the expression node most recently +** inserted into the tree. This function adds pNew, which is always a binary +** operator node, into the expression tree based on the relative precedence +** of pNew and the existing nodes of the tree. This may result in the head +** of the tree changing, in which case *ppHead is set to the new root node. +*/ +static void insertBinaryOperator( + Fts3Expr **ppHead, /* Pointer to the root node of a tree */ + Fts3Expr *pPrev, /* Node most recently inserted into the tree */ + Fts3Expr *pNew /* New binary node to insert into expression tree */ +){ + Fts3Expr *pSplit = pPrev; + while( pSplit->pParent && opPrecedence(pSplit->pParent)<=opPrecedence(pNew) ){ + pSplit = pSplit->pParent; + } + + if( pSplit->pParent ){ + assert( pSplit->pParent->pRight==pSplit ); + pSplit->pParent->pRight = pNew; + pNew->pParent = pSplit->pParent; + }else{ + *ppHead = pNew; + } + pNew->pLeft = pSplit; + pSplit->pParent = pNew; +} + +/* +** Parse the fts3 query expression found in buffer z, length n. This function +** returns either when the end of the buffer is reached or an unmatched +** closing bracket - ')' - is encountered. +** +** If successful, SQLITE_OK is returned, *ppExpr is set to point to the +** parsed form of the expression and *pnConsumed is set to the number of +** bytes read from buffer z. Otherwise, *ppExpr is set to 0 and SQLITE_NOMEM +** (out of memory error) or SQLITE_ERROR (parse error) is returned. +*/ +static int fts3ExprParse( + ParseContext *pParse, /* fts3 query parse context */ + const char *z, int n, /* Text of MATCH query */ + Fts3Expr **ppExpr, /* OUT: Parsed query structure */ + int *pnConsumed /* OUT: Number of bytes consumed */ +){ + Fts3Expr *pRet = 0; + Fts3Expr *pPrev = 0; + Fts3Expr *pNotBranch = 0; /* Only used in legacy parse mode */ + int nIn = n; + const char *zIn = z; + int rc = SQLITE_OK; + int isRequirePhrase = 1; + + while( rc==SQLITE_OK ){ + Fts3Expr *p = 0; + int nByte = 0; + rc = getNextNode(pParse, zIn, nIn, &p, &nByte); + if( rc==SQLITE_OK ){ + int isPhrase; + + if( !sqlite3_fts3_enable_parentheses + && p->eType==FTSQUERY_PHRASE && pParse->isNot + ){ + /* Create an implicit NOT operator. */ + Fts3Expr *pNot = fts3MallocZero(sizeof(Fts3Expr)); + if( !pNot ){ + sqlite3Fts3ExprFree(p); + rc = SQLITE_NOMEM; + goto exprparse_out; + } + pNot->eType = FTSQUERY_NOT; + pNot->pRight = p; + if( pNotBranch ){ + pNot->pLeft = pNotBranch; + } + pNotBranch = pNot; + p = pPrev; + }else{ + int eType = p->eType; + isPhrase = (eType==FTSQUERY_PHRASE || p->pLeft); + + /* The isRequirePhrase variable is set to true if a phrase or + ** an expression contained in parenthesis is required. If a + ** binary operator (AND, OR, NOT or NEAR) is encounted when + ** isRequirePhrase is set, this is a syntax error. + */ + if( !isPhrase && isRequirePhrase ){ + sqlite3Fts3ExprFree(p); + rc = SQLITE_ERROR; + goto exprparse_out; + } + + if( isPhrase && !isRequirePhrase ){ + /* Insert an implicit AND operator. */ + Fts3Expr *pAnd; + assert( pRet && pPrev ); + pAnd = fts3MallocZero(sizeof(Fts3Expr)); + if( !pAnd ){ + sqlite3Fts3ExprFree(p); + rc = SQLITE_NOMEM; + goto exprparse_out; + } + pAnd->eType = FTSQUERY_AND; + insertBinaryOperator(&pRet, pPrev, pAnd); + pPrev = pAnd; + } + + /* This test catches attempts to make either operand of a NEAR + ** operator something other than a phrase. For example, either of + ** the following: + ** + ** (bracketed expression) NEAR phrase + ** phrase NEAR (bracketed expression) + ** + ** Return an error in either case. + */ + if( pPrev && ( + (eType==FTSQUERY_NEAR && !isPhrase && pPrev->eType!=FTSQUERY_PHRASE) + || (eType!=FTSQUERY_PHRASE && isPhrase && pPrev->eType==FTSQUERY_NEAR) + )){ + sqlite3Fts3ExprFree(p); + rc = SQLITE_ERROR; + goto exprparse_out; + } + + if( isPhrase ){ + if( pRet ){ + assert( pPrev && pPrev->pLeft && pPrev->pRight==0 ); + pPrev->pRight = p; + p->pParent = pPrev; + }else{ + pRet = p; + } + }else{ + insertBinaryOperator(&pRet, pPrev, p); + } + isRequirePhrase = !isPhrase; + } + assert( nByte>0 ); + } + assert( rc!=SQLITE_OK || (nByte>0 && nByte<=nIn) ); + nIn -= nByte; + zIn += nByte; + pPrev = p; + } + + if( rc==SQLITE_DONE && pRet && isRequirePhrase ){ + rc = SQLITE_ERROR; + } + + if( rc==SQLITE_DONE ){ + rc = SQLITE_OK; + if( !sqlite3_fts3_enable_parentheses && pNotBranch ){ + if( !pRet ){ + rc = SQLITE_ERROR; + }else{ + Fts3Expr *pIter = pNotBranch; + while( pIter->pLeft ){ + pIter = pIter->pLeft; + } + pIter->pLeft = pRet; + pRet = pNotBranch; + } + } + } + *pnConsumed = n - nIn; + +exprparse_out: + if( rc!=SQLITE_OK ){ + sqlite3Fts3ExprFree(pRet); + sqlite3Fts3ExprFree(pNotBranch); + pRet = 0; + } + *ppExpr = pRet; + return rc; +} + +/* +** Parameters z and n contain a pointer to and length of a buffer containing +** an fts3 query expression, respectively. This function attempts to parse the +** query expression and create a tree of Fts3Expr structures representing the +** parsed expression. If successful, *ppExpr is set to point to the head +** of the parsed expression tree and SQLITE_OK is returned. If an error +** occurs, either SQLITE_NOMEM (out-of-memory error) or SQLITE_ERROR (parse +** error) is returned and *ppExpr is set to 0. +** +** If parameter n is a negative number, then z is assumed to point to a +** nul-terminated string and the length is determined using strlen(). +** +** The first parameter, pTokenizer, is passed the fts3 tokenizer module to +** use to normalize query tokens while parsing the expression. The azCol[] +** array, which is assumed to contain nCol entries, should contain the names +** of each column in the target fts3 table, in order from left to right. +** Column names must be nul-terminated strings. +** +** The iDefaultCol parameter should be passed the index of the table column +** that appears on the left-hand-side of the MATCH operator (the default +** column to match against for tokens for which a column name is not explicitly +** specified as part of the query string), or -1 if tokens may by default +** match any table column. +*/ +int sqlite3Fts3ExprParse( + sqlite3_tokenizer *pTokenizer, /* Tokenizer module */ + char **azCol, /* Array of column names for fts3 table */ + int bFts4, /* True to allow FTS4-only syntax */ + int nCol, /* Number of entries in azCol[] */ + int iDefaultCol, /* Default column to query */ + const char *z, int n, /* Text of MATCH query */ + Fts3Expr **ppExpr /* OUT: Parsed query structure */ +){ + int nParsed; + int rc; + ParseContext sParse; + sParse.pTokenizer = pTokenizer; + sParse.azCol = (const char **)azCol; + sParse.nCol = nCol; + sParse.iDefaultCol = iDefaultCol; + sParse.nNest = 0; + sParse.bFts4 = bFts4; + if( z==0 ){ + *ppExpr = 0; + return SQLITE_OK; + } + if( n<0 ){ + n = (int)strlen(z); + } + rc = fts3ExprParse(&sParse, z, n, ppExpr, &nParsed); + + /* Check for mismatched parenthesis */ + if( rc==SQLITE_OK && sParse.nNest ){ + rc = SQLITE_ERROR; + sqlite3Fts3ExprFree(*ppExpr); + *ppExpr = 0; + } + + return rc; +} + +/* +** Free a parsed fts3 query expression allocated by sqlite3Fts3ExprParse(). +*/ +void sqlite3Fts3ExprFree(Fts3Expr *p){ + if( p ){ + assert( p->eType==FTSQUERY_PHRASE || p->pPhrase==0 ); + sqlite3Fts3ExprFree(p->pLeft); + sqlite3Fts3ExprFree(p->pRight); + sqlite3Fts3EvalPhraseCleanup(p->pPhrase); + sqlite3_free(p->aMI); + sqlite3_free(p); + } +} + +/**************************************************************************** +***************************************************************************** +** Everything after this point is just test code. +*/ + +#ifdef SQLITE_TEST + +#include + +/* +** Function to query the hash-table of tokenizers (see README.tokenizers). +*/ +static int queryTestTokenizer( + sqlite3 *db, + const char *zName, + const sqlite3_tokenizer_module **pp +){ + int rc; + sqlite3_stmt *pStmt; + const char zSql[] = "SELECT fts3_tokenizer(?)"; + + *pp = 0; + rc = sqlite3_prepare_v2(db, zSql, -1, &pStmt, 0); + if( rc!=SQLITE_OK ){ + return rc; + } + + sqlite3_bind_text(pStmt, 1, zName, -1, SQLITE_STATIC); + if( SQLITE_ROW==sqlite3_step(pStmt) ){ + if( sqlite3_column_type(pStmt, 0)==SQLITE_BLOB ){ + memcpy((void *)pp, sqlite3_column_blob(pStmt, 0), sizeof(*pp)); + } + } + + return sqlite3_finalize(pStmt); +} + +/* +** Return a pointer to a buffer containing a text representation of the +** expression passed as the first argument. The buffer is obtained from +** sqlite3_malloc(). It is the responsibility of the caller to use +** sqlite3_free() to release the memory. If an OOM condition is encountered, +** NULL is returned. +** +** If the second argument is not NULL, then its contents are prepended to +** the returned expression text and then freed using sqlite3_free(). +*/ +static char *exprToString(Fts3Expr *pExpr, char *zBuf){ + switch( pExpr->eType ){ + case FTSQUERY_PHRASE: { + Fts3Phrase *pPhrase = pExpr->pPhrase; + int i; + zBuf = sqlite3_mprintf( + "%zPHRASE %d 0", zBuf, pPhrase->iColumn); + for(i=0; zBuf && inToken; i++){ + zBuf = sqlite3_mprintf("%z %.*s%s", zBuf, + pPhrase->aToken[i].n, pPhrase->aToken[i].z, + (pPhrase->aToken[i].isPrefix?"+":"") + ); + } + return zBuf; + } + + case FTSQUERY_NEAR: + zBuf = sqlite3_mprintf("%zNEAR/%d ", zBuf, pExpr->nNear); + break; + case FTSQUERY_NOT: + zBuf = sqlite3_mprintf("%zNOT ", zBuf); + break; + case FTSQUERY_AND: + zBuf = sqlite3_mprintf("%zAND ", zBuf); + break; + case FTSQUERY_OR: + zBuf = sqlite3_mprintf("%zOR ", zBuf); + break; + } + + if( zBuf ) zBuf = sqlite3_mprintf("%z{", zBuf); + if( zBuf ) zBuf = exprToString(pExpr->pLeft, zBuf); + if( zBuf ) zBuf = sqlite3_mprintf("%z} {", zBuf); + + if( zBuf ) zBuf = exprToString(pExpr->pRight, zBuf); + if( zBuf ) zBuf = sqlite3_mprintf("%z}", zBuf); + + return zBuf; +} + +/* +** This is the implementation of a scalar SQL function used to test the +** expression parser. It should be called as follows: +** +** fts3_exprtest(, , , ...); +** +** The first argument, , is the name of the fts3 tokenizer used +** to parse the query expression (see README.tokenizers). The second argument +** is the query expression to parse. Each subsequent argument is the name +** of a column of the fts3 table that the query expression may refer to. +** For example: +** +** SELECT fts3_exprtest('simple', 'Bill col2:Bloggs', 'col1', 'col2'); +*/ +static void fts3ExprTest( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + sqlite3_tokenizer_module const *pModule = 0; + sqlite3_tokenizer *pTokenizer = 0; + int rc; + char **azCol = 0; + const char *zExpr; + int nExpr; + int nCol; + int ii; + Fts3Expr *pExpr; + char *zBuf = 0; + sqlite3 *db = sqlite3_context_db_handle(context); + + if( argc<3 ){ + sqlite3_result_error(context, + "Usage: fts3_exprtest(tokenizer, expr, col1, ...", -1 + ); + return; + } + + rc = queryTestTokenizer(db, + (const char *)sqlite3_value_text(argv[0]), &pModule); + if( rc==SQLITE_NOMEM ){ + sqlite3_result_error_nomem(context); + goto exprtest_out; + }else if( !pModule ){ + sqlite3_result_error(context, "No such tokenizer module", -1); + goto exprtest_out; + } + + rc = pModule->xCreate(0, 0, &pTokenizer); + assert( rc==SQLITE_NOMEM || rc==SQLITE_OK ); + if( rc==SQLITE_NOMEM ){ + sqlite3_result_error_nomem(context); + goto exprtest_out; + } + pTokenizer->pModule = pModule; + + zExpr = (const char *)sqlite3_value_text(argv[1]); + nExpr = sqlite3_value_bytes(argv[1]); + nCol = argc-2; + azCol = (char **)sqlite3_malloc(nCol*sizeof(char *)); + if( !azCol ){ + sqlite3_result_error_nomem(context); + goto exprtest_out; + } + for(ii=0; iixDestroy(pTokenizer); + } + sqlite3_free(azCol); +} + +/* +** Register the query expression parser test function fts3_exprtest() +** with database connection db. +*/ +int sqlite3Fts3ExprInitTestInterface(sqlite3* db){ + return sqlite3_create_function( + db, "fts3_exprtest", -1, SQLITE_UTF8, 0, fts3ExprTest, 0, 0 + ); +} + +#endif +#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) */ diff --git a/ext/fts3/fts3_hash.c b/ext/fts3/fts3_hash.c new file mode 100644 index 0000000..57c59b5 --- /dev/null +++ b/ext/fts3/fts3_hash.c @@ -0,0 +1,383 @@ +/* +** 2001 September 22 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This is the implementation of generic hash-tables used in SQLite. +** We've modified it slightly to serve as a standalone hash table +** implementation for the full-text indexing module. +*/ + +/* +** The code in this file is only compiled if: +** +** * The FTS3 module is being built as an extension +** (in which case SQLITE_CORE is not defined), or +** +** * The FTS3 module is being built into the core of +** SQLite (in which case SQLITE_ENABLE_FTS3 is defined). +*/ +#include "fts3Int.h" +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) + +#include +#include +#include + +#include "fts3_hash.h" + +/* +** Malloc and Free functions +*/ +static void *fts3HashMalloc(int n){ + void *p = sqlite3_malloc(n); + if( p ){ + memset(p, 0, n); + } + return p; +} +static void fts3HashFree(void *p){ + sqlite3_free(p); +} + +/* Turn bulk memory into a hash table object by initializing the +** fields of the Hash structure. +** +** "pNew" is a pointer to the hash table that is to be initialized. +** keyClass is one of the constants +** FTS3_HASH_BINARY or FTS3_HASH_STRING. The value of keyClass +** determines what kind of key the hash table will use. "copyKey" is +** true if the hash table should make its own private copy of keys and +** false if it should just use the supplied pointer. +*/ +void sqlite3Fts3HashInit(Fts3Hash *pNew, char keyClass, char copyKey){ + assert( pNew!=0 ); + assert( keyClass>=FTS3_HASH_STRING && keyClass<=FTS3_HASH_BINARY ); + pNew->keyClass = keyClass; + pNew->copyKey = copyKey; + pNew->first = 0; + pNew->count = 0; + pNew->htsize = 0; + pNew->ht = 0; +} + +/* Remove all entries from a hash table. Reclaim all memory. +** Call this routine to delete a hash table or to reset a hash table +** to the empty state. +*/ +void sqlite3Fts3HashClear(Fts3Hash *pH){ + Fts3HashElem *elem; /* For looping over all elements of the table */ + + assert( pH!=0 ); + elem = pH->first; + pH->first = 0; + fts3HashFree(pH->ht); + pH->ht = 0; + pH->htsize = 0; + while( elem ){ + Fts3HashElem *next_elem = elem->next; + if( pH->copyKey && elem->pKey ){ + fts3HashFree(elem->pKey); + } + fts3HashFree(elem); + elem = next_elem; + } + pH->count = 0; +} + +/* +** Hash and comparison functions when the mode is FTS3_HASH_STRING +*/ +static int fts3StrHash(const void *pKey, int nKey){ + const char *z = (const char *)pKey; + int h = 0; + if( nKey<=0 ) nKey = (int) strlen(z); + while( nKey > 0 ){ + h = (h<<3) ^ h ^ *z++; + nKey--; + } + return h & 0x7fffffff; +} +static int fts3StrCompare(const void *pKey1, int n1, const void *pKey2, int n2){ + if( n1!=n2 ) return 1; + return strncmp((const char*)pKey1,(const char*)pKey2,n1); +} + +/* +** Hash and comparison functions when the mode is FTS3_HASH_BINARY +*/ +static int fts3BinHash(const void *pKey, int nKey){ + int h = 0; + const char *z = (const char *)pKey; + while( nKey-- > 0 ){ + h = (h<<3) ^ h ^ *(z++); + } + return h & 0x7fffffff; +} +static int fts3BinCompare(const void *pKey1, int n1, const void *pKey2, int n2){ + if( n1!=n2 ) return 1; + return memcmp(pKey1,pKey2,n1); +} + +/* +** Return a pointer to the appropriate hash function given the key class. +** +** The C syntax in this function definition may be unfamilar to some +** programmers, so we provide the following additional explanation: +** +** The name of the function is "ftsHashFunction". The function takes a +** single parameter "keyClass". The return value of ftsHashFunction() +** is a pointer to another function. Specifically, the return value +** of ftsHashFunction() is a pointer to a function that takes two parameters +** with types "const void*" and "int" and returns an "int". +*/ +static int (*ftsHashFunction(int keyClass))(const void*,int){ + if( keyClass==FTS3_HASH_STRING ){ + return &fts3StrHash; + }else{ + assert( keyClass==FTS3_HASH_BINARY ); + return &fts3BinHash; + } +} + +/* +** Return a pointer to the appropriate hash function given the key class. +** +** For help in interpreted the obscure C code in the function definition, +** see the header comment on the previous function. +*/ +static int (*ftsCompareFunction(int keyClass))(const void*,int,const void*,int){ + if( keyClass==FTS3_HASH_STRING ){ + return &fts3StrCompare; + }else{ + assert( keyClass==FTS3_HASH_BINARY ); + return &fts3BinCompare; + } +} + +/* Link an element into the hash table +*/ +static void fts3HashInsertElement( + Fts3Hash *pH, /* The complete hash table */ + struct _fts3ht *pEntry, /* The entry into which pNew is inserted */ + Fts3HashElem *pNew /* The element to be inserted */ +){ + Fts3HashElem *pHead; /* First element already in pEntry */ + pHead = pEntry->chain; + if( pHead ){ + pNew->next = pHead; + pNew->prev = pHead->prev; + if( pHead->prev ){ pHead->prev->next = pNew; } + else { pH->first = pNew; } + pHead->prev = pNew; + }else{ + pNew->next = pH->first; + if( pH->first ){ pH->first->prev = pNew; } + pNew->prev = 0; + pH->first = pNew; + } + pEntry->count++; + pEntry->chain = pNew; +} + + +/* Resize the hash table so that it cantains "new_size" buckets. +** "new_size" must be a power of 2. The hash table might fail +** to resize if sqliteMalloc() fails. +** +** Return non-zero if a memory allocation error occurs. +*/ +static int fts3Rehash(Fts3Hash *pH, int new_size){ + struct _fts3ht *new_ht; /* The new hash table */ + Fts3HashElem *elem, *next_elem; /* For looping over existing elements */ + int (*xHash)(const void*,int); /* The hash function */ + + assert( (new_size & (new_size-1))==0 ); + new_ht = (struct _fts3ht *)fts3HashMalloc( new_size*sizeof(struct _fts3ht) ); + if( new_ht==0 ) return 1; + fts3HashFree(pH->ht); + pH->ht = new_ht; + pH->htsize = new_size; + xHash = ftsHashFunction(pH->keyClass); + for(elem=pH->first, pH->first=0; elem; elem = next_elem){ + int h = (*xHash)(elem->pKey, elem->nKey) & (new_size-1); + next_elem = elem->next; + fts3HashInsertElement(pH, &new_ht[h], elem); + } + return 0; +} + +/* This function (for internal use only) locates an element in an +** hash table that matches the given key. The hash for this key has +** already been computed and is passed as the 4th parameter. +*/ +static Fts3HashElem *fts3FindElementByHash( + const Fts3Hash *pH, /* The pH to be searched */ + const void *pKey, /* The key we are searching for */ + int nKey, + int h /* The hash for this key. */ +){ + Fts3HashElem *elem; /* Used to loop thru the element list */ + int count; /* Number of elements left to test */ + int (*xCompare)(const void*,int,const void*,int); /* comparison function */ + + if( pH->ht ){ + struct _fts3ht *pEntry = &pH->ht[h]; + elem = pEntry->chain; + count = pEntry->count; + xCompare = ftsCompareFunction(pH->keyClass); + while( count-- && elem ){ + if( (*xCompare)(elem->pKey,elem->nKey,pKey,nKey)==0 ){ + return elem; + } + elem = elem->next; + } + } + return 0; +} + +/* Remove a single entry from the hash table given a pointer to that +** element and a hash on the element's key. +*/ +static void fts3RemoveElementByHash( + Fts3Hash *pH, /* The pH containing "elem" */ + Fts3HashElem* elem, /* The element to be removed from the pH */ + int h /* Hash value for the element */ +){ + struct _fts3ht *pEntry; + if( elem->prev ){ + elem->prev->next = elem->next; + }else{ + pH->first = elem->next; + } + if( elem->next ){ + elem->next->prev = elem->prev; + } + pEntry = &pH->ht[h]; + if( pEntry->chain==elem ){ + pEntry->chain = elem->next; + } + pEntry->count--; + if( pEntry->count<=0 ){ + pEntry->chain = 0; + } + if( pH->copyKey && elem->pKey ){ + fts3HashFree(elem->pKey); + } + fts3HashFree( elem ); + pH->count--; + if( pH->count<=0 ){ + assert( pH->first==0 ); + assert( pH->count==0 ); + fts3HashClear(pH); + } +} + +Fts3HashElem *sqlite3Fts3HashFindElem( + const Fts3Hash *pH, + const void *pKey, + int nKey +){ + int h; /* A hash on key */ + int (*xHash)(const void*,int); /* The hash function */ + + if( pH==0 || pH->ht==0 ) return 0; + xHash = ftsHashFunction(pH->keyClass); + assert( xHash!=0 ); + h = (*xHash)(pKey,nKey); + assert( (pH->htsize & (pH->htsize-1))==0 ); + return fts3FindElementByHash(pH,pKey,nKey, h & (pH->htsize-1)); +} + +/* +** Attempt to locate an element of the hash table pH with a key +** that matches pKey,nKey. Return the data for this element if it is +** found, or NULL if there is no match. +*/ +void *sqlite3Fts3HashFind(const Fts3Hash *pH, const void *pKey, int nKey){ + Fts3HashElem *pElem; /* The element that matches key (if any) */ + + pElem = sqlite3Fts3HashFindElem(pH, pKey, nKey); + return pElem ? pElem->data : 0; +} + +/* Insert an element into the hash table pH. The key is pKey,nKey +** and the data is "data". +** +** If no element exists with a matching key, then a new +** element is created. A copy of the key is made if the copyKey +** flag is set. NULL is returned. +** +** If another element already exists with the same key, then the +** new data replaces the old data and the old data is returned. +** The key is not copied in this instance. If a malloc fails, then +** the new data is returned and the hash table is unchanged. +** +** If the "data" parameter to this function is NULL, then the +** element corresponding to "key" is removed from the hash table. +*/ +void *sqlite3Fts3HashInsert( + Fts3Hash *pH, /* The hash table to insert into */ + const void *pKey, /* The key */ + int nKey, /* Number of bytes in the key */ + void *data /* The data */ +){ + int hraw; /* Raw hash value of the key */ + int h; /* the hash of the key modulo hash table size */ + Fts3HashElem *elem; /* Used to loop thru the element list */ + Fts3HashElem *new_elem; /* New element added to the pH */ + int (*xHash)(const void*,int); /* The hash function */ + + assert( pH!=0 ); + xHash = ftsHashFunction(pH->keyClass); + assert( xHash!=0 ); + hraw = (*xHash)(pKey, nKey); + assert( (pH->htsize & (pH->htsize-1))==0 ); + h = hraw & (pH->htsize-1); + elem = fts3FindElementByHash(pH,pKey,nKey,h); + if( elem ){ + void *old_data = elem->data; + if( data==0 ){ + fts3RemoveElementByHash(pH,elem,h); + }else{ + elem->data = data; + } + return old_data; + } + if( data==0 ) return 0; + if( (pH->htsize==0 && fts3Rehash(pH,8)) + || (pH->count>=pH->htsize && fts3Rehash(pH, pH->htsize*2)) + ){ + pH->count = 0; + return data; + } + assert( pH->htsize>0 ); + new_elem = (Fts3HashElem*)fts3HashMalloc( sizeof(Fts3HashElem) ); + if( new_elem==0 ) return data; + if( pH->copyKey && pKey!=0 ){ + new_elem->pKey = fts3HashMalloc( nKey ); + if( new_elem->pKey==0 ){ + fts3HashFree(new_elem); + return data; + } + memcpy((void*)new_elem->pKey, pKey, nKey); + }else{ + new_elem->pKey = (void*)pKey; + } + new_elem->nKey = nKey; + pH->count++; + assert( pH->htsize>0 ); + assert( (pH->htsize & (pH->htsize-1))==0 ); + h = hraw & (pH->htsize-1); + fts3HashInsertElement(pH, &pH->ht[h], new_elem); + new_elem->data = data; + return 0; +} + +#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) */ diff --git a/ext/fts3/fts3_hash.h b/ext/fts3/fts3_hash.h new file mode 100644 index 0000000..399f515 --- /dev/null +++ b/ext/fts3/fts3_hash.h @@ -0,0 +1,112 @@ +/* +** 2001 September 22 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This is the header file for the generic hash-table implemenation +** used in SQLite. We've modified it slightly to serve as a standalone +** hash table implementation for the full-text indexing module. +** +*/ +#ifndef _FTS3_HASH_H_ +#define _FTS3_HASH_H_ + +/* Forward declarations of structures. */ +typedef struct Fts3Hash Fts3Hash; +typedef struct Fts3HashElem Fts3HashElem; + +/* A complete hash table is an instance of the following structure. +** The internals of this structure are intended to be opaque -- client +** code should not attempt to access or modify the fields of this structure +** directly. Change this structure only by using the routines below. +** However, many of the "procedures" and "functions" for modifying and +** accessing this structure are really macros, so we can't really make +** this structure opaque. +*/ +struct Fts3Hash { + char keyClass; /* HASH_INT, _POINTER, _STRING, _BINARY */ + char copyKey; /* True if copy of key made on insert */ + int count; /* Number of entries in this table */ + Fts3HashElem *first; /* The first element of the array */ + int htsize; /* Number of buckets in the hash table */ + struct _fts3ht { /* the hash table */ + int count; /* Number of entries with this hash */ + Fts3HashElem *chain; /* Pointer to first entry with this hash */ + } *ht; +}; + +/* Each element in the hash table is an instance of the following +** structure. All elements are stored on a single doubly-linked list. +** +** Again, this structure is intended to be opaque, but it can't really +** be opaque because it is used by macros. +*/ +struct Fts3HashElem { + Fts3HashElem *next, *prev; /* Next and previous elements in the table */ + void *data; /* Data associated with this element */ + void *pKey; int nKey; /* Key associated with this element */ +}; + +/* +** There are 2 different modes of operation for a hash table: +** +** FTS3_HASH_STRING pKey points to a string that is nKey bytes long +** (including the null-terminator, if any). Case +** is respected in comparisons. +** +** FTS3_HASH_BINARY pKey points to binary data nKey bytes long. +** memcmp() is used to compare keys. +** +** A copy of the key is made if the copyKey parameter to fts3HashInit is 1. +*/ +#define FTS3_HASH_STRING 1 +#define FTS3_HASH_BINARY 2 + +/* +** Access routines. To delete, insert a NULL pointer. +*/ +void sqlite3Fts3HashInit(Fts3Hash *pNew, char keyClass, char copyKey); +void *sqlite3Fts3HashInsert(Fts3Hash*, const void *pKey, int nKey, void *pData); +void *sqlite3Fts3HashFind(const Fts3Hash*, const void *pKey, int nKey); +void sqlite3Fts3HashClear(Fts3Hash*); +Fts3HashElem *sqlite3Fts3HashFindElem(const Fts3Hash *, const void *, int); + +/* +** Shorthand for the functions above +*/ +#define fts3HashInit sqlite3Fts3HashInit +#define fts3HashInsert sqlite3Fts3HashInsert +#define fts3HashFind sqlite3Fts3HashFind +#define fts3HashClear sqlite3Fts3HashClear +#define fts3HashFindElem sqlite3Fts3HashFindElem + +/* +** Macros for looping over all elements of a hash table. The idiom is +** like this: +** +** Fts3Hash h; +** Fts3HashElem *p; +** ... +** for(p=fts3HashFirst(&h); p; p=fts3HashNext(p)){ +** SomeStructure *pData = fts3HashData(p); +** // do something with pData +** } +*/ +#define fts3HashFirst(H) ((H)->first) +#define fts3HashNext(E) ((E)->next) +#define fts3HashData(E) ((E)->data) +#define fts3HashKey(E) ((E)->pKey) +#define fts3HashKeysize(E) ((E)->nKey) + +/* +** Number of entries in a hash table +*/ +#define fts3HashCount(H) ((H)->count) + +#endif /* _FTS3_HASH_H_ */ diff --git a/ext/fts3/fts3_icu.c b/ext/fts3/fts3_icu.c new file mode 100644 index 0000000..a10a55d --- /dev/null +++ b/ext/fts3/fts3_icu.c @@ -0,0 +1,258 @@ +/* +** 2007 June 22 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file implements a tokenizer for fts3 based on the ICU library. +*/ +#include "fts3Int.h" +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) +#ifdef SQLITE_ENABLE_ICU + +#include +#include +#include "fts3_tokenizer.h" + +#include +#include +#include +#include + +typedef struct IcuTokenizer IcuTokenizer; +typedef struct IcuCursor IcuCursor; + +struct IcuTokenizer { + sqlite3_tokenizer base; + char *zLocale; +}; + +struct IcuCursor { + sqlite3_tokenizer_cursor base; + + UBreakIterator *pIter; /* ICU break-iterator object */ + int nChar; /* Number of UChar elements in pInput */ + UChar *aChar; /* Copy of input using utf-16 encoding */ + int *aOffset; /* Offsets of each character in utf-8 input */ + + int nBuffer; + char *zBuffer; + + int iToken; +}; + +/* +** Create a new tokenizer instance. +*/ +static int icuCreate( + int argc, /* Number of entries in argv[] */ + const char * const *argv, /* Tokenizer creation arguments */ + sqlite3_tokenizer **ppTokenizer /* OUT: Created tokenizer */ +){ + IcuTokenizer *p; + int n = 0; + + if( argc>0 ){ + n = strlen(argv[0])+1; + } + p = (IcuTokenizer *)sqlite3_malloc(sizeof(IcuTokenizer)+n); + if( !p ){ + return SQLITE_NOMEM; + } + memset(p, 0, sizeof(IcuTokenizer)); + + if( n ){ + p->zLocale = (char *)&p[1]; + memcpy(p->zLocale, argv[0], n); + } + + *ppTokenizer = (sqlite3_tokenizer *)p; + + return SQLITE_OK; +} + +/* +** Destroy a tokenizer +*/ +static int icuDestroy(sqlite3_tokenizer *pTokenizer){ + IcuTokenizer *p = (IcuTokenizer *)pTokenizer; + sqlite3_free(p); + return SQLITE_OK; +} + +/* +** Prepare to begin tokenizing a particular string. The input +** string to be tokenized is pInput[0..nBytes-1]. A cursor +** used to incrementally tokenize this string is returned in +** *ppCursor. +*/ +static int icuOpen( + sqlite3_tokenizer *pTokenizer, /* The tokenizer */ + const char *zInput, /* Input string */ + int nInput, /* Length of zInput in bytes */ + sqlite3_tokenizer_cursor **ppCursor /* OUT: Tokenization cursor */ +){ + IcuTokenizer *p = (IcuTokenizer *)pTokenizer; + IcuCursor *pCsr; + + const int32_t opt = U_FOLD_CASE_DEFAULT; + UErrorCode status = U_ZERO_ERROR; + int nChar; + + UChar32 c; + int iInput = 0; + int iOut = 0; + + *ppCursor = 0; + + if( nInput<0 ){ + nInput = strlen(zInput); + } + nChar = nInput+1; + pCsr = (IcuCursor *)sqlite3_malloc( + sizeof(IcuCursor) + /* IcuCursor */ + nChar * sizeof(UChar) + /* IcuCursor.aChar[] */ + (nChar+1) * sizeof(int) /* IcuCursor.aOffset[] */ + ); + if( !pCsr ){ + return SQLITE_NOMEM; + } + memset(pCsr, 0, sizeof(IcuCursor)); + pCsr->aChar = (UChar *)&pCsr[1]; + pCsr->aOffset = (int *)&pCsr->aChar[nChar]; + + pCsr->aOffset[iOut] = iInput; + U8_NEXT(zInput, iInput, nInput, c); + while( c>0 ){ + int isError = 0; + c = u_foldCase(c, opt); + U16_APPEND(pCsr->aChar, iOut, nChar, c, isError); + if( isError ){ + sqlite3_free(pCsr); + return SQLITE_ERROR; + } + pCsr->aOffset[iOut] = iInput; + + if( iInputpIter = ubrk_open(UBRK_WORD, p->zLocale, pCsr->aChar, iOut, &status); + if( !U_SUCCESS(status) ){ + sqlite3_free(pCsr); + return SQLITE_ERROR; + } + pCsr->nChar = iOut; + + ubrk_first(pCsr->pIter); + *ppCursor = (sqlite3_tokenizer_cursor *)pCsr; + return SQLITE_OK; +} + +/* +** Close a tokenization cursor previously opened by a call to icuOpen(). +*/ +static int icuClose(sqlite3_tokenizer_cursor *pCursor){ + IcuCursor *pCsr = (IcuCursor *)pCursor; + ubrk_close(pCsr->pIter); + sqlite3_free(pCsr->zBuffer); + sqlite3_free(pCsr); + return SQLITE_OK; +} + +/* +** Extract the next token from a tokenization cursor. +*/ +static int icuNext( + sqlite3_tokenizer_cursor *pCursor, /* Cursor returned by simpleOpen */ + const char **ppToken, /* OUT: *ppToken is the token text */ + int *pnBytes, /* OUT: Number of bytes in token */ + int *piStartOffset, /* OUT: Starting offset of token */ + int *piEndOffset, /* OUT: Ending offset of token */ + int *piPosition /* OUT: Position integer of token */ +){ + IcuCursor *pCsr = (IcuCursor *)pCursor; + + int iStart = 0; + int iEnd = 0; + int nByte = 0; + + while( iStart==iEnd ){ + UChar32 c; + + iStart = ubrk_current(pCsr->pIter); + iEnd = ubrk_next(pCsr->pIter); + if( iEnd==UBRK_DONE ){ + return SQLITE_DONE; + } + + while( iStartaChar, iWhite, pCsr->nChar, c); + if( u_isspace(c) ){ + iStart = iWhite; + }else{ + break; + } + } + assert(iStart<=iEnd); + } + + do { + UErrorCode status = U_ZERO_ERROR; + if( nByte ){ + char *zNew = sqlite3_realloc(pCsr->zBuffer, nByte); + if( !zNew ){ + return SQLITE_NOMEM; + } + pCsr->zBuffer = zNew; + pCsr->nBuffer = nByte; + } + + u_strToUTF8( + pCsr->zBuffer, pCsr->nBuffer, &nByte, /* Output vars */ + &pCsr->aChar[iStart], iEnd-iStart, /* Input vars */ + &status /* Output success/failure */ + ); + } while( nByte>pCsr->nBuffer ); + + *ppToken = pCsr->zBuffer; + *pnBytes = nByte; + *piStartOffset = pCsr->aOffset[iStart]; + *piEndOffset = pCsr->aOffset[iEnd]; + *piPosition = pCsr->iToken++; + + return SQLITE_OK; +} + +/* +** The set of routines that implement the simple tokenizer +*/ +static const sqlite3_tokenizer_module icuTokenizerModule = { + 0, /* iVersion */ + icuCreate, /* xCreate */ + icuDestroy, /* xCreate */ + icuOpen, /* xOpen */ + icuClose, /* xClose */ + icuNext, /* xNext */ +}; + +/* +** Set *ppModule to point at the implementation of the ICU tokenizer. +*/ +void sqlite3Fts3IcuTokenizerModule( + sqlite3_tokenizer_module const**ppModule +){ + *ppModule = &icuTokenizerModule; +} + +#endif /* defined(SQLITE_ENABLE_ICU) */ +#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) */ diff --git a/ext/fts3/fts3_porter.c b/ext/fts3/fts3_porter.c new file mode 100644 index 0000000..148c570 --- /dev/null +++ b/ext/fts3/fts3_porter.c @@ -0,0 +1,645 @@ +/* +** 2006 September 30 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** Implementation of the full-text-search tokenizer that implements +** a Porter stemmer. +*/ + +/* +** The code in this file is only compiled if: +** +** * The FTS3 module is being built as an extension +** (in which case SQLITE_CORE is not defined), or +** +** * The FTS3 module is being built into the core of +** SQLite (in which case SQLITE_ENABLE_FTS3 is defined). +*/ +#include "fts3Int.h" +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) + +#include +#include +#include +#include + +#include "fts3_tokenizer.h" + +/* +** Class derived from sqlite3_tokenizer +*/ +typedef struct porter_tokenizer { + sqlite3_tokenizer base; /* Base class */ +} porter_tokenizer; + +/* +** Class derived from sqlit3_tokenizer_cursor +*/ +typedef struct porter_tokenizer_cursor { + sqlite3_tokenizer_cursor base; + const char *zInput; /* input we are tokenizing */ + int nInput; /* size of the input */ + int iOffset; /* current position in zInput */ + int iToken; /* index of next token to be returned */ + char *zToken; /* storage for current token */ + int nAllocated; /* space allocated to zToken buffer */ +} porter_tokenizer_cursor; + + +/* +** Create a new tokenizer instance. +*/ +static int porterCreate( + int argc, const char * const *argv, + sqlite3_tokenizer **ppTokenizer +){ + porter_tokenizer *t; + + UNUSED_PARAMETER(argc); + UNUSED_PARAMETER(argv); + + t = (porter_tokenizer *) sqlite3_malloc(sizeof(*t)); + if( t==NULL ) return SQLITE_NOMEM; + memset(t, 0, sizeof(*t)); + *ppTokenizer = &t->base; + return SQLITE_OK; +} + +/* +** Destroy a tokenizer +*/ +static int porterDestroy(sqlite3_tokenizer *pTokenizer){ + sqlite3_free(pTokenizer); + return SQLITE_OK; +} + +/* +** Prepare to begin tokenizing a particular string. The input +** string to be tokenized is zInput[0..nInput-1]. A cursor +** used to incrementally tokenize this string is returned in +** *ppCursor. +*/ +static int porterOpen( + sqlite3_tokenizer *pTokenizer, /* The tokenizer */ + const char *zInput, int nInput, /* String to be tokenized */ + sqlite3_tokenizer_cursor **ppCursor /* OUT: Tokenization cursor */ +){ + porter_tokenizer_cursor *c; + + UNUSED_PARAMETER(pTokenizer); + + c = (porter_tokenizer_cursor *) sqlite3_malloc(sizeof(*c)); + if( c==NULL ) return SQLITE_NOMEM; + + c->zInput = zInput; + if( zInput==0 ){ + c->nInput = 0; + }else if( nInput<0 ){ + c->nInput = (int)strlen(zInput); + }else{ + c->nInput = nInput; + } + c->iOffset = 0; /* start tokenizing at the beginning */ + c->iToken = 0; + c->zToken = NULL; /* no space allocated, yet. */ + c->nAllocated = 0; + + *ppCursor = &c->base; + return SQLITE_OK; +} + +/* +** Close a tokenization cursor previously opened by a call to +** porterOpen() above. +*/ +static int porterClose(sqlite3_tokenizer_cursor *pCursor){ + porter_tokenizer_cursor *c = (porter_tokenizer_cursor *) pCursor; + sqlite3_free(c->zToken); + sqlite3_free(c); + return SQLITE_OK; +} +/* +** Vowel or consonant +*/ +static const char cType[] = { + 0, 1, 1, 1, 0, 1, 1, 1, 0, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 0, + 1, 1, 1, 2, 1 +}; + +/* +** isConsonant() and isVowel() determine if their first character in +** the string they point to is a consonant or a vowel, according +** to Porter ruls. +** +** A consonate is any letter other than 'a', 'e', 'i', 'o', or 'u'. +** 'Y' is a consonant unless it follows another consonant, +** in which case it is a vowel. +** +** In these routine, the letters are in reverse order. So the 'y' rule +** is that 'y' is a consonant unless it is followed by another +** consonent. +*/ +static int isVowel(const char*); +static int isConsonant(const char *z){ + int j; + char x = *z; + if( x==0 ) return 0; + assert( x>='a' && x<='z' ); + j = cType[x-'a']; + if( j<2 ) return j; + return z[1]==0 || isVowel(z + 1); +} +static int isVowel(const char *z){ + int j; + char x = *z; + if( x==0 ) return 0; + assert( x>='a' && x<='z' ); + j = cType[x-'a']; + if( j<2 ) return 1-j; + return isConsonant(z + 1); +} + +/* +** Let any sequence of one or more vowels be represented by V and let +** C be sequence of one or more consonants. Then every word can be +** represented as: +** +** [C] (VC){m} [V] +** +** In prose: A word is an optional consonant followed by zero or +** vowel-consonant pairs followed by an optional vowel. "m" is the +** number of vowel consonant pairs. This routine computes the value +** of m for the first i bytes of a word. +** +** Return true if the m-value for z is 1 or more. In other words, +** return true if z contains at least one vowel that is followed +** by a consonant. +** +** In this routine z[] is in reverse order. So we are really looking +** for an instance of of a consonant followed by a vowel. +*/ +static int m_gt_0(const char *z){ + while( isVowel(z) ){ z++; } + if( *z==0 ) return 0; + while( isConsonant(z) ){ z++; } + return *z!=0; +} + +/* Like mgt0 above except we are looking for a value of m which is +** exactly 1 +*/ +static int m_eq_1(const char *z){ + while( isVowel(z) ){ z++; } + if( *z==0 ) return 0; + while( isConsonant(z) ){ z++; } + if( *z==0 ) return 0; + while( isVowel(z) ){ z++; } + if( *z==0 ) return 1; + while( isConsonant(z) ){ z++; } + return *z==0; +} + +/* Like mgt0 above except we are looking for a value of m>1 instead +** or m>0 +*/ +static int m_gt_1(const char *z){ + while( isVowel(z) ){ z++; } + if( *z==0 ) return 0; + while( isConsonant(z) ){ z++; } + if( *z==0 ) return 0; + while( isVowel(z) ){ z++; } + if( *z==0 ) return 0; + while( isConsonant(z) ){ z++; } + return *z!=0; +} + +/* +** Return TRUE if there is a vowel anywhere within z[0..n-1] +*/ +static int hasVowel(const char *z){ + while( isConsonant(z) ){ z++; } + return *z!=0; +} + +/* +** Return TRUE if the word ends in a double consonant. +** +** The text is reversed here. So we are really looking at +** the first two characters of z[]. +*/ +static int doubleConsonant(const char *z){ + return isConsonant(z) && z[0]==z[1]; +} + +/* +** Return TRUE if the word ends with three letters which +** are consonant-vowel-consonent and where the final consonant +** is not 'w', 'x', or 'y'. +** +** The word is reversed here. So we are really checking the +** first three letters and the first one cannot be in [wxy]. +*/ +static int star_oh(const char *z){ + return + isConsonant(z) && + z[0]!='w' && z[0]!='x' && z[0]!='y' && + isVowel(z+1) && + isConsonant(z+2); +} + +/* +** If the word ends with zFrom and xCond() is true for the stem +** of the word that preceeds the zFrom ending, then change the +** ending to zTo. +** +** The input word *pz and zFrom are both in reverse order. zTo +** is in normal order. +** +** Return TRUE if zFrom matches. Return FALSE if zFrom does not +** match. Not that TRUE is returned even if xCond() fails and +** no substitution occurs. +*/ +static int stem( + char **pz, /* The word being stemmed (Reversed) */ + const char *zFrom, /* If the ending matches this... (Reversed) */ + const char *zTo, /* ... change the ending to this (not reversed) */ + int (*xCond)(const char*) /* Condition that must be true */ +){ + char *z = *pz; + while( *zFrom && *zFrom==*z ){ z++; zFrom++; } + if( *zFrom!=0 ) return 0; + if( xCond && !xCond(z) ) return 1; + while( *zTo ){ + *(--z) = *(zTo++); + } + *pz = z; + return 1; +} + +/* +** This is the fallback stemmer used when the porter stemmer is +** inappropriate. The input word is copied into the output with +** US-ASCII case folding. If the input word is too long (more +** than 20 bytes if it contains no digits or more than 6 bytes if +** it contains digits) then word is truncated to 20 or 6 bytes +** by taking 10 or 3 bytes from the beginning and end. +*/ +static void copy_stemmer(const char *zIn, int nIn, char *zOut, int *pnOut){ + int i, mx, j; + int hasDigit = 0; + for(i=0; i='A' && c<='Z' ){ + zOut[i] = c - 'A' + 'a'; + }else{ + if( c>='0' && c<='9' ) hasDigit = 1; + zOut[i] = c; + } + } + mx = hasDigit ? 3 : 10; + if( nIn>mx*2 ){ + for(j=mx, i=nIn-mx; i=(int)sizeof(zReverse)-7 ){ + /* The word is too big or too small for the porter stemmer. + ** Fallback to the copy stemmer */ + copy_stemmer(zIn, nIn, zOut, pnOut); + return; + } + for(i=0, j=sizeof(zReverse)-6; i='A' && c<='Z' ){ + zReverse[j] = c + 'a' - 'A'; + }else if( c>='a' && c<='z' ){ + zReverse[j] = c; + }else{ + /* The use of a character not in [a-zA-Z] means that we fallback + ** to the copy stemmer */ + copy_stemmer(zIn, nIn, zOut, pnOut); + return; + } + } + memset(&zReverse[sizeof(zReverse)-5], 0, 5); + z = &zReverse[j+1]; + + + /* Step 1a */ + if( z[0]=='s' ){ + if( + !stem(&z, "sess", "ss", 0) && + !stem(&z, "sei", "i", 0) && + !stem(&z, "ss", "ss", 0) + ){ + z++; + } + } + + /* Step 1b */ + z2 = z; + if( stem(&z, "dee", "ee", m_gt_0) ){ + /* Do nothing. The work was all in the test */ + }else if( + (stem(&z, "gni", "", hasVowel) || stem(&z, "de", "", hasVowel)) + && z!=z2 + ){ + if( stem(&z, "ta", "ate", 0) || + stem(&z, "lb", "ble", 0) || + stem(&z, "zi", "ize", 0) ){ + /* Do nothing. The work was all in the test */ + }else if( doubleConsonant(z) && (*z!='l' && *z!='s' && *z!='z') ){ + z++; + }else if( m_eq_1(z) && star_oh(z) ){ + *(--z) = 'e'; + } + } + + /* Step 1c */ + if( z[0]=='y' && hasVowel(z+1) ){ + z[0] = 'i'; + } + + /* Step 2 */ + switch( z[1] ){ + case 'a': + stem(&z, "lanoita", "ate", m_gt_0) || + stem(&z, "lanoit", "tion", m_gt_0); + break; + case 'c': + stem(&z, "icne", "ence", m_gt_0) || + stem(&z, "icna", "ance", m_gt_0); + break; + case 'e': + stem(&z, "rezi", "ize", m_gt_0); + break; + case 'g': + stem(&z, "igol", "log", m_gt_0); + break; + case 'l': + stem(&z, "ilb", "ble", m_gt_0) || + stem(&z, "illa", "al", m_gt_0) || + stem(&z, "iltne", "ent", m_gt_0) || + stem(&z, "ile", "e", m_gt_0) || + stem(&z, "ilsuo", "ous", m_gt_0); + break; + case 'o': + stem(&z, "noitazi", "ize", m_gt_0) || + stem(&z, "noita", "ate", m_gt_0) || + stem(&z, "rota", "ate", m_gt_0); + break; + case 's': + stem(&z, "msila", "al", m_gt_0) || + stem(&z, "ssenevi", "ive", m_gt_0) || + stem(&z, "ssenluf", "ful", m_gt_0) || + stem(&z, "ssensuo", "ous", m_gt_0); + break; + case 't': + stem(&z, "itila", "al", m_gt_0) || + stem(&z, "itivi", "ive", m_gt_0) || + stem(&z, "itilib", "ble", m_gt_0); + break; + } + + /* Step 3 */ + switch( z[0] ){ + case 'e': + stem(&z, "etaci", "ic", m_gt_0) || + stem(&z, "evita", "", m_gt_0) || + stem(&z, "ezila", "al", m_gt_0); + break; + case 'i': + stem(&z, "itici", "ic", m_gt_0); + break; + case 'l': + stem(&z, "laci", "ic", m_gt_0) || + stem(&z, "luf", "", m_gt_0); + break; + case 's': + stem(&z, "ssen", "", m_gt_0); + break; + } + + /* Step 4 */ + switch( z[1] ){ + case 'a': + if( z[0]=='l' && m_gt_1(z+2) ){ + z += 2; + } + break; + case 'c': + if( z[0]=='e' && z[2]=='n' && (z[3]=='a' || z[3]=='e') && m_gt_1(z+4) ){ + z += 4; + } + break; + case 'e': + if( z[0]=='r' && m_gt_1(z+2) ){ + z += 2; + } + break; + case 'i': + if( z[0]=='c' && m_gt_1(z+2) ){ + z += 2; + } + break; + case 'l': + if( z[0]=='e' && z[2]=='b' && (z[3]=='a' || z[3]=='i') && m_gt_1(z+4) ){ + z += 4; + } + break; + case 'n': + if( z[0]=='t' ){ + if( z[2]=='a' ){ + if( m_gt_1(z+3) ){ + z += 3; + } + }else if( z[2]=='e' ){ + stem(&z, "tneme", "", m_gt_1) || + stem(&z, "tnem", "", m_gt_1) || + stem(&z, "tne", "", m_gt_1); + } + } + break; + case 'o': + if( z[0]=='u' ){ + if( m_gt_1(z+2) ){ + z += 2; + } + }else if( z[3]=='s' || z[3]=='t' ){ + stem(&z, "noi", "", m_gt_1); + } + break; + case 's': + if( z[0]=='m' && z[2]=='i' && m_gt_1(z+3) ){ + z += 3; + } + break; + case 't': + stem(&z, "eta", "", m_gt_1) || + stem(&z, "iti", "", m_gt_1); + break; + case 'u': + if( z[0]=='s' && z[2]=='o' && m_gt_1(z+3) ){ + z += 3; + } + break; + case 'v': + case 'z': + if( z[0]=='e' && z[2]=='i' && m_gt_1(z+3) ){ + z += 3; + } + break; + } + + /* Step 5a */ + if( z[0]=='e' ){ + if( m_gt_1(z+1) ){ + z++; + }else if( m_eq_1(z+1) && !star_oh(z+1) ){ + z++; + } + } + + /* Step 5b */ + if( m_gt_1(z) && z[0]=='l' && z[1]=='l' ){ + z++; + } + + /* z[] is now the stemmed word in reverse order. Flip it back + ** around into forward order and return. + */ + *pnOut = i = (int)strlen(z); + zOut[i] = 0; + while( *z ){ + zOut[--i] = *(z++); + } +} + +/* +** Characters that can be part of a token. We assume any character +** whose value is greater than 0x80 (any UTF character) can be +** part of a token. In other words, delimiters all must have +** values of 0x7f or lower. +*/ +static const char porterIdChar[] = { +/* x0 x1 x2 x3 x4 x5 x6 x7 x8 x9 xA xB xC xD xE xF */ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, /* 3x */ + 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /* 4x */ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 1, /* 5x */ + 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /* 6x */ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, /* 7x */ +}; +#define isDelim(C) (((ch=C)&0x80)==0 && (ch<0x30 || !porterIdChar[ch-0x30])) + +/* +** Extract the next token from a tokenization cursor. The cursor must +** have been opened by a prior call to porterOpen(). +*/ +static int porterNext( + sqlite3_tokenizer_cursor *pCursor, /* Cursor returned by porterOpen */ + const char **pzToken, /* OUT: *pzToken is the token text */ + int *pnBytes, /* OUT: Number of bytes in token */ + int *piStartOffset, /* OUT: Starting offset of token */ + int *piEndOffset, /* OUT: Ending offset of token */ + int *piPosition /* OUT: Position integer of token */ +){ + porter_tokenizer_cursor *c = (porter_tokenizer_cursor *) pCursor; + const char *z = c->zInput; + + while( c->iOffsetnInput ){ + int iStartOffset, ch; + + /* Scan past delimiter characters */ + while( c->iOffsetnInput && isDelim(z[c->iOffset]) ){ + c->iOffset++; + } + + /* Count non-delimiter characters. */ + iStartOffset = c->iOffset; + while( c->iOffsetnInput && !isDelim(z[c->iOffset]) ){ + c->iOffset++; + } + + if( c->iOffset>iStartOffset ){ + int n = c->iOffset-iStartOffset; + if( n>c->nAllocated ){ + char *pNew; + c->nAllocated = n+20; + pNew = sqlite3_realloc(c->zToken, c->nAllocated); + if( !pNew ) return SQLITE_NOMEM; + c->zToken = pNew; + } + porter_stemmer(&z[iStartOffset], n, c->zToken, pnBytes); + *pzToken = c->zToken; + *piStartOffset = iStartOffset; + *piEndOffset = c->iOffset; + *piPosition = c->iToken++; + return SQLITE_OK; + } + } + return SQLITE_DONE; +} + +/* +** The set of routines that implement the porter-stemmer tokenizer +*/ +static const sqlite3_tokenizer_module porterTokenizerModule = { + 0, + porterCreate, + porterDestroy, + porterOpen, + porterClose, + porterNext, +}; + +/* +** Allocate a new porter tokenizer. Return a pointer to the new +** tokenizer in *ppModule +*/ +void sqlite3Fts3PorterTokenizerModule( + sqlite3_tokenizer_module const**ppModule +){ + *ppModule = &porterTokenizerModule; +} + +#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) */ diff --git a/ext/fts3/fts3_snippet.c b/ext/fts3/fts3_snippet.c new file mode 100644 index 0000000..23ef25c --- /dev/null +++ b/ext/fts3/fts3_snippet.c @@ -0,0 +1,1500 @@ +/* +** 2009 Oct 23 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +*/ + +#include "fts3Int.h" +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) + +#include +#include + +/* +** Characters that may appear in the second argument to matchinfo(). +*/ +#define FTS3_MATCHINFO_NPHRASE 'p' /* 1 value */ +#define FTS3_MATCHINFO_NCOL 'c' /* 1 value */ +#define FTS3_MATCHINFO_NDOC 'n' /* 1 value */ +#define FTS3_MATCHINFO_AVGLENGTH 'a' /* nCol values */ +#define FTS3_MATCHINFO_LENGTH 'l' /* nCol values */ +#define FTS3_MATCHINFO_LCS 's' /* nCol values */ +#define FTS3_MATCHINFO_HITS 'x' /* 3*nCol*nPhrase values */ + +/* +** The default value for the second argument to matchinfo(). +*/ +#define FTS3_MATCHINFO_DEFAULT "pcx" + + +/* +** Used as an fts3ExprIterate() context when loading phrase doclists to +** Fts3Expr.aDoclist[]/nDoclist. +*/ +typedef struct LoadDoclistCtx LoadDoclistCtx; +struct LoadDoclistCtx { + Fts3Cursor *pCsr; /* FTS3 Cursor */ + int nPhrase; /* Number of phrases seen so far */ + int nToken; /* Number of tokens seen so far */ +}; + +/* +** The following types are used as part of the implementation of the +** fts3BestSnippet() routine. +*/ +typedef struct SnippetIter SnippetIter; +typedef struct SnippetPhrase SnippetPhrase; +typedef struct SnippetFragment SnippetFragment; + +struct SnippetIter { + Fts3Cursor *pCsr; /* Cursor snippet is being generated from */ + int iCol; /* Extract snippet from this column */ + int nSnippet; /* Requested snippet length (in tokens) */ + int nPhrase; /* Number of phrases in query */ + SnippetPhrase *aPhrase; /* Array of size nPhrase */ + int iCurrent; /* First token of current snippet */ +}; + +struct SnippetPhrase { + int nToken; /* Number of tokens in phrase */ + char *pList; /* Pointer to start of phrase position list */ + int iHead; /* Next value in position list */ + char *pHead; /* Position list data following iHead */ + int iTail; /* Next value in trailing position list */ + char *pTail; /* Position list data following iTail */ +}; + +struct SnippetFragment { + int iCol; /* Column snippet is extracted from */ + int iPos; /* Index of first token in snippet */ + u64 covered; /* Mask of query phrases covered */ + u64 hlmask; /* Mask of snippet terms to highlight */ +}; + +/* +** This type is used as an fts3ExprIterate() context object while +** accumulating the data returned by the matchinfo() function. +*/ +typedef struct MatchInfo MatchInfo; +struct MatchInfo { + Fts3Cursor *pCursor; /* FTS3 Cursor */ + int nCol; /* Number of columns in table */ + int nPhrase; /* Number of matchable phrases in query */ + sqlite3_int64 nDoc; /* Number of docs in database */ + u32 *aMatchinfo; /* Pre-allocated buffer */ +}; + + + +/* +** The snippet() and offsets() functions both return text values. An instance +** of the following structure is used to accumulate those values while the +** functions are running. See fts3StringAppend() for details. +*/ +typedef struct StrBuffer StrBuffer; +struct StrBuffer { + char *z; /* Pointer to buffer containing string */ + int n; /* Length of z in bytes (excl. nul-term) */ + int nAlloc; /* Allocated size of buffer z in bytes */ +}; + + +/* +** This function is used to help iterate through a position-list. A position +** list is a list of unique integers, sorted from smallest to largest. Each +** element of the list is represented by an FTS3 varint that takes the value +** of the difference between the current element and the previous one plus +** two. For example, to store the position-list: +** +** 4 9 113 +** +** the three varints: +** +** 6 7 106 +** +** are encoded. +** +** When this function is called, *pp points to the start of an element of +** the list. *piPos contains the value of the previous entry in the list. +** After it returns, *piPos contains the value of the next element of the +** list and *pp is advanced to the following varint. +*/ +static void fts3GetDeltaPosition(char **pp, int *piPos){ + int iVal; + *pp += sqlite3Fts3GetVarint32(*pp, &iVal); + *piPos += (iVal-2); +} + +/* +** Helper function for fts3ExprIterate() (see below). +*/ +static int fts3ExprIterate2( + Fts3Expr *pExpr, /* Expression to iterate phrases of */ + int *piPhrase, /* Pointer to phrase counter */ + int (*x)(Fts3Expr*,int,void*), /* Callback function to invoke for phrases */ + void *pCtx /* Second argument to pass to callback */ +){ + int rc; /* Return code */ + int eType = pExpr->eType; /* Type of expression node pExpr */ + + if( eType!=FTSQUERY_PHRASE ){ + assert( pExpr->pLeft && pExpr->pRight ); + rc = fts3ExprIterate2(pExpr->pLeft, piPhrase, x, pCtx); + if( rc==SQLITE_OK && eType!=FTSQUERY_NOT ){ + rc = fts3ExprIterate2(pExpr->pRight, piPhrase, x, pCtx); + } + }else{ + rc = x(pExpr, *piPhrase, pCtx); + (*piPhrase)++; + } + return rc; +} + +/* +** Iterate through all phrase nodes in an FTS3 query, except those that +** are part of a sub-tree that is the right-hand-side of a NOT operator. +** For each phrase node found, the supplied callback function is invoked. +** +** If the callback function returns anything other than SQLITE_OK, +** the iteration is abandoned and the error code returned immediately. +** Otherwise, SQLITE_OK is returned after a callback has been made for +** all eligible phrase nodes. +*/ +static int fts3ExprIterate( + Fts3Expr *pExpr, /* Expression to iterate phrases of */ + int (*x)(Fts3Expr*,int,void*), /* Callback function to invoke for phrases */ + void *pCtx /* Second argument to pass to callback */ +){ + int iPhrase = 0; /* Variable used as the phrase counter */ + return fts3ExprIterate2(pExpr, &iPhrase, x, pCtx); +} + +/* +** This is an fts3ExprIterate() callback used while loading the doclists +** for each phrase into Fts3Expr.aDoclist[]/nDoclist. See also +** fts3ExprLoadDoclists(). +*/ +static int fts3ExprLoadDoclistsCb(Fts3Expr *pExpr, int iPhrase, void *ctx){ + int rc = SQLITE_OK; + Fts3Phrase *pPhrase = pExpr->pPhrase; + LoadDoclistCtx *p = (LoadDoclistCtx *)ctx; + + UNUSED_PARAMETER(iPhrase); + + p->nPhrase++; + p->nToken += pPhrase->nToken; + + return rc; +} + +/* +** Load the doclists for each phrase in the query associated with FTS3 cursor +** pCsr. +** +** If pnPhrase is not NULL, then *pnPhrase is set to the number of matchable +** phrases in the expression (all phrases except those directly or +** indirectly descended from the right-hand-side of a NOT operator). If +** pnToken is not NULL, then it is set to the number of tokens in all +** matchable phrases of the expression. +*/ +static int fts3ExprLoadDoclists( + Fts3Cursor *pCsr, /* Fts3 cursor for current query */ + int *pnPhrase, /* OUT: Number of phrases in query */ + int *pnToken /* OUT: Number of tokens in query */ +){ + int rc; /* Return Code */ + LoadDoclistCtx sCtx = {0,0,0}; /* Context for fts3ExprIterate() */ + sCtx.pCsr = pCsr; + rc = fts3ExprIterate(pCsr->pExpr, fts3ExprLoadDoclistsCb, (void *)&sCtx); + if( pnPhrase ) *pnPhrase = sCtx.nPhrase; + if( pnToken ) *pnToken = sCtx.nToken; + return rc; +} + +static int fts3ExprPhraseCountCb(Fts3Expr *pExpr, int iPhrase, void *ctx){ + (*(int *)ctx)++; + UNUSED_PARAMETER(pExpr); + UNUSED_PARAMETER(iPhrase); + return SQLITE_OK; +} +static int fts3ExprPhraseCount(Fts3Expr *pExpr){ + int nPhrase = 0; + (void)fts3ExprIterate(pExpr, fts3ExprPhraseCountCb, (void *)&nPhrase); + return nPhrase; +} + +/* +** Advance the position list iterator specified by the first two +** arguments so that it points to the first element with a value greater +** than or equal to parameter iNext. +*/ +static void fts3SnippetAdvance(char **ppIter, int *piIter, int iNext){ + char *pIter = *ppIter; + if( pIter ){ + int iIter = *piIter; + + while( iIteriCurrent<0 ){ + /* The SnippetIter object has just been initialized. The first snippet + ** candidate always starts at offset 0 (even if this candidate has a + ** score of 0.0). + */ + pIter->iCurrent = 0; + + /* Advance the 'head' iterator of each phrase to the first offset that + ** is greater than or equal to (iNext+nSnippet). + */ + for(i=0; inPhrase; i++){ + SnippetPhrase *pPhrase = &pIter->aPhrase[i]; + fts3SnippetAdvance(&pPhrase->pHead, &pPhrase->iHead, pIter->nSnippet); + } + }else{ + int iStart; + int iEnd = 0x7FFFFFFF; + + for(i=0; inPhrase; i++){ + SnippetPhrase *pPhrase = &pIter->aPhrase[i]; + if( pPhrase->pHead && pPhrase->iHeadiHead; + } + } + if( iEnd==0x7FFFFFFF ){ + return 1; + } + + pIter->iCurrent = iStart = iEnd - pIter->nSnippet + 1; + for(i=0; inPhrase; i++){ + SnippetPhrase *pPhrase = &pIter->aPhrase[i]; + fts3SnippetAdvance(&pPhrase->pHead, &pPhrase->iHead, iEnd+1); + fts3SnippetAdvance(&pPhrase->pTail, &pPhrase->iTail, iStart); + } + } + + return 0; +} + +/* +** Retrieve information about the current candidate snippet of snippet +** iterator pIter. +*/ +static void fts3SnippetDetails( + SnippetIter *pIter, /* Snippet iterator */ + u64 mCovered, /* Bitmask of phrases already covered */ + int *piToken, /* OUT: First token of proposed snippet */ + int *piScore, /* OUT: "Score" for this snippet */ + u64 *pmCover, /* OUT: Bitmask of phrases covered */ + u64 *pmHighlight /* OUT: Bitmask of terms to highlight */ +){ + int iStart = pIter->iCurrent; /* First token of snippet */ + int iScore = 0; /* Score of this snippet */ + int i; /* Loop counter */ + u64 mCover = 0; /* Mask of phrases covered by this snippet */ + u64 mHighlight = 0; /* Mask of tokens to highlight in snippet */ + + for(i=0; inPhrase; i++){ + SnippetPhrase *pPhrase = &pIter->aPhrase[i]; + if( pPhrase->pTail ){ + char *pCsr = pPhrase->pTail; + int iCsr = pPhrase->iTail; + + while( iCsr<(iStart+pIter->nSnippet) ){ + int j; + u64 mPhrase = (u64)1 << i; + u64 mPos = (u64)1 << (iCsr - iStart); + assert( iCsr>=iStart ); + if( (mCover|mCovered)&mPhrase ){ + iScore++; + }else{ + iScore += 1000; + } + mCover |= mPhrase; + + for(j=0; jnToken; j++){ + mHighlight |= (mPos>>j); + } + + if( 0==(*pCsr & 0x0FE) ) break; + fts3GetDeltaPosition(&pCsr, &iCsr); + } + } + } + + /* Set the output variables before returning. */ + *piToken = iStart; + *piScore = iScore; + *pmCover = mCover; + *pmHighlight = mHighlight; +} + +/* +** This function is an fts3ExprIterate() callback used by fts3BestSnippet(). +** Each invocation populates an element of the SnippetIter.aPhrase[] array. +*/ +static int fts3SnippetFindPositions(Fts3Expr *pExpr, int iPhrase, void *ctx){ + SnippetIter *p = (SnippetIter *)ctx; + SnippetPhrase *pPhrase = &p->aPhrase[iPhrase]; + char *pCsr; + + pPhrase->nToken = pExpr->pPhrase->nToken; + + pCsr = sqlite3Fts3EvalPhrasePoslist(p->pCsr, pExpr, p->iCol); + if( pCsr ){ + int iFirst = 0; + pPhrase->pList = pCsr; + fts3GetDeltaPosition(&pCsr, &iFirst); + assert( iFirst>=0 ); + pPhrase->pHead = pCsr; + pPhrase->pTail = pCsr; + pPhrase->iHead = iFirst; + pPhrase->iTail = iFirst; + }else{ + assert( pPhrase->pList==0 && pPhrase->pHead==0 && pPhrase->pTail==0 ); + } + + return SQLITE_OK; +} + +/* +** Select the fragment of text consisting of nFragment contiguous tokens +** from column iCol that represent the "best" snippet. The best snippet +** is the snippet with the highest score, where scores are calculated +** by adding: +** +** (a) +1 point for each occurence of a matchable phrase in the snippet. +** +** (b) +1000 points for the first occurence of each matchable phrase in +** the snippet for which the corresponding mCovered bit is not set. +** +** The selected snippet parameters are stored in structure *pFragment before +** returning. The score of the selected snippet is stored in *piScore +** before returning. +*/ +static int fts3BestSnippet( + int nSnippet, /* Desired snippet length */ + Fts3Cursor *pCsr, /* Cursor to create snippet for */ + int iCol, /* Index of column to create snippet from */ + u64 mCovered, /* Mask of phrases already covered */ + u64 *pmSeen, /* IN/OUT: Mask of phrases seen */ + SnippetFragment *pFragment, /* OUT: Best snippet found */ + int *piScore /* OUT: Score of snippet pFragment */ +){ + int rc; /* Return Code */ + int nList; /* Number of phrases in expression */ + SnippetIter sIter; /* Iterates through snippet candidates */ + int nByte; /* Number of bytes of space to allocate */ + int iBestScore = -1; /* Best snippet score found so far */ + int i; /* Loop counter */ + + memset(&sIter, 0, sizeof(sIter)); + + /* Iterate through the phrases in the expression to count them. The same + ** callback makes sure the doclists are loaded for each phrase. + */ + rc = fts3ExprLoadDoclists(pCsr, &nList, 0); + if( rc!