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
# an inventory delta was accumulated without creating a new
209
basis_id = self.basis_delta_revision
210
self.inv_sha1 = self.repository.add_inventory_by_delta(
211
basis_id, self._basis_delta, self._new_revision_id,
214
if self.new_inventory.root is None:
215
raise AssertionError('Root entry should be supplied to'
216
' record_entry_contents, as of bzr 0.10.')
217
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
218
self.new_inventory.revision_id = self._new_revision_id
219
self.inv_sha1 = self.repository.add_inventory(
220
self._new_revision_id,
224
return self._new_revision_id
226
def _gen_revision_id(self):
227
"""Return new revision-id."""
228
return generate_ids.gen_revision_id(self._config.username(),
231
def _generate_revision_if_needed(self):
232
"""Create a revision id if None was supplied.
234
If the repository can not support user-specified revision ids
235
they should override this function and raise CannotSetRevisionId
236
if _new_revision_id is not None.
238
:raises: CannotSetRevisionId
240
if self._new_revision_id is None:
241
self._new_revision_id = self._gen_revision_id()
242
self.random_revid = True
244
self.random_revid = False
246
def _heads(self, file_id, revision_ids):
247
"""Calculate the graph heads for revision_ids in the graph of file_id.
249
This can use either a per-file graph or a global revision graph as we
250
have an identity relationship between the two graphs.
252
return self.__heads(revision_ids)
254
def _check_root(self, ie, parent_invs, tree):
255
"""Helper for record_entry_contents.
257
:param ie: An entry being added.
258
:param parent_invs: The inventories of the parent revisions of the
260
:param tree: The tree that is being committed.
262
# In this revision format, root entries have no knit or weave When
263
# serializing out to disk and back in root.revision is always
265
ie.revision = self._new_revision_id
267
def _require_root_change(self, tree):
268
"""Enforce an appropriate root object change.
270
This is called once when record_iter_changes is called, if and only if
271
the root was not in the delta calculated by record_iter_changes.
273
:param tree: The tree which is being committed.
275
# NB: if there are no parents then this method is not called, so no
276
# need to guard on parents having length.
277
entry = entry_factory['directory'](tree.path2id(''), '',
279
entry.revision = self._new_revision_id
280
self._basis_delta.append(('', '', entry.file_id, entry))
282
def _get_delta(self, ie, basis_inv, path):
283
"""Get a delta against the basis inventory for ie."""
284
if ie.file_id not in basis_inv:
286
result = (None, path, ie.file_id, ie)
287
self._basis_delta.append(result)
289
elif ie != basis_inv[ie.file_id]:
291
# TODO: avoid tis id2path call.
292
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
293
self._basis_delta.append(result)
299
def get_basis_delta(self):
300
"""Return the complete inventory delta versus the basis inventory.
302
This has been built up with the calls to record_delete and
303
record_entry_contents. The client must have already called
304
will_record_deletes() to indicate that they will be generating a
307
:return: An inventory delta, suitable for use with apply_delta, or
308
Repository.add_inventory_by_delta, etc.
310
if not self._recording_deletes:
311
raise AssertionError("recording deletes not activated.")
312
return self._basis_delta
314
def record_delete(self, path, file_id):
315
"""Record that a delete occured against a basis tree.
317
This is an optional API - when used it adds items to the basis_delta
318
being accumulated by the commit builder. It cannot be called unless the
319
method will_record_deletes() has been called to inform the builder that
320
a delta is being supplied.
322
:param path: The path of the thing deleted.
323
:param file_id: The file id that was deleted.
325
if not self._recording_deletes:
326
raise AssertionError("recording deletes not activated.")
327
delta = (path, None, file_id, None)
328
self._basis_delta.append(delta)
329
self._any_changes = True
332
def will_record_deletes(self):
333
"""Tell the commit builder that deletes are being notified.
335
This enables the accumulation of an inventory delta; for the resulting
336
commit to be valid, deletes against the basis MUST be recorded via
337
builder.record_delete().
339
self._recording_deletes = True
341
basis_id = self.parents[0]
343
basis_id = _mod_revision.NULL_REVISION
344
self.basis_delta_revision = basis_id
346
def record_entry_contents(self, ie, parent_invs, path, tree,
348
"""Record the content of ie from tree into the commit if needed.
350
Side effect: sets ie.revision when unchanged
352
:param ie: An inventory entry present in the commit.
353
:param parent_invs: The inventories of the parent revisions of the
355
:param path: The path the entry is at in the tree.
356
:param tree: The tree which contains this entry and should be used to
358
:param content_summary: Summary data from the tree about the paths
359
content - stat, length, exec, sha/link target. This is only
360
accessed when the entry has a revision of None - that is when it is
361
a candidate to commit.
362
:return: A tuple (change_delta, version_recorded, fs_hash).
363
change_delta is an inventory_delta change for this entry against
364
the basis tree of the commit, or None if no change occured against
366
version_recorded is True if a new version of the entry has been
367
recorded. For instance, committing a merge where a file was only
368
changed on the other side will return (delta, False).
369
fs_hash is either None, or the hash details for the path (currently
370
a tuple of the contents sha1 and the statvalue returned by
371
tree.get_file_with_stat()).
373
if self.new_inventory.root is None:
374
if ie.parent_id is not None:
375
raise errors.RootMissing()
376
self._check_root(ie, parent_invs, tree)
377
if ie.revision is None:
378
kind = content_summary[0]
380
# ie is carried over from a prior commit
382
# XXX: repository specific check for nested tree support goes here - if
383
# the repo doesn't want nested trees we skip it ?
384
if (kind == 'tree-reference' and
385
not self.repository._format.supports_tree_reference):
386
# mismatch between commit builder logic and repository:
387
# this needs the entry creation pushed down into the builder.
388
raise NotImplementedError('Missing repository subtree support.')
389
self.new_inventory.add(ie)
391
# TODO: slow, take it out of the inner loop.
393
basis_inv = parent_invs[0]
395
basis_inv = Inventory(root_id=None)
397
# ie.revision is always None if the InventoryEntry is considered
398
# for committing. We may record the previous parents revision if the
399
# content is actually unchanged against a sole head.
400
if ie.revision is not None:
401
if not self._versioned_root and path == '':
402
# repositories that do not version the root set the root's
403
# revision to the new commit even when no change occurs (more
404
# specifically, they do not record a revision on the root; and
405
# the rev id is assigned to the root during deserialisation -
406
# this masks when a change may have occurred against the basis.
407
# To match this we always issue a delta, because the revision
408
# of the root will always be changing.
409
if ie.file_id in basis_inv:
410
delta = (basis_inv.id2path(ie.file_id), path,
414
delta = (None, path, ie.file_id, ie)
415
self._basis_delta.append(delta)
416
return delta, False, None
418
# we don't need to commit this, because the caller already
419
# determined that an existing revision of this file is
420
# appropriate. If its not being considered for committing then
421
# it and all its parents to the root must be unaltered so
422
# no-change against the basis.
423
if ie.revision == self._new_revision_id:
424
raise AssertionError("Impossible situation, a skipped "
425
"inventory entry (%r) claims to be modified in this "
426
"commit (%r).", (ie, self._new_revision_id))
427
return None, False, None
428
# XXX: Friction: parent_candidates should return a list not a dict
429
# so that we don't have to walk the inventories again.
430
parent_candiate_entries = ie.parent_candidates(parent_invs)
431
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
433
for inv in parent_invs:
434
if ie.file_id in inv:
435
old_rev = inv[ie.file_id].revision
436
if old_rev in head_set:
437
heads.append(inv[ie.file_id].revision)
438
head_set.remove(inv[ie.file_id].revision)
441
# now we check to see if we need to write a new record to the
443
# We write a new entry unless there is one head to the ancestors, and
444
# the kind-derived content is unchanged.
446
# Cheapest check first: no ancestors, or more the one head in the
447
# ancestors, we write a new node.
451
# There is a single head, look it up for comparison
452
parent_entry = parent_candiate_entries[heads[0]]
453
# if the non-content specific data has changed, we'll be writing a
455
if (parent_entry.parent_id != ie.parent_id or
456
parent_entry.name != ie.name):
458
# now we need to do content specific checks:
460
# if the kind changed the content obviously has
461
if kind != parent_entry.kind:
