[pkg] unify client and server into a single python package

We have been discussing about this merge for a while.
Its main goal is to simplify things: code navigation, but also
packaging.

The rationale is that the code is more cohesive in this way, and there's
only one source package to install.

Dependencies that are only for the server or the client will not be
installed by default, and they are expected to be provided by the
environment. There are setuptools extras defined for the client and the
server.

Debianization is still expected to split the single source package into
3 binaries.

Another avantage is that the documentation can now install a single
package with a single step, and therefore include the docstrings into
the generated docs.

- Resolves: #8896

1 files changed, 812 insertions, 0 deletions

diff --git a/src/leap/soledad/common/couch/__init__.py b/src/leap/soledad/common/couch/__init__.py
new file mode 100644
index 00000000..2343e849
--- /dev/null
+++ b/src/leap/soledad/common/couch/__init__.py
@@ -0,0 +1,812 @@
+# -*- coding: utf-8 -*-
+# __init__.py
+# Copyright (C) 2015 LEAP
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program. If not, see <http://www.gnu.org/licenses/>.
+
+
+"""A U1DB backend that uses CouchDB as its persistence layer."""
+
+
+import json
+import copy
+import re
+import uuid
+import binascii
+
+from six import StringIO
+from six.moves.urllib.parse import urljoin
+from contextlib import contextmanager
+
+
+from couchdb.client import Server, Database
+from couchdb.http import (
+    ResourceConflict,
+    ResourceNotFound,
+    Session,
+    urljoin as couch_urljoin,
+    Resource,
+)
+from leap.soledad.common.l2db.errors import (
+    DatabaseDoesNotExist,
+    InvalidGeneration,
+    RevisionConflict,
+)
+from leap.soledad.common.l2db.remote import http_app
+
+
+from .support import MultipartWriter
+from leap.soledad.common.errors import InvalidURLError
+from leap.soledad.common.document import ServerDocument
+from leap.soledad.common.backend import SoledadBackend
+
+
+COUCH_TIMEOUT = 120  # timeout for transfers between Soledad server and Couch
+
+
+def list_users_dbs(couch_url):
+    """
+    Retrieves a list with all databases that starts with 'user-' on CouchDB.
+    Those databases belongs to users. So, the list will contain all the
+    database names in the form of 'user-{uuid4}'.
+
+    :param couch_url: The couch url with needed credentials
+    :type couch_url: str
+
+    :return: The list of all database names from users.
+    :rtype: [str]
+    """
+    with couch_server(couch_url) as server:
+        users = [dbname for dbname in server if dbname.startswith('user-')]
+    return users
+
+
+# monkey-patch the u1db http app to use ServerDocument
+http_app.Document = ServerDocument
+
+
+@contextmanager
+def couch_server(url):
+    """
+    Provide a connection to a couch server and cleanup after use.
+
+    For database creation and deletion we use an ephemeral connection to the
+    couch server. That connection has to be properly closed, so we provide it
+    as a context manager.
+
+    :param url: The URL of the Couch server.
+    :type url: str
+    """
+    session = Session(timeout=COUCH_TIMEOUT)
+    server = Server(url=url, full_commit=False, session=session)
+    yield server
+
+
+def _get_gen_doc_id(gen):
+    return 'gen-%s' % str(gen).zfill(10)
+
+
+GENERATION_KEY = 'gen'
+TRANSACTION_ID_KEY = 'trans_id'
+REPLICA_UID_KEY = 'replica_uid'
+DOC_ID_KEY = 'doc_id'
+SCHEMA_VERSION_KEY = 'schema_version'
+
+CONFIG_DOC_ID = '_local/config'
+SYNC_DOC_ID_PREFIX = '_local/sync_'
+SCHEMA_VERSION = 1
+
+
+class CouchDatabase(object):
+    """
+    Holds CouchDB related code.
+    This class gives methods to encapsulate database operations and hide
+    CouchDB details from backend code.