=SQLITE_OK ){ + return rc; + } + + /* Now that it is known how many phrases there are, allocate and zero + ** the required space using malloc(). + */ + nByte = sizeof(SnippetPhrase) * nList; + sIter.aPhrase = (SnippetPhrase *)sqlite3_malloc(nByte); + if( !sIter.aPhrase ){ + return SQLITE_NOMEM; + } + memset(sIter.aPhrase, 0, nByte); + + /* Initialize the contents of the SnippetIter object. Then iterate through + ** the set of phrases in the expression to populate the aPhrase[] array. + */ + sIter.pCsr = pCsr; + sIter.iCol = iCol; + sIter.nSnippet = nSnippet; + sIter.nPhrase = nList; + sIter.iCurrent = -1; + (void)fts3ExprIterate(pCsr->pExpr, fts3SnippetFindPositions, (void *)&sIter); + + /* Set the *pmSeen output variable. */ + for(i=0; iiCol = iCol; + while( !fts3SnippetNextCandidate(&sIter) ){ + int iPos; + int iScore; + u64 mCover; + u64 mHighlight; + fts3SnippetDetails(&sIter, mCovered, &iPos, &iScore, &mCover, &mHighlight); + assert( iScore>=0 ); + if( iScore>iBestScore ){ + pFragment->iPos = iPos; + pFragment->hlmask = mHighlight; + pFragment->covered = mCover; + iBestScore = iScore; + } + } + + sqlite3_free(sIter.aPhrase); + *piScore = iBestScore; + return SQLITE_OK; +} + + +/* +** Append a string to the string-buffer passed as the first argument. +** +** If nAppend is negative, then the length of the string zAppend is +** determined using strlen(). +*/ +static int fts3StringAppend( + StrBuffer *pStr, /* Buffer to append to */ + const char *zAppend, /* Pointer to data to append to buffer */ + int nAppend /* Size of zAppend in bytes (or -1) */ +){ + if( nAppend<0 ){ + nAppend = (int)strlen(zAppend); + } + + /* If there is insufficient space allocated at StrBuffer.z, use realloc() + ** to grow the buffer until so that it is big enough to accomadate the + ** appended data. + */ + if( pStr->n+nAppend+1>=pStr->nAlloc ){ + int nAlloc = pStr->nAlloc+nAppend+100; + char *zNew = sqlite3_realloc(pStr->z, nAlloc); + if( !zNew ){ + return SQLITE_NOMEM; + } + pStr->z = zNew; + pStr->nAlloc = nAlloc; + } + + /* Append the data to the string buffer. */ + memcpy(&pStr->z[pStr->n], zAppend, nAppend); + pStr->n += nAppend; + pStr->z[pStr->n] = '\0'; + + return SQLITE_OK; +} + +/* +** The fts3BestSnippet() function often selects snippets that end with a +** query term. That is, the final term of the snippet is always a term +** that requires highlighting. For example, if 'X' is a highlighted term +** and '.' is a non-highlighted term, BestSnippet() may select: +** +** ........X.....X +** +** This function "shifts" the beginning of the snippet forward in the +** document so that there are approximately the same number of +** non-highlighted terms to the right of the final highlighted term as there +** are to the left of the first highlighted term. For example, to this: +** +** ....X.....X.... +** +** This is done as part of extracting the snippet text, not when selecting +** the snippet. Snippet selection is done based on doclists only, so there +** is no way for fts3BestSnippet() to know whether or not the document +** actually contains terms that follow the final highlighted term. +*/ +static int fts3SnippetShift( + Fts3Table *pTab, /* FTS3 table snippet comes from */ + int nSnippet, /* Number of tokens desired for snippet */ + const char *zDoc, /* Document text to extract snippet from */ + int nDoc, /* Size of buffer zDoc in bytes */ + int *piPos, /* IN/OUT: First token of snippet */ + u64 *pHlmask /* IN/OUT: Mask of tokens to highlight */ +){ + u64 hlmask = *pHlmask; /* Local copy of initial highlight-mask */ + + if( hlmask ){ + int nLeft; /* Tokens to the left of first highlight */ + int nRight; /* Tokens to the right of last highlight */ + int nDesired; /* Ideal number of tokens to shift forward */ + + for(nLeft=0; !(hlmask & ((u64)1 << nLeft)); nLeft++); + for(nRight=0; !(hlmask & ((u64)1 << (nSnippet-1-nRight))); nRight++); + nDesired = (nLeft-nRight)/2; + + /* Ideally, the start of the snippet should be pushed forward in the + ** document nDesired tokens. This block checks if there are actually + ** nDesired tokens to the right of the snippet. If so, *piPos and + ** *pHlMask are updated to shift the snippet nDesired tokens to the + ** right. Otherwise, the snippet is shifted by the number of tokens + ** available. + */ + if( nDesired>0 ){ + int nShift; /* Number of tokens to shift snippet by */ + int iCurrent = 0; /* Token counter */ + int rc; /* Return Code */ + sqlite3_tokenizer_module *pMod; + sqlite3_tokenizer_cursor *pC; + pMod = (sqlite3_tokenizer_module *)pTab->pTokenizer->pModule; + + /* Open a cursor on zDoc/nDoc. Check if there are (nSnippet+nDesired) + ** or more tokens in zDoc/nDoc. + */ + rc = pMod->xOpen(pTab->pTokenizer, zDoc, nDoc, &pC); + if( rc!=SQLITE_OK ){ + return rc; + } + pC->pTokenizer = pTab->pTokenizer; + while( rc==SQLITE_OK && iCurrent<(nSnippet+nDesired) ){ + const char *ZDUMMY; int DUMMY1, DUMMY2, DUMMY3; + rc = pMod->xNext(pC, &ZDUMMY, &DUMMY1, &DUMMY2, &DUMMY3, &iCurrent); + } + pMod->xClose(pC); + if( rc!=SQLITE_OK && rc!=SQLITE_DONE ){ return rc; } + + nShift = (rc==SQLITE_DONE)+iCurrent-nSnippet; + assert( nShift<=nDesired ); + if( nShift>0 ){ + *piPos += nShift; + *pHlmask = hlmask >> nShift; + } + } + } + return SQLITE_OK; +} + +/* +** Extract the snippet text for fragment pFragment from cursor pCsr and +** append it to string buffer pOut. +*/ +static int fts3SnippetText( + Fts3Cursor *pCsr, /* FTS3 Cursor */ + SnippetFragment *pFragment, /* Snippet to extract */ + int iFragment, /* Fragment number */ + int isLast, /* True for final fragment in snippet */ + int nSnippet, /* Number of tokens in extracted snippet */ + const char *zOpen, /* String inserted before highlighted term */ + const char *zClose, /* String inserted after highlighted term */ + const char *zEllipsis, /* String inserted between snippets */ + StrBuffer *pOut /* Write output here */ +){ + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + int rc; /* Return code */ + const char *zDoc; /* Document text to extract snippet from */ + int nDoc; /* Size of zDoc in bytes */ + int iCurrent = 0; /* Current token number of document */ + int iEnd = 0; /* Byte offset of end of current token */ + int isShiftDone = 0; /* True after snippet is shifted */ + int iPos = pFragment->iPos; /* First token of snippet */ + u64 hlmask = pFragment->hlmask; /* Highlight-mask for snippet */ + int iCol = pFragment->iCol+1; /* Query column to extract text from */ + sqlite3_tokenizer_module *pMod; /* Tokenizer module methods object */ + sqlite3_tokenizer_cursor *pC; /* Tokenizer cursor open on zDoc/nDoc */ + const char *ZDUMMY; /* Dummy argument used with tokenizer */ + int DUMMY1; /* Dummy argument used with tokenizer */ + + zDoc = (const char *)sqlite3_column_text(pCsr->pStmt, iCol); + if( zDoc==0 ){ + if( sqlite3_column_type(pCsr->pStmt, iCol)!=SQLITE_NULL ){ + return SQLITE_NOMEM; + } + return SQLITE_OK; + } + nDoc = sqlite3_column_bytes(pCsr->pStmt, iCol); + + /* Open a token cursor on the document. */ + pMod = (sqlite3_tokenizer_module *)pTab->pTokenizer->pModule; + rc = pMod->xOpen(pTab->pTokenizer, zDoc, nDoc, &pC); + if( rc!=SQLITE_OK ){ + return rc; + } + pC->pTokenizer = pTab->pTokenizer; + + while( rc==SQLITE_OK ){ + int iBegin; /* Offset in zDoc of start of token */ + int iFin; /* Offset in zDoc of end of token */ + int isHighlight; /* True for highlighted terms */ + + rc = pMod->xNext(pC, &ZDUMMY, &DUMMY1, &iBegin, &iFin, &iCurrent); + if( rc!=SQLITE_OK ){ + if( rc==SQLITE_DONE ){ + /* Special case - the last token of the snippet is also the last token + ** of the column. Append any punctuation that occurred between the end + ** of the previous token and the end of the document to the output. + ** Then break out of the loop. */ + rc = fts3StringAppend(pOut, &zDoc[iEnd], -1); + } + break; + } + if( iCurrent0 || iFragment>0) ){ + rc = fts3StringAppend(pOut, zEllipsis, -1); + } + if( rc!=SQLITE_OK || iCurrent=(iPos+nSnippet) ){ + if( isLast ){ + rc = fts3StringAppend(pOut, zEllipsis, -1); + } + break; + } + + /* Set isHighlight to true if this term should be highlighted. */ + isHighlight = (hlmask & ((u64)1 << (iCurrent-iPos)))!=0; + + if( iCurrent>iPos ) rc = fts3StringAppend(pOut, &zDoc[iEnd], iBegin-iEnd); + if( rc==SQLITE_OK && isHighlight ) rc = fts3StringAppend(pOut, zOpen, -1); + if( rc==SQLITE_OK ) rc = fts3StringAppend(pOut, &zDoc[iBegin], iFin-iBegin); + if( rc==SQLITE_OK && isHighlight ) rc = fts3StringAppend(pOut, zClose, -1); + + iEnd = iFin; + } + + pMod->xClose(pC); + return rc; +} + + +/* +** This function is used to count the entries in a column-list (a +** delta-encoded list of term offsets within a single column of a single +** row). When this function is called, *ppCollist should point to the +** beginning of the first varint in the column-list (the varint that +** contains the position of the first matching term in the column data). +** Before returning, *ppCollist is set to point to the first byte after +** the last varint in the column-list (either the 0x00 signifying the end +** of the position-list, or the 0x01 that precedes the column number of +** the next column in the position-list). +** +** The number of elements in the column-list is returned. +*/ +static int fts3ColumnlistCount(char **ppCollist){ + char *pEnd = *ppCollist; + char c = 0; + int nEntry = 0; + + /* A column-list is terminated by either a 0x01 or 0x00. */ + while( 0xFE & (*pEnd | c) ){ + c = *pEnd++ & 0x80; + if( !c ) nEntry++; + } + + *ppCollist = pEnd; + return nEntry; +} + +/* +** fts3ExprIterate() callback used to collect the "global" matchinfo stats +** for a single query. +** +** fts3ExprIterate() callback to load the 'global' elements of a +** FTS3_MATCHINFO_HITS matchinfo array. The global stats are those elements +** of the matchinfo array that are constant for all rows returned by the +** current query. +** +** Argument pCtx is actually a pointer to a struct of type MatchInfo. This +** function populates Matchinfo.aMatchinfo[] as follows: +** +** for(iCol=0; iColpCursor, pExpr, &p->aMatchinfo[3*iPhrase*p->nCol] + ); +} + +/* +** fts3ExprIterate() callback used to collect the "local" part of the +** FTS3_MATCHINFO_HITS array. The local stats are those elements of the +** array that are different for each row returned by the query. +*/ +static int fts3ExprLocalHitsCb( + Fts3Expr *pExpr, /* Phrase expression node */ + int iPhrase, /* Phrase number */ + void *pCtx /* Pointer to MatchInfo structure */ +){ + MatchInfo *p = (MatchInfo *)pCtx; + int iStart = iPhrase * p->nCol * 3; + int i; + + for(i=0; inCol; i++){ + char *pCsr; + pCsr = sqlite3Fts3EvalPhrasePoslist(p->pCursor, pExpr, i); + if( pCsr ){ + p->aMatchinfo[iStart+i*3] = fts3ColumnlistCount(&pCsr); + }else{ + p->aMatchinfo[iStart+i*3] = 0; + } + } + + return SQLITE_OK; +} + +static int fts3MatchinfoCheck( + Fts3Table *pTab, + char cArg, + char **pzErr +){ + if( (cArg==FTS3_MATCHINFO_NPHRASE) + || (cArg==FTS3_MATCHINFO_NCOL) + || (cArg==FTS3_MATCHINFO_NDOC && pTab->bHasStat) + || (cArg==FTS3_MATCHINFO_AVGLENGTH && pTab->bHasStat) + || (cArg==FTS3_MATCHINFO_LENGTH && pTab->bHasDocsize) + || (cArg==FTS3_MATCHINFO_LCS) + || (cArg==FTS3_MATCHINFO_HITS) + ){ + return SQLITE_OK; + } + *pzErr = sqlite3_mprintf("unrecognized matchinfo request: %c", cArg); + return SQLITE_ERROR; +} + +static int fts3MatchinfoSize(MatchInfo *pInfo, char cArg){ + int nVal; /* Number of integers output by cArg */ + + switch( cArg ){ + case FTS3_MATCHINFO_NDOC: + case FTS3_MATCHINFO_NPHRASE: + case FTS3_MATCHINFO_NCOL: + nVal = 1; + break; + + case FTS3_MATCHINFO_AVGLENGTH: + case FTS3_MATCHINFO_LENGTH: + case FTS3_MATCHINFO_LCS: + nVal = pInfo->nCol; + break; + + default: + assert( cArg==FTS3_MATCHINFO_HITS ); + nVal = pInfo->nCol * pInfo->nPhrase * 3; + break; + } + + return nVal; +} + +static int fts3MatchinfoSelectDoctotal( + Fts3Table *pTab, + sqlite3_stmt **ppStmt, + sqlite3_int64 *pnDoc, + const char **paLen +){ + sqlite3_stmt *pStmt; + const char *a; + sqlite3_int64 nDoc; + + if( !*ppStmt ){ + int rc = sqlite3Fts3SelectDoctotal(pTab, ppStmt); + if( rc!=SQLITE_OK ) return rc; + } + pStmt = *ppStmt; + assert( sqlite3_data_count(pStmt)==1 ); + + a = sqlite3_column_blob(pStmt, 0); + a += sqlite3Fts3GetVarint(a, &nDoc); + if( nDoc==0 ) return FTS_CORRUPT_VTAB; + *pnDoc = (u32)nDoc; + + if( paLen ) *paLen = a; + return SQLITE_OK; +} + +/* +** An instance of the following structure is used to store state while +** iterating through a multi-column position-list corresponding to the +** hits for a single phrase on a single row in order to calculate the +** values for a matchinfo() FTS3_MATCHINFO_LCS request. +*/ +typedef struct LcsIterator LcsIterator; +struct LcsIterator { + Fts3Expr *pExpr; /* Pointer to phrase expression */ + int iPosOffset; /* Tokens count up to end of this phrase */ + char *pRead; /* Cursor used to iterate through aDoclist */ + int iPos; /* Current position */ +}; + +/* +** If LcsIterator.iCol is set to the following value, the iterator has +** finished iterating through all offsets for all columns. +*/ +#define LCS_ITERATOR_FINISHED 0x7FFFFFFF; + +static int fts3MatchinfoLcsCb( + Fts3Expr *pExpr, /* Phrase expression node */ + int iPhrase, /* Phrase number (numbered from zero) */ + void *pCtx /* Pointer to MatchInfo structure */ +){ + LcsIterator *aIter = (LcsIterator *)pCtx; + aIter[iPhrase].pExpr = pExpr; + return SQLITE_OK; +} + +/* +** Advance the iterator passed as an argument to the next position. Return +** 1 if the iterator is at EOF or if it now points to the start of the +** position list for the next column. +*/ +static int fts3LcsIteratorAdvance(LcsIterator *pIter){ + char *pRead = pIter->pRead; + sqlite3_int64 iRead; + int rc = 0; + + pRead += sqlite3Fts3GetVarint(pRead, &iRead); + if( iRead==0 || iRead==1 ){ + pRead = 0; + rc = 1; + }else{ + pIter->iPos += (int)(iRead-2); + } + + pIter->pRead = pRead; + return rc; +} + +/* +** This function implements the FTS3_MATCHINFO_LCS matchinfo() flag. +** +** If the call is successful, the longest-common-substring lengths for each +** column are written into the first nCol elements of the pInfo->aMatchinfo[] +** array before returning. SQLITE_OK is returned in this case. +** +** Otherwise, if an error occurs, an SQLite error code is returned and the +** data written to the first nCol elements of pInfo->aMatchinfo[] is +** undefined. +*/ +static int fts3MatchinfoLcs(Fts3Cursor *pCsr, MatchInfo *pInfo){ + LcsIterator *aIter; + int i; + int iCol; + int nToken = 0; + + /* Allocate and populate the array of LcsIterator objects. The array + ** contains one element for each matchable phrase in the query. + **/ + aIter = sqlite3_malloc(sizeof(LcsIterator) * pCsr->nPhrase); + if( !aIter ) return SQLITE_NOMEM; + memset(aIter, 0, sizeof(LcsIterator) * pCsr->nPhrase); + (void)fts3ExprIterate(pCsr->pExpr, fts3MatchinfoLcsCb, (void*)aIter); + + for(i=0; inPhrase; i++){ + LcsIterator *pIter = &aIter[i]; + nToken -= pIter->pExpr->pPhrase->nToken; + pIter->iPosOffset = nToken; + } + + for(iCol=0; iColnCol; iCol++){ + int nLcs = 0; /* LCS value for this column */ + int nLive = 0; /* Number of iterators in aIter not at EOF */ + + for(i=0; inPhrase; i++){ + LcsIterator *pIt = &aIter[i]; + pIt->pRead = sqlite3Fts3EvalPhrasePoslist(pCsr, pIt->pExpr, iCol); + if( pIt->pRead ){ + pIt->iPos = pIt->iPosOffset; + fts3LcsIteratorAdvance(&aIter[i]); + nLive++; + } + } + + while( nLive>0 ){ + LcsIterator *pAdv = 0; /* The iterator to advance by one position */ + int nThisLcs = 0; /* LCS for the current iterator positions */ + + for(i=0; inPhrase; i++){ + LcsIterator *pIter = &aIter[i]; + if( pIter->pRead==0 ){ + /* This iterator is already at EOF for this column. */ + nThisLcs = 0; + }else{ + if( pAdv==0 || pIter->iPosiPos ){ + pAdv = pIter; + } + if( nThisLcs==0 || pIter->iPos==pIter[-1].iPos ){ + nThisLcs++; + }else{ + nThisLcs = 1; + } + if( nThisLcs>nLcs ) nLcs = nThisLcs; + } + } + if( fts3LcsIteratorAdvance(pAdv) ) nLive--; + } + + pInfo->aMatchinfo[iCol] = nLcs; + } + + sqlite3_free(aIter); + return SQLITE_OK; +} + +/* +** Populate the buffer pInfo->aMatchinfo[] with an array of integers to +** be returned by the matchinfo() function. Argument zArg contains the +** format string passed as the second argument to matchinfo (or the +** default value "pcx" if no second argument was specified). The format +** string has already been validated and the pInfo->aMatchinfo[] array +** is guaranteed to be large enough for the output. +** +** If bGlobal is true, then populate all fields of the matchinfo() output. +** If it is false, then assume that those fields that do not change between +** rows (i.e. FTS3_MATCHINFO_NPHRASE, NCOL, NDOC, AVGLENGTH and part of HITS) +** have already been populated. +** +** Return SQLITE_OK if successful, or an SQLite error code if an error +** occurs. If a value other than SQLITE_OK is returned, the state the +** pInfo->aMatchinfo[] buffer is left in is undefined. +*/ +static int fts3MatchinfoValues( + Fts3Cursor *pCsr, /* FTS3 cursor object */ + int bGlobal, /* True to grab the global stats */ + MatchInfo *pInfo, /* Matchinfo context object */ + const char *zArg /* Matchinfo format string */ +){ + int rc = SQLITE_OK; + int i; + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + sqlite3_stmt *pSelect = 0; + + for(i=0; rc==SQLITE_OK && zArg[i]; i++){ + + switch( zArg[i] ){ + case FTS3_MATCHINFO_NPHRASE: + if( bGlobal ) pInfo->aMatchinfo[0] = pInfo->nPhrase; + break; + + case FTS3_MATCHINFO_NCOL: + if( bGlobal ) pInfo->aMatchinfo[0] = pInfo->nCol; + break; + + case FTS3_MATCHINFO_NDOC: + if( bGlobal ){ + sqlite3_int64 nDoc = 0; + rc = fts3MatchinfoSelectDoctotal(pTab, &pSelect, &nDoc, 0); + pInfo->aMatchinfo[0] = (u32)nDoc; + } + break; + + case FTS3_MATCHINFO_AVGLENGTH: + if( bGlobal ){ + sqlite3_int64 nDoc; /* Number of rows in table */ + const char *a; /* Aggregate column length array */ + + rc = fts3MatchinfoSelectDoctotal(pTab, &pSelect, &nDoc, &a); + if( rc==SQLITE_OK ){ + int iCol; + for(iCol=0; iColnCol; iCol++){ + u32 iVal; + sqlite3_int64 nToken; + a += sqlite3Fts3GetVarint(a, &nToken); + iVal = (u32)(((u32)(nToken&0xffffffff)+nDoc/2)/nDoc); + pInfo->aMatchinfo[iCol] = iVal; + } + } + } + break; + + case FTS3_MATCHINFO_LENGTH: { + sqlite3_stmt *pSelectDocsize = 0; + rc = sqlite3Fts3SelectDocsize(pTab, pCsr->iPrevId, &pSelectDocsize); + if( rc==SQLITE_OK ){ + int iCol; + const char *a = sqlite3_column_blob(pSelectDocsize, 0); + for(iCol=0; iColnCol; iCol++){ + sqlite3_int64 nToken; + a += sqlite3Fts3GetVarint(a, &nToken); + pInfo->aMatchinfo[iCol] = (u32)nToken; + } + } + sqlite3_reset(pSelectDocsize); + break; + } + + case FTS3_MATCHINFO_LCS: + rc = fts3ExprLoadDoclists(pCsr, 0, 0); + if( rc==SQLITE_OK ){ + rc = fts3MatchinfoLcs(pCsr, pInfo); + } + break; + + default: { + Fts3Expr *pExpr; + assert( zArg[i]==FTS3_MATCHINFO_HITS ); + pExpr = pCsr->pExpr; + rc = fts3ExprLoadDoclists(pCsr, 0, 0); + if( rc!=SQLITE_OK ) break; + if( bGlobal ){ + if( pCsr->pDeferred ){ + rc = fts3MatchinfoSelectDoctotal(pTab, &pSelect, &pInfo->nDoc, 0); + if( rc!=SQLITE_OK ) break; + } + rc = fts3ExprIterate(pExpr, fts3ExprGlobalHitsCb,(void*)pInfo); + if( rc!=SQLITE_OK ) break; + } + (void)fts3ExprIterate(pExpr, fts3ExprLocalHitsCb,(void*)pInfo); + break; + } + } + + pInfo->aMatchinfo += fts3MatchinfoSize(pInfo, zArg[i]); + } + + sqlite3_reset(pSelect); + return rc; +} + + +/* +** Populate pCsr->aMatchinfo[] with data for the current row. The +** 'matchinfo' data is an array of 32-bit unsigned integers (C type u32). +*/ +static int fts3GetMatchinfo( + Fts3Cursor *pCsr, /* FTS3 Cursor object */ + const char *zArg /* Second argument to matchinfo() function */ +){ + MatchInfo sInfo; + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + int rc = SQLITE_OK; + int bGlobal = 0; /* Collect 'global' stats as well as local */ + + memset(&sInfo, 0, sizeof(MatchInfo)); + sInfo.pCursor = pCsr; + sInfo.nCol = pTab->nColumn; + + /* If there is cached matchinfo() data, but the format string for the + ** cache does not match the format string for this request, discard + ** the cached data. */ + if( pCsr->zMatchinfo && strcmp(pCsr->zMatchinfo, zArg) ){ + assert( pCsr->aMatchinfo ); + sqlite3_free(pCsr->aMatchinfo); + pCsr->zMatchinfo = 0; + pCsr->aMatchinfo = 0; + } + + /* If Fts3Cursor.aMatchinfo[] is NULL, then this is the first time the + ** matchinfo function has been called for this query. In this case + ** allocate the array used to accumulate the matchinfo data and + ** initialize those elements that are constant for every row. + */ + if( pCsr->aMatchinfo==0 ){ + int nMatchinfo = 0; /* Number of u32 elements in match-info */ + int nArg; /* Bytes in zArg */ + int i; /* Used to iterate through zArg */ + + /* Determine the number of phrases in the query */ + pCsr->nPhrase = fts3ExprPhraseCount(pCsr->pExpr); + sInfo.nPhrase = pCsr->nPhrase; + + /* Determine the number of integers in the buffer returned by this call. */ + for(i=0; zArg[i]; i++){ + nMatchinfo += fts3MatchinfoSize(&sInfo, zArg[i]); + } + + /* Allocate space for Fts3Cursor.aMatchinfo[] and Fts3Cursor.zMatchinfo. */ + nArg = (int)strlen(zArg); + pCsr->aMatchinfo = (u32 *)sqlite3_malloc(sizeof(u32)*nMatchinfo + nArg + 1); + if( !pCsr->aMatchinfo ) return SQLITE_NOMEM; + + pCsr->zMatchinfo = (char *)&pCsr->aMatchinfo[nMatchinfo]; + pCsr->nMatchinfo = nMatchinfo; + memcpy(pCsr->zMatchinfo, zArg, nArg+1); + memset(pCsr->aMatchinfo, 0, sizeof(u32)*nMatchinfo); + pCsr->isMatchinfoNeeded = 1; + bGlobal = 1; + } + + sInfo.aMatchinfo = pCsr->aMatchinfo; + sInfo.nPhrase = pCsr->nPhrase; + if( pCsr->isMatchinfoNeeded ){ + rc = fts3MatchinfoValues(pCsr, bGlobal, &sInfo, zArg); + pCsr->isMatchinfoNeeded = 0; + } + + return rc; +} + +/* +** Implementation of snippet() function. +*/ +void sqlite3Fts3Snippet( + sqlite3_context *pCtx, /* SQLite function call context */ + Fts3Cursor *pCsr, /* Cursor object */ + const char *zStart, /* Snippet start text - "" */ + const char *zEnd, /* Snippet end text - "" */ + const char *zEllipsis, /* Snippet ellipsis text - "..." */ + int iCol, /* Extract snippet from this column */ + int nToken /* Approximate number of tokens in snippet */ +){ + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + int rc = SQLITE_OK; + int i; + StrBuffer res = {0, 0, 0}; + + /* The returned text includes up to four fragments of text extracted from + ** the data in the current row. The first iteration of the for(...) loop + ** below attempts to locate a single fragment of text nToken tokens in + ** size that contains at least one instance of all phrases in the query + ** expression that appear in the current row. If such a fragment of text + ** cannot be found, the second iteration of the loop attempts to locate + ** a pair of fragments, and so on. + */ + int nSnippet = 0; /* Number of fragments in this snippet */ + SnippetFragment aSnippet[4]; /* Maximum of 4 fragments per snippet */ + int nFToken = -1; /* Number of tokens in each fragment */ + + if( !pCsr->pExpr ){ + sqlite3_result_text(pCtx, "", 0, SQLITE_STATIC); + return; + } + + for(nSnippet=1; 1; nSnippet++){ + + int iSnip; /* Loop counter 0..nSnippet-1 */ + u64 mCovered = 0; /* Bitmask of phrases covered by snippet */ + u64 mSeen = 0; /* Bitmask of phrases seen by BestSnippet() */ + + if( nToken>=0 ){ + nFToken = (nToken+nSnippet-1) / nSnippet; + }else{ + nFToken = -1 * nToken; + } + + for(iSnip=0; iSnipnColumn; iRead++){ + SnippetFragment sF = {0, 0, 0, 0}; + int iS; + if( iCol>=0 && iRead!=iCol ) continue; + + /* Find the best snippet of nFToken tokens in column iRead. */ + rc = fts3BestSnippet(nFToken, pCsr, iRead, mCovered, &mSeen, &sF, &iS); + if( rc!=SQLITE_OK ){ + goto snippet_out; + } + if( iS>iBestScore ){ + *pFragment = sF; + iBestScore = iS; + } + } + + mCovered |= pFragment->covered; + } + + /* If all query phrases seen by fts3BestSnippet() are present in at least + ** one of the nSnippet snippet fragments, break out of the loop. + */ + assert( (mCovered&mSeen)==mCovered ); + if( mSeen==mCovered || nSnippet==SizeofArray(aSnippet) ) break; + } + + assert( nFToken>0 ); + + for(i=0; ipCsr, pExpr, p->iCol); + nTerm = pExpr->pPhrase->nToken; + if( pList ){ + fts3GetDeltaPosition(&pList, &iPos); + assert( iPos>=0 ); + } + + for(iTerm=0; iTermaTerm[p->iTerm++]; + pT->iOff = nTerm-iTerm-1; + pT->pList = pList; + pT->iPos = iPos; + } + + return SQLITE_OK; +} + +/* +** Implementation of offsets() function. +*/ +void sqlite3Fts3Offsets( + sqlite3_context *pCtx, /* SQLite function call context */ + Fts3Cursor *pCsr /* Cursor object */ +){ + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + sqlite3_tokenizer_module const *pMod = pTab->pTokenizer->pModule; + const char *ZDUMMY; /* Dummy argument used with xNext() */ + int NDUMMY; /* Dummy argument used with xNext() */ + int rc; /* Return Code */ + int nToken; /* Number of tokens in query */ + int iCol; /* Column currently being processed */ + StrBuffer res = {0, 0, 0}; /* Result string */ + TermOffsetCtx sCtx; /* Context for fts3ExprTermOffsetInit() */ + + if( !pCsr->pExpr ){ + sqlite3_result_text(pCtx, "", 0, SQLITE_STATIC); + return; + } + + memset(&sCtx, 0, sizeof(sCtx)); + assert( pCsr->isRequireSeek==0 ); + + /* Count the number of terms in the query */ + rc = fts3ExprLoadDoclists(pCsr, 0, &nToken); + if( rc!=SQLITE_OK ) goto offsets_out; + + /* Allocate the array of TermOffset iterators. */ + sCtx.aTerm = (TermOffset *)sqlite3_malloc(sizeof(TermOffset)*nToken); + if( 0==sCtx.aTerm ){ + rc = SQLITE_NOMEM; + goto offsets_out; + } + sCtx.iDocid = pCsr->iPrevId; + sCtx.pCsr = pCsr; + + /* Loop through the table columns, appending offset information to + ** string-buffer res for each column. + */ + for(iCol=0; iColnColumn; iCol++){ + sqlite3_tokenizer_cursor *pC; /* Tokenizer cursor */ + int iStart; + int iEnd; + int iCurrent; + const char *zDoc; + int nDoc; + + /* Initialize the contents of sCtx.aTerm[] for column iCol. There is + ** no way that this operation can fail, so the return code from + ** fts3ExprIterate() can be discarded. + */ + sCtx.iCol = iCol; + sCtx.iTerm = 0; + (void)fts3ExprIterate(pCsr->pExpr, fts3ExprTermOffsetInit, (void *)&sCtx); + + /* Retreive the text stored in column iCol. If an SQL NULL is stored + ** in column iCol, jump immediately to the next iteration of the loop. + ** If an OOM occurs while retrieving the data (this can happen if SQLite + ** needs to transform the data from utf-16 to utf-8), return SQLITE_NOMEM + ** to the caller. + */ + zDoc = (const char *)sqlite3_column_text(pCsr->pStmt, iCol+1); + nDoc = sqlite3_column_bytes(pCsr->pStmt, iCol+1); + if( zDoc==0 ){ + if( sqlite3_column_type(pCsr->pStmt, iCol+1)==SQLITE_NULL ){ + continue; + } + rc = SQLITE_NOMEM; + goto offsets_out; + } + + /* Initialize a tokenizer iterator to iterate through column iCol. */ + rc = pMod->xOpen(pTab->pTokenizer, zDoc, nDoc, &pC); + if( rc!