463
# Stat cache fingerprint feedback for the caller - None as we usually
464
# don't generate one.
467
if content_summary[2] is None:
468
raise ValueError("Files must not have executable = None")
470
if (# if the file length changed we have to store:
471
parent_entry.text_size != content_summary[1] or
472
# if the exec bit has changed we have to store:
473
parent_entry.executable != content_summary[2]):
475
elif parent_entry.text_sha1 == content_summary[3]:
476
# all meta and content is unchanged (using a hash cache
477
# hit to check the sha)
478
ie.revision = parent_entry.revision
479
ie.text_size = parent_entry.text_size
480
ie.text_sha1 = parent_entry.text_sha1
481
ie.executable = parent_entry.executable
482
return self._get_delta(ie, basis_inv, path), False, None
484
# Either there is only a hash change(no hash cache entry,
485
# or same size content change), or there is no change on
487
# Provide the parent's hash to the store layer, so that the
488
# content is unchanged we will not store a new node.
489
nostore_sha = parent_entry.text_sha1
491
# We want to record a new node regardless of the presence or
492
# absence of a content change in the file.
494
ie.executable = content_summary[2]
495
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
497
lines = file_obj.readlines()
501
ie.text_sha1, ie.text_size = self._add_text_to_weave(
502
ie.file_id, lines, heads, nostore_sha)
503
# Let the caller know we generated a stat fingerprint.
504
fingerprint = (ie.text_sha1, stat_value)
505
except errors.ExistingContent:
506
# Turns out that the file content was unchanged, and we were
507
# only going to store a new node if it was changed. Carry over
509
ie.revision = parent_entry.revision
510
ie.text_size = parent_entry.text_size
511
ie.text_sha1 = parent_entry.text_sha1
512
ie.executable = parent_entry.executable
513
return self._get_delta(ie, basis_inv, path), False, None
514
elif kind == 'directory':
516
# all data is meta here, nothing specific to directory, so
518
ie.revision = parent_entry.revision
519
return self._get_delta(ie, basis_inv, path), False, None
521
self._add_text_to_weave(ie.file_id, lines, heads, None)
522
elif kind == 'symlink':
523
current_link_target = content_summary[3]
525
# symlink target is not generic metadata, check if it has
527
if current_link_target != parent_entry.symlink_target:
530
# unchanged, carry over.
531
ie.revision = parent_entry.revision
532
ie.symlink_target = parent_entry.symlink_target
533
return self._get_delta(ie, basis_inv, path), False, None
534
ie.symlink_target = current_link_target
536
self._add_text_to_weave(ie.file_id, lines, heads, None)
537
elif kind == 'tree-reference':
539
if content_summary[3] != parent_entry.reference_revision:
542
# unchanged, carry over.
543
ie.reference_revision = parent_entry.reference_revision
544
ie.revision = parent_entry.revision
545
return self._get_delta(ie, basis_inv, path), False, None
546
ie.reference_revision = content_summary[3]
548
self._add_text_to_weave(ie.file_id, lines, heads, None)
550
raise NotImplementedError('unknown kind')
551
ie.revision = self._new_revision_id
552
self._any_changes = True
553
return self._get_delta(ie, basis_inv, path), True, fingerprint
555
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
556
_entry_factory=entry_factory):
557
"""Record a new tree via iter_changes.
559
:param tree: The tree to obtain text contents from for changed objects.
560
:param basis_revision_id: The revision id of the tree the iter_changes
561
has been generated against. Currently assumed to be the same
562
as self.parents[0] - if it is not, errors may occur.
563
:param iter_changes: An iter_changes iterator with the changes to apply
564
to basis_revision_id. The iterator must not include any items with
565
a current kind of None - missing items must be either filtered out
566
or errored-on beefore record_iter_changes sees the item.
567
:param _entry_factory: Private method to bind entry_factory locally for
569
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
572
# Create an inventory delta based on deltas between all the parents and
573
# deltas between all the parent inventories. We use inventory delta's
574
# between the inventory objects because iter_changes masks
575
# last-changed-field only changes.
577
# file_id -> change map, change is fileid, paths, changed, versioneds,
578
# parents, names, kinds, executables
580
# {file_id -> revision_id -> inventory entry, for entries in parent
581
# trees that are not parents[0]
585
revtrees = list(self.repository.revision_trees(self.parents))
586
except errors.NoSuchRevision:
587
# one or more ghosts, slow path.
589
for revision_id in self.parents:
591
revtrees.append(self.repository.revision_tree(revision_id))
592
except errors.NoSuchRevision:
594
basis_revision_id = _mod_revision.NULL_REVISION
596
revtrees.append(self.repository.revision_tree(
597
_mod_revision.NULL_REVISION))
598
# The basis inventory from a repository
600
basis_inv = revtrees[0].inventory
602
basis_inv = self.repository.revision_tree(
603
_mod_revision.NULL_REVISION).inventory
604
if len(self.parents) > 0:
605
if basis_revision_id != self.parents[0] and not ghost_basis:
607
"arbitrary basis parents not yet supported with merges")
608
for revtree in revtrees[1:]:
609
for change in revtree.inventory._make_delta(basis_inv):
610
if change[1] is None:
611
# Not present in this parent.
613
if change[2] not in merged_ids:
614
if change[0] is not None:
615
basis_entry = basis_inv[change[2]]
616
merged_ids[change[2]] = [
618
basis_entry.revision,
621
parent_entries[change[2]] = {
623
basis_entry.revision:basis_entry,
625
change[3].revision:change[3],
628
merged_ids[change[2]] = [change[3].revision]
629
parent_entries[change[2]] = {change[3].revision:change[3]}
631
merged_ids[change[2]].append(change[3].revision)
632
parent_entries[change[2]][change[3].revision] = change[3]
635
# Setup the changes from the tree:
636
# changes maps file_id -> (change, [parent revision_ids])
638
for change in iter_changes:
639
# This probably looks up in basis_inv way to much.
640
if change[1][0] is not None:
641
head_candidate = [basis_inv[change[0]].revision]
644
changes[change[0]] = change, merged_ids.get(change[0],
646
unchanged_merged = set(merged_ids) - set(changes)
647
# Extend the changes dict with synthetic changes to record merges of
649
for file_id in unchanged_merged:
650
# Record a merged version of these items that did not change vs the
651
# basis. This can be either identical parallel changes, or a revert
652
# of a specific file after a merge. The recorded content will be
653
# that of the current tree (which is the same as the basis), but
654
# the per-file graph will reflect a merge.
655
# NB:XXX: We are reconstructing path information we had, this
656
# should be preserved instead.
657
# inv delta change: (file_id, (path_in_source, path_in_target),
658
# changed_content, versioned, parent, name, kind,
661
basis_entry = basis_inv[file_id]
662
except errors.NoSuchId:
663
# a change from basis->some_parents but file_id isn't in basis
664
# so was new in the merge, which means it must have changed
665
# from basis -> current, and as it hasn't the add was reverted
666
# by the user. So we discard this change.
670
(basis_inv.id2path(file_id), tree.id2path(file_id)),
672
(basis_entry.parent_id, basis_entry.parent_id),
673
(basis_entry.name, basis_entry.name),
674
(basis_entry.kind, basis_entry.kind),
675
(basis_entry.executable, basis_entry.executable))
676
changes[file_id] = (change, merged_ids[file_id])
677
# changes contains tuples with the change and a set of inventory
678
# candidates for the file.
680
# old_path, new_path, file_id, new_inventory_entry
681
seen_root = False # Is the root in the basis delta?
682
inv_delta = self._basis_delta
683
modified_rev = self._new_revision_id
684
for change, head_candidates in changes.values():
685
if change[3][1]: # versioned in target.
