From 8bd863936ead4243f58fb99e11d1221e1af0a71e Mon Sep 17 00:00:00 2001 From: Micah Anderson Date: Wed, 15 Jan 2014 18:04:21 +0000 Subject: remove unnecessary embedded things: scons, rebar --- .../scons-local-2.0.1/SCons/Scanner/Fortran.py | 316 --------------------- 1 file changed, 316 deletions(-) delete mode 100644 couchjs/scons/scons-local-2.0.1/SCons/Scanner/Fortran.py (limited to 'couchjs/scons/scons-local-2.0.1/SCons/Scanner/Fortran.py') diff --git a/couchjs/scons/scons-local-2.0.1/SCons/Scanner/Fortran.py b/couchjs/scons/scons-local-2.0.1/SCons/Scanner/Fortran.py deleted file mode 100644 index 63d34e49..00000000 --- a/couchjs/scons/scons-local-2.0.1/SCons/Scanner/Fortran.py +++ /dev/null @@ -1,316 +0,0 @@ -"""SCons.Scanner.Fortran - -This module implements the dependency scanner for Fortran code. - -""" - -# -# Copyright (c) 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010 The SCons Foundation -# -# Permission is hereby granted, free of charge, to any person obtaining -# a copy of this software and associated documentation files (the -# "Software"), to deal in the Software without restriction, including -# without limitation the rights to use, copy, modify, merge, publish, -# distribute, sublicense, and/or sell copies of the Software, and to -# permit persons to whom the Software is furnished to do so, subject to -# the following conditions: -# -# The above copyright notice and this permission notice shall be included -# in all copies or substantial portions of the Software. -# -# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY -# KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE -# WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND -# NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE -# LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION -# OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION -# WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - -__revision__ = "src/engine/SCons/Scanner/Fortran.py 5134 2010/08/16 23:02:40 bdeegan" - -import re - -import SCons.Node -import SCons.Node.FS -import SCons.Scanner -import SCons.Util -import SCons.Warnings - -class F90Scanner(SCons.Scanner.Classic): - """ - A Classic Scanner subclass for Fortran source files which takes - into account both USE and INCLUDE statements. This scanner will - work for both F77 and F90 (and beyond) compilers. - - Currently, this scanner assumes that the include files do not contain - USE statements. To enable the ability to deal with USE statements - in include files, add logic right after the module names are found - to loop over each include file, search for and locate each USE - statement, and append each module name to the list of dependencies. - Caching the search results in a common dictionary somewhere so that - the same include file is not searched multiple times would be a - smart thing to do. - """ - - def __init__(self, name, suffixes, path_variable, - use_regex, incl_regex, def_regex, *args, **kw): - - self.cre_use = re.compile(use_regex, re.M) - self.cre_incl = re.compile(incl_regex, re.M) - self.cre_def = re.compile(def_regex, re.M) - - def _scan(node, env, path, self=self): - node = node.rfile() - - if not node.exists(): - return [] - - return self.scan(node, env, path) - - kw['function'] = _scan - kw['path_function'] = SCons.Scanner.FindPathDirs(path_variable) - kw['recursive'] = 1 - kw['skeys'] = suffixes - kw['name'] = name - - SCons.Scanner.Current.__init__(self, *args, **kw) - - def scan(self, node, env, path=()): - - # cache the includes list in node so we only scan it once: - if node.includes != None: - mods_and_includes = node.includes - else: - # retrieve all included filenames - includes = self.cre_incl.findall(node.get_text_contents()) - # retrieve all USE'd module names - modules = self.cre_use.findall(node.get_text_contents()) - # retrieve all defined module names - defmodules = self.cre_def.findall(node.get_text_contents()) - - # Remove all USE'd module names that are defined in the same file - # (case-insensitively) - d = {} - for m in defmodules: - d[m.lower()] = 1 - modules = [m for m in modules if m.lower() not in d] - - # Convert module name to a .mod filename - suffix = env.subst('$FORTRANMODSUFFIX') - modules = [x.lower() + suffix for