=SQLITE_OK ) goto offsets_out; + pC->pTokenizer = pTab->pTokenizer; + + rc = pMod->xNext(pC, &ZDUMMY, &NDUMMY, &iStart, &iEnd, &iCurrent); + while( rc==SQLITE_OK ){ + int i; /* Used to loop through terms */ + int iMinPos = 0x7FFFFFFF; /* Position of next token */ + TermOffset *pTerm = 0; /* TermOffset associated with next token */ + + for(i=0; ipList && (pT->iPos-pT->iOff)iPos-pT->iOff; + pTerm = pT; + } + } + + if( !pTerm ){ + /* All offsets for this column have been gathered. */ + rc = SQLITE_DONE; + }else{ + assert( iCurrent<=iMinPos ); + if( 0==(0xFE&*pTerm->pList) ){ + pTerm->pList = 0; + }else{ + fts3GetDeltaPosition(&pTerm->pList, &pTerm->iPos); + } + while( rc==SQLITE_OK && iCurrentxNext(pC, &ZDUMMY, &NDUMMY, &iStart, &iEnd, &iCurrent); + } + if( rc==SQLITE_OK ){ + char aBuffer[64]; + sqlite3_snprintf(sizeof(aBuffer), aBuffer, + "%d %d %d %d ", iCol, pTerm-sCtx.aTerm, iStart, iEnd-iStart + ); + rc = fts3StringAppend(&res, aBuffer, -1); + }else if( rc==SQLITE_DONE && pTab->zContentTbl==0 ){ + rc = FTS_CORRUPT_VTAB; + } + } + } + if( rc==SQLITE_DONE ){ + rc = SQLITE_OK; + } + + pMod->xClose(pC); + if( rc!=SQLITE_OK ) goto offsets_out; + } + + offsets_out: + sqlite3_free(sCtx.aTerm); + assert( rc!=SQLITE_DONE ); + sqlite3Fts3SegmentsClose(pTab); + if( rc!=SQLITE_OK ){ + sqlite3_result_error_code(pCtx, rc); + sqlite3_free(res.z); + }else{ + sqlite3_result_text(pCtx, res.z, res.n-1, sqlite3_free); + } + return; +} + +/* +** Implementation of matchinfo() function. +*/ +void sqlite3Fts3Matchinfo( + sqlite3_context *pContext, /* Function call context */ + Fts3Cursor *pCsr, /* FTS3 table cursor */ + const char *zArg /* Second arg to matchinfo() function */ +){ + Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab; + int rc; + int i; + const char *zFormat; + + if( zArg ){ + for(i=0; zArg[i]; i++){ + char *zErr = 0; + if( fts3MatchinfoCheck(pTab, zArg[i], &zErr) ){ + sqlite3_result_error(pContext, zErr, -1); + sqlite3_free(zErr); + return; + } + } + zFormat = zArg; + }else{ + zFormat = FTS3_MATCHINFO_DEFAULT; + } + + if( !pCsr->pExpr ){ + sqlite3_result_blob(pContext, "", 0, SQLITE_STATIC); + return; + } + + /* Retrieve matchinfo() data. */ + rc = fts3GetMatchinfo(pCsr, zFormat); + sqlite3Fts3SegmentsClose(pTab); + + if( rc!=SQLITE_OK ){ + sqlite3_result_error_code(pContext, rc); + }else{ + int n = pCsr->nMatchinfo * sizeof(u32); + sqlite3_result_blob(pContext, pCsr->aMatchinfo, n, SQLITE_TRANSIENT); + } +} + +#endif diff --git a/ext/fts3/fts3_term.c b/ext/fts3/fts3_term.c new file mode 100644 index 0000000..d3eb690 --- /dev/null +++ b/ext/fts3/fts3_term.c @@ -0,0 +1,369 @@ +/* +** 2011 Jan 27 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This file is not part of the production FTS code. It is only used for +** testing. It contains a virtual table implementation that provides direct +** access to the full-text index of an FTS table. +*/ + +#include "fts3Int.h" +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) +#ifdef SQLITE_TEST + +#include +#include +#include + +typedef struct Fts3termTable Fts3termTable; +typedef struct Fts3termCursor Fts3termCursor; + +struct Fts3termTable { + sqlite3_vtab base; /* Base class used by SQLite core */ + int iIndex; /* Index for Fts3Table.aIndex[] */ + Fts3Table *pFts3Tab; +}; + +struct Fts3termCursor { + sqlite3_vtab_cursor base; /* Base class used by SQLite core */ + Fts3MultiSegReader csr; /* Must be right after "base" */ + Fts3SegFilter filter; + + int isEof; /* True if cursor is at EOF */ + char *pNext; + + sqlite3_int64 iRowid; /* Current 'rowid' value */ + sqlite3_int64 iDocid; /* Current 'docid' value */ + int iCol; /* Current 'col' value */ + int iPos; /* Current 'pos' value */ +}; + +/* +** Schema of the terms table. +*/ +#define FTS3_TERMS_SCHEMA "CREATE TABLE x(term, docid, col, pos)" + +/* +** This function does all the work for both the xConnect and xCreate methods. +** These tables have no persistent representation of their own, so xConnect +** and xCreate are identical operations. +*/ +static int fts3termConnectMethod( + sqlite3 *db, /* Database connection */ + void *pCtx, /* Non-zero for an fts4prefix table */ + int argc, /* Number of elements in argv array */ + const char * const *argv, /* xCreate/xConnect argument array */ + sqlite3_vtab **ppVtab, /* OUT: New sqlite3_vtab object */ + char **pzErr /* OUT: sqlite3_malloc'd error message */ +){ + char const *zDb; /* Name of database (e.g. "main") */ + char const *zFts3; /* Name of fts3 table */ + int nDb; /* Result of strlen(zDb) */ + int nFts3; /* Result of strlen(zFts3) */ + int nByte; /* Bytes of space to allocate here */ + int rc; /* value returned by declare_vtab() */ + Fts3termTable *p; /* Virtual table object to return */ + int iIndex = 0; + + if( argc==5 ){ + iIndex = atoi(argv[4]); + argc--; + } + + /* The user should specify a single argument - the name of an fts3 table. */ + if( argc!=4 ){ + *pzErr = sqlite3_mprintf( + "wrong number of arguments to fts4term constructor" + ); + return SQLITE_ERROR; + } + + zDb = argv[1]; + nDb = strlen(zDb); + zFts3 = argv[3]; + nFts3 = strlen(zFts3); + + rc = sqlite3_declare_vtab(db, FTS3_TERMS_SCHEMA); + if( rc!=SQLITE_OK ) return rc; + + nByte = sizeof(Fts3termTable) + sizeof(Fts3Table) + nDb + nFts3 + 2; + p = (Fts3termTable *)sqlite3_malloc(nByte); + if( !p ) return SQLITE_NOMEM; + memset(p, 0, nByte); + + p->pFts3Tab = (Fts3Table *)&p[1]; + p->pFts3Tab->zDb = (char *)&p->pFts3Tab[1]; + p->pFts3Tab->zName = &p->pFts3Tab->zDb[nDb+1]; + p->pFts3Tab->db = db; + p->pFts3Tab->nIndex = iIndex+1; + p->iIndex = iIndex; + + memcpy((char *)p->pFts3Tab->zDb, zDb, nDb); + memcpy((char *)p->pFts3Tab->zName, zFts3, nFts3); + sqlite3Fts3Dequote((char *)p->pFts3Tab->zName); + + *ppVtab = (sqlite3_vtab *)p; + return SQLITE_OK; +} + +/* +** This function does the work for both the xDisconnect and xDestroy methods. +** These tables have no persistent representation of their own, so xDisconnect +** and xDestroy are identical operations. +*/ +static int fts3termDisconnectMethod(sqlite3_vtab *pVtab){ + Fts3termTable *p = (Fts3termTable *)pVtab; + Fts3Table *pFts3 = p->pFts3Tab; + int i; + + /* Free any prepared statements held */ + for(i=0; iaStmt); i++){ + sqlite3_finalize(pFts3->aStmt[i]); + } + sqlite3_free(pFts3->zSegmentsTbl); + sqlite3_free(p); + return SQLITE_OK; +} + +#define FTS4AUX_EQ_CONSTRAINT 1 +#define FTS4AUX_GE_CONSTRAINT 2 +#define FTS4AUX_LE_CONSTRAINT 4 + +/* +** xBestIndex - Analyze a WHERE and ORDER BY clause. +*/ +static int fts3termBestIndexMethod( + sqlite3_vtab *pVTab, + sqlite3_index_info *pInfo +){ + UNUSED_PARAMETER(pVTab); + + /* This vtab naturally does "ORDER BY term, docid, col, pos". */ + if( pInfo->nOrderBy ){ + int i; + for(i=0; inOrderBy; i++){ + if( pInfo->aOrderBy[i].iColumn!=i || pInfo->aOrderBy[i].desc ) break; + } + if( i==pInfo->nOrderBy ){ + pInfo->orderByConsumed = 1; + } + } + + return SQLITE_OK; +} + +/* +** xOpen - Open a cursor. +*/ +static int fts3termOpenMethod(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCsr){ + Fts3termCursor *pCsr; /* Pointer to cursor object to return */ + + UNUSED_PARAMETER(pVTab); + + pCsr = (Fts3termCursor *)sqlite3_malloc(sizeof(Fts3termCursor)); + if( !pCsr ) return SQLITE_NOMEM; + memset(pCsr, 0, sizeof(Fts3termCursor)); + + *ppCsr = (sqlite3_vtab_cursor *)pCsr; + return SQLITE_OK; +} + +/* +** xClose - Close a cursor. +*/ +static int fts3termCloseMethod(sqlite3_vtab_cursor *pCursor){ + Fts3Table *pFts3 = ((Fts3termTable *)pCursor->pVtab)->pFts3Tab; + Fts3termCursor *pCsr = (Fts3termCursor *)pCursor; + + sqlite3Fts3SegmentsClose(pFts3); + sqlite3Fts3SegReaderFinish(&pCsr->csr); + sqlite3_free(pCsr); + return SQLITE_OK; +} + +/* +** xNext - Advance the cursor to the next row, if any. +*/ +static int fts3termNextMethod(sqlite3_vtab_cursor *pCursor){ + Fts3termCursor *pCsr = (Fts3termCursor *)pCursor; + Fts3Table *pFts3 = ((Fts3termTable *)pCursor->pVtab)->pFts3Tab; + int rc; + sqlite3_int64 v; + + /* Increment our pretend rowid value. */ + pCsr->iRowid++; + + /* Advance to the next term in the full-text index. */ + if( pCsr->csr.aDoclist==0 + || pCsr->pNext>=&pCsr->csr.aDoclist[pCsr->csr.nDoclist-1] + ){ + rc = sqlite3Fts3SegReaderStep(pFts3, &pCsr->csr); + if( rc!=SQLITE_ROW ){ + pCsr->isEof = 1; + return rc; + } + + pCsr->iCol = 0; + pCsr->iPos = 0; + pCsr->iDocid = 0; + pCsr->pNext = pCsr->csr.aDoclist; + + /* Read docid */ + pCsr->pNext += sqlite3Fts3GetVarint(pCsr->pNext, &pCsr->iDocid); + } + + pCsr->pNext += sqlite3Fts3GetVarint(pCsr->pNext, &v); + if( v==0 ){ + pCsr->pNext += sqlite3Fts3GetVarint(pCsr->pNext, &v); + pCsr->iDocid += v; + pCsr->pNext += sqlite3Fts3GetVarint(pCsr->pNext, &v); + pCsr->iCol = 0; + pCsr->iPos = 0; + } + + if( v==1 ){ + pCsr->pNext += sqlite3Fts3GetVarint(pCsr->pNext, &v); + pCsr->iCol += v; + pCsr->iPos = 0; + pCsr->pNext += sqlite3Fts3GetVarint(pCsr->pNext, &v); + } + + pCsr->iPos += (v - 2); + + return SQLITE_OK; +} + +/* +** xFilter - Initialize a cursor to point at the start of its data. +*/ +static int fts3termFilterMethod( + sqlite3_vtab_cursor *pCursor, /* The cursor used for this query */ + int idxNum, /* Strategy index */ + const char *idxStr, /* Unused */ + int nVal, /* Number of elements in apVal */ + sqlite3_value **apVal /* Arguments for the indexing scheme */ +){ + Fts3termCursor *pCsr = (Fts3termCursor *)pCursor; + Fts3termTable *p = (Fts3termTable *)pCursor->pVtab; + Fts3Table *pFts3 = p->pFts3Tab; + int rc; + + UNUSED_PARAMETER(nVal); + UNUSED_PARAMETER(idxNum); + UNUSED_PARAMETER(idxStr); + UNUSED_PARAMETER(apVal); + + assert( idxStr==0 && idxNum==0 ); + + /* In case this cursor is being reused, close and zero it. */ + testcase(pCsr->filter.zTerm); + sqlite3Fts3SegReaderFinish(&pCsr->csr); + memset(&pCsr->csr, 0, ((u8*)&pCsr[1]) - (u8*)&pCsr->csr); + + pCsr->filter.flags = FTS3_SEGMENT_REQUIRE_POS|FTS3_SEGMENT_IGNORE_EMPTY; + pCsr->filter.flags |= FTS3_SEGMENT_SCAN; + + rc = sqlite3Fts3SegReaderCursor(pFts3, p->iIndex, FTS3_SEGCURSOR_ALL, + pCsr->filter.zTerm, pCsr->filter.nTerm, 0, 1, &pCsr->csr + ); + if( rc==SQLITE_OK ){ + rc = sqlite3Fts3SegReaderStart(pFts3, &pCsr->csr, &pCsr->filter); + } + if( rc==SQLITE_OK ){ + rc = fts3termNextMethod(pCursor); + } + return rc; +} + +/* +** xEof - Return true if the cursor is at EOF, or false otherwise. +*/ +static int fts3termEofMethod(sqlite3_vtab_cursor *pCursor){ + Fts3termCursor *pCsr = (Fts3termCursor *)pCursor; + return pCsr->isEof; +} + +/* +** xColumn - Return a column value. +*/ +static int fts3termColumnMethod( + sqlite3_vtab_cursor *pCursor, /* Cursor to retrieve value from */ + sqlite3_context *pCtx, /* Context for sqlite3_result_xxx() calls */ + int iCol /* Index of column to read value from */ +){ + Fts3termCursor *p = (Fts3termCursor *)pCursor; + + assert( iCol>=0 && iCol<=3 ); + switch( iCol ){ + case 0: + sqlite3_result_text(pCtx, p->csr.zTerm, p->csr.nTerm, SQLITE_TRANSIENT); + break; + case 1: + sqlite3_result_int64(pCtx, p->iDocid); + break; + case 2: + sqlite3_result_int64(pCtx, p->iCol); + break; + default: + sqlite3_result_int64(pCtx, p->iPos); + break; + } + + return SQLITE_OK; +} + +/* +** xRowid - Return the current rowid for the cursor. +*/ +static int fts3termRowidMethod( + sqlite3_vtab_cursor *pCursor, /* Cursor to retrieve value from */ + sqlite_int64 *pRowid /* OUT: Rowid value */ +){ + Fts3termCursor *pCsr = (Fts3termCursor *)pCursor; + *pRowid = pCsr->iRowid; + return SQLITE_OK; +} + +/* +** Register the fts3term module with database connection db. Return SQLITE_OK +** if successful or an error code if sqlite3_create_module() fails. +*/ +int sqlite3Fts3InitTerm(sqlite3 *db){ + static const sqlite3_module fts3term_module = { + 0, /* iVersion */ + fts3termConnectMethod, /* xCreate */ + fts3termConnectMethod, /* xConnect */ + fts3termBestIndexMethod, /* xBestIndex */ + fts3termDisconnectMethod, /* xDisconnect */ + fts3termDisconnectMethod, /* xDestroy */ + fts3termOpenMethod, /* xOpen */ + fts3termCloseMethod, /* xClose */ + fts3termFilterMethod, /* xFilter */ + fts3termNextMethod, /* xNext */ + fts3termEofMethod, /* xEof */ + fts3termColumnMethod, /* xColumn */ + fts3termRowidMethod, /* xRowid */ + 0, /* xUpdate */ + 0, /* xBegin */ + 0, /* xSync */ + 0, /* xCommit */ + 0, /* xRollback */ + 0, /* xFindFunction */ + 0 /* xRename */ + }; + int rc; /* Return code */ + + rc = sqlite3_create_module(db, "fts4term", &fts3term_module, 0); + return rc; +} + +#endif +#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) */ diff --git a/ext/fts3/fts3_test.c b/ext/fts3/fts3_test.c new file mode 100644 index 0000000..72735f3 --- /dev/null +++ b/ext/fts3/fts3_test.c @@ -0,0 +1,324 @@ +/* +** 2011 Jun 13 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This file is not part of the production FTS code. It is only used for +** testing. It contains a Tcl command that can be used to test if a document +** matches an FTS NEAR expression. +*/ + +#include +#include +#include + +#ifdef SQLITE_TEST + +/* Required so that the "ifdef SQLITE_ENABLE_FTS3" below works */ +#include "fts3Int.h" + +#define NM_MAX_TOKEN 12 + +typedef struct NearPhrase NearPhrase; +typedef struct NearDocument NearDocument; +typedef struct NearToken NearToken; + +struct NearDocument { + int nToken; /* Length of token in bytes */ + NearToken *aToken; /* Token array */ +}; + +struct NearToken { + int n; /* Length of token in bytes */ + const char *z; /* Pointer to token string */ +}; + +struct NearPhrase { + int nNear; /* Preceding NEAR value */ + int nToken; /* Number of tokens in this phrase */ + NearToken aToken[NM_MAX_TOKEN]; /* Array of tokens in this phrase */ +}; + +static int nm_phrase_match( + NearPhrase *p, + NearToken *aToken +){ + int ii; + + for(ii=0; iinToken; ii++){ + NearToken *pToken = &p->aToken[ii]; + if( pToken->n>0 && pToken->z[pToken->n-1]=='*' ){ + if( aToken[ii].n<(pToken->n-1) ) return 0; + if( memcmp(aToken[ii].z, pToken->z, pToken->n-1) ) return 0; + }else{ + if( aToken[ii].n!=pToken->n ) return 0; + if( memcmp(aToken[ii].z, pToken->z, pToken->n) ) return 0; + } + } + + return 1; +} + +static int nm_near_chain( + int iDir, /* Direction to iterate through aPhrase[] */ + NearDocument *pDoc, /* Document to match against */ + int iPos, /* Position at which iPhrase was found */ + int nPhrase, /* Size of phrase array */ + NearPhrase *aPhrase, /* Phrase array */ + int iPhrase /* Index of phrase found */ +){ + int iStart; + int iStop; + int ii; + int nNear; + int iPhrase2; + NearPhrase *p; + NearPhrase *pPrev; + + assert( iDir==1 || iDir==-1 ); + + if( iDir==1 ){ + if( (iPhrase+1)==nPhrase ) return 1; + nNear = aPhrase[iPhrase+1].nNear; + }else{ + if( iPhrase==0 ) return 1; + nNear = aPhrase[iPhrase].nNear; + } + pPrev = &aPhrase[iPhrase]; + iPhrase2 = iPhrase+iDir; + p = &aPhrase[iPhrase2]; + + iStart = iPos - nNear - p->nToken; + iStop = iPos + nNear + pPrev->nToken; + + if( iStart<0 ) iStart = 0; + if( iStop > pDoc->nToken - p->nToken ) iStop = pDoc->nToken - p->nToken; + + for(ii=iStart; ii<=iStop; ii++){ + if( nm_phrase_match(p, &pDoc->aToken[ii]) ){ + if( nm_near_chain(iDir, pDoc, ii, nPhrase, aPhrase, iPhrase2) ) return 1; + } + } + + return 0; +} + +static int nm_match_count( + NearDocument *pDoc, /* Document to match against */ + int nPhrase, /* Size of phrase array */ + NearPhrase *aPhrase, /* Phrase array */ + int iPhrase /* Index of phrase to count matches for */ +){ + int nOcc = 0; + int ii; + NearPhrase *p = &aPhrase[iPhrase]; + + for(ii=0; ii<(pDoc->nToken + 1 - p->nToken); ii++){ + if( nm_phrase_match(p, &pDoc->aToken[ii]) ){ + /* Test forward NEAR chain (i>iPhrase) */ + if( 0==nm_near_chain(1, pDoc, ii, nPhrase, aPhrase, iPhrase) ) continue; + + /* Test reverse NEAR chain (iNM_MAX_TOKEN ){ + Tcl_AppendResult(interp, "Too many tokens in phrase", 0); + rc = TCL_ERROR; + goto near_match_out; + } + for(jj=0; jjz = Tcl_GetStringFromObj(apToken[jj], &pT->n); + } + aPhrase[ii].nToken = nToken; + } + for(ii=1; ii0)); + + near_match_out: + ckfree((char *)aPhrase); + ckfree((char *)doc.aToken); + return rc; +} + +/* +** Tclcmd: fts3_configure_incr_load ?CHUNKSIZE THRESHOLD? +** +** Normally, FTS uses hard-coded values to determine the minimum doclist +** size eligible for incremental loading, and the size of the chunks loaded +** when a doclist is incrementally loaded. This command allows the built-in +** values to be overridden for testing purposes. +** +** If present, the first argument is the chunksize in bytes to load doclists +** in. The second argument is the minimum doclist size in bytes to use +** incremental loading with. +** +** Whether or not the arguments are present, this command returns a list of +** two integers - the initial chunksize and threshold when the command is +** invoked. This can be used to restore the default behaviour after running +** tests. For example: +** +** # Override incr-load settings for testing: +** set cfg [fts3_configure_incr_load $new_chunksize $new_threshold] +** +** .... run tests .... +** +** # Restore initial incr-load settings: +** eval fts3_configure_incr_load $cfg +*/ +static int fts3_configure_incr_load_cmd( + ClientData clientData, + Tcl_Interp *interp, + int objc, + Tcl_Obj *CONST objv[] +){ +#ifdef SQLITE_ENABLE_FTS3 + extern int test_fts3_node_chunksize; + extern int test_fts3_node_chunk_threshold; + Tcl_Obj *pRet; + + if( objc!=1 && objc!=3 ){ + Tcl_WrongNumArgs(interp, 1, objv, "?CHUNKSIZE THRESHOLD?"); + return TCL_ERROR; + } + + pRet = Tcl_NewObj(); + Tcl_IncrRefCount(pRet); + Tcl_ListObjAppendElement( + interp, pRet, Tcl_NewIntObj(test_fts3_node_chunksize)); + Tcl_ListObjAppendElement( + interp, pRet, Tcl_NewIntObj(test_fts3_node_chunk_threshold)); + + if( objc==3 ){ + int iArg1; + int iArg2; + if( Tcl_GetIntFromObj(interp, objv[1], &iArg1) + || Tcl_GetIntFromObj(interp, objv[2], &iArg2) + ){ + Tcl_DecrRefCount(pRet); + return TCL_ERROR; + } + test_fts3_node_chunksize = iArg1; + test_fts3_node_chunk_threshold = iArg2; + } + + Tcl_SetObjResult(interp, pRet); + Tcl_DecrRefCount(pRet); +#endif + return TCL_OK; +} + +int Sqlitetestfts3_Init(Tcl_Interp *interp){ + Tcl_CreateObjCommand(interp, "fts3_near_match", fts3_near_match_cmd, 0, 0); + Tcl_CreateObjCommand(interp, + "fts3_configure_incr_load", fts3_configure_incr_load_cmd, 0, 0 + ); + return TCL_OK; +} +#endif /* ifdef SQLITE_TEST */ diff --git a/ext/fts3/fts3_tokenizer.c b/ext/fts3/fts3_tokenizer.c new file mode 100644 index 0000000..6494bb9 --- /dev/null +++ b/ext/fts3/fts3_tokenizer.c @@ -0,0 +1,489 @@ +/* +** 2007 June 22 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This is part of an SQLite module implementing full-text search. +** This particular file implements the generic tokenizer interface. +*/ + +/* +** The code in this file is only compiled if: +** +** * The FTS3 module is being built as an extension +** (in which case SQLITE_CORE is not defined), or +** +** * The FTS3 module is being built into the core of +** SQLite (in which case SQLITE_ENABLE_FTS3 is defined). +*/ +#include "fts3Int.h" +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) + +#include +#include + +/* +** Implementation of the SQL scalar function for accessing the underlying +** hash table. This function may be called as follows: +** +** SELECT (); +** SELECT (, ); +** +** where is the name passed as the second argument +** to the sqlite3Fts3InitHashTable() function (e.g. 'fts3_tokenizer'). +** +** If the argument is specified, it must be a blob value +** containing a pointer to be stored as the hash data corresponding +** to the string . If is not specified, then +** the string must already exist in the has table. Otherwise, +** an error is returned. +** +** Whether or not the argument is specified, the value returned +** is a blob containing the pointer stored as the hash data corresponding +** to string (after the hash-table is updated, if applicable). +*/ +static void scalarFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + Fts3Hash *pHash; + void *pPtr = 0; + const unsigned char *zName; + int nName; + + assert( argc==1 || argc==2 ); + + pHash = (Fts3Hash *)sqlite3_user_data(context); + + zName = sqlite3_value_text(argv[0]); + nName = sqlite3_value_bytes(argv[0])+1; + + if( argc==2 ){ + void *pOld; + int n = sqlite3_value_bytes(argv[1]); + if( n!=sizeof(pPtr) ){ + sqlite3_result_error(context, "argument type mismatch", -1); + return; + } + pPtr = *(void **)sqlite3_value_blob(argv[1]); + pOld = sqlite3Fts3HashInsert(pHash, (void *)zName, nName, pPtr); + if( pOld==pPtr ){ + sqlite3_result_error(context, "out of memory", -1); + return; + } + }else{ + pPtr = sqlite3Fts3HashFind(pHash, zName, nName); + if( !pPtr ){ + char *zErr = sqlite3_mprintf("unknown tokenizer: %s", zName); + sqlite3_result_error(context, zErr, -1); + sqlite3_free(zErr); + return; + } + } + + sqlite3_result_blob(context, (void *)&pPtr, sizeof(pPtr), SQLITE_TRANSIENT); +} + +int sqlite3Fts3IsIdChar(char c){ + static const char isFtsIdChar[] = { + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 0x */ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 1x */ + 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 2x */ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, /* 3x */ + 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /* 4x */ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 1, /* 5x */ + 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /* 6x */ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, /* 7x */ + }; + return (c&0x80 || isFtsIdChar[(int)(c)]); +} + +const char *sqlite3Fts3NextToken(const char *zStr, int *pn){ + const char *z1; + const char *z2 = 0; + + /* Find the start of the next token. */ + z1 = zStr; + while( z2==0 ){ + char c = *z1; + switch( c ){ + case '\0': return 0; /* No more tokens here */ + case '\'': + case '"': + case '`': { + z2 = z1; + while( *++z2 && (*z2!=c || *++z2==c) ); + break; + } + case '[': + z2 = &z1[1]; + while( *z2 && z2[0]!=']' ) z2++; + if( *z2 ) z2++; + break; + + default: + if( sqlite3Fts3IsIdChar(*z1) ){ + z2 = &z1[1]; + while( sqlite3Fts3IsIdChar(*z2) ) z2++; + }else{ + z1++; + } + } + } + + *pn = (int)(z2-z1); + return z1; +} + +int sqlite3Fts3InitTokenizer( + Fts3Hash *pHash, /* Tokenizer hash table */ + const char *zArg, /* Tokenizer name */ + sqlite3_tokenizer **ppTok, /* OUT: Tokenizer (if applicable) */ + char **pzErr /* OUT: Set to malloced error message */ +){ + int rc; + char *z = (char *)zArg; + int n = 0; + char *zCopy; + char *zEnd; /* Pointer to nul-term of zCopy */ + sqlite3_tokenizer_module *m; + + zCopy = sqlite3_mprintf("%s", zArg); + if( !zCopy ) return SQLITE_NOMEM; + zEnd = &zCopy[strlen(zCopy)]; + + z = (char *)sqlite3Fts3NextToken(zCopy, &n); + z[n] = '\0'; + sqlite3Fts3Dequote(z); + + m = (sqlite3_tokenizer_module *)sqlite3Fts3HashFind(pHash,z,(int)strlen(z)+1); + if( !m ){ + *pzErr = sqlite3_mprintf("unknown tokenizer: %s", z); + rc = SQLITE_ERROR; + }else{ + char const **aArg = 0; + int iArg = 0; + z = &z[n+1]; + while( zxCreate(iArg, aArg, ppTok); + assert( rc!=SQLITE_OK || *ppTok ); + if( rc!=SQLITE_OK ){ + *pzErr = sqlite3_mprintf("unknown tokenizer"); + }else{ + (*ppTok)->pModule = m; + } + sqlite3_free((void *)aArg); + } + + sqlite3_free(zCopy); + return rc; +} + + +#ifdef SQLITE_TEST + +#include +#include + +/* +** Implementation of a special SQL scalar function for testing tokenizers +** designed to be used in concert with the Tcl testing framework. This +** function must be called with two arguments: +** +** SELECT (, ); +** SELECT (, ); +** +** where is the name passed as the second argument +** to the sqlite3Fts3InitHashTable() function (e.g. 'fts3_tokenizer') +** concatenated with the string '_test' (e.g. 'fts3_tokenizer_test'). +** +** The return value is a string that may be interpreted as a Tcl +** list. For each token in the , three elements are +** added to the returned list. The first is the token position, the +** second is the token text (folded, stemmed, etc.) and the third is the +** substring of associated with the token. For example, +** using the built-in "simple" tokenizer: +** +** SELECT fts_tokenizer_test('simple', 'I don't see how'); +** +** will return the string: +** +** "{0 i I 1 dont don't 2 see see 3 how how}" +** +*/ +static void testFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + Fts3Hash *pHash; + sqlite3_tokenizer_module *p; + sqlite3_tokenizer *pTokenizer = 0; + sqlite3_tokenizer_cursor *pCsr = 0; + + const char *zErr = 0; + + const char *zName; + int nName; + const char *zInput; + int nInput; + + const char *zArg = 0; + + const char *zToken; + int nToken; + int iStart; + int iEnd; + int iPos; + + Tcl_Obj *pRet; + + assert( argc==2 || argc==3 ); + + nName = sqlite3_value_bytes(argv[0]); + zName = (const char *)sqlite3_value_text(argv[0]); + nInput = sqlite3_value_bytes(argv[argc-1]); + zInput = (const char *)sqlite3_value_text(argv[argc-1]); + + if( argc==3 ){ + zArg = (const char *)sqlite3_value_text(argv[1]); + } + + pHash = (Fts3Hash *)sqlite3_user_data(context); + p = (sqlite3_tokenizer_module *)sqlite3Fts3HashFind(pHash, zName, nName+1); + + if( !p ){ + char *zErr = sqlite3_mprintf("unknown tokenizer: %s", zName); + sqlite3_result_error(context, zErr, -1); + sqlite3_free(zErr); + return; + } + + pRet = Tcl_NewObj(); + Tcl_IncrRefCount(pRet); + + if( SQLITE_OK!=p->xCreate(zArg ? 1 : 0, &zArg, &pTokenizer) ){ + zErr = "error in xCreate()"; + goto finish; + } + pTokenizer->pModule = p; + if( SQLITE_OK!=p->xOpen(pTokenizer, zInput, nInput, &pCsr) ){ + zErr = "error in xOpen()"; + goto finish; + } + pCsr->pTokenizer = pTokenizer; + + while( SQLITE_OK==p->xNext(pCsr, &zToken, &nToken, &iStart, &iEnd, &iPos) ){ + Tcl_ListObjAppendElement(0, pRet, Tcl_NewIntObj(iPos)); + Tcl_ListObjAppendElement(0, pRet, Tcl_NewStringObj(zToken, nToken)); + zToken = &zInput[iStart]; + nToken = iEnd-iStart; + Tcl_ListObjAppendElement(0, pRet, Tcl_NewStringObj(zToken, nToken)); + } + + if( SQLITE_OK!=p->xClose(pCsr) ){ + zErr = "error in xClose()"; + goto finish; + } + if( SQLITE_OK!=p->xDestroy(pTokenizer) ){ + zErr = "error in xDestroy()"; + goto finish; + } + +finish: + if( zErr ){ + sqlite3_result_error(context, zErr, -1); + }else{ + sqlite3_result_text(context, Tcl_GetString(pRet), -1, SQLITE_TRANSIENT); + } + Tcl_DecrRefCount(pRet); +} + +static +int registerTokenizer( + sqlite3 *db, + char *zName, + const sqlite3_tokenizer_module *p +){ + int rc; + sqlite3_stmt *pStmt; + const char zSql[] = "SELECT fts3_tokenizer(?, ?)"; + + rc = sqlite3_prepare_v2(db, zSql, -1, &pStmt, 0); + if( rc!=SQLITE_OK ){ + return rc; + } + + sqlite3_bind_text(pStmt, 1, zName, -1, SQLITE_STATIC); + sqlite3_bind_blob(pStmt, 2, &p, sizeof(p), SQLITE_STATIC); + sqlite3_step(pStmt); + + return sqlite3_finalize(pStmt); +} + +static +int queryTokenizer( + sqlite3 *db, + char *zName, + const sqlite3_tokenizer_module **pp +){ + int rc; + sqlite3_stmt *pStmt; + const char zSql[] = "SELECT fts3_tokenizer(?)"; + + *pp = 0; + rc = sqlite3_prepare_v2(db, zSql, -1, &pStmt, 0); + if( rc!