+    """
+
+    @classmethod
+    def open_database(cls, url, create, replica_uid=None,
+                      database_security=None):
+        """
+        Open a U1DB database using CouchDB as backend.
+
+        :param url: the url of the database replica
+        :type url: str
+        :param create: should the replica be created if it does not exist?
+        :type create: bool
+        :param replica_uid: an optional unique replica identifier
+        :type replica_uid: str
+        :param database_security: security rules as CouchDB security doc
+        :type database_security: dict
+
+        :return: the database instance
+        :rtype: SoledadBackend
+
+        :raise DatabaseDoesNotExist: Raised if database does not exist.
+        """
+        # get database from url
+        m = re.match('(^https?://[^/]+)/(.+)$', url)
+        if not m:
+            raise InvalidURLError
+        url = m.group(1)
+        dbname = m.group(2)
+        with couch_server(url) as server:
+            if dbname not in server:
+                if create:
+                    server.create(dbname)
+                else:
+                    raise DatabaseDoesNotExist()
+        db = cls(url, dbname, ensure_security=create,
+                 database_security=database_security)
+        return SoledadBackend(
+            db, replica_uid=replica_uid)
+
+    def __init__(self, url, dbname, ensure_security=False,
+                 database_security=None):
+        """
+        :param url: Couch server URL with necessary credentials
+        :type url: string
+        :param dbname: Couch database name
+        :type dbname: string
+        :param ensure_security: will PUT a _security ddoc if set
+        :type ensure_security: bool
+        :param database_security: security rules as CouchDB security doc
+        :type database_security: dict
+        """
+        self._session = Session(timeout=COUCH_TIMEOUT)
+        self._url = url
+        self._dbname = dbname
+        self._database = self.get_couch_database(url, dbname)
+        self.batching = False
+        self.batch_generation = None
+        self.batch_docs = {}
+        if ensure_security:
+            self.ensure_security_ddoc(database_security)
+
+    def batch_start(self):
+        self.batching = True
+        self.batch_generation = self.get_generation_info()
+        ids = set(row.id for row in self._database.view('_all_docs'))
+        self.batched_ids = ids
+
+    def batch_end(self):
+        self.batching = False
+        self.batch_generation = None
+        self.__perform_batch()
+
+    def get_couch_database(self, url, dbname):
+        """
+        Generate a couchdb.Database instance given a url and dbname.
+
+        :param url: CouchDB's server url with credentials
+        :type url: str
+        :param dbname: Database name
+        :type dbname: str
+
+        :return: couch library database instance
+        :rtype: couchdb.Database
+
+        :raise DatabaseDoesNotExist: Raised if database does not exist.
+        """
+        try:
+            return Database(
+                urljoin(url, dbname),
+                self._session)
+        except ResourceNotFound:
+            raise DatabaseDoesNotExist()
+
+    def ensure_security_ddoc(self, security_config=None):
+        """
+        Make sure that only soledad user is able to access this database as
+        an unprivileged member, meaning that administration access will
+        be forbidden even inside an user database.
+        The goal is to make sure that only the lowest access level is given
+        to the unprivileged CouchDB user set on the server process.
+        This is achieved by creating a _security design document, see:
+        http://docs.couchdb.org/en/latest/api/database/security.html
+
+        :param security_config: security configuration parsed from conf file
+        :type security_config: dict
+        """
+        security_config = security_config or {}
+        security = self._database.resource.get_json('_security')[2]
+        security['members'] = {'names': [], 'roles': []}
+        security['members']['names'] = security_config.get('members',
+                                                           ['soledad'])
+        security['members']['roles'] = security_config.get('members_roles', [])
+        security['admins'] = {'names': [], 'roles': []}
+        security['admins']['names'] = security_config.get('admins', [])
+        security['admins']['roles'] = security_config.get('admins_roles', [])
+        self._database.resource.put_json('_security', body=security)
+
+    def delete_database(self):
+        """
+        Delete a U1DB CouchDB database.
+        """
+        with couch_server(self._url) as server:
+            del(server[self._dbname])
+
+    def set_replica_uid(self, replica_uid):
+        """
+        Force the replica uid to be set.