686
# Several things may be happening here:
687
# We may have a fork in the per-file graph
688
# - record a change with the content from tree
689
# We may have a change against < all trees
690
# - carry over the tree that hasn't changed
691
# We may have a change against all trees
692
# - record the change with the content from tree
695
entry = _entry_factory[kind](file_id, change[5][1],
697
head_set = self._heads(change[0], set(head_candidates))
700
for head_candidate in head_candidates:
701
if head_candidate in head_set:
702
heads.append(head_candidate)
703
head_set.remove(head_candidate)
706
# Could be a carry-over situation:
707
parent_entry_revs = parent_entries.get(file_id, None)
708
if parent_entry_revs:
709
parent_entry = parent_entry_revs.get(heads[0], None)
712
if parent_entry is None:
713
# The parent iter_changes was called against is the one
714
# that is the per-file head, so any change is relevant
715
# iter_changes is valid.
716
carry_over_possible = False
718
# could be a carry over situation
719
# A change against the basis may just indicate a merge,
720
# we need to check the content against the source of the
721
# merge to determine if it was changed after the merge
723
if (parent_entry.kind != entry.kind or
724
parent_entry.parent_id != entry.parent_id or
725
parent_entry.name != entry.name):
726
# Metadata common to all entries has changed
727
# against per-file parent
728
carry_over_possible = False
730
carry_over_possible = True
731
# per-type checks for changes against the parent_entry
734
# Cannot be a carry-over situation
735
carry_over_possible = False
736
# Populate the entry in the delta
738
# XXX: There is still a small race here: If someone reverts the content of a file
739
# after iter_changes examines and decides it has changed,
740
# we will unconditionally record a new version even if some
741
# other process reverts it while commit is running (with
742
# the revert happening after iter_changes did it's
745
entry.executable = True
747
entry.executable = False
748
if (carry_over_possible and
749
parent_entry.executable == entry.executable):
750
# Check the file length, content hash after reading
752
nostore_sha = parent_entry.text_sha1
755
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
757
lines = file_obj.readlines()
761
entry.text_sha1, entry.text_size = self._add_text_to_weave(
762
file_id, lines, heads, nostore_sha)
763
yield file_id, change[1][1], (entry.text_sha1, stat_value)
764
except errors.ExistingContent:
765
# No content change against a carry_over parent
766
# Perhaps this should also yield a fs hash update?
768
entry.text_size = parent_entry.text_size
769
entry.text_sha1 = parent_entry.text_sha1
770
elif kind == 'symlink':
772
entry.symlink_target = tree.get_symlink_target(file_id)
773
if (carry_over_possible and
774
parent_entry.symlink_target == entry.symlink_target):
777
self._add_text_to_weave(change[0], [], heads, None)
778
elif kind == 'directory':
779
if carry_over_possible:
782
# Nothing to set on the entry.
783
# XXX: split into the Root and nonRoot versions.
784
if change[1][1] != '' or self.repository.supports_rich_root():
785
self._add_text_to_weave(change[0], [], heads, None)
786
elif kind == 'tree-reference':
787
if not self.repository._format.supports_tree_reference:
788
# This isn't quite sane as an error, but we shouldn't
789
# ever see this code path in practice: tree's don't
790
# permit references when the repo doesn't support tree
792
raise errors.UnsupportedOperation(tree.add_reference,
794
entry.reference_revision = \
795
tree.get_reference_revision(change[0])
796
if (carry_over_possible and
797
parent_entry.reference_revision == reference_revision):
800
self._add_text_to_weave(change[0], [], heads, None)
802
raise AssertionError('unknown kind %r' % kind)
804
entry.revision = modified_rev
806
entry.revision = parent_entry.revision
809
new_path = change[1][1]
810
inv_delta.append((change[1][0], new_path, change[0], entry))
813
self.new_inventory = None
815
self._any_changes = True
817
# housekeeping root entry changes do not affect no-change commits.
818
self._require_root_change(tree)
819
self.basis_delta_revision = basis_revision_id
821
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
822
# Note: as we read the content directly from the tree, we know its not
823
# been turned into unicode or badly split - but a broken tree
824
# implementation could give us bad output from readlines() so this is
825
# not a guarantee of safety. What would be better is always checking
826
# the content during test suite execution. RBC 20070912
827
parent_keys = tuple((file_id, parent) for parent in parents)
828
return self.repository.texts.add_lines(
829
(file_id, self._new_revision_id), parent_keys, new_lines,
830
nostore_sha=nostore_sha, random_id=self.random_revid,
831
check_content=False)[0:2]
834
class RootCommitBuilder(CommitBuilder):
835
"""This commitbuilder actually records the root id"""
837
# the root entry gets versioned properly by this builder.
838
_versioned_root = True
840
def _check_root(self, ie, parent_invs, tree):
841
"""Helper for record_entry_contents.
843
:param ie: An entry being added.
844
:param parent_invs: The inventories of the parent revisions of the
846
:param tree: The tree that is being committed.
849
def _require_root_change(self, tree):
850
"""Enforce an appropriate root object change.
852
This is called once when record_iter_changes is called, if and only if
853
the root was not in the delta calculated by record_iter_changes.
855
:param tree: The tree which is being committed.
857
# versioned roots do not change unless the tree found a change.
860
######################################################################
863
class Repository(object):
864
"""Repository holding history for one or more branches.
866
The repository holds and retrieves historical information including
867
revisions and file history. It's normally accessed only by the Branch,
868
which views a particular line of development through that history.
870
The Repository builds on top of some byte storage facilies (the revisions,
871
signatures, inventories, texts and chk_bytes attributes) and a Transport,
872
which respectively provide byte storage and a means to access the (possibly
875
The byte storage facilities are addressed via tuples, which we refer to
876
as 'keys' throughout the code base. Revision_keys, inventory_keys and
877
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
878
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
879
byte string made up of a hash identifier and a hash value.
880
We use this interface because it allows low friction with the underlying
881
code that implements disk indices, network encoding and other parts of
884
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
885
the serialised revisions for the repository. This can be used to obtain
886
revision graph information or to access raw serialised revisions.
887
The result of trying to insert data into the repository via this store
888
is undefined: it should be considered read-only except for implementors
890
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
891
the serialised signatures for the repository. This can be used to
892
obtain access to raw serialised signatures. The result of trying to
893
insert data into the repository via this store is undefined: it should
894
be considered read-only except for implementors of repositories.
895
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
896
the serialised inventories for the repository. This can be used to
897
obtain unserialised inventories. The result of trying to insert data
898
into the repository via this store is undefined: it should be
899
considered read-only except for implementors of repositories.
900
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
901
texts of files and directories for the repository. This can be used to
902
obtain file texts or file graphs. Note that Repository.iter_file_bytes
903
is usually a better interface for accessing file texts.
904
The result of trying to insert data into the repository via this store
905
is undefined: it should be considered read-only except for implementors
907
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
908
any data the repository chooses to store or have indexed by its hash.
909
The result of trying to insert data into the repository via this store
910
is undefined: it should be considered read-only except for implementors
912
:ivar _transport: Transport for file access to repository, typically
913
pointing to .bzr/repository.
916
# What class to use for a CommitBuilder. Often its simpler to change this
917
# in a Repository class subclass rather than to override
918
# get_commit_builder.
919
_commit_builder_class = CommitBuilder
920
# The search regex used by xml based repositories to determine what things
921
# where changed in a single commit.
922
_file_ids_altered_regex = lazy_regex.lazy_compile(
923
r'file_id="(?P<file_id>[^"]+)"'
924
r'.* revision="(?P<revision_id>[^"]+)"'
927
def abort_write_group(self, suppress_errors=False):
928
"""Commit the contents accrued within the current write group.
930
:param suppress_errors: if true, abort_write_group will catch and log
931
unexpected errors that happen during the abort, rather than
932
allowing them to propagate. Defaults to False.
934
:seealso: start_write_group.
936
if self._write_group is not self.get_transaction():
937
# has an unlock or relock occured ?
938
raise errors.BzrError(
939
'mismatched lock context and write group. %r, %r' %
940
(self._write_group, self.get_transaction()))
942
self._abort_write_group()
943
except Exception, exc:
944
self._write_group = None
945
if not suppress_errors:
947
mutter('abort_write_group failed')
948
log_exception_quietly()
949
note('bzr: ERROR (ignored): %s', exc)
950
self._write_group = None
952
def _abort_write_group(self):
953
"""Template method for per-repository write group cleanup.