=SQLITE_OK ){ + return rc; + } + + sqlite3_bind_text(pStmt, 1, zName, -1, SQLITE_STATIC); + if( SQLITE_ROW==sqlite3_step(pStmt) ){ + if( sqlite3_column_type(pStmt, 0)==SQLITE_BLOB ){ + memcpy((void *)pp, sqlite3_column_blob(pStmt, 0), sizeof(*pp)); + } + } + + return sqlite3_finalize(pStmt); +} + +void sqlite3Fts3SimpleTokenizerModule(sqlite3_tokenizer_module const**ppModule); + +/* +** Implementation of the scalar function fts3_tokenizer_internal_test(). +** This function is used for testing only, it is not included in the +** build unless SQLITE_TEST is defined. +** +** The purpose of this is to test that the fts3_tokenizer() function +** can be used as designed by the C-code in the queryTokenizer and +** registerTokenizer() functions above. These two functions are repeated +** in the README.tokenizer file as an example, so it is important to +** test them. +** +** To run the tests, evaluate the fts3_tokenizer_internal_test() scalar +** function with no arguments. An assert() will fail if a problem is +** detected. i.e.: +** +** SELECT fts3_tokenizer_internal_test(); +** +*/ +static void intTestFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + int rc; + const sqlite3_tokenizer_module *p1; + const sqlite3_tokenizer_module *p2; + sqlite3 *db = (sqlite3 *)sqlite3_user_data(context); + + UNUSED_PARAMETER(argc); + UNUSED_PARAMETER(argv); + + /* Test the query function */ + sqlite3Fts3SimpleTokenizerModule(&p1); + rc = queryTokenizer(db, "simple", &p2); + assert( rc==SQLITE_OK ); + assert( p1==p2 ); + rc = queryTokenizer(db, "nosuchtokenizer", &p2); + assert( rc==SQLITE_ERROR ); + assert( p2==0 ); + assert( 0==strcmp(sqlite3_errmsg(db), "unknown tokenizer: nosuchtokenizer") ); + + /* Test the storage function */ + rc = registerTokenizer(db, "nosuchtokenizer", p1); + assert( rc==SQLITE_OK ); + rc = queryTokenizer(db, "nosuchtokenizer", &p2); + assert( rc==SQLITE_OK ); + assert( p2==p1 ); + + sqlite3_result_text(context, "ok", -1, SQLITE_STATIC); +} + +#endif + +/* +** Set up SQL objects in database db used to access the contents of +** the hash table pointed to by argument pHash. The hash table must +** been initialised to use string keys, and to take a private copy +** of the key when a value is inserted. i.e. by a call similar to: +** +** sqlite3Fts3HashInit(pHash, FTS3_HASH_STRING, 1); +** +** This function adds a scalar function (see header comment above +** scalarFunc() in this file for details) and, if ENABLE_TABLE is +** defined at compilation time, a temporary virtual table (see header +** comment above struct HashTableVtab) to the database schema. Both +** provide read/write access to the contents of *pHash. +** +** The third argument to this function, zName, is used as the name +** of both the scalar and, if created, the virtual table. +*/ +int sqlite3Fts3InitHashTable( + sqlite3 *db, + Fts3Hash *pHash, + const char *zName +){ + int rc = SQLITE_OK; + void *p = (void *)pHash; + const int any = SQLITE_ANY; + +#ifdef SQLITE_TEST + char *zTest = 0; + char *zTest2 = 0; + void *pdb = (void *)db; + zTest = sqlite3_mprintf("%s_test", zName); + zTest2 = sqlite3_mprintf("%s_internal_test", zName); + if( !zTest || !zTest2 ){ + rc = SQLITE_NOMEM; + } +#endif + + if( SQLITE_OK==rc ){ + rc = sqlite3_create_function(db, zName, 1, any, p, scalarFunc, 0, 0); + } + if( SQLITE_OK==rc ){ + rc = sqlite3_create_function(db, zName, 2, any, p, scalarFunc, 0, 0); + } +#ifdef SQLITE_TEST + if( SQLITE_OK==rc ){ + rc = sqlite3_create_function(db, zTest, 2, any, p, testFunc, 0, 0); + } + if( SQLITE_OK==rc ){ + rc = sqlite3_create_function(db, zTest, 3, any, p, testFunc, 0, 0); + } + if( SQLITE_OK==rc ){ + rc = sqlite3_create_function(db, zTest2, 0, any, pdb, intTestFunc, 0, 0); + } +#endif + +#ifdef SQLITE_TEST + sqlite3_free(zTest); + sqlite3_free(zTest2); +#endif + + return rc; +} + +#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) */ diff --git a/ext/fts3/fts3_tokenizer.h b/ext/fts3/fts3_tokenizer.h new file mode 100644 index 0000000..6156445 --- /dev/null +++ b/ext/fts3/fts3_tokenizer.h @@ -0,0 +1,152 @@ +/* +** 2006 July 10 +** +** The author disclaims copyright to this source code. +** +************************************************************************* +** Defines the interface to tokenizers used by fulltext-search. There +** are three basic components: +** +** sqlite3_tokenizer_module is a singleton defining the tokenizer +** interface functions. This is essentially the class structure for +** tokenizers. +** +** sqlite3_tokenizer is used to define a particular tokenizer, perhaps +** including customization information defined at creation time. +** +** sqlite3_tokenizer_cursor is generated by a tokenizer to generate +** tokens from a particular input. +*/ +#ifndef _FTS3_TOKENIZER_H_ +#define _FTS3_TOKENIZER_H_ + +/* TODO(shess) Only used for SQLITE_OK and SQLITE_DONE at this time. +** If tokenizers are to be allowed to call sqlite3_*() functions, then +** we will need a way to register the API consistently. +*/ +#include "sqlite3.h" + +/* +** Structures used by the tokenizer interface. When a new tokenizer +** implementation is registered, the caller provides a pointer to +** an sqlite3_tokenizer_module containing pointers to the callback +** functions that make up an implementation. +** +** When an fts3 table is created, it passes any arguments passed to +** the tokenizer clause of the CREATE VIRTUAL TABLE statement to the +** sqlite3_tokenizer_module.xCreate() function of the requested tokenizer +** implementation. The xCreate() function in turn returns an +** sqlite3_tokenizer structure representing the specific tokenizer to +** be used for the fts3 table (customized by the tokenizer clause arguments). +** +** To tokenize an input buffer, the sqlite3_tokenizer_module.xOpen() +** method is called. It returns an sqlite3_tokenizer_cursor object +** that may be used to tokenize a specific input buffer based on +** the tokenization rules supplied by a specific sqlite3_tokenizer +** object. +*/ +typedef struct sqlite3_tokenizer_module sqlite3_tokenizer_module; +typedef struct sqlite3_tokenizer sqlite3_tokenizer; +typedef struct sqlite3_tokenizer_cursor sqlite3_tokenizer_cursor; + +struct sqlite3_tokenizer_module { + + /* + ** Structure version. Should always be set to 0. + */ + int iVersion; + + /* + ** Create a new tokenizer. The values in the argv[] array are the + ** arguments passed to the "tokenizer" clause of the CREATE VIRTUAL + ** TABLE statement that created the fts3 table. For example, if + ** the following SQL is executed: + ** + ** CREATE .. USING fts3( ... , tokenizer arg1 arg2) + ** + ** then argc is set to 2, and the argv[] array contains pointers + ** to the strings "arg1" and "arg2". + ** + ** This method should return either SQLITE_OK (0), or an SQLite error + ** code. If SQLITE_OK is returned, then *ppTokenizer should be set + ** to point at the newly created tokenizer structure. The generic + ** sqlite3_tokenizer.pModule variable should not be initialised by + ** this callback. The caller will do so. + */ + int (*xCreate)( + int argc, /* Size of argv array */ + const char *const*argv, /* Tokenizer argument strings */ + sqlite3_tokenizer **ppTokenizer /* OUT: Created tokenizer */ + ); + + /* + ** Destroy an existing tokenizer. The fts3 module calls this method + ** exactly once for each successful call to xCreate(). + */ + int (*xDestroy)(sqlite3_tokenizer *pTokenizer); + + /* + ** Create a tokenizer cursor to tokenize an input buffer. The caller + ** is responsible for ensuring that the input buffer remains valid + ** until the cursor is closed (using the xClose() method). + */ + int (*xOpen)( + sqlite3_tokenizer *pTokenizer, /* Tokenizer object */ + const char *pInput, int nBytes, /* Input buffer */ + sqlite3_tokenizer_cursor **ppCursor /* OUT: Created tokenizer cursor */ + ); + + /* + ** Destroy an existing tokenizer cursor. The fts3 module calls this + ** method exactly once for each successful call to xOpen(). + */ + int (*xClose)(sqlite3_tokenizer_cursor *pCursor); + + /* + ** Retrieve the next token from the tokenizer cursor pCursor. This + ** method should either return SQLITE_OK and set the values of the + ** "OUT" variables identified below, or SQLITE_DONE to indicate that + ** the end of the buffer has been reached, or an SQLite error code. + ** + ** *ppToken should be set to point at a buffer containing the + ** normalized version of the token (i.e. after any case-folding and/or + ** stemming has been performed). *pnBytes should be set to the length + ** of this buffer in bytes. The input text that generated the token is + ** identified by the byte offsets returned in *piStartOffset and + ** *piEndOffset. *piStartOffset should be set to the index of the first + ** byte of the token in the input buffer. *piEndOffset should be set + ** to the index of the first byte just past the end of the token in + ** the input buffer. + ** + ** The buffer *ppToken is set to point at is managed by the tokenizer + ** implementation. It is only required to be valid until the next call + ** to xNext() or xClose(). + */ + /* TODO(shess) current implementation requires pInput to be + ** nul-terminated. This should either be fixed, or pInput/nBytes + ** should be converted to zInput. + */ + int (*xNext)( + sqlite3_tokenizer_cursor *pCursor, /* Tokenizer cursor */ + const char **ppToken, int *pnBytes, /* OUT: Normalized text for token */ + int *piStartOffset, /* OUT: Byte offset of token in input buffer */ + int *piEndOffset, /* OUT: Byte offset of end of token in input buffer */ + int *piPosition /* OUT: Number of tokens returned before this one */ + ); +}; + +struct sqlite3_tokenizer { + const sqlite3_tokenizer_module *pModule; /* The module for this tokenizer */ + /* Tokenizer implementations will typically add additional fields */ +}; + +struct sqlite3_tokenizer_cursor { + sqlite3_tokenizer *pTokenizer; /* Tokenizer for this cursor. */ + /* Tokenizer implementations will typically add additional fields */ +}; + +int fts3_global_term_cnt(int iTerm, int iCol); +int fts3_term_cnt(int iTerm, int iCol); + + +#endif /* _FTS3_TOKENIZER_H_ */ diff --git a/ext/fts3/fts3_tokenizer1.c b/ext/fts3/fts3_tokenizer1.c new file mode 100644 index 0000000..d11a499 --- /dev/null +++ b/ext/fts3/fts3_tokenizer1.c @@ -0,0 +1,233 @@ +/* +** 2006 Oct 10 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** Implementation of the "simple" full-text-search tokenizer. +*/ + +/* +** The code in this file is only compiled if: +** +** * The FTS3 module is being built as an extension +** (in which case SQLITE_CORE is not defined), or +** +** * The FTS3 module is being built into the core of +** SQLite (in which case SQLITE_ENABLE_FTS3 is defined). +*/ +#include "fts3Int.h" +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) + +#include +#include +#include +#include + +#include "fts3_tokenizer.h" + +typedef struct simple_tokenizer { + sqlite3_tokenizer base; + char delim[128]; /* flag ASCII delimiters */ +} simple_tokenizer; + +typedef struct simple_tokenizer_cursor { + sqlite3_tokenizer_cursor base; + const char *pInput; /* input we are tokenizing */ + int nBytes; /* size of the input */ + int iOffset; /* current position in pInput */ + int iToken; /* index of next token to be returned */ + char *pToken; /* storage for current token */ + int nTokenAllocated; /* space allocated to zToken buffer */ +} simple_tokenizer_cursor; + + +static int simpleDelim(simple_tokenizer *t, unsigned char c){ + return c<0x80 && t->delim[c]; +} +static int fts3_isalnum(int x){ + return (x>='0' && x<='9') || (x>='A' && x<='Z') || (x>='a' && x<='z'); +} + +/* +** Create a new tokenizer instance. +*/ +static int simpleCreate( + int argc, const char * const *argv, + sqlite3_tokenizer **ppTokenizer +){ + simple_tokenizer *t; + + t = (simple_tokenizer *) sqlite3_malloc(sizeof(*t)); + if( t==NULL ) return SQLITE_NOMEM; + memset(t, 0, sizeof(*t)); + + /* TODO(shess) Delimiters need to remain the same from run to run, + ** else we need to reindex. One solution would be a meta-table to + ** track such information in the database, then we'd only want this + ** information on the initial create. + */ + if( argc>1 ){ + int i, n = (int)strlen(argv[1]); + for(i=0; i=0x80 ){ + sqlite3_free(t); + return SQLITE_ERROR; + } + t->delim[ch] = 1; + } + } else { + /* Mark non-alphanumeric ASCII characters as delimiters */ + int i; + for(i=1; i<0x80; i++){ + t->delim[i] = !fts3_isalnum(i) ? -1 : 0; + } + } + + *ppTokenizer = &t->base; + return SQLITE_OK; +} + +/* +** Destroy a tokenizer +*/ +static int simpleDestroy(sqlite3_tokenizer *pTokenizer){ + sqlite3_free(pTokenizer); + return SQLITE_OK; +} + +/* +** Prepare to begin tokenizing a particular string. The input +** string to be tokenized is pInput[0..nBytes-1]. A cursor +** used to incrementally tokenize this string is returned in +** *ppCursor. +*/ +static int simpleOpen( + sqlite3_tokenizer *pTokenizer, /* The tokenizer */ + const char *pInput, int nBytes, /* String to be tokenized */ + sqlite3_tokenizer_cursor **ppCursor /* OUT: Tokenization cursor */ +){ + simple_tokenizer_cursor *c; + + UNUSED_PARAMETER(pTokenizer); + + c = (simple_tokenizer_cursor *) sqlite3_malloc(sizeof(*c)); + if( c==NULL ) return SQLITE_NOMEM; + + c->pInput = pInput; + if( pInput==0 ){ + c->nBytes = 0; + }else if( nBytes<0 ){ + c->nBytes = (int)strlen(pInput); + }else{ + c->nBytes = nBytes; + } + c->iOffset = 0; /* start tokenizing at the beginning */ + c->iToken = 0; + c->pToken = NULL; /* no space allocated, yet. */ + c->nTokenAllocated = 0; + + *ppCursor = &c->base; + return SQLITE_OK; +} + +/* +** Close a tokenization cursor previously opened by a call to +** simpleOpen() above. +*/ +static int simpleClose(sqlite3_tokenizer_cursor *pCursor){ + simple_tokenizer_cursor *c = (simple_tokenizer_cursor *) pCursor; + sqlite3_free(c->pToken); + sqlite3_free(c); + return SQLITE_OK; +} + +/* +** Extract the next token from a tokenization cursor. The cursor must +** have been opened by a prior call to simpleOpen(). +*/ +static int simpleNext( + sqlite3_tokenizer_cursor *pCursor, /* Cursor returned by simpleOpen */ + const char **ppToken, /* OUT: *ppToken is the token text */ + int *pnBytes, /* OUT: Number of bytes in token */ + int *piStartOffset, /* OUT: Starting offset of token */ + int *piEndOffset, /* OUT: Ending offset of token */ + int *piPosition /* OUT: Position integer of token */ +){ + simple_tokenizer_cursor *c = (simple_tokenizer_cursor *) pCursor; + simple_tokenizer *t = (simple_tokenizer *) pCursor->pTokenizer; + unsigned char *p = (unsigned char *)c->pInput; + + while( c->iOffsetnBytes ){ + int iStartOffset; + + /* Scan past delimiter characters */ + while( c->iOffsetnBytes && simpleDelim(t, p[c->iOffset]) ){ + c->iOffset++; + } + + /* Count non-delimiter characters. */ + iStartOffset = c->iOffset; + while( c->iOffsetnBytes && !simpleDelim(t, p[c->iOffset]) ){ + c->iOffset++; + } + + if( c->iOffset>iStartOffset ){ + int i, n = c->iOffset-iStartOffset; + if( n>c->nTokenAllocated ){ + char *pNew; + c->nTokenAllocated = n+20; + pNew = sqlite3_realloc(c->pToken, c->nTokenAllocated); + if( !pNew ) return SQLITE_NOMEM; + c->pToken = pNew; + } + for(i=0; ipToken[i] = (char)((ch>='A' && ch<='Z') ? ch-'A'+'a' : ch); + } + *ppToken = c->pToken; + *pnBytes = n; + *piStartOffset = iStartOffset; + *piEndOffset = c->iOffset; + *piPosition = c->iToken++; + + return SQLITE_OK; + } + } + return SQLITE_DONE; +} + +/* +** The set of routines that implement the simple tokenizer +*/ +static const sqlite3_tokenizer_module simpleTokenizerModule = { + 0, + simpleCreate, + simpleDestroy, + simpleOpen, + simpleClose, + simpleNext, +}; + +/* +** Allocate a new simple tokenizer. Return a pointer to the new +** tokenizer in *ppModule +*/ +void sqlite3Fts3SimpleTokenizerModule( + sqlite3_tokenizer_module const**ppModule +){ + *ppModule = &simpleTokenizerModule; +} + +#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) */ diff --git a/ext/fts3/fts3_write.c b/ext/fts3/fts3_write.c new file mode 100644 index 0000000..2904a9a --- /dev/null +++ b/ext/fts3/fts3_write.c @@ -0,0 +1,3397 @@ +/* +** 2009 Oct 23 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +****************************************************************************** +** +** This file is part of the SQLite FTS3 extension module. Specifically, +** this file contains code to insert, update and delete rows from FTS3 +** tables. It also contains code to merge FTS3 b-tree segments. Some +** of the sub-routines used to merge segments are also used by the query +** code in fts3.c. +*/ + +#include "fts3Int.h" +#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) + +#include +#include +#include + +/* +** When full-text index nodes are loaded from disk, the buffer that they +** are loaded into has the following number of bytes of padding at the end +** of it. i.e. if a full-text index node is 900 bytes in size, then a buffer +** of 920 bytes is allocated for it. +** +** This means that if we have a pointer into a buffer containing node data, +** it is always safe to read up to two varints from it without risking an +** overread, even if the node data is corrupted. +*/ +#define FTS3_NODE_PADDING (FTS3_VARINT_MAX*2) + +/* +** Under certain circumstances, b-tree nodes (doclists) can be loaded into +** memory incrementally instead of all at once. This can be a big performance +** win (reduced IO and CPU) if SQLite stops calling the virtual table xNext() +** method before retrieving all query results (as may happen, for example, +** if a query has a LIMIT clause). +** +** Incremental loading is used for b-tree nodes FTS3_NODE_CHUNK_THRESHOLD +** bytes and larger. Nodes are loaded in chunks of FTS3_NODE_CHUNKSIZE bytes. +** The code is written so that the hard lower-limit for each of these values +** is 1. Clearly such small values would be inefficient, but can be useful +** for testing purposes. +** +** If this module is built with SQLITE_TEST defined, these constants may +** be overridden at runtime for testing purposes. File fts3_test.c contains +** a Tcl interface to read and write the values. +*/ +#ifdef SQLITE_TEST +int test_fts3_node_chunksize = (4*1024); +int test_fts3_node_chunk_threshold = (4*1024)*4; +# define FTS3_NODE_CHUNKSIZE test_fts3_node_chunksize +# define FTS3_NODE_CHUNK_THRESHOLD test_fts3_node_chunk_threshold +#else +# define FTS3_NODE_CHUNKSIZE (4*1024) +# define FTS3_NODE_CHUNK_THRESHOLD (FTS3_NODE_CHUNKSIZE*4) +#endif + +typedef struct PendingList PendingList; +typedef struct SegmentNode SegmentNode; +typedef struct SegmentWriter SegmentWriter; + +/* +** An instance of the following data structure is used to build doclists +** incrementally. See function fts3PendingListAppend() for details. +*/ +struct PendingList { + int nData; + char *aData; + int nSpace; + sqlite3_int64 iLastDocid; + sqlite3_int64 iLastCol; + sqlite3_int64 iLastPos; +}; + + +/* +** Each cursor has a (possibly empty) linked list of the following objects. +*/ +struct Fts3DeferredToken { + Fts3PhraseToken *pToken; /* Pointer to corresponding expr token */ + int iCol; /* Column token must occur in */ + Fts3DeferredToken *pNext; /* Next in list of deferred tokens */ + PendingList *pList; /* Doclist is assembled here */ +}; + +/* +** An instance of this structure is used to iterate through the terms on +** a contiguous set of segment b-tree leaf nodes. Although the details of +** this structure are only manipulated by code in this file, opaque handles +** of type Fts3SegReader* are also used by code in fts3.c to iterate through +** terms when querying the full-text index. See functions: +** +** sqlite3Fts3SegReaderNew() +** sqlite3Fts3SegReaderFree() +** sqlite3Fts3SegReaderIterate() +** +** Methods used to manipulate Fts3SegReader structures: +** +** fts3SegReaderNext() +** fts3SegReaderFirstDocid() +** fts3SegReaderNextDocid() +*/ +struct Fts3SegReader { + int iIdx; /* Index within level, or 0x7FFFFFFF for PT */ + + sqlite3_int64 iStartBlock; /* Rowid of first leaf block to traverse */ + sqlite3_int64 iLeafEndBlock; /* Rowid of final leaf block to traverse */ + sqlite3_int64 iEndBlock; /* Rowid of final block in segment (or 0) */ + sqlite3_int64 iCurrentBlock; /* Current leaf block (or 0) */ + + char *aNode; /* Pointer to node data (or NULL) */ + int nNode; /* Size of buffer at aNode (or 0) */ + int nPopulate; /* If >0, bytes of buffer aNode[] loaded */ + sqlite3_blob *pBlob; /* If not NULL, blob handle to read node */ + + Fts3HashElem **ppNextElem; + + /* Variables set by fts3SegReaderNext(). These may be read directly + ** by the caller. They are valid from the time SegmentReaderNew() returns + ** until SegmentReaderNext() returns something other than SQLITE_OK + ** (i.e. SQLITE_DONE). + */ + int nTerm; /* Number of bytes in current term */ + char *zTerm; /* Pointer to current term */ + int nTermAlloc; /* Allocated size of zTerm buffer */ + char *aDoclist; /* Pointer to doclist of current entry */ + int nDoclist; /* Size of doclist in current entry */ + + /* The following variables are used by fts3SegReaderNextDocid() to iterate + ** through the current doclist (aDoclist/nDoclist). + */ + char *pOffsetList; + int nOffsetList; /* For descending pending seg-readers only */ + sqlite3_int64 iDocid; +}; + +#define fts3SegReaderIsPending(p) ((p)->ppNextElem!=0) +#define fts3SegReaderIsRootOnly(p) ((p)->aNode==(char *)&(p)[1]) + +/* +** An instance of this structure is used to create a segment b-tree in the +** database. The internal details of this type are only accessed by the +** following functions: +** +** fts3SegWriterAdd() +** fts3SegWriterFlush() +** fts3SegWriterFree() +*/ +struct SegmentWriter { + SegmentNode *pTree; /* Pointer to interior tree structure */ + sqlite3_int64 iFirst; /* First slot in %_segments written */ + sqlite3_int64 iFree; /* Next free slot in %_segments */ + char *zTerm; /* Pointer to previous term buffer */ + int nTerm; /* Number of bytes in zTerm */ + int nMalloc; /* Size of malloc'd buffer at zMalloc */ + char *zMalloc; /* Malloc'd space (possibly) used for zTerm */ + int nSize; /* Size of allocation at aData */ + int nData; /* Bytes of data in aData */ + char *aData; /* Pointer to block from malloc() */ +}; + +/* +** Type SegmentNode is used by the following three functions to create +** the interior part of the segment b+-tree structures (everything except +** the leaf nodes). These functions and type are only ever used by code +** within the fts3SegWriterXXX() family of functions described above. +** +** fts3NodeAddTerm() +** fts3NodeWrite() +** fts3NodeFree() +** +** When a b+tree is written to the database (either as a result of a merge +** or the pending-terms table being flushed), leaves are written into the +** database file as soon as they are completely populated. The interior of +** the tree is assembled in memory and written out only once all leaves have +** been populated and stored. This is Ok, as the b+-tree fanout is usually +** very large, meaning that the interior of the tree consumes relatively +** little memory. +*/ +struct SegmentNode { + SegmentNode *pParent; /* Parent node (or NULL for root node) */ + SegmentNode *pRight; /* Pointer to right-sibling */ + SegmentNode *pLeftmost; /* Pointer to left-most node of this depth */ + int nEntry; /* Number of terms written to node so far */ + char *zTerm; /* Pointer to previous term buffer */ + int nTerm; /* Number of bytes in zTerm */ + int nMalloc; /* Size of malloc'd buffer at zMalloc */ + char *zMalloc; /* Malloc'd space (possibly) used for zTerm */ + int nData; /* Bytes of valid data so far */ + char *aData; /* Node data */ +}; + +/* +** Valid values for the second argument to fts3SqlStmt(). +*/ +#define SQL_DELETE_CONTENT 0 +#define SQL_IS_EMPTY 1 +#define SQL_DELETE_ALL_CONTENT 2 +#define SQL_DELETE_ALL_SEGMENTS 3 +#define SQL_DELETE_ALL_SEGDIR 4 +#define SQL_DELETE_ALL_DOCSIZE 5 +#define SQL_DELETE_ALL_STAT 6 +#define SQL_SELECT_CONTENT_BY_ROWID 7 +#define SQL_NEXT_SEGMENT_INDEX 8 +#define SQL_INSERT_SEGMENTS 9 +#define SQL_NEXT_SEGMENTS_ID 10 +#define SQL_INSERT_SEGDIR 11 +#define SQL_SELECT_LEVEL 12 +#define SQL_SELECT_LEVEL_RANGE 13 +#define SQL_SELECT_LEVEL_COUNT 14 +#define SQL_SELECT_SEGDIR_MAX_LEVEL 15 +#define SQL_DELETE_SEGDIR_LEVEL 16 +#define SQL_DELETE_SEGMENTS_RANGE 17 +#define SQL_CONTENT_INSERT 18 +#define SQL_DELETE_DOCSIZE 19 +#define SQL_REPLACE_DOCSIZE 20 +#define SQL_SELECT_DOCSIZE 21 +#define SQL_SELECT_DOCTOTAL 22 +#define SQL_REPLACE_DOCTOTAL 23 + +#define SQL_SELECT_ALL_PREFIX_LEVEL 24 +#define SQL_DELETE_ALL_TERMS_SEGDIR 25 + +#define SQL_DELETE_SEGDIR_RANGE 26 + +/* +** This function is used to obtain an SQLite prepared statement handle +** for the statement identified by the second argument. If successful, +** *pp is set to the requested statement handle and SQLITE_OK returned. +** Otherwise, an SQLite error code is returned and *pp is set to 0. +** +** If argument apVal is not NULL, then it must point to an array with +** at least as many entries as the requested statement has bound +** parameters. The values are bound to the statements parameters before +** returning. +*/ +static int fts3SqlStmt( + Fts3Table *p, /* Virtual table handle */ + int eStmt, /* One of the SQL_XXX constants above */ + sqlite3_stmt **pp, /* OUT: Statement handle */ + sqlite3_value **apVal /* Values to bind to statement */ +){ + const char *azSql[] = { +/* 0 */ "DELETE FROM %Q.'%q_content' WHERE rowid = ?", +/* 1 */ "SELECT NOT EXISTS(SELECT docid FROM %Q.'%q_content' WHERE rowid!=?)", +/* 2 */ "DELETE FROM %Q.'%q_content'", +/* 3 */ "DELETE FROM %Q.'%q_segments'", +/* 4 */ "DELETE FROM %Q.'%q_segdir'", +/* 5 */ "DELETE FROM %Q.'%q_docsize'", +/* 6 */ "DELETE FROM %Q.'%q_stat'", +/* 7 */ "SELECT %s WHERE rowid=?", +/* 8 */ "SELECT (SELECT max(idx) FROM %Q.'%q_segdir' WHERE level = ?) + 1", +/* 9 */ "INSERT INTO %Q.'%q_segments'(blockid, block) VALUES(?, ?)", +/* 10 */ "SELECT coalesce((SELECT max(blockid) FROM %Q.'%q_segments') + 1, 1)", +/* 11 */ "INSERT INTO %Q.'%q_segdir' VALUES(?,?,?,?,?,?)", + + /* Return segments in order from oldest to newest.*/ +/* 12 */ "SELECT idx, start_block, leaves_end_block, end_block, root " + "FROM %Q.'%q_segdir' WHERE level = ? ORDER BY idx ASC", +/* 13 */ "SELECT idx, start_block, leaves_end_block, end_block, root " + "FROM %Q.'%q_segdir' WHERE level BETWEEN ? AND ?" + "ORDER BY level DESC, idx ASC", + +/* 14 */ "SELECT count(*) FROM %Q.'%q_segdir' WHERE level = ?", +/* 15 */ "SELECT max(level) FROM %Q.'%q_segdir' WHERE level BETWEEN ? AND ?", + +/* 16 */ "DELETE FROM %Q.'%q_segdir' WHERE level = ?", +/* 17 */ "DELETE FROM %Q.'%q_segments' WHERE blockid BETWEEN ? AND ?", +/* 18 */ "INSERT INTO %Q.'%q_content' VALUES(%s)", +/* 19 */ "DELETE FROM %Q.'%q_docsize' WHERE docid = ?", +/* 20 */ "REPLACE INTO %Q.'%q_docsize' VALUES(?,?)", +/* 21 */ "SELECT size FROM %Q.'%q_docsize' WHERE docid=?", +/* 22 */ "SELECT value FROM %Q.'%q_stat' WHERE id=0", +/* 23 */ "REPLACE INTO %Q.'%q_stat' VALUES(0,?)", +/* 24 */ "", +/* 25 */ "", + +/* 26 */ "DELETE FROM %Q.'%q_segdir' WHERE level BETWEEN ? AND ?", + + }; + int rc = SQLITE_OK; + sqlite3_stmt *pStmt; + + assert( SizeofArray(azSql)==SizeofArray(p->aStmt) ); + assert( eStmt=0 ); + + pStmt = p->aStmt[eStmt]; + if( !pStmt ){ + char *zSql; + if( eStmt==SQL_CONTENT_INSERT ){ + zSql = sqlite3_mprintf(azSql[eStmt], p->zDb, p->zName, p->zWriteExprlist); + }else if( eStmt==SQL_SELECT_CONTENT_BY_ROWID ){ + zSql = sqlite3_mprintf(azSql[eStmt], p->zReadExprlist); + }else{ + zSql = sqlite3_mprintf(azSql[eStmt], p->zDb, p->zName); + } + if( !zSql ){ + rc = SQLITE_NOMEM; + }else{ + rc = sqlite3_prepare_v2(p->db, zSql, -1, &pStmt, NULL); + sqlite3_free(zSql); + assert( rc==SQLITE_OK || pStmt==0 ); + p->aStmt[eStmt] = pStmt; + } + } + if( apVal ){ + int i; + int nParam = sqlite3_bind_parameter_count(pStmt); + for(i=0; rc==SQLITE_OK && izContentTbl==0 ){ + rc = fts3SqlStmt(p, SQL_SELECT_CONTENT_BY_ROWID, &pStmt, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_null(pStmt, 1); + sqlite3_step(pStmt); + rc = sqlite3_reset(pStmt); + } + }else{ + rc = SQLITE_OK; + } + + return rc; +} + +/* +** Set *ppStmt to a statement handle that may be used to iterate through +** all rows in the %_segdir table, from oldest to newest. If successful, +** return SQLITE_OK. If an error occurs while preparing the statement, +** return an SQLite error code. +** +** There is only ever one instance of this SQL statement compiled for +** each FTS3 table. +** +** The statement returns the following columns from the %_segdir table: +** +** 0: idx +** 1: start_block +** 2: leaves_end_block +** 3: end_block +** 4: root +*/ +int sqlite3Fts3AllSegdirs( + Fts3Table *p, /* FTS3 table */ + int iIndex, /* Index for p->aIndex[] */ + int iLevel, /* Level to select */ + sqlite3_stmt **ppStmt /* OUT: Compiled statement */ +){ + int rc; + sqlite3_stmt *pStmt = 0; + + assert( iLevel==FTS3_SEGCURSOR_ALL || iLevel>=0 ); + assert( iLevel=0 && iIndexnIndex ); + + if( iLevel<0 ){ + /* "SELECT * FROM %_segdir WHERE level BETWEEN ? AND ? ORDER BY ..." */ + rc = fts3SqlStmt(p, SQL_SELECT_LEVEL_RANGE, &pStmt, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int(pStmt, 1, iIndex*FTS3_SEGDIR_MAXLEVEL); + sqlite3_bind_int(pStmt, 2, (iIndex+1)*FTS3_SEGDIR_MAXLEVEL-1); + } + }else{ + /* "SELECT * FROM %_segdir WHERE level = ? ORDER BY ..." */ + rc = fts3SqlStmt(p, SQL_SELECT_LEVEL, &pStmt, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int(pStmt, 1, iLevel+iIndex*FTS3_SEGDIR_MAXLEVEL); + } + } + *ppStmt = pStmt; + return rc; +} + + +/* +** Append a single varint to a PendingList buffer. SQLITE_OK is returned +** if successful, or an SQLite error code otherwise. +** +** This function also serves to allocate the PendingList structure itself. +** For example, to create a new PendingList structure containing two +** varints: +** +** PendingList *p = 0; +** fts3PendingListAppendVarint(&p, 1); +** fts3PendingListAppendVarint(&p, 2); +*/ +static int fts3PendingListAppendVarint( + PendingList **pp, /* IN/OUT: Pointer to PendingList struct */ + sqlite3_int64 i /* Value to append to data */ +){ + PendingList *p = *pp; + + /* Allocate or grow the PendingList as required. */ + if( !p ){ + p = sqlite3_malloc(sizeof(*p) + 100); + if( !p ){ + return SQLITE_NOMEM; + } + p->nSpace = 100; + p->aData = (char *)&p[1]; + p->nData = 0; + } + else if( p->nData+FTS3_VARINT_MAX+1>p->nSpace ){ + int nNew = p->nSpace * 2; + p = sqlite3_realloc(p, sizeof(*p) + nNew); + if( !p ){ + sqlite3_free(*pp); + *pp = 0; + return SQLITE_NOMEM; + } + p->nSpace = nNew; + p->aData = (char *)&p[1]; + } + + /* Append the new serialized varint to the end of the list. */ + p->nData += sqlite3Fts3PutVarint(&p->aData[p->nData], i); + p->aData[p->nData] = '\0'; + *pp = p; + return SQLITE_OK; +} + +/* +** Add a docid/column/position entry to a PendingList structure. Non-zero +** is returned if the structure is sqlite3_realloced as part of adding +** the entry. Otherwise, zero. +** +** If an OOM error occurs, *pRc is set to SQLITE_NOMEM before returning. +** Zero is always returned in this case. Otherwise, if no OOM error occurs, +** it is set to SQLITE_OK. +*/ +static int fts3PendingListAppend( + PendingList **pp, /* IN/OUT: PendingList structure */ + sqlite3_int64 iDocid, /* Docid for entry to add */ + sqlite3_int64 iCol, /* Column for entry to add */ + sqlite3_int64 iPos, /* Position of term for entry to add */ + int *pRc /* OUT: Return code */ +){ + PendingList *p = *pp; + int rc = SQLITE_OK; + + assert( !p || p->iLastDocid<=iDocid ); + + if( !p || p->iLastDocid!