+
+        :param replica_uid: The new replica uid.
+        :type replica_uid: str
+        """
+        try:
+            # set on existent config document
+            doc = self._database[CONFIG_DOC_ID]
+            doc[REPLICA_UID_KEY] = replica_uid
+        except ResourceNotFound:
+            # or create the config document
+            doc = {
+                '_id': CONFIG_DOC_ID,
+                REPLICA_UID_KEY: replica_uid,
+                SCHEMA_VERSION_KEY: SCHEMA_VERSION,
+            }
+        self._database.save(doc)
+
+    def get_replica_uid(self):
+        """
+        Get the replica uid.
+
+        :return: The replica uid.
+        :rtype: str
+        """
+        try:
+            # grab replica_uid from server
+            doc = self._database[CONFIG_DOC_ID]
+            replica_uid = doc[REPLICA_UID_KEY]
+            return replica_uid
+        except ResourceNotFound:
+            # create a unique replica_uid
+            replica_uid = uuid.uuid4().hex
+            self.set_replica_uid(replica_uid)
+            return replica_uid
+
+    def close(self):
+        self._database = None
+
+    def get_all_docs(self, include_deleted=False):
+        """
+        Get the JSON content for all documents in the database.
+
+        :param include_deleted: If set to True, deleted documents will be
+                                returned with empty content. Otherwise deleted
+                                documents will not be included in the results.
+        :type include_deleted: bool
+
+        :return: (generation, [ServerDocument])
+            The current generation of the database, followed by a list of all
+            the documents in the database.
+        :rtype: (int, [ServerDocument])
+        """
+
+        generation, _ = self.get_generation_info()
+        results = list(
+            self.get_docs(None, True, include_deleted))
+        return (generation, results)
+
+    def get_docs(self, doc_ids, check_for_conflicts=True,
+                 include_deleted=False, read_content=True):
+        """
+        Get the JSON content for many documents.
+
+        Use couch's `_all_docs` view to get the documents indicated in
+        `doc_ids`,
+
+        :param doc_ids: A list of document identifiers or None for all.
+        :type doc_ids: list
+        :param check_for_conflicts: If set to False, then the conflict check
+                                    will be skipped, and 'None' will be
+                                    returned instead of True/False.
+        :type check_for_conflicts: bool
+        :param include_deleted: If set to True, deleted documents will be
+                                returned with empty content. Otherwise deleted
+                                documents will not be included in the results.
+
+        :return: iterable giving the Document object for each document id
+                 in matching doc_ids order.
+        :rtype: iterable
+        """
+        params = {'include_docs': 'true', 'attachments': 'false'}
+        if doc_ids is not None:
+            params['keys'] = doc_ids
+        view = self._database.view("_all_docs", **params)
+        for row in view.rows:
+            result = copy.deepcopy(row['doc'])
+            for file_name in result.get('_attachments', {}).keys():
+                data = self._database.get_attachment(result, file_name)
+                if data:
+                    if read_content:
+                        data = data.read()
+                    result['_attachments'][file_name] = {'data': data}
+            doc = self.__parse_doc_from_couch(
+                result, result['_id'],
+                check_for_conflicts=check_for_conflicts, decode=False)
+            # filter out non-u1db or deleted documents
+            if not doc or (not include_deleted and doc.is_tombstone()):
+                continue
+            yield doc
+
+    def get_doc(self, doc_id, check_for_conflicts=False):
+        """
+        Extract the document from storage.
+
+        This can return None if the document doesn't exist.
+
+        :param doc_id: The unique document identifier
+        :type doc_id: str
+        :param check_for_conflicts: If set to False, then the conflict check
+                                    will be skipped.
+        :type check_for_conflicts: bool
+
+        :return: The document.