955
This is called during abort before the write group is considered to be
956
finished and should cleanup any internal state accrued during the write
957
group. There is no requirement that data handed to the repository be
958
*not* made available - this is not a rollback - but neither should any
959
attempt be made to ensure that data added is fully commited. Abort is
960
invoked when an error has occured so futher disk or network operations
961
may not be possible or may error and if possible should not be
965
def add_fallback_repository(self, repository):
966
"""Add a repository to use for looking up data not held locally.
968
:param repository: A repository.
970
if not self._format.supports_external_lookups:
971
raise errors.UnstackableRepositoryFormat(self._format, self.base)
973
# This repository will call fallback.unlock() when we transition to
974
# the unlocked state, so we make sure to increment the lock count
975
repository.lock_read()
976
self._check_fallback_repository(repository)
977
self._fallback_repositories.append(repository)
978
self.texts.add_fallback_versioned_files(repository.texts)
979
self.inventories.add_fallback_versioned_files(repository.inventories)
980
self.revisions.add_fallback_versioned_files(repository.revisions)
981
self.signatures.add_fallback_versioned_files(repository.signatures)
982
if self.chk_bytes is not None:
983
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
985
def _check_fallback_repository(self, repository):
986
"""Check that this repository can fallback to repository safely.
988
Raise an error if not.
990
:param repository: A repository to fallback to.
992
return InterRepository._assert_same_model(self, repository)
994
def add_inventory(self, revision_id, inv, parents):
995
"""Add the inventory inv to the repository as revision_id.
997
:param parents: The revision ids of the parents that revision_id
998
is known to have and are in the repository already.
1000
:returns: The validator(which is a sha1 digest, though what is sha'd is
1001
repository format specific) of the serialized inventory.
1003
if not self.is_in_write_group():
1004
raise AssertionError("%r not in write group" % (self,))
1005
_mod_revision.check_not_reserved_id(revision_id)
1006
if not (inv.revision_id is None or inv.revision_id == revision_id):
1007
raise AssertionError(
1008
"Mismatch between inventory revision"
1009
" id and insertion revid (%r, %r)"
1010
% (inv.revision_id, revision_id))
1011
if inv.root is None:
1012
raise AssertionError()
1013
return self._add_inventory_checked(revision_id, inv, parents)
1015
def _add_inventory_checked(self, revision_id, inv, parents):
1016
"""Add inv to the repository after checking the inputs.
1018
This function can be overridden to allow different inventory styles.
1020
:seealso: add_inventory, for the contract.
1022
inv_lines = self._serialise_inventory_to_lines(inv)
1023
return self._inventory_add_lines(revision_id, parents,
1024
inv_lines, check_content=False)
1026
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1027
parents, basis_inv=None, propagate_caches=False):
1028
"""Add a new inventory expressed as a delta against another revision.
1030
:param basis_revision_id: The inventory id the delta was created
1031
against. (This does not have to be a direct parent.)
1032
:param delta: The inventory delta (see Inventory.apply_delta for
1034
:param new_revision_id: The revision id that the inventory is being
1036
:param parents: The revision ids of the parents that revision_id is
1037
known to have and are in the repository already. These are supplied
1038
for repositories that depend on the inventory graph for revision
1039
graph access, as well as for those that pun ancestry with delta
1041
:param basis_inv: The basis inventory if it is already known,
1043
:param propagate_caches: If True, the caches for this inventory are
1044
copied to and updated for the result if possible.
1046
:returns: (validator, new_inv)
1047
The validator(which is a sha1 digest, though what is sha'd is
1048
repository format specific) of the serialized inventory, and the
1049
resulting inventory.
1051
if not self.is_in_write_group():
1052
raise AssertionError("%r not in write group" % (self,))
1053
_mod_revision.check_not_reserved_id(new_revision_id)
1054
basis_tree = self.revision_tree(basis_revision_id)
1055
basis_tree.lock_read()
1057
# Note that this mutates the inventory of basis_tree, which not all
1058
# inventory implementations may support: A better idiom would be to
1059
# return a new inventory, but as there is no revision tree cache in
1060
# repository this is safe for now - RBC 20081013
1061
if basis_inv is None:
1062
basis_inv = basis_tree.inventory
1063
basis_inv.apply_delta(delta)
1064
basis_inv.revision_id = new_revision_id
1065
return (self.add_inventory(new_revision_id, basis_inv, parents),
1070
def _inventory_add_lines(self, revision_id, parents, lines,
1071
check_content=True):
1072
"""Store lines in inv_vf and return the sha1 of the inventory."""
1073
parents = [(parent,) for parent in parents]
1074
return self.inventories.add_lines((revision_id,), parents, lines,
1075
check_content=check_content)[0]
1077
def add_revision(self, revision_id, rev, inv=None, config=None):
1078
"""Add rev to the revision store as revision_id.
1080
:param revision_id: the revision id to use.
1081
:param rev: The revision object.
1082
:param inv: The inventory for the revision. if None, it will be looked
1083
up in the inventory storer
1084
:param config: If None no digital signature will be created.
1085
If supplied its signature_needed method will be used
1086
to determine if a signature should be made.
1088
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1090
_mod_revision.check_not_reserved_id(revision_id)
1091
if config is not None and config.signature_needed():
1093
inv = self.get_inventory(revision_id)
1094
plaintext = Testament(rev, inv).as_short_text()
1095
self.store_revision_signature(
1096
gpg.GPGStrategy(config), plaintext, revision_id)
1097
# check inventory present
1098
if not self.inventories.get_parent_map([(revision_id,)]):
1100
raise errors.WeaveRevisionNotPresent(revision_id,
1103
# yes, this is not suitable for adding with ghosts.
1104
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1107
key = (revision_id,)
1108
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1109
self._add_revision(rev)
1111
def _add_revision(self, revision):
1112
text = self._serializer.write_revision_to_string(revision)
1113
key = (revision.revision_id,)
1114
parents = tuple((parent,) for parent in revision.parent_ids)
1115
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1117
def all_revision_ids(self):
1118
"""Returns a list of all the revision ids in the repository.
1120
This is conceptually deprecated because code should generally work on
1121
the graph reachable from a particular revision, and ignore any other
1122
revisions that might be present. There is no direct replacement
1125
if 'evil' in debug.debug_flags:
1126
mutter_callsite(2, "all_revision_ids is linear with history.")
1127
return self._all_revision_ids()
1129
def _all_revision_ids(self):
1130
"""Returns a list of all the revision ids in the repository.
1132
These are in as much topological order as the underlying store can
1135
raise NotImplementedError(self._all_revision_ids)
1137
def break_lock(self):
1138
"""Break a lock if one is present from another instance.
1140
Uses the ui factory to ask for confirmation if the lock may be from
1143
self.control_files.break_lock()
1146
def _eliminate_revisions_not_present(self, revision_ids):
1147
"""Check every revision id in revision_ids to see if we have it.
1149
Returns a set of the present revisions.
1152
graph = self.get_graph()
1153
parent_map = graph.get_parent_map(revision_ids)
1154
# The old API returned a list, should this actually be a set?
1155
return parent_map.keys()
1158
def create(a_bzrdir):
1159
"""Construct the current default format repository in a_bzrdir."""
1160
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1162
def __init__(self, _format, a_bzrdir, control_files):
1163
"""instantiate a Repository.
1165
:param _format: The format of the repository on disk.
1166
:param a_bzrdir: The BzrDir of the repository.
1168
In the future we will have a single api for all stores for
1169
getting file texts, inventories and revisions, then
1170
this construct will accept instances of those things.
1172
super(Repository, self).__init__()
1173
self._format = _format
1174
# the following are part of the public API for Repository:
1175
self.bzrdir = a_bzrdir
1176
self.control_files = control_files
1177
self._transport = control_files._transport
1178
self.base = self._transport.base
1180
self._reconcile_does_inventory_gc = True
1181
self._reconcile_fixes_text_parents = False
1182
self._reconcile_backsup_inventory = True
1183
# not right yet - should be more semantically clear ?
1185
# TODO: make sure to construct the right store classes, etc, depending
1186
# on whether escaping is required.
1187
self._warn_if_deprecated()
1188
self._write_group = None
1189
# Additional places to query for data.
1190
self._fallback_repositories = []
1191
# An InventoryEntry cache, used during deserialization
1192
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1195
return '%s(%r)' % (self.__class__.__name__,
1198
def has_same_location(self, other):
1199
"""Returns a boolean indicating if this repository is at the same
1200
location as another repository.
