1
# Copyright (C) 2005, 2006, 2007, 2008, 2009 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
from bzrlib.lazy_import import lazy_import
18
lazy_import(globals(), """
39
revision as _mod_revision,
45
from bzrlib.bundle import serializer
46
from bzrlib.revisiontree import RevisionTree
47
from bzrlib.store.versioned import VersionedFileStore
48
from bzrlib.testament import Testament
51
from bzrlib.decorators import needs_read_lock, needs_write_lock
52
from bzrlib.inter import InterObject
53
from bzrlib.inventory import (
59
from bzrlib import registry
60
from bzrlib.symbol_versioning import (
63
from bzrlib.trace import (
64
log_exception_quietly, note, mutter, mutter_callsite, warning)
67
# Old formats display a warning, but only once
68
_deprecation_warning_done = False
71
class CommitBuilder(object):
72
"""Provides an interface to build up a commit.
74
This allows describing a tree to be committed without needing to
75
know the internals of the format of the repository.
78
# all clients should supply tree roots.
79
record_root_entry = True
80
# the default CommitBuilder does not manage trees whose root is versioned.
81
_versioned_root = False
83
def __init__(self, repository, parents, config, timestamp=None,
84
timezone=None, committer=None, revprops=None,
86
"""Initiate a CommitBuilder.
88
:param repository: Repository to commit to.
89
:param parents: Revision ids of the parents of the new revision.
90
:param config: Configuration to use.
91
:param timestamp: Optional timestamp recorded for commit.
92
:param timezone: Optional timezone for timestamp.
93
:param committer: Optional committer to set for commit.
94
:param revprops: Optional dictionary of revision properties.
95
:param revision_id: Optional revision id.
100
self._committer = self._config.username()
102
self._committer = committer
104
self.new_inventory = Inventory(None)
105
self._new_revision_id = revision_id
106
self.parents = parents
107
self.repository = repository
110
if revprops is not None:
111
self._validate_revprops(revprops)
112
self._revprops.update(revprops)
114
if timestamp is None:
115
timestamp = time.time()
116
# Restrict resolution to 1ms
117
self._timestamp = round(timestamp, 3)
120
self._timezone = osutils.local_time_offset()
122
self._timezone = int(timezone)
124
self._generate_revision_if_needed()
125
self.__heads = graph.HeadsCache(repository.get_graph()).heads
126
self._basis_delta = []
127
# API compatibility, older code that used CommitBuilder did not call
128
# .record_delete(), which means the delta that is computed would not be
129
# valid. Callers that will call record_delete() should call
130
# .will_record_deletes() to indicate that.
131
self._recording_deletes = False
132
# memo'd check for no-op commits.
133
self._any_changes = False
135
def any_changes(self):
136
"""Return True if any entries were changed.
138
This includes merge-only changes. It is the core for the --unchanged
141
:return: True if any changes have occured.
143
return self._any_changes
145
def _validate_unicode_text(self, text, context):
146
"""Verify things like commit messages don't have bogus characters."""
148
raise ValueError('Invalid value for %s: %r' % (context, text))
150
def _validate_revprops(self, revprops):
151
for key, value in revprops.iteritems():
152
# We know that the XML serializers do not round trip '\r'
153
# correctly, so refuse to accept them
154
if not isinstance(value, basestring):
155
raise ValueError('revision property (%s) is not a valid'
156
' (unicode) string: %r' % (key, value))
157
self._validate_unicode_text(value,
158
'revision property (%s)' % (key,))
160
def commit(self, message):
161
"""Make the actual commit.
163
:return: The revision id of the recorded revision.
165
self._validate_unicode_text(message, 'commit message')
166
rev = _mod_revision.Revision(
167
timestamp=self._timestamp,
168
timezone=self._timezone,
169
committer=self._committer,
171
inventory_sha1=self.inv_sha1,
172
revision_id=self._new_revision_id,
173
properties=self._revprops)
174
rev.parent_ids = self.parents
175
self.repository.add_revision(self._new_revision_id, rev,
176
self.new_inventory, self._config)
177
self.repository.commit_write_group()
178
return self._new_revision_id
181
"""Abort the commit that is being built.
183
self.repository.abort_write_group()
185
def revision_tree(self):
186
"""Return the tree that was just committed.
188
After calling commit() this can be called to get a RevisionTree
189
representing the newly committed tree. This is preferred to
190
calling Repository.revision_tree() because that may require
191
deserializing the inventory, while we already have a copy in
194
if self.new_inventory is None:
195
self.new_inventory = self.repository.get_inventory(
196
self._new_revision_id)
197
return RevisionTree(self.repository, self.new_inventory,
198
self._new_revision_id)
200
def finish_inventory(self):
201
"""Tell the builder that the inventory is finished.
203
:return: The inventory id in the repository, which can be used with
204
repository.get_inventory.
206
if self.new_inventory is None:
207
# XXX: Using these asserts causes test failures. However, at least
208
# "self._recording_deletes" seems like a useful check to do,
209
# as it ensure the delta is completely valid. Most likely this
210
# just exposes that the test suite isn't using CommitBuilder
212
# if (not self.repository._format._commit_inv_deltas
213
# or not self._recording_deletes):
214
# raise AssertionError('new_inventory is None, but we did not'
215
# ' set the flag that the repository format supports'
216
# ' partial inventory generation.')
217
# an inventory delta was accumulated without creating a new
219
basis_id = self.basis_delta_revision
220
self.inv_sha1 = self.repository.add_inventory_by_delta(
221
basis_id, self._basis_delta, self._new_revision_id,
224
if self.new_inventory.root is None:
225
raise AssertionError('Root entry should be supplied to'
226
' record_entry_contents, as of bzr 0.10.')
227
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
228
self.new_inventory.revision_id = self._new_revision_id
229
self.inv_sha1 = self.repository.add_inventory(
230
self._new_revision_id,
234
return self._new_revision_id
236
def _gen_revision_id(self):
237
"""Return new revision-id."""
238
return generate_ids.gen_revision_id(self._config.username(),
241
def _generate_revision_if_needed(self):
242
"""Create a revision id if None was supplied.
244
If the repository can not support user-specified revision ids
245
they should override this function and raise CannotSetRevisionId
246
if _new_revision_id is not None.
248
:raises: CannotSetRevisionId
250
if self._new_revision_id is None:
251
self._new_revision_id = self._gen_revision_id()
252
self.random_revid = True
254
self.random_revid = False
256
def _heads(self, file_id, revision_ids):
257
"""Calculate the graph heads for revision_ids in the graph of file_id.
259
This can use either a per-file graph or a global revision graph as we
260
have an identity relationship between the two graphs.
262
return self.__heads(revision_ids)
264
def _check_root(self, ie, parent_invs, tree):
265
"""Helper for record_entry_contents.
267
:param ie: An entry being added.
268
:param parent_invs: The inventories of the parent revisions of the
270
:param tree: The tree that is being committed.
272
# In this revision format, root entries have no knit or weave When
273
# serializing out to disk and back in root.revision is always
275
ie.revision = self._new_revision_id
277
def _require_root_change(self, tree):
278
"""Enforce an appropriate root object change.
280
This is called once when record_iter_changes is called, if and only if
281
the root was not in the delta calculated by record_iter_changes.
283
:param tree: The tree which is being committed.
285
# NB: if there are no parents then this method is not called, so no
286
# need to guard on parents having length.
287
entry = entry_factory['directory'](tree.path2id(''), '',
289
entry.revision = self._new_revision_id
290
self._basis_delta.append(('', '', entry.file_id, entry))
292
def _get_delta(self, ie, basis_inv, path):
293
"""Get a delta against the basis inventory for ie."""
294
if ie.file_id not in basis_inv:
296
result = (None, path, ie.file_id, ie)
297
self._basis_delta.append(result)
299
elif ie != basis_inv[ie.file_id]:
301
# TODO: avoid tis id2path call.
302
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
303
self._basis_delta.append(result)
309
def get_basis_delta(self):
310
"""Return the complete inventory delta versus the basis inventory.
312
This has been built up with the calls to record_delete and
313
record_entry_contents. The client must have already called
314
will_record_deletes() to indicate that they will be generating a
317
:return: An inventory delta, suitable for use with apply_delta, or
318
Repository.add_inventory_by_delta, etc.
320
if not self._recording_deletes:
321
raise AssertionError("recording deletes not activated.")
322
return self._basis_delta
324
def record_delete(self, path, file_id):
325
"""Record that a delete occured against a basis tree.
327
This is an optional API - when used it adds items to the basis_delta
328
being accumulated by the commit builder. It cannot be called unless the
329
method will_record_deletes() has been called to inform the builder that
330
a delta is being supplied.
332
:param path: The path of the thing deleted.
333
:param file_id: The file id that was deleted.
335
if not self._recording_deletes:
336
raise AssertionError("recording deletes not activated.")
337
delta = (path, None, file_id, None)
338
self._basis_delta.append(delta)
339
self._any_changes = True
342
def will_record_deletes(self):
343
"""Tell the commit builder that deletes are being notified.
345
This enables the accumulation of an inventory delta; for the resulting
346
commit to be valid, deletes against the basis MUST be recorded via
347
builder.record_delete().
349
self._recording_deletes = True
351
basis_id = self.parents[0]
353
basis_id = _mod_revision.NULL_REVISION
354
self.basis_delta_revision = basis_id
356
def record_entry_contents(self, ie, parent_invs, path, tree,
358
"""Record the content of ie from tree into the commit if needed.
360
Side effect: sets ie.revision when unchanged
362
:param ie: An inventory entry present in the commit.
363
:param parent_invs: The inventories of the parent revisions of the
365
:param path: The path the entry is at in the tree.
366
:param tree: The tree which contains this entry and should be used to
368
:param content_summary: Summary data from the tree about the paths
369
content - stat, length, exec, sha/link target. This is only
370
accessed when the entry has a revision of None - that is when it is
371
a candidate to commit.
372
:return: A tuple (change_delta, version_recorded, fs_hash).
373
change_delta is an inventory_delta change for this entry against
374
the basis tree of the commit, or None if no change occured against
376
version_recorded is True if a new version of the entry has been
377
recorded. For instance, committing a merge where a file was only
378
changed on the other side will return (delta, False).
379
fs_hash is either None, or the hash details for the path (currently
380
a tuple of the contents sha1 and the statvalue returned by
381
tree.get_file_with_stat()).
383
if self.new_inventory.root is None:
384
if ie.parent_id is not None:
385
raise errors.RootMissing()
386
self._check_root(ie, parent_invs, tree)
387
if ie.revision is None:
388
kind = content_summary[0]
390
# ie is carried over from a prior commit
392
# XXX: repository specific check for nested tree support goes here - if
393
# the repo doesn't want nested trees we skip it ?
394
if (kind == 'tree-reference' and
395
not self.repository._format.supports_tree_reference):
396
# mismatch between commit builder logic and repository:
397
# this needs the entry creation pushed down into the builder.
398
raise NotImplementedError('Missing repository subtree support.')
399
self.new_inventory.add(ie)
401
# TODO: slow, take it out of the inner loop.
403
basis_inv = parent_invs[0]
405
basis_inv = Inventory(root_id=None)
407
# ie.revision is always None if the InventoryEntry is considered
408
# for committing. We may record the previous parents revision if the
409
# content is actually unchanged against a sole head.
410
if ie.revision is not None:
411
if not self._versioned_root and path == '':
412
# repositories that do not version the root set the root's
413
# revision to the new commit even when no change occurs (more
414
# specifically, they do not record a revision on the root; and
415
# the rev id is assigned to the root during deserialisation -
416
# this masks when a change may have occurred against the basis.
417
# To match this we always issue a delta, because the revision
418
# of the root will always be changing.
419
if ie.file_id in basis_inv:
420
delta = (basis_inv.id2path(ie.file_id), path,
424
delta = (None, path, ie.file_id, ie)
425
self._basis_delta.append(delta)
426
return delta, False, None
428
# we don't need to commit this, because the caller already
429
# determined that an existing revision of this file is
430
# appropriate. If its not being considered for committing then
431
# it and all its parents to the root must be unaltered so
432
# no-change against the basis.
433
if ie.revision == self._new_revision_id:
434
raise AssertionError("Impossible situation, a skipped "
435
"inventory entry (%r) claims to be modified in this "
436
"commit (%r).", (ie, self._new_revision_id))
437
return None, False, None
438
# XXX: Friction: parent_candidates should return a list not a dict
439
# so that we don't have to walk the inventories again.
440
parent_candiate_entries = ie.parent_candidates(parent_invs)
441
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
443
for inv in parent_invs:
444
if ie.file_id in inv:
445
old_rev = inv[ie.file_id].revision
446
if old_rev in head_set:
447
heads.append(inv[ie.file_id].revision)
448
head_set.remove(inv[ie.file_id].revision)
451
# now we check to see if we need to write a new record to the
453
# We write a new entry unless there is one head to the ancestors, and
454
# the kind-derived content is unchanged.
456
# Cheapest check first: no ancestors, or more the one head in the
457
# ancestors, we write a new node.
461
# There is a single head, look it up for comparison
462
parent_entry = parent_candiate_entries[heads[0]]
463
# if the non-content specific data has changed, we'll be writing a
465
if (parent_entry.parent_id != ie.parent_id or
466
parent_entry.name != ie.name):
468
# now we need to do content specific checks:
470
# if the kind changed the content obviously has
471
if kind != parent_entry.kind:
473
# Stat cache fingerprint feedback for the caller - None as we usually
474
# don't generate one.
477
if content_summary[2] is None:
478
raise ValueError("Files must not have executable = None")
480
if (# if the file length changed we have to store:
481
parent_entry.text_size != content_summary[1] or
482
# if the exec bit has changed we have to store:
483
parent_entry.executable != content_summary[2]):
485
elif parent_entry.text_sha1 == content_summary[3]:
486
# all meta and content is unchanged (using a hash cache
487
# hit to check the sha)
488
ie.revision = parent_entry.revision
489
ie.text_size = parent_entry.text_size
490
ie.text_sha1 = parent_entry.text_sha1
491
ie.executable = parent_entry.executable
492
return self._get_delta(ie, basis_inv, path), False, None
494
# Either there is only a hash change(no hash cache entry,
495
# or same size content change), or there is no change on
497
# Provide the parent's hash to the store layer, so that the
498
# content is unchanged we will not store a new node.
499
nostore_sha = parent_entry.text_sha1
501
# We want to record a new node regardless of the presence or
502
# absence of a content change in the file.
504
ie.executable = content_summary[2]
505
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
507
lines = file_obj.readlines()
511
ie.text_sha1, ie.text_size = self._add_text_to_weave(
512
ie.file_id, lines, heads, nostore_sha)
513
# Let the caller know we generated a stat fingerprint.
514
fingerprint = (ie.text_sha1, stat_value)
515
except errors.ExistingContent:
516
# Turns out that the file content was unchanged, and we were
517
# only going to store a new node if it was changed. Carry over
519
ie.revision = parent_entry.revision
520
ie.text_size = parent_entry.text_size
521
ie.text_sha1 = parent_entry.text_sha1
522
ie.executable = parent_entry.executable
523
return self._get_delta(ie, basis_inv, path), False, None
524
elif kind == 'directory':
526
# all data is meta here, nothing specific to directory, so
528
ie.revision = parent_entry.revision
529
return self._get_delta(ie, basis_inv, path), False, None
531
self._add_text_to_weave(ie.file_id, lines, heads, None)
532
elif kind == 'symlink':
533
current_link_target = content_summary[3]
535
# symlink target is not generic metadata, check if it has
537
if current_link_target != parent_entry.symlink_target:
540
# unchanged, carry over.
541
ie.revision = parent_entry.revision
542
ie.symlink_target = parent_entry.symlink_target
543
return self._get_delta(ie, basis_inv, path), False, None
544
ie.symlink_target = current_link_target
546
self._add_text_to_weave(ie.file_id, lines, heads, None)
547
elif kind == 'tree-reference':
549
if content_summary[3] != parent_entry.reference_revision:
552
# unchanged, carry over.
553
ie.reference_revision = parent_entry.reference_revision
554
ie.revision = parent_entry.revision
555
return self._get_delta(ie, basis_inv, path), False, None
556
ie.reference_revision = content_summary[3]
558
self._add_text_to_weave(ie.file_id, lines, heads, None)
560
raise NotImplementedError('unknown kind')
561
ie.revision = self._new_revision_id
562
self._any_changes = True
563
return self._get_delta(ie, basis_inv, path), True, fingerprint
565
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
566
_entry_factory=entry_factory):
567
"""Record a new tree via iter_changes.
569
:param tree: The tree to obtain text contents from for changed objects.
570
:param basis_revision_id: The revision id of the tree the iter_changes
571
has been generated against. Currently assumed to be the same
572
as self.parents[0] - if it is not, errors may occur.
573
:param iter_changes: An iter_changes iterator with the changes to apply
574
to basis_revision_id. The iterator must not include any items with
575
a current kind of None - missing items must be either filtered out
576
or errored-on beefore record_iter_changes sees the item.