+        :rtype: ServerDocument
+        """
+        doc_from_batch = self.__check_batch_before_get(doc_id)
+        if doc_from_batch:
+            return doc_from_batch
+        if self.batching and doc_id not in self.batched_ids:
+            return None
+        if doc_id not in self._database:
+            return None
+        # get document with all attachments (u1db content and eventual
+        # conflicts)
+        result = self.json_from_resource([doc_id], attachments=True)
+        return self.__parse_doc_from_couch(result, doc_id, check_for_conflicts)
+
+    def __check_batch_before_get(self, doc_id):
+        """
+        If doc_id is staged for batching, then we need to commit the batch
+        before going ahead. This avoids consistency problems, like trying to
+        get a document that isn't persisted and processing like it is missing.
+
+        :param doc_id: The unique document identifier
+        :type doc_id: str
+        """
+        if doc_id in self.batch_docs:
+            couch_doc = self.batch_docs[doc_id]
+            rev = self.__perform_batch(doc_id)
+            couch_doc['_rev'] = rev
+            self.batched_ids.add(doc_id)
+            return self.__parse_doc_from_couch(couch_doc, doc_id, True)
+        return None
+
+    def __perform_batch(self, doc_id=None):
+        status = self._database.update(self.batch_docs.values())
+        rev = None
+        for ok, stored_doc_id, rev_or_error in status:
+            if not ok:
+                error = rev_or_error
+                if type(error) is ResourceConflict:
+                    raise RevisionConflict
+                raise error
+            elif doc_id == stored_doc_id:
+                rev = rev_or_error
+        self.batch_docs.clear()
+        return rev
+
+    def __parse_doc_from_couch(self, result, doc_id,
+                               check_for_conflicts=False, decode=True):
+        # restrict to u1db documents
+        if 'u1db_rev' not in result:
+            return None
+        doc = ServerDocument(doc_id, result['u1db_rev'])
+        # set contents or make tombstone
+        if '_attachments' not in result \
+                or 'u1db_content' not in result['_attachments']:
+            doc.make_tombstone()
+        elif decode:
+            doc.content = json.loads(
+                binascii.a2b_base64(
+                    result['_attachments']['u1db_content']['data']))
+        else:
+            doc._json = result['_attachments']['u1db_content']['data']
+        # determine if there are conflicts
+        if check_for_conflicts \
+                and '_attachments' in result \
+                and 'u1db_conflicts' in result['_attachments']:
+            if decode:
+                conflicts = binascii.a2b_base64(
+                    result['_attachments']['u1db_conflicts']['data'])
+            else:
+                conflicts = result['_attachments']['u1db_conflicts']['data']
+            conflicts = json.loads(conflicts)
+            doc.set_conflicts(self._build_conflicts(doc.doc_id, conflicts))
+        # store couch revision
+        doc.couch_rev = result['_rev']
+        return doc
+
+    def _build_conflicts(self, doc_id, attached_conflicts):
+        """
+        Build the conflicted documents list from the conflicts attachment
+        fetched from a couch document.
+
+        :param attached_conflicts: The document's conflicts as fetched from a
+                                   couch document attachment.
+        :type attached_conflicts: dict
+        """
+        conflicts = []
+        for doc_rev, content in attached_conflicts:
+            doc = ServerDocument(doc_id, doc_rev)
+            if content is None:
+                doc.make_tombstone()
+            else:
+                doc.content = content
+            conflicts.append(doc)
+        return conflicts
+
+    def get_trans_id_for_gen(self, generation):
+        """
+        Get the transaction id corresponding to a particular generation.
+
+        :param generation: The generation for which to get the transaction id.
+        :type generation: int
+
+        :return: The transaction id for C{generation}.
+        :rtype: str
+
+        :raise InvalidGeneration: Raised when the generation does not exist.
+        """
+        if generation == 0:
+            return ''
+        log = self._get_transaction_log(start=generation, end=generation)
+        if not log:
+            raise InvalidGeneration
+        _, _, trans_id = log[0]
+        return trans_id
+
+    def get_replica_gen_and_trans_id(self, other_replica_uid):
+        """
+        Return the last known generation and transaction id for the other db
+        replica.
