shadow-repos: use safer way to destroy shadow repositories.

we had reported errors on removal of shadow repos. Not reproduced, however
suspecting are filesystem sync/symlink race-conditions on shared storage.

End result were existing shadow-repo directories that tricked rhodecode
into thinking shadow repos is existing, but infact it was a dummy
structure semi-removed. Using shutil.move we ENSURE rhodecode doesn't
read those back even if removal fails.

rhodecode/lib/vcs/backends/base.py

@@ -1,1710 +1,1730 b''
 # -*- coding: utf-8 -*-
 
 # Copyright (C) 2014-2018 RhodeCode GmbH
 #
 # This program is free software: you can redistribute it and/or modify
 # it under the terms of the GNU Affero General Public License, version 3
 # (only), as published by the Free Software Foundation.
 #
 # 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 Affero General Public License
 # along with this program.  If not, see <http://www.gnu.org/licenses/>.
 #
 # This program is dual-licensed. If you wish to learn more about the
 # RhodeCode Enterprise Edition, including its added features, Support services,
 # and proprietary license terms, please see https://rhodecode.com/licenses/
 
 """
 Base module for all VCS systems
 """
 
 import collections
 import datetime
 import fnmatch
 import itertools
 import logging
 import os
 import re
 import time
 import warnings
+import shutil
 
 from zope.cachedescriptors.property import Lazy as LazyProperty
 
 from rhodecode.lib.utils2 import safe_str, safe_unicode
 from rhodecode.lib.vcs import connection
 from rhodecode.lib.vcs.utils import author_name, author_email
 from rhodecode.lib.vcs.conf import settings
 from rhodecode.lib.vcs.exceptions import (
     CommitError, EmptyRepositoryError, NodeAlreadyAddedError,
     NodeAlreadyChangedError, NodeAlreadyExistsError, NodeAlreadyRemovedError,
     NodeDoesNotExistError, NodeNotChangedError, VCSError,
     ImproperArchiveTypeError, BranchDoesNotExistError, CommitDoesNotExistError,
     RepositoryError)
 
 
 log = logging.getLogger(__name__)
 
 
 FILEMODE_DEFAULT = 0100644
 FILEMODE_EXECUTABLE = 0100755
 
 Reference = collections.namedtuple('Reference', ('type', 'name', 'commit_id'))
 MergeResponse = collections.namedtuple(
     'MergeResponse',
     ('possible', 'executed', 'merge_ref', 'failure_reason'))
 
 
 class MergeFailureReason(object):
     """
     Enumeration with all the reasons why the server side merge could fail.
 
     DO NOT change the number of the reasons, as they may be stored in the
     database.
 
     Changing the name of a reason is acceptable and encouraged to deprecate old
     reasons.
     """
 
     # Everything went well.
     NONE = 0
 
     # An unexpected exception was raised. Check the logs for more details.
     UNKNOWN = 1
 
     # The merge was not successful, there are conflicts.
     MERGE_FAILED = 2
 
     # The merge succeeded but we could not push it to the target repository.
     PUSH_FAILED = 3
 
     # The specified target is not a head in the target repository.
     TARGET_IS_NOT_HEAD = 4
 
     # The source repository contains more branches than the target. Pushing
     # the merge will create additional branches in the target.
     HG_SOURCE_HAS_MORE_BRANCHES = 5
 
     # The target reference has multiple heads. That does not allow to correctly
     # identify the target location. This could only happen for mercurial
     # branches.
     HG_TARGET_HAS_MULTIPLE_HEADS = 6
 
     # The target repository is locked
     TARGET_IS_LOCKED = 7
 
     # Deprecated, use MISSING_TARGET_REF or MISSING_SOURCE_REF instead.
     # A involved commit could not be found.
     _DEPRECATED_MISSING_COMMIT = 8
 
     # The target repo reference is missing.
     MISSING_TARGET_REF = 9
 
     # The source repo reference is missing.
     MISSING_SOURCE_REF = 10
 
     # The merge was not successful, there are conflicts related to sub
     # repositories.
     SUBREPO_MERGE_FAILED = 11
 
 
 class UpdateFailureReason(object):
     """
     Enumeration with all the reasons why the pull request update could fail.
 
     DO NOT change the number of the reasons, as they may be stored in the
     database.
 
     Changing the name of a reason is acceptable and encouraged to deprecate old
     reasons.
     """
 
     # Everything went well.
     NONE = 0
 
     # An unexpected exception was raised. Check the logs for more details.
     UNKNOWN = 1
 
     # The pull request is up to date.
     NO_CHANGE = 2
 
     # The pull request has a reference type that is not supported for update.
     WRONG_REF_TYPE = 3
 
     # Update failed because the target reference is missing.
     MISSING_TARGET_REF = 4
 
     # Update failed because the source reference is missing.
     MISSING_SOURCE_REF = 5
 
 
 class BaseRepository(object):
     """
     Base Repository for final backends
 
     .. attribute:: DEFAULT_BRANCH_NAME
 
        name of default branch (i.e. "trunk" for svn, "master" for git etc.
 
     .. attribute:: commit_ids
 
        list of all available commit ids, in ascending order
 
     .. attribute:: path
 
        absolute path to the repository
 
     .. attribute:: bookmarks
 
        Mapping from name to :term:`Commit ID` of the bookmark. Empty in case
        there are no bookmarks or the backend implementation does not support
        bookmarks.
 
     .. attribute:: tags
 
        Mapping from name to :term:`Commit ID` of the tag.
 
     """
 
     DEFAULT_BRANCH_NAME = None
     DEFAULT_CONTACT = u"Unknown"
     DEFAULT_DESCRIPTION = u"unknown"
     EMPTY_COMMIT_ID = '0' * 40
 
     path = None
     _remote = None
 
     def __init__(self, repo_path, config=None, create=False, **kwargs):
         """
         Initializes repository. Raises RepositoryError if repository could
         not be find at the given ``repo_path`` or directory at ``repo_path``
         exists and ``create`` is set to True.
 
         :param repo_path: local path of the repository
         :param config: repository configuration
         :param create=False: if set to True, would try to create repository.
         :param src_url=None: if set, should be proper url from which repository
           would be cloned; requires ``create`` parameter to be set to True -
           raises RepositoryError if src_url is set and create evaluates to
           False
         """
         raise NotImplementedError
 
     def __repr__(self):
         return '<%s at %s>' % (self.__class__.__name__, self.path)
 
     def __len__(self):
         return self.count()
 
     def __eq__(self, other):
         same_instance = isinstance(other, self.__class__)
         return same_instance and other.path == self.path
 
     def __ne__(self, other):
         return not self.__eq__(other)
 
     def get_create_shadow_cache_pr_path(self, db_repo):
         path = db_repo.cached_diffs_dir
         if not os.path.exists(path):
             os.makedirs(path, 0755)
         return path
 
     @classmethod
     def get_default_config(cls, default=None):
         config = Config()
         if default and isinstance(default, list):
             for section, key, val in default:
                 config.set(section, key, val)
         return config
 
     @LazyProperty
     def EMPTY_COMMIT(self):
         return EmptyCommit(self.EMPTY_COMMIT_ID)
 
     @LazyProperty
     def alias(self):
         for k, v in settings.BACKENDS.items():
             if v.split('.')[-1] == str(self.__class__.__name__):
                 return k
 
     @LazyProperty
     def name(self):
         return safe_unicode(os.path.basename(self.path))
 
     @LazyProperty
     def description(self):
         raise NotImplementedError
 
     def refs(self):
         """
         returns a `dict` with branches, bookmarks, tags, and closed_branches
         for this repository
         """
         return dict(
             branches=self.branches,
             branches_closed=self.branches_closed,
             tags=self.tags,
             bookmarks=self.bookmarks
         )
 
     @LazyProperty
     def branches(self):
         """
         A `dict` which maps branch names to commit ids.
         """
         raise NotImplementedError
 
     @LazyProperty
     def branches_closed(self):
         """
         A `dict` which maps tags names to commit ids.
         """
         raise NotImplementedError
 
     @LazyProperty
     def bookmarks(self):
         """
         A `dict` which maps tags names to commit ids.
         """
         raise NotImplementedError
 
     @LazyProperty
     def tags(self):
         """
         A `dict` which maps tags names to commit ids.
         """
         raise NotImplementedError
 
     @LazyProperty
     def size(self):
         """
         Returns combined size in bytes for all repository files
         """
         tip = self.get_commit()
         return tip.size
 
     def size_at_commit(self, commit_id):
         commit = self.get_commit(commit_id)
         return commit.size
 
     def is_empty(self):
         return not bool(self.commit_ids)
 
     @staticmethod
     def check_url(url, config):
         """
         Function will check given url and try to verify if it's a valid
         link.
         """
         raise NotImplementedError
 
     @staticmethod
     def is_valid_repository(path):
         """
         Check if given `path` contains a valid repository of this backend
         """
         raise NotImplementedError
 
     # ==========================================================================
     # COMMITS
     # ==========================================================================
 
     def get_commit(self, commit_id=None, commit_idx=None, pre_load=None):
         """
         Returns instance of `BaseCommit` class. If `commit_id` and `commit_idx`
         are both None, most recent commit is returned.
 
         :param pre_load: Optional. List of commit attributes to load.
 
         :raises ``EmptyRepositoryError``: if there are no commits
         """
         raise NotImplementedError
 
     def __iter__(self):
         for commit_id in self.commit_ids:
             yield self.get_commit(commit_id=commit_id)
 
     def get_commits(
             self, start_id=None, end_id=None, start_date=None, end_date=None,
             branch_name=None, show_hidden=False, pre_load=None):
         """
         Returns iterator of `BaseCommit` objects from start to end
         not inclusive. This should behave just like a list, ie. end is not
         inclusive.
 
         :param start_id: None or str, must be a valid commit id
         :param end_id: None or str, must be a valid commit id
         :param start_date:
         :param end_date:
         :param branch_name:
         :param show_hidden:
         :param pre_load:
         """
         raise NotImplementedError
 
     def __getitem__(self, key):
         """
         Allows index based access to the commit objects of this repository.
         """
         pre_load = ["author", "branch", "date", "message", "parents"]
         if isinstance(key, slice):
             return self._get_range(key, pre_load)
         return self.get_commit(commit_idx=key, pre_load=pre_load)
 
     def _get_range(self, slice_obj, pre_load):
         for commit_id in self.commit_ids.__getitem__(slice_obj):
             yield self.get_commit(commit_id=commit_id, pre_load=pre_load)
 
     def count(self):
         return len(self.commit_ids)
 
     def tag(self, name, user, commit_id=None, message=None, date=None, **opts):
         """
         Creates and returns a tag for the given ``commit_id``.
 
         :param name: name for new tag
         :param user: full username, i.e.: "Joe Doe <[e-mail]>"
         :param commit_id: commit id for which new tag would be created
         :param message: message of the tag's commit
         :param date: date of tag's commit
 
         :raises TagAlreadyExistError: if tag with same name already exists
         """
         raise NotImplementedError
 
     def remove_tag(self, name, user, message=None, date=None):
         """
         Removes tag with the given ``name``.
 
         :param name: name of the tag to be removed
         :param user: full username, i.e.: "Joe Doe <[e-mail]>"
         :param message: message of the tag's removal commit
         :param date: date of tag's removal commit
 
         :raises TagDoesNotExistError: if tag with given name does not exists
         """
         raise NotImplementedError
 
     def get_diff(
             self, commit1, commit2, path=None, ignore_whitespace=False,
             context=3, path1=None):
         """
         Returns (git like) *diff*, as plain text. Shows changes introduced by
         `commit2` since `commit1`.
 
         :param commit1: Entry point from which diff is shown. Can be
           ``self.EMPTY_COMMIT`` - in this case, patch showing all
           the changes since empty state of the repository until `commit2`
         :param commit2: Until which commit changes should be shown.
         :param path: Can be set to a path of a file to create a diff of that
           file. If `path1` is also set, this value is only associated to
           `commit2`.
         :param ignore_whitespace: If set to ``True``, would not show whitespace
           changes. Defaults to ``False``.
         :param context: How many lines before/after changed lines should be
           shown. Defaults to ``3``.
         :param path1: Can be set to a path to associate with `commit1`. This
           parameter works only for backends which support diff generation for
           different paths. Other backends will raise a `ValueError` if `path1`
           is set and has a different value than `path`.
         :param file_path: filter this diff by given path pattern
         """
         raise NotImplementedError
 
     def strip(self, commit_id, branch=None):
         """
         Strip given commit_id from the repository
         """
         raise NotImplementedError
 
     def get_common_ancestor(self, commit_id1, commit_id2, repo2):
         """
         Return a latest common ancestor commit if one exists for this repo
         `commit_id1` vs `commit_id2` from `repo2`.
 
         :param commit_id1: Commit it from this repository to use as a
            target for the comparison.
         :param commit_id2: Source commit id to use for comparison.
         :param repo2: Source repository to use for comparison.
         """
         raise NotImplementedError
 
     def compare(self, commit_id1, commit_id2, repo2, merge, pre_load=None):
         """
         Compare this repository's revision `commit_id1` with `commit_id2`.
 
         Returns a tuple(commits, ancestor) that would be merged from
         `commit_id2`. Doing a normal compare (``merge=False``), ``None``
         will be returned as ancestor.
 
         :param commit_id1: Commit it from this repository to use as a
            target for the comparison.
         :param commit_id2: Source commit id to use for comparison.
         :param repo2: Source repository to use for comparison.
         :param merge: If set to ``True`` will do a merge compare which also
            returns the common ancestor.
         :param pre_load: Optional. List of commit attributes to load.
         """
         raise NotImplementedError
 
     def merge(self, target_ref, source_repo, source_ref, workspace_id,
               user_name='', user_email='', message='', dry_run=False,
               use_rebase=False, close_branch=False):
         """
         Merge the revisions specified in `source_ref` from `source_repo`
         onto the `target_ref` of this repository.
 
         `source_ref` and `target_ref` are named tupls with the following
         fields `type`, `name` and `commit_id`.
 
         Returns a MergeResponse named tuple with the following fields
         'possible', 'executed', 'source_commit', 'target_commit',
         'merge_commit'.
 