577
:param _entry_factory: Private method to bind entry_factory locally for
579
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
582
# Create an inventory delta based on deltas between all the parents and
583
# deltas between all the parent inventories. We use inventory delta's
584
# between the inventory objects because iter_changes masks
585
# last-changed-field only changes.
587
# file_id -> change map, change is fileid, paths, changed, versioneds,
588
# parents, names, kinds, executables
590
# {file_id -> revision_id -> inventory entry, for entries in parent
591
# trees that are not parents[0]
595
revtrees = list(self.repository.revision_trees(self.parents))
596
except errors.NoSuchRevision:
597
# one or more ghosts, slow path.
599
for revision_id in self.parents:
601
revtrees.append(self.repository.revision_tree(revision_id))
602
except errors.NoSuchRevision:
604
basis_revision_id = _mod_revision.NULL_REVISION
606
revtrees.append(self.repository.revision_tree(
607
_mod_revision.NULL_REVISION))
608
# The basis inventory from a repository
610
basis_inv = revtrees[0].inventory
612
basis_inv = self.repository.revision_tree(
613
_mod_revision.NULL_REVISION).inventory
614
if len(self.parents) > 0:
615
if basis_revision_id != self.parents[0] and not ghost_basis:
617
"arbitrary basis parents not yet supported with merges")
618
for revtree in revtrees[1:]:
619
for change in revtree.inventory._make_delta(basis_inv):
620
if change[1] is None:
621
# Not present in this parent.
623
if change[2] not in merged_ids:
624
if change[0] is not None:
625
basis_entry = basis_inv[change[2]]
626
merged_ids[change[2]] = [
628
basis_entry.revision,
631
parent_entries[change[2]] = {
633
basis_entry.revision:basis_entry,
635
change[3].revision:change[3],
638
merged_ids[change[2]] = [change[3].revision]
639
parent_entries[change[2]] = {change[3].revision:change[3]}
641
merged_ids[change[2]].append(change[3].revision)
642
parent_entries[change[2]][change[3].revision] = change[3]
645
# Setup the changes from the tree:
646
# changes maps file_id -> (change, [parent revision_ids])
648
for change in iter_changes:
649
# This probably looks up in basis_inv way to much.
650
if change[1][0] is not None:
651
head_candidate = [basis_inv[change[0]].revision]
654
changes[change[0]] = change, merged_ids.get(change[0],
656
unchanged_merged = set(merged_ids) - set(changes)
657
# Extend the changes dict with synthetic changes to record merges of
659
for file_id in unchanged_merged:
660
# Record a merged version of these items that did not change vs the
661
# basis. This can be either identical parallel changes, or a revert
662
# of a specific file after a merge. The recorded content will be
663
# that of the current tree (which is the same as the basis), but
664
# the per-file graph will reflect a merge.
665
# NB:XXX: We are reconstructing path information we had, this
666
# should be preserved instead.
667
# inv delta change: (file_id, (path_in_source, path_in_target),
668
# changed_content, versioned, parent, name, kind,
671
basis_entry = basis_inv[file_id]
672
except errors.NoSuchId:
673
# a change from basis->some_parents but file_id isn't in basis
674
# so was new in the merge, which means it must have changed
675
# from basis -> current, and as it hasn't the add was reverted
676
# by the user. So we discard this change.
680
(basis_inv.id2path(file_id), tree.id2path(file_id)),
682
(basis_entry.parent_id, basis_entry.parent_id),
683
(basis_entry.name, basis_entry.name),
684
(basis_entry.kind, basis_entry.kind),
685
(basis_entry.executable, basis_entry.executable))
686
changes[file_id] = (change, merged_ids[file_id])
687
# changes contains tuples with the change and a set of inventory
688
# candidates for the file.
690
# old_path, new_path, file_id, new_inventory_entry
691
seen_root = False # Is the root in the basis delta?
692
inv_delta = self._basis_delta
693
modified_rev = self._new_revision_id
694
for change, head_candidates in changes.values():
695
if change[3][1]: # versioned in target.
696
# Several things may be happening here:
697
# We may have a fork in the per-file graph
698
# - record a change with the content from tree
699
# We may have a change against < all trees
700
# - carry over the tree that hasn't changed
701
# We may have a change against all trees
702
# - record the change with the content from tree
705
entry = _entry_factory[kind](file_id, change[5][1],
707
head_set = self._heads(change[0], set(head_candidates))
710
for head_candidate in head_candidates:
711
if head_candidate in head_set:
712
heads.append(head_candidate)
713
head_set.remove(head_candidate)
716
# Could be a carry-over situation:
717
parent_entry_revs = parent_entries.get(file_id, None)
718
if parent_entry_revs:
719
parent_entry = parent_entry_revs.get(heads[0], None)
722
if parent_entry is None:
723
# The parent iter_changes was called against is the one
724
# that is the per-file head, so any change is relevant
725
# iter_changes is valid.
726
carry_over_possible = False
728
# could be a carry over situation
729
# A change against the basis may just indicate a merge,
730
# we need to check the content against the source of the
731
# merge to determine if it was changed after the merge
733
if (parent_entry.kind != entry.kind or
734
parent_entry.parent_id != entry.parent_id or
735
parent_entry.name != entry.name):
736
# Metadata common to all entries has changed
737
# against per-file parent
738
carry_over_possible = False
740
carry_over_possible = True
741
# per-type checks for changes against the parent_entry
744
# Cannot be a carry-over situation
745
carry_over_possible = False
746
# Populate the entry in the delta
748
# XXX: There is still a small race here: If someone reverts the content of a file
749
# after iter_changes examines and decides it has changed,
750
# we will unconditionally record a new version even if some
751
# other process reverts it while commit is running (with
752
# the revert happening after iter_changes did it's
755
entry.executable = True
757
entry.executable = False
758
if (carry_over_possible and
759
parent_entry.executable == entry.executable):
760
# Check the file length, content hash after reading
762
nostore_sha = parent_entry.text_sha1
765
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
767
lines = file_obj.readlines()
771
entry.text_sha1, entry.text_size = self._add_text_to_weave(
772
file_id, lines, heads, nostore_sha)
773
yield file_id, change[1][1], (entry.text_sha1, stat_value)
774
except errors.ExistingContent:
775
# No content change against a carry_over parent
776
# Perhaps this should also yield a fs hash update?
778
entry.text_size = parent_entry.text_size
779
entry.text_sha1 = parent_entry.text_sha1
780
elif kind == 'symlink':
782
entry.symlink_target = tree.get_symlink_target(file_id)
783
if (carry_over_possible and
784
parent_entry.symlink_target == entry.symlink_target):
787
self._add_text_to_weave(change[0], [], heads, None)
788
elif kind == 'directory':
789
if carry_over_possible:
792
# Nothing to set on the entry.
793
# XXX: split into the Root and nonRoot versions.
794
if change[1][1] != '' or self.repository.supports_rich_root():
795
self._add_text_to_weave(change[0], [], heads, None)
796
elif kind == 'tree-reference':
797
if not self.repository._format.supports_tree_reference:
798
# This isn't quite sane as an error, but we shouldn't
799
# ever see this code path in practice: tree's don't
800
# permit references when the repo doesn't support tree
802
raise errors.UnsupportedOperation(tree.add_reference,
804
entry.reference_revision = \
805
tree.get_reference_revision(change[0])
806
if (carry_over_possible and
807
parent_entry.reference_revision == reference_revision):
810
self._add_text_to_weave(change[0], [], heads, None)
812
raise AssertionError('unknown kind %r' % kind)
814
entry.revision = modified_rev
816
entry.revision = parent_entry.revision
819
new_path = change[1][1]
820
inv_delta.append((change[1][0], new_path, change[0], entry))
823
self.new_inventory = None
825
self._any_changes = True
827
# housekeeping root entry changes do not affect no-change commits.
828
self._require_root_change(tree)
829
self.basis_delta_revision = basis_revision_id
831
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
832
# Note: as we read the content directly from the tree, we know its not
833
# been turned into unicode or badly split - but a broken tree
834
# implementation could give us bad output from readlines() so this is
835
# not a guarantee of safety. What would be better is always checking
836
# the content during test suite execution. RBC 20070912
837
parent_keys = tuple((file_id, parent) for parent in parents)
838
return self.repository.texts.add_lines(
839
(file_id, self._new_revision_id), parent_keys, new_lines,
840
nostore_sha=nostore_sha, random_id=self.random_revid,
841
check_content=False)[0:2]
844
class RootCommitBuilder(CommitBuilder):
845
"""This commitbuilder actually records the root id"""
847
# the root entry gets versioned properly by this builder.
848
_versioned_root = True
850
def _check_root(self, ie, parent_invs, tree):
851
"""Helper for record_entry_contents.
853
:param ie: An entry being added.
854
:param parent_invs: The inventories of the parent revisions of the
856
:param tree: The tree that is being committed.
859
def _require_root_change(self, tree):
860
"""Enforce an appropriate root object change.
862
This is called once when record_iter_changes is called, if and only if
863
the root was not in the delta calculated by record_iter_changes.
865
:param tree: The tree which is being committed.
867
# versioned roots do not change unless the tree found a change.
870
######################################################################
873
class Repository(object):
874
"""Repository holding history for one or more branches.
876
The repository holds and retrieves historical information including
877
revisions and file history. It's normally accessed only by the Branch,
878
which views a particular line of development through that history.
880
The Repository builds on top of some byte storage facilies (the revisions,
881
signatures, inventories, texts and chk_bytes attributes) and a Transport,
882
which respectively provide byte storage and a means to access the (possibly
885
The byte storage facilities are addressed via tuples, which we refer to
886
as 'keys' throughout the code base. Revision_keys, inventory_keys and
887
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
888
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
889
byte string made up of a hash identifier and a hash value.
890
We use this interface because it allows low friction with the underlying
891
code that implements disk indices, network encoding and other parts of
894
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
895
the serialised revisions for the repository. This can be used to obtain
896
revision graph information or to access raw serialised revisions.
897
The result of trying to insert data into the repository via this store
898
is undefined: it should be considered read-only except for implementors
900
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
901
the serialised signatures for the repository. This can be used to
902
obtain access to raw serialised signatures. The result of trying to
903
insert data into the repository via this store is undefined: it should
904
be considered read-only except for implementors of repositories.
905
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
906
the serialised inventories for the repository. This can be used to
907
obtain unserialised inventories. The result of trying to insert data
908
into the repository via this store is undefined: it should be
909
considered read-only except for implementors of repositories.
910
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
911
texts of files and directories for the repository. This can be used to
912
obtain file texts or file graphs. Note that Repository.iter_file_bytes
913
is usually a better interface for accessing file texts.
914
The result of trying to insert data into the repository via this store
915
is undefined: it should be considered read-only except for implementors
917
:ivar chk_bytes: A bzrlib.versionedfile.VersioedFiles instance containing
918
any data the repository chooses to store or have indexed by its hash.
919
The result of trying to insert data into the repository via this store
920
is undefined: it should be considered read-only except for implementors
922
:ivar _transport: Transport for file access to repository, typically
923
pointing to .bzr/repository.
926
# What class to use for a CommitBuilder. Often its simpler to change this
927
# in a Repository class subclass rather than to override
928
# get_commit_builder.
929
_commit_builder_class = CommitBuilder
930
# The search regex used by xml based repositories to determine what things
931
# where changed in a single commit.
932
_file_ids_altered_regex = lazy_regex.lazy_compile(
933
r'file_id="(?P<file_id>[^"]+)"'
934
r'.* revision="(?P<revision_id>[^"]+)"'
937
def abort_write_group(self, suppress_errors=False):
938
"""Commit the contents accrued within the current write group.
940
:param suppress_errors: if true, abort_write_group will catch and log
941
unexpected errors that happen during the abort, rather than
942
allowing them to propagate. Defaults to False.
944
:seealso: start_write_group.
946
if self._write_group is not self.get_transaction():
947
# has an unlock or relock occured ?
948
raise errors.BzrError(
949
'mismatched lock context and write group. %r, %r' %
950
(self._write_group, self.get_transaction()))
952
self._abort_write_group()
953
except Exception, exc:
954
self._write_group = None
955
if not suppress_errors:
957
mutter('abort_write_group failed')
958
log_exception_quietly()
959
note('bzr: ERROR (ignored): %s', exc)
960
self._write_group = None
962
def _abort_write_group(self):
963
"""Template method for per-repository write group cleanup.
965
This is called during abort before the write group is considered to be
966
finished and should cleanup any internal state accrued during the write
967
group. There is no requirement that data handed to the repository be
968
*not* made available - this is not a rollback - but neither should any
969
attempt be made to ensure that data added is fully commited. Abort is
970
invoked when an error has occured so futher disk or network operations
971
may not be possible or may error and if possible should not be
975
def add_fallback_repository(self, repository):
976
"""Add a repository to use for looking up data not held locally.
978
:param repository: A repository.
980
if not self._format.supports_external_lookups:
981
raise errors.UnstackableRepositoryFormat(self._format, self.base)
982
self._check_fallback_repository(repository)
983
self._fallback_repositories.append(repository)
984
self.texts.add_fallback_versioned_files(repository.texts)
985
self.inventories.add_fallback_versioned_files(repository.inventories)
986
self.revisions.add_fallback_versioned_files(repository.revisions)
987
self.signatures.add_fallback_versioned_files(repository.signatures)
988
if self.chk_bytes is not None:
989
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
990
self._fetch_order = 'topological'
992
def _check_fallback_repository(self, repository):
993
"""Check that this repository can fallback to repository safely.
995
Raise an error if not.
997
:param repository: A repository to fallback to.
999
return InterRepository._assert_same_model(self, repository)
1001
def add_inventory(self, revision_id, inv, parents):
1002
"""Add the inventory inv to the repository as revision_id.
1004
:param parents: The revision ids of the parents that revision_id
1005
is known to have and are in the repository already.
1007
:returns: The validator(which is a sha1 digest, though what is sha'd is
1008
repository format specific) of the serialized inventory.
1010
if not self.is_in_write_group():
1011
raise AssertionError("%r not in write group" % (self,))
1012
_mod_revision.check_not_reserved_id(revision_id)
1013
if not (inv.revision_id is None or inv.revision_id == revision_id):
1014
raise AssertionError(
1015
"Mismatch between inventory revision"
1016
" id and insertion revid (%r, %r)"
1017
% (inv.revision_id, revision_id))
1018
if inv.root is None:
1019
raise AssertionError()
1020
return self._add_inventory_checked(revision_id, inv, parents)
1022
def _add_inventory_checked(self, revision_id, inv, parents):
1023
"""Add inv to the repository after checking the inputs.
1025
This function can be overridden to allow different inventory styles.
1027
:seealso: add_inventory, for the contract.
1029
inv_lines = self._serialise_inventory_to_lines(inv)
1030
return self._inventory_add_lines(revision_id, parents,
1031
inv_lines, check_content=False)
1033
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1034
parents, basis_inv=None, propagate_caches=False):
1035
"""Add a new inventory expressed as a delta against another revision.
1037
:param basis_revision_id: The inventory id the delta was created
1038
against. (This does not have to be a direct parent.)
1039
:param delta: The inventory delta (see Inventory.apply_delta for
1041
:param new_revision_id: The revision id that the inventory is being
1043
:param parents: The revision ids of the parents that revision_id is
1044
known to have and are in the repository already. These are supplied
1045
for repositories that depend on the inventory graph for revision
1046
graph access, as well as for those that pun ancestry with delta
1048
:param basis_inv: The basis inventory if it is already known,
1050
:param propagate_caches: If True, the caches for this inventory are
1051
copied to and updated for the result if possible.
1053
:returns: (validator, new_inv)
1054
The validator(which is a sha1 digest, though what is sha'd is
1055
repository format specific) of the serialized inventory, and the
1056
resulting inventory.
1058
if not self.is_in_write_group():
1059
raise AssertionError("%r not in write group" % (self,))
1060
_mod_revision.check_not_reserved_id(new_revision_id)
1061
basis_tree = self.revision_tree(basis_revision_id)
1062
basis_tree.lock_read()
1064
# Note that this mutates the inventory of basis_tree, which not all
1065
# inventory implementations may support: A better idiom would be to
1066
# return a new inventory, but as there is no revision tree cache in
1067
# repository this is safe for now - RBC 20081013
1068
if basis_inv is None:
1069
basis_inv = basis_tree.inventory
1070
basis_inv.apply_delta(delta)
1071
basis_inv.revision_id = new_revision_id
1072
return (self.add_inventory(new_revision_id, basis_inv, parents),
1077
def _inventory_add_lines(self, revision_id, parents, lines,
1078
check_content=True):
1079
"""Store lines in inv_vf and return the sha1 of the inventory."""
1080
parents = [(parent,) for parent in parents]
1081
return self.inventories.add_lines((revision_id,), parents, lines,
1082
check_content=check_content)[0]
1084
def add_revision(self, revision_id, rev, inv=None, config=None):
1085
"""Add rev to the revision store as revision_id.
1087
:param revision_id: the revision id to use.
1088
:param rev: The revision object.
1089
:param inv: The inventory for the revision. if None, it will be looked
1090
up in the inventory storer
1091
:param config: If None no digital signature will be created.
1092
If supplied its signature_needed method will be used
1093
to determine if a signature should be made.
1095
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1097
_mod_revision.check_not_reserved_id(revision_id)
1098
if config is not None and config.signature_needed():
1100
inv = self.get_inventory(revision_id)
1101
plaintext = Testament(rev, inv).as_short_text()
1102
self.store_revision_signature(
1103
gpg.GPGStrategy(config), plaintext, revision_id)
1104
# check inventory present
1105
if not self.inventories.get_parent_map([(revision_id,)]):
1107
raise errors.WeaveRevisionNotPresent(revision_id,
1110
# yes, this is not suitable for adding with ghosts.
1111
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1114
key = (revision_id,)
1115
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1116
self._add_revision(rev)
1118
def _add_revision(self, revision):
1119
text = self._serializer.write_revision_to_string(revision)
1120
key = (revision.revision_id,)
1121
parents = tuple((parent,) for parent in revision.parent_ids)
1122
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1124
def all_revision_ids(self):
1125
"""Returns a list of all the revision ids in the repository.
1127
This is conceptually deprecated because code should generally work on
1128
the graph reachable from a particular revision, and ignore any other
1129
revisions that might be present. There is no direct replacement
1132
if 'evil' in debug.debug_flags:
1133
mutter_callsite(2, "all_revision_ids is linear with history.")
1134
return self._all_revision_ids()
1136
def _all_revision_ids(self):
1137
"""Returns a list of all the revision ids in the repository.