+
+        When you do a synchronization with another replica, the Database keeps
+        track of what generation the other database replica was at, and what
+        the associated transaction id was.  This is used to determine what data
+        needs to be sent, and if two databases are claiming to be the same
+        replica.
+
+        :param other_replica_uid: The identifier for the other replica.
+        :type other_replica_uid: str
+
+        :return: A tuple containing the generation and transaction id we
+                 encountered during synchronization. If we've never
+                 synchronized with the replica, this is (0, '').
+        :rtype: (int, str)
+        """
+        doc_id = '%s%s' % (SYNC_DOC_ID_PREFIX, other_replica_uid)
+        try:
+            doc = self._database[doc_id]
+        except ResourceNotFound:
+            doc = {
+                '_id': doc_id,
+                GENERATION_KEY: 0,
+                REPLICA_UID_KEY: str(other_replica_uid),
+                TRANSACTION_ID_KEY: '',
+            }
+            self._database.save(doc)
+        gen, trans_id = doc[GENERATION_KEY], doc[TRANSACTION_ID_KEY]
+        return gen, trans_id
+
+    def get_doc_conflicts(self, doc_id, couch_rev=None):
+        """
+        Get the conflicted versions of a document.
+
+        If the C{couch_rev} parameter is not None, conflicts for a specific
+        document's couch revision are returned.
+
+        :param couch_rev: The couch document revision.
+        :type couch_rev: str
+
+        :return: A list of conflicted versions of the document.
+        :rtype: list
+        """
+        # request conflicts attachment from server
+        params = {}
+        conflicts = []
+        if couch_rev is not None:
+            params['rev'] = couch_rev  # restric document's couch revision
+        else:
+            # TODO: move into resource logic!
+            first_entry = self.get_doc(doc_id, check_for_conflicts=True)
+            conflicts.append(first_entry)
+
+        try:
+            response = self.json_from_resource([doc_id, 'u1db_conflicts'],
+                                               **params)
+            return conflicts + self._build_conflicts(
+                doc_id, json.loads(response.read()))
+        except ResourceNotFound:
+            return []
+
+    def set_replica_gen_and_trans_id(
+            self, other_replica_uid, other_generation, other_transaction_id):
+        """
+        Set the last-known generation and transaction id for the other
+        database replica.
+
+        We have just performed some synchronization, and we want to track what
+        generation the other replica was at. See also
+        get_replica_gen_and_trans_id.
+
+        :param other_replica_uid: The U1DB identifier for the other replica.
+        :type other_replica_uid: str
+        :param other_generation: The generation number for the other replica.
+        :type other_generation: int
+        :param other_transaction_id: The transaction id associated with the
+                                     generation.
+        :type other_transaction_id: str
+        """
+        doc_id = '%s%s' % (SYNC_DOC_ID_PREFIX, other_replica_uid)
+        try:
+            doc = self._database[doc_id]
+        except ResourceNotFound:
+            doc = {'_id': doc_id}
+        doc[GENERATION_KEY] = other_generation
+        doc[TRANSACTION_ID_KEY] = other_transaction_id
+        self._database.save(doc)
+
+    def get_transaction_log(self):
+        """
+        This is only for the test suite, it is not part of the api.
+
+        :return: The complete transaction log.
+        :rtype: [(str, str)]
+        """
+        log = self._get_transaction_log()
+        return map(lambda i: (i[1], i[2]), log)
+
+    def _get_gen_docs(
+            self, start=0, end=9999999999, descending=None, limit=None):
+        params = {}
+        if descending:
+            params['descending'] = 'true'
+            # honor couch way of traversing the view tree in reverse order
+            start, end = end, start
+        params['startkey'] = _get_gen_doc_id(start)
+        params['endkey'] = _get_gen_doc_id(end)
+        params['include_docs'] = 'true'
+        if limit:
+            params['limit'] = limit
+        view = self._database.view("_all_docs", **params)
+        return view.rows
+
+    def _get_transaction_log(self, start=0, end=9999999999):
+        # get current gen and trans_id
+        rows = self._get_gen_docs(start=start, end=end)
+        log = []
+        for row in rows:
+            doc = row['doc']
+            log.append((
+                doc[GENERATION_KEY],
+                doc[DOC_ID_KEY],
+                doc[TRANSACTION_ID_KEY]))
+        return log
+
+    def whats_changed(self, old_generation=0):
+        """
+        Return a list of documents that have changed since old_generation.