1202
This might return False even when two repository objects are accessing
1203
the same physical repository via different URLs.
1205
if self.__class__ is not other.__class__:
1207
return (self._transport.base == other._transport.base)
1209
def is_in_write_group(self):
1210
"""Return True if there is an open write group.
1212
:seealso: start_write_group.
1214
return self._write_group is not None
1216
def is_locked(self):
1217
return self.control_files.is_locked()
1219
def is_write_locked(self):
1220
"""Return True if this object is write locked."""
1221
return self.is_locked() and self.control_files._lock_mode == 'w'
1223
def lock_write(self, token=None):
1224
"""Lock this repository for writing.
1226
This causes caching within the repository obejct to start accumlating
1227
data during reads, and allows a 'write_group' to be obtained. Write
1228
groups must be used for actual data insertion.
1230
:param token: if this is already locked, then lock_write will fail
1231
unless the token matches the existing lock.
1232
:returns: a token if this instance supports tokens, otherwise None.
1233
:raises TokenLockingNotSupported: when a token is given but this
1234
instance doesn't support using token locks.
1235
:raises MismatchedToken: if the specified token doesn't match the token
1236
of the existing lock.
1237
:seealso: start_write_group.
1239
A token should be passed in if you know that you have locked the object
1240
some other way, and need to synchronise this object's state with that
1243
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1245
locked = self.is_locked()
1246
result = self.control_files.lock_write(token=token)
1248
for repo in self._fallback_repositories:
1249
# Writes don't affect fallback repos
1251
self._refresh_data()
1254
def lock_read(self):
1255
locked = self.is_locked()
1256
self.control_files.lock_read()
1258
for repo in self._fallback_repositories:
1260
self._refresh_data()
1262
def get_physical_lock_status(self):
1263
return self.control_files.get_physical_lock_status()
1265
def leave_lock_in_place(self):
1266
"""Tell this repository not to release the physical lock when this
1269
If lock_write doesn't return a token, then this method is not supported.
1271
self.control_files.leave_in_place()
1273
def dont_leave_lock_in_place(self):
1274
"""Tell this repository to release the physical lock when this
1275
object is unlocked, even if it didn't originally acquire it.
1277
If lock_write doesn't return a token, then this method is not supported.
1279
self.control_files.dont_leave_in_place()
1282
def gather_stats(self, revid=None, committers=None):
1283
"""Gather statistics from a revision id.
1285
:param revid: The revision id to gather statistics from, if None, then
1286
no revision specific statistics are gathered.
1287
:param committers: Optional parameter controlling whether to grab
1288
a count of committers from the revision specific statistics.
1289
:return: A dictionary of statistics. Currently this contains:
1290
committers: The number of committers if requested.
1291
firstrev: A tuple with timestamp, timezone for the penultimate left
1292
most ancestor of revid, if revid is not the NULL_REVISION.
1293
latestrev: A tuple with timestamp, timezone for revid, if revid is
1294
not the NULL_REVISION.
1295
revisions: The total revision count in the repository.
1296
size: An estimate disk size of the repository in bytes.
1299
if revid and committers:
1300
result['committers'] = 0
1301
if revid and revid != _mod_revision.NULL_REVISION:
1303
all_committers = set()
1304
revisions = self.get_ancestry(revid)
1305
# pop the leading None
1307
first_revision = None
1309
# ignore the revisions in the middle - just grab first and last
1310
revisions = revisions[0], revisions[-1]
1311
for revision in self.get_revisions(revisions):
1312
if not first_revision:
1313
first_revision = revision
1315
all_committers.add(revision.committer)
1316
last_revision = revision
1318
result['committers'] = len(all_committers)
1319
result['firstrev'] = (first_revision.timestamp,
1320
first_revision.timezone)
1321
result['latestrev'] = (last_revision.timestamp,
1322
last_revision.timezone)
1324
# now gather global repository information
1325
# XXX: This is available for many repos regardless of listability.
1326
if self.bzrdir.root_transport.listable():
1327
# XXX: do we want to __define len__() ?
1328
# Maybe the versionedfiles object should provide a different
1329
# method to get the number of keys.
1330
result['revisions'] = len(self.revisions.keys())
1331
# result['size'] = t
1334
def find_branches(self, using=False):
1335
"""Find branches underneath this repository.
1337
This will include branches inside other branches.
1339
:param using: If True, list only branches using this repository.
1341
if using and not self.is_shared():
1343
return [self.bzrdir.open_branch()]
1344
except errors.NotBranchError:
1346
class Evaluator(object):
1349
self.first_call = True
1351
def __call__(self, bzrdir):
1352
# On the first call, the parameter is always the bzrdir
1353
# containing the current repo.
1354
if not self.first_call:
1356
repository = bzrdir.open_repository()
1357
except errors.NoRepositoryPresent:
1360
return False, (None, repository)
1361
self.first_call = False
1363
value = (bzrdir.open_branch(), None)
1364
except errors.NotBranchError:
1365
value = (None, None)
1369
for branch, repository in bzrdir.BzrDir.find_bzrdirs(
1370
self.bzrdir.root_transport, evaluate=Evaluator()):
1371
if branch is not None:
1372
branches.append(branch)
1373
if not using and repository is not None:
1374
branches.extend(repository.find_branches())
1378
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
1379
"""Return the revision ids that other has that this does not.
1381
These are returned in topological order.
1383
revision_id: only return revision ids included by revision_id.
1385
return InterRepository.get(other, self).search_missing_revision_ids(
1386
revision_id, find_ghosts)
1390
"""Open the repository rooted at base.
1392
For instance, if the repository is at URL/.bzr/repository,
1393
Repository.open(URL) -> a Repository instance.
1395
control = bzrdir.BzrDir.open(base)
1396
return control.open_repository()
1398
def copy_content_into(self, destination, revision_id=None):
1399
"""Make a complete copy of the content in self into destination.
1401
This is a destructive operation! Do not use it on existing
1404
return InterRepository.get(self, destination).copy_content(revision_id)
1406
def commit_write_group(self):
1407
"""Commit the contents accrued within the current write group.
1409
:seealso: start_write_group.
1411
if self._write_group is not self.get_transaction():
1412
# has an unlock or relock occured ?
1413
raise errors.BzrError('mismatched lock context %r and '
1415
(self.get_transaction(), self._write_group))
1416
result = self._commit_write_group()
1417
self._write_group = None
1420
def _commit_write_group(self):
1421
"""Template method for per-repository write group cleanup.
1423
This is called before the write group is considered to be
1424
finished and should ensure that all data handed to the repository
1425
for writing during the write group is safely committed (to the
1426
extent possible considering file system caching etc).
1429
def suspend_write_group(self):
1430
raise errors.UnsuspendableWriteGroup(self)
1432
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1433
"""Return the keys of missing inventory parents for revisions added in
1436
A revision is not complete if the inventory delta for that revision
1437
cannot be calculated. Therefore if the parent inventories of a
1438
revision are not present, the revision is incomplete, and e.g. cannot
1439
be streamed by a smart server. This method finds missing inventory
1440
parents for revisions added in this write group.