=iDocid ){ + sqlite3_int64 iDelta = iDocid - (p ? p->iLastDocid : 0); + if( p ){ + assert( p->nDatanSpace ); + assert( p->aData[p->nData]==0 ); + p->nData++; + } + if( SQLITE_OK!=(rc = fts3PendingListAppendVarint(&p, iDelta)) ){ + goto pendinglistappend_out; + } + p->iLastCol = -1; + p->iLastPos = 0; + p->iLastDocid = iDocid; + } + if( iCol>0 && p->iLastCol!=iCol ){ + if( SQLITE_OK!=(rc = fts3PendingListAppendVarint(&p, 1)) + || SQLITE_OK!=(rc = fts3PendingListAppendVarint(&p, iCol)) + ){ + goto pendinglistappend_out; + } + p->iLastCol = iCol; + p->iLastPos = 0; + } + if( iCol>=0 ){ + assert( iPos>p->iLastPos || (iPos==0 && p->iLastPos==0) ); + rc = fts3PendingListAppendVarint(&p, 2+iPos-p->iLastPos); + if( rc==SQLITE_OK ){ + p->iLastPos = iPos; + } + } + + pendinglistappend_out: + *pRc = rc; + if( p!=*pp ){ + *pp = p; + return 1; + } + return 0; +} + +/* +** Free a PendingList object allocated by fts3PendingListAppend(). +*/ +static void fts3PendingListDelete(PendingList *pList){ + sqlite3_free(pList); +} + +/* +** Add an entry to one of the pending-terms hash tables. +*/ +static int fts3PendingTermsAddOne( + Fts3Table *p, + int iCol, + int iPos, + Fts3Hash *pHash, /* Pending terms hash table to add entry to */ + const char *zToken, + int nToken +){ + PendingList *pList; + int rc = SQLITE_OK; + + pList = (PendingList *)fts3HashFind(pHash, zToken, nToken); + if( pList ){ + p->nPendingData -= (pList->nData + nToken + sizeof(Fts3HashElem)); + } + if( fts3PendingListAppend(&pList, p->iPrevDocid, iCol, iPos, &rc) ){ + if( pList==fts3HashInsert(pHash, zToken, nToken, pList) ){ + /* Malloc failed while inserting the new entry. This can only + ** happen if there was no previous entry for this token. + */ + assert( 0==fts3HashFind(pHash, zToken, nToken) ); + sqlite3_free(pList); + rc = SQLITE_NOMEM; + } + } + if( rc==SQLITE_OK ){ + p->nPendingData += (pList->nData + nToken + sizeof(Fts3HashElem)); + } + return rc; +} + +/* +** Tokenize the nul-terminated string zText and add all tokens to the +** pending-terms hash-table. The docid used is that currently stored in +** p->iPrevDocid, and the column is specified by argument iCol. +** +** If successful, SQLITE_OK is returned. Otherwise, an SQLite error code. +*/ +static int fts3PendingTermsAdd( + Fts3Table *p, /* Table into which text will be inserted */ + const char *zText, /* Text of document to be inserted */ + int iCol, /* Column into which text is being inserted */ + u32 *pnWord /* OUT: Number of tokens inserted */ +){ + int rc; + int iStart; + int iEnd; + int iPos; + int nWord = 0; + + char const *zToken; + int nToken; + + sqlite3_tokenizer *pTokenizer = p->pTokenizer; + sqlite3_tokenizer_module const *pModule = pTokenizer->pModule; + sqlite3_tokenizer_cursor *pCsr; + int (*xNext)(sqlite3_tokenizer_cursor *pCursor, + const char**,int*,int*,int*,int*); + + assert( pTokenizer && pModule ); + + /* If the user has inserted a NULL value, this function may be called with + ** zText==0. In this case, add zero token entries to the hash table and + ** return early. */ + if( zText==0 ){ + *pnWord = 0; + return SQLITE_OK; + } + + rc = pModule->xOpen(pTokenizer, zText, -1, &pCsr); + if( rc!=SQLITE_OK ){ + return rc; + } + pCsr->pTokenizer = pTokenizer; + + xNext = pModule->xNext; + while( SQLITE_OK==rc + && SQLITE_OK==(rc = xNext(pCsr, &zToken, &nToken, &iStart, &iEnd, &iPos)) + ){ + int i; + if( iPos>=nWord ) nWord = iPos+1; + + /* Positions cannot be negative; we use -1 as a terminator internally. + ** Tokens must have a non-zero length. + */ + if( iPos<0 || !zToken || nToken<=0 ){ + rc = SQLITE_ERROR; + break; + } + + /* Add the term to the terms index */ + rc = fts3PendingTermsAddOne( + p, iCol, iPos, &p->aIndex[0].hPending, zToken, nToken + ); + + /* Add the term to each of the prefix indexes that it is not too + ** short for. */ + for(i=1; rc==SQLITE_OK && inIndex; i++){ + struct Fts3Index *pIndex = &p->aIndex[i]; + if( nTokennPrefix ) continue; + rc = fts3PendingTermsAddOne( + p, iCol, iPos, &pIndex->hPending, zToken, pIndex->nPrefix + ); + } + } + + pModule->xClose(pCsr); + *pnWord = nWord; + return (rc==SQLITE_DONE ? SQLITE_OK : rc); +} + +/* +** Calling this function indicates that subsequent calls to +** fts3PendingTermsAdd() are to add term/position-list pairs for the +** contents of the document with docid iDocid. +*/ +static int fts3PendingTermsDocid(Fts3Table *p, sqlite_int64 iDocid){ + /* TODO(shess) Explore whether partially flushing the buffer on + ** forced-flush would provide better performance. I suspect that if + ** we ordered the doclists by size and flushed the largest until the + ** buffer was half empty, that would let the less frequent terms + ** generate longer doclists. + */ + if( iDocid<=p->iPrevDocid || p->nPendingData>p->nMaxPendingData ){ + int rc = sqlite3Fts3PendingTermsFlush(p); + if( rc!=SQLITE_OK ) return rc; + } + p->iPrevDocid = iDocid; + return SQLITE_OK; +} + +/* +** Discard the contents of the pending-terms hash tables. +*/ +void sqlite3Fts3PendingTermsClear(Fts3Table *p){ + int i; + for(i=0; inIndex; i++){ + Fts3HashElem *pElem; + Fts3Hash *pHash = &p->aIndex[i].hPending; + for(pElem=fts3HashFirst(pHash); pElem; pElem=fts3HashNext(pElem)){ + PendingList *pList = (PendingList *)fts3HashData(pElem); + fts3PendingListDelete(pList); + } + fts3HashClear(pHash); + } + p->nPendingData = 0; +} + +/* +** This function is called by the xUpdate() method as part of an INSERT +** operation. It adds entries for each term in the new record to the +** pendingTerms hash table. +** +** Argument apVal is the same as the similarly named argument passed to +** fts3InsertData(). Parameter iDocid is the docid of the new row. +*/ +static int fts3InsertTerms(Fts3Table *p, sqlite3_value **apVal, u32 *aSz){ + int i; /* Iterator variable */ + for(i=2; inColumn+2; i++){ + const char *zText = (const char *)sqlite3_value_text(apVal[i]); + int rc = fts3PendingTermsAdd(p, zText, i-2, &aSz[i-2]); + if( rc!=SQLITE_OK ){ + return rc; + } + aSz[p->nColumn] += sqlite3_value_bytes(apVal[i]); + } + return SQLITE_OK; +} + +/* +** This function is called by the xUpdate() method for an INSERT operation. +** The apVal parameter is passed a copy of the apVal argument passed by +** SQLite to the xUpdate() method. i.e: +** +** apVal[0] Not used for INSERT. +** apVal[1] rowid +** apVal[2] Left-most user-defined column +** ... +** apVal[p->nColumn+1] Right-most user-defined column +** apVal[p->nColumn+2] Hidden column with same name as table +** apVal[p->nColumn+3] Hidden "docid" column (alias for rowid) +*/ +static int fts3InsertData( + Fts3Table *p, /* Full-text table */ + sqlite3_value **apVal, /* Array of values to insert */ + sqlite3_int64 *piDocid /* OUT: Docid for row just inserted */ +){ + int rc; /* Return code */ + sqlite3_stmt *pContentInsert; /* INSERT INTO %_content VALUES(...) */ + + if( p->zContentTbl ){ + sqlite3_value *pRowid = apVal[p->nColumn+3]; + if( sqlite3_value_type(pRowid)==SQLITE_NULL ){ + pRowid = apVal[1]; + } + if( sqlite3_value_type(pRowid)!=SQLITE_INTEGER ){ + return SQLITE_CONSTRAINT; + } + *piDocid = sqlite3_value_int64(pRowid); + return SQLITE_OK; + } + + /* Locate the statement handle used to insert data into the %_content + ** table. The SQL for this statement is: + ** + ** INSERT INTO %_content VALUES(?, ?, ?, ...) + ** + ** The statement features N '?' variables, where N is the number of user + ** defined columns in the FTS3 table, plus one for the docid field. + */ + rc = fts3SqlStmt(p, SQL_CONTENT_INSERT, &pContentInsert, &apVal[1]); + if( rc!=SQLITE_OK ){ + return rc; + } + + /* There is a quirk here. The users INSERT statement may have specified + ** a value for the "rowid" field, for the "docid" field, or for both. + ** Which is a problem, since "rowid" and "docid" are aliases for the + ** same value. For example: + ** + ** INSERT INTO fts3tbl(rowid, docid) VALUES(1, 2); + ** + ** In FTS3, this is an error. It is an error to specify non-NULL values + ** for both docid and some other rowid alias. + */ + if( SQLITE_NULL!=sqlite3_value_type(apVal[3+p->nColumn]) ){ + if( SQLITE_NULL==sqlite3_value_type(apVal[0]) + && SQLITE_NULL!=sqlite3_value_type(apVal[1]) + ){ + /* A rowid/docid conflict. */ + return SQLITE_ERROR; + } + rc = sqlite3_bind_value(pContentInsert, 1, apVal[3+p->nColumn]); + if( rc!=SQLITE_OK ) return rc; + } + + /* Execute the statement to insert the record. Set *piDocid to the + ** new docid value. + */ + sqlite3_step(pContentInsert); + rc = sqlite3_reset(pContentInsert); + + *piDocid = sqlite3_last_insert_rowid(p->db); + return rc; +} + + + +/* +** Remove all data from the FTS3 table. Clear the hash table containing +** pending terms. +*/ +static int fts3DeleteAll(Fts3Table *p, int bContent){ + int rc = SQLITE_OK; /* Return code */ + + /* Discard the contents of the pending-terms hash table. */ + sqlite3Fts3PendingTermsClear(p); + + /* Delete everything from the shadow tables. Except, leave %_content as + ** is if bContent is false. */ + assert( p->zContentTbl==0 || bContent==0 ); + if( bContent ) fts3SqlExec(&rc, p, SQL_DELETE_ALL_CONTENT, 0); + fts3SqlExec(&rc, p, SQL_DELETE_ALL_SEGMENTS, 0); + fts3SqlExec(&rc, p, SQL_DELETE_ALL_SEGDIR, 0); + if( p->bHasDocsize ){ + fts3SqlExec(&rc, p, SQL_DELETE_ALL_DOCSIZE, 0); + } + if( p->bHasStat ){ + fts3SqlExec(&rc, p, SQL_DELETE_ALL_STAT, 0); + } + return rc; +} + +/* +** The first element in the apVal[] array is assumed to contain the docid +** (an integer) of a row about to be deleted. Remove all terms from the +** full-text index. +*/ +static void fts3DeleteTerms( + int *pRC, /* Result code */ + Fts3Table *p, /* The FTS table to delete from */ + sqlite3_value *pRowid, /* The docid to be deleted */ + u32 *aSz /* Sizes of deleted document written here */ +){ + int rc; + sqlite3_stmt *pSelect; + + if( *pRC ) return; + rc = fts3SqlStmt(p, SQL_SELECT_CONTENT_BY_ROWID, &pSelect, &pRowid); + if( rc==SQLITE_OK ){ + if( SQLITE_ROW==sqlite3_step(pSelect) ){ + int i; + for(i=1; i<=p->nColumn; i++){ + const char *zText = (const char *)sqlite3_column_text(pSelect, i); + rc = fts3PendingTermsAdd(p, zText, -1, &aSz[i-1]); + if( rc!=SQLITE_OK ){ + sqlite3_reset(pSelect); + *pRC = rc; + return; + } + aSz[p->nColumn] += sqlite3_column_bytes(pSelect, i); + } + } + rc = sqlite3_reset(pSelect); + }else{ + sqlite3_reset(pSelect); + } + *pRC = rc; +} + +/* +** Forward declaration to account for the circular dependency between +** functions fts3SegmentMerge() and fts3AllocateSegdirIdx(). +*/ +static int fts3SegmentMerge(Fts3Table *, int, int); + +/* +** This function allocates a new level iLevel index in the segdir table. +** Usually, indexes are allocated within a level sequentially starting +** with 0, so the allocated index is one greater than the value returned +** by: +** +** SELECT max(idx) FROM %_segdir WHERE level = :iLevel +** +** However, if there are already FTS3_MERGE_COUNT indexes at the requested +** level, they are merged into a single level (iLevel+1) segment and the +** allocated index is 0. +** +** If successful, *piIdx is set to the allocated index slot and SQLITE_OK +** returned. Otherwise, an SQLite error code is returned. +*/ +static int fts3AllocateSegdirIdx( + Fts3Table *p, + int iIndex, /* Index for p->aIndex */ + int iLevel, + int *piIdx +){ + int rc; /* Return Code */ + sqlite3_stmt *pNextIdx; /* Query for next idx at level iLevel */ + int iNext = 0; /* Result of query pNextIdx */ + + /* Set variable iNext to the next available segdir index at level iLevel. */ + rc = fts3SqlStmt(p, SQL_NEXT_SEGMENT_INDEX, &pNextIdx, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int(pNextIdx, 1, iIndex*FTS3_SEGDIR_MAXLEVEL + iLevel); + if( SQLITE_ROW==sqlite3_step(pNextIdx) ){ + iNext = sqlite3_column_int(pNextIdx, 0); + } + rc = sqlite3_reset(pNextIdx); + } + + if( rc==SQLITE_OK ){ + /* If iNext is FTS3_MERGE_COUNT, indicating that level iLevel is already + ** full, merge all segments in level iLevel into a single iLevel+1 + ** segment and allocate (newly freed) index 0 at level iLevel. Otherwise, + ** if iNext is less than FTS3_MERGE_COUNT, allocate index iNext. + */ + if( iNext>=FTS3_MERGE_COUNT ){ + rc = fts3SegmentMerge(p, iIndex, iLevel); + *piIdx = 0; + }else{ + *piIdx = iNext; + } + } + + return rc; +} + +/* +** The %_segments table is declared as follows: +** +** CREATE TABLE %_segments(blockid INTEGER PRIMARY KEY, block BLOB) +** +** This function reads data from a single row of the %_segments table. The +** specific row is identified by the iBlockid parameter. If paBlob is not +** NULL, then a buffer is allocated using sqlite3_malloc() and populated +** with the contents of the blob stored in the "block" column of the +** identified table row is. Whether or not paBlob is NULL, *pnBlob is set +** to the size of the blob in bytes before returning. +** +** If an error occurs, or the table does not contain the specified row, +** an SQLite error code is returned. Otherwise, SQLITE_OK is returned. If +** paBlob is non-NULL, then it is the responsibility of the caller to +** eventually free the returned buffer. +** +** This function may leave an open sqlite3_blob* handle in the +** Fts3Table.pSegments variable. This handle is reused by subsequent calls +** to this function. The handle may be closed by calling the +** sqlite3Fts3SegmentsClose() function. Reusing a blob handle is a handy +** performance improvement, but the blob handle should always be closed +** before control is returned to the user (to prevent a lock being held +** on the database file for longer than necessary). Thus, any virtual table +** method (xFilter etc.) that may directly or indirectly call this function +** must call sqlite3Fts3SegmentsClose() before returning. +*/ +int sqlite3Fts3ReadBlock( + Fts3Table *p, /* FTS3 table handle */ + sqlite3_int64 iBlockid, /* Access the row with blockid=$iBlockid */ + char **paBlob, /* OUT: Blob data in malloc'd buffer */ + int *pnBlob, /* OUT: Size of blob data */ + int *pnLoad /* OUT: Bytes actually loaded */ +){ + int rc; /* Return code */ + + /* pnBlob must be non-NULL. paBlob may be NULL or non-NULL. */ + assert( pnBlob); + + if( p->pSegments ){ + rc = sqlite3_blob_reopen(p->pSegments, iBlockid); + }else{ + if( 0==p->zSegmentsTbl ){ + p->zSegmentsTbl = sqlite3_mprintf("%s_segments", p->zName); + if( 0==p->zSegmentsTbl ) return SQLITE_NOMEM; + } + rc = sqlite3_blob_open( + p->db, p->zDb, p->zSegmentsTbl, "block", iBlockid, 0, &p->pSegments + ); + } + + if( rc==SQLITE_OK ){ + int nByte = sqlite3_blob_bytes(p->pSegments); + *pnBlob = nByte; + if( paBlob ){ + char *aByte = sqlite3_malloc(nByte + FTS3_NODE_PADDING); + if( !aByte ){ + rc = SQLITE_NOMEM; + }else{ + if( pnLoad && nByte>(FTS3_NODE_CHUNK_THRESHOLD) ){ + nByte = FTS3_NODE_CHUNKSIZE; + *pnLoad = nByte; + } + rc = sqlite3_blob_read(p->pSegments, aByte, nByte, 0); + memset(&aByte[nByte], 0, FTS3_NODE_PADDING); + if( rc!=SQLITE_OK ){ + sqlite3_free(aByte); + aByte = 0; + } + } + *paBlob = aByte; + } + } + + return rc; +} + +/* +** Close the blob handle at p->pSegments, if it is open. See comments above +** the sqlite3Fts3ReadBlock() function for details. +*/ +void sqlite3Fts3SegmentsClose(Fts3Table *p){ + sqlite3_blob_close(p->pSegments); + p->pSegments = 0; +} + +static int fts3SegReaderIncrRead(Fts3SegReader *pReader){ + int nRead; /* Number of bytes to read */ + int rc; /* Return code */ + + nRead = MIN(pReader->nNode - pReader->nPopulate, FTS3_NODE_CHUNKSIZE); + rc = sqlite3_blob_read( + pReader->pBlob, + &pReader->aNode[pReader->nPopulate], + nRead, + pReader->nPopulate + ); + + if( rc==SQLITE_OK ){ + pReader->nPopulate += nRead; + memset(&pReader->aNode[pReader->nPopulate], 0, FTS3_NODE_PADDING); + if( pReader->nPopulate==pReader->nNode ){ + sqlite3_blob_close(pReader->pBlob); + pReader->pBlob = 0; + pReader->nPopulate = 0; + } + } + return rc; +} + +static int fts3SegReaderRequire(Fts3SegReader *pReader, char *pFrom, int nByte){ + int rc = SQLITE_OK; + assert( !pReader->pBlob + || (pFrom>=pReader->aNode && pFrom<&pReader->aNode[pReader->nNode]) + ); + while( pReader->pBlob && rc==SQLITE_OK + && (pFrom - pReader->aNode + nByte)>pReader->nPopulate + ){ + rc = fts3SegReaderIncrRead(pReader); + } + return rc; +} + +/* +** Move the iterator passed as the first argument to the next term in the +** segment. If successful, SQLITE_OK is returned. If there is no next term, +** SQLITE_DONE. Otherwise, an SQLite error code. +*/ +static int fts3SegReaderNext( + Fts3Table *p, + Fts3SegReader *pReader, + int bIncr +){ + int rc; /* Return code of various sub-routines */ + char *pNext; /* Cursor variable */ + int nPrefix; /* Number of bytes in term prefix */ + int nSuffix; /* Number of bytes in term suffix */ + + if( !pReader->aDoclist ){ + pNext = pReader->aNode; + }else{ + pNext = &pReader->aDoclist[pReader->nDoclist]; + } + + if( !pNext || pNext>=&pReader->aNode[pReader->nNode] ){ + + if( fts3SegReaderIsPending(pReader) ){ + Fts3HashElem *pElem = *(pReader->ppNextElem); + if( pElem==0 ){ + pReader->aNode = 0; + }else{ + PendingList *pList = (PendingList *)fts3HashData(pElem); + pReader->zTerm = (char *)fts3HashKey(pElem); + pReader->nTerm = fts3HashKeysize(pElem); + pReader->nNode = pReader->nDoclist = pList->nData + 1; + pReader->aNode = pReader->aDoclist = pList->aData; + pReader->ppNextElem++; + assert( pReader->aNode ); + } + return SQLITE_OK; + } + + if( !fts3SegReaderIsRootOnly(pReader) ){ + sqlite3_free(pReader->aNode); + sqlite3_blob_close(pReader->pBlob); + pReader->pBlob = 0; + } + pReader->aNode = 0; + + /* If iCurrentBlock>=iLeafEndBlock, this is an EOF condition. All leaf + ** blocks have already been traversed. */ + assert( pReader->iCurrentBlock<=pReader->iLeafEndBlock ); + if( pReader->iCurrentBlock>=pReader->iLeafEndBlock ){ + return SQLITE_OK; + } + + rc = sqlite3Fts3ReadBlock( + p, ++pReader->iCurrentBlock, &pReader->aNode, &pReader->nNode, + (bIncr ? &pReader->nPopulate : 0) + ); + if( rc!=SQLITE_OK ) return rc; + assert( pReader->pBlob==0 ); + if( bIncr && pReader->nPopulatenNode ){ + pReader->pBlob = p->pSegments; + p->pSegments = 0; + } + pNext = pReader->aNode; + } + + assert( !fts3SegReaderIsPending(pReader) ); + + rc = fts3SegReaderRequire(pReader, pNext, FTS3_VARINT_MAX*2); + if( rc!=SQLITE_OK ) return rc; + + /* Because of the FTS3_NODE_PADDING bytes of padding, the following is + ** safe (no risk of overread) even if the node data is corrupted. */ + pNext += sqlite3Fts3GetVarint32(pNext, &nPrefix); + pNext += sqlite3Fts3GetVarint32(pNext, &nSuffix); + if( nPrefix<0 || nSuffix<=0 + || &pNext[nSuffix]>&pReader->aNode[pReader->nNode] + ){ + return FTS_CORRUPT_VTAB; + } + + if( nPrefix+nSuffix>pReader->nTermAlloc ){ + int nNew = (nPrefix+nSuffix)*2; + char *zNew = sqlite3_realloc(pReader->zTerm, nNew); + if( !zNew ){ + return SQLITE_NOMEM; + } + pReader->zTerm = zNew; + pReader->nTermAlloc = nNew; + } + + rc = fts3SegReaderRequire(pReader, pNext, nSuffix+FTS3_VARINT_MAX); + if( rc!=SQLITE_OK ) return rc; + + memcpy(&pReader->zTerm[nPrefix], pNext, nSuffix); + pReader->nTerm = nPrefix+nSuffix; + pNext += nSuffix; + pNext += sqlite3Fts3GetVarint32(pNext, &pReader->nDoclist); + pReader->aDoclist = pNext; + pReader->pOffsetList = 0; + + /* Check that the doclist does not appear to extend past the end of the + ** b-tree node. And that the final byte of the doclist is 0x00. If either + ** of these statements is untrue, then the data structure is corrupt. + */ + if( &pReader->aDoclist[pReader->nDoclist]>&pReader->aNode[pReader->nNode] + || (pReader->nPopulate==0 && pReader->aDoclist[pReader->nDoclist-1]) + ){ + return FTS_CORRUPT_VTAB; + } + return SQLITE_OK; +} + +/* +** Set the SegReader to point to the first docid in the doclist associated +** with the current term. +*/ +static int fts3SegReaderFirstDocid(Fts3Table *pTab, Fts3SegReader *pReader){ + int rc = SQLITE_OK; + assert( pReader->aDoclist ); + assert( !pReader->pOffsetList ); + if( pTab->bDescIdx && fts3SegReaderIsPending(pReader) ){ + u8 bEof = 0; + pReader->iDocid = 0; + pReader->nOffsetList = 0; + sqlite3Fts3DoclistPrev(0, + pReader->aDoclist, pReader->nDoclist, &pReader->pOffsetList, + &pReader->iDocid, &pReader->nOffsetList, &bEof + ); + }else{ + rc = fts3SegReaderRequire(pReader, pReader->aDoclist, FTS3_VARINT_MAX); + if( rc==SQLITE_OK ){ + int n = sqlite3Fts3GetVarint(pReader->aDoclist, &pReader->iDocid); + pReader->pOffsetList = &pReader->aDoclist[n]; + } + } + return rc; +} + +/* +** Advance the SegReader to point to the next docid in the doclist +** associated with the current term. +** +** If arguments ppOffsetList and pnOffsetList are not NULL, then +** *ppOffsetList is set to point to the first column-offset list +** in the doclist entry (i.e. immediately past the docid varint). +** *pnOffsetList is set to the length of the set of column-offset +** lists, not including the nul-terminator byte. For example: +*/ +static int fts3SegReaderNextDocid( + Fts3Table *pTab, + Fts3SegReader *pReader, /* Reader to advance to next docid */ + char **ppOffsetList, /* OUT: Pointer to current position-list */ + int *pnOffsetList /* OUT: Length of *ppOffsetList in bytes */ +){ + int rc = SQLITE_OK; + char *p = pReader->pOffsetList; + char c = 0; + + assert( p ); + + if( pTab->bDescIdx && fts3SegReaderIsPending(pReader) ){ + /* A pending-terms seg-reader for an FTS4 table that uses order=desc. + ** Pending-terms doclists are always built up in ascending order, so + ** we have to iterate through them backwards here. */ + u8 bEof = 0; + if( ppOffsetList ){ + *ppOffsetList = pReader->pOffsetList; + *pnOffsetList = pReader->nOffsetList - 1; + } + sqlite3Fts3DoclistPrev(0, + pReader->aDoclist, pReader->nDoclist, &p, &pReader->iDocid, + &pReader->nOffsetList, &bEof + ); + if( bEof ){ + pReader->pOffsetList = 0; + }else{ + pReader->pOffsetList = p; + } + }else{ + char *pEnd = &pReader->aDoclist[pReader->nDoclist]; + + /* Pointer p currently points at the first byte of an offset list. The + ** following block advances it to point one byte past the end of + ** the same offset list. */ + while( 1 ){ + + /* The following line of code (and the "p++" below the while() loop) is + ** normally all that is required to move pointer p to the desired + ** position. The exception is if this node is being loaded from disk + ** incrementally and pointer "p" now points to the first byte passed + ** the populated part of pReader->aNode[]. + */ + while( *p | c ) c = *p++ & 0x80; + assert( *p==0 ); + + if( pReader->pBlob==0 || p<&pReader->aNode[pReader->nPopulate] ) break; + rc = fts3SegReaderIncrRead(pReader); + if( rc!=SQLITE_OK ) return rc; + } + p++; + + /* If required, populate the output variables with a pointer to and the + ** size of the previous offset-list. + */ + if( ppOffsetList ){ + *ppOffsetList = pReader->pOffsetList; + *pnOffsetList = (int)(p - pReader->pOffsetList - 1); + } + + while( p=pEnd ){ + pReader->pOffsetList = 0; + }else{ + rc = fts3SegReaderRequire(pReader, p, FTS3_VARINT_MAX); + if( rc==SQLITE_OK ){ + sqlite3_int64 iDelta; + pReader->pOffsetList = p + sqlite3Fts3GetVarint(p, &iDelta); + if( pTab->bDescIdx ){ + pReader->iDocid -= iDelta; + }else{ + pReader->iDocid += iDelta; + } + } + } + } + + return SQLITE_OK; +} + + +int sqlite3Fts3MsrOvfl( + Fts3Cursor *pCsr, + Fts3MultiSegReader *pMsr, + int *pnOvfl +){ + Fts3Table *p = (Fts3Table*)pCsr->base.pVtab; + int nOvfl = 0; + int ii; + int rc = SQLITE_OK; + int pgsz = p->nPgsz; + + assert( p->bHasStat ); + assert( pgsz>0 ); + + for(ii=0; rc==SQLITE_OK && iinSegment; ii++){ + Fts3SegReader *pReader = pMsr->apSegment[ii]; + if( !fts3SegReaderIsPending(pReader) + && !fts3SegReaderIsRootOnly(pReader) + ){ + sqlite3_int64 jj; + for(jj=pReader->iStartBlock; jj<=pReader->iLeafEndBlock; jj++){ + int nBlob; + rc = sqlite3Fts3ReadBlock(p, jj, 0, &nBlob, 0); + if( rc!=SQLITE_OK ) break; + if( (nBlob+35)>pgsz ){ + nOvfl += (nBlob + 34)/pgsz; + } + } + } + } + *pnOvfl = nOvfl; + return rc; +} + +/* +** Free all allocations associated with the iterator passed as the +** second argument. +*/ +void sqlite3Fts3SegReaderFree(Fts3SegReader *pReader){ + if( pReader && !fts3SegReaderIsPending(pReader) ){ + sqlite3_free(pReader->zTerm); + if( !fts3SegReaderIsRootOnly(pReader) ){ + sqlite3_free(pReader->aNode); + sqlite3_blob_close(pReader->pBlob); + } + } + sqlite3_free(pReader); +} + +/* +** Allocate a new SegReader object. +*/ +int sqlite3Fts3SegReaderNew( + int iAge, /* Segment "age". */ + sqlite3_int64 iStartLeaf, /* First leaf to traverse */ + sqlite3_int64 iEndLeaf, /* Final leaf to traverse */ + sqlite3_int64 iEndBlock, /* Final block of segment */ + const char *zRoot, /* Buffer containing root node */ + int nRoot, /* Size of buffer containing root node */ + Fts3SegReader **ppReader /* OUT: Allocated Fts3SegReader */ +){ + int rc = SQLITE_OK; /* Return code */ + Fts3SegReader *pReader; /* Newly allocated SegReader object */ + int nExtra = 0; /* Bytes to allocate segment root node */ + + assert( iStartLeaf<=iEndLeaf ); + if( iStartLeaf==0 ){ + nExtra = nRoot + FTS3_NODE_PADDING; + } + + pReader = (Fts3SegReader *)sqlite3_malloc(sizeof(Fts3SegReader) + nExtra); + if( !pReader ){ + return SQLITE_NOMEM; + } + memset(pReader, 0, sizeof(Fts3SegReader)); + pReader->iIdx = iAge; + pReader->iStartBlock = iStartLeaf; + pReader->iLeafEndBlock = iEndLeaf; + pReader->iEndBlock = iEndBlock; + + if( nExtra ){ + /* The entire segment is stored in the root node. */ + pReader->aNode = (char *)&pReader[1]; + pReader->nNode = nRoot; + memcpy(pReader->aNode, zRoot, nRoot); + memset(&pReader->aNode[nRoot], 0, FTS3_NODE_PADDING); + }else{ + pReader->iCurrentBlock = iStartLeaf-1; + } + + if( rc==SQLITE_OK ){ + *ppReader = pReader; + }else{ + sqlite3Fts3SegReaderFree(pReader); + } + return rc; +} + +/* +** This is a comparison function used as a qsort() callback when sorting +** an array of pending terms by term. This occurs as part of flushing +** the contents of the pending-terms hash table to the database. +*/ +static int fts3CompareElemByTerm(const void *lhs, const void *rhs){ + char *z1 = fts3HashKey(*(Fts3HashElem **)lhs); + char *z2 = fts3HashKey(*(Fts3HashElem **)rhs); + int n1 = fts3HashKeysize(*(Fts3HashElem **)lhs); + int n2 = fts3HashKeysize(*(Fts3HashElem **)rhs); + + int n = (n1aIndex */ + const char *zTerm, /* Term to search for */ + int nTerm, /* Size of buffer zTerm */ + int bPrefix, /* True for a prefix iterator */ + Fts3SegReader **ppReader /* OUT: SegReader for pending-terms */ +){ + Fts3SegReader *pReader = 0; /* Fts3SegReader object to return */ + Fts3HashElem **aElem = 0; /* Array of term hash entries to scan */ + int nElem = 0; /* Size of array at aElem */ + int rc = SQLITE_OK; /* Return Code */ + Fts3Hash *pHash; + + pHash = &p->aIndex[iIndex].hPending; + if( bPrefix ){ + int nAlloc = 0; /* Size of allocated array at aElem */ + Fts3HashElem *pE = 0; /* Iterator variable */ + + for(pE=fts3HashFirst(pHash); pE; pE=fts3HashNext(pE)){ + char *zKey = (char *)fts3HashKey(pE); + int nKey = fts3HashKeysize(pE); + if( nTerm==0 || (nKey>=nTerm && 0==memcmp(zKey, zTerm, nTerm)) ){ + if( nElem==nAlloc ){ + Fts3HashElem **aElem2; + nAlloc += 16; + aElem2 = (Fts3HashElem **)sqlite3_realloc( + aElem, nAlloc*sizeof(Fts3HashElem *) + ); + if( !aElem2 ){ + rc = SQLITE_NOMEM; + nElem = 0; + break; + } + aElem = aElem2; + } + + aElem[nElem++] = pE; + } + } + + /* If more than one term matches the prefix, sort the Fts3HashElem + ** objects in term order using qsort(). This uses the same comparison + ** callback as is used when flushing terms to disk. + */ + if( nElem>1 ){ + qsort(aElem, nElem, sizeof(Fts3HashElem *), fts3CompareElemByTerm); + } + + }else{ + /* The query is a simple term lookup that matches at most one term in + ** the index. All that is required is a straight hash-lookup. */ + Fts3HashElem *pE = fts3HashFindElem(pHash, zTerm, nTerm); + if( pE ){ + aElem = &pE; + nElem = 1; + } + } + + if( nElem>0 ){ + int nByte = sizeof(Fts3SegReader) + (nElem+1)*sizeof(Fts3HashElem *); + pReader = (Fts3SegReader *)sqlite3_malloc(nByte); + if( !pReader ){ + rc = SQLITE_NOMEM; + }else{ + memset(pReader, 0, nByte); + pReader->iIdx = 0x7FFFFFFF; + pReader->ppNextElem = (Fts3HashElem **)&pReader[1]; + memcpy(pReader->ppNextElem, aElem, nElem*sizeof(Fts3HashElem *)); + } + } + + if( bPrefix ){ + sqlite3_free(aElem); + } + *ppReader = pReader; + return rc; +} + +/* +** Compare the entries pointed to by two Fts3SegReader structures. +** Comparison is as follows: +** +** 1) EOF is greater than not EOF. +** +** 2) The current terms (if any) are compared using memcmp(). If one +** term is a prefix of another, the longer term is considered the +** larger. +** +** 3) By segment age. An older segment is considered larger. +*/ +static int fts3SegReaderCmp(Fts3SegReader *pLhs, Fts3SegReader *pRhs){ + int rc; + if( pLhs->aNode && pRhs->aNode ){ + int rc2 = pLhs->nTerm - pRhs->nTerm; + if( rc2<0 ){ + rc = memcmp(pLhs->zTerm, pRhs->zTerm, pLhs->nTerm); + }else{ + rc = memcmp(pLhs->zTerm, pRhs->zTerm, pRhs->nTerm); + } + if( rc==0 ){ + rc = rc2; + } + }else{ + rc = (pLhs->aNode==0) - (pRhs->aNode==0); + } + if( rc==0 ){ + rc = pRhs->iIdx - pLhs->iIdx; + } + assert( rc!