+
+        :param old_generation: The generation of the database in the old
+                               state.
+        :type old_generation: int
+
+        :return: (generation, trans_id, [(doc_id, generation, trans_id),...])
+                 The current generation of the database, its associated
+                 transaction id, and a list of of changed documents since
+                 old_generation, represented by tuples with for each document
+                 its doc_id and the generation and transaction id corresponding
+                 to the last intervening change and sorted by generation (old
+                 changes first)
+        :rtype: (int, str, [(str, int, str)])
+        """
+        changes = []
+        cur_generation, last_trans_id = self.get_generation_info()
+        relevant_tail = self._get_transaction_log(start=old_generation + 1)
+        seen = set()
+        for generation, doc_id, trans_id in reversed(relevant_tail):
+            if doc_id not in seen:
+                changes.append((doc_id, generation, trans_id))
+                seen.add(doc_id)
+        changes.reverse()
+        return (cur_generation, last_trans_id, changes)
+
+    def get_generation_info(self):
+        """
+        Return the current generation.
+
+        :return: A tuple containing the current generation and transaction id.
+        :rtype: (int, str)
+        """
+        if self.batching and self.batch_generation:
+            return self.batch_generation
+        rows = self._get_gen_docs(descending=True, limit=1)
+        if not rows:
+            return 0, ''
+        gen_doc = rows.pop()['doc']
+        return gen_doc[GENERATION_KEY], gen_doc[TRANSACTION_ID_KEY]
+
+    def json_from_resource(self, doc_path, **kwargs):
+        """
+        Get a resource from it's path and gets a doc's JSON using provided
+        parameters.
+
+        :param doc_path: The path to resource.
+        :type doc_path: [str]
+
+        :return: The request's data parsed from JSON to a dict.
+        :rtype: dict
+        """
+        if doc_path is not None:
+            resource = self._database.resource(*doc_path)
+        else:
+            resource = self._database.resource()
+        _, _, data = resource.get_json(**kwargs)
+        return data
+
+    def _allocate_new_generation(self, doc_id, transaction_id, save=True):
+        """
+        Allocate a new generation number for a document modification.
+
+        We need to allocate a new generation to this document modification by
+        creating a new gen doc. In order to avoid concurrent database updates
+        from allocating the same new generation, we will try to create the
+        document until we succeed, meaning that no other piece of code holds
+        the same generation number as ours.
+
+        The loop below would only be executed more than once if:
+
+          1. there's more than one thread trying to modify the user's database,
+             and
+
+          2. the execution of getting the current generation and saving the gen
+             doc different threads get interleaved (one of them will succeed
+             and the others will fail and try again).
+
+        Number 1 only happens when more than one user device is syncing at the
+        same time. Number 2 depends on not-so-frequent coincidence of
+        code execution.
+
+        Also, in the race between threads for a generation number there's
+        always one thread that wins. so if there are N threads in the race, the
+        expected number of repetitions of the loop for each thread would be
+        N/2. If N is equal to the number of devices that the user has, the
+        number of possible repetitions of the loop should always be low.
+        """
+        while True:
+            try:
+                # add the gen document
+                gen, _ = self.get_generation_info()
+                new_gen = gen + 1
+                gen_doc = {
+                    '_id': _get_gen_doc_id(new_gen),
+                    GENERATION_KEY: new_gen,
+                    DOC_ID_KEY: doc_id,
+                    TRANSACTION_ID_KEY: transaction_id,
+                }
+                if save:
+                    self._database.save(gen_doc)
+                break  # succeeded allocating a new generation, proceed
+            except ResourceConflict:
+                pass  # try again!