1442
if not self._format.supports_external_lookups:
1443
# This is only an issue for stacked repositories
1445
if not self.is_in_write_group():
1446
raise AssertionError('not in a write group')
1448
# XXX: We assume that every added revision already has its
1449
# corresponding inventory, so we only check for parent inventories that
1450
# might be missing, rather than all inventories.
1451
parents = set(self.revisions._index.get_missing_parents())
1452
parents.discard(_mod_revision.NULL_REVISION)
1453
unstacked_inventories = self.inventories._index
1454
present_inventories = unstacked_inventories.get_parent_map(
1455
key[-1:] for key in parents)
1456
parents.difference_update(present_inventories)
1457
if len(parents) == 0:
1458
# No missing parent inventories.
1460
if not check_for_missing_texts:
1461
return set(('inventories', rev_id) for (rev_id,) in parents)
1462
# Ok, now we have a list of missing inventories. But these only matter
1463
# if the inventories that reference them are missing some texts they
1464
# appear to introduce.
1465
# XXX: Texts referenced by all added inventories need to be present,
1466
# but at the moment we're only checking for texts referenced by
1467
# inventories at the graph's edge.
1468
key_deps = self.revisions._index._key_dependencies
1469
key_deps.add_keys(present_inventories)
1470
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1471
file_ids = self.fileids_altered_by_revision_ids(referrers)
1472
missing_texts = set()
1473
for file_id, version_ids in file_ids.iteritems():
1474
missing_texts.update(
1475
(file_id, version_id) for version_id in version_ids)
1476
present_texts = self.texts.get_parent_map(missing_texts)
1477
missing_texts.difference_update(present_texts)
1478
if not missing_texts:
1479
# No texts are missing, so all revisions and their deltas are
1482
# Alternatively the text versions could be returned as the missing
1483
# keys, but this is likely to be less data.
1484
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1487
def refresh_data(self):
1488
"""Re-read any data needed to to synchronise with disk.
1490
This method is intended to be called after another repository instance
1491
(such as one used by a smart server) has inserted data into the
1492
repository. It may not be called during a write group, but may be
1493
called at any other time.
1495
if self.is_in_write_group():
1496
raise errors.InternalBzrError(
1497
"May not refresh_data while in a write group.")
1498
self._refresh_data()
1500
def resume_write_group(self, tokens):
1501
if not self.is_write_locked():
1502
raise errors.NotWriteLocked(self)
1503
if self._write_group:
1504
raise errors.BzrError('already in a write group')
1505
self._resume_write_group(tokens)
1506
# so we can detect unlock/relock - the write group is now entered.
1507
self._write_group = self.get_transaction()
1509
def _resume_write_group(self, tokens):
1510
raise errors.UnsuspendableWriteGroup(self)
1512
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1514
"""Fetch the content required to construct revision_id from source.
1516
If revision_id is None and fetch_spec is None, then all content is
1519
fetch() may not be used when the repository is in a write group -
1520
either finish the current write group before using fetch, or use
1521
fetch before starting the write group.
1523
:param find_ghosts: Find and copy revisions in the source that are
1524
ghosts in the target (and not reachable directly by walking out to
1525
the first-present revision in target from revision_id).
1526
:param revision_id: If specified, all the content needed for this
1527
revision ID will be copied to the target. Fetch will determine for
1528
itself which content needs to be copied.
1529
:param fetch_spec: If specified, a SearchResult or
1530
PendingAncestryResult that describes which revisions to copy. This
1531
allows copying multiple heads at once. Mutually exclusive with
1534
if fetch_spec is not None and revision_id is not None:
1535
raise AssertionError(
1536
"fetch_spec and revision_id are mutually exclusive.")
1537
if self.is_in_write_group():
1538
raise errors.InternalBzrError(
1539
"May not fetch while in a write group.")
1540
# fast path same-url fetch operations
1541
if self.has_same_location(source) and fetch_spec is None:
1542
# check that last_revision is in 'from' and then return a
1544
if (revision_id is not None and
1545
not _mod_revision.is_null(revision_id)):
1546
self.get_revision(revision_id)
1548
# if there is no specific appropriate InterRepository, this will get
1549
# the InterRepository base class, which raises an
1550
# IncompatibleRepositories when asked to fetch.
1551
inter = InterRepository.get(source, self)
1552
return inter.fetch(revision_id=revision_id, pb=pb,
1553
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1555
def create_bundle(self, target, base, fileobj, format=None):
1556
return serializer.write_bundle(self, target, base, fileobj, format)
1558
def get_commit_builder(self, branch, parents, config, timestamp=None,
1559
timezone=None, committer=None, revprops=None,
1561
"""Obtain a CommitBuilder for this repository.
1563
:param branch: Branch to commit to.
1564
:param parents: Revision ids of the parents of the new revision.
1565
:param config: Configuration to use.
1566
:param timestamp: Optional timestamp recorded for commit.
1567
:param timezone: Optional timezone for timestamp.
1568
:param committer: Optional committer to set for commit.
1569
:param revprops: Optional dictionary of revision properties.
1570
:param revision_id: Optional revision id.
1572
result = self._commit_builder_class(self, parents, config,
1573
timestamp, timezone, committer, revprops, revision_id)
1574
self.start_write_group()
1578
if (self.control_files._lock_count == 1 and
1579
self.control_files._lock_mode == 'w'):
1580
if self._write_group is not None:
1581
self.abort_write_group()
1582
self.control_files.unlock()
1583
raise errors.BzrError(
1584
'Must end write groups before releasing write locks.')
1585
self.control_files.unlock()
1586
if self.control_files._lock_count == 0:
1587
self._inventory_entry_cache.clear()
1588
for repo in self._fallback_repositories:
1592
def clone(self, a_bzrdir, revision_id=None):
1593
"""Clone this repository into a_bzrdir using the current format.
1595
Currently no check is made that the format of this repository and
1596
the bzrdir format are compatible. FIXME RBC 20060201.
1598
:return: The newly created destination repository.
1600
# TODO: deprecate after 0.16; cloning this with all its settings is
1601
# probably not very useful -- mbp 20070423
1602
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1603
self.copy_content_into(dest_repo, revision_id)
1606
def start_write_group(self):
1607
"""Start a write group in the repository.
1609
Write groups are used by repositories which do not have a 1:1 mapping
1610
between file ids and backend store to manage the insertion of data from
1611
both fetch and commit operations.
1613
A write lock is required around the start_write_group/commit_write_group
1614
for the support of lock-requiring repository formats.
1616
One can only insert data into a repository inside a write group.
1620
if not self.is_write_locked():
1621
raise errors.NotWriteLocked(self)
1622
if self._write_group:
1623
raise errors.BzrError('already in a write group')
1624
self._start_write_group()
1625
# so we can detect unlock/relock - the write group is now entered.
1626
self._write_group = self.get_transaction()
1628
def _start_write_group(self):
1629
"""Template method for per-repository write group startup.
1631
This is called before the write group is considered to be
1636
def sprout(self, to_bzrdir, revision_id=None):
1637
"""Create a descendent repository for new development.
1639
Unlike clone, this does not copy the settings of the repository.
1641
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1642
dest_repo.fetch(self, revision_id=revision_id)
1645
def _create_sprouting_repo(self, a_bzrdir, shared):
1646
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1647
# use target default format.
1648
dest_repo = a_bzrdir.create_repository()
1650
# Most control formats need the repository to be specifically
1651
# created, but on some old all-in-one formats it's not needed
1653
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1654
except errors.UninitializableFormat:
1655
dest_repo = a_bzrdir.open_repository()
1658
def _get_sink(self):
1659
"""Return a sink for streaming into this repository."""
1660
return StreamSink(self)
1662
def _get_source(self, to_format):
1663
"""Return a source for streaming from this repository."""
1664
return StreamSource(self, to_format)
1667
def has_revision(self, revision_id):
1668
"""True if this repository has a copy of the revision."""
1669
return revision_id in self.has_revisions((revision_id,))
1672
def has_revisions(self, revision_ids):
1673
"""Probe to find out the presence of multiple revisions.
1675
:param revision_ids: An iterable of revision_ids.
1676
:return: A set of the revision_ids that were present.
1678
parent_map = self.revisions.get_parent_map(
1679
[(rev_id,) for rev_id in revision_ids])
1681
if _mod_revision.NULL_REVISION in revision_ids:
1682
result.add(_mod_revision.NULL_REVISION)
1683
result.update([key[0] for key in parent_map])
1687
def get_revision(self, revision_id):
1688
"""Return the Revision object for a named revision."""
1689
return self.get_revisions([revision_id])[0]
1692
def get_revision_reconcile(self, revision_id):
1693
"""'reconcile' helper routine that allows access to a revision always.
1695
This variant of get_revision does not cross check the weave graph
1696
against the revision one as get_revision does: but it should only
1697
be used by reconcile, or reconcile-alike commands that are correcting
1698
or testing the revision graph.