         :param target_ref: `target_ref` points to the commit on top of which
             the `source_ref` should be merged.
         :param source_repo: The repository that contains the commits to be
             merged.
         :param source_ref: `source_ref` points to the topmost commit from
             the `source_repo` which should be merged.
         :param workspace_id: `workspace_id` unique identifier.
         :param user_name: Merge commit `user_name`.
         :param user_email: Merge commit `user_email`.
         :param message: Merge commit `message`.
         :param dry_run: If `True` the merge will not take place.
         :param use_rebase: If `True` commits from the source will be rebased
             on top of the target instead of being merged.
         :param close_branch: If `True` branch will be close before merging it
         """
         if dry_run:
             message = message or 'dry_run_merge_message'
             user_email = user_email or '[e-mail]'
             user_name = user_name or 'Dry-Run User'
         else:
             if not user_name:
                 raise ValueError('user_name cannot be empty')
             if not user_email:
                 raise ValueError('user_email cannot be empty')
             if not message:
                 raise ValueError('message cannot be empty')
 
         shadow_repository_path = self._maybe_prepare_merge_workspace(
             workspace_id, target_ref, source_ref)
 
         try:
             return self._merge_repo(
                 shadow_repository_path, target_ref, source_repo,
                 source_ref, message, user_name, user_email, dry_run=dry_run,
                 use_rebase=use_rebase, close_branch=close_branch)
         except RepositoryError:
             log.exception(
                 'Unexpected failure when running merge, dry-run=%s',
                 dry_run)
             return MergeResponse(
                 False, False, None, MergeFailureReason.UNKNOWN)
 
     def _merge_repo(self, shadow_repository_path, target_ref,
                     source_repo, source_ref, merge_message,
                     merger_name, merger_email, dry_run=False,
                     use_rebase=False, close_branch=False):
         """Internal implementation of merge."""
         raise NotImplementedError
 
     def _maybe_prepare_merge_workspace(self, workspace_id, target_ref, source_ref):
         """
         Create the merge workspace.
 
         :param workspace_id: `workspace_id` unique identifier.
         """
         raise NotImplementedError
 
+    def _get_shadow_repository_path(self, workspace_id):
+        raise NotImplementedError
+
     def cleanup_merge_workspace(self, workspace_id):
         """
         Remove merge workspace.
 
         This function MUST not fail in case there is no workspace associated to
         the given `workspace_id`.
 
         :param workspace_id: `workspace_id` unique identifier.
         """
-        raise NotImplementedError
+        shadow_repository_path = self._get_shadow_repository_path(workspace_id)
+        shadow_repository_path_del = '{}.{}.delete'.format(
+            shadow_repository_path, time.time())
+
+        # move the shadow repo, so it never conflicts with the one used.
+        # we use this method because shutil.rmtree had some edge case problems
+        # removing symlinked repositories
+        if not os.path.isdir(shadow_repository_path):
+            return
+
+        shutil.move(shadow_repository_path, shadow_repository_path_del)
+        try:
+            shutil.rmtree(shadow_repository_path_del, ignore_errors=False)
+        except Exception:
+            log.exception('Failed to gracefully remove shadow repo under %s',
+                          shadow_repository_path_del)
+            shutil.rmtree(shadow_repository_path_del, ignore_errors=True)
 
     # ========== #
     # COMMIT API #
     # ========== #
 
     @LazyProperty
     def in_memory_commit(self):
         """
         Returns :class:`InMemoryCommit` object for this repository.
         """
         raise NotImplementedError
 
     # ======================== #
     # UTILITIES FOR SUBCLASSES #
     # ======================== #
 
     def _validate_diff_commits(self, commit1, commit2):
         """
         Validates that the given commits are related to this repository.
 
         Intended as a utility for sub classes to have a consistent validation
         of input parameters in methods like :meth:`get_diff`.
         """
         self._validate_commit(commit1)
         self._validate_commit(commit2)
         if (isinstance(commit1, EmptyCommit) and
                 isinstance(commit2, EmptyCommit)):
             raise ValueError("Cannot compare two empty commits")
 
     def _validate_commit(self, commit):
         if not isinstance(commit, BaseCommit):
             raise TypeError(
                 "%s is not of type BaseCommit" % repr(commit))
         if commit.repository != self and not isinstance(commit, EmptyCommit):
             raise ValueError(
                 "Commit %s must be a valid commit from this repository %s, "
                 "related to this repository instead %s." %
                 (commit, self, commit.repository))
 
     def _validate_commit_id(self, commit_id):
         if not isinstance(commit_id, basestring):
             raise TypeError("commit_id must be a string value")
 
     def _validate_commit_idx(self, commit_idx):
         if not isinstance(commit_idx, (int, long)):
             raise TypeError("commit_idx must be a numeric value")
 
     def _validate_branch_name(self, branch_name):
         if branch_name and branch_name not in self.branches_all:
             msg = ("Branch %s not found in %s" % (branch_name, self))
             raise BranchDoesNotExistError(msg)
 
     #
     # Supporting deprecated API parts
     # TODO: johbo: consider to move this into a mixin
     #
 
     @property
     def EMPTY_CHANGESET(self):
         warnings.warn(
             "Use EMPTY_COMMIT or EMPTY_COMMIT_ID instead", DeprecationWarning)
         return self.EMPTY_COMMIT_ID
 
     @property
     def revisions(self):
         warnings.warn("Use commits attribute instead", DeprecationWarning)
         return self.commit_ids
 
     @revisions.setter
     def revisions(self, value):
         warnings.warn("Use commits attribute instead", DeprecationWarning)
         self.commit_ids = value
 
     def get_changeset(self, revision=None, pre_load=None):
         warnings.warn("Use get_commit instead", DeprecationWarning)
         commit_id = None
         commit_idx = None
         if isinstance(revision, basestring):
             commit_id = revision
         else:
             commit_idx = revision
         return self.get_commit(
             commit_id=commit_id, commit_idx=commit_idx, pre_load=pre_load)
 
     def get_changesets(
             self, start=None, end=None, start_date=None, end_date=None,
             branch_name=None, pre_load=None):
         warnings.warn("Use get_commits instead", DeprecationWarning)
         start_id = self._revision_to_commit(start)
         end_id = self._revision_to_commit(end)
         return self.get_commits(
             start_id=start_id, end_id=end_id, start_date=start_date,
             end_date=end_date, branch_name=branch_name, pre_load=pre_load)
 
     def _revision_to_commit(self, revision):
         """
         Translates a revision to a commit_id
 
         Helps to support the old changeset based API which allows to use
         commit ids and commit indices interchangeable.
         """
         if revision is None:
             return revision
 
         if isinstance(revision, basestring):
             commit_id = revision
         else:
             commit_id = self.commit_ids[revision]
         return commit_id
 
     @property
     def in_memory_changeset(self):
         warnings.warn("Use in_memory_commit instead", DeprecationWarning)
         return self.in_memory_commit
 
     def get_path_permissions(self, username):
         """
         Returns a path permission checker or None if not supported
 
         :param username: session user name
         :return: an instance of BasePathPermissionChecker or None
         """
         return None
 
     def install_hooks(self, force=False):
         return self._remote.install_hooks(force)
 
 
 class BaseCommit(object):
     """
     Each backend should implement it's commit representation.
 
     **Attributes**
 
         ``repository``
             repository object within which commit exists
 
         ``id``
             The commit id, may be ``raw_id`` or i.e. for mercurial's tip
             just ``tip``.
 
         ``raw_id``
             raw commit representation (i.e. full 40 length sha for git
             backend)
 
         ``short_id``
             shortened (if apply) version of ``raw_id``; it would be simple
             shortcut for ``raw_id[:12]`` for git/mercurial backends or same
             as ``raw_id`` for subversion
 
         ``idx``
             commit index
 
         ``files``
             list of ``FileNode`` (``Node`` with NodeKind.FILE) objects
 
         ``dirs``
             list of ``DirNode`` (``Node`` with NodeKind.DIR) objects
 
         ``nodes``
             combined list of ``Node`` objects
 
         ``author``
             author of the commit, as unicode
 
         ``message``
             message of the commit, as unicode
 
         ``parents``
             list of parent commits
 
     """
 
     branch = None
     """
     Depending on the backend this should be set to the branch name of the
     commit. Backends not supporting branches on commits should leave this
     value as ``None``.
     """
 
     _ARCHIVE_PREFIX_TEMPLATE = b'{repo_name}-{short_id}'
     """
     This template is used to generate a default prefix for repository archives
     if no prefix has been specified.
     """
 
     def __str__(self):
         return '<%s at %s:%s>' % (
             self.__class__.__name__, self.idx, self.short_id)
 
     def __repr__(self):
         return self.__str__()
 
     def __unicode__(self):
         return u'%s:%s' % (self.idx, self.short_id)
 
     def __eq__(self, other):
         same_instance = isinstance(other, self.__class__)
         return same_instance and self.raw_id == other.raw_id
 
     def __json__(self):
         parents = []
         try:
             for parent in self.parents:
                 parents.append({'raw_id': parent.raw_id})
         except NotImplementedError:
             # empty commit doesn't have parents implemented
             pass
 
         return {
             'short_id': self.short_id,
             'raw_id': self.raw_id,
             'revision': self.idx,
             'message': self.message,
             'date': self.date,
             'author': self.author,
             'parents': parents,
             'branch': self.branch
         }
 
     def __getstate__(self):
         d = self.__dict__.copy()
         d.pop('_remote', None)
         d.pop('repository', None)
         return d
 
     def _get_refs(self):
         return {
             'branches': [self.branch] if self.branch else [],
             'bookmarks': getattr(self, 'bookmarks', []),
             'tags': self.tags
         }
 
     @LazyProperty
     def last(self):
         """
         ``True`` if this is last commit in repository, ``False``
         otherwise; trying to access this attribute while there is no
         commits would raise `EmptyRepositoryError`
         """
         if self.repository is None:
             raise CommitError("Cannot check if it's most recent commit")
         return self.raw_id == self.repository.commit_ids[-1]
 
     @LazyProperty
     def parents(self):
         """
         Returns list of parent commits.
         """
         raise NotImplementedError
 
     @property
     def merge(self):
         """
         Returns boolean if commit is a merge.
         """
         return len(self.parents) > 1
 
     @LazyProperty
     def children(self):
         """
         Returns list of child commits.
         """
         raise NotImplementedError
 
     @LazyProperty
     def id(self):
         """
         Returns string identifying this commit.
         """
         raise NotImplementedError
 
     @LazyProperty
     def raw_id(self):
         """
         Returns raw string identifying this commit.
         """
         raise NotImplementedError
 
     @LazyProperty
     def short_id(self):
         """
         Returns shortened version of ``raw_id`` attribute, as string,
         identifying this commit, useful for presentation to users.
         """
         raise NotImplementedError
 
     @LazyProperty
     def idx(self):
         """
         Returns integer identifying this commit.
         """
         raise NotImplementedError
 
     @LazyProperty
     def committer(self):
         """
         Returns committer for this commit
         """
         raise NotImplementedError
 
     @LazyProperty
     def committer_name(self):
         """
         Returns committer name for this commit
         """
 
         return author_name(self.committer)
 
     @LazyProperty
     def committer_email(self):
         """
         Returns committer email address for this commit
         """
 
         return author_email(self.committer)
 
     @LazyProperty
     def author(self):
         """
         Returns author for this commit
         """
 
         raise NotImplementedError
 
     @LazyProperty
     def author_name(self):
         """
         Returns author name for this commit
         """
 
         return author_name(self.author)
 
     @LazyProperty
     def author_email(self):
         """
         Returns author email address for this commit
         """
 
         return author_email(self.author)
 
     def get_file_mode(self, path):
         """
         Returns stat mode of the file at `path`.
         """
         raise NotImplementedError
 
     def is_link(self, path):
         """
         Returns ``True`` if given `path` is a symlink
         """
         raise NotImplementedError
 
     def get_file_content(self, path):
         """
         Returns content of the file at the given `path`.
         """
         raise NotImplementedError
 
     def get_file_size(self, path):
         """
         Returns size of the file at the given `path`.
         """
         raise NotImplementedError
 
     def get_file_commit(self, path, pre_load=None):
         """
         Returns last commit of the file at the given `path`.
 
         :param pre_load: Optional. List of commit attributes to load.
         """
         commits = self.get_file_history(path, limit=1, pre_load=pre_load)
         if not commits:
             raise RepositoryError(
                 'Failed to fetch history for path {}. '
                 'Please check if such path exists in your repository'.format(
                     path))
         return commits[0]
 
     def get_file_history(self, path, limit=None, pre_load=None):
         """
         Returns history of file as reversed list of :class:`BaseCommit`
         objects for which file at given `path` has been modified.
 
         :param limit: Optional. Allows to limit the size of the returned
            history. This is intended as a hint to the underlying backend, so
            that it can apply optimizations depending on the limit.
         :param pre_load: Optional. List of commit attributes to load.
         """
         raise NotImplementedError
 
     def get_file_annotate(self, path, pre_load=None):
         """
         Returns a generator of four element tuples with
         lineno, sha, commit lazy loader and line
 
         :param pre_load: Optional. List of commit attributes to load.
         """
         raise NotImplementedError
 
     def get_nodes(self, path):
         """
         Returns combined ``DirNode`` and ``FileNode`` objects list representing
         state of commit at the given ``path``.
 
         :raises ``CommitError``: if node at the given ``path`` is not
           instance of ``DirNode``
         """
         raise NotImplementedError
 
     def get_node(self, path):
         """
         Returns ``Node`` object from the given ``path``.
 
         :raises ``NodeDoesNotExistError``: if there is no node at the given
           ``path``
         """
         raise NotImplementedError
 
     def get_largefile_node(self, path):
         """
         Returns the path to largefile from Mercurial/Git-lfs storage.
         or None if it's not a largefile node
         """
         return None
 
     def archive_repo(self, file_path, kind='tgz', subrepos=None,
                      prefix=None, write_metadata=False, mtime=None):
         """
         Creates an archive containing the contents of the repository.
 