1139
These are in as much topological order as the underlying store can
1142
raise NotImplementedError(self._all_revision_ids)
1144
def break_lock(self):
1145
"""Break a lock if one is present from another instance.
1147
Uses the ui factory to ask for confirmation if the lock may be from
1150
self.control_files.break_lock()
1153
def _eliminate_revisions_not_present(self, revision_ids):
1154
"""Check every revision id in revision_ids to see if we have it.
1156
Returns a set of the present revisions.
1159
graph = self.get_graph()
1160
parent_map = graph.get_parent_map(revision_ids)
1161
# The old API returned a list, should this actually be a set?
1162
return parent_map.keys()
1165
def create(a_bzrdir):
1166
"""Construct the current default format repository in a_bzrdir."""
1167
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1169
def __init__(self, _format, a_bzrdir, control_files):
1170
"""instantiate a Repository.
1172
:param _format: The format of the repository on disk.
1173
:param a_bzrdir: The BzrDir of the repository.
1175
In the future we will have a single api for all stores for
1176
getting file texts, inventories and revisions, then
1177
this construct will accept instances of those things.
1179
super(Repository, self).__init__()
1180
self._format = _format
1181
# the following are part of the public API for Repository:
1182
self.bzrdir = a_bzrdir
1183
self.control_files = control_files
1184
self._transport = control_files._transport
1185
self.base = self._transport.base
1187
self._reconcile_does_inventory_gc = True
1188
self._reconcile_fixes_text_parents = False
1189
self._reconcile_backsup_inventory = True
1190
# not right yet - should be more semantically clear ?
1192
# TODO: make sure to construct the right store classes, etc, depending
1193
# on whether escaping is required.
1194
self._warn_if_deprecated()
1195
self._write_group = None
1196
# Additional places to query for data.
1197
self._fallback_repositories = []
1198
# An InventoryEntry cache, used during deserialization
1199
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1202
return '%s(%r)' % (self.__class__.__name__,
1205
def has_same_location(self, other):
1206
"""Returns a boolean indicating if this repository is at the same
1207
location as another repository.
1209
This might return False even when two repository objects are accessing
1210
the same physical repository via different URLs.
1212
if self.__class__ is not other.__class__:
1214
return (self._transport.base == other._transport.base)
1216
def is_in_write_group(self):
1217
"""Return True if there is an open write group.
1219
:seealso: start_write_group.
1221
return self._write_group is not None
1223
def is_locked(self):
1224
return self.control_files.is_locked()
1226
def is_write_locked(self):
1227
"""Return True if this object is write locked."""
1228
return self.is_locked() and self.control_files._lock_mode == 'w'
1230
def lock_write(self, token=None):
1231
"""Lock this repository for writing.
1233
This causes caching within the repository obejct to start accumlating
1234
data during reads, and allows a 'write_group' to be obtained. Write
1235
groups must be used for actual data insertion.
1237
:param token: if this is already locked, then lock_write will fail
1238
unless the token matches the existing lock.
1239
:returns: a token if this instance supports tokens, otherwise None.
1240
:raises TokenLockingNotSupported: when a token is given but this
1241
instance doesn't support using token locks.
1242
:raises MismatchedToken: if the specified token doesn't match the token
1243
of the existing lock.
1244
:seealso: start_write_group.
1246
A token should be passed in if you know that you have locked the object
1247
some other way, and need to synchronise this object's state with that
1250
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1252
locked = self.is_locked()
1253
result = self.control_files.lock_write(token=token)
1254
for repo in self._fallback_repositories:
1255
# Writes don't affect fallback repos
1258
self._refresh_data()
1261
def lock_read(self):
1262
locked = self.is_locked()
1263
self.control_files.lock_read()
1264
for repo in self._fallback_repositories:
1267
self._refresh_data()
1269
def get_physical_lock_status(self):
1270
return self.control_files.get_physical_lock_status()
1272
def leave_lock_in_place(self):
1273
"""Tell this repository not to release the physical lock when this
1276
If lock_write doesn't return a token, then this method is not supported.
1278
self.control_files.leave_in_place()
1280
def dont_leave_lock_in_place(self):
1281
"""Tell this repository to release the physical lock when this
1282
object is unlocked, even if it didn't originally acquire it.
1284
If lock_write doesn't return a token, then this method is not supported.
1286
self.control_files.dont_leave_in_place()
1289
def gather_stats(self, revid=None, committers=None):
1290
"""Gather statistics from a revision id.
1292
:param revid: The revision id to gather statistics from, if None, then
1293
no revision specific statistics are gathered.
1294
:param committers: Optional parameter controlling whether to grab
1295
a count of committers from the revision specific statistics.
1296
:return: A dictionary of statistics. Currently this contains:
1297
committers: The number of committers if requested.
1298
firstrev: A tuple with timestamp, timezone for the penultimate left
1299
most ancestor of revid, if revid is not the NULL_REVISION.
1300
latestrev: A tuple with timestamp, timezone for revid, if revid is
1301
not the NULL_REVISION.
1302
revisions: The total revision count in the repository.
1303
size: An estimate disk size of the repository in bytes.
1306
if revid and committers:
1307
result['committers'] = 0
1308
if revid and revid != _mod_revision.NULL_REVISION:
1310
all_committers = set()
1311
revisions = self.get_ancestry(revid)
1312
# pop the leading None
1314
first_revision = None
1316
# ignore the revisions in the middle - just grab first and last
1317
revisions = revisions[0], revisions[-1]
1318
for revision in self.get_revisions(revisions):
1319
if not first_revision:
1320
first_revision = revision
1322
all_committers.add(revision.committer)
1323
last_revision = revision
1325
result['committers'] = len(all_committers)
1326
result['firstrev'] = (first_revision.timestamp,
1327
first_revision.timezone)
1328
result['latestrev'] = (last_revision.timestamp,
1329
last_revision.timezone)
1331
# now gather global repository information
1332
# XXX: This is available for many repos regardless of listability.
1333
if self.bzrdir.root_transport.listable():
1334
# XXX: do we want to __define len__() ?
1335
# Maybe the versionedfiles object should provide a different
1336
# method to get the number of keys.
1337
result['revisions'] = len(self.revisions.keys())
1338
# result['size'] = t
1341
def find_branches(self, using=False):
1342
"""Find branches underneath this repository.
1344
This will include branches inside other branches.
1346
:param using: If True, list only branches using this repository.
1348
if using and not self.is_shared():
1350
return [self.bzrdir.open_branch()]
1351
except errors.NotBranchError:
1353
class Evaluator(object):
1356
self.first_call = True
1358
def __call__(self, bzrdir):
1359
# On the first call, the parameter is always the bzrdir
1360
# containing the current repo.
1361
if not self.first_call:
1363
repository = bzrdir.open_repository()
1364
except errors.NoRepositoryPresent:
1367
return False, (None, repository)
1368
self.first_call = False
1370
value = (bzrdir.open_branch(), None)
1371
except errors.NotBranchError:
1372
value = (None, None)
1376
for branch, repository in bzrdir.BzrDir.find_bzrdirs(
1377
self.bzrdir.root_transport, evaluate=Evaluator()):
1378
if branch is not None:
1379
branches.append(branch)
1380
if not using and repository is not None:
1381
branches.extend(repository.find_branches())
1385
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
1386
"""Return the revision ids that other has that this does not.
1388
These are returned in topological order.
1390
revision_id: only return revision ids included by revision_id.
1392
return InterRepository.get(other, self).search_missing_revision_ids(
1393
revision_id, find_ghosts)
1397
"""Open the repository rooted at base.
1399
For instance, if the repository is at URL/.bzr/repository,
1400
Repository.open(URL) -> a Repository instance.
1402
control = bzrdir.BzrDir.open(base)
1403
return control.open_repository()
1405
def copy_content_into(self, destination, revision_id=None):
1406
"""Make a complete copy of the content in self into destination.
1408
This is a destructive operation! Do not use it on existing
1411
return InterRepository.get(self, destination).copy_content(revision_id)
1413
def commit_write_group(self):
1414
"""Commit the contents accrued within the current write group.
1416
:seealso: start_write_group.
1418
if self._write_group is not self.get_transaction():
1419
# has an unlock or relock occured ?
1420
raise errors.BzrError('mismatched lock context %r and '
1422
(self.get_transaction(), self._write_group))
1423
self._commit_write_group()
1424
self._write_group = None
1426
def _commit_write_group(self):
1427
"""Template method for per-repository write group cleanup.
1429
This is called before the write group is considered to be
1430
finished and should ensure that all data handed to the repository
1431
for writing during the write group is safely committed (to the
1432
extent possible considering file system caching etc).
1435
def suspend_write_group(self):
1436
raise errors.UnsuspendableWriteGroup(self)
1438
def refresh_data(self):
1439
"""Re-read any data needed to to synchronise with disk.
1441
This method is intended to be called after another repository instance
1442
(such as one used by a smart server) has inserted data into the
1443
repository. It may not be called during a write group, but may be
1444
called at any other time.
1446
if self.is_in_write_group():
1447
raise errors.InternalBzrError(
1448
"May not refresh_data while in a write group.")
1449
self._refresh_data()
1451
def resume_write_group(self, tokens):
1452
if not self.is_write_locked():
1453
raise errors.NotWriteLocked(self)
1454
if self._write_group:
1455
raise errors.BzrError('already in a write group')
1456
self._resume_write_group(tokens)
1457
# so we can detect unlock/relock - the write group is now entered.
1458
self._write_group = self.get_transaction()
1460
def _resume_write_group(self, tokens):
1461
raise errors.UnsuspendableWriteGroup(self)
1463
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1465
"""Fetch the content required to construct revision_id from source.
1467
If revision_id is None and fetch_spec is None, then all content is
1470
fetch() may not be used when the repository is in a write group -
1471
either finish the current write group before using fetch, or use
1472
fetch before starting the write group.
1474
:param find_ghosts: Find and copy revisions in the source that are
1475
ghosts in the target (and not reachable directly by walking out to
1476
the first-present revision in target from revision_id).
1477
:param revision_id: If specified, all the content needed for this
1478
revision ID will be copied to the target. Fetch will determine for
1479
itself which content needs to be copied.
1480
:param fetch_spec: If specified, a SearchResult or
1481
PendingAncestryResult that describes which revisions to copy. This
1482
allows copying multiple heads at once. Mutually exclusive with
1485
if fetch_spec is not None and revision_id is not None:
1486
raise AssertionError(
1487
"fetch_spec and revision_id are mutually exclusive.")
1488
if self.is_in_write_group():
1489
raise errors.InternalBzrError(
1490
"May not fetch while in a write group.")
1491
# fast path same-url fetch operations
1492
if self.has_same_location(source) and fetch_spec is None:
1493
# check that last_revision is in 'from' and then return a
1495
if (revision_id is not None and
1496
not _mod_revision.is_null(revision_id)):
1497
self.get_revision(revision_id)
1499
# if there is no specific appropriate InterRepository, this will get
1500
# the InterRepository base class, which raises an
1501
# IncompatibleRepositories when asked to fetch.
1502
inter = InterRepository.get(source, self)
1503
return inter.fetch(revision_id=revision_id, pb=pb,
1504
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1506
def create_bundle(self, target, base, fileobj, format=None):
1507
return serializer.write_bundle(self, target, base, fileobj, format)
1509
def get_commit_builder(self, branch, parents, config, timestamp=None,
1510
timezone=None, committer=None, revprops=None,
1512
"""Obtain a CommitBuilder for this repository.
1514
:param branch: Branch to commit to.
1515
:param parents: Revision ids of the parents of the new revision.
1516
:param config: Configuration to use.
1517
:param timestamp: Optional timestamp recorded for commit.
1518
:param timezone: Optional timezone for timestamp.
1519
:param committer: Optional committer to set for commit.
1520
:param revprops: Optional dictionary of revision properties.
1521
:param revision_id: Optional revision id.
1523
result = self._commit_builder_class(self, parents, config,
1524
timestamp, timezone, committer, revprops, revision_id)
1525
self.start_write_group()
1529
if (self.control_files._lock_count == 1 and
1530
self.control_files._lock_mode == 'w'):
1531
if self._write_group is not None:
1532
self.abort_write_group()
1533
self.control_files.unlock()
1534
raise errors.BzrError(
1535
'Must end write groups before releasing write locks.')
1536
self.control_files.unlock()
1537
if self.control_files._lock_count == 0:
1538
self._inventory_entry_cache.clear()
1539
for repo in self._fallback_repositories:
1543
def clone(self, a_bzrdir, revision_id=None):
1544
"""Clone this repository into a_bzrdir using the current format.
1546
Currently no check is made that the format of this repository and
1547
the bzrdir format are compatible. FIXME RBC 20060201.
1549
:return: The newly created destination repository.
1551
# TODO: deprecate after 0.16; cloning this with all its settings is
1552
# probably not very useful -- mbp 20070423
1553
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1554
self.copy_content_into(dest_repo, revision_id)
1557
def start_write_group(self):
1558
"""Start a write group in the repository.
1560
Write groups are used by repositories which do not have a 1:1 mapping
1561
between file ids and backend store to manage the insertion of data from
1562
both fetch and commit operations.
1564
A write lock is required around the start_write_group/commit_write_group
1565
for the support of lock-requiring repository formats.
1567
One can only insert data into a repository inside a write group.
1571
if not self.is_write_locked():
1572
raise errors.NotWriteLocked(self)
1573
if self._write_group:
1574
raise errors.BzrError('already in a write group')
1575
self._start_write_group()
1576
# so we can detect unlock/relock - the write group is now entered.
1577
self._write_group = self.get_transaction()
1579
def _start_write_group(self):
1580
"""Template method for per-repository write group startup.
1582
This is called before the write group is considered to be
1587
def sprout(self, to_bzrdir, revision_id=None):
1588
"""Create a descendent repository for new development.
1590
Unlike clone, this does not copy the settings of the repository.
1592
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1593
dest_repo.fetch(self, revision_id=revision_id)
1596
def _create_sprouting_repo(self, a_bzrdir, shared):
1597
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1598
# use target default format.
1599
dest_repo = a_bzrdir.create_repository()
1601
# Most control formats need the repository to be specifically
1602
# created, but on some old all-in-one formats it's not needed
1604
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1605
except errors.UninitializableFormat:
1606
dest_repo = a_bzrdir.open_repository()
1609
def _get_sink(self):
1610
"""Return a sink for streaming into this repository."""
1611
return StreamSink(self)
1613
def _get_source(self, to_format):
1614
"""Return a source for streaming from this repository."""
1615
return StreamSource(self, to_format)
1618
def has_revision(self, revision_id):
1619
"""True if this repository has a copy of the revision."""
1620
return revision_id in self.has_revisions((revision_id,))
1623
def has_revisions(self, revision_ids):
1624
"""Probe to find out the presence of multiple revisions.
1626
:param revision_ids: An iterable of revision_ids.
1627
:return: A set of the revision_ids that were present.
1629
parent_map = self.revisions.get_parent_map(
1630
[(rev_id,) for rev_id in revision_ids])
1632
if _mod_revision.NULL_REVISION in revision_ids:
1633
result.add(_mod_revision.NULL_REVISION)
1634
result.update([key[0] for key in parent_map])
1638
def get_revision(self, revision_id):
1639
"""Return the Revision object for a named revision."""
1640
return self.get_revisions([revision_id])[0]
1643
def get_revision_reconcile(self, revision_id):
1644
"""'reconcile' helper routine that allows access to a revision always.
1646
This variant of get_revision does not cross check the weave graph
1647
against the revision one as get_revision does: but it should only
1648
be used by reconcile, or reconcile-alike commands that are correcting
1649
or testing the revision graph.
1651
return self._get_revisions([revision_id])[0]
1654
def get_revisions(self, revision_ids):
1655
"""Get many revisions at once."""
1656
return self._get_revisions(revision_ids)
1659
def _get_revisions(self, revision_ids):
1660
"""Core work logic to get many revisions without sanity checks."""
1661
for rev_id in revision_ids:
1662
if not rev_id or not isinstance(rev_id, basestring):
1663
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1664
keys = [(key,) for key in revision_ids]
1665
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1667
for record in stream:
1668
if record.storage_kind == 'absent':
1669
raise errors.NoSuchRevision(self, record.key[0])
1670
text = record.get_bytes_as('fulltext')
1671
rev = self._serializer.read_revision_from_string(text)
1672
revs[record.key[0]] = rev
1673
return [revs[revid] for revid in revision_ids]
1676
def get_revision_xml(self, revision_id):
1677
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1678
# would have already do it.
1679
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1680
rev = self.get_revision(revision_id)
1681
rev_tmp = cStringIO.StringIO()
1682
# the current serializer..
1683
self._serializer.write_revision(rev, rev_tmp)
1685
return rev_tmp.getvalue()
1687
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1688
"""Produce a generator of revision deltas.
1690
Note that the input is a sequence of REVISIONS, not revision_ids.
1691
Trees will be held in memory until the generator exits.
1692
Each delta is relative to the revision's lefthand predecessor.
1694
:param specific_fileids: if not None, the result is filtered
1695
so that only those file-ids, their parents and their
1696
children are included.
1698
# Get the revision-ids of interest
1699
required_trees = set()
1700
for revision in revisions:
1701
required_trees.add(revision.revision_id)
1702
required_trees.update(revision.parent_ids[:1])
1704
# Get the matching filtered trees. Note that it's more
1705
# efficient to pass filtered trees to changes_from() rather
1706
# than doing the filtering afterwards. changes_from() could
1707
# arguably do the filtering itself but it's path-based, not
1708
# file-id based, so filtering before or afterwards is
1710
if specific_fileids is None:
1711
trees = dict((t.get_revision_id(), t) for
1712
t in self.revision_trees(required_trees))
1714
trees = dict((t.get_revision_id(), t) for
1715
t in self._filtered_revision_trees(required_trees,
1718
# Calculate the deltas
1719
for revision in revisions:
1720
if not revision.parent_ids:
1721
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1723
old_tree = trees[revision.parent_ids[0]]
1724
yield trees[revision.revision_id].changes_from(old_tree)
1727
def get_revision_delta(self, revision_id, specific_fileids=None):
1728
"""Return the delta for one revision.