+        return gen_doc
+
+    def save_document(self, old_doc, doc, transaction_id):
+        """
+        Put the document in the Couch backend database.
+
+        Note that C{old_doc} must have been fetched with the parameter
+        C{check_for_conflicts} equal to True, so we can properly update the
+        new document using the conflict information from the old one.
+
+        :param old_doc: The old document version.
+        :type old_doc: ServerDocument
+        :param doc: The document to be put.
+        :type doc: ServerDocument
+
+        :raise RevisionConflict: Raised when trying to update a document but
+                                 couch revisions mismatch.
+        """
+        attachments = {}  # we save content and conflicts as attachments
+        parts = []  # and we put it using couch's multipart PUT
+        # save content as attachment
+        if doc.is_tombstone() is False:
+            content = doc.get_json()
+            attachments['u1db_content'] = {
+                'follows': True,
+                'content_type': 'application/octet-stream',
+                'length': len(content),
+            }
+            parts.append(content)
+
+        # save conflicts as attachment
+        if doc.has_conflicts is True:
+            conflicts = json.dumps(
+                map(lambda cdoc: (cdoc.rev, cdoc.content),
+                    doc.get_conflicts()))
+            attachments['u1db_conflicts'] = {
+                'follows': True,
+                'content_type': 'application/octet-stream',
+                'length': len(conflicts),
+            }
+            parts.append(conflicts)
+
+        # build the couch document
+        couch_doc = {
+            '_id': doc.doc_id,
+            'u1db_rev': doc.rev,
+            '_attachments': attachments,
+        }
+        # if we are updating a doc we have to add the couch doc revision
+        if old_doc is not None and hasattr(old_doc, 'couch_rev'):
+            couch_doc['_rev'] = old_doc.couch_rev
+        # prepare the multipart PUT
+        if not self.batching:
+            buf = StringIO()
+            envelope = MultipartWriter(buf)
+            # the order in which attachments are described inside the
+            # serialization of the couch document must match the order in
+            # which they are actually written in the multipart structure.
+            # Because of that, we use `sorted_keys=True` in the json
+            # serialization (so "u1db_conflicts" comes before
+            # "u1db_content" on the couch document attachments
+            # description), and also reverse the order of the parts before
+            # writing them, so the "conflict" part is written before the
+            # "content" part.
+            envelope.add(
+                'application/json',
+                json.dumps(couch_doc, sort_keys=True))
+            parts.reverse()
+            for part in parts:
+                envelope.add('application/octet-stream', part)
+            envelope.close()
+            # try to save and fail if there's a revision conflict
+            try:
+                resource = self._new_resource()
+                resource.put_json(
+                    doc.doc_id, body=str(buf.getvalue()),
+                    headers=envelope.headers)
+            except ResourceConflict:
+                raise RevisionConflict()
+            self._allocate_new_generation(doc.doc_id, transaction_id)
+        else:
+            for name, attachment in attachments.items():
+                del attachment['follows']
+                del attachment['length']
+                index = 0 if name is 'u1db_content' else 1
+                attachment['data'] = binascii.b2a_base64(
+                    parts[index]).strip()
+            couch_doc['_attachments'] = attachments
+            gen_doc = self._allocate_new_generation(
+                doc.doc_id, transaction_id, save=False)
+            self.batch_docs[doc.doc_id] = couch_doc
+            self.batch_docs[gen_doc['_id']] = gen_doc
+            last_gen, last_trans_id = self.batch_generation
+            self.batch_generation = (last_gen + 1, transaction_id)
+
+    def _new_resource(self, *path):
+        """
+        Return a new resource for accessing a couch database.
+
+        :return: A resource for accessing a couch database.
+        :rtype: couchdb.http.Resource
+        """
+        # Workaround for: https://leap.se/code/issues/5448
+        url = couch_urljoin(self._database.resource.url, *path)
+        resource = Resource(url, Session(timeout=COUCH_TIMEOUT))
+        resource.credentials = self._database.resource.credentials
+        resource.headers = self._database.resource.headers.copy()
+        return resource