         :param file_path: path to the file which to create the archive.
         :param kind: one of following: ``"tbz2"``, ``"tgz"``, ``"zip"``.
         :param prefix: name of root directory in archive.
             Default is repository name and commit's short_id joined with dash:
             ``"{repo_name}-{short_id}"``.
         :param write_metadata: write a metadata file into archive.
         :param mtime: custom modification time for archive creation, defaults
             to time.time() if not given.
 
         :raise VCSError: If prefix has a problem.
         """
         allowed_kinds = settings.ARCHIVE_SPECS.keys()
         if kind not in allowed_kinds:
             raise ImproperArchiveTypeError(
                 'Archive kind (%s) not supported use one of %s' %
                 (kind, allowed_kinds))
 
         prefix = self._validate_archive_prefix(prefix)
 
         mtime = mtime or time.mktime(self.date.timetuple())
 
         file_info = []
         cur_rev = self.repository.get_commit(commit_id=self.raw_id)
         for _r, _d, files in cur_rev.walk('/'):
             for f in files:
                 f_path = os.path.join(prefix, f.path)
                 file_info.append(
                     (f_path, f.mode, f.is_link(), f.raw_bytes))
 
         if write_metadata:
             metadata = [
                 ('repo_name', self.repository.name),
                 ('rev', self.raw_id),
                 ('create_time', mtime),
                 ('branch', self.branch),
                 ('tags', ','.join(self.tags)),
             ]
             meta = ["%s:%s" % (f_name, value) for f_name, value in metadata]
             file_info.append(('.archival.txt', 0644, False, '\n'.join(meta)))
 
         connection.Hg.archive_repo(file_path, mtime, file_info, kind)
 
     def _validate_archive_prefix(self, prefix):
         if prefix is None:
             prefix = self._ARCHIVE_PREFIX_TEMPLATE.format(
                 repo_name=safe_str(self.repository.name),
                 short_id=self.short_id)
         elif not isinstance(prefix, str):
             raise ValueError("prefix not a bytes object: %s" % repr(prefix))
         elif prefix.startswith('/'):
             raise VCSError("Prefix cannot start with leading slash")
         elif prefix.strip() == '':
             raise VCSError("Prefix cannot be empty")
         return prefix
 
     @LazyProperty
     def root(self):
         """
         Returns ``RootNode`` object for this commit.
         """
         return self.get_node('')
 
     def next(self, branch=None):
         """
         Returns next commit from current, if branch is gives it will return
         next commit belonging to this branch
 
         :param branch: show commits within the given named branch
         """
         indexes = xrange(self.idx + 1, self.repository.count())
         return self._find_next(indexes, branch)
 
     def prev(self, branch=None):
         """
         Returns previous commit from current, if branch is gives it will
         return previous commit belonging to this branch
 
         :param branch: show commit within the given named branch
         """
         indexes = xrange(self.idx - 1, -1, -1)
         return self._find_next(indexes, branch)
 
     def _find_next(self, indexes, branch=None):
         if branch and self.branch != branch:
             raise VCSError('Branch option used on commit not belonging '
                            'to that branch')
 
         for next_idx in indexes:
             commit = self.repository.get_commit(commit_idx=next_idx)
             if branch and branch != commit.branch:
                 continue
             return commit
         raise CommitDoesNotExistError
 
     def diff(self, ignore_whitespace=True, context=3):
         """
         Returns a `Diff` object representing the change made by this commit.
         """
         parent = (
             self.parents[0] if self.parents else self.repository.EMPTY_COMMIT)
         diff = self.repository.get_diff(
             parent, self,
             ignore_whitespace=ignore_whitespace,
             context=context)
         return diff
 
     @LazyProperty
     def added(self):
         """
         Returns list of added ``FileNode`` objects.
         """
         raise NotImplementedError
 
     @LazyProperty
     def changed(self):
         """
         Returns list of modified ``FileNode`` objects.
         """
         raise NotImplementedError
 
     @LazyProperty
     def removed(self):
         """
         Returns list of removed ``FileNode`` objects.
         """
         raise NotImplementedError
 
     @LazyProperty
     def size(self):
         """
         Returns total number of bytes from contents of all filenodes.
         """
         return sum((node.size for node in self.get_filenodes_generator()))
 
     def walk(self, topurl=''):
         """
         Similar to os.walk method. Insted of filesystem it walks through
         commit starting at given ``topurl``.  Returns generator of tuples
         (topnode, dirnodes, filenodes).
         """
         topnode = self.get_node(topurl)
         if not topnode.is_dir():
             return
         yield (topnode, topnode.dirs, topnode.files)
         for dirnode in topnode.dirs:
             for tup in self.walk(dirnode.path):
                 yield tup
 
     def get_filenodes_generator(self):
         """
         Returns generator that yields *all* file nodes.
         """
         for topnode, dirs, files in self.walk():
             for node in files:
                 yield node
 
     #
     # Utilities for sub classes to support consistent behavior
     #
 
     def no_node_at_path(self, path):
         return NodeDoesNotExistError(
             u"There is no file nor directory at the given path: "
             u"`%s` at commit %s" % (safe_unicode(path), self.short_id))
 
     def _fix_path(self, path):
         """
         Paths are stored without trailing slash so we need to get rid off it if
         needed.
         """
         return path.rstrip('/')
 
     #
     # Deprecated API based on changesets
     #
 
     @property
     def revision(self):
         warnings.warn("Use idx instead", DeprecationWarning)
         return self.idx
 
     @revision.setter
     def revision(self, value):
         warnings.warn("Use idx instead", DeprecationWarning)
         self.idx = value
 
     def get_file_changeset(self, path):
         warnings.warn("Use get_file_commit instead", DeprecationWarning)
         return self.get_file_commit(path)
 
 
 class BaseChangesetClass(type):
 
     def __instancecheck__(self, instance):
         return isinstance(instance, BaseCommit)
 
 
 class BaseChangeset(BaseCommit):
 
     __metaclass__ = BaseChangesetClass
 
     def __new__(cls, *args, **kwargs):
         warnings.warn(
             "Use BaseCommit instead of BaseChangeset", DeprecationWarning)
         return super(BaseChangeset, cls).__new__(cls, *args, **kwargs)
 
 
 class BaseInMemoryCommit(object):
     """
     Represents differences between repository's state (most recent head) and
     changes made *in place*.
 
     **Attributes**
 
         ``repository``
             repository object for this in-memory-commit
 
         ``added``
             list of ``FileNode`` objects marked as *added*
 
         ``changed``
             list of ``FileNode`` objects marked as *changed*
 
         ``removed``
             list of ``FileNode`` or ``RemovedFileNode`` objects marked to be
             *removed*
 
         ``parents``
             list of :class:`BaseCommit` instances representing parents of
             in-memory commit. Should always be 2-element sequence.
 
     """
 
     def __init__(self, repository):
         self.repository = repository
         self.added = []
         self.changed = []
         self.removed = []
         self.parents = []
 
     def add(self, *filenodes):
         """
         Marks given ``FileNode`` objects as *to be committed*.
 
         :raises ``NodeAlreadyExistsError``: if node with same path exists at
            latest commit
         :raises ``NodeAlreadyAddedError``: if node with same path is already
            marked as *added*
         """
         # Check if not already marked as *added* first
         for node in filenodes:
             if node.path in (n.path for n in self.added):
                 raise NodeAlreadyAddedError(
                     "Such FileNode %s is already marked for addition"
                     % node.path)
         for node in filenodes:
             self.added.append(node)
 
     def change(self, *filenodes):
         """
         Marks given ``FileNode`` objects to be *changed* in next commit.
 
         :raises ``EmptyRepositoryError``: if there are no commits yet
         :raises ``NodeAlreadyExistsError``: if node with same path is already
            marked to be *changed*
         :raises ``NodeAlreadyRemovedError``: if node with same path is already
            marked to be *removed*
         :raises ``NodeDoesNotExistError``: if node doesn't exist in latest
            commit
         :raises ``NodeNotChangedError``: if node hasn't really be changed
         """
         for node in filenodes:
             if node.path in (n.path for n in self.removed):
                 raise NodeAlreadyRemovedError(
                     "Node at %s is already marked as removed" % node.path)
         try:
             self.repository.get_commit()
         except EmptyRepositoryError:
             raise EmptyRepositoryError(
                 "Nothing to change - try to *add* new nodes rather than "
                 "changing them")
         for node in filenodes:
             if node.path in (n.path for n in self.changed):
                 raise NodeAlreadyChangedError(
                     "Node at '%s' is already marked as changed" % node.path)
             self.changed.append(node)
 
     def remove(self, *filenodes):
         """
         Marks given ``FileNode`` (or ``RemovedFileNode``) objects to be
         *removed* in next commit.
 
         :raises ``NodeAlreadyRemovedError``: if node has been already marked to
            be *removed*
         :raises ``NodeAlreadyChangedError``: if node has been already marked to
            be *changed*
         """
         for node in filenodes:
             if node.path in (n.path for n in self.removed):
                 raise NodeAlreadyRemovedError(
                     "Node is already marked to for removal at %s" % node.path)
             if node.path in (n.path for n in self.changed):
                 raise NodeAlreadyChangedError(
                     "Node is already marked to be changed at %s" % node.path)
             # We only mark node as *removed* - real removal is done by
             # commit method
             self.removed.append(node)
 
     def reset(self):
         """
         Resets this instance to initial state (cleans ``added``, ``changed``
         and ``removed`` lists).
         """
         self.added = []
         self.changed = []
         self.removed = []
         self.parents = []
 
     def get_ipaths(self):
         """
         Returns generator of paths from nodes marked as added, changed or
         removed.
         """
         for node in itertools.chain(self.added, self.changed, self.removed):
             yield node.path
 
     def get_paths(self):
         """
         Returns list of paths from nodes marked as added, changed or removed.
         """
         return list(self.get_ipaths())
 
     def check_integrity(self, parents=None):
         """
         Checks in-memory commit's integrity. Also, sets parents if not
         already set.
 
         :raises CommitError: if any error occurs (i.e.
           ``NodeDoesNotExistError``).
         """
         if not self.parents:
             parents = parents or []
             if len(parents) == 0:
                 try:
                     parents = [self.repository.get_commit(), None]
                 except EmptyRepositoryError:
                     parents = [None, None]
             elif len(parents) == 1:
                 parents += [None]
             self.parents = parents
 
         # Local parents, only if not None
         parents = [p for p in self.parents if p]
 
         # Check nodes marked as added
         for p in parents:
             for node in self.added:
                 try:
                     p.get_node(node.path)
                 except NodeDoesNotExistError:
                     pass
                 else:
                     raise NodeAlreadyExistsError(
                         "Node `%s` already exists at %s" % (node.path, p))
 
         # Check nodes marked as changed
         missing = set(self.changed)
         not_changed = set(self.changed)
         if self.changed and not parents:
             raise NodeDoesNotExistError(str(self.changed[0].path))
         for p in parents:
             for node in self.changed:
                 try:
                     old = p.get_node(node.path)
                     missing.remove(node)
                     # if content actually changed, remove node from not_changed
                     if old.content != node.content:
                         not_changed.remove(node)
                 except NodeDoesNotExistError:
                     pass
         if self.changed and missing:
             raise NodeDoesNotExistError(
                 "Node `%s` marked as modified but missing in parents: %s"
                 % (node.path, parents))
 
         if self.changed and not_changed:
             raise NodeNotChangedError(
                 "Node `%s` wasn't actually changed (parents: %s)"
                 % (not_changed.pop().path, parents))
 
         # Check nodes marked as removed
         if self.removed and not parents:
             raise NodeDoesNotExistError(
                 "Cannot remove node at %s as there "
                 "were no parents specified" % self.removed[0].path)
         really_removed = set()
         for p in parents:
             for node in self.removed:
                 try:
                     p.get_node(node.path)
                     really_removed.add(node)
                 except CommitError:
                     pass
         not_removed = set(self.removed) - really_removed
         if not_removed:
             # TODO: johbo: This code branch does not seem to be covered
             raise NodeDoesNotExistError(
                 "Cannot remove node at %s from "
                 "following parents: %s" % (not_removed, parents))
 
     def commit(
             self, message, author, parents=None, branch=None, date=None,
             **kwargs):
         """
         Performs in-memory commit (doesn't check workdir in any way) and
         returns newly created :class:`BaseCommit`. Updates repository's
         attribute `commits`.
 
         .. note::
 
            While overriding this method each backend's should call
            ``self.check_integrity(parents)`` in the first place.
 
         :param message: message of the commit
         :param author: full username, i.e. "Joe Doe <[e-mail]>"
         :param parents: single parent or sequence of parents from which commit
            would be derived
         :param date: ``datetime.datetime`` instance. Defaults to
            ``datetime.datetime.now()``.
         :param branch: branch name, as string. If none given, default backend's
            branch would be used.
 