1730
The delta is relative to the left-hand predecessor of the
1733
:param specific_fileids: if not None, the result is filtered
1734
so that only those file-ids, their parents and their
1735
children are included.
1737
r = self.get_revision(revision_id)
1738
return list(self.get_deltas_for_revisions([r],
1739
specific_fileids=specific_fileids))[0]
1742
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1743
signature = gpg_strategy.sign(plaintext)
1744
self.add_signature_text(revision_id, signature)
1747
def add_signature_text(self, revision_id, signature):
1748
self.signatures.add_lines((revision_id,), (),
1749
osutils.split_lines(signature))
1751
def find_text_key_references(self):
1752
"""Find the text key references within the repository.
1754
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1755
to whether they were referred to by the inventory of the
1756
revision_id that they contain. The inventory texts from all present
1757
revision ids are assessed to generate this report.
1759
revision_keys = self.revisions.keys()
1760
w = self.inventories
1761
pb = ui.ui_factory.nested_progress_bar()
1763
return self._find_text_key_references_from_xml_inventory_lines(
1764
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1768
def _find_text_key_references_from_xml_inventory_lines(self,
1770
"""Core routine for extracting references to texts from inventories.
1772
This performs the translation of xml lines to revision ids.
1774
:param line_iterator: An iterator of lines, origin_version_id
1775
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1776
to whether they were referred to by the inventory of the
1777
revision_id that they contain. Note that if that revision_id was
1778
not part of the line_iterator's output then False will be given -
1779
even though it may actually refer to that key.
1781
if not self._serializer.support_altered_by_hack:
1782
raise AssertionError(
1783
"_find_text_key_references_from_xml_inventory_lines only "
1784
"supported for branches which store inventory as unnested xml"
1785
", not on %r" % self)
1788
# this code needs to read every new line in every inventory for the
1789
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
1790
# not present in one of those inventories is unnecessary but not
1791
# harmful because we are filtering by the revision id marker in the
1792
# inventory lines : we only select file ids altered in one of those
1793
# revisions. We don't need to see all lines in the inventory because
1794
# only those added in an inventory in rev X can contain a revision=X
1796
unescape_revid_cache = {}
1797
unescape_fileid_cache = {}
1799
# jam 20061218 In a big fetch, this handles hundreds of thousands
1800
# of lines, so it has had a lot of inlining and optimizing done.
1801
# Sorry that it is a little bit messy.
1802
# Move several functions to be local variables, since this is a long
1804
search = self._file_ids_altered_regex.search
1805
unescape = _unescape_xml
1806
setdefault = result.setdefault
1807
for line, line_key in line_iterator:
1808
match = search(line)
1811
# One call to match.group() returning multiple items is quite a
1812
# bit faster than 2 calls to match.group() each returning 1
1813
file_id, revision_id = match.group('file_id', 'revision_id')
1815
# Inlining the cache lookups helps a lot when you make 170,000
1816
# lines and 350k ids, versus 8.4 unique ids.
1817
# Using a cache helps in 2 ways:
1818
# 1) Avoids unnecessary decoding calls
1819
# 2) Re-uses cached strings, which helps in future set and
1821
# (2) is enough that removing encoding entirely along with
1822
# the cache (so we are using plain strings) results in no
1823
# performance improvement.
1825
revision_id = unescape_revid_cache[revision_id]
1827
unescaped = unescape(revision_id)
1828
unescape_revid_cache[revision_id] = unescaped
1829
revision_id = unescaped
1831
# Note that unconditionally unescaping means that we deserialise
1832
# every fileid, which for general 'pull' is not great, but we don't
1833
# really want to have some many fulltexts that this matters anyway.
1836
file_id = unescape_fileid_cache[file_id]
1838
unescaped = unescape(file_id)
1839
unescape_fileid_cache[file_id] = unescaped
1842
key = (file_id, revision_id)
1843
setdefault(key, False)
1844
if revision_id == line_key[-1]:
1848
def _inventory_xml_lines_for_keys(self, keys):
1849
"""Get a line iterator of the sort needed for findind references.
1851
Not relevant for non-xml inventory repositories.
1853
Ghosts in revision_keys are ignored.
1855
:param revision_keys: The revision keys for the inventories to inspect.
1856
:return: An iterator over (inventory line, revid) for the fulltexts of
1857
all of the xml inventories specified by revision_keys.
1859
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1860
for record in stream:
1861
if record.storage_kind != 'absent':
1862
chunks = record.get_bytes_as('chunked')
1863
revid = record.key[-1]
1864
lines = osutils.chunks_to_lines(chunks)
1868
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1870
"""Helper routine for fileids_altered_by_revision_ids.
1872
This performs the translation of xml lines to revision ids.
1874
:param line_iterator: An iterator of lines, origin_version_id
1875
:param revision_ids: The revision ids to filter for. This should be a
1876
set or other type which supports efficient __contains__ lookups, as
1877
the revision id from each parsed line will be looked up in the
1878
revision_ids filter.
1879
:return: a dictionary mapping altered file-ids to an iterable of
1880
revision_ids. Each altered file-ids has the exact revision_ids that
1881
altered it listed explicitly.
1883
seen = set(self._find_text_key_references_from_xml_inventory_lines(
1884
line_iterator).iterkeys())
1885
# Note that revision_ids are revision keys.
1886
parent_maps = self.revisions.get_parent_map(revision_ids)
1888
map(parents.update, parent_maps.itervalues())
1889
parents.difference_update(revision_ids)
1890
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
1891
self._inventory_xml_lines_for_keys(parents)))
1892
new_keys = seen - parent_seen
1894
setdefault = result.setdefault
1895
for key in new_keys:
1896
setdefault(key[0], set()).add(key[-1])
1899
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
1900
"""Find the file ids and versions affected by revisions.
1902
:param revisions: an iterable containing revision ids.
1903
:param _inv_weave: The inventory weave from this repository or None.
1904
If None, the inventory weave will be opened automatically.
1905
:return: a dictionary mapping altered file-ids to an iterable of
1906
revision_ids. Each altered file-ids has the exact revision_ids that
1907
altered it listed explicitly.
1909
selected_keys = set((revid,) for revid in revision_ids)
1910
w = _inv_weave or self.inventories
1911
pb = ui.ui_factory.nested_progress_bar()
1913
return self._find_file_ids_from_xml_inventory_lines(
1914
w.iter_lines_added_or_present_in_keys(
1915
selected_keys, pb=pb),
1920
def iter_files_bytes(self, desired_files):
1921
"""Iterate through file versions.
1923
Files will not necessarily be returned in the order they occur in
1924
desired_files. No specific order is guaranteed.
1926
Yields pairs of identifier, bytes_iterator. identifier is an opaque
1927
value supplied by the caller as part of desired_files. It should
1928
uniquely identify the file version in the caller's context. (Examples:
1929
an index number or a TreeTransform trans_id.)
1931
bytes_iterator is an iterable of bytestrings for the file. The
1932
kind of iterable and length of the bytestrings are unspecified, but for
1933
this implementation, it is a list of bytes produced by
1934
VersionedFile.get_record_stream().
1936
:param desired_files: a list of (file_id, revision_id, identifier)
1940
for file_id, revision_id, callable_data in desired_files:
1941
text_keys[(file_id, revision_id)] = callable_data
1942
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
1943
if record.storage_kind == 'absent':
1944
raise errors.RevisionNotPresent(record.key, self)
1945
yield text_keys[record.key], record.get_bytes_as('chunked')
1947
def _generate_text_key_index(self, text_key_references=None,
1949
"""Generate a new text key index for the repository.
1951
This is an expensive function that will take considerable time to run.
1953
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
1954
list of parents, also text keys. When a given key has no parents,
1955
the parents list will be [NULL_REVISION].
1957
# All revisions, to find inventory parents.
1958
if ancestors is None:
1959
graph = self.get_graph()
1960
ancestors = graph.get_parent_map(self.all_revision_ids())
1961
if text_key_references is None:
1962
text_key_references = self.find_text_key_references()
1963
pb = ui.ui_factory.nested_progress_bar()
1965
return self._do_generate_text_key_index(ancestors,
1966
text_key_references, pb)
1970
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
1971
"""Helper for _generate_text_key_index to avoid deep nesting."""
1972
revision_order = tsort.topo_sort(ancestors)
1973
invalid_keys = set()
1975
for revision_id in revision_order:
1976
revision_keys[revision_id] = set()
1977
text_count = len(text_key_references)
1978
# a cache of the text keys to allow reuse; costs a dict of all the
1979
# keys, but saves a 2-tuple for every child of a given key.
1981
for text_key, valid in text_key_references.iteritems():
1983
invalid_keys.add(text_key)
1985
revision_keys[text_key[1]].add(text_key)
1986
text_key_cache[text_key] = text_key
1987
del text_key_references
1989
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
1990
NULL_REVISION = _mod_revision.NULL_REVISION
1991
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
1992
# too small for large or very branchy trees. However, for 55K path
1993
# trees, it would be easy to use too much memory trivially. Ideally we
1994
# could gauge this by looking at available real memory etc, but this is
1995
# always a tricky proposition.
1996
inventory_cache = lru_cache.LRUCache(10)
1997
batch_size = 10 # should be ~150MB on a 55K path tree
1998
batch_count = len(revision_order) / batch_size + 1
2000
pb.update("Calculating text parents", processed_texts, text_count)
2001
for offset in xrange(batch_count):
2002
to_query = revision_order[offset * batch_size:(offset + 1) *
2006
for rev_tree in self.revision_trees(to_query):
2007
revision_id = rev_tree.get_revision_id()
2008
parent_ids = ancestors[revision_id]
2009
for text_key in revision_keys[revision_id]:
2010
pb.update("Calculating text parents", processed_texts)
2011
processed_texts += 1
2012
candidate_parents = []
2013
for parent_id in parent_ids:
2014
parent_text_key = (text_key[0], parent_id)
2016
check_parent = parent_text_key not in \
2017
revision_keys[parent_id]
2019
# the parent parent_id is a ghost:
2020
check_parent = False
2021
# truncate the derived graph against this ghost.
2022
parent_text_key = None
2024
# look at the parent commit details inventories to
2025
# determine possible candidates in the per file graph.
2028
inv = inventory_cache[parent_id]
2030
inv = self.revision_tree(parent_id).inventory
2031
inventory_cache[parent_id] = inv
2033
parent_entry = inv[text_key[0]]
2034
except (KeyError, errors.NoSuchId):
2036
if parent_entry is not None:
2038
text_key[0], parent_entry.revision)
2040
parent_text_key = None
2041
if parent_text_key is not None:
2042
candidate_parents.append(
2043
text_key_cache[parent_text_key])
2044
parent_heads = text_graph.heads(candidate_parents)
2045
new_parents = list(parent_heads)
2046
new_parents.sort(key=lambda x:candidate_parents.index(x))
2047
if new_parents == []:
2048
new_parents = [NULL_REVISION]
2049
text_index[text_key] = new_parents
2051
for text_key in invalid_keys:
2052
text_index[text_key] = [NULL_REVISION]
2055
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2056
"""Get an iterable listing the keys of all the data introduced by a set
2059
The keys will be ordered so that the corresponding items can be safely
2060
fetched and inserted in that order.
2062
:returns: An iterable producing tuples of (knit-kind, file-id,
2063
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2064
'revisions'. file-id is None unless knit-kind is 'file'.
2066
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2069
for result in self._find_non_file_keys_to_fetch(revision_ids):
2072
def _find_file_keys_to_fetch(self, revision_ids, pb):
2073
# XXX: it's a bit weird to control the inventory weave caching in this
2074
# generator. Ideally the caching would be done in fetch.py I think. Or
2075
# maybe this generator should explicitly have the contract that it
2076
# should not be iterated until the previously yielded item has been
2078
inv_w = self.inventories
2080
# file ids that changed
2081
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2083
num_file_ids = len(file_ids)
2084
for file_id, altered_versions in file_ids.iteritems():
2086
pb.update("fetch texts", count, num_file_ids)
2088
yield ("file", file_id, altered_versions)
2090
def _find_non_file_keys_to_fetch(self, revision_ids):
2092
yield ("inventory", None, revision_ids)
2095
# XXX: Note ATM no callers actually pay attention to this return
2096
# instead they just use the list of revision ids and ignore
2097
# missing sigs. Consider removing this work entirely
2098
revisions_with_signatures = set(self.signatures.get_parent_map(
2099
[(r,) for r in revision_ids]))
2100
revisions_with_signatures = set(
2101
[r for (r,) in revisions_with_signatures])
2102
revisions_with_signatures.intersection_update(revision_ids)
2103
yield ("signatures", None, revisions_with_signatures)
2106
yield ("revisions", None, revision_ids)
2109
def get_inventory(self, revision_id):
2110
"""Get Inventory object by revision id."""
2111
return self.iter_inventories([revision_id]).next()
2113
def iter_inventories(self, revision_ids):
2114
"""Get many inventories by revision_ids.
2116
This will buffer some or all of the texts used in constructing the
2117
inventories in memory, but will only parse a single inventory at a
2120
:param revision_ids: The expected revision ids of the inventories.
2121
:return: An iterator of inventories.
2123
if ((None in revision_ids)
2124
or (_mod_revision.NULL_REVISION in revision_ids)):
2125
raise ValueError('cannot get null revision inventory')
2126
return self._iter_inventories(revision_ids)
2128
def _iter_inventories(self, revision_ids):
2129
"""single-document based inventory iteration."""
2130
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2131
yield self.deserialise_inventory(revision_id, text)
2133
def _iter_inventory_xmls(self, revision_ids):
2134
keys = [(revision_id,) for revision_id in revision_ids]
2135
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2137
for record in stream:
2138
if record.storage_kind != 'absent':
2139
text_chunks[record.key] = record.get_bytes_as('chunked')
2141
raise errors.NoSuchRevision(self, record.key)
2143
chunks = text_chunks.pop(key)
2144
yield ''.join(chunks), key[-1]
2146
def deserialise_inventory(self, revision_id, xml):
2147
"""Transform the xml into an inventory object.
2149
:param revision_id: The expected revision id of the inventory.
2150
:param xml: A serialised inventory.
2152
result = self._serializer.read_inventory_from_string(xml, revision_id,
2153
entry_cache=self._inventory_entry_cache)
2154
if result.revision_id != revision_id:
2155
raise AssertionError('revision id mismatch %s != %s' % (
2156
result.revision_id, revision_id))
2159
def serialise_inventory(self, inv):
2160
return self._serializer.write_inventory_to_string(inv)
2162
def _serialise_inventory_to_lines(self, inv):
2163
return self._serializer.write_inventory_to_lines(inv)
2165
def get_serializer_format(self):
2166
return self._serializer.format_num
2169
def get_inventory_xml(self, revision_id):
2170
"""Get inventory XML as a file object."""
2171
texts = self._iter_inventory_xmls([revision_id])
2173
text, revision_id = texts.next()
2174
except StopIteration:
2175
raise errors.HistoryMissing(self, 'inventory', revision_id)
2179
def get_inventory_sha1(self, revision_id):
2180
"""Return the sha1 hash of the inventory entry
2182
return self.get_revision(revision_id).inventory_sha1
2184
def iter_reverse_revision_history(self, revision_id):
2185
"""Iterate backwards through revision ids in the lefthand history
2187
:param revision_id: The revision id to start with. All its lefthand
2188
ancestors will be traversed.
2190
graph = self.get_graph()
2191
next_id = revision_id
2193
if next_id in (None, _mod_revision.NULL_REVISION):
2196
# Note: The following line may raise KeyError in the event of
2197
# truncated history. We decided not to have a try:except:raise
2198
# RevisionNotPresent here until we see a use for it, because of the
2199
# cost in an inner loop that is by its very nature O(history).
2200
# Robert Collins 20080326
2201
parents = graph.get_parent_map([next_id])[next_id]
2202
if len(parents) == 0:
2205
next_id = parents[0]
2208
def get_revision_inventory(self, revision_id):
2209
"""Return inventory of a past revision."""
2210
# TODO: Unify this with get_inventory()
2211
# bzr 0.0.6 and later imposes the constraint that the inventory_id
2212
# must be the same as its revision, so this is trivial.
2213
if revision_id is None:
2214
# This does not make sense: if there is no revision,
2215
# then it is the current tree inventory surely ?!
2216
# and thus get_root_id() is something that looks at the last
2217
# commit on the branch, and the get_root_id is an inventory check.
2218
raise NotImplementedError
2219
# return Inventory(self.get_root_id())
2221
return self.get_inventory(revision_id)
2223
def is_shared(self):
2224
"""Return True if this repository is flagged as a shared repository."""
2225
raise NotImplementedError(self.is_shared)
2228
def reconcile(self, other=None, thorough=False):
2229
"""Reconcile this repository."""
2230
from bzrlib.reconcile import RepoReconciler
2231
reconciler = RepoReconciler(self, thorough=thorough)
2232
reconciler.reconcile()
2235
def _refresh_data(self):
2236
"""Helper called from lock_* to ensure coherency with disk.
2238
The default implementation does nothing; it is however possible
2239
for repositories to maintain loaded indices across multiple locks
2240
by checking inside their implementation of this method to see
2241
whether their indices are still valid. This depends of course on
2242
the disk format being validatable in this manner. This method is
2243
also called by the refresh_data() public interface to cause a refresh
2244
to occur while in a write lock so that data inserted by a smart server
2245
push operation is visible on the client's instance of the physical
2250
def revision_tree(self, revision_id):
2251
"""Return Tree for a revision on this branch.