x in modules] - # Remove unique items from the list - mods_and_includes = SCons.Util.unique(includes+modules) - node.includes = mods_and_includes - - # This is a hand-coded DSU (decorate-sort-undecorate, or - # Schwartzian transform) pattern. The sort key is the raw name - # of the file as specifed on the USE or INCLUDE line, which lets - # us keep the sort order constant regardless of whether the file - # is actually found in a Repository or locally. - nodes = [] - source_dir = node.get_dir() - if callable(path): - path = path() - for dep in mods_and_includes: - n, i = self.find_include(dep, source_dir, path) - - if n is None: - SCons.Warnings.warn(SCons.Warnings.DependencyWarning, - "No dependency generated for file: %s (referenced by: %s) -- file not found" % (i, node)) - else: - sortkey = self.sort_key(dep) - nodes.append((sortkey, n)) - - return [pair[1] for pair in sorted(nodes)] - -def FortranScan(path_variable="FORTRANPATH"): - """Return a prototype Scanner instance for scanning source files - for Fortran USE & INCLUDE statements""" - -# The USE statement regex matches the following: -# -# USE module_name -# USE :: module_name -# USE, INTRINSIC :: module_name -# USE, NON_INTRINSIC :: module_name -# -# Limitations -# -# -- While the regex can handle multiple USE statements on one line, -# it cannot properly handle them if they are commented out. -# In either of the following cases: -# -# ! USE mod_a ; USE mod_b [entire line is commented out] -# USE mod_a ! ; USE mod_b [in-line comment of second USE statement] -# -# the second module name (mod_b) will be picked up as a dependency -# even though it should be ignored. The only way I can see -# to rectify this would be to modify the scanner to eliminate -# the call to re.findall, read in the contents of the file, -# treating the comment character as an end-of-line character -# in addition to the normal linefeed, loop over each line, -# weeding out the comments, and looking for the USE statements. -# One advantage to this is that the regex passed to the scanner -# would no longer need to match a semicolon. -# -# -- I question whether or not we need to detect dependencies to -# INTRINSIC modules because these are built-in to the compiler. -# If we consider them a dependency, will SCons look for them, not -# find them, and kill the build? Or will we there be standard -# compiler-specific directories we will need to point to so the -# compiler and SCons can locate the proper object and mod files? - -# Here is a breakdown of the regex: -# -# (?i) : regex is case insensitive -# ^ : start of line -# (?: : group a collection of regex symbols without saving the match as a "group" -# ^|; : matches either the start of the line or a semicolon - semicolon -# ) : end the unsaved grouping -# \s* : any amount of white space -# USE : match the string USE, case insensitive -# (?: : group a collection of regex symbols without saving the match as a "group" -# \s+| : match one or more whitespace OR .... (the next entire grouped set of regex symbols) -# (?: : group a collection of regex symbols without saving the match as a "group" -# (?: : establish another unsaved grouping of regex symbols -# \s* : any amount of white space -# , : match a comma -# \s* : any amount of white space -# (?:NON_)? : optionally match the prefix NON_, case insensitive -# INTRINSIC : match the string INTRINSIC, case insensitive -# )? : optionally match the ", INTRINSIC/NON_INTRINSIC" grouped expression -# \s* : any amount of white space -# :: : match a double colon that must appear after the INTRINSIC/NON_INTRINSIC attribute -# ) : end the unsaved grouping -# ) : end the unsaved grouping -# \s* : match any amount of white space -# (\w+) : match the module name that is being USE'd -# -# - use_regex = "(?i)(?:^|;)\s*USE(?:\s+|(?:(?:\s*,\s*(?:NON_)?INTRINSIC)?