1700
return self._get_revisions([revision_id])[0]
1703
def get_revisions(self, revision_ids):
1704
"""Get many revisions at once."""
1705
return self._get_revisions(revision_ids)
1708
def _get_revisions(self, revision_ids):
1709
"""Core work logic to get many revisions without sanity checks."""
1710
for rev_id in revision_ids:
1711
if not rev_id or not isinstance(rev_id, basestring):
1712
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1713
keys = [(key,) for key in revision_ids]
1714
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1716
for record in stream:
1717
if record.storage_kind == 'absent':
1718
raise errors.NoSuchRevision(self, record.key[0])
1719
text = record.get_bytes_as('fulltext')
1720
rev = self._serializer.read_revision_from_string(text)
1721
revs[record.key[0]] = rev
1722
return [revs[revid] for revid in revision_ids]
1725
def get_revision_xml(self, revision_id):
1726
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1727
# would have already do it.
1728
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1729
# TODO: this can't just be replaced by:
1730
# return self._serializer.write_revision_to_string(
1731
# self.get_revision(revision_id))
1732
# as cStringIO preservers the encoding unlike write_revision_to_string
1733
# or some other call down the path.
1734
rev = self.get_revision(revision_id)
1735
rev_tmp = cStringIO.StringIO()
1736
# the current serializer..
1737
self._serializer.write_revision(rev, rev_tmp)
1739
return rev_tmp.getvalue()
1741
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1742
"""Produce a generator of revision deltas.
1744
Note that the input is a sequence of REVISIONS, not revision_ids.
1745
Trees will be held in memory until the generator exits.
1746
Each delta is relative to the revision's lefthand predecessor.
1748
:param specific_fileids: if not None, the result is filtered
1749
so that only those file-ids, their parents and their
1750
children are included.
1752
# Get the revision-ids of interest
1753
required_trees = set()
1754
for revision in revisions:
1755
required_trees.add(revision.revision_id)
1756
required_trees.update(revision.parent_ids[:1])
1758
# Get the matching filtered trees. Note that it's more
1759
# efficient to pass filtered trees to changes_from() rather
1760
# than doing the filtering afterwards. changes_from() could
1761
# arguably do the filtering itself but it's path-based, not
1762
# file-id based, so filtering before or afterwards is
1764
if specific_fileids is None:
1765
trees = dict((t.get_revision_id(), t) for
1766
t in self.revision_trees(required_trees))
1768
trees = dict((t.get_revision_id(), t) for
1769
t in self._filtered_revision_trees(required_trees,
1772
# Calculate the deltas
1773
for revision in revisions:
1774
if not revision.parent_ids:
1775
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1777
old_tree = trees[revision.parent_ids[0]]
1778
yield trees[revision.revision_id].changes_from(old_tree)
1781
def get_revision_delta(self, revision_id, specific_fileids=None):
1782
"""Return the delta for one revision.
1784
The delta is relative to the left-hand predecessor of the
1787
:param specific_fileids: if not None, the result is filtered
1788
so that only those file-ids, their parents and their
1789
children are included.
1791
r = self.get_revision(revision_id)
1792
return list(self.get_deltas_for_revisions([r],
1793
specific_fileids=specific_fileids))[0]
1796
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1797
signature = gpg_strategy.sign(plaintext)
1798
self.add_signature_text(revision_id, signature)
1801
def add_signature_text(self, revision_id, signature):
1802
self.signatures.add_lines((revision_id,), (),
1803
osutils.split_lines(signature))
1805
def find_text_key_references(self):
1806
"""Find the text key references within the repository.
1808
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1809
to whether they were referred to by the inventory of the
1810
revision_id that they contain. The inventory texts from all present
1811
revision ids are assessed to generate this report.
1813
revision_keys = self.revisions.keys()
1814
w = self.inventories
1815
pb = ui.ui_factory.nested_progress_bar()
1817
return self._find_text_key_references_from_xml_inventory_lines(
1818
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1822
def _find_text_key_references_from_xml_inventory_lines(self,
1824
"""Core routine for extracting references to texts from inventories.
1826
This performs the translation of xml lines to revision ids.
1828
:param line_iterator: An iterator of lines, origin_version_id
1829
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1830
to whether they were referred to by the inventory of the
1831
revision_id that they contain. Note that if that revision_id was
1832
not part of the line_iterator's output then False will be given -
1833
even though it may actually refer to that key.
1835
if not self._serializer.support_altered_by_hack:
1836
raise AssertionError(
1837
"_find_text_key_references_from_xml_inventory_lines only "
1838
"supported for branches which store inventory as unnested xml"
1839
", not on %r" % self)
1842
# this code needs to read every new line in every inventory for the
1843
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
1844
# not present in one of those inventories is unnecessary but not
1845
# harmful because we are filtering by the revision id marker in the
1846
# inventory lines : we only select file ids altered in one of those
1847
# revisions. We don't need to see all lines in the inventory because
1848
# only those added in an inventory in rev X can contain a revision=X
1850
unescape_revid_cache = {}
1851
unescape_fileid_cache = {}
1853
# jam 20061218 In a big fetch, this handles hundreds of thousands
1854
# of lines, so it has had a lot of inlining and optimizing done.
1855
# Sorry that it is a little bit messy.
1856
# Move several functions to be local variables, since this is a long
1858
search = self._file_ids_altered_regex.search
1859
unescape = _unescape_xml
1860
setdefault = result.setdefault
1861
for line, line_key in line_iterator:
1862
match = search(line)
1865
# One call to match.group() returning multiple items is quite a
1866
# bit faster than 2 calls to match.group() each returning 1
1867
file_id, revision_id = match.group('file_id', 'revision_id')
1869
# Inlining the cache lookups helps a lot when you make 170,000
1870
# lines and 350k ids, versus 8.4 unique ids.
1871
# Using a cache helps in 2 ways:
1872
# 1) Avoids unnecessary decoding calls
1873
# 2) Re-uses cached strings, which helps in future set and
1875
# (2) is enough that removing encoding entirely along with
1876
# the cache (so we are using plain strings) results in no
1877
# performance improvement.
1879
revision_id = unescape_revid_cache[revision_id]
1881
unescaped = unescape(revision_id)
1882
unescape_revid_cache[revision_id] = unescaped
1883
revision_id = unescaped
1885
# Note that unconditionally unescaping means that we deserialise
1886
# every fileid, which for general 'pull' is not great, but we don't
1887
# really want to have some many fulltexts that this matters anyway.
1890
file_id = unescape_fileid_cache[file_id]
1892
unescaped = unescape(file_id)
1893
unescape_fileid_cache[file_id] = unescaped
1896
key = (file_id, revision_id)
1897
setdefault(key, False)
1898
if revision_id == line_key[-1]:
1902
def _inventory_xml_lines_for_keys(self, keys):
1903
"""Get a line iterator of the sort needed for findind references.
1905
Not relevant for non-xml inventory repositories.
1907
Ghosts in revision_keys are ignored.
1909
:param revision_keys: The revision keys for the inventories to inspect.
1910
:return: An iterator over (inventory line, revid) for the fulltexts of
1911
all of the xml inventories specified by revision_keys.
1913
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1914
for record in stream:
1915
if record.storage_kind != 'absent':
1916
chunks = record.get_bytes_as('chunked')
1917
revid = record.key[-1]
1918
lines = osutils.chunks_to_lines(chunks)
1922
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1924
"""Helper routine for fileids_altered_by_revision_ids.
1926
This performs the translation of xml lines to revision ids.
1928
:param line_iterator: An iterator of lines, origin_version_id
1929
:param revision_keys: The revision ids to filter for. This should be a
1930
set or other type which supports efficient __contains__ lookups, as
1931
the revision key from each parsed line will be looked up in the
1932
revision_keys filter.
1933
:return: a dictionary mapping altered file-ids to an iterable of
1934
revision_ids. Each altered file-ids has the exact revision_ids that
1935
altered it listed explicitly.