2253
`revision_id` may be NULL_REVISION for the empty tree revision.
2255
revision_id = _mod_revision.ensure_null(revision_id)
2256
# TODO: refactor this to use an existing revision object
2257
# so we don't need to read it in twice.
2258
if revision_id == _mod_revision.NULL_REVISION:
2259
return RevisionTree(self, Inventory(root_id=None),
2260
_mod_revision.NULL_REVISION)
2262
inv = self.get_revision_inventory(revision_id)
2263
return RevisionTree(self, inv, revision_id)
2265
def revision_trees(self, revision_ids):
2266
"""Return Trees for revisions in this repository.
2268
:param revision_ids: a sequence of revision-ids;
2269
a revision-id may not be None or 'null:'
2271
inventories = self.iter_inventories(revision_ids)
2272
for inv in inventories:
2273
yield RevisionTree(self, inv, inv.revision_id)
2275
def _filtered_revision_trees(self, revision_ids, file_ids):
2276
"""Return Tree for a revision on this branch with only some files.
2278
:param revision_ids: a sequence of revision-ids;
2279
a revision-id may not be None or 'null:'
2280
:param file_ids: if not None, the result is filtered
2281
so that only those file-ids, their parents and their
2282
children are included.
2284
inventories = self.iter_inventories(revision_ids)
2285
for inv in inventories:
2286
# Should we introduce a FilteredRevisionTree class rather
2287
# than pre-filter the inventory here?
2288
filtered_inv = inv.filter(file_ids)
2289
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2292
def get_ancestry(self, revision_id, topo_sorted=True):
2293
"""Return a list of revision-ids integrated by a revision.
2295
The first element of the list is always None, indicating the origin
2296
revision. This might change when we have history horizons, or
2297
perhaps we should have a new API.
2299
This is topologically sorted.
2301
if _mod_revision.is_null(revision_id):
2303
if not self.has_revision(revision_id):
2304
raise errors.NoSuchRevision(self, revision_id)
2305
graph = self.get_graph()
2307
search = graph._make_breadth_first_searcher([revision_id])
2310
found, ghosts = search.next_with_ghosts()
2311
except StopIteration:
2314
if _mod_revision.NULL_REVISION in keys:
2315
keys.remove(_mod_revision.NULL_REVISION)
2317
parent_map = graph.get_parent_map(keys)
2318
keys = tsort.topo_sort(parent_map)
2319
return [None] + list(keys)
2322
"""Compress the data within the repository.
2324
This operation only makes sense for some repository types. For other
2325
types it should be a no-op that just returns.
2327
This stub method does not require a lock, but subclasses should use
2328
@needs_write_lock as this is a long running call its reasonable to
2329
implicitly lock for the user.
2332
def get_transaction(self):
2333
return self.control_files.get_transaction()
2335
def get_parent_map(self, revision_ids):
2336
"""See graph._StackedParentsProvider.get_parent_map"""
2337
# revisions index works in keys; this just works in revisions
2338
# therefore wrap and unwrap
2341
for revision_id in revision_ids:
2342
if revision_id == _mod_revision.NULL_REVISION:
2343
result[revision_id] = ()
2344
elif revision_id is None:
2345
raise ValueError('get_parent_map(None) is not valid')
2347
query_keys.append((revision_id ,))
2348
for ((revision_id,), parent_keys) in \
2349
self.revisions.get_parent_map(query_keys).iteritems():
2351
result[revision_id] = tuple(parent_revid
2352
for (parent_revid,) in parent_keys)
2354
result[revision_id] = (_mod_revision.NULL_REVISION,)
2357
def _make_parents_provider(self):
2360
def get_graph(self, other_repository=None):
2361
"""Return the graph walker for this repository format"""
2362
parents_provider = self._make_parents_provider()
2363
if (other_repository is not None and
2364
not self.has_same_location(other_repository)):
2365
parents_provider = graph._StackedParentsProvider(
2366
[parents_provider, other_repository._make_parents_provider()])
2367
return graph.Graph(parents_provider)
2369
def _get_versioned_file_checker(self, text_key_references=None):
2370
"""Return an object suitable for checking versioned files.
2372
:param text_key_references: if non-None, an already built
2373
dictionary mapping text keys ((fileid, revision_id) tuples)
2374
to whether they were referred to by the inventory of the
2375
revision_id that they contain. If None, this will be
2378
return _VersionedFileChecker(self,
2379
text_key_references=text_key_references)
2381
def revision_ids_to_search_result(self, result_set):
2382
"""Convert a set of revision ids to a graph SearchResult."""
2383
result_parents = set()
2384
for parents in self.get_graph().get_parent_map(
2385
result_set).itervalues():
2386
result_parents.update(parents)
2387
included_keys = result_set.intersection(result_parents)
2388
start_keys = result_set.difference(included_keys)
2389
exclude_keys = result_parents.difference(result_set)
2390
result = graph.SearchResult(start_keys, exclude_keys,
2391
len(result_set), result_set)
2395
def set_make_working_trees(self, new_value):
2396
"""Set the policy flag for making working trees when creating branches.
2398
This only applies to branches that use this repository.
2400
The default is 'True'.
2401
:param new_value: True to restore the default, False to disable making
2404
raise NotImplementedError(self.set_make_working_trees)
2406
def make_working_trees(self):
2407
"""Returns the policy for making working trees on new branches."""
2408
raise NotImplementedError(self.make_working_trees)
2411
def sign_revision(self, revision_id, gpg_strategy):
2412
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2413
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2416
def has_signature_for_revision_id(self, revision_id):
2417
"""Query for a revision signature for revision_id in the repository."""
2418
if not self.has_revision(revision_id):
2419
raise errors.NoSuchRevision(self, revision_id)
2420
sig_present = (1 == len(
2421
self.signatures.get_parent_map([(revision_id,)])))
2425
def get_signature_text(self, revision_id):
2426
"""Return the text for a signature."""
2427
stream = self.signatures.get_record_stream([(revision_id,)],
2429
record = stream.next()
2430
if record.storage_kind == 'absent':
2431
raise errors.NoSuchRevision(self, revision_id)
2432
return record.get_bytes_as('fulltext')
2435
def check(self, revision_ids=None):
2436
"""Check consistency of all history of given revision_ids.
2438
Different repository implementations should override _check().
2440
:param revision_ids: A non-empty list of revision_ids whose ancestry
2441
will be checked. Typically the last revision_id of a branch.
2443
return self._check(revision_ids)
2445
def _check(self, revision_ids):
2446
result = check.Check(self)
2450
def _warn_if_deprecated(self):
2451
global _deprecation_warning_done
2452
if _deprecation_warning_done:
2454
_deprecation_warning_done = True
2455
warning("Format %s for %s is deprecated - please use 'bzr upgrade' to get better performance"
2456
% (self._format, self.bzrdir.transport.base))
2458
def supports_rich_root(self):
2459
return self._format.rich_root_data
2461
def _check_ascii_revisionid(self, revision_id, method):
2462
"""Private helper for ascii-only repositories."""
2463
# weave repositories refuse to store revisionids that are non-ascii.
2464
if revision_id is not None:
2465
# weaves require ascii revision ids.
2466
if isinstance(revision_id, unicode):
2468
revision_id.encode('ascii')
2469
except UnicodeEncodeError:
2470
raise errors.NonAsciiRevisionId(method, self)
2473
revision_id.decode('ascii')
2474
except UnicodeDecodeError:
2475
raise errors.NonAsciiRevisionId(method, self)
2477
def revision_graph_can_have_wrong_parents(self):
2478
"""Is it possible for this repository to have a revision graph with
2481
If True, then this repository must also implement
2482
_find_inconsistent_revision_parents so that check and reconcile can
2483
check for inconsistencies before proceeding with other checks that may
2484
depend on the revision index being consistent.
2486
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2489
# remove these delegates a while after bzr 0.15
2490
def __make_delegated(name, from_module):
2491
def _deprecated_repository_forwarder():
2492
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2493
% (name, from_module),
2496
m = __import__(from_module, globals(), locals(), [name])
2498
return getattr(m, name)
2499
except AttributeError:
2500
raise AttributeError('module %s has no name %s'
2502
globals()[name] = _deprecated_repository_forwarder
2505
'AllInOneRepository',
2506
'WeaveMetaDirRepository',
2507
'PreSplitOutRepositoryFormat',
2508
'RepositoryFormat4',
2509
'RepositoryFormat5',
2510
'RepositoryFormat6',
2511
'RepositoryFormat7',
2513
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2517
'RepositoryFormatKnit',
2518
'RepositoryFormatKnit1',
2520
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2523
def install_revision(repository, rev, revision_tree):
2524
"""Install all revision data into a repository."""
2525
install_revisions(repository, [(rev, revision_tree, None)])
2528
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2529
"""Install all revision data into a repository.
2531
Accepts an iterable of revision, tree, signature tuples. The signature
2534
repository.start_write_group()
2536
inventory_cache = lru_cache.LRUCache(10)
2537
for n, (revision, revision_tree, signature) in enumerate(iterable):
2538
_install_revision(repository, revision, revision_tree, signature,
2541
pb.update('Transferring revisions', n + 1, num_revisions)
2543
repository.abort_write_group()
2546
repository.commit_write_group()
2549
def _install_revision(repository, rev, revision_tree, signature,
2551
"""Install all revision data into a repository."""
2552
present_parents = []
2554
for p_id in rev.parent_ids:
2555
if repository.has_revision(p_id):
2556
present_parents.append(p_id)
2557
parent_trees[p_id] = repository.revision_tree(p_id)
2559
parent_trees[p_id] = repository.revision_tree(
2560
_mod_revision.NULL_REVISION)
2562
inv = revision_tree.inventory
2563
entries = inv.iter_entries()
2564
# backwards compatibility hack: skip the root id.
2565
if not repository.supports_rich_root():
2566
path, root = entries.next()
2567
if root.revision != rev.revision_id:
2568
raise errors.IncompatibleRevision(repr(repository))
2570
for path, ie in entries:
2571
text_keys[(ie.file_id, ie.revision)] = ie
2572
text_parent_map = repository.texts.get_parent_map(text_keys)
2573
missing_texts = set(text_keys) - set(text_parent_map)
2574
# Add the texts that are not already present
2575
for text_key in missing_texts:
2576
ie = text_keys[text_key]
2578
# FIXME: TODO: The following loop overlaps/duplicates that done by
2579
# commit to determine parents. There is a latent/real bug here where
2580
# the parents inserted are not those commit would do - in particular
2581
# they are not filtered by heads(). RBC, AB
2582
for revision, tree in parent_trees.iteritems():
2583
if ie.file_id not in tree:
2585
parent_id = tree.inventory[ie.file_id].revision
2586
if parent_id in text_parents:
2588
text_parents.append((ie.file_id, parent_id))
2589
lines = revision_tree.get_file(ie.file_id).readlines()
2590
repository.texts.add_lines(text_key, text_parents, lines)
2592
# install the inventory
2593
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2594
# Cache this inventory
2595
inventory_cache[rev.revision_id] = inv
2597
basis_inv = inventory_cache[rev.parent_ids[0]]
2599
repository.add_inventory(rev.revision_id, inv, present_parents)
2601
delta = inv._make_delta(basis_inv)
2602
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2603
rev.revision_id, present_parents)
2605
repository.add_inventory(rev.revision_id, inv, present_parents)
2606
except errors.RevisionAlreadyPresent:
2608
if signature is not None:
2609
repository.add_signature_text(rev.revision_id, signature)
2610
repository.add_revision(rev.revision_id, rev, inv)
2613
class MetaDirRepository(Repository):
2614
"""Repositories in the new meta-dir layout.
2616
:ivar _transport: Transport for access to repository control files,
2617
typically pointing to .bzr/repository.
2620
def __init__(self, _format, a_bzrdir, control_files):
2621
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2622
self._transport = control_files._transport
2624
def is_shared(self):
2625
"""Return True if this repository is flagged as a shared repository."""
2626
return self._transport.has('shared-storage')
2629
def set_make_working_trees(self, new_value):
2630
"""Set the policy flag for making working trees when creating branches.
2632
This only applies to branches that use this repository.
2634
The default is 'True'.
2635
:param new_value: True to restore the default, False to disable making
2640
self._transport.delete('no-working-trees')
2641
except errors.NoSuchFile:
2644
self._transport.put_bytes('no-working-trees', '',
2645
mode=self.bzrdir._get_file_mode())
2647
def make_working_trees(self):
2648
"""Returns the policy for making working trees on new branches."""
2649
return not self._transport.has('no-working-trees')
2652
class MetaDirVersionedFileRepository(MetaDirRepository):
2653
"""Repositories in a meta-dir, that work via versioned file objects."""
2655
def __init__(self, _format, a_bzrdir, control_files):
2656
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2660
network_format_registry = registry.FormatRegistry()
2661
"""Registry of formats indexed by their network name.
2663
The network name for a repository format is an identifier that can be used when
2664
referring to formats with smart server operations. See
2665
RepositoryFormat.network_name() for more detail.
2669
format_registry = registry.FormatRegistry(network_format_registry)
2670
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
2672
This can contain either format instances themselves, or classes/factories that
2673
can be called to obtain one.
2677
#####################################################################
2678
# Repository Formats
2680
class RepositoryFormat(object):
2681
"""A repository format.
2683
Formats provide four things:
2684
* An initialization routine to construct repository data on disk.
2685
* a optional format string which is used when the BzrDir supports
2687
* an open routine which returns a Repository instance.
2688
* A network name for referring to the format in smart server RPC
2691
There is one and only one Format subclass for each on-disk format. But
2692
there can be one Repository subclass that is used for several different
2693
formats. The _format attribute on a Repository instance can be used to
2694
determine the disk format.
2696
Formats are placed in a registry by their format string for reference
2697
during opening. These should be subclasses of RepositoryFormat for
2700
Once a format is deprecated, just deprecate the initialize and open
2701
methods on the format class. Do not deprecate the object, as the
2702
object may be created even when a repository instnace hasn't been
2705
Common instance attributes:
2706
_matchingbzrdir - the bzrdir format that the repository format was
2707
originally written to work with. This can be used if manually
2708
constructing a bzrdir and repository, or more commonly for test suite
2712
# Set to True or False in derived classes. True indicates that the format
2713
# supports ghosts gracefully.
2714
supports_ghosts = None
2715
# Can this repository be given external locations to lookup additional
2716
# data. Set to True or False in derived classes.
2717
supports_external_lookups = None
2718
# Does this format support CHK bytestring lookups. Set to True or False in
2720
supports_chks = None
2721
# Should commit add an inventory, or an inventory delta to the repository.
2722
_commit_inv_deltas = True
2723
# What order should fetch operations request streams in?
2724
# The default is unordered as that is the cheapest for an origin to
2726
_fetch_order = 'unordered'
2727
# Does this repository format use deltas that can be fetched as-deltas ?
2728
# (E.g. knits, where the knit deltas can be transplanted intact.
2729
# We default to False, which will ensure that enough data to get
2730
# a full text out of any fetch stream will be grabbed.
2731
_fetch_uses_deltas = False
2732
# Should fetch trigger a reconcile after the fetch? Only needed for
2733
# some repository formats that can suffer internal inconsistencies.
2734
_fetch_reconcile = False
2735
# Does this format have < O(tree_size) delta generation. Used to hint what
2736
# code path for commit, amongst other things.
2740
return "<%s>" % self.__class__.__name__
2742
def __eq__(self, other):
2743
# format objects are generally stateless
2744
return isinstance(other, self.__class__)
2746
def __ne__(self, other):
2747
return not self == other
2750
def find_format(klass, a_bzrdir):
2751
"""Return the format for the repository object in a_bzrdir.
2753
This is used by bzr native formats that have a "format" file in
2754
the repository. Other methods may be used by different types of
2758
transport = a_bzrdir.get_repository_transport(None)
2759
format_string = transport.get("format").read()
2760
return format_registry.get(format_string)
2761
except errors.NoSuchFile:
2762
raise errors.NoRepositoryPresent(a_bzrdir)
2764
raise errors.UnknownFormatError(format=format_string,
2768
def register_format(klass, format):
2769
format_registry.register(format.get_format_string(), format)
2772
def unregister_format(klass, format):
2773
format_registry.remove(format.get_format_string())
2776
def get_default_format(klass):
2777
"""Return the current default format."""
2778
from bzrlib import bzrdir
2779
return bzrdir.format_registry.make_bzrdir('default').repository_format
2781
def get_format_string(self):
2782
"""Return the ASCII format string that identifies this format.
2784
Note that in pre format ?? repositories the format string is
2785
not permitted nor written to disk.
2787
raise NotImplementedError(self.get_format_string)
2789
def get_format_description(self):
2790
"""Return the short description for this format."""
2791
raise NotImplementedError(self.get_format_description)
2793
# TODO: this shouldn't be in the base class, it's specific to things that
2794
# use weaves or knits -- mbp 20070207
2795
def _get_versioned_file_store(self,
2800
versionedfile_class=None,
2801
versionedfile_kwargs={},
2803
if versionedfile_class is None:
2804
versionedfile_class = self._versionedfile_class
2805
weave_transport = control_files._transport.clone(name)
2806
dir_mode = control_files._dir_mode
2807
file_mode = control_files._file_mode
2808
return VersionedFileStore(weave_transport, prefixed=prefixed,
2810
file_mode=file_mode,
2811
versionedfile_class=versionedfile_class,
2812
versionedfile_kwargs=versionedfile_kwargs,
2815
def initialize(self, a_bzrdir, shared=False):
2816
"""Initialize a repository of this format in a_bzrdir.
2818
:param a_bzrdir: The bzrdir to put the new repository in it.
2819
:param shared: The repository should be initialized as a sharable one.
2820
:returns: The new repository object.
2822
This may raise UninitializableFormat if shared repository are not
2823
compatible the a_bzrdir.