         :raises ``CommitError``: if any error occurs while committing
         """
         raise NotImplementedError
 
 
 class BaseInMemoryChangesetClass(type):
 
     def __instancecheck__(self, instance):
         return isinstance(instance, BaseInMemoryCommit)
 
 
 class BaseInMemoryChangeset(BaseInMemoryCommit):
 
     __metaclass__ = BaseInMemoryChangesetClass
 
     def __new__(cls, *args, **kwargs):
         warnings.warn(
             "Use BaseCommit instead of BaseInMemoryCommit", DeprecationWarning)
         return super(BaseInMemoryChangeset, cls).__new__(cls, *args, **kwargs)
 
 
 class EmptyCommit(BaseCommit):
     """
     An dummy empty commit. It's possible to pass hash when creating
     an EmptyCommit
     """
 
     def __init__(
             self, commit_id='0' * 40, repo=None, alias=None, idx=-1,
             message='', author='', date=None):
         self._empty_commit_id = commit_id
         # TODO: johbo: Solve idx parameter, default value does not make
         # too much sense
         self.idx = idx
         self.message = message
         self.author = author
         self.date = date or datetime.datetime.fromtimestamp(0)
         self.repository = repo
         self.alias = alias
 
     @LazyProperty
     def raw_id(self):
         """
         Returns raw string identifying this commit, useful for web
         representation.
         """
 
         return self._empty_commit_id
 
     @LazyProperty
     def branch(self):
         if self.alias:
             from rhodecode.lib.vcs.backends import get_backend
             return get_backend(self.alias).DEFAULT_BRANCH_NAME
 
     @LazyProperty
     def short_id(self):
         return self.raw_id[:12]
 
     @LazyProperty
     def id(self):
         return self.raw_id
 
     def get_file_commit(self, path):
         return self
 
     def get_file_content(self, path):
         return u''
 
     def get_file_size(self, path):
         return 0
 
 
 class EmptyChangesetClass(type):
 
     def __instancecheck__(self, instance):
         return isinstance(instance, EmptyCommit)
 
 
 class EmptyChangeset(EmptyCommit):
 
     __metaclass__ = EmptyChangesetClass
 
     def __new__(cls, *args, **kwargs):
         warnings.warn(
             "Use EmptyCommit instead of EmptyChangeset", DeprecationWarning)
         return super(EmptyCommit, cls).__new__(cls, *args, **kwargs)
 
     def __init__(self, cs='0' * 40, repo=None, requested_revision=None,
                  alias=None, revision=-1, message='', author='', date=None):
         if requested_revision is not None:
             warnings.warn(
                 "Parameter requested_revision not supported anymore",
                 DeprecationWarning)
         super(EmptyChangeset, self).__init__(
             commit_id=cs, repo=repo, alias=alias, idx=revision,
             message=message, author=author, date=date)
 
         @property
         def revision(self):
             warnings.warn("Use idx instead", DeprecationWarning)
             return self.idx
 
         @revision.setter
         def revision(self, value):
             warnings.warn("Use idx instead", DeprecationWarning)
             self.idx = value
 
 
 class EmptyRepository(BaseRepository):
     def __init__(self, repo_path=None, config=None, create=False, **kwargs):
         pass
 
     def get_diff(self, *args, **kwargs):
         from rhodecode.lib.vcs.backends.git.diff import GitDiff
         return GitDiff('')
 
 
 class CollectionGenerator(object):
 
     def __init__(self, repo, commit_ids, collection_size=None, pre_load=None):
         self.repo = repo
         self.commit_ids = commit_ids
         # TODO: (oliver) this isn't currently hooked up
         self.collection_size = None
         self.pre_load = pre_load
 
     def __len__(self):
         if self.collection_size is not None:
             return self.collection_size
         return self.commit_ids.__len__()
 
     def __iter__(self):
         for commit_id in self.commit_ids:
             # TODO: johbo: Mercurial passes in commit indices or commit ids
             yield self._commit_factory(commit_id)
 
     def _commit_factory(self, commit_id):
         """
         Allows backends to override the way commits are generated.
         """
         return self.repo.get_commit(commit_id=commit_id,
                                     pre_load=self.pre_load)
 
     def __getslice__(self, i, j):
         """
         Returns an iterator of sliced repository
         """
         commit_ids = self.commit_ids[i:j]
         return self.__class__(
             self.repo, commit_ids, pre_load=self.pre_load)
 
     def __repr__(self):
         return '<CollectionGenerator[len:%s]>' % (self.__len__())
 
 
 class Config(object):
     """
     Represents the configuration for a repository.
 
     The API is inspired by :class:`ConfigParser.ConfigParser` from the
     standard library. It implements only the needed subset.
     """
 
     def __init__(self):
         self._values = {}
 
     def copy(self):
         clone = Config()
         for section, values in self._values.items():
             clone._values[section] = values.copy()
         return clone
 
     def __repr__(self):
         return '<Config(%s sections) at %s>' % (
             len(self._values), hex(id(self)))
 
     def items(self, section):
         return self._values.get(section, {}).iteritems()
 
     def get(self, section, option):
         return self._values.get(section, {}).get(option)
 
     def set(self, section, option, value):
         section_values = self._values.setdefault(section, {})
         section_values[option] = value
 
     def clear_section(self, section):
         self._values[section] = {}
 
     def serialize(self):
         """
         Creates a list of three tuples (section, key, value) representing
         this config object.
         """
         items = []
         for section in self._values:
             for option, value in self._values[section].items():
                 items.append(
                     (safe_str(section), safe_str(option), safe_str(value)))
         return items
 
 
 class Diff(object):
     """
     Represents a diff result from a repository backend.
 
     Subclasses have to provide a backend specific value for
     :attr:`_header_re` and :attr:`_meta_re`.
     """
     _meta_re = None
     _header_re = None
 
     def __init__(self, raw_diff):
         self.raw = raw_diff
 
     def chunks(self):
         """
         split the diff in chunks of separate --git a/file b/file chunks
         to make diffs consistent we must prepend with \n, and make sure
         we can detect last chunk as this was also has special rule
         """
 
         diff_parts = ('\n' + self.raw).split('\ndiff --git')
         header = diff_parts[0]
 
         if self._meta_re:
             match = self._meta_re.match(header)
 
         chunks = diff_parts[1:]
         total_chunks = len(chunks)
 
         return (
             DiffChunk(chunk, self, cur_chunk == total_chunks)
             for cur_chunk, chunk in enumerate(chunks, start=1))
 
 
 class DiffChunk(object):
 
     def __init__(self, chunk, diff, last_chunk):
         self._diff = diff
 
         # since we split by \ndiff --git that part is lost from original diff
         # we need to re-apply it at the end, EXCEPT ! if it's last chunk
         if not last_chunk:
             chunk += '\n'
 
         match = self._diff._header_re.match(chunk)
         self.header = match.groupdict()
         self.diff = chunk[match.end():]
         self.raw = chunk
 
 
 class BasePathPermissionChecker(object):
 
     @staticmethod
     def create_from_patterns(includes, excludes):
         if includes and '*' in includes and not excludes:
             return AllPathPermissionChecker()
         elif excludes and '*' in excludes:
             return NonePathPermissionChecker()
         else:
             return PatternPathPermissionChecker(includes, excludes)
 
     @property
     def has_full_access(self):
         raise NotImplemented()
 
     def has_access(self, path):
         raise NotImplemented()
 
 
 class AllPathPermissionChecker(BasePathPermissionChecker):
 
     @property
     def has_full_access(self):
         return True
 
     def has_access(self, path):
         return True
 
 
 class NonePathPermissionChecker(BasePathPermissionChecker):
 
     @property
     def has_full_access(self):
         return False
 
     def has_access(self, path):
         return False
 
 
 class PatternPathPermissionChecker(BasePathPermissionChecker):
 
     def __init__(self, includes, excludes):
         self.includes = includes
         self.excludes = excludes
         self.includes_re = [] if not includes else [
             re.compile(fnmatch.translate(pattern)) for pattern in includes]
         self.excludes_re = [] if not excludes else [
             re.compile(fnmatch.translate(pattern)) for pattern in excludes]
 
     @property
     def has_full_access(self):
         return '*' in self.includes and not self.excludes
 
     def has_access(self, path):
         for regex in self.excludes_re:
             if regex.match(path):
                 return False
         for regex in self.includes_re:
             if regex.match(path):
                 return True
         return False

rhodecode/lib/vcs/backends/git/repository.py

@@ -1,989 +1,984 b''
 # -*- coding: utf-8 -*-
 
 # Copyright (C) 2014-2018 RhodeCode GmbH
 #
 # This program is free software: you can redistribute it and/or modify
 # it under the terms of the GNU Affero General Public License, version 3
 # (only), as published by the Free Software Foundation.
 #
 # 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 Affero General Public License
 # along with this program.  If not, see <http://www.gnu.org/licenses/>.
 #
 # This program is dual-licensed. If you wish to learn more about the
 # RhodeCode Enterprise Edition, including its added features, Support services,
 # and proprietary license terms, please see https://rhodecode.com/licenses/
 
 """
 GIT repository module
 """
 
 import logging
 import os
 import re
-import shutil
 
 from zope.cachedescriptors.property import Lazy as LazyProperty
 
 from rhodecode.lib.compat import OrderedDict
 from rhodecode.lib.datelib import (
     utcdate_fromtimestamp, makedate, date_astimestamp)
 from rhodecode.lib.utils import safe_unicode, safe_str
 from rhodecode.lib.vcs import connection, path as vcspath
 from rhodecode.lib.vcs.backends.base import (
     BaseRepository, CollectionGenerator, Config, MergeResponse,
     MergeFailureReason, Reference)
 from rhodecode.lib.vcs.backends.git.commit import GitCommit
 from rhodecode.lib.vcs.backends.git.diff import GitDiff
 from rhodecode.lib.vcs.backends.git.inmemory import GitInMemoryCommit
 from rhodecode.lib.vcs.exceptions import (
     CommitDoesNotExistError, EmptyRepositoryError,
     RepositoryError, TagAlreadyExistError, TagDoesNotExistError, VCSError)
 
 
 SHA_PATTERN = re.compile(r'^[[0-9a-fA-F]{12}|[0-9a-fA-F]{40}]$')
 
 log = logging.getLogger(__name__)
 
 
 class GitRepository(BaseRepository):
     """
     Git repository backend.
     """
     DEFAULT_BRANCH_NAME = 'master'
 
     contact = BaseRepository.DEFAULT_CONTACT
 
     def __init__(self, repo_path, config=None, create=False, src_url=None,
                  update_after_clone=False, with_wire=None, bare=False):
 
         self.path = safe_str(os.path.abspath(repo_path))
         self.config = config if config else self.get_default_config()
         self._remote = connection.Git(
             self.path, self.config, with_wire=with_wire)
 
         self._init_repo(create, src_url, update_after_clone, bare)
 
         # caches
         self._commit_ids = {}
 
     @LazyProperty
     def bare(self):
         return self._remote.bare()
 
     @LazyProperty
     def head(self):
         return self._remote.head()
 
     @LazyProperty
     def commit_ids(self):
         """
         Returns list of commit ids, in ascending order.  Being lazy
         attribute allows external tools to inject commit ids from cache.
         """
         commit_ids = self._get_all_commit_ids()
         self._rebuild_cache(commit_ids)
         return commit_ids
 
     def _rebuild_cache(self, commit_ids):
         self._commit_ids = dict((commit_id, index)
                                 for index, commit_id in enumerate(commit_ids))
 
     def run_git_command(self, cmd, **opts):
         """
         Runs given ``cmd`` as git command and returns tuple
         (stdout, stderr).
 
         :param cmd: git command to be executed
         :param opts: env options to pass into Subprocess command
         """
         if not isinstance(cmd, list):
             raise ValueError('cmd must be a list, got %s instead' % type(cmd))
 
         out, err = self._remote.run_git_command(cmd, **opts)
         if err:
             log.debug('Stderr output of git command "%s":\n%s', cmd, err)
         return out, err
 
     @staticmethod
     def check_url(url, config):
         """
         Function will check given url and try to verify if it's a valid
         link. Sometimes it may happened that git will issue basic
         auth request that can cause whole API to hang when used from python
         or other external calls.
 
         On failures it'll raise urllib2.HTTPError, exception is also thrown
         when the return code is non 200
         """
         # check first if it's not an url
         if os.path.isdir(url) or url.startswith('file:'):
             return True
 
         if '+' in url.split('://', 1)[0]:
             url = url.split('+', 1)[1]
 
         # Request the _remote to verify the url
         return connection.Git.check_url(url, config.serialize())
 
     @staticmethod
     def is_valid_repository(path):
         if os.path.isdir(os.path.join(path, '.git')):
             return True
         # check case of bare repository
         try:
             GitRepository(path)
             return True
         except VCSError:
             pass
         return False
 
     def _init_repo(self, create, src_url=None, update_after_clone=False,
                    bare=False):
         if create and os.path.exists(self.path):
             raise RepositoryError(
                 "Cannot create repository at %s, location already exist"
                 % self.path)
 
         try:
             if create and src_url:
                 GitRepository.check_url(src_url, self.config)
                 self.clone(src_url, update_after_clone, bare)
             elif create:
                 os.makedirs(self.path, mode=0755)
 
                 if bare:
                     self._remote.init_bare()
                 else:
                     self._remote.init()
             else:
                 if not self._remote.assert_correct_path():
                     raise RepositoryError(
                         'Path "%s" does not contain a Git repository' %
                         (self.path,))
 
         # TODO: johbo: check if we have to translate the OSError here
         except OSError as err:
             raise RepositoryError(err)
 
     def _get_all_commit_ids(self, filters=None):
         # we must check if this repo is not empty, since later command
         # fails if it is. And it's cheaper to ask than throw the subprocess
         # errors
         try:
             self._remote.head()
         except KeyError:
             return []
 
         rev_filter = ['--branches', '--tags']
         extra_filter = []
 
         if filters:
             if filters.get('since'):
                 extra_filter.append('--since=%s' % (filters['since']))
             if filters.get('until'):
                 extra_filter.append('--until=%s' % (filters['until']))
             if filters.get('branch_name'):
                 rev_filter = ['--tags']
                 extra_filter.append(filters['branch_name'])
         rev_filter.extend(extra_filter)
 
             # if filters.get('start') or filters.get('end'):
             #     # skip is offset, max-count is limit
             #     if filters.get('start'):
             #         extra_filter += ' --skip=%s' % filters['start']
             #     if filters.get('end'):
             #         extra_filter += ' --max-count=%s' % (filters['end'] - (filters['start'] or 0))
 
         cmd = ['rev-list', '--reverse', '--date-order'] + rev_filter
         try:
             output, __ = self.run_git_command(cmd)
         except RepositoryError:
             # Can be raised for empty repositories
             return []
         return output.splitlines()
 
     def _get_commit_id(self, commit_id_or_idx):
         def is_null(value):
             return len(value) == commit_id_or_idx.count('0')
 
         if self.is_empty():
             raise EmptyRepositoryError("There are no commits yet")
 
         if commit_id_or_idx in (None, '', 'tip', 'HEAD', 'head', -1):
             return self.commit_ids[-1]
 
         is_bstr = isinstance(commit_id_or_idx, (str, unicode))
         if ((is_bstr and commit_id_or_idx.isdigit() and len(commit_id_or_idx) < 12)
             or isinstance(commit_id_or_idx, int) or is_null(commit_id_or_idx)):
             try:
                 commit_id_or_idx = self.commit_ids[int(commit_id_or_idx)]
             except Exception:
                 msg = "Commit %s does not exist for %s" % (
                     commit_id_or_idx, self)
                 raise CommitDoesNotExistError(msg)
 
         elif is_bstr:
             # check full path ref, eg. refs/heads/master
             ref_id = self._refs.get(commit_id_or_idx)
             if ref_id:
                 return ref_id
 