\s*::))\s*(\w+)" - - -# The INCLUDE statement regex matches the following: -# -# INCLUDE 'some_Text' -# INCLUDE "some_Text" -# INCLUDE "some_Text" ; INCLUDE "some_Text" -# INCLUDE kind_"some_Text" -# INCLUDE kind_'some_Text" -# -# where some_Text can include any alphanumeric and/or special character -# as defined by the Fortran 2003 standard. -# -# Limitations: -# -# -- The Fortran standard dictates that a " or ' in the INCLUDE'd -# string must be represented as a "" or '', if the quotes that wrap -# the entire string are either a ' or ", respectively. While the -# regular expression below can detect the ' or " characters just fine, -# the scanning logic, presently is unable to detect them and reduce -# them to a single instance. This probably isn't an issue since, -# in practice, ' or " are not generally used in filenames. -# -# -- This regex will not properly deal with multiple INCLUDE statements -# when the entire line has been commented out, ala -# -# ! INCLUDE 'some_file' ; INCLUDE 'some_file' -# -# In such cases, it will properly ignore the first INCLUDE file, -# but will actually still pick up the second. Interestingly enough, -# the regex will properly deal with these cases: -# -# INCLUDE 'some_file' -# INCLUDE 'some_file' !; INCLUDE 'some_file' -# -# To get around the above limitation, the FORTRAN programmer could -# simply comment each INCLUDE statement separately, like this -# -# ! INCLUDE 'some_file' !; INCLUDE 'some_file' -# -# The way I see it, the only way to get around this limitation would -# be to modify the scanning logic to replace the calls to re.findall -# with a custom loop that processes each line separately, throwing -# away fully commented out lines before attempting to match against -# the INCLUDE syntax. -# -# Here is a breakdown of the regex: -# -# (?i) : regex is case insensitive -# (?: : begin a non-saving group that matches the following: -# ^ : either the start of the line -# | : or -# ['">]\s*; : a semicolon that follows a single quote, -# double quote or greater than symbol (with any -# amount of whitespace in between). This will -# allow the regex to match multiple INCLUDE -# statements per line (although it also requires -# the positive lookahead assertion that is -# used below). It will even properly deal with -# (i.e. ignore) cases in which the additional -# INCLUDES are part of an in-line comment, ala -# " INCLUDE 'someFile' ! ; INCLUDE 'someFile2' " -# ) : end of non-saving group -# \s* : any amount of white space -# INCLUDE : match the string INCLUDE, case insensitive -# \s+ : match one or more white space characters -# (?\w+_)? : match the optional "kind-param _" prefix allowed by the standard -# [<"'] : match the include delimiter - an apostrophe, double quote, or less than symbol -# (.+?) : match one or more characters that make up -# the included path and file name and save it -# in a group. The Fortran standard allows for -# any non-control character to be used. The dot -# operator will pick up any character, including -# control codes, but I can't conceive of anyone -# putting control codes in their file names. -# The question mark indicates it is non-greedy so -# that regex will match only up to the next quote, -# double quote, or greater than symbol -# (?=["'>]) : positive lookahead assertion to match the include -# delimiter - an apostrophe, double quote, or -# greater than symbol. This level of complexity -# is required so that the include delimiter is -# not consumed by the match, thus allowing the -# sub-regex discussed above to uniquely match a -# set of semicolon-separated INCLUDE statements -# (as allowed by the F2003 standard) - - include_regex = """(?i)(?:^|['">]\s*;)\s*INCLUDE\s+(?:\w+_)?[<"'](.+?)(?=["'>])""" - -# The MODULE statement regex finds module definitions by matching -# the following: -# -# MODULE module_name -# -# but *not* the following: -# -# MODULE PROCEDURE procedure_name -# -# Here is a breakdown of the regex: -# -# (?i) : regex is case insensitive -# ^\s* : any amount of white space -# MODULE : match the string MODULE, case insensitive -# \s+ : match one or more white space characters -# (?!PROCEDURE) : but *don't* match if the next word matches -# PROCEDURE (negative lookahead assertion), -# case insensitive -# (\w+) : match one or more alphanumeric characters -# that make up the defined module name and -# save it in a group - - def_regex = """(?i)^\s*MODULE\s+(?!PROCEDURE)(\w+)""" - - scanner = F90Scanner("FortranScan", - "$FORTRANSUFFIXES", - path_variable, - use_regex, - include_regex, - def_regex) - return scanner - -# Local Variables: -# tab-width:4 -# indent-tabs-mode:nil -# End: -# vim: set expandtab tabstop=4 shiftwidth=4: -- cgit v1.2.3