2825
raise NotImplementedError(self.initialize)
2827
def is_supported(self):
2828
"""Is this format supported?
2830
Supported formats must be initializable and openable.
2831
Unsupported formats may not support initialization or committing or
2832
some other features depending on the reason for not being supported.
2836
def network_name(self):
2837
"""A simple byte string uniquely identifying this format for RPC calls.
2839
MetaDir repository formats use their disk format string to identify the
2840
repository over the wire. All in one formats such as bzr < 0.8, and
2841
foreign formats like svn/git and hg should use some marker which is
2842
unique and immutable.
2844
raise NotImplementedError(self.network_name)
2846
def check_conversion_target(self, target_format):
2847
raise NotImplementedError(self.check_conversion_target)
2849
def open(self, a_bzrdir, _found=False):
2850
"""Return an instance of this format for the bzrdir a_bzrdir.
2852
_found is a private parameter, do not use it.
2854
raise NotImplementedError(self.open)
2857
class MetaDirRepositoryFormat(RepositoryFormat):
2858
"""Common base class for the new repositories using the metadir layout."""
2860
rich_root_data = False
2861
supports_tree_reference = False
2862
supports_external_lookups = False
2865
def _matchingbzrdir(self):
2866
matching = bzrdir.BzrDirMetaFormat1()
2867
matching.repository_format = self
2871
super(MetaDirRepositoryFormat, self).__init__()
2873
def _create_control_files(self, a_bzrdir):
2874
"""Create the required files and the initial control_files object."""
2875
# FIXME: RBC 20060125 don't peek under the covers
2876
# NB: no need to escape relative paths that are url safe.
2877
repository_transport = a_bzrdir.get_repository_transport(self)
2878
control_files = lockable_files.LockableFiles(repository_transport,
2879
'lock', lockdir.LockDir)
2880
control_files.create_lock()
2881
return control_files
2883
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
2884
"""Upload the initial blank content."""
2885
control_files = self._create_control_files(a_bzrdir)
2886
control_files.lock_write()
2887
transport = control_files._transport
2889
utf8_files += [('shared-storage', '')]
2891
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
2892
for (filename, content_stream) in files:
2893
transport.put_file(filename, content_stream,
2894
mode=a_bzrdir._get_file_mode())
2895
for (filename, content_bytes) in utf8_files:
2896
transport.put_bytes_non_atomic(filename, content_bytes,
2897
mode=a_bzrdir._get_file_mode())
2899
control_files.unlock()
2901
def network_name(self):
2902
"""Metadir formats have matching disk and network format strings."""
2903
return self.get_format_string()
2906
# Pre-0.8 formats that don't have a disk format string (because they are
2907
# versioned by the matching control directory). We use the control directories
2908
# disk format string as a key for the network_name because they meet the
2909
# constraints (simple string, unique, immmutable).
2910
network_format_registry.register_lazy(
2911
"Bazaar-NG branch, format 5\n",
2912
'bzrlib.repofmt.weaverepo',
2913
'RepositoryFormat5',
2915
network_format_registry.register_lazy(
2916
"Bazaar-NG branch, format 6\n",
2917
'bzrlib.repofmt.weaverepo',
2918
'RepositoryFormat6',
2921
# formats which have no format string are not discoverable or independently
2922
# creatable on disk, so are not registered in format_registry. They're
2923
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
2924
# needed, it's constructed directly by the BzrDir. Non-native formats where
2925
# the repository is not separately opened are similar.
2927
format_registry.register_lazy(
2928
'Bazaar-NG Repository format 7',
2929
'bzrlib.repofmt.weaverepo',
2933
format_registry.register_lazy(
2934
'Bazaar-NG Knit Repository Format 1',
2935
'bzrlib.repofmt.knitrepo',
2936
'RepositoryFormatKnit1',
2939
format_registry.register_lazy(
2940
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
2941
'bzrlib.repofmt.knitrepo',
2942
'RepositoryFormatKnit3',
2945
format_registry.register_lazy(
2946
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
2947
'bzrlib.repofmt.knitrepo',
2948
'RepositoryFormatKnit4',
2951
# Pack-based formats. There is one format for pre-subtrees, and one for
2952
# post-subtrees to allow ease of testing.
2953
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
2954
format_registry.register_lazy(
2955
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
2956
'bzrlib.repofmt.pack_repo',
2957
'RepositoryFormatKnitPack1',
2959
format_registry.register_lazy(
2960
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
2961
'bzrlib.repofmt.pack_repo',
2962
'RepositoryFormatKnitPack3',
2964
format_registry.register_lazy(
2965
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
2966
'bzrlib.repofmt.pack_repo',
2967
'RepositoryFormatKnitPack4',
2969
format_registry.register_lazy(
2970
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
2971
'bzrlib.repofmt.pack_repo',
2972
'RepositoryFormatKnitPack5',
2974
format_registry.register_lazy(
2975
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
2976
'bzrlib.repofmt.pack_repo',
2977
'RepositoryFormatKnitPack5RichRoot',
2979
format_registry.register_lazy(
2980
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
2981
'bzrlib.repofmt.pack_repo',
2982
'RepositoryFormatKnitPack5RichRootBroken',
2984
format_registry.register_lazy(
2985
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
2986
'bzrlib.repofmt.pack_repo',
2987
'RepositoryFormatKnitPack6',
2989
format_registry.register_lazy(
2990
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
2991
'bzrlib.repofmt.pack_repo',
2992
'RepositoryFormatKnitPack6RichRoot',
2995
# Development formats.
2996
# 1.7->1.8 go below here
2997
format_registry.register_lazy(
2998
"Bazaar development format 2 (needs bzr.dev from before 1.8)\n",
2999
'bzrlib.repofmt.pack_repo',
3000
'RepositoryFormatPackDevelopment2',
3002
format_registry.register_lazy(
3003
("Bazaar development format 2 with subtree support "
3004
"(needs bzr.dev from before 1.8)\n"),
3005
'bzrlib.repofmt.pack_repo',
3006
'RepositoryFormatPackDevelopment2Subtree',
3008
# 1.9->1.110 go below here
3009
format_registry.register_lazy(
3010
# merge-bbc-dev4-to-bzr.dev
3011
"Bazaar development format 5 (needs bzr.dev from before 1.13)\n",
3012
'bzrlib.repofmt.pack_repo',
3013
'RepositoryFormatPackDevelopment5',
3015
format_registry.register_lazy(
3016
# merge-bbc-dev4-to-bzr.dev
3017
("Bazaar development format 5 with subtree support"
3018
" (needs bzr.dev from before 1.13)\n"),
3019
'bzrlib.repofmt.pack_repo',
3020
'RepositoryFormatPackDevelopment5Subtree',
3022
format_registry.register_lazy(
3023
# merge-bbc-dev4-to-bzr.dev
3024
('Bazaar development format 5 hash 16'
3025
' (needs bzr.dev from before 1.13)\n'),
3026
'bzrlib.repofmt.pack_repo',
3027
'RepositoryFormatPackDevelopment5Hash16',
3029
format_registry.register_lazy(
3030
# merge-bbc-dev4-to-bzr.dev
3031
('Bazaar development format 5 hash 255'
3032
' (needs bzr.dev from before 1.13)\n'),
3033
'bzrlib.repofmt.pack_repo',
3034
'RepositoryFormatPackDevelopment5Hash255',
3036
format_registry.register_lazy(
3037
'Bazaar development format - hash16chk+gc rich-root (needs bzr.dev from 1.14)\n',
3038
'bzrlib.repofmt.groupcompress_repo',
3039
'RepositoryFormatPackGCCHK16',
3041
format_registry.register_lazy(
3042
'Bazaar development format - hash255chk+gc rich-root (needs bzr.dev from 1.14)\n',
3043
'bzrlib.repofmt.groupcompress_repo',
3044
'RepositoryFormatPackGCCHK255',
3046
format_registry.register_lazy(
3047
'Bazaar development format - hash255chk+gc rich-root bigpage (needs bzr.dev from 1.14)\n',
3048
'bzrlib.repofmt.groupcompress_repo',
3049
'RepositoryFormatPackGCCHK255Big',
3053
class InterRepository(InterObject):
3054
"""This class represents operations taking place between two repositories.
3056
Its instances have methods like copy_content and fetch, and contain
3057
references to the source and target repositories these operations can be
3060
Often we will provide convenience methods on 'repository' which carry out
3061
operations with another repository - they will always forward to
3062
InterRepository.get(other).method_name(parameters).
3065
_walk_to_common_revisions_batch_size = 50
3067
"""The available optimised InterRepository types."""
3070
def copy_content(self, revision_id=None):
3071
"""Make a complete copy of the content in self into destination.
3073
This is a destructive operation! Do not use it on existing
3076
:param revision_id: Only copy the content needed to construct
3077
revision_id and its parents.
3080
self.target.set_make_working_trees(self.source.make_working_trees())
3081
except NotImplementedError:
3083
self.target.fetch(self.source, revision_id=revision_id)
3086
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3088
"""Fetch the content required to construct revision_id.
3090
The content is copied from self.source to self.target.
3092
:param revision_id: if None all content is copied, if NULL_REVISION no
3094
:param pb: optional progress bar to use for progress reports. If not
3095
provided a default one will be created.
3098
from bzrlib.fetch import RepoFetcher
3099
f = RepoFetcher(to_repository=self.target,
3100
from_repository=self.source,
3101
last_revision=revision_id,
3102
fetch_spec=fetch_spec,
3103
pb=pb, find_ghosts=find_ghosts)
3105
def _walk_to_common_revisions(self, revision_ids):
3106
"""Walk out from revision_ids in source to revisions target has.
3108
:param revision_ids: The start point for the search.
3109
:return: A set of revision ids.
3111
target_graph = self.target.get_graph()
3112
revision_ids = frozenset(revision_ids)
3113
# Fast path for the case where all the revisions are already in the
3115
# (Although this does incur an extra round trip for the
3116
# fairly common case where the target doesn't already have the revision
3118
if set(target_graph.get_parent_map(revision_ids)) == revision_ids:
3119
return graph.SearchResult(revision_ids, set(), 0, set())
3120
missing_revs = set()
3121
source_graph = self.source.get_graph()
3122
# ensure we don't pay silly lookup costs.
3123
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3124
null_set = frozenset([_mod_revision.NULL_REVISION])
3125
searcher_exhausted = False
3129
# Iterate the searcher until we have enough next_revs
3130
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3132
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3133
next_revs.update(next_revs_part)
3134
ghosts.update(ghosts_part)
3135
except StopIteration:
3136
searcher_exhausted = True
3138
# If there are ghosts in the source graph, and the caller asked for
3139
# them, make sure that they are present in the target.
3140
# We don't care about other ghosts as we can't fetch them and
3141
# haven't been asked to.
3142
ghosts_to_check = set(revision_ids.intersection(ghosts))
3143
revs_to_get = set(next_revs).union(ghosts_to_check)
3145
have_revs = set(target_graph.get_parent_map(revs_to_get))
3146
# we always have NULL_REVISION present.
3147
have_revs = have_revs.union(null_set)
3148
# Check if the target is missing any ghosts we need.
3149
ghosts_to_check.difference_update(have_revs)
3151
# One of the caller's revision_ids is a ghost in both the
3152
# source and the target.
3153
raise errors.NoSuchRevision(
3154
self.source, ghosts_to_check.pop())
3155
missing_revs.update(next_revs - have_revs)
3156
# Because we may have walked past the original stop point, make
3157
# sure everything is stopped
3158
stop_revs = searcher.find_seen_ancestors(have_revs)
3159
searcher.stop_searching_any(stop_revs)
3160
if searcher_exhausted:
3162
return searcher.get_result()
3165
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3166
"""Return the revision ids that source has that target does not.
3168
:param revision_id: only return revision ids included by this
3170
:param find_ghosts: If True find missing revisions in deep history
3171
rather than just finding the surface difference.
3172
:return: A bzrlib.graph.SearchResult.
3174
# stop searching at found target revisions.
3175
if not find_ghosts and revision_id is not None:
3176
return self._walk_to_common_revisions([revision_id])
3177
# generic, possibly worst case, slow code path.
3178
target_ids = set(self.target.all_revision_ids())
3179
if revision_id is not None:
3180
source_ids = self.source.get_ancestry(revision_id)
3181
if source_ids[0] is not None:
3182
raise AssertionError()
3185
source_ids = self.source.all_revision_ids()
3186
result_set = set(source_ids).difference(target_ids)
3187
return self.source.revision_ids_to_search_result(result_set)
3190
def _same_model(source, target):
3191
"""True if source and target have the same data representation.
3193
Note: this is always called on the base class; overriding it in a
3194
subclass will have no effect.
3197
InterRepository._assert_same_model(source, target)
3199
except errors.IncompatibleRepositories, e:
3203
def _assert_same_model(source, target):
3204
"""Raise an exception if two repositories do not use the same model.
3206
if source.supports_rich_root() != target.supports_rich_root():
3207
raise errors.IncompatibleRepositories(source, target,
3208
"different rich-root support")
3209
if source._serializer != target._serializer:
3210
raise errors.IncompatibleRepositories(source, target,
3211
"different serializers")
3214
class InterSameDataRepository(InterRepository):
3215
"""Code for converting between repositories that represent the same data.
3217
Data format and model must match for this to work.
3221
def _get_repo_format_to_test(self):
3222
"""Repository format for testing with.
3224
InterSameData can pull from subtree to subtree and from non-subtree to
3225
non-subtree, so we test this with the richest repository format.
3227
from bzrlib.repofmt import knitrepo
3228
return knitrepo.RepositoryFormatKnit3()
3231
def is_compatible(source, target):
3232
return InterRepository._same_model(source, target)
3235
class InterWeaveRepo(InterSameDataRepository):
3236
"""Optimised code paths between Weave based repositories.
3238
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3239
implemented lazy inter-object optimisation.
3243
def _get_repo_format_to_test(self):
3244
from bzrlib.repofmt import weaverepo
3245
return weaverepo.RepositoryFormat7()
3248
def is_compatible(source, target):
3249
"""Be compatible with known Weave formats.
3251
We don't test for the stores being of specific types because that
3252
could lead to confusing results, and there is no need to be
3255
from bzrlib.repofmt.weaverepo import (
3261
return (isinstance(source._format, (RepositoryFormat5,
3263
RepositoryFormat7)) and
3264
isinstance(target._format, (RepositoryFormat5,
3266
RepositoryFormat7)))
3267
except AttributeError:
3271
def copy_content(self, revision_id=None):
3272
"""See InterRepository.copy_content()."""
3273
# weave specific optimised path:
3275
self.target.set_make_working_trees(self.source.make_working_trees())
3276
except (errors.RepositoryUpgradeRequired, NotImplemented):
3278
# FIXME do not peek!
3279
if self.source._transport.listable():
3280
pb = ui.ui_factory.nested_progress_bar()
3282
self.target.texts.insert_record_stream(
3283
self.source.texts.get_record_stream(
3284
self.source.texts.keys(), 'topological', False))
3285
pb.update('copying inventory', 0, 1)
3286
self.target.inventories.insert_record_stream(
3287
self.source.inventories.get_record_stream(
3288
self.source.inventories.keys(), 'topological', False))
3289
self.target.signatures.insert_record_stream(
3290
self.source.signatures.get_record_stream(
3291
self.source.signatures.keys(),
3293
self.target.revisions.insert_record_stream(
3294
self.source.revisions.get_record_stream(
3295
self.source.revisions.keys(),
3296
'topological', True))
3300
self.target.fetch(self.source, revision_id=revision_id)
3303
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3304
"""See InterRepository.missing_revision_ids()."""
3305
# we want all revisions to satisfy revision_id in source.
3306
# but we don't want to stat every file here and there.
3307
# we want then, all revisions other needs to satisfy revision_id
3308
# checked, but not those that we have locally.
3309
# so the first thing is to get a subset of the revisions to
3310
# satisfy revision_id in source, and then eliminate those that
3311
# we do already have.
3312
# this is slow on high latency connection to self, but as as this
3313
# disk format scales terribly for push anyway due to rewriting
3314
# inventory.weave, this is considered acceptable.
3316
if revision_id is not None:
3317
source_ids = self.source.get_ancestry(revision_id)
3318
if source_ids[0] is not None:
3319
raise AssertionError()
3322
source_ids = self.source._all_possible_ids()
3323
source_ids_set = set(source_ids)
3324
# source_ids is the worst possible case we may need to pull.
3325
# now we want to filter source_ids against what we actually
3326
# have in target, but don't try to check for existence where we know
3327
# we do not have a revision as that would be pointless.
3328
target_ids = set(self.target._all_possible_ids())
3329
possibly_present_revisions = target_ids.intersection(source_ids_set)
3330
actually_present_revisions = set(
3331
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3332
required_revisions = source_ids_set.difference(actually_present_revisions)
3333
if revision_id is not None:
3334
# we used get_ancestry to determine source_ids then we are assured all
3335
# revisions referenced are present as they are installed in topological order.
3336
# and the tip revision was validated by get_ancestry.
3337
result_set = required_revisions
3339
# if we just grabbed the possibly available ids, then
3340
# we only have an estimate of whats available and need to validate
3341
# that against the revision records.
3343
self.source._eliminate_revisions_not_present(required_revisions))
3344
return self.source.revision_ids_to_search_result(result_set)
3347
class InterKnitRepo(InterSameDataRepository):
3348
"""Optimised code paths between Knit based repositories."""
3351
def _get_repo_format_to_test(self):
3352
from bzrlib.repofmt import knitrepo
3353
return knitrepo.RepositoryFormatKnit1()
3356
def is_compatible(source, target):
3357
"""Be compatible with known Knit formats.