             # check branch name
             branch_ids = self.branches.values()
             ref_id = self._refs.get('refs/heads/%s' % commit_id_or_idx)
             if ref_id:
                 return ref_id
 
             # check tag name
             ref_id = self._refs.get('refs/tags/%s' % commit_id_or_idx)
             if ref_id:
                 return ref_id
 
             if (not SHA_PATTERN.match(commit_id_or_idx) or
                     commit_id_or_idx not in self.commit_ids):
                 msg = "Commit %s does not exist for %s" % (
                     commit_id_or_idx, self)
                 raise CommitDoesNotExistError(msg)
 
         # Ensure we return full id
         if not SHA_PATTERN.match(str(commit_id_or_idx)):
             raise CommitDoesNotExistError(
                 "Given commit id %s not recognized" % commit_id_or_idx)
         return commit_id_or_idx
 
     def get_hook_location(self):
         """
         returns absolute path to location where hooks are stored
         """
         loc = os.path.join(self.path, 'hooks')
         if not self.bare:
             loc = os.path.join(self.path, '.git', 'hooks')
         return loc
 
     @LazyProperty
     def last_change(self):
         """
         Returns last change made on this repository as
         `datetime.datetime` object.
         """
         try:
             return self.get_commit().date
         except RepositoryError:
             tzoffset = makedate()[1]
             return utcdate_fromtimestamp(self._get_fs_mtime(), tzoffset)
 
     def _get_fs_mtime(self):
         idx_loc = '' if self.bare else '.git'
         # fallback to filesystem
         in_path = os.path.join(self.path, idx_loc, "index")
         he_path = os.path.join(self.path, idx_loc, "HEAD")
         if os.path.exists(in_path):
             return os.stat(in_path).st_mtime
         else:
             return os.stat(he_path).st_mtime
 
     @LazyProperty
     def description(self):
         description = self._remote.get_description()
         return safe_unicode(description or self.DEFAULT_DESCRIPTION)
 
     def _get_refs_entries(self, prefix='', reverse=False, strip_prefix=True):
         if self.is_empty():
             return OrderedDict()
 
         result = []
         for ref, sha in self._refs.iteritems():
             if ref.startswith(prefix):
                 ref_name = ref
                 if strip_prefix:
                     ref_name = ref[len(prefix):]
                 result.append((safe_unicode(ref_name), sha))
 
         def get_name(entry):
             return entry[0]
 
         return OrderedDict(sorted(result, key=get_name, reverse=reverse))
 
     def _get_branches(self):
         return self._get_refs_entries(prefix='refs/heads/', strip_prefix=True)
 
     @LazyProperty
     def branches(self):
         return self._get_branches()
 
     @LazyProperty
     def branches_closed(self):
         return {}
 
     @LazyProperty
     def bookmarks(self):
         return {}
 
     @LazyProperty
     def branches_all(self):
         all_branches = {}
         all_branches.update(self.branches)
         all_branches.update(self.branches_closed)
         return all_branches
 
     @LazyProperty
     def tags(self):
         return self._get_tags()
 
     def _get_tags(self):
         return self._get_refs_entries(
             prefix='refs/tags/', strip_prefix=True, reverse=True)
 
     def tag(self, name, user, commit_id=None, message=None, date=None,
             **kwargs):
         # TODO: fix this method to apply annotated tags correct with message
         """
         Creates and returns a tag for the given ``commit_id``.
 
         :param name: name for new tag
         :param user: full username, i.e.: "Joe Doe <[e-mail]>"
         :param commit_id: commit id for which new tag would be created
         :param message: message of the tag's commit
         :param date: date of tag's commit
 
         :raises TagAlreadyExistError: if tag with same name already exists
         """
         if name in self.tags:
             raise TagAlreadyExistError("Tag %s already exists" % name)
         commit = self.get_commit(commit_id=commit_id)
         message = message or "Added tag %s for commit %s" % (
             name, commit.raw_id)
         self._remote.set_refs('refs/tags/%s' % name, commit._commit['id'])
 
         self._refs = self._get_refs()
         self.tags = self._get_tags()
         return commit
 
     def remove_tag(self, name, user, message=None, date=None):
         """
         Removes tag with the given ``name``.
 
         :param name: name of the tag to be removed
         :param user: full username, i.e.: "Joe Doe <[e-mail]>"
         :param message: message of the tag's removal commit
         :param date: date of tag's removal commit
 
         :raises TagDoesNotExistError: if tag with given name does not exists
         """
         if name not in self.tags:
             raise TagDoesNotExistError("Tag %s does not exist" % name)
         tagpath = vcspath.join(
             self._remote.get_refs_path(), 'refs', 'tags', name)
         try:
             os.remove(tagpath)
             self._refs = self._get_refs()
             self.tags = self._get_tags()
         except OSError as e:
             raise RepositoryError(e.strerror)
 
     def _get_refs(self):
         return self._remote.get_refs()
 
     @LazyProperty
     def _refs(self):
         return self._get_refs()
 
     @property
     def _ref_tree(self):
         node = tree = {}
         for ref, sha in self._refs.iteritems():
             path = ref.split('/')
             for bit in path[:-1]:
                 node = node.setdefault(bit, {})
             node[path[-1]] = sha
             node = tree
         return tree
 
     def get_remote_ref(self, ref_name):
         ref_key = 'refs/remotes/origin/{}'.format(safe_str(ref_name))
         try:
             return self._refs[ref_key]
         except Exception:
             return
 
     def get_commit(self, commit_id=None, commit_idx=None, pre_load=None):
         """
         Returns `GitCommit` object representing commit from git repository
         at the given `commit_id` or head (most recent commit) if None given.
         """
         if commit_id is not None:
             self._validate_commit_id(commit_id)
         elif commit_idx is not None:
             self._validate_commit_idx(commit_idx)
             commit_id = commit_idx
         commit_id = self._get_commit_id(commit_id)
         try:
             # Need to call remote to translate id for tagging scenario
             commit_id = self._remote.get_object(commit_id)["commit_id"]
             idx = self._commit_ids[commit_id]
         except KeyError:
             raise RepositoryError("Cannot get object with id %s" % commit_id)
 
         return GitCommit(self, commit_id, idx, pre_load=pre_load)
 
     def get_commits(
             self, start_id=None, end_id=None, start_date=None, end_date=None,
             branch_name=None, show_hidden=False, pre_load=None):
         """
         Returns generator of `GitCommit` objects from start to end (both
         are inclusive), in ascending date order.
 
         :param start_id: None, str(commit_id)
         :param end_id: None, str(commit_id)
         :param start_date: if specified, commits with commit date less than
           ``start_date`` would be filtered out from returned set
         :param end_date: if specified, commits with commit date greater than
           ``end_date`` would be filtered out from returned set
         :param branch_name: if specified, commits not reachable from given
           branch would be filtered out from returned set
         :param show_hidden: Show hidden commits such as obsolete or hidden from
             Mercurial evolve
         :raise BranchDoesNotExistError: If given `branch_name` does not
             exist.
         :raise CommitDoesNotExistError: If commits for given `start` or
           `end` could not be found.
 
         """
         if self.is_empty():
             raise EmptyRepositoryError("There are no commits yet")
         self._validate_branch_name(branch_name)
 
         if start_id is not None:
             self._validate_commit_id(start_id)
         if end_id is not None:
             self._validate_commit_id(end_id)
 
         start_raw_id = self._get_commit_id(start_id)
         start_pos = self._commit_ids[start_raw_id] if start_id else None
         end_raw_id = self._get_commit_id(end_id)
         end_pos = max(0, self._commit_ids[end_raw_id]) if end_id else None
 
         if None not in [start_id, end_id] and start_pos > end_pos:
             raise RepositoryError(
                 "Start commit '%s' cannot be after end commit '%s'" %
                 (start_id, end_id))
 
         if end_pos is not None:
             end_pos += 1
 
         filter_ = []
         if branch_name:
             filter_.append({'branch_name': branch_name})
         if start_date and not end_date:
             filter_.append({'since': start_date})
         if end_date and not start_date:
             filter_.append({'until': end_date})
         if start_date and end_date:
             filter_.append({'since': start_date})
             filter_.append({'until': end_date})
 
         # if start_pos or end_pos:
         #     filter_.append({'start': start_pos})
         #     filter_.append({'end': end_pos})
 
         if filter_:
             revfilters = {
                 'branch_name': branch_name,
                 'since': start_date.strftime('%m/%d/%y %H:%M:%S') if start_date else None,
                 'until': end_date.strftime('%m/%d/%y %H:%M:%S') if end_date else None,
                 'start': start_pos,
                 'end': end_pos,
             }
             commit_ids = self._get_all_commit_ids(filters=revfilters)
 
             # pure python stuff, it's slow due to walker walking whole repo
             # def get_revs(walker):
             #     for walker_entry in walker:
             #         yield walker_entry.commit.id
             # revfilters = {}
             # commit_ids = list(reversed(list(get_revs(self._repo.get_walker(**revfilters)))))
         else:
             commit_ids = self.commit_ids
 
         if start_pos or end_pos:
             commit_ids = commit_ids[start_pos: end_pos]
 
         return CollectionGenerator(self, commit_ids, pre_load=pre_load)
 
     def get_diff(
             self, commit1, commit2, path='', ignore_whitespace=False,
             context=3, path1=None):
         """
         Returns (git like) *diff*, as plain text. Shows changes introduced by
         ``commit2`` since ``commit1``.
 
         :param commit1: Entry point from which diff is shown. Can be
           ``self.EMPTY_COMMIT`` - in this case, patch showing all
           the changes since empty state of the repository until ``commit2``
         :param commit2: Until which commits changes should be shown.
         :param ignore_whitespace: If set to ``True``, would not show whitespace
           changes. Defaults to ``False``.
         :param context: How many lines before/after changed lines should be
           shown. Defaults to ``3``.
         """
         self._validate_diff_commits(commit1, commit2)
         if path1 is not None and path1 != path:
             raise ValueError("Diff of two different paths not supported.")
 
         flags = [
             '-U%s' % context, '--full-index', '--binary', '-p',
             '-M', '--abbrev=40']
         if ignore_whitespace:
             flags.append('-w')
 
         if commit1 == self.EMPTY_COMMIT:
             cmd = ['show'] + flags + [commit2.raw_id]
         else:
             cmd = ['diff'] + flags + [commit1.raw_id, commit2.raw_id]
 
         if path:
             cmd.extend(['--', path])
 
         stdout, __ = self.run_git_command(cmd)
         # If we used 'show' command, strip first few lines (until actual diff
         # starts)
         if commit1 == self.EMPTY_COMMIT:
             lines = stdout.splitlines()
             x = 0
             for line in lines:
                 if line.startswith('diff'):
                     break
                 x += 1
             # Append new line just like 'diff' command do
             stdout = '\n'.join(lines[x:]) + '\n'
         return GitDiff(stdout)
 
     def strip(self, commit_id, branch_name):
         commit = self.get_commit(commit_id=commit_id)
         if commit.merge:
             raise Exception('Cannot reset to merge commit')
 
         # parent is going to be the new head now
         commit = commit.parents[0]
         self._remote.set_refs('refs/heads/%s' % branch_name, commit.raw_id)
 
         self.commit_ids = self._get_all_commit_ids()
         self._rebuild_cache(self.commit_ids)
 
     def get_common_ancestor(self, commit_id1, commit_id2, repo2):
         if commit_id1 == commit_id2:
             return commit_id1
 
         if self != repo2:
             commits = self._remote.get_missing_revs(
                 commit_id1, commit_id2, repo2.path)
             if commits:
                 commit = repo2.get_commit(commits[-1])
                 if commit.parents:
                     ancestor_id = commit.parents[0].raw_id
                 else:
                     ancestor_id = None
             else:
                 # no commits from other repo, ancestor_id is the commit_id2
                 ancestor_id = commit_id2
         else:
             output, __ = self.run_git_command(
                 ['merge-base', commit_id1, commit_id2])
             ancestor_id = re.findall(r'[0-9a-fA-F]{40}', output)[0]
 
         return ancestor_id
 
     def compare(self, commit_id1, commit_id2, repo2, merge, pre_load=None):
         repo1 = self
         ancestor_id = None
 
         if commit_id1 == commit_id2:
             commits = []
         elif repo1 != repo2:
             missing_ids = self._remote.get_missing_revs(commit_id1, commit_id2,
                                                         repo2.path)
             commits = [
                 repo2.get_commit(commit_id=commit_id, pre_load=pre_load)
                 for commit_id in reversed(missing_ids)]
         else:
             output, __ = repo1.run_git_command(
                 ['log', '--reverse', '--pretty=format: %H', '-s',
                  '%s..%s' % (commit_id1, commit_id2)])
             commits = [
                 repo1.get_commit(commit_id=commit_id, pre_load=pre_load)
                 for commit_id in re.findall(r'[0-9a-fA-F]{40}', output)]
 
         return commits
 
     @LazyProperty
     def in_memory_commit(self):
         """
         Returns ``GitInMemoryCommit`` object for this repository.
         """
         return GitInMemoryCommit(self)
 
     def clone(self, url, update_after_clone=True, bare=False):
         """
         Tries to clone commits from external location.
 
         :param update_after_clone: If set to ``False``, git won't checkout
           working directory
         :param bare: If set to ``True``, repository would be cloned into
           *bare* git repository (no working directory at all).
         """
         # init_bare and init expect empty dir created to proceed
         if not os.path.exists(self.path):
             os.mkdir(self.path)
 
         if bare:
             self._remote.init_bare()
         else:
             self._remote.init()
 
         deferred = '^{}'
         valid_refs = ('refs/heads', 'refs/tags', 'HEAD')
 
         return self._remote.clone(
             url, deferred, valid_refs, update_after_clone)
 
     def pull(self, url, commit_ids=None):
         """
         Tries to pull changes from external location. We use fetch here since
         pull in get does merges and we want to be compatible with hg backend so
         pull == fetch in this case
         """
         self.fetch(url, commit_ids=commit_ids)
 
     def fetch(self, url, commit_ids=None):
         """
         Tries to fetch changes from external location.
         """
         refs = None
 
         if commit_ids is not None:
             remote_refs = self._remote.get_remote_refs(url)
             refs = [
                 ref for ref in remote_refs if remote_refs[ref] in commit_ids]
         self._remote.fetch(url, refs=refs)
 
     def push(self, url):
         refs = None
         self._remote.sync_push(url, refs=refs)
 
     def set_refs(self, ref_name, commit_id):
         self._remote.set_refs(ref_name, commit_id)
 
     def remove_ref(self, ref_name):
         self._remote.remove_ref(ref_name)
 
     def _update_server_info(self):
         """
         runs gits update-server-info command in this repo instance
         """
         self._remote.update_server_info()
 
     def _current_branch(self):
         """
         Return the name of the current branch.
 