=0 ); + return rc; +} + +/* +** A different comparison function for SegReader structures. In this +** version, it is assumed that each SegReader points to an entry in +** a doclist for identical terms. Comparison is made as follows: +** +** 1) EOF (end of doclist in this case) is greater than not EOF. +** +** 2) By current docid. +** +** 3) By segment age. An older segment is considered larger. +*/ +static int fts3SegReaderDoclistCmp(Fts3SegReader *pLhs, Fts3SegReader *pRhs){ + int rc = (pLhs->pOffsetList==0)-(pRhs->pOffsetList==0); + if( rc==0 ){ + if( pLhs->iDocid==pRhs->iDocid ){ + rc = pRhs->iIdx - pLhs->iIdx; + }else{ + rc = (pLhs->iDocid > pRhs->iDocid) ? 1 : -1; + } + } + assert( pLhs->aNode && pRhs->aNode ); + return rc; +} +static int fts3SegReaderDoclistCmpRev(Fts3SegReader *pLhs, Fts3SegReader *pRhs){ + int rc = (pLhs->pOffsetList==0)-(pRhs->pOffsetList==0); + if( rc==0 ){ + if( pLhs->iDocid==pRhs->iDocid ){ + rc = pRhs->iIdx - pLhs->iIdx; + }else{ + rc = (pLhs->iDocid < pRhs->iDocid) ? 1 : -1; + } + } + assert( pLhs->aNode && pRhs->aNode ); + return rc; +} + +/* +** Compare the term that the Fts3SegReader object passed as the first argument +** points to with the term specified by arguments zTerm and nTerm. +** +** If the pSeg iterator is already at EOF, return 0. Otherwise, return +** -ve if the pSeg term is less than zTerm/nTerm, 0 if the two terms are +** equal, or +ve if the pSeg term is greater than zTerm/nTerm. +*/ +static int fts3SegReaderTermCmp( + Fts3SegReader *pSeg, /* Segment reader object */ + const char *zTerm, /* Term to compare to */ + int nTerm /* Size of term zTerm in bytes */ +){ + int res = 0; + if( pSeg->aNode ){ + if( pSeg->nTerm>nTerm ){ + res = memcmp(pSeg->zTerm, zTerm, nTerm); + }else{ + res = memcmp(pSeg->zTerm, zTerm, pSeg->nTerm); + } + if( res==0 ){ + res = pSeg->nTerm-nTerm; + } + } + return res; +} + +/* +** Argument apSegment is an array of nSegment elements. It is known that +** the final (nSegment-nSuspect) members are already in sorted order +** (according to the comparison function provided). This function shuffles +** the array around until all entries are in sorted order. +*/ +static void fts3SegReaderSort( + Fts3SegReader **apSegment, /* Array to sort entries of */ + int nSegment, /* Size of apSegment array */ + int nSuspect, /* Unsorted entry count */ + int (*xCmp)(Fts3SegReader *, Fts3SegReader *) /* Comparison function */ +){ + int i; /* Iterator variable */ + + assert( nSuspect<=nSegment ); + + if( nSuspect==nSegment ) nSuspect--; + for(i=nSuspect-1; i>=0; i--){ + int j; + for(j=i; j<(nSegment-1); j++){ + Fts3SegReader *pTmp; + if( xCmp(apSegment[j], apSegment[j+1])<0 ) break; + pTmp = apSegment[j+1]; + apSegment[j+1] = apSegment[j]; + apSegment[j] = pTmp; + } + } + +#ifndef NDEBUG + /* Check that the list really is sorted now. */ + for(i=0; i<(nSuspect-1); i++){ + assert( xCmp(apSegment[i], apSegment[i+1])<0 ); + } +#endif +} + +/* +** Insert a record into the %_segments table. +*/ +static int fts3WriteSegment( + Fts3Table *p, /* Virtual table handle */ + sqlite3_int64 iBlock, /* Block id for new block */ + char *z, /* Pointer to buffer containing block data */ + int n /* Size of buffer z in bytes */ +){ + sqlite3_stmt *pStmt; + int rc = fts3SqlStmt(p, SQL_INSERT_SEGMENTS, &pStmt, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int64(pStmt, 1, iBlock); + sqlite3_bind_blob(pStmt, 2, z, n, SQLITE_STATIC); + sqlite3_step(pStmt); + rc = sqlite3_reset(pStmt); + } + return rc; +} + +/* +** Insert a record into the %_segdir table. +*/ +static int fts3WriteSegdir( + Fts3Table *p, /* Virtual table handle */ + int iLevel, /* Value for "level" field */ + int iIdx, /* Value for "idx" field */ + sqlite3_int64 iStartBlock, /* Value for "start_block" field */ + sqlite3_int64 iLeafEndBlock, /* Value for "leaves_end_block" field */ + sqlite3_int64 iEndBlock, /* Value for "end_block" field */ + char *zRoot, /* Blob value for "root" field */ + int nRoot /* Number of bytes in buffer zRoot */ +){ + sqlite3_stmt *pStmt; + int rc = fts3SqlStmt(p, SQL_INSERT_SEGDIR, &pStmt, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int(pStmt, 1, iLevel); + sqlite3_bind_int(pStmt, 2, iIdx); + sqlite3_bind_int64(pStmt, 3, iStartBlock); + sqlite3_bind_int64(pStmt, 4, iLeafEndBlock); + sqlite3_bind_int64(pStmt, 5, iEndBlock); + sqlite3_bind_blob(pStmt, 6, zRoot, nRoot, SQLITE_STATIC); + sqlite3_step(pStmt); + rc = sqlite3_reset(pStmt); + } + return rc; +} + +/* +** Return the size of the common prefix (if any) shared by zPrev and +** zNext, in bytes. For example, +** +** fts3PrefixCompress("abc", 3, "abcdef", 6) // returns 3 +** fts3PrefixCompress("abX", 3, "abcdef", 6) // returns 2 +** fts3PrefixCompress("abX", 3, "Xbcdef", 6) // returns 0 +*/ +static int fts3PrefixCompress( + const char *zPrev, /* Buffer containing previous term */ + int nPrev, /* Size of buffer zPrev in bytes */ + const char *zNext, /* Buffer containing next term */ + int nNext /* Size of buffer zNext in bytes */ +){ + int n; + UNUSED_PARAMETER(nNext); + for(n=0; nnData; /* Current size of node in bytes */ + int nReq = nData; /* Required space after adding zTerm */ + int nPrefix; /* Number of bytes of prefix compression */ + int nSuffix; /* Suffix length */ + + nPrefix = fts3PrefixCompress(pTree->zTerm, pTree->nTerm, zTerm, nTerm); + nSuffix = nTerm-nPrefix; + + nReq += sqlite3Fts3VarintLen(nPrefix)+sqlite3Fts3VarintLen(nSuffix)+nSuffix; + if( nReq<=p->nNodeSize || !pTree->zTerm ){ + + if( nReq>p->nNodeSize ){ + /* An unusual case: this is the first term to be added to the node + ** and the static node buffer (p->nNodeSize bytes) is not large + ** enough. Use a separately malloced buffer instead This wastes + ** p->nNodeSize bytes, but since this scenario only comes about when + ** the database contain two terms that share a prefix of almost 2KB, + ** this is not expected to be a serious problem. + */ + assert( pTree->aData==(char *)&pTree[1] ); + pTree->aData = (char *)sqlite3_malloc(nReq); + if( !pTree->aData ){ + return SQLITE_NOMEM; + } + } + + if( pTree->zTerm ){ + /* There is no prefix-length field for first term in a node */ + nData += sqlite3Fts3PutVarint(&pTree->aData[nData], nPrefix); + } + + nData += sqlite3Fts3PutVarint(&pTree->aData[nData], nSuffix); + memcpy(&pTree->aData[nData], &zTerm[nPrefix], nSuffix); + pTree->nData = nData + nSuffix; + pTree->nEntry++; + + if( isCopyTerm ){ + if( pTree->nMalloczMalloc, nTerm*2); + if( !zNew ){ + return SQLITE_NOMEM; + } + pTree->nMalloc = nTerm*2; + pTree->zMalloc = zNew; + } + pTree->zTerm = pTree->zMalloc; + memcpy(pTree->zTerm, zTerm, nTerm); + pTree->nTerm = nTerm; + }else{ + pTree->zTerm = (char *)zTerm; + pTree->nTerm = nTerm; + } + return SQLITE_OK; + } + } + + /* If control flows to here, it was not possible to append zTerm to the + ** current node. Create a new node (a right-sibling of the current node). + ** If this is the first node in the tree, the term is added to it. + ** + ** Otherwise, the term is not added to the new node, it is left empty for + ** now. Instead, the term is inserted into the parent of pTree. If pTree + ** has no parent, one is created here. + */ + pNew = (SegmentNode *)sqlite3_malloc(sizeof(SegmentNode) + p->nNodeSize); + if( !pNew ){ + return SQLITE_NOMEM; + } + memset(pNew, 0, sizeof(SegmentNode)); + pNew->nData = 1 + FTS3_VARINT_MAX; + pNew->aData = (char *)&pNew[1]; + + if( pTree ){ + SegmentNode *pParent = pTree->pParent; + rc = fts3NodeAddTerm(p, &pParent, isCopyTerm, zTerm, nTerm); + if( pTree->pParent==0 ){ + pTree->pParent = pParent; + } + pTree->pRight = pNew; + pNew->pLeftmost = pTree->pLeftmost; + pNew->pParent = pParent; + pNew->zMalloc = pTree->zMalloc; + pNew->nMalloc = pTree->nMalloc; + pTree->zMalloc = 0; + }else{ + pNew->pLeftmost = pNew; + rc = fts3NodeAddTerm(p, &pNew, isCopyTerm, zTerm, nTerm); + } + + *ppTree = pNew; + return rc; +} + +/* +** Helper function for fts3NodeWrite(). +*/ +static int fts3TreeFinishNode( + SegmentNode *pTree, + int iHeight, + sqlite3_int64 iLeftChild +){ + int nStart; + assert( iHeight>=1 && iHeight<128 ); + nStart = FTS3_VARINT_MAX - sqlite3Fts3VarintLen(iLeftChild); + pTree->aData[nStart] = (char)iHeight; + sqlite3Fts3PutVarint(&pTree->aData[nStart+1], iLeftChild); + return nStart; +} + +/* +** Write the buffer for the segment node pTree and all of its peers to the +** database. Then call this function recursively to write the parent of +** pTree and its peers to the database. +** +** Except, if pTree is a root node, do not write it to the database. Instead, +** set output variables *paRoot and *pnRoot to contain the root node. +** +** If successful, SQLITE_OK is returned and output variable *piLast is +** set to the largest blockid written to the database (or zero if no +** blocks were written to the db). Otherwise, an SQLite error code is +** returned. +*/ +static int fts3NodeWrite( + Fts3Table *p, /* Virtual table handle */ + SegmentNode *pTree, /* SegmentNode handle */ + int iHeight, /* Height of this node in tree */ + sqlite3_int64 iLeaf, /* Block id of first leaf node */ + sqlite3_int64 iFree, /* Block id of next free slot in %_segments */ + sqlite3_int64 *piLast, /* OUT: Block id of last entry written */ + char **paRoot, /* OUT: Data for root node */ + int *pnRoot /* OUT: Size of root node in bytes */ +){ + int rc = SQLITE_OK; + + if( !pTree->pParent ){ + /* Root node of the tree. */ + int nStart = fts3TreeFinishNode(pTree, iHeight, iLeaf); + *piLast = iFree-1; + *pnRoot = pTree->nData - nStart; + *paRoot = &pTree->aData[nStart]; + }else{ + SegmentNode *pIter; + sqlite3_int64 iNextFree = iFree; + sqlite3_int64 iNextLeaf = iLeaf; + for(pIter=pTree->pLeftmost; pIter && rc==SQLITE_OK; pIter=pIter->pRight){ + int nStart = fts3TreeFinishNode(pIter, iHeight, iNextLeaf); + int nWrite = pIter->nData - nStart; + + rc = fts3WriteSegment(p, iNextFree, &pIter->aData[nStart], nWrite); + iNextFree++; + iNextLeaf += (pIter->nEntry+1); + } + if( rc==SQLITE_OK ){ + assert( iNextLeaf==iFree ); + rc = fts3NodeWrite( + p, pTree->pParent, iHeight+1, iFree, iNextFree, piLast, paRoot, pnRoot + ); + } + } + + return rc; +} + +/* +** Free all memory allocations associated with the tree pTree. +*/ +static void fts3NodeFree(SegmentNode *pTree){ + if( pTree ){ + SegmentNode *p = pTree->pLeftmost; + fts3NodeFree(p->pParent); + while( p ){ + SegmentNode *pRight = p->pRight; + if( p->aData!=(char *)&p[1] ){ + sqlite3_free(p->aData); + } + assert( pRight==0 || p->zMalloc==0 ); + sqlite3_free(p->zMalloc); + sqlite3_free(p); + p = pRight; + } + } +} + +/* +** Add a term to the segment being constructed by the SegmentWriter object +** *ppWriter. When adding the first term to a segment, *ppWriter should +** be passed NULL. This function will allocate a new SegmentWriter object +** and return it via the input/output variable *ppWriter in this case. +** +** If successful, SQLITE_OK is returned. Otherwise, an SQLite error code. +*/ +static int fts3SegWriterAdd( + Fts3Table *p, /* Virtual table handle */ + SegmentWriter **ppWriter, /* IN/OUT: SegmentWriter handle */ + int isCopyTerm, /* True if buffer zTerm must be copied */ + const char *zTerm, /* Pointer to buffer containing term */ + int nTerm, /* Size of term in bytes */ + const char *aDoclist, /* Pointer to buffer containing doclist */ + int nDoclist /* Size of doclist in bytes */ +){ + int nPrefix; /* Size of term prefix in bytes */ + int nSuffix; /* Size of term suffix in bytes */ + int nReq; /* Number of bytes required on leaf page */ + int nData; + SegmentWriter *pWriter = *ppWriter; + + if( !pWriter ){ + int rc; + sqlite3_stmt *pStmt; + + /* Allocate the SegmentWriter structure */ + pWriter = (SegmentWriter *)sqlite3_malloc(sizeof(SegmentWriter)); + if( !pWriter ) return SQLITE_NOMEM; + memset(pWriter, 0, sizeof(SegmentWriter)); + *ppWriter = pWriter; + + /* Allocate a buffer in which to accumulate data */ + pWriter->aData = (char *)sqlite3_malloc(p->nNodeSize); + if( !pWriter->aData ) return SQLITE_NOMEM; + pWriter->nSize = p->nNodeSize; + + /* Find the next free blockid in the %_segments table */ + rc = fts3SqlStmt(p, SQL_NEXT_SEGMENTS_ID, &pStmt, 0); + if( rc!=SQLITE_OK ) return rc; + if( SQLITE_ROW==sqlite3_step(pStmt) ){ + pWriter->iFree = sqlite3_column_int64(pStmt, 0); + pWriter->iFirst = pWriter->iFree; + } + rc = sqlite3_reset(pStmt); + if( rc!=SQLITE_OK ) return rc; + } + nData = pWriter->nData; + + nPrefix = fts3PrefixCompress(pWriter->zTerm, pWriter->nTerm, zTerm, nTerm); + nSuffix = nTerm-nPrefix; + + /* Figure out how many bytes are required by this new entry */ + nReq = sqlite3Fts3VarintLen(nPrefix) + /* varint containing prefix size */ + sqlite3Fts3VarintLen(nSuffix) + /* varint containing suffix size */ + nSuffix + /* Term suffix */ + sqlite3Fts3VarintLen(nDoclist) + /* Size of doclist */ + nDoclist; /* Doclist data */ + + if( nData>0 && nData+nReq>p->nNodeSize ){ + int rc; + + /* The current leaf node is full. Write it out to the database. */ + rc = fts3WriteSegment(p, pWriter->iFree++, pWriter->aData, nData); + if( rc!=SQLITE_OK ) return rc; + + /* Add the current term to the interior node tree. The term added to + ** the interior tree must: + ** + ** a) be greater than the largest term on the leaf node just written + ** to the database (still available in pWriter->zTerm), and + ** + ** b) be less than or equal to the term about to be added to the new + ** leaf node (zTerm/nTerm). + ** + ** In other words, it must be the prefix of zTerm 1 byte longer than + ** the common prefix (if any) of zTerm and pWriter->zTerm. + */ + assert( nPrefixpTree, isCopyTerm, zTerm, nPrefix+1); + if( rc!=SQLITE_OK ) return rc; + + nData = 0; + pWriter->nTerm = 0; + + nPrefix = 0; + nSuffix = nTerm; + nReq = 1 + /* varint containing prefix size */ + sqlite3Fts3VarintLen(nTerm) + /* varint containing suffix size */ + nTerm + /* Term suffix */ + sqlite3Fts3VarintLen(nDoclist) + /* Size of doclist */ + nDoclist; /* Doclist data */ + } + + /* If the buffer currently allocated is too small for this entry, realloc + ** the buffer to make it large enough. + */ + if( nReq>pWriter->nSize ){ + char *aNew = sqlite3_realloc(pWriter->aData, nReq); + if( !aNew ) return SQLITE_NOMEM; + pWriter->aData = aNew; + pWriter->nSize = nReq; + } + assert( nData+nReq<=pWriter->nSize ); + + /* Append the prefix-compressed term and doclist to the buffer. */ + nData += sqlite3Fts3PutVarint(&pWriter->aData[nData], nPrefix); + nData += sqlite3Fts3PutVarint(&pWriter->aData[nData], nSuffix); + memcpy(&pWriter->aData[nData], &zTerm[nPrefix], nSuffix); + nData += nSuffix; + nData += sqlite3Fts3PutVarint(&pWriter->aData[nData], nDoclist); + memcpy(&pWriter->aData[nData], aDoclist, nDoclist); + pWriter->nData = nData + nDoclist; + + /* Save the current term so that it can be used to prefix-compress the next. + ** If the isCopyTerm parameter is true, then the buffer pointed to by + ** zTerm is transient, so take a copy of the term data. Otherwise, just + ** store a copy of the pointer. + */ + if( isCopyTerm ){ + if( nTerm>pWriter->nMalloc ){ + char *zNew = sqlite3_realloc(pWriter->zMalloc, nTerm*2); + if( !zNew ){ + return SQLITE_NOMEM; + } + pWriter->nMalloc = nTerm*2; + pWriter->zMalloc = zNew; + pWriter->zTerm = zNew; + } + assert( pWriter->zTerm==pWriter->zMalloc ); + memcpy(pWriter->zTerm, zTerm, nTerm); + }else{ + pWriter->zTerm = (char *)zTerm; + } + pWriter->nTerm = nTerm; + + return SQLITE_OK; +} + +/* +** Flush all data associated with the SegmentWriter object pWriter to the +** database. This function must be called after all terms have been added +** to the segment using fts3SegWriterAdd(). If successful, SQLITE_OK is +** returned. Otherwise, an SQLite error code. +*/ +static int fts3SegWriterFlush( + Fts3Table *p, /* Virtual table handle */ + SegmentWriter *pWriter, /* SegmentWriter to flush to the db */ + int iLevel, /* Value for 'level' column of %_segdir */ + int iIdx /* Value for 'idx' column of %_segdir */ +){ + int rc; /* Return code */ + if( pWriter->pTree ){ + sqlite3_int64 iLast = 0; /* Largest block id written to database */ + sqlite3_int64 iLastLeaf; /* Largest leaf block id written to db */ + char *zRoot = NULL; /* Pointer to buffer containing root node */ + int nRoot = 0; /* Size of buffer zRoot */ + + iLastLeaf = pWriter->iFree; + rc = fts3WriteSegment(p, pWriter->iFree++, pWriter->aData, pWriter->nData); + if( rc==SQLITE_OK ){ + rc = fts3NodeWrite(p, pWriter->pTree, 1, + pWriter->iFirst, pWriter->iFree, &iLast, &zRoot, &nRoot); + } + if( rc==SQLITE_OK ){ + rc = fts3WriteSegdir( + p, iLevel, iIdx, pWriter->iFirst, iLastLeaf, iLast, zRoot, nRoot); + } + }else{ + /* The entire tree fits on the root node. Write it to the segdir table. */ + rc = fts3WriteSegdir( + p, iLevel, iIdx, 0, 0, 0, pWriter->aData, pWriter->nData); + } + return rc; +} + +/* +** Release all memory held by the SegmentWriter object passed as the +** first argument. +*/ +static void fts3SegWriterFree(SegmentWriter *pWriter){ + if( pWriter ){ + sqlite3_free(pWriter->aData); + sqlite3_free(pWriter->zMalloc); + fts3NodeFree(pWriter->pTree); + sqlite3_free(pWriter); + } +} + +/* +** The first value in the apVal[] array is assumed to contain an integer. +** This function tests if there exist any documents with docid values that +** are different from that integer. i.e. if deleting the document with docid +** pRowid would mean the FTS3 table were empty. +** +** If successful, *pisEmpty is set to true if the table is empty except for +** document pRowid, or false otherwise, and SQLITE_OK is returned. If an +** error occurs, an SQLite error code is returned. +*/ +static int fts3IsEmpty(Fts3Table *p, sqlite3_value *pRowid, int *pisEmpty){ + sqlite3_stmt *pStmt; + int rc; + if( p->zContentTbl ){ + /* If using the content=xxx option, assume the table is never empty */ + *pisEmpty = 0; + rc = SQLITE_OK; + }else{ + rc = fts3SqlStmt(p, SQL_IS_EMPTY, &pStmt, &pRowid); + if( rc==SQLITE_OK ){ + if( SQLITE_ROW==sqlite3_step(pStmt) ){ + *pisEmpty = sqlite3_column_int(pStmt, 0); + } + rc = sqlite3_reset(pStmt); + } + } + return rc; +} + +/* +** Set *pnMax to the largest segment level in the database for the index +** iIndex. +** +** Segment levels are stored in the 'level' column of the %_segdir table. +** +** Return SQLITE_OK if successful, or an SQLite error code if not. +*/ +static int fts3SegmentMaxLevel(Fts3Table *p, int iIndex, int *pnMax){ + sqlite3_stmt *pStmt; + int rc; + assert( iIndex>=0 && iIndexnIndex ); + + /* Set pStmt to the compiled version of: + ** + ** SELECT max(level) FROM %Q.'%q_segdir' WHERE level BETWEEN ? AND ? + ** + ** (1024 is actually the value of macro FTS3_SEGDIR_PREFIXLEVEL_STR). + */ + rc = fts3SqlStmt(p, SQL_SELECT_SEGDIR_MAX_LEVEL, &pStmt, 0); + if( rc!=SQLITE_OK ) return rc; + sqlite3_bind_int(pStmt, 1, iIndex*FTS3_SEGDIR_MAXLEVEL); + sqlite3_bind_int(pStmt, 2, (iIndex+1)*FTS3_SEGDIR_MAXLEVEL - 1); + if( SQLITE_ROW==sqlite3_step(pStmt) ){ + *pnMax = sqlite3_column_int(pStmt, 0); + } + return sqlite3_reset(pStmt); +} + +/* +** This function is used after merging multiple segments into a single large +** segment to delete the old, now redundant, segment b-trees. Specifically, +** it: +** +** 1) Deletes all %_segments entries for the segments associated with +** each of the SegReader objects in the array passed as the third +** argument, and +** +** 2) deletes all %_segdir entries with level iLevel, or all %_segdir +** entries regardless of level if (iLevel<0). +** +** SQLITE_OK is returned if successful, otherwise an SQLite error code. +*/ +static int fts3DeleteSegdir( + Fts3Table *p, /* Virtual table handle */ + int iIndex, /* Index for p->aIndex */ + int iLevel, /* Level of %_segdir entries to delete */ + Fts3SegReader **apSegment, /* Array of SegReader objects */ + int nReader /* Size of array apSegment */ +){ + int rc; /* Return Code */ + int i; /* Iterator variable */ + sqlite3_stmt *pDelete; /* SQL statement to delete rows */ + + rc = fts3SqlStmt(p, SQL_DELETE_SEGMENTS_RANGE, &pDelete, 0); + for(i=0; rc==SQLITE_OK && iiStartBlock ){ + sqlite3_bind_int64(pDelete, 1, pSegment->iStartBlock); + sqlite3_bind_int64(pDelete, 2, pSegment->iEndBlock); + sqlite3_step(pDelete); + rc = sqlite3_reset(pDelete); + } + } + if( rc!=SQLITE_OK ){ + return rc; + } + + assert( iLevel>=0 || iLevel==FTS3_SEGCURSOR_ALL ); + if( iLevel==FTS3_SEGCURSOR_ALL ){ + rc = fts3SqlStmt(p, SQL_DELETE_SEGDIR_RANGE, &pDelete, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int(pDelete, 1, iIndex*FTS3_SEGDIR_MAXLEVEL); + sqlite3_bind_int(pDelete, 2, (iIndex+1) * FTS3_SEGDIR_MAXLEVEL - 1); + } + }else{ + rc = fts3SqlStmt(p, SQL_DELETE_SEGDIR_LEVEL, &pDelete, 0); + if( rc==SQLITE_OK ){ + sqlite3_bind_int(pDelete, 1, iIndex*FTS3_SEGDIR_MAXLEVEL + iLevel); + } + } + + if( rc==SQLITE_OK ){ + sqlite3_step(pDelete); + rc = sqlite3_reset(pDelete); + } + + return rc; +} + +/* +** When this function is called, buffer *ppList (size *pnList bytes) contains +** a position list that may (or may not) feature multiple columns. This +** function adjusts the pointer *ppList and the length *pnList so that they +** identify the subset of the position list that corresponds to column iCol. +** +** If there are no entries in the input position list for column iCol, then +** *pnList is set to zero before returning. +*/ +static void fts3ColumnFilter( + int iCol, /* Column to filter on */ + char **ppList, /* IN/OUT: Pointer to position list */ + int *pnList /* IN/OUT: Size of buffer *ppList in bytes */ +){ + char *pList = *ppList; + int nList = *pnList; + char *pEnd = &pList[nList]; + int iCurrent = 0; + char *p = pList; + + assert( iCol>=0 ); + while( 1 ){ + char c = 0; + while( ppMsr->nBuffer ){ + char *pNew; + pMsr->nBuffer = nList*2; + pNew = (char *)sqlite3_realloc(pMsr->aBuffer, pMsr->nBuffer); + if( !pNew ) return SQLITE_NOMEM; + pMsr->aBuffer = pNew; + } + + memcpy(pMsr->aBuffer, pList, nList); + return SQLITE_OK; +} + +int sqlite3Fts3MsrIncrNext( + Fts3Table *p, /* Virtual table handle */ + Fts3MultiSegReader *pMsr, /* Multi-segment-reader handle */ + sqlite3_int64 *piDocid, /* OUT: Docid value */ + char **paPoslist, /* OUT: Pointer to position list */ + int *pnPoslist /* OUT: Size of position list in bytes */ +){ + int nMerge = pMsr->nAdvance; + Fts3SegReader **apSegment = pMsr->apSegment; + int (*xCmp)(Fts3SegReader *, Fts3SegReader *) = ( + p->bDescIdx ? fts3SegReaderDoclistCmpRev : fts3SegReaderDoclistCmp + ); + + if( nMerge==0 ){ + *paPoslist = 0; + return SQLITE_OK; + } + + while( 1 ){ + Fts3SegReader *pSeg; + pSeg = pMsr->apSegment[0]; + + if( pSeg->pOffsetList==0 ){ + *paPoslist = 0; + break; + }else{ + int rc; + char *pList; + int nList; + int j; + sqlite3_int64 iDocid = apSegment[0]->iDocid; + + rc = fts3SegReaderNextDocid(p, apSegment[0], &pList, &nList); + j = 1; + while( rc==SQLITE_OK + && jpOffsetList + && apSegment[j]->iDocid==iDocid + ){ + rc = fts3SegReaderNextDocid(p, apSegment[j], 0, 0); + j++; + } + if( rc!=SQLITE_OK ) return rc; + fts3SegReaderSort(pMsr->apSegment, nMerge, j, xCmp); + + if( pMsr->iColFilter>=0 ){ + fts3ColumnFilter(pMsr->iColFilter, &pList, &nList); + } + + if( nList>0 ){ + if( fts3SegReaderIsPending(apSegment[0]) ){ + rc = fts3MsrBufferData(pMsr, pList, nList+1); + if( rc!=SQLITE_OK ) return rc; + *paPoslist = pMsr->aBuffer; + assert( (pMsr->aBuffer[nList] & 0xFE)==0x00 ); + }else{ + *paPoslist = pList; + } + *piDocid = iDocid; + *pnPoslist = nList; + break; + } + } + } + + return SQLITE_OK; +} + +static int fts3SegReaderStart( + Fts3Table *p, /* Virtual table handle */ + Fts3MultiSegReader *pCsr, /* Cursor object */ + const char *zTerm, /* Term searched for (or NULL) */ + int nTerm /* Length of zTerm in bytes */ +){ + int i; + int nSeg = pCsr->nSegment; + + /* If the Fts3SegFilter defines a specific term (or term prefix) to search + ** for, then advance each segment iterator until it points to a term of + ** equal or greater value than the specified term. This prevents many + ** unnecessary merge/sort operations for the case where single segment + ** b-tree leaf nodes contain more than one term. + */ + for(i=0; pCsr->bRestart==0 && inSegment; i++){ + Fts3SegReader *pSeg = pCsr->apSegment[i]; + do { + int rc = fts3SegReaderNext(p, pSeg, 0); + if( rc!=SQLITE_OK ) return rc; + }while( zTerm && fts3SegReaderTermCmp(pSeg, zTerm, nTerm)<0 ); + } + fts3SegReaderSort(pCsr->apSegment, nSeg, nSeg, fts3SegReaderCmp); + + return SQLITE_OK; +} + +int sqlite3Fts3SegReaderStart( + Fts3Table *p, /* Virtual table handle */ + Fts3MultiSegReader *pCsr, /* Cursor object */ + Fts3SegFilter *pFilter /* Restrictions on range of iteration */ +){ + pCsr->pFilter = pFilter; + return fts3SegReaderStart(p, pCsr, pFilter->zTerm, pFilter->nTerm); +} + +int sqlite3Fts3MsrIncrStart( + Fts3Table *p, /* Virtual table handle */ + Fts3MultiSegReader *pCsr, /* Cursor object */ + int iCol, /* Column to match on. */ + const char *zTerm, /* Term to iterate through a doclist for */ + int nTerm /* Number of bytes in zTerm */ +){ + int i; + int rc; + int nSegment = pCsr->nSegment; + int (*xCmp)(Fts3SegReader *, Fts3SegReader *) = ( + p->bDescIdx ? fts3SegReaderDoclistCmpRev : fts3SegReaderDoclistCmp + ); + + assert( pCsr->pFilter==0 ); + assert( zTerm && nTerm>0 ); + + /* Advance each segment iterator until it points to the term zTerm/nTerm. */ + rc = fts3SegReaderStart(p, pCsr, zTerm, nTerm); + if( rc!=SQLITE_OK ) return rc; + + /* Determine how many of the segments actually point to zTerm/nTerm. */ + for(i=0; iapSegment[i]; + if( !pSeg->aNode || fts3SegReaderTermCmp(pSeg, zTerm, nTerm) ){ + break; + } + } + pCsr->nAdvance = i; + + /* Advance each of the segments to point to the first docid. */ + for(i=0; inAdvance; i++){ + rc = fts3SegReaderFirstDocid(p, pCsr->apSegment[i]); + if( rc!=SQLITE_OK ) return rc; + } + fts3SegReaderSort(pCsr->apSegment, i, i, xCmp); + + assert( iCol<0 || iColnColumn ); + pCsr->iColFilter = iCol; + + return SQLITE_OK; +} + +/* +** This function is called on a MultiSegReader that has been started using +** sqlite3Fts3MsrIncrStart(). One or more calls to MsrIncrNext() may also +** have been made. Calling this function puts the MultiSegReader in such +** a state that if the next two calls are: +** +** sqlite3Fts3SegReaderStart() +** sqlite3Fts3SegReaderStep() +** +** then the entire doclist for the term is available in +** MultiSegReader.aDoclist/nDoclist. +*/ +int sqlite3Fts3MsrIncrRestart(Fts3MultiSegReader *pCsr){ + int i; /* Used to iterate through segment-readers */ + + assert( pCsr->zTerm==0 ); + assert( pCsr->nTerm==0 ); + assert( pCsr->aDoclist==0 ); + assert( pCsr->nDoclist==0 ); + + pCsr->nAdvance = 0; + pCsr->bRestart = 1; + for(i=0; inSegment; i++){ + pCsr->apSegment[i]->pOffsetList = 0; + pCsr->apSegment[i]->nOffsetList = 0; + pCsr->apSegment[i]->iDocid = 0; + } + + return SQLITE_OK; +} + + +int sqlite3Fts3SegReaderStep( + Fts3Table *p, /* Virtual table handle */ + Fts3MultiSegReader *pCsr /* Cursor object */ +){ + int rc = SQLITE_OK; + + int isIgnoreEmpty = (pCsr->pFilter->flags & FTS3_SEGMENT_IGNORE_EMPTY); + int isRequirePos = (pCsr->pFilter->flags & FTS3_SEGMENT_REQUIRE_POS); + int isColFilter = (pCsr->pFilter->flags & FTS3_SEGMENT_COLUMN_FILTER); + int isPrefix = (pCsr->pFilter->flags & FTS3_SEGMENT_PREFIX); + int isScan = (pCsr->pFilter->flags & FTS3_SEGMENT_SCAN); + int isFirst = (pCsr->pFilter->flags & FTS3_SEGMENT_FIRST); + + Fts3SegReader **apSegment = pCsr->apSegment; + int nSegment = pCsr->nSegment; + Fts3SegFilter *pFilter = pCsr->pFilter; + int (*xCmp)(Fts3SegReader *, Fts3SegReader *) = ( + p->bDescIdx ? fts3SegReaderDoclistCmpRev : fts3SegReaderDoclistCmp + ); + + if( pCsr->nSegment==0 ) return SQLITE_OK; + + do { + int nMerge; + int i; + + /* Advance the first pCsr->nAdvance entries in the apSegment[] array + ** forward. Then sort the list in order of current term again. + */ + for(i=0; inAdvance; i++){ + rc = fts3SegReaderNext(p, apSegment[i], 0); + if( rc!