3359
We don't test for the stores being of specific types because that
3360
could lead to confusing results, and there is no need to be
3363
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3365
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3366
isinstance(target._format, RepositoryFormatKnit))
3367
except AttributeError:
3369
return are_knits and InterRepository._same_model(source, target)
3372
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3373
"""See InterRepository.missing_revision_ids()."""
3374
if revision_id is not None:
3375
source_ids = self.source.get_ancestry(revision_id)
3376
if source_ids[0] is not None:
3377
raise AssertionError()
3380
source_ids = self.source.all_revision_ids()
3381
source_ids_set = set(source_ids)
3382
# source_ids is the worst possible case we may need to pull.
3383
# now we want to filter source_ids against what we actually
3384
# have in target, but don't try to check for existence where we know
3385
# we do not have a revision as that would be pointless.
3386
target_ids = set(self.target.all_revision_ids())
3387
possibly_present_revisions = target_ids.intersection(source_ids_set)
3388
actually_present_revisions = set(
3389
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3390
required_revisions = source_ids_set.difference(actually_present_revisions)
3391
if revision_id is not None:
3392
# we used get_ancestry to determine source_ids then we are assured all
3393
# revisions referenced are present as they are installed in topological order.
3394
# and the tip revision was validated by get_ancestry.
3395
result_set = required_revisions
3397
# if we just grabbed the possibly available ids, then
3398
# we only have an estimate of whats available and need to validate
3399
# that against the revision records.
3401
self.source._eliminate_revisions_not_present(required_revisions))
3402
return self.source.revision_ids_to_search_result(result_set)
3405
class InterPackRepo(InterSameDataRepository):
3406
"""Optimised code paths between Pack based repositories."""
3409
def _get_repo_format_to_test(self):
3410
from bzrlib.repofmt import pack_repo
3411
return pack_repo.RepositoryFormatKnitPack1()
3414
def is_compatible(source, target):
3415
"""Be compatible with known Pack formats.
3417
We don't test for the stores being of specific types because that
3418
could lead to confusing results, and there is no need to be
3421
Do not support CHK based repositories at this point.
3423
from bzrlib.repofmt.pack_repo import RepositoryFormatPack
3424
# XXX: This format is scheduled for termination
3425
# from bzrlib.repofmt.groupcompress_repo import (
3426
# RepositoryFormatPackGCPlain,
3429
are_packs = (isinstance(source._format, RepositoryFormatPack) and
3430
isinstance(target._format, RepositoryFormatPack))
3431
except AttributeError:
3435
# if (isinstance(source._format, RepositoryFormatPackGCPlain)
3436
# or isinstance(target._format, RepositoryFormatPackGCPlain)):
3438
return (InterRepository._same_model(source, target) and
3439
not source._format.supports_chks)
3442
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3444
"""See InterRepository.fetch()."""
3445
if (len(self.source._fallback_repositories) > 0 or
3446
len(self.target._fallback_repositories) > 0):
3447
# The pack layer is not aware of fallback repositories, so when
3448
# fetching from a stacked repository or into a stacked repository
3449
# we use the generic fetch logic which uses the VersionedFiles
3450
# attributes on repository.
3451
from bzrlib.fetch import RepoFetcher
3452
fetcher = RepoFetcher(self.target, self.source, revision_id,
3453
pb, find_ghosts, fetch_spec=fetch_spec)
3454
if fetch_spec is not None:
3455
if len(list(fetch_spec.heads)) != 1:
3456
raise AssertionError(
3457
"InterPackRepo.fetch doesn't support "
3458
"fetching multiple heads yet.")
3459
revision_id = list(fetch_spec.heads)[0]
3461
if revision_id is None:
3463
# everything to do - use pack logic
3464
# to fetch from all packs to one without
3465
# inventory parsing etc, IFF nothing to be copied is in the target.
3467
source_revision_ids = frozenset(self.source.all_revision_ids())
3468
revision_ids = source_revision_ids - \
3469
frozenset(self.target.get_parent_map(source_revision_ids))
3470
revision_keys = [(revid,) for revid in revision_ids]
3471
index = self.target._pack_collection.revision_index.combined_index
3472
present_revision_ids = set(item[1][0] for item in
3473
index.iter_entries(revision_keys))
3474
revision_ids = set(revision_ids) - present_revision_ids
3475
# implementing the TODO will involve:
3476
# - detecting when all of a pack is selected
3477
# - avoiding as much as possible pre-selection, so the
3478
# more-core routines such as create_pack_from_packs can filter in
3479
# a just-in-time fashion. (though having a HEADS list on a
3480
# repository might make this a lot easier, because we could
3481
# sensibly detect 'new revisions' without doing a full index scan.
3482
elif _mod_revision.is_null(revision_id):
3487
revision_ids = self.search_missing_revision_ids(revision_id,
3488
find_ghosts=find_ghosts).get_keys()
3489
except errors.NoSuchRevision:
3490
raise errors.InstallFailed([revision_id])
3491
if len(revision_ids) == 0:
3493
return self._pack(self.source, self.target, revision_ids)
3495
def _pack(self, source, target, revision_ids):
3496
from bzrlib.repofmt.pack_repo import Packer
3497
packs = source._pack_collection.all_packs()
3498
pack = Packer(self.target._pack_collection, packs, '.fetch',
3499
revision_ids).pack()
3500
if pack is not None:
3501
self.target._pack_collection._save_pack_names()
3502
copied_revs = pack.get_revision_count()
3503
# Trigger an autopack. This may duplicate effort as we've just done
3504
# a pack creation, but for now it is simpler to think about as
3505
# 'upload data, then repack if needed'.
3506
self.target._pack_collection.autopack()
3507
return (copied_revs, [])
3512
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3513
"""See InterRepository.missing_revision_ids().
3515
:param find_ghosts: Find ghosts throughout the ancestry of
3518
if not find_ghosts and revision_id is not None:
3519
return self._walk_to_common_revisions([revision_id])
3520
elif revision_id is not None:
3521
# Find ghosts: search for revisions pointing from one repository to
3522
# the other, and vice versa, anywhere in the history of revision_id.
3523
graph = self.target.get_graph(other_repository=self.source)
3524
searcher = graph._make_breadth_first_searcher([revision_id])
3528
next_revs, ghosts = searcher.next_with_ghosts()
3529
except StopIteration:
3531
if revision_id in ghosts:
3532
raise errors.NoSuchRevision(self.source, revision_id)
3533
found_ids.update(next_revs)
3534
found_ids.update(ghosts)
3535
found_ids = frozenset(found_ids)
3536
# Double query here: should be able to avoid this by changing the
3537
# graph api further.
3538
result_set = found_ids - frozenset(
3539
self.target.get_parent_map(found_ids))
3541
source_ids = self.source.all_revision_ids()
3542
# source_ids is the worst possible case we may need to pull.
3543
# now we want to filter source_ids against what we actually
3544
# have in target, but don't try to check for existence where we know
3545
# we do not have a revision as that would be pointless.
3546
target_ids = set(self.target.all_revision_ids())
3547
result_set = set(source_ids).difference(target_ids)
3548
return self.source.revision_ids_to_search_result(result_set)
3551
class InterDifferingSerializer(InterRepository):
3554
def _get_repo_format_to_test(self):
3558
def is_compatible(source, target):
3559
"""Be compatible with Knit2 source and Knit3 target"""
3560
# This is redundant with format.check_conversion_target(), however that
3561
# raises an exception, and we just want to say "False" as in we won't
3562
# support converting between these formats.
3563
if source.supports_rich_root() and not target.supports_rich_root():
3565
if (source._format.supports_tree_reference
3566
and not target._format.supports_tree_reference):
3570
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3571
"""Get the best delta and base for this revision.
3573
:return: (basis_id, delta)
3575
possible_trees = [(parent_id, cache[parent_id])
3576
for parent_id in parent_ids
3577
if parent_id in cache]
3578
if len(possible_trees) == 0:
3579
# There either aren't any parents, or the parents aren't in the
3580
# cache, so just use the last converted tree
3581
possible_trees.append((basis_id, cache[basis_id]))
3583
for basis_id, basis_tree in possible_trees:
3584
delta = tree.inventory._make_delta(basis_tree.inventory)
3585
deltas.append((len(delta), basis_id, delta))
3587
return deltas[0][1:]
3589
def _get_parent_keys(self, root_key, parent_map):
3590
"""Get the parent keys for a given root id."""
3591
root_id, rev_id = root_key
3592
# Include direct parents of the revision, but only if they used
3595
for parent_id in parent_map[rev_id]:
3596
if parent_id == _mod_revision.NULL_REVISION:
3598
if parent_id not in self._revision_id_to_root_id:
3599
# We probably didn't read this revision, go spend the
3600
# extra effort to actually check
3602
tree = self.source.revision_tree(parent_id)
3603
except errors.NoSuchRevision:
3604
# Ghost, fill out _revision_id_to_root_id in case we
3605
# encounter this again.
3606
# But set parent_root_id to None since we don't really know
3607
parent_root_id = None
3609
parent_root_id = tree.get_root_id()
3610
self._revision_id_to_root_id[parent_id] = None
3612
parent_root_id = self._revision_id_to_root_id[parent_id]
3613
if root_id == parent_root_id or parent_root_id is None:
3614
parent_keys.append((root_id, parent_id))
3615
return tuple(parent_keys)
3617
def _new_root_data_stream(self, root_keys_to_create, parent_map):
3618
for root_key in root_keys_to_create:
3619
parent_keys = self._get_parent_keys(root_key, parent_map)
3620
yield versionedfile.FulltextContentFactory(root_key,
3621
parent_keys, None, '')
3623
def _fetch_batch(self, revision_ids, basis_id, cache):
3624
"""Fetch across a few revisions.
3626
:param revision_ids: The revisions to copy
3627
:param basis_id: The revision_id of a tree that must be in cache, used
3628
as a basis for delta when no other base is available
3629
:param cache: A cache of RevisionTrees that we can use.
3630
:return: The revision_id of the last converted tree. The RevisionTree
3631
for it will be in cache
3633
# Walk though all revisions; get inventory deltas, copy referenced
3634
# texts that delta references, insert the delta, revision and
3636
root_keys_to_create = set()
3639
pending_revisions = []
3640
parent_map = self.source.get_parent_map(revision_ids)
3641
# NB: This fails with dubious inventory data (when inv A has rev OLD
3642
# for file F, and in B, after A, has rev A for file F) when A and B are
3643
# in different groups.
3644
for tree in self.source.revision_trees(revision_ids):
3645
current_revision_id = tree.get_revision_id()
3646
parent_ids = parent_map.get(current_revision_id, ())
3647
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3649
if self._converting_to_rich_root:
3650
self._revision_id_to_root_id[current_revision_id] = \
3652
# Find text entries that need to be copied
3653
for old_path, new_path, file_id, entry in delta:
3654
if new_path is not None:
3657
if not self.target.supports_rich_root():
3658
# The target doesn't support rich root, so we don't
3661
if self._converting_to_rich_root:
3662
# This can't be copied normally, we have to insert
3664
root_keys_to_create.add((file_id, entry.revision))
3666
text_keys.add((file_id, entry.revision))
3667
revision = self.source.get_revision(current_revision_id)
3668
pending_deltas.append((basis_id, delta,
3669
current_revision_id, revision.parent_ids))
3670
pending_revisions.append(revision)
3671
cache[current_revision_id] = tree
3672
basis_id = current_revision_id
3674
from_texts = self.source.texts
3675
to_texts = self.target.texts
3676
if root_keys_to_create:
3677
root_stream = self._new_root_data_stream(root_keys_to_create,
3679
to_texts.insert_record_stream(root_stream)
3680
to_texts.insert_record_stream(from_texts.get_record_stream(
3681
text_keys, self.target._format._fetch_order,
3682
not self.target._format._fetch_uses_deltas))
3684
for delta in pending_deltas:
3685
self.target.add_inventory_by_delta(*delta)
3686
# insert signatures and revisions
3687
for revision in pending_revisions:
3689
signature = self.source.get_signature_text(
3690
revision.revision_id)
3691
self.target.add_signature_text(revision.revision_id,
3693
except errors.NoSuchRevision:
3695
self.target.add_revision(revision.revision_id, revision)
3698
def _fetch_all_revisions(self, revision_ids, pb):
3699
"""Fetch everything for the list of revisions.
3701
:param revision_ids: The list of revisions to fetch. Must be in
3703
:param pb: A ProgressBar
3706
basis_id, basis_tree = self._get_basis(revision_ids[0])
3708
cache = lru_cache.LRUCache(100)
3709
cache[basis_id] = basis_tree
3710
del basis_tree # We don't want to hang on to it here
3711
for offset in range(0, len(revision_ids), batch_size):
3712
self.target.start_write_group()
3714
pb.update('Transferring revisions', offset,
3716
batch = revision_ids[offset:offset+batch_size]
3717
basis_id = self._fetch_batch(batch, basis_id, cache)
3719
self.target.abort_write_group()
3722
self.target.commit_write_group()
3723
pb.update('Transferring revisions', len(revision_ids),
3727
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3729
"""See InterRepository.fetch()."""
3730
if fetch_spec is not None:
3731
raise AssertionError("Not implemented yet...")
3732
if (not self.source.supports_rich_root()
3733
and self.target.supports_rich_root()):
3734
self._converting_to_rich_root = True
3735
self._revision_id_to_root_id = {}
3737
self._converting_to_rich_root = False
3738
revision_ids = self.target.search_missing_revision_ids(self.source,
3739
revision_id, find_ghosts=find_ghosts).get_keys()
3740
if not revision_ids:
3742
revision_ids = tsort.topo_sort(
3743
self.source.get_graph().get_parent_map(revision_ids))
3744
if not revision_ids:
3746
# Walk though all revisions; get inventory deltas, copy referenced
3747
# texts that delta references, insert the delta, revision and
3749
first_rev = self.source.get_revision(revision_ids[0])
3751
my_pb = ui.ui_factory.nested_progress_bar()
3754
symbol_versioning.warn(
3755
symbol_versioning.deprecated_in((1, 14, 0))
3756
% "pb parameter to fetch()")
3759
self._fetch_all_revisions(revision_ids, pb)
3761
if my_pb is not None:
3763
return len(revision_ids), 0
3765
def _get_basis(self, first_revision_id):
3766
"""Get a revision and tree which exists in the target.
3768
This assumes that first_revision_id is selected for transmission
3769
because all other ancestors are already present. If we can't find an
3770
ancestor we fall back to NULL_REVISION since we know that is safe.
3772
:return: (basis_id, basis_tree)
3774
first_rev = self.source.get_revision(first_revision_id)
3776
basis_id = first_rev.parent_ids[0]
3777
# only valid as a basis if the target has it
3778
self.target.get_revision(basis_id)
3779
# Try to get a basis tree - if its a ghost it will hit the
3780
# NoSuchRevision case.
3781
basis_tree = self.source.revision_tree(basis_id)
3782
except (IndexError, errors.NoSuchRevision):
3783
basis_id = _mod_revision.NULL_REVISION
3784
basis_tree = self.source.revision_tree(basis_id)
3785
return basis_id, basis_tree
3788
InterRepository.register_optimiser(InterDifferingSerializer)
3789
InterRepository.register_optimiser(InterSameDataRepository)
3790
InterRepository.register_optimiser(InterWeaveRepo)
3791
InterRepository.register_optimiser(InterKnitRepo)
3792
InterRepository.register_optimiser(InterPackRepo)
3795
class CopyConverter(object):
3796
"""A repository conversion tool which just performs a copy of the content.
3798
This is slow but quite reliable.
3801
def __init__(self, target_format):
3802
"""Create a CopyConverter.
3804
:param target_format: The format the resulting repository should be.
3806
self.target_format = target_format
3808
def convert(self, repo, pb):
3809
"""Perform the conversion of to_convert, giving feedback via pb.
3811
:param to_convert: The disk object to convert.
3812
:param pb: a progress bar to use for progress information.
3817
# this is only useful with metadir layouts - separated repo content.
3818
# trigger an assertion if not such
3819
repo._format.get_format_string()
3820
self.repo_dir = repo.bzrdir
3821
self.step('Moving repository to repository.backup')
3822
self.repo_dir.transport.move('repository', 'repository.backup')
3823
backup_transport = self.repo_dir.transport.clone('repository.backup')
3824
repo._format.check_conversion_target(self.target_format)
3825
self.source_repo = repo._format.open(self.repo_dir,
3827
_override_transport=backup_transport)
3828
self.step('Creating new repository')
3829
converted = self.target_format.initialize(self.repo_dir,
3830
self.source_repo.is_shared())
3831
converted.lock_write()
3833
self.step('Copying content into repository.')
3834
self.source_repo.copy_content_into(converted)
3837
self.step('Deleting old repository content.')
3838
self.repo_dir.transport.delete_tree('repository.backup')
3839
self.pb.note('repository converted')
3841
def step(self, message):
3842
"""Update the pb by a step."""
3844
self.pb.update(message, self.count, self.total)
3856
def _unescaper(match, _map=_unescape_map):
3857
code = match.group(1)
3861
if not code.startswith('#'):
3863
return unichr(int(code[1:])).encode('utf8')
3869
def _unescape_xml(data):
3870
"""Unescape predefined XML entities in a string of data."""
3872
if _unescape_re is None:
3873
_unescape_re = re.compile('\&([^;]*);')
3874
return _unescape_re.sub(_unescaper, data)
3877
class _VersionedFileChecker(object):
3879
def __init__(self, repository, text_key_references=None):
3880
self.repository = repository
3881
self.text_index = self.repository._generate_text_key_index(
3882
text_key_references=text_key_references)
3884
def calculate_file_version_parents(self, text_key):
3885
"""Calculate the correct parents for a file version according to
3888
parent_keys = self.text_index[text_key]
3889
if parent_keys == [_mod_revision.NULL_REVISION]:
3891
return tuple(parent_keys)
3893
def check_file_version_parents(self, texts, progress_bar=None):
3894
"""Check the parents stored in a versioned file are correct.