         It only works for non bare repositories (i.e. repositories with a
         working copy)
         """
         if self.bare:
             raise RepositoryError('Bare git repos do not have active branches')
 
         if self.is_empty():
             return None
 
         stdout, _ = self.run_git_command(['rev-parse', '--abbrev-ref', 'HEAD'])
         return stdout.strip()
 
     def _checkout(self, branch_name, create=False, force=False):
         """
         Checkout a branch in the working directory.
 
         It tries to create the branch if create is True, failing if the branch
         already exists.
 
         It only works for non bare repositories (i.e. repositories with a
         working copy)
         """
         if self.bare:
             raise RepositoryError('Cannot checkout branches in a bare git repo')
 
         cmd = ['checkout']
         if force:
             cmd.append('-f')
         if create:
             cmd.append('-b')
         cmd.append(branch_name)
         self.run_git_command(cmd, fail_on_stderr=False)
 
     def _identify(self):
         """
         Return the current state of the working directory.
         """
         if self.bare:
             raise RepositoryError('Bare git repos do not have active branches')
 
         if self.is_empty():
             return None
 
         stdout, _ = self.run_git_command(['rev-parse', 'HEAD'])
         return stdout.strip()
 
     def _local_clone(self, clone_path, branch_name, source_branch=None):
         """
         Create a local clone of the current repo.
         """
         # N.B.(skreft): the --branch option is required as otherwise the shallow
         # clone will only fetch the active branch.
         cmd = ['clone', '--branch', branch_name,
                self.path, os.path.abspath(clone_path)]
 
         self.run_git_command(cmd, fail_on_stderr=False)
 
         # if we get the different source branch, make sure we also fetch it for
         # merge conditions
         if source_branch and source_branch != branch_name:
             # check if the ref exists.
             shadow_repo = GitRepository(os.path.abspath(clone_path))
             if shadow_repo.get_remote_ref(source_branch):
                 cmd = ['fetch', self.path, source_branch]
                 self.run_git_command(cmd, fail_on_stderr=False)
 
     def _local_fetch(self, repository_path, branch_name):
         """
         Fetch a branch from a local repository.
         """
         repository_path = os.path.abspath(repository_path)
         if repository_path == self.path:
             raise ValueError('Cannot fetch from the same repository')
 
         cmd = ['fetch', '--no-tags', repository_path, branch_name]
         self.run_git_command(cmd, fail_on_stderr=False)
 
     def _last_fetch_heads(self):
         """
         Return the last fetched heads that need merging.
 
         The algorithm is defined at
         https://github.com/git/git/blob/v2.1.3/git-pull.sh#L283
         """
         if not self.bare:
             fetch_heads_path = os.path.join(self.path, '.git', 'FETCH_HEAD')
         else:
             fetch_heads_path = os.path.join(self.path, 'FETCH_HEAD')
 
         heads = []
         with open(fetch_heads_path) as f:
             for line in f:
                 if '    not-for-merge   ' in line:
                     continue
                 line = re.sub('\t.*', '', line, flags=re.DOTALL)
                 heads.append(line)
 
         return heads
 
     def _get_shadow_instance(self, shadow_repository_path, enable_hooks=False):
         return GitRepository(shadow_repository_path)
 
     def _local_pull(self, repository_path, branch_name, ff_only=True):
         """
         Pull a branch from a local repository.
         """
         if self.bare:
             raise RepositoryError('Cannot pull into a bare git repository')
         # N.B.(skreft): The --ff-only option is to make sure this is a
         # fast-forward (i.e., we are only pulling new changes and there are no
         # conflicts with our current branch)
         # Additionally, that option needs to go before --no-tags, otherwise git
         # pull complains about it being an unknown flag.
         cmd = ['pull']
         if ff_only:
             cmd.append('--ff-only')
         cmd.extend(['--no-tags', repository_path, branch_name])
         self.run_git_command(cmd, fail_on_stderr=False)
 
     def _local_merge(self, merge_message, user_name, user_email, heads):
         """
         Merge the given head into the checked out branch.
 
         It will force a merge commit.
 
         Currently it raises an error if the repo is empty, as it is not possible
         to create a merge commit in an empty repo.
 
         :param merge_message: The message to use for the merge commit.
         :param heads: the heads to merge.
         """
         if self.bare:
             raise RepositoryError('Cannot merge into a bare git repository')
 
         if not heads:
             return
 
         if self.is_empty():
             # TODO(skreft): do somehting more robust in this case.
             raise RepositoryError(
                 'Do not know how to merge into empty repositories yet')
 
         # N.B.(skreft): the --no-ff option is used to enforce the creation of a
         # commit message. We also specify the user who is doing the merge.
         cmd = ['-c', 'user.name=%s' % safe_str(user_name),
                '-c', 'user.email=%s' % safe_str(user_email),
                'merge', '--no-ff', '-m', safe_str(merge_message)]
         cmd.extend(heads)
         try:
             self.run_git_command(cmd, fail_on_stderr=False)
         except RepositoryError:
             # Cleanup any merge leftovers
             self.run_git_command(['merge', '--abort'], fail_on_stderr=False)
             raise
 
     def _local_push(
             self, source_branch, repository_path, target_branch,
             enable_hooks=False, rc_scm_data=None):
         """
         Push the source_branch to the given repository and target_branch.
 
         Currently it if the target_branch is not master and the target repo is
         empty, the push will work, but then GitRepository won't be able to find
         the pushed branch or the commits. As the HEAD will be corrupted (i.e.,
         pointing to master, which does not exist).
 
         It does not run the hooks in the target repo.
         """
         # TODO(skreft): deal with the case in which the target repo is empty,
         # and the target_branch is not master.
         target_repo = GitRepository(repository_path)
         if (not target_repo.bare and
                 target_repo._current_branch() == target_branch):
             # Git prevents pushing to the checked out branch, so simulate it by
             # pulling into the target repository.
             target_repo._local_pull(self.path, source_branch)
         else:
             cmd = ['push', os.path.abspath(repository_path),
                    '%s:%s' % (source_branch, target_branch)]
             gitenv = {}
             if rc_scm_data:
                 gitenv.update({'RC_SCM_DATA': rc_scm_data})
 
             if not enable_hooks:
                 gitenv['RC_SKIP_HOOKS'] = '1'
             self.run_git_command(cmd, fail_on_stderr=False, extra_env=gitenv)
 
     def _get_new_pr_branch(self, source_branch, target_branch):
         prefix = 'pr_%s-%s_' % (source_branch, target_branch)
         pr_branches = []
         for branch in self.branches:
             if branch.startswith(prefix):
                 pr_branches.append(int(branch[len(prefix):]))
 
         if not pr_branches:
             branch_id = 0
         else:
             branch_id = max(pr_branches) + 1
 
         return '%s%d' % (prefix, branch_id)
 
     def _merge_repo(self, shadow_repository_path, target_ref,
                     source_repo, source_ref, merge_message,
                     merger_name, merger_email, dry_run=False,
                     use_rebase=False, close_branch=False):
         if target_ref.commit_id != self.branches[target_ref.name]:
             return MergeResponse(
                 False, False, None, MergeFailureReason.TARGET_IS_NOT_HEAD)
 
         shadow_repo = GitRepository(shadow_repository_path)
         # checkout source, if it's different. Otherwise we could not
         # fetch proper commits for merge testing
         if source_ref.name != target_ref.name:
             if shadow_repo.get_remote_ref(source_ref.name):
                 shadow_repo._checkout(source_ref.name, force=True)
 
         # checkout target
         shadow_repo._checkout(target_ref.name, force=True)
         shadow_repo._local_pull(self.path, target_ref.name)
 
         # Need to reload repo to invalidate the cache, or otherwise we cannot
         # retrieve the last target commit.
         shadow_repo = GitRepository(shadow_repository_path)
         if target_ref.commit_id != shadow_repo.branches[target_ref.name]:
             return MergeResponse(
                 False, False, None, MergeFailureReason.TARGET_IS_NOT_HEAD)
 
         pr_branch = shadow_repo._get_new_pr_branch(
             source_ref.name, target_ref.name)
         log.debug('using pull-request merge branch: `%s`', pr_branch)
         shadow_repo._checkout(pr_branch, create=True)
         try:
             shadow_repo._local_fetch(source_repo.path, source_ref.name)
         except RepositoryError:
             log.exception('Failure when doing local fetch on git shadow repo')
             return MergeResponse(
                 False, False, None, MergeFailureReason.MISSING_SOURCE_REF)
 
         merge_ref = None
         merge_failure_reason = MergeFailureReason.NONE
         try:
             shadow_repo._local_merge(merge_message, merger_name, merger_email,
                                      [source_ref.commit_id])
             merge_possible = True
 
             # Need to reload repo to invalidate the cache, or otherwise we
             # cannot retrieve the merge commit.
             shadow_repo = GitRepository(shadow_repository_path)
             merge_commit_id = shadow_repo.branches[pr_branch]
 
             # Set a reference pointing to the merge commit. This reference may
             # be used to easily identify the last successful merge commit in
             # the shadow repository.
             shadow_repo.set_refs('refs/heads/pr-merge', merge_commit_id)
             merge_ref = Reference('branch', 'pr-merge', merge_commit_id)
         except RepositoryError:
             log.exception('Failure when doing local merge on git shadow repo')
             merge_possible = False
             merge_failure_reason = MergeFailureReason.MERGE_FAILED
 
         if merge_possible and not dry_run:
             try:
                 shadow_repo._local_push(
                     pr_branch, self.path, target_ref.name, enable_hooks=True,
                     rc_scm_data=self.config.get('rhodecode', 'RC_SCM_DATA'))
                 merge_succeeded = True
             except RepositoryError:
                 log.exception(
                     'Failure when doing local push on git shadow repo')
                 merge_succeeded = False
                 merge_failure_reason = MergeFailureReason.PUSH_FAILED
         else:
             merge_succeeded = False
 
         return MergeResponse(
             merge_possible, merge_succeeded, merge_ref,
             merge_failure_reason)
 
     def _get_shadow_repository_path(self, workspace_id):
         # The name of the shadow repository must start with '.', so it is
         # skipped by 'rhodecode.lib.utils.get_filesystem_repos'.
         return os.path.join(
             os.path.dirname(self.path),
             '.__shadow_%s_%s' % (os.path.basename(self.path), workspace_id))
 
     def _maybe_prepare_merge_workspace(self, workspace_id, target_ref, source_ref):
         shadow_repository_path = self._get_shadow_repository_path(workspace_id)
         if not os.path.exists(shadow_repository_path):
             self._local_clone(
                 shadow_repository_path, target_ref.name, source_ref.name)
 
         return shadow_repository_path
-
-    def cleanup_merge_workspace(self, workspace_id):
-        shadow_repository_path = self._get_shadow_repository_path(workspace_id)
-        shutil.rmtree(shadow_repository_path, ignore_errors=True)

rhodecode/lib/vcs/backends/hg/repository.py

@@ -1,936 +1,931 b''
 # -*- coding: utf-8 -*-
 
 # Copyright (C) 2014-2018 RhodeCode GmbH
 #
 # This program is free software: you can redistribute it and/or modify
 # it under the terms of the GNU Affero General Public License, version 3
 # (only), as published by the Free Software Foundation.
 #
 # 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 Affero General Public License
 # along with this program.  If not, see <http://www.gnu.org/licenses/>.
 #
 # This program is dual-licensed. If you wish to learn more about the
 # RhodeCode Enterprise Edition, including its added features, Support services,
 # and proprietary license terms, please see https://rhodecode.com/licenses/
 
 """
 HG repository module
 """
 import os
 import logging
 import binascii
-import shutil
 import urllib
 
 from zope.cachedescriptors.property import Lazy as LazyProperty
 
 from rhodecode.lib.compat import OrderedDict
 from rhodecode.lib.datelib import (
     date_to_timestamp_plus_offset, utcdate_fromtimestamp, makedate)
 from rhodecode.lib.utils import safe_unicode, safe_str
 from rhodecode.lib.vcs import connection, exceptions
 from rhodecode.lib.vcs.backends.base import (
     BaseRepository, CollectionGenerator, Config, MergeResponse,
     MergeFailureReason, Reference, BasePathPermissionChecker)
 from rhodecode.lib.vcs.backends.hg.commit import MercurialCommit
 from rhodecode.lib.vcs.backends.hg.diff import MercurialDiff
 from rhodecode.lib.vcs.backends.hg.inmemory import MercurialInMemoryCommit
 from rhodecode.lib.vcs.exceptions import (
     EmptyRepositoryError, RepositoryError, TagAlreadyExistError,
     TagDoesNotExistError, CommitDoesNotExistError, SubrepoMergeError)
 from rhodecode.lib.vcs.compat import configparser
 
 hexlify = binascii.hexlify
 nullid = "\0" * 20
 
 log = logging.getLogger(__name__)
 
 
 class MercurialRepository(BaseRepository):
     """
     Mercurial repository backend
     """
     DEFAULT_BRANCH_NAME = 'default'
 
     def __init__(self, repo_path, config=None, create=False, src_url=None,
                  update_after_clone=False, with_wire=None):
         """
         Raises RepositoryError if repository could not be find at the given
         ``repo_path``.
 