=SQLITE_OK ) return rc; + } + fts3SegReaderSort(apSegment, nSegment, pCsr->nAdvance, fts3SegReaderCmp); + pCsr->nAdvance = 0; + + /* If all the seg-readers are at EOF, we're finished. return SQLITE_OK. */ + assert( rc==SQLITE_OK ); + if( apSegment[0]->aNode==0 ) break; + + pCsr->nTerm = apSegment[0]->nTerm; + pCsr->zTerm = apSegment[0]->zTerm; + + /* If this is a prefix-search, and if the term that apSegment[0] points + ** to does not share a suffix with pFilter->zTerm/nTerm, then all + ** required callbacks have been made. In this case exit early. + ** + ** Similarly, if this is a search for an exact match, and the first term + ** of segment apSegment[0] is not a match, exit early. + */ + if( pFilter->zTerm && !isScan ){ + if( pCsr->nTermnTerm + || (!isPrefix && pCsr->nTerm>pFilter->nTerm) + || memcmp(pCsr->zTerm, pFilter->zTerm, pFilter->nTerm) + ){ + break; + } + } + + nMerge = 1; + while( nMergeaNode + && apSegment[nMerge]->nTerm==pCsr->nTerm + && 0==memcmp(pCsr->zTerm, apSegment[nMerge]->zTerm, pCsr->nTerm) + ){ + nMerge++; + } + + assert( isIgnoreEmpty || (isRequirePos && !isColFilter) ); + if( nMerge==1 + && !isIgnoreEmpty + && !isFirst + && (p->bDescIdx==0 || fts3SegReaderIsPending(apSegment[0])==0) + ){ + pCsr->nDoclist = apSegment[0]->nDoclist; + if( fts3SegReaderIsPending(apSegment[0]) ){ + rc = fts3MsrBufferData(pCsr, apSegment[0]->aDoclist, pCsr->nDoclist); + pCsr->aDoclist = pCsr->aBuffer; + }else{ + pCsr->aDoclist = apSegment[0]->aDoclist; + } + if( rc==SQLITE_OK ) rc = SQLITE_ROW; + }else{ + int nDoclist = 0; /* Size of doclist */ + sqlite3_int64 iPrev = 0; /* Previous docid stored in doclist */ + + /* The current term of the first nMerge entries in the array + ** of Fts3SegReader objects is the same. The doclists must be merged + ** and a single term returned with the merged doclist. + */ + for(i=0; ipOffsetList ){ + int j; /* Number of segments that share a docid */ + char *pList; + int nList; + int nByte; + sqlite3_int64 iDocid = apSegment[0]->iDocid; + fts3SegReaderNextDocid(p, apSegment[0], &pList, &nList); + j = 1; + while( jpOffsetList + && apSegment[j]->iDocid==iDocid + ){ + fts3SegReaderNextDocid(p, apSegment[j], 0, 0); + j++; + } + + if( isColFilter ){ + fts3ColumnFilter(pFilter->iCol, &pList, &nList); + } + + if( !isIgnoreEmpty || nList>0 ){ + + /* Calculate the 'docid' delta value to write into the merged + ** doclist. */ + sqlite3_int64 iDelta; + if( p->bDescIdx && nDoclist>0 ){ + iDelta = iPrev - iDocid; + }else{ + iDelta = iDocid - iPrev; + } + assert( iDelta>0 || (nDoclist==0 && iDelta==iDocid) ); + assert( nDoclist>0 || iDelta==iDocid ); + + nByte = sqlite3Fts3VarintLen(iDelta) + (isRequirePos?nList+1:0); + if( nDoclist+nByte>pCsr->nBuffer ){ + char *aNew; + pCsr->nBuffer = (nDoclist+nByte)*2; + aNew = sqlite3_realloc(pCsr->aBuffer, pCsr->nBuffer); + if( !aNew ){ + return SQLITE_NOMEM; + } + pCsr->aBuffer = aNew; + } + + if( isFirst ){ + char *a = &pCsr->aBuffer[nDoclist]; + int nWrite; + + nWrite = sqlite3Fts3FirstFilter(iDelta, pList, nList, a); + if( nWrite ){ + iPrev = iDocid; + nDoclist += nWrite; + } + }else{ + nDoclist += sqlite3Fts3PutVarint(&pCsr->aBuffer[nDoclist], iDelta); + iPrev = iDocid; + if( isRequirePos ){ + memcpy(&pCsr->aBuffer[nDoclist], pList, nList); + nDoclist += nList; + pCsr->aBuffer[nDoclist++] = '\0'; + } + } + } + + fts3SegReaderSort(apSegment, nMerge, j, xCmp); + } + if( nDoclist>0 ){ + pCsr->aDoclist = pCsr->aBuffer; + pCsr->nDoclist = nDoclist; + rc = SQLITE_ROW; + } + } + pCsr->nAdvance = nMerge; + }while( rc==SQLITE_OK ); + + return rc; +} + + +void sqlite3Fts3SegReaderFinish( + Fts3MultiSegReader *pCsr /* Cursor object */ +){ + if( pCsr ){ + int i; + for(i=0; inSegment; i++){ + sqlite3Fts3SegReaderFree(pCsr->apSegment[i]); + } + sqlite3_free(pCsr->apSegment); + sqlite3_free(pCsr->aBuffer); + + pCsr->nSegment = 0; + pCsr->apSegment = 0; + pCsr->aBuffer = 0; + } +} + +/* +** Merge all level iLevel segments in the database into a single +** iLevel+1 segment. Or, if iLevel<0, merge all segments into a +** single segment with a level equal to the numerically largest level +** currently present in the database. +** +** If this function is called with iLevel<0, but there is only one +** segment in the database, SQLITE_DONE is returned immediately. +** Otherwise, if successful, SQLITE_OK is returned. If an error occurs, +** an SQLite error code is returned. +*/ +static int fts3SegmentMerge(Fts3Table *p, int iIndex, int iLevel){ + int rc; /* Return code */ + int iIdx = 0; /* Index of new segment */ + int iNewLevel = 0; /* Level/index to create new segment at */ + SegmentWriter *pWriter = 0; /* Used to write the new, merged, segment */ + Fts3SegFilter filter; /* Segment term filter condition */ + Fts3MultiSegReader csr; /* Cursor to iterate through level(s) */ + int bIgnoreEmpty = 0; /* True to ignore empty segments */ + + assert( iLevel==FTS3_SEGCURSOR_ALL + || iLevel==FTS3_SEGCURSOR_PENDING + || iLevel>=0 + ); + assert( iLevel=0 && iIndexnIndex ); + + rc = sqlite3Fts3SegReaderCursor(p, iIndex, iLevel, 0, 0, 1, 0, &csr); + if( rc!=SQLITE_OK || csr.nSegment==0 ) goto finished; + + if( iLevel==FTS3_SEGCURSOR_ALL ){ + /* This call is to merge all segments in the database to a single + ** segment. The level of the new segment is equal to the the numerically + ** greatest segment level currently present in the database for this + ** index. The idx of the new segment is always 0. */ + if( csr.nSegment==1 ){ + rc = SQLITE_DONE; + goto finished; + } + rc = fts3SegmentMaxLevel(p, iIndex, &iNewLevel); + bIgnoreEmpty = 1; + + }else if( iLevel==FTS3_SEGCURSOR_PENDING ){ + iNewLevel = iIndex * FTS3_SEGDIR_MAXLEVEL; + rc = fts3AllocateSegdirIdx(p, iIndex, 0, &iIdx); + }else{ + /* This call is to merge all segments at level iLevel. find the next + ** available segment index at level iLevel+1. The call to + ** fts3AllocateSegdirIdx() will merge the segments at level iLevel+1 to + ** a single iLevel+2 segment if necessary. */ + rc = fts3AllocateSegdirIdx(p, iIndex, iLevel+1, &iIdx); + iNewLevel = iIndex * FTS3_SEGDIR_MAXLEVEL + iLevel+1; + } + if( rc!=SQLITE_OK ) goto finished; + assert( csr.nSegment>0 ); + assert( iNewLevel>=(iIndex*FTS3_SEGDIR_MAXLEVEL) ); + assert( iNewLevel<((iIndex+1)*FTS3_SEGDIR_MAXLEVEL) ); + + memset(&filter, 0, sizeof(Fts3SegFilter)); + filter.flags = FTS3_SEGMENT_REQUIRE_POS; + filter.flags |= (bIgnoreEmpty ? FTS3_SEGMENT_IGNORE_EMPTY : 0); + + rc = sqlite3Fts3SegReaderStart(p, &csr, &filter); + while( SQLITE_OK==rc ){ + rc = sqlite3Fts3SegReaderStep(p, &csr); + if( rc!=SQLITE_ROW ) break; + rc = fts3SegWriterAdd(p, &pWriter, 1, + csr.zTerm, csr.nTerm, csr.aDoclist, csr.nDoclist); + } + if( rc!=SQLITE_OK ) goto finished; + assert( pWriter ); + + if( iLevel!=FTS3_SEGCURSOR_PENDING ){ + rc = fts3DeleteSegdir(p, iIndex, iLevel, csr.apSegment, csr.nSegment); + if( rc!=SQLITE_OK ) goto finished; + } + rc = fts3SegWriterFlush(p, pWriter, iNewLevel, iIdx); + + finished: + fts3SegWriterFree(pWriter); + sqlite3Fts3SegReaderFinish(&csr); + return rc; +} + + +/* +** Flush the contents of pendingTerms to level 0 segments. +*/ +int sqlite3Fts3PendingTermsFlush(Fts3Table *p){ + int rc = SQLITE_OK; + int i; + for(i=0; rc==SQLITE_OK && inIndex; i++){ + rc = fts3SegmentMerge(p, i, FTS3_SEGCURSOR_PENDING); + if( rc==SQLITE_DONE ) rc = SQLITE_OK; + } + sqlite3Fts3PendingTermsClear(p); + return rc; +} + +/* +** Encode N integers as varints into a blob. +*/ +static void fts3EncodeIntArray( + int N, /* The number of integers to encode */ + u32 *a, /* The integer values */ + char *zBuf, /* Write the BLOB here */ + int *pNBuf /* Write number of bytes if zBuf[] used here */ +){ + int i, j; + for(i=j=0; iiPrevDocid. The sizes are encoded as +** a blob of varints. +*/ +static void fts3InsertDocsize( + int *pRC, /* Result code */ + Fts3Table *p, /* Table into which to insert */ + u32 *aSz /* Sizes of each column, in tokens */ +){ + char *pBlob; /* The BLOB encoding of the document size */ + int nBlob; /* Number of bytes in the BLOB */ + sqlite3_stmt *pStmt; /* Statement used to insert the encoding */ + int rc; /* Result code from subfunctions */ + + if( *pRC ) return; + pBlob = sqlite3_malloc( 10*p->nColumn ); + if( pBlob==0 ){ + *pRC = SQLITE_NOMEM; + return; + } + fts3EncodeIntArray(p->nColumn, aSz, pBlob, &nBlob); + rc = fts3SqlStmt(p, SQL_REPLACE_DOCSIZE, &pStmt, 0); + if( rc ){ + sqlite3_free(pBlob); + *pRC = rc; + return; + } + sqlite3_bind_int64(pStmt, 1, p->iPrevDocid); + sqlite3_bind_blob(pStmt, 2, pBlob, nBlob, sqlite3_free); + sqlite3_step(pStmt); + *pRC = sqlite3_reset(pStmt); +} + +/* +** Record 0 of the %_stat table contains a blob consisting of N varints, +** where N is the number of user defined columns in the fts3 table plus +** two. If nCol is the number of user defined columns, then values of the +** varints are set as follows: +** +** Varint 0: Total number of rows in the table. +** +** Varint 1..nCol: For each column, the total number of tokens stored in +** the column for all rows of the table. +** +** Varint 1+nCol: The total size, in bytes, of all text values in all +** columns of all rows of the table. +** +*/ +static void fts3UpdateDocTotals( + int *pRC, /* The result code */ + Fts3Table *p, /* Table being updated */ + u32 *aSzIns, /* Size increases */ + u32 *aSzDel, /* Size decreases */ + int nChng /* Change in the number of documents */ +){ + char *pBlob; /* Storage for BLOB written into %_stat */ + int nBlob; /* Size of BLOB written into %_stat */ + u32 *a; /* Array of integers that becomes the BLOB */ + sqlite3_stmt *pStmt; /* Statement for reading and writing */ + int i; /* Loop counter */ + int rc; /* Result code from subfunctions */ + + const int nStat = p->nColumn+2; + + if( *pRC ) return; + a = sqlite3_malloc( (sizeof(u32)+10)*nStat ); + if( a==0 ){ + *pRC = SQLITE_NOMEM; + return; + } + pBlob = (char*)&a[nStat]; + rc = fts3SqlStmt(p, SQL_SELECT_DOCTOTAL, &pStmt, 0); + if( rc ){ + sqlite3_free(a); + *pRC = rc; + return; + } + if( sqlite3_step(pStmt)==SQLITE_ROW ){ + fts3DecodeIntArray(nStat, a, + sqlite3_column_blob(pStmt, 0), + sqlite3_column_bytes(pStmt, 0)); + }else{ + memset(a, 0, sizeof(u32)*(nStat) ); + } + sqlite3_reset(pStmt); + if( nChng<0 && a[0]<(u32)(-nChng) ){ + a[0] = 0; + }else{ + a[0] += nChng; + } + for(i=0; inColumn+1; i++){ + u32 x = a[i+1]; + if( x+aSzIns[i] < aSzDel[i] ){ + x = 0; + }else{ + x = x + aSzIns[i] - aSzDel[i]; + } + a[i+1] = x; + } + fts3EncodeIntArray(nStat, a, pBlob, &nBlob); + rc = fts3SqlStmt(p, SQL_REPLACE_DOCTOTAL, &pStmt, 0); + if( rc ){ + sqlite3_free(a); + *pRC = rc; + return; + } + sqlite3_bind_blob(pStmt, 1, pBlob, nBlob, SQLITE_STATIC); + sqlite3_step(pStmt); + *pRC = sqlite3_reset(pStmt); + sqlite3_free(a); +} + +static int fts3DoOptimize(Fts3Table *p, int bReturnDone){ + int i; + int bSeenDone = 0; + int rc = SQLITE_OK; + for(i=0; rc==SQLITE_OK && inIndex; i++){ + rc = fts3SegmentMerge(p, i, FTS3_SEGCURSOR_ALL); + if( rc==SQLITE_DONE ){ + bSeenDone = 1; + rc = SQLITE_OK; + } + } + sqlite3Fts3SegmentsClose(p); + sqlite3Fts3PendingTermsClear(p); + + return (rc==SQLITE_OK && bReturnDone && bSeenDone) ? SQLITE_DONE : rc; +} + +/* +** This function is called when the user executes the following statement: +** +** INSERT INTO () VALUES('rebuild'); +** +** The entire FTS index is discarded and rebuilt. If the table is one +** created using the content=xxx option, then the new index is based on +** the current contents of the xxx table. Otherwise, it is rebuilt based +** on the contents of the %_content table. +*/ +static int fts3DoRebuild(Fts3Table *p){ + int rc; /* Return Code */ + + rc = fts3DeleteAll(p, 0); + if( rc==SQLITE_OK ){ + u32 *aSz = 0; + u32 *aSzIns = 0; + u32 *aSzDel = 0; + sqlite3_stmt *pStmt = 0; + int nEntry = 0; + + /* Compose and prepare an SQL statement to loop through the content table */ + char *zSql = sqlite3_mprintf("SELECT %s" , p->zReadExprlist); + if( !zSql ){ + rc = SQLITE_NOMEM; + }else{ + rc = sqlite3_prepare_v2(p->db, zSql, -1, &pStmt, 0); + sqlite3_free(zSql); + } + + if( rc==SQLITE_OK ){ + int nByte = sizeof(u32) * (p->nColumn+1)*3; + aSz = (u32 *)sqlite3_malloc(nByte); + if( aSz==0 ){ + rc = SQLITE_NOMEM; + }else{ + memset(aSz, 0, nByte); + aSzIns = &aSz[p->nColumn+1]; + aSzDel = &aSzIns[p->nColumn+1]; + } + } + + while( rc==SQLITE_OK && SQLITE_ROW==sqlite3_step(pStmt) ){ + int iCol; + rc = fts3PendingTermsDocid(p, sqlite3_column_int64(pStmt, 0)); + aSz[p->nColumn] = 0; + for(iCol=0; rc==SQLITE_OK && iColnColumn; iCol++){ + const char *z = (const char *) sqlite3_column_text(pStmt, iCol+1); + rc = fts3PendingTermsAdd(p, z, iCol, &aSz[iCol]); + aSz[p->nColumn] += sqlite3_column_bytes(pStmt, iCol+1); + } + if( p->bHasDocsize ){ + fts3InsertDocsize(&rc, p, aSz); + } + if( rc!=SQLITE_OK ){ + sqlite3_finalize(pStmt); + pStmt = 0; + }else{ + nEntry++; + for(iCol=0; iCol<=p->nColumn; iCol++){ + aSzIns[iCol] += aSz[iCol]; + } + } + } + if( p->bHasStat ){ + fts3UpdateDocTotals(&rc, p, aSzIns, aSzDel, nEntry); + } + sqlite3_free(aSz); + + if( pStmt ){ + int rc2 = sqlite3_finalize(pStmt); + if( rc==SQLITE_OK ){ + rc = rc2; + } + } + } + + return rc; +} + +/* +** Handle a 'special' INSERT of the form: +** +** "INSERT INTO tbl(tbl) VALUES()" +** +** Argument pVal contains the result of . Currently the only +** meaningful value to insert is the text 'optimize'. +*/ +static int fts3SpecialInsert(Fts3Table *p, sqlite3_value *pVal){ + int rc; /* Return Code */ + const char *zVal = (const char *)sqlite3_value_text(pVal); + int nVal = sqlite3_value_bytes(pVal); + + if( !zVal ){ + return SQLITE_NOMEM; + }else if( nVal==8 && 0==sqlite3_strnicmp(zVal, "optimize", 8) ){ + rc = fts3DoOptimize(p, 0); + }else if( nVal==7 && 0==sqlite3_strnicmp(zVal, "rebuild", 7) ){ + rc = fts3DoRebuild(p); +#ifdef SQLITE_TEST + }else if( nVal>9 && 0==sqlite3_strnicmp(zVal, "nodesize=", 9) ){ + p->nNodeSize = atoi(&zVal[9]); + rc = SQLITE_OK; + }else if( nVal>11 && 0==sqlite3_strnicmp(zVal, "maxpending=", 9) ){ + p->nMaxPendingData = atoi(&zVal[11]); + rc = SQLITE_OK; +#endif + }else{ + rc = SQLITE_ERROR; + } + + return rc; +} + +/* +** Delete all cached deferred doclists. Deferred doclists are cached +** (allocated) by the sqlite3Fts3CacheDeferredDoclists() function. +*/ +void sqlite3Fts3FreeDeferredDoclists(Fts3Cursor *pCsr){ + Fts3DeferredToken *pDef; + for(pDef=pCsr->pDeferred; pDef; pDef=pDef->pNext){ + fts3PendingListDelete(pDef->pList); + pDef->pList = 0; + } +} + +/* +** Free all entries in the pCsr->pDeffered list. Entries are added to +** this list using sqlite3Fts3DeferToken(). +*/ +void sqlite3Fts3FreeDeferredTokens(Fts3Cursor *pCsr){ + Fts3DeferredToken *pDef; + Fts3DeferredToken *pNext; + for(pDef=pCsr->pDeferred; pDef; pDef=pNext){ + pNext = pDef->pNext; + fts3PendingListDelete(pDef->pList); + sqlite3_free(pDef); + } + pCsr->pDeferred = 0; +} + +/* +** Generate deferred-doclists for all tokens in the pCsr->pDeferred list +** based on the row that pCsr currently points to. +** +** A deferred-doclist is like any other doclist with position information +** included, except that it only contains entries for a single row of the +** table, not for all rows. +*/ +int sqlite3Fts3CacheDeferredDoclists(Fts3Cursor *pCsr){ + int rc = SQLITE_OK; /* Return code */ + if( pCsr->pDeferred ){ + int i; /* Used to iterate through table columns */ + sqlite3_int64 iDocid; /* Docid of the row pCsr points to */ + Fts3DeferredToken *pDef; /* Used to iterate through deferred tokens */ + + Fts3Table *p = (Fts3Table *)pCsr->base.pVtab; + sqlite3_tokenizer *pT = p->pTokenizer; + sqlite3_tokenizer_module const *pModule = pT->pModule; + + assert( pCsr->isRequireSeek==0 ); + iDocid = sqlite3_column_int64(pCsr->pStmt, 0); + + for(i=0; inColumn && rc==SQLITE_OK; i++){ + const char *zText = (const char *)sqlite3_column_text(pCsr->pStmt, i+1); + sqlite3_tokenizer_cursor *pTC = 0; + + rc = pModule->xOpen(pT, zText, -1, &pTC); + while( rc==SQLITE_OK ){ + char const *zToken; /* Buffer containing token */ + int nToken; /* Number of bytes in token */ + int iDum1, iDum2; /* Dummy variables */ + int iPos; /* Position of token in zText */ + + pTC->pTokenizer = pT; + rc = pModule->xNext(pTC, &zToken, &nToken, &iDum1, &iDum2, &iPos); + for(pDef=pCsr->pDeferred; pDef && rc==SQLITE_OK; pDef=pDef->pNext){ + Fts3PhraseToken *pPT = pDef->pToken; + if( (pDef->iCol>=p->nColumn || pDef->iCol==i) + && (pPT->bFirst==0 || iPos==0) + && (pPT->n==nToken || (pPT->isPrefix && pPT->nz, pPT->n)) + ){ + fts3PendingListAppend(&pDef->pList, iDocid, i, iPos, &rc); + } + } + } + if( pTC ) pModule->xClose(pTC); + if( rc==SQLITE_DONE ) rc = SQLITE_OK; + } + + for(pDef=pCsr->pDeferred; pDef && rc==SQLITE_OK; pDef=pDef->pNext){ + if( pDef->pList ){ + rc = fts3PendingListAppendVarint(&pDef->pList, 0); + } + } + } + + return rc; +} + +int sqlite3Fts3DeferredTokenList( + Fts3DeferredToken *p, + char **ppData, + int *pnData +){ + char *pRet; + int nSkip; + sqlite3_int64 dummy; + + *ppData = 0; + *pnData = 0; + + if( p->pList==0 ){ + return SQLITE_OK; + } + + pRet = (char *)sqlite3_malloc(p->pList->nData); + if( !pRet ) return SQLITE_NOMEM; + + nSkip = sqlite3Fts3GetVarint(p->pList->aData, &dummy); + *pnData = p->pList->nData - nSkip; + *ppData = pRet; + + memcpy(pRet, &p->pList->aData[nSkip], *pnData); + return SQLITE_OK; +} + +/* +** Add an entry for token pToken to the pCsr->pDeferred list. +*/ +int sqlite3Fts3DeferToken( + Fts3Cursor *pCsr, /* Fts3 table cursor */ + Fts3PhraseToken *pToken, /* Token to defer */ + int iCol /* Column that token must appear in (or -1) */ +){ + Fts3DeferredToken *pDeferred; + pDeferred = sqlite3_malloc(sizeof(*pDeferred)); + if( !pDeferred ){ + return SQLITE_NOMEM; + } + memset(pDeferred, 0, sizeof(*pDeferred)); + pDeferred->pToken = pToken; + pDeferred->pNext = pCsr->pDeferred; + pDeferred->iCol = iCol; + pCsr->pDeferred = pDeferred; + + assert( pToken->pDeferred==0 ); + pToken->pDeferred = pDeferred; + + return SQLITE_OK; +} + +/* +** SQLite value pRowid contains the rowid of a row that may or may not be +** present in the FTS3 table. If it is, delete it and adjust the contents +** of subsiduary data structures accordingly. +*/ +static int fts3DeleteByRowid( + Fts3Table *p, + sqlite3_value *pRowid, + int *pnDoc, + u32 *aSzDel +){ + int isEmpty = 0; + int rc = fts3IsEmpty(p, pRowid, &isEmpty); + if( rc==SQLITE_OK ){ + if( isEmpty ){ + /* Deleting this row means the whole table is empty. In this case + ** delete the contents of all three tables and throw away any + ** data in the pendingTerms hash table. */ + rc = fts3DeleteAll(p, 1); + *pnDoc = *pnDoc - 1; + }else{ + sqlite3_int64 iRemove = sqlite3_value_int64(pRowid); + rc = fts3PendingTermsDocid(p, iRemove); + fts3DeleteTerms(&rc, p, pRowid, aSzDel); + if( p->zContentTbl==0 ){ + fts3SqlExec(&rc, p, SQL_DELETE_CONTENT, &pRowid); + if( sqlite3_changes(p->db) ) *pnDoc = *pnDoc - 1; + }else{ + *pnDoc = *pnDoc - 1; + } + if( p->bHasDocsize ){ + fts3SqlExec(&rc, p, SQL_DELETE_DOCSIZE, &pRowid); + } + } + } + + return rc; +} + +/* +** This function does the work for the xUpdate method of FTS3 virtual +** tables. +*/ +int sqlite3Fts3UpdateMethod( + sqlite3_vtab *pVtab, /* FTS3 vtab object */ + int nArg, /* Size of argument array */ + sqlite3_value **apVal, /* Array of arguments */ + sqlite_int64 *pRowid /* OUT: The affected (or effected) rowid */ +){ + Fts3Table *p = (Fts3Table *)pVtab; + int rc = SQLITE_OK; /* Return Code */ + int isRemove = 0; /* True for an UPDATE or DELETE */ + u32 *aSzIns = 0; /* Sizes of inserted documents */ + u32 *aSzDel; /* Sizes of deleted documents */ + int nChng = 0; /* Net change in number of documents */ + int bInsertDone = 0; + + assert( p->pSegments==0 ); + + /* Check for a "special" INSERT operation. One of the form: + ** + ** INSERT INTO xyz(xyz) VALUES('command'); + */ + if( nArg>1 + && sqlite3_value_type(apVal[0])==SQLITE_NULL + && sqlite3_value_type(apVal[p->nColumn+2])!=SQLITE_NULL + ){ + rc = fts3SpecialInsert(p, apVal[p->nColumn+2]); + goto update_out; + } + + /* Allocate space to hold the change in document sizes */ + aSzIns = sqlite3_malloc( sizeof(aSzIns[0])*(p->nColumn+1)*2 ); + if( aSzIns==0 ){ + rc = SQLITE_NOMEM; + goto update_out; + } + aSzDel = &aSzIns[p->nColumn+1]; + memset(aSzIns, 0, sizeof(aSzIns[0])*(p->nColumn+1)*2); + + /* If this is an INSERT operation, or an UPDATE that modifies the rowid + ** value, then this operation requires constraint handling. + ** + ** If the on-conflict mode is REPLACE, this means that the existing row + ** should be deleted from the database before inserting the new row. Or, + ** if the on-conflict mode is other than REPLACE, then this method must + ** detect the conflict and return SQLITE_CONSTRAINT before beginning to + ** modify the database file. + */ + if( nArg>1 && p->zContentTbl==0 ){ + /* Find the value object that holds the new rowid value. */ + sqlite3_value *pNewRowid = apVal[3+p->nColumn]; + if( sqlite3_value_type(pNewRowid)==SQLITE_NULL ){ + pNewRowid = apVal[1]; + } + + if( sqlite3_value_type(pNewRowid)!=SQLITE_NULL && ( + sqlite3_value_type(apVal[0])==SQLITE_NULL + || sqlite3_value_int64(apVal[0])!=sqlite3_value_int64(pNewRowid) + )){ + /* The new rowid is not NULL (in this case the rowid will be + ** automatically assigned and there is no chance of a conflict), and + ** the statement is either an INSERT or an UPDATE that modifies the + ** rowid column. So if the conflict mode is REPLACE, then delete any + ** existing row with rowid=pNewRowid. + ** + ** Or, if the conflict mode is not REPLACE, insert the new record into + ** the %_content table. If we hit the duplicate rowid constraint (or any + ** other error) while doing so, return immediately. + ** + ** This branch may also run if pNewRowid contains a value that cannot + ** be losslessly converted to an integer. In this case, the eventual + ** call to fts3InsertData() (either just below or further on in this + ** function) will return SQLITE_MISMATCH. If fts3DeleteByRowid is + ** invoked, it will delete zero rows (since no row will have + ** docid=$pNewRowid if $pNewRowid is not an integer value). + */ + if( sqlite3_vtab_on_conflict(p->db)==SQLITE_REPLACE ){ + rc = fts3DeleteByRowid(p, pNewRowid, &nChng, aSzDel); + }else{ + rc = fts3InsertData(p, apVal, pRowid); + bInsertDone = 1; + } + } + } + if( rc!=SQLITE_OK ){ + goto update_out; + } + + /* If this is a DELETE or UPDATE operation, remove the old record. */ + if( sqlite3_value_type(apVal[0])!=SQLITE_NULL ){ + assert( sqlite3_value_type(apVal[0])==SQLITE_INTEGER ); + rc = fts3DeleteByRowid(p, apVal[0], &nChng, aSzDel); + isRemove = 1; + } + + /* If this is an INSERT or UPDATE operation, insert the new record. */ + if( nArg>1 && rc==SQLITE_OK ){ + if( bInsertDone==0 ){ + rc = fts3InsertData(p, apVal, pRowid); + if( rc==SQLITE_CONSTRAINT && p->zContentTbl==0 ){ + rc = FTS_CORRUPT_VTAB; + } + } + if( rc==SQLITE_OK && (!isRemove || *pRowid!=p->iPrevDocid ) ){ + rc = fts3PendingTermsDocid(p, *pRowid); + } + if( rc==SQLITE_OK ){ + assert( p->iPrevDocid==*pRowid ); + rc = fts3InsertTerms(p, apVal, aSzIns); + } + if( p->bHasDocsize ){ + fts3InsertDocsize(&rc, p, aSzIns); + } + nChng++; + } + + if( p->bHasStat ){ + fts3UpdateDocTotals(&rc, p, aSzIns, aSzDel, nChng); + } + + update_out: + sqlite3_free(aSzIns); + sqlite3Fts3SegmentsClose(p); + return rc; +} + +/* +** Flush any data in the pending-terms hash table to disk. If successful, +** merge all segments in the database (including the new segment, if +** there was any data to flush) into a single segment. +*/ +int sqlite3Fts3Optimize(Fts3Table *p){ + int rc; + rc = sqlite3_exec(p->db, "SAVEPOINT fts3", 0, 0, 0); + if( rc==SQLITE_OK ){ + rc = fts3DoOptimize(p, 1); + if( rc==SQLITE_OK || rc==SQLITE_DONE ){ + int rc2 = sqlite3_exec(p->db, "RELEASE fts3", 0, 0, 0); + if( rc2!=SQLITE_OK ) rc = rc2; + }else{ + sqlite3_exec(p->db, "ROLLBACK TO fts3", 0, 0, 0); + sqlite3_exec(p->db, "RELEASE fts3", 0, 0, 0); + } + } + sqlite3Fts3SegmentsClose(p); + return rc; +} + +#endif diff --git a/ext/fts3/fts3speed.tcl b/ext/fts3/fts3speed.tcl new file mode 100644 index 0000000..377cb19 --- /dev/null +++ b/ext/fts3/fts3speed.tcl @@ -0,0 +1,122 @@ + + +#-------------------------------------------------------------------------- +# This script contains several sub-programs used to test FTS3/FTS4 +# performance. It does not run the queries directly, but generates SQL +# scripts that can be run using the shell tool. +# +# The following cases are tested: +# +# 1. Inserting documents into an FTS3 table. +# 2. Optimizing an FTS3 table (i.e. "INSERT INTO t1 VALUES('optimize')"). +# 3. Deleting documents from an FTS3 table. +# 4. Querying FTS3 tables. +# + +# Number of tokens in vocabulary. And number of tokens in each document. +# +set VOCAB_SIZE 2000 +set DOC_SIZE 100 + +set NUM_INSERTS 100000 +set NUM_SELECTS 1000 + +# Force everything in this script to be deterministic. +# +expr {srand(0)} + +proc usage {} { + puts stderr "Usage: $::argv0 " + exit -1 +} + +proc sql {sql} { + puts $::fd $sql +} + + +# Return a list of $nWord randomly generated tokens each between 2 and 10 +# characters in length. +# +proc build_vocab {nWord} { + set ret [list] + set chars [list a b c d e f g h i j k l m n o p q r s t u v w x y z] + for {set i 0} {$i<$nWord} {incr i} { + set len [expr {int((rand()*9.0)+2)}] + set term "" + for {set j 0} {$j<$len} {incr j} { + append term [lindex $chars [expr {int(rand()*[llength $chars])}]] + } + lappend ret $term + } + set ret +} + +proc select_term {} { + set n [llength $::vocab] + set t [expr int(rand()*$n*3)] + if {$t>=2*$n} { set t [expr {($t-2*$n)/100}] } + if {$t>=$n} { set t [expr {($t-$n)/10}] } + lindex $::vocab $t +} + +proc select_doc {nTerm} { + set ret [list] + for {set i 0} {$i<$nTerm} {incr i} { + lappend ret [select_term] + } + set ret +} + +proc test_1 {nInsert} { + sql "PRAGMA synchronous = OFF;" + sql "DROP TABLE IF EXISTS t1;" + sql "CREATE VIRTUAL TABLE t1 USING fts4;" + for {set i 0} {$i < $nInsert} {incr i} { + set doc [select_doc $::DOC_SIZE] + sql "INSERT INTO t1 VALUES('$doc');" + } +} + +proc test_2 {} { + sql "INSERT INTO t1(t1) VALUES('optimize');" +} + +proc test_3 {nSelect} { + for {set i 0} {$i < $nSelect} {incr i} { + sql "SELECT count(*) FROM t1 WHERE t1 MATCH '[select_term]';" + } +} + +proc test_4 {nSelect} { + for {set i 0} {$i < $nSelect} {incr i} { + sql "SELECT count(*) FROM t1 WHERE t1 MATCH '[select_term] [select_term]';" + } +} + +if {[llength $argv]!=0} usage + +set ::vocab [build_vocab $::VOCAB_SIZE] + +set ::fd [open fts3speed_insert.sql w] +test_1 $NUM_INSERTS +close $::fd + +set ::fd [open fts3speed_select.sql w] +test_3 $NUM_SELECTS +close $::fd + +set ::fd [open fts3speed_select2.sql w] +test_4 $NUM_SELECTS +close $::fd + +set ::fd [open fts3speed_optimize.sql w] +test_2 +close $::fd + +puts "Success. Created files:" +puts " fts3speed_insert.sql" +puts " fts3speed_select.sql" +puts " fts3speed_select2.sql" +puts " fts3speed_optimize.sql" + diff --git a/ext/fts3/mkfts3amal.tcl b/ext/fts3/mkfts3amal.tcl new file mode 100644 index 0000000..0590487 --- /dev/null +++ b/ext/fts3/mkfts3amal.tcl @@ -0,0 +1,115 @@ +#!/usr/bin/tclsh +# +# This script builds a single C code file holding all of FTS3 code. +# The name of the output file is fts3amal.c. To build this file, +# first do: +# +# make target_source +# +# The make target above moves all of the source code files into +# a subdirectory named "tsrc". (This script expects to find the files +# there and will not work if they are not found.) +# +# After the "tsrc" directory has been created and populated, run +# this script: +# +# tclsh mkfts3amal.tcl +# +# The amalgamated FTS3 code will be written into fts3amal.c +# + +# Open the output file and write a header comment at the beginning +# of the file. +# +set out [open fts3amal.c w] +set today [clock format [clock seconds] -format "%Y-%m-%d %H:%M:%S UTC" -gmt 1] +puts $out [subst \ +{/****************************************************************************** +** This file is an amalgamation of separate C source files from the SQLite +** Full Text Search extension 2 (fts3). By combining all the individual C +** code files into this single large file, the entire code can be compiled +** as a one translation unit. This allows many compilers to do optimizations +** that would not be possible if the files were compiled separately. It also +** makes the code easier to import into other projects. +** +** This amalgamation was generated on $today. +*/}] + +# These are the header files used by FTS3. The first time any of these +# files are seen in a #include statement in the C code, include the complete +# text of the file in-line. The file only needs to be included once. +# +foreach hdr { + fts3.h + fts3_hash.h + fts3_tokenizer.h + sqlite3.h + sqlite3ext.h +} { + set available_hdr($hdr) 1 +} + +# 78 stars used for comment formatting. +set s78 \ +{*****************************************************************************} + +# Insert a comment into the code +# +proc section_comment {text} { + global out s78 + set n [string length $text] + set nstar [expr {60 - $n}] + set stars [string range $s78 0 $nstar] + puts $out "/************** $text $stars/" +} + +# Read the source file named $filename and write it into the +# sqlite3.c output file. If any #include statements are seen, +# process them approprately. +# +proc copy_file {filename} { + global seen_hdr available_hdr out + set tail [file tail $filename] + section_comment "Begin file $tail" + set in [open $filename r] + while {![eof $in]} { + set line [gets $in] + if {[regexp {^#\s*include\s+["<]([^">]+)[">]} $line all hdr]} { + if {[info exists available_hdr($hdr)]} { + if {$available_hdr($hdr)} { + section_comment "Include $hdr in the middle of $tail" + copy_file tsrc/$hdr + section_comment "Continuing where we left off in $tail" + } + } elseif {![info exists seen_hdr($hdr)]} { + set seen_hdr($hdr) 1 + puts $out $line + } + } elseif {[regexp {^#ifdef __cplusplus} $line]} { + puts $out "#if 0" + } elseif {[regexp {^#line} $line]} { + # Skip #line directives. + } else { + puts $out $line + } + } + close $in + section_comment "End of $tail" +} + + +# Process the source files. Process files containing commonly +# used subroutines first in order to help the compiler find +# inlining opportunities. +# +foreach file { + fts3.c + fts3_hash.c + fts3_porter.c + fts3_tokenizer.c + fts3_tokenizer1.c +} { + copy_file tsrc/$file +} + +close $out -- cgit v1.2.3