3896
It also detects file versions that are not referenced by their
3897
corresponding revision's inventory.
3899
:returns: A tuple of (wrong_parents, dangling_file_versions).
3900
wrong_parents is a dict mapping {revision_id: (stored_parents,
3901
correct_parents)} for each revision_id where the stored parents
3902
are not correct. dangling_file_versions is a set of (file_id,
3903
revision_id) tuples for versions that are present in this versioned
3904
file, but not used by the corresponding inventory.
3907
self.file_ids = set([file_id for file_id, _ in
3908
self.text_index.iterkeys()])
3909
# text keys is now grouped by file_id
3910
n_weaves = len(self.file_ids)
3911
files_in_revisions = {}
3912
revisions_of_files = {}
3913
n_versions = len(self.text_index)
3914
progress_bar.update('loading text store', 0, n_versions)
3915
parent_map = self.repository.texts.get_parent_map(self.text_index)
3916
# On unlistable transports this could well be empty/error...
3917
text_keys = self.repository.texts.keys()
3918
unused_keys = frozenset(text_keys) - set(self.text_index)
3919
for num, key in enumerate(self.text_index.iterkeys()):
3920
if progress_bar is not None:
3921
progress_bar.update('checking text graph', num, n_versions)
3922
correct_parents = self.calculate_file_version_parents(key)
3924
knit_parents = parent_map[key]
3925
except errors.RevisionNotPresent:
3928
if correct_parents != knit_parents:
3929
wrong_parents[key] = (knit_parents, correct_parents)
3930
return wrong_parents, unused_keys
3933
def _old_get_graph(repository, revision_id):
3934
"""DO NOT USE. That is all. I'm serious."""
3935
graph = repository.get_graph()
3936
revision_graph = dict(((key, value) for key, value in
3937
graph.iter_ancestry([revision_id]) if value is not None))
3938
return _strip_NULL_ghosts(revision_graph)
3941
def _strip_NULL_ghosts(revision_graph):
3942
"""Also don't use this. more compatibility code for unmigrated clients."""
3943
# Filter ghosts, and null:
3944
if _mod_revision.NULL_REVISION in revision_graph:
3945
del revision_graph[_mod_revision.NULL_REVISION]
3946
for key, parents in revision_graph.items():
3947
revision_graph[key] = tuple(parent for parent in parents if parent
3949
return revision_graph
3952
class StreamSink(object):
3953
"""An object that can insert a stream into a repository.
3955
This interface handles the complexity of reserialising inventories and
3956
revisions from different formats, and allows unidirectional insertion into
3957
stacked repositories without looking for the missing basis parents
3961
def __init__(self, target_repo):
3962
self.target_repo = target_repo
3964
def insert_stream(self, stream, src_format, resume_tokens):
3965
"""Insert a stream's content into the target repository.
3967
:param src_format: a bzr repository format.
3969
:return: a list of resume tokens and an iterable of keys additional
3970
items required before the insertion can be completed.
3972
self.target_repo.lock_write()
3975
self.target_repo.resume_write_group(resume_tokens)
3977
self.target_repo.start_write_group()
3979
# locked_insert_stream performs a commit|suspend.
3980
return self._locked_insert_stream(stream, src_format)
3982
self.target_repo.abort_write_group(suppress_errors=True)
3985
self.target_repo.unlock()
3987
def _locked_insert_stream(self, stream, src_format):
3988
to_serializer = self.target_repo._format._serializer
3989
src_serializer = src_format._serializer
3990
if to_serializer == src_serializer:
3991
# If serializers match and the target is a pack repository, set the
3992
# write cache size on the new pack. This avoids poor performance
3993
# on transports where append is unbuffered (such as
3994
# RemoteTransport). This is safe to do because nothing should read
3995
# back from the target repository while a stream with matching
3996
# serialization is being inserted.
3997
# The exception is that a delta record from the source that should
3998
# be a fulltext may need to be expanded by the target (see
3999
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4000
# explicitly flush any buffered writes first in that rare case.
4002
new_pack = self.target_repo._pack_collection._new_pack
4003
except AttributeError:
4004
# Not a pack repository
4007
new_pack.set_write_cache_size(1024*1024)
4008
for substream_type, substream in stream:
4009
if substream_type == 'texts':
4010
self.target_repo.texts.insert_record_stream(substream)
4011
elif substream_type == 'inventories':
4012
if src_serializer == to_serializer:
4013
self.target_repo.inventories.insert_record_stream(
4016
self._extract_and_insert_inventories(
4017
substream, src_serializer)
4018
elif substream_type == 'chk_bytes':
4019
# XXX: This doesn't support conversions, as it assumes the
4020
# conversion was done in the fetch code.
4021
self.target_repo.chk_bytes.insert_record_stream(substream)
4022
elif substream_type == 'revisions':
4023
# This may fallback to extract-and-insert more often than
4024
# required if the serializers are different only in terms of
4026
if src_serializer == to_serializer:
4027
self.target_repo.revisions.insert_record_stream(
4030
self._extract_and_insert_revisions(substream,
4032
elif substream_type == 'signatures':
4033
self.target_repo.signatures.insert_record_stream(substream)
4035
raise AssertionError('kaboom! %s' % (substream_type,))
4037
missing_keys = set()
4038
for prefix, versioned_file in (
4039
('texts', self.target_repo.texts),
4040
('inventories', self.target_repo.inventories),
4041
('revisions', self.target_repo.revisions),
4042
('signatures', self.target_repo.signatures),
4044
missing_keys.update((prefix,) + key for key in
4045
versioned_file.get_missing_compression_parent_keys())
4046
except NotImplementedError:
4047
# cannot even attempt suspending, and missing would have failed
4048
# during stream insertion.
4049
missing_keys = set()
4052
# suspend the write group and tell the caller what we is
4053
# missing. We know we can suspend or else we would not have
4054
# entered this code path. (All repositories that can handle
4055
# missing keys can handle suspending a write group).
4056
write_group_tokens = self.target_repo.suspend_write_group()
4057
return write_group_tokens, missing_keys
4058
self.target_repo.commit_write_group()
4061
def _extract_and_insert_inventories(self, substream, serializer):
4062
"""Generate a new inventory versionedfile in target, converting data.
4064
The inventory is retrieved from the source, (deserializing it), and
4065
stored in the target (reserializing it in a different format).
4067
for record in substream:
4068
bytes = record.get_bytes_as('fulltext')
4069
revision_id = record.key[0]
4070
inv = serializer.read_inventory_from_string(bytes, revision_id)
4071
parents = [key[0] for key in record.parents]
4072
self.target_repo.add_inventory(revision_id, inv, parents)
4074
def _extract_and_insert_revisions(self, substream, serializer):
4075
for record in substream:
4076
bytes = record.get_bytes_as('fulltext')
4077
revision_id = record.key[0]
4078
rev = serializer.read_revision_from_string(bytes)
4079
if rev.revision_id != revision_id:
4080
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4081
self.target_repo.add_revision(revision_id, rev)
4084
if self.target_repo._format._fetch_reconcile:
4085
self.target_repo.reconcile()
4088
class StreamSource(object):
4089
"""A source of a stream for fetching between repositories."""
4091
def __init__(self, from_repository, to_format):
4092
"""Create a StreamSource streaming from from_repository."""
4093
self.from_repository = from_repository
4094
self.to_format = to_format
4096
def delta_on_metadata(self):
4097
"""Return True if delta's are permitted on metadata streams.
4099
That is on revisions and signatures.
4101
src_serializer = self.from_repository._format._serializer
4102
target_serializer = self.to_format._serializer
4103
return (self.to_format._fetch_uses_deltas and
4104
src_serializer == target_serializer)
4106
def _fetch_revision_texts(self, revs):
4107
# fetch signatures first and then the revision texts
4108
# may need to be a InterRevisionStore call here.
4109
from_sf = self.from_repository.signatures
4110
# A missing signature is just skipped.
4111
keys = [(rev_id,) for rev_id in revs]
4112
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4114
self.to_format._fetch_order,
4115
not self.to_format._fetch_uses_deltas))
4116
# If a revision has a delta, this is actually expanded inside the
4117
# insert_record_stream code now, which is an alternate fix for
4119
from_rf = self.from_repository.revisions
4120
revisions = from_rf.get_record_stream(
4122
self.to_format._fetch_order,
4123
not self.delta_on_metadata())
4124
return [('signatures', signatures), ('revisions', revisions)]
4126
def _generate_root_texts(self, revs):
4127
"""This will be called by __fetch between fetching weave texts and
4128
fetching the inventory weave.
4130
Subclasses should override this if they need to generate root texts
4131
after fetching weave texts.
4133
if self._rich_root_upgrade():
4135
return bzrlib.fetch.Inter1and2Helper(
4136
self.from_repository).generate_root_texts(revs)
4140
def get_stream(self, search):
4142
revs = search.get_keys()
4143
graph = self.from_repository.get_graph()
4144
revs = list(graph.iter_topo_order(revs))
4145
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4147
for knit_kind, file_id, revisions in data_to_fetch:
4148
if knit_kind != phase:
4150
# Make a new progress bar for this phase
4151
if knit_kind == "file":
4152
# Accumulate file texts
4153
text_keys.extend([(file_id, revision) for revision in
4155
elif knit_kind == "inventory":
4156
# Now copy the file texts.
4157
from_texts = self.from_repository.texts
4158
yield ('texts', from_texts.get_record_stream(
4159
text_keys, self.to_format._fetch_order,
4160
not self.to_format._fetch_uses_deltas))
4161
# Cause an error if a text occurs after we have done the
4164
# Before we process the inventory we generate the root
4165
# texts (if necessary) so that the inventories references
4167
for _ in self._generate_root_texts(revs):
4169
# NB: This currently reopens the inventory weave in source;
4170
# using a single stream interface instead would avoid this.
4171
from_weave = self.from_repository.inventories
4172
# we fetch only the referenced inventories because we do not
4173
# know for unselected inventories whether all their required
4174
# texts are present in the other repository - it could be
4176
for info in self._get_inventory_stream(revs):
4178
elif knit_kind == "signatures":
4179
# Nothing to do here; this will be taken care of when
4180
# _fetch_revision_texts happens.
4182
elif knit_kind == "revisions":
4183
for record in self._fetch_revision_texts(revs):
4186
raise AssertionError("Unknown knit kind %r" % knit_kind)
4188
def get_stream_for_missing_keys(self, missing_keys):
4189
# missing keys can only occur when we are byte copying and not
4190
# translating (because translation means we don't send
4191
# unreconstructable deltas ever).
4193
keys['texts'] = set()
4194
keys['revisions'] = set()
4195
keys['inventories'] = set()
4196
keys['signatures'] = set()
4197
for key in missing_keys:
4198
keys[key[0]].add(key[1:])
4199
if len(keys['revisions']):
4200
# If we allowed copying revisions at this point, we could end up
4201
# copying a revision without copying its required texts: a
4202
# violation of the requirements for repository integrity.
4203
raise AssertionError(
4204
'cannot copy revisions to fill in missing deltas %s' % (
4205
keys['revisions'],))
4206
for substream_kind, keys in keys.iteritems():
4207
vf = getattr(self.from_repository, substream_kind)
4208
# Ask for full texts always so that we don't need more round trips
4209
# after this stream.
4210
stream = vf.get_record_stream(keys,
4211
self.to_format._fetch_order, True)
4212
yield substream_kind, stream
4214
def inventory_fetch_order(self):
4215
if self._rich_root_upgrade():
4216
return 'topological'
4218
return self.to_format._fetch_order
4220
def _rich_root_upgrade(self):
4221
return (not self.from_repository._format.rich_root_data and
4222
self.to_format.rich_root_data)
4224
def _get_inventory_stream(self, revision_ids):
4225
from_format = self.from_repository._format
4226
if (from_format.supports_chks and self.to_format.supports_chks
4227
and (from_format._serializer == self.to_format._serializer)):
4228
# Both sides support chks, and they use the same serializer, so it
4229
# is safe to transmit the chk pages and inventory pages across
4231
return self._get_chk_inventory_stream(revision_ids)
4232
elif (not from_format.supports_chks):
4233
# Source repository doesn't support chks. So we can transmit the
4234
# inventories 'as-is' and either they are just accepted on the
4235
# target, or the Sink will properly convert it.
4236
return self._get_simple_inventory_stream(revision_ids)
4238
# XXX: Hack to make not-chk->chk fetch: copy the inventories as
4239
# inventories. Note that this should probably be done somehow
4240
# as part of bzrlib.repository.StreamSink. Except JAM couldn't
4241
# figure out how a non-chk repository could possibly handle
4242
# deserializing an inventory stream from a chk repo, as it
4243
# doesn't have a way to understand individual pages.
4244
return self._get_convertable_inventory_stream(revision_ids)
4246
def _get_simple_inventory_stream(self, revision_ids):
4247
from_weave = self.from_repository.inventories
4248
yield ('inventories', from_weave.get_record_stream(
4249
[(rev_id,) for rev_id in revision_ids],
4250
self.inventory_fetch_order(),
4251
not self.delta_on_metadata()))
4253
def _get_chk_inventory_stream(self, revision_ids):
4254
"""Fetch the inventory texts, along with the associated chk maps."""
4255
# We want an inventory outside of the search set, so that we can filter
4256
# out uninteresting chk pages. For now we use
4257
# _find_revision_outside_set, but if we had a Search with cut_revs, we
4258
# could use that instead.
4259
start_rev_id = self.from_repository._find_revision_outside_set(
4261
start_rev_key = (start_rev_id,)
4262
inv_keys_to_fetch = [(rev_id,) for rev_id in revision_ids]
4263
if start_rev_id != _mod_revision.NULL_REVISION:
4264
inv_keys_to_fetch.append((start_rev_id,))
4265
# Any repo that supports chk_bytes must also support out-of-order
4266
# insertion. At least, that is how we expect it to work
4267
# We use get_record_stream instead of iter_inventories because we want
4268
# to be able to insert the stream as well. We could instead fetch
4269
# allowing deltas, and then iter_inventories, but we don't know whether
4270
# source or target is more 'local' anway.
4271
inv_stream = self.from_repository.inventories.get_record_stream(
4272
inv_keys_to_fetch, 'unordered',
4273
True) # We need them as full-texts so we can find their references
4274
uninteresting_chk_roots = set()
4275
interesting_chk_roots = set()
4276
def filter_inv_stream(inv_stream):
4277
for idx, record in enumerate(inv_stream):
4278
### child_pb.update('fetch inv', idx, len(inv_keys_to_fetch))
4279
bytes = record.get_bytes_as('fulltext')
4280
chk_inv = inventory.CHKInventory.deserialise(
4281
self.from_repository.chk_bytes, bytes, record.key)
4282
if record.key == start_rev_key:
4283
uninteresting_chk_roots.add(chk_inv.id_to_entry.key())
4284
p_id_map = chk_inv.parent_id_basename_to_file_id
4285
if p_id_map is not None:
4286
uninteresting_chk_roots.add(p_id_map.key())
4289
interesting_chk_roots.add(chk_inv.id_to_entry.key())
4290
p_id_map = chk_inv.parent_id_basename_to_file_id
4291
if p_id_map is not None:
4292
interesting_chk_roots.add(p_id_map.key())
4293
### pb.update('fetch inventory', 0, 2)
4294
yield ('inventories', filter_inv_stream(inv_stream))
4295
# Now that we have worked out all of the interesting root nodes, grab
4296
# all of the interesting pages and insert them
4297
### pb.update('fetch inventory', 1, 2)
4298
interesting = chk_map.iter_interesting_nodes(
4299
self.from_repository.chk_bytes, interesting_chk_roots,
4300
uninteresting_chk_roots)
4301
def to_stream_adapter():
4302
"""Adapt the iter_interesting_nodes result to a single stream.
4304
iter_interesting_nodes returns records as it processes them, along
4305
with keys. However, we only want to return the records themselves.
4307
for record, items in interesting:
4308
if record is not None:
4310
# XXX: We could instead call get_record_stream(records.keys())
4311
# ATM, this will always insert the records as fulltexts, and
4312
# requires that you can hang on to records once you have gone
4313
# on to the next one. Further, it causes the target to
4314
# recompress the data. Testing shows it to be faster than
4315
# requesting the records again, though.
4316
yield ('chk_bytes', to_stream_adapter())
4317
### pb.update('fetch inventory', 2, 2)
4319
def _get_convertable_inventory_stream(self, revision_ids):
4320
# XXX: One of source or target is using chks, and they don't have
4321
# compatible serializations. The StreamSink code expects to be
4322
# able to convert on the target, so we need to put
4323
# bytes-on-the-wire that can be converted
4324
yield ('inventories', self._stream_invs_as_fulltexts(revision_ids))
4326
def _stream_invs_as_fulltexts(self, revision_ids):
4327
from_repo = self.from_repository
4328
from_serializer = from_repo._format._serializer
4329
revision_keys = [(rev_id,) for rev_id in revision_ids]
4330
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4331
for inv in self.from_repository.iter_inventories(revision_ids):
4332
# XXX: This is a bit hackish, but it works. Basically,
4333
# CHKSerializer 'accidentally' supports
4334
# read/write_inventory_to_string, even though that is never
4335
# the format that is stored on disk. It *does* give us a
4336
# single string representation for an inventory, so live with
4338
# This would be far better if we had a 'serialized inventory
4339
# delta' form. Then we could use 'inventory._make_delta', and
4340
# transmit that. This would both be faster to generate, and
4341
# result in fewer bytes-on-the-wire.
4342
as_bytes = from_serializer.write_inventory_to_string(inv)
4343
key = (inv.revision_id,)
4344
parent_keys = parent_map.get(key, ())
4345
yield versionedfile.FulltextContentFactory(
4346
key, parent_keys, None, as_bytes)
added
removed
bzrlib/repository.py