         :param repo_path: local path of the repository
         :param config: config object containing the repo configuration
         :param create=False: if set to True, would try to create repository if
            it does not exist rather than raising exception
         :param src_url=None: would try to clone repository from given location
         :param update_after_clone=False: sets update of working copy after
            making a clone
         """
 
         self.path = safe_str(os.path.abspath(repo_path))
         # mercurial since 4.4.X requires certain configuration to be present
         # because sometimes we init the repos with config we need to meet
         # special requirements
         self.config = config if config else self.get_default_config(
             default=[('extensions', 'largefiles', '1')])
 
         self._remote = connection.Hg(
             self.path, self.config, with_wire=with_wire)
 
         self._init_repo(create, src_url, update_after_clone)
 
         # caches
         self._commit_ids = {}
 
     @LazyProperty
     def commit_ids(self):
         """
         Returns list of commit ids, in ascending order.  Being lazy
         attribute allows external tools to inject shas from cache.
         """
         commit_ids = self._get_all_commit_ids()
         self._rebuild_cache(commit_ids)
         return commit_ids
 
     def _rebuild_cache(self, commit_ids):
         self._commit_ids = dict((commit_id, index)
                                 for index, commit_id in enumerate(commit_ids))
 
     @LazyProperty
     def branches(self):
         return self._get_branches()
 
     @LazyProperty
     def branches_closed(self):
         return self._get_branches(active=False, closed=True)
 
     @LazyProperty
     def branches_all(self):
         all_branches = {}
         all_branches.update(self.branches)
         all_branches.update(self.branches_closed)
         return all_branches
 
     def _get_branches(self, active=True, closed=False):
         """
         Gets branches for this repository
         Returns only not closed active branches by default
 
         :param active: return also active branches
         :param closed: return also closed branches
 
         """
         if self.is_empty():
             return {}
 
         def get_name(ctx):
             return ctx[0]
 
         _branches = [(safe_unicode(n), hexlify(h),) for n, h in
                      self._remote.branches(active, closed).items()]
 
         return OrderedDict(sorted(_branches, key=get_name, reverse=False))
 
     @LazyProperty
     def tags(self):
         """
         Gets tags for this repository
         """
         return self._get_tags()
 
     def _get_tags(self):
         if self.is_empty():
             return {}
 
         def get_name(ctx):
             return ctx[0]
 
         _tags = [(safe_unicode(n), hexlify(h),) for n, h in
                  self._remote.tags().items()]
 
         return OrderedDict(sorted(_tags, key=get_name, reverse=True))
 
     def tag(self, name, user, commit_id=None, message=None, date=None,
             **kwargs):
         """
         Creates and returns a tag for the given ``commit_id``.
 
         :param name: name for new tag
         :param user: full username, i.e.: "Joe Doe <[e-mail]>"
         :param commit_id: commit id for which new tag would be created
         :param message: message of the tag's commit
         :param date: date of tag's commit
 
         :raises TagAlreadyExistError: if tag with same name already exists
         """
         if name in self.tags:
             raise TagAlreadyExistError("Tag %s already exists" % name)
         commit = self.get_commit(commit_id=commit_id)
         local = kwargs.setdefault('local', False)
 
         if message is None:
             message = "Added tag %s for commit %s" % (name, commit.short_id)
 
         date, tz = date_to_timestamp_plus_offset(date)
 
         self._remote.tag(
             name, commit.raw_id, message, local, user, date, tz)
         self._remote.invalidate_vcs_cache()
 
         # Reinitialize tags
         self.tags = self._get_tags()
         tag_id = self.tags[name]
 
         return self.get_commit(commit_id=tag_id)
 
     def remove_tag(self, name, user, message=None, date=None):
         """
         Removes tag with the given `name`.
 
         :param name: name of the tag to be removed
         :param user: full username, i.e.: "Joe Doe <[e-mail]>"
         :param message: message of the tag's removal commit
         :param date: date of tag's removal commit
 
         :raises TagDoesNotExistError: if tag with given name does not exists
         """
         if name not in self.tags:
             raise TagDoesNotExistError("Tag %s does not exist" % name)
         if message is None:
             message = "Removed tag %s" % name
         local = False
 
         date, tz = date_to_timestamp_plus_offset(date)
 
         self._remote.tag(name, nullid, message, local, user, date, tz)
         self._remote.invalidate_vcs_cache()
         self.tags = self._get_tags()
 
     @LazyProperty
     def bookmarks(self):
         """
         Gets bookmarks for this repository
         """
         return self._get_bookmarks()
 
     def _get_bookmarks(self):
         if self.is_empty():
             return {}
 
         def get_name(ctx):
             return ctx[0]
 
         _bookmarks = [
             (safe_unicode(n), hexlify(h)) for n, h in
             self._remote.bookmarks().items()]
 
         return OrderedDict(sorted(_bookmarks, key=get_name))
 
     def _get_all_commit_ids(self):
         return self._remote.get_all_commit_ids('visible')
 
     def get_diff(
             self, commit1, commit2, path='', ignore_whitespace=False,
             context=3, path1=None):
         """
         Returns (git like) *diff*, as plain text. Shows changes introduced by
         `commit2` since `commit1`.
 
         :param commit1: Entry point from which diff is shown. Can be
           ``self.EMPTY_COMMIT`` - in this case, patch showing all
           the changes since empty state of the repository until `commit2`
         :param commit2: Until which commit changes should be shown.
         :param ignore_whitespace: If set to ``True``, would not show whitespace
           changes. Defaults to ``False``.
         :param context: How many lines before/after changed lines should be
           shown. Defaults to ``3``.
         """
         self._validate_diff_commits(commit1, commit2)
         if path1 is not None and path1 != path:
             raise ValueError("Diff of two different paths not supported.")
 
         if path:
             file_filter = [self.path, path]
         else:
             file_filter = None
 
         diff = self._remote.diff(
             commit1.raw_id, commit2.raw_id, file_filter=file_filter,
             opt_git=True, opt_ignorews=ignore_whitespace,
             context=context)
         return MercurialDiff(diff)
 
     def strip(self, commit_id, branch=None):
         self._remote.strip(commit_id, update=False, backup="none")
 
         self._remote.invalidate_vcs_cache()
         self.commit_ids = self._get_all_commit_ids()
         self._rebuild_cache(self.commit_ids)
 
     def verify(self):
         verify = self._remote.verify()
 
         self._remote.invalidate_vcs_cache()
         return verify
 
     def get_common_ancestor(self, commit_id1, commit_id2, repo2):
         if commit_id1 == commit_id2:
             return commit_id1
 
         ancestors = self._remote.revs_from_revspec(
             "ancestor(id(%s), id(%s))", commit_id1, commit_id2,
             other_path=repo2.path)
         return repo2[ancestors[0]].raw_id if ancestors else None
 
     def compare(self, commit_id1, commit_id2, repo2, merge, pre_load=None):
         if commit_id1 == commit_id2:
             commits = []
         else:
             if merge:
                 indexes = self._remote.revs_from_revspec(
                     "ancestors(id(%s)) - ancestors(id(%s)) - id(%s)",
                     commit_id2, commit_id1, commit_id1, other_path=repo2.path)
             else:
                 indexes = self._remote.revs_from_revspec(
                     "id(%s)..id(%s) - id(%s)", commit_id1, commit_id2,
                     commit_id1, other_path=repo2.path)
 
             commits = [repo2.get_commit(commit_idx=idx, pre_load=pre_load)
                        for idx in indexes]
 
         return commits
 
     @staticmethod
     def check_url(url, config):
         """
         Function will check given url and try to verify if it's a valid
         link. Sometimes it may happened that mercurial will issue basic
         auth request that can cause whole API to hang when used from python
         or other external calls.
 
         On failures it'll raise urllib2.HTTPError, exception is also thrown
         when the return code is non 200
         """
         # check first if it's not an local url
         if os.path.isdir(url) or url.startswith('file:'):
             return True
 
         # Request the _remote to verify the url
         return connection.Hg.check_url(url, config.serialize())
 
     @staticmethod
     def is_valid_repository(path):
         return os.path.isdir(os.path.join(path, '.hg'))
 
     def _init_repo(self, create, src_url=None, update_after_clone=False):
         """
         Function will check for mercurial repository in given path. If there
         is no repository in that path it will raise an exception unless
         `create` parameter is set to True - in that case repository would
         be created.
 
         If `src_url` is given, would try to clone repository from the
         location at given clone_point. Additionally it'll make update to
         working copy accordingly to `update_after_clone` flag.
         """
         if create and os.path.exists(self.path):
             raise RepositoryError(
                 "Cannot create repository at %s, location already exist"
                 % self.path)
 
         if src_url:
             url = str(self._get_url(src_url))
             MercurialRepository.check_url(url, self.config)
 
             self._remote.clone(url, self.path, update_after_clone)
 
             # Don't try to create if we've already cloned repo
             create = False
 
         if create:
             os.makedirs(self.path, mode=0755)
 
         self._remote.localrepository(create)
 
     @LazyProperty
     def in_memory_commit(self):
         return MercurialInMemoryCommit(self)
 
     @LazyProperty
     def description(self):
         description = self._remote.get_config_value(
             'web', 'description', untrusted=True)
         return safe_unicode(description or self.DEFAULT_DESCRIPTION)
 
     @LazyProperty
     def contact(self):
         contact = (
             self._remote.get_config_value("web", "contact") or
             self._remote.get_config_value("ui", "username"))
         return safe_unicode(contact or self.DEFAULT_CONTACT)
 
     @LazyProperty
     def last_change(self):
         """
         Returns last change made on this repository as
         `datetime.datetime` object.
         """
         try:
             return self.get_commit().date
         except RepositoryError:
             tzoffset = makedate()[1]
             return utcdate_fromtimestamp(self._get_fs_mtime(), tzoffset)
 
     def _get_fs_mtime(self):
         # fallback to filesystem
         cl_path = os.path.join(self.path, '.hg', "00changelog.i")
         st_path = os.path.join(self.path, '.hg', "store")
         if os.path.exists(cl_path):
             return os.stat(cl_path).st_mtime
         else:
             return os.stat(st_path).st_mtime
 
     def _sanitize_commit_idx(self, idx):
         # Note: Mercurial has ``int(-1)`` reserved as not existing id_or_idx
         # number. A `long` is treated in the correct way though. So we convert
         # `int` to `long` here to make sure it is handled correctly.
         if isinstance(idx, int):
             return long(idx)
         return idx
 
     def _get_url(self, url):
         """
         Returns normalized url. If schema is not given, would fall
         to filesystem
         (``file:///``) schema.
         """
         url = url.encode('utf8')
         if url != 'default' and '://' not in url:
             url = "file:" + urllib.pathname2url(url)
         return url
 
     def get_hook_location(self):
         """
         returns absolute path to location where hooks are stored
         """
         return os.path.join(self.path, '.hg', '.hgrc')
 
     def get_commit(self, commit_id=None, commit_idx=None, pre_load=None):
         """
         Returns ``MercurialCommit`` object representing repository's
         commit at the given `commit_id` or `commit_idx`.
         """
         if self.is_empty():
             raise EmptyRepositoryError("There are no commits yet")
 
         if commit_id is not None:
             self._validate_commit_id(commit_id)
             try:
                 idx = self._commit_ids[commit_id]
                 return MercurialCommit(self, commit_id, idx, pre_load=pre_load)
             except KeyError:
                 pass
         elif commit_idx is not None:
             self._validate_commit_idx(commit_idx)
             commit_idx = self._sanitize_commit_idx(commit_idx)
             try:
                 id_ = self.commit_ids[commit_idx]
                 if commit_idx < 0:
                     commit_idx += len(self.commit_ids)
                 return MercurialCommit(
                     self, id_, commit_idx, pre_load=pre_load)
             except IndexError:
                 commit_id = commit_idx
         else:
             commit_id = "tip"
 
         # TODO Paris: Ugly hack to "serialize" long for msgpack
         if isinstance(commit_id, long):
             commit_id = float(commit_id)
 
         if isinstance(commit_id, unicode):
             commit_id = safe_str(commit_id)
 
         try:
             raw_id, idx = self._remote.lookup(commit_id, both=True)
         except CommitDoesNotExistError:
             msg = "Commit %s does not exist for %s" % (
                 commit_id, self)
             raise CommitDoesNotExistError(msg)
 
         return MercurialCommit(self, raw_id, idx, pre_load=pre_load)
 
     def get_commits(
             self, start_id=None, end_id=None, start_date=None, end_date=None,
             branch_name=None, show_hidden=False, pre_load=None):
         """
         Returns generator of ``MercurialCommit`` objects from start to end
         (both are inclusive)
 
         :param start_id: None, str(commit_id)
         :param end_id: None, str(commit_id)
         :param start_date: if specified, commits with commit date less than
           ``start_date`` would be filtered out from returned set
         :param end_date: if specified, commits with commit date greater than
           ``end_date`` would be filtered out from returned set
         :param branch_name: if specified, commits not reachable from given
           branch would be filtered out from returned set
         :param show_hidden: Show hidden commits such as obsolete or hidden from
             Mercurial evolve
         :raise BranchDoesNotExistError: If given ``branch_name`` does not
             exist.
         :raise CommitDoesNotExistError: If commit for given ``start`` or
           ``end`` could not be found.
         """
         # actually we should check now if it's not an empty repo
         branch_ancestors = False
         if self.is_empty():
             raise EmptyRepositoryError("There are no commits yet")
         self._validate_branch_name(branch_name)
 
         if start_id is not None:
             self._validate_commit_id(start_id)
             c_start = self.get_commit(commit_id=start_id)
             start_pos = self._commit_ids[c_start.raw_id]
         else:
             start_pos = None
 
         if end_id is not None:
             self._validate_commit_id(end_id)
             c_end = self.get_commit(commit_id=end_id)
             end_pos = max(0, self._commit_ids[c_end.raw_id])
         else:
             end_pos = None
 
         if None not in [start_id, end_id] and start_pos > end_pos:
             raise RepositoryError(
                 "Start commit '%s' cannot be after end commit '%s'" %
                 (start_id, end_id))
 
         if end_pos is not None:
             end_pos += 1
 
         commit_filter = []
 
         if branch_name and not branch_ancestors:
             commit_filter.append('branch("%s")' % (branch_name,))
         elif branch_name and branch_ancestors:
             commit_filter.append('ancestors(branch("%s"))' % (branch_name,))
 
         if start_date and not end_date:
             commit_filter.append('date(">%s")' % (start_date,))
         if end_date and not start_date:
             commit_filter.append('date("<%s")' % (end_date,))
         if start_date and end_date:
             commit_filter.append(
                 'date(">%s") and date("<%s")' % (start_date, end_date))
 
         if not show_hidden:
             commit_filter.append('not obsolete()')
             commit_filter.append('not hidden()')
 
         # TODO: johbo: Figure out a simpler way for this solution
         collection_generator = CollectionGenerator
         if commit_filter:
             commit_filter = ' and '.join(map(safe_str, commit_filter))
             revisions = self._remote.rev_range([commit_filter])
             collection_generator = MercurialIndexBasedCollectionGenerator
         else:
             revisions = self.commit_ids
 
         if start_pos or end_pos:
             revisions = revisions[start_pos:end_pos]
 
         return collection_generator(self, revisions, pre_load=pre_load)
 
     def pull(self, url, commit_ids=None):
         """
         Tries to pull changes from external location.
 
         :param commit_ids: Optional. Can be set to a list of commit ids
            which shall be pulled from the other repository.
         """
         url = self._get_url(url)
         self._remote.pull(url, commit_ids=commit_ids)
         self._remote.invalidate_vcs_cache()
 
     def push(self, url):
         url = self._get_url(url)
         self._remote.sync_push(url)
 
     def _local_clone(self, clone_path):
         """
         Create a local clone of the current repo.
         """
         self._remote.clone(self.path, clone_path, update_after_clone=True,
                            hooks=False)
 
     def _update(self, revision, clean=False):
         """
         Update the working copy to the specified revision.
         """
         log.debug('Doing checkout to commit: `%s` for %s', revision, self)
         self._remote.update(revision, clean=clean)
 
     def _identify(self):
         """
         Return the current state of the working directory.
         """
         return self._remote.identify().strip().rstrip('+')
 
     def _heads(self, branch=None):
         """
         Return the commit ids of the repository heads.
         """
         return self._remote.heads(branch=branch).strip().split(' ')
 
     def _ancestor(self, revision1, revision2):
         """
         Return the common ancestor of the two revisions.
         """
         return self._remote.ancestor(revision1, revision2)
 
     def _local_push(
             self, revision, repository_path, push_branches=False,
             enable_hooks=False):
         """
         Push the given revision to the specified repository.
 
         :param push_branches: allow to create branches in the target repo.
         """
         self._remote.push(
             [revision], repository_path, hooks=enable_hooks,
             push_branches=push_branches)
 
     def _local_merge(self, target_ref, merge_message, user_name, user_email,
                      source_ref, use_rebase=False, dry_run=False):
         """
         Merge the given source_revision into the checked out revision.
 
         Returns the commit id of the merge and a boolean indicating if the
         commit needs to be pushed.
         """
         self._update(target_ref.commit_id)
 
         ancestor = self._ancestor(target_ref.commit_id, source_ref.commit_id)
         is_the_same_branch = self._is_the_same_branch(target_ref, source_ref)
 
         if ancestor == source_ref.commit_id:
             # Nothing to do, the changes were already integrated
             return target_ref.commit_id, False
 
         elif ancestor == target_ref.commit_id and is_the_same_branch:
             # In this case we should force a commit message
             return source_ref.commit_id, True
 
         if use_rebase:
             try:
                 bookmark_name = 'rcbook%s%s' % (source_ref.commit_id,
                                                 target_ref.commit_id)
                 self.bookmark(bookmark_name, revision=source_ref.commit_id)
                 self._remote.rebase(
                     source=source_ref.commit_id, dest=target_ref.commit_id)
                 self._remote.invalidate_vcs_cache()
                 self._update(bookmark_name)
                 return self._identify(), True
             except RepositoryError:
                 # The rebase-abort may raise another exception which 'hides'
                 # the original one, therefore we log it here.
                 log.exception('Error while rebasing shadow repo during merge.')
 
                 # Cleanup any rebase leftovers
                 self._remote.invalidate_vcs_cache()
                 self._remote.rebase(abort=True)
                 self._remote.invalidate_vcs_cache()
                 self._remote.update(clean=True)
                 raise
         else:
             try:
                 self._remote.merge(source_ref.commit_id)
                 self._remote.invalidate_vcs_cache()
                 self._remote.commit(
                     message=safe_str(merge_message),
                     username=safe_str('%s <%s>' % (user_name, user_email)))
                 self._remote.invalidate_vcs_cache()
                 return self._identify(), True
             except RepositoryError:
                 # Cleanup any merge leftovers
                 self._remote.update(clean=True)
                 raise
 
     def _local_close(self, target_ref, user_name, user_email,
                      source_ref, close_message=''):
         """
         Close the branch of the given source_revision
 
         Returns the commit id of the close and a boolean indicating if the
         commit needs to be pushed.
         """
         self._update(source_ref.commit_id)
         message = close_message or "Closing branch: `{}`".format(source_ref.name)
         try:
             self._remote.commit(
                 message=safe_str(message),
                 username=safe_str('%s <%s>' % (user_name, user_email)),
                 close_branch=True)
             self._remote.invalidate_vcs_cache()
             return self._identify(), True
         except RepositoryError:
             # Cleanup any commit leftovers
             self._remote.update(clean=True)
             raise
 
     def _is_the_same_branch(self, target_ref, source_ref):
         return (
             self._get_branch_name(target_ref) ==
             self._get_branch_name(source_ref))
 
     def _get_branch_name(self, ref):
         if ref.type == 'branch':
             return ref.name
         return self._remote.ctx_branch(ref.commit_id)
 
     def _get_shadow_repository_path(self, workspace_id):
         # The name of the shadow repository must start with '.', so it is
         # skipped by 'rhodecode.lib.utils.get_filesystem_repos'.
         return os.path.join(
             os.path.dirname(self.path),
             '.__shadow_%s_%s' % (os.path.basename(self.path), workspace_id))
 
     def _maybe_prepare_merge_workspace(self, workspace_id, unused_target_ref, unused_source_ref):
         shadow_repository_path = self._get_shadow_repository_path(workspace_id)
         if not os.path.exists(shadow_repository_path):
             self._local_clone(shadow_repository_path)
             log.debug(
                 'Prepared shadow repository in %s', shadow_repository_path)
 
         return shadow_repository_path
 
-    def cleanup_merge_workspace(self, workspace_id):
-        shadow_repository_path = self._get_shadow_repository_path(workspace_id)
-        shutil.rmtree(shadow_repository_path, ignore_errors=True)
-
     def _merge_repo(self, shadow_repository_path, target_ref,
                     source_repo, source_ref, merge_message,
                     merger_name, merger_email, dry_run=False,
                     use_rebase=False, close_branch=False):
 
         log.debug('Executing merge_repo with %s strategy, dry_run mode:%s',
                   'rebase' if use_rebase else 'merge', dry_run)
         if target_ref.commit_id not in self._heads():
             return MergeResponse(
                 False, False, None, MergeFailureReason.TARGET_IS_NOT_HEAD)
 
         try:
             if (target_ref.type == 'branch' and
                     len(self._heads(target_ref.name)) != 1):
                 return MergeResponse(
                     False, False, None,
                     MergeFailureReason.HG_TARGET_HAS_MULTIPLE_HEADS)
         except CommitDoesNotExistError:
             log.exception('Failure when looking up branch heads on hg target')
             return MergeResponse(
                 False, False, None, MergeFailureReason.MISSING_TARGET_REF)
 
         shadow_repo = self._get_shadow_instance(shadow_repository_path)
 
         log.debug('Pulling in target reference %s', target_ref)
         self._validate_pull_reference(target_ref)
         shadow_repo._local_pull(self.path, target_ref)
         try:
             log.debug('Pulling in source reference %s', source_ref)
             source_repo._validate_pull_reference(source_ref)
             shadow_repo._local_pull(source_repo.path, source_ref)
         except CommitDoesNotExistError:
             log.exception('Failure when doing local pull on hg shadow repo')
             return MergeResponse(
                 False, False, None, MergeFailureReason.MISSING_SOURCE_REF)
 
         merge_ref = None
         merge_commit_id = None
         close_commit_id = None
         merge_failure_reason = MergeFailureReason.NONE
 
         # enforce that close branch should be used only in case we source from
         # an actual Branch
         close_branch = close_branch and source_ref.type == 'branch'
 
         # don't allow to close branch if source and target are the same
         close_branch = close_branch and source_ref.name != target_ref.name
 
         needs_push_on_close = False
         if close_branch and not use_rebase and not dry_run:
             try:
                 close_commit_id, needs_push_on_close = shadow_repo._local_close(
                     target_ref, merger_name, merger_email, source_ref)
                 merge_possible = True
             except RepositoryError:
                 log.exception(
                     'Failure when doing close branch on hg shadow repo')
                 merge_possible = False
                 merge_failure_reason = MergeFailureReason.MERGE_FAILED
         else:
             merge_possible = True
 
         needs_push = False
         if merge_possible:
             try:
                 merge_commit_id, needs_push = shadow_repo._local_merge(
                     target_ref, merge_message, merger_name, merger_email,
                     source_ref, use_rebase=use_rebase, dry_run=dry_run)
                 merge_possible = True
 
                 # read the state of the close action, if it
                 # maybe required a push
                 needs_push = needs_push or needs_push_on_close
 
                 # Set a bookmark pointing to the merge commit. This bookmark
                 # may be used to easily identify the last successful merge
                 # commit in the shadow repository.
                 shadow_repo.bookmark('pr-merge', revision=merge_commit_id)
                 merge_ref = Reference('book', 'pr-merge', merge_commit_id)
             except SubrepoMergeError:
                 log.exception(
                     'Subrepo merge error during local merge on hg shadow repo.')
                 merge_possible = False
                 merge_failure_reason = MergeFailureReason.SUBREPO_MERGE_FAILED
                 needs_push = False
             except RepositoryError:
                 log.exception('Failure when doing local merge on hg shadow repo')
                 merge_possible = False
                 merge_failure_reason = MergeFailureReason.MERGE_FAILED
                 needs_push = False
 
         if merge_possible and not dry_run:
             if needs_push:
                 # In case the target is a bookmark, update it, so after pushing
                 # the bookmarks is also updated in the target.
                 if target_ref.type == 'book':
                     shadow_repo.bookmark(
                         target_ref.name, revision=merge_commit_id)
                 try:
                     shadow_repo_with_hooks = self._get_shadow_instance(
                         shadow_repository_path,
                         enable_hooks=True)
                     # This is the actual merge action, we push from shadow
                     # into origin.
                     # Note: the push_branches option will push any new branch
                     # defined in the source repository to the target. This may
                     # be dangerous as branches are permanent in Mercurial.
                     # This feature was requested in issue #441.
                     shadow_repo_with_hooks._local_push(
                         merge_commit_id, self.path, push_branches=True,
                         enable_hooks=True)
 
                     # maybe we also need to push the close_commit_id
                     if close_commit_id:
                         shadow_repo_with_hooks._local_push(
                             close_commit_id, self.path, push_branches=True,
                             enable_hooks=True)
                     merge_succeeded = True
                 except RepositoryError:
                     log.exception(
                         'Failure when doing local push from the shadow '
                         'repository to the target repository.')
                     merge_succeeded = False
                     merge_failure_reason = MergeFailureReason.PUSH_FAILED
             else:
                 merge_succeeded = True
         else:
             merge_succeeded = False
 
         return MergeResponse(
             merge_possible, merge_succeeded, merge_ref, merge_failure_reason)
 
     def _get_shadow_instance(
             self, shadow_repository_path, enable_hooks=False):
         config = self.config.copy()
         if not enable_hooks:
             config.clear_section('hooks')
         return MercurialRepository(shadow_repository_path, config)
 
     def _validate_pull_reference(self, reference):
         if not (reference.name in self.bookmarks or
                 reference.name in self.branches or
                 self.get_commit(reference.commit_id)):
             raise CommitDoesNotExistError(
                 'Unknown branch, bookmark or commit id')
 
     def _local_pull(self, repository_path, reference):
         """
         Fetch a branch, bookmark or commit from a local repository.
         """
         repository_path = os.path.abspath(repository_path)
         if repository_path == self.path:
             raise ValueError('Cannot pull from the same repository')
 
         reference_type_to_option_name = {
             'book': 'bookmark',
             'branch': 'branch',
         }
         option_name = reference_type_to_option_name.get(
             reference.type, 'revision')
 
         if option_name == 'revision':
             ref = reference.commit_id
         else:
             ref = reference.name
 
         options = {option_name: [ref]}
         self._remote.pull_cmd(repository_path, hooks=False, **options)
         self._remote.invalidate_vcs_cache()
 
     def bookmark(self, bookmark, revision=None):
         if isinstance(bookmark, unicode):
             bookmark = safe_str(bookmark)
         self._remote.bookmark(bookmark, revision=revision)
         self._remote.invalidate_vcs_cache()
 
     def get_path_permissions(self, username):
         hgacl_file = os.path.join(self.path, '.hg/hgacl')
 
         def read_patterns(suffix):
             svalue = None
             try:
                 svalue = hgacl.get('narrowhgacl', username + suffix)
             except configparser.NoOptionError:
                 try:
                     svalue = hgacl.get('narrowhgacl', 'default' + suffix)
                 except configparser.NoOptionError:
                     pass
             if not svalue:
                 return None
             result = ['/']
             for pattern in svalue.split():
                 result.append(pattern)
                 if '*' not in pattern and '?' not in pattern:
                     result.append(pattern + '/*')
             return result
 
         if os.path.exists(hgacl_file):
             try:
                 hgacl = configparser.RawConfigParser()
                 hgacl.read(hgacl_file)
 
                 includes = read_patterns('.includes')
                 excludes = read_patterns('.excludes')
                 return BasePathPermissionChecker.create_from_patterns(
                     includes, excludes)
             except BaseException as e:
                 msg = 'Cannot read ACL settings from {} on {}: {}'.format(
                     hgacl_file, self.name, e)
                 raise exceptions.RepositoryRequirementError(msg)
         else:
             return None
 
 
 class MercurialIndexBasedCollectionGenerator(CollectionGenerator):
 
     def _commit_factory(self, commit_id):
         return self.repo.get_commit(
             commit_idx=commit_id, pre_load=self.pre_load)