1937
seen = set(self._find_text_key_references_from_xml_inventory_lines(
1938
line_iterator).iterkeys())
1939
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
1940
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
1941
self._inventory_xml_lines_for_keys(parent_keys)))
1942
new_keys = seen - parent_seen
1944
setdefault = result.setdefault
1945
for key in new_keys:
1946
setdefault(key[0], set()).add(key[-1])
1949
def _find_parent_ids_of_revisions(self, revision_ids):
1950
"""Find all parent ids that are mentioned in the revision graph.
1952
:return: set of revisions that are parents of revision_ids which are
1953
not part of revision_ids themselves
1955
parent_map = self.get_parent_map(revision_ids)
1957
map(parent_ids.update, parent_map.itervalues())
1958
parent_ids.difference_update(revision_ids)
1959
parent_ids.discard(_mod_revision.NULL_REVISION)
1962
def _find_parent_keys_of_revisions(self, revision_keys):
1963
"""Similar to _find_parent_ids_of_revisions, but used with keys.
1965
:param revision_keys: An iterable of revision_keys.
1966
:return: The parents of all revision_keys that are not already in
1969
parent_map = self.revisions.get_parent_map(revision_keys)
1971
map(parent_keys.update, parent_map.itervalues())
1972
parent_keys.difference_update(revision_keys)
1973
parent_keys.discard(_mod_revision.NULL_REVISION)
1976
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
1977
"""Find the file ids and versions affected by revisions.
1979
:param revisions: an iterable containing revision ids.
1980
:param _inv_weave: The inventory weave from this repository or None.
1981
If None, the inventory weave will be opened automatically.
1982
:return: a dictionary mapping altered file-ids to an iterable of
1983
revision_ids. Each altered file-ids has the exact revision_ids that
1984
altered it listed explicitly.
1986
selected_keys = set((revid,) for revid in revision_ids)
1987
w = _inv_weave or self.inventories
1988
pb = ui.ui_factory.nested_progress_bar()
1990
return self._find_file_ids_from_xml_inventory_lines(
1991
w.iter_lines_added_or_present_in_keys(
1992
selected_keys, pb=pb),
1997
def iter_files_bytes(self, desired_files):
1998
"""Iterate through file versions.
2000
Files will not necessarily be returned in the order they occur in
2001
desired_files. No specific order is guaranteed.
2003
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2004
value supplied by the caller as part of desired_files. It should
2005
uniquely identify the file version in the caller's context. (Examples:
2006
an index number or a TreeTransform trans_id.)
2008
bytes_iterator is an iterable of bytestrings for the file. The
2009
kind of iterable and length of the bytestrings are unspecified, but for
2010
this implementation, it is a list of bytes produced by
2011
VersionedFile.get_record_stream().
2013
:param desired_files: a list of (file_id, revision_id, identifier)
2017
for file_id, revision_id, callable_data in desired_files:
2018
text_keys[(file_id, revision_id)] = callable_data
2019
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2020
if record.storage_kind == 'absent':
2021
raise errors.RevisionNotPresent(record.key, self)
2022
yield text_keys[record.key], record.get_bytes_as('chunked')
2024
def _generate_text_key_index(self, text_key_references=None,
2026
"""Generate a new text key index for the repository.
2028
This is an expensive function that will take considerable time to run.
2030
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2031
list of parents, also text keys. When a given key has no parents,
2032
the parents list will be [NULL_REVISION].
2034
# All revisions, to find inventory parents.
2035
if ancestors is None:
2036
graph = self.get_graph()
2037
ancestors = graph.get_parent_map(self.all_revision_ids())
2038
if text_key_references is None:
2039
text_key_references = self.find_text_key_references()
2040
pb = ui.ui_factory.nested_progress_bar()
2042
return self._do_generate_text_key_index(ancestors,
2043
text_key_references, pb)
2047
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2048
"""Helper for _generate_text_key_index to avoid deep nesting."""
2049
revision_order = tsort.topo_sort(ancestors)
2050
invalid_keys = set()
2052
for revision_id in revision_order:
2053
revision_keys[revision_id] = set()
2054
text_count = len(text_key_references)
2055
# a cache of the text keys to allow reuse; costs a dict of all the
2056
# keys, but saves a 2-tuple for every child of a given key.
2058
for text_key, valid in text_key_references.iteritems():
2060
invalid_keys.add(text_key)
2062
revision_keys[text_key[1]].add(text_key)
2063
text_key_cache[text_key] = text_key
2064
del text_key_references
2066
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2067
NULL_REVISION = _mod_revision.NULL_REVISION
2068
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2069
# too small for large or very branchy trees. However, for 55K path
2070
# trees, it would be easy to use too much memory trivially. Ideally we
2071
# could gauge this by looking at available real memory etc, but this is
2072
# always a tricky proposition.
2073
inventory_cache = lru_cache.LRUCache(10)
2074
batch_size = 10 # should be ~150MB on a 55K path tree
2075
batch_count = len(revision_order) / batch_size + 1
2077
pb.update("Calculating text parents", processed_texts, text_count)
2078
for offset in xrange(batch_count):
2079
to_query = revision_order[offset * batch_size:(offset + 1) *
2083
for rev_tree in self.revision_trees(to_query):
2084
revision_id = rev_tree.get_revision_id()
2085
parent_ids = ancestors[revision_id]
2086
for text_key in revision_keys[revision_id]:
2087
pb.update("Calculating text parents", processed_texts)
2088
processed_texts += 1
2089
candidate_parents = []
2090
for parent_id in parent_ids:
2091
parent_text_key = (text_key[0], parent_id)
2093
check_parent = parent_text_key not in \
2094
revision_keys[parent_id]
2096
# the parent parent_id is a ghost:
2097
check_parent = False
2098
# truncate the derived graph against this ghost.
2099
parent_text_key = None
2101
# look at the parent commit details inventories to
2102
# determine possible candidates in the per file graph.
2105
inv = inventory_cache[parent_id]
2107
inv = self.revision_tree(parent_id).inventory
2108
inventory_cache[parent_id] = inv
2110
parent_entry = inv[text_key[0]]
2111
except (KeyError, errors.NoSuchId):
2113
if parent_entry is not None:
2115
text_key[0], parent_entry.revision)
2117
parent_text_key = None
2118
if parent_text_key is not None:
2119
candidate_parents.append(
2120
text_key_cache[parent_text_key])
2121
parent_heads = text_graph.heads(candidate_parents)
2122
new_parents = list(parent_heads)
2123
new_parents.sort(key=lambda x:candidate_parents.index(x))
2124
if new_parents == []:
2125
new_parents = [NULL_REVISION]
2126
text_index[text_key] = new_parents
2128
for text_key in invalid_keys:
2129
text_index[text_key] = [NULL_REVISION]
2132
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2133
"""Get an iterable listing the keys of all the data introduced by a set
2136
The keys will be ordered so that the corresponding items can be safely
2137
fetched and inserted in that order.
2139
:returns: An iterable producing tuples of (knit-kind, file-id,
2140
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2141
'revisions'. file-id is None unless knit-kind is 'file'.
2143
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2146
for result in self._find_non_file_keys_to_fetch(revision_ids):
2149
def _find_file_keys_to_fetch(self, revision_ids, pb):
2150
# XXX: it's a bit weird to control the inventory weave caching in this
2151
# generator. Ideally the caching would be done in fetch.py I think. Or
2152
# maybe this generator should explicitly have the contract that it
2153
# should not be iterated until the previously yielded item has been
2155
inv_w = self.inventories
2157
# file ids that changed
2158
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2160
num_file_ids = len(file_ids)
2161
for file_id, altered_versions in file_ids.iteritems():
2163
pb.update("fetch texts", count, num_file_ids)
2165
yield ("file", file_id, altered_versions)
2167
def _find_non_file_keys_to_fetch(self, revision_ids):
2169
yield ("inventory", None, revision_ids)
2172
# XXX: Note ATM no callers actually pay attention to this return
2173
# instead they just use the list of revision ids and ignore
2174
# missing sigs. Consider removing this work entirely
2175
revisions_with_signatures = set(self.signatures.get_parent_map(
2176
[(r,) for r in revision_ids]))
2177
revisions_with_signatures = set(
2178
[r for (r,) in revisions_with_signatures])
2179
revisions_with_signatures.intersection_update(revision_ids)
2180
yield ("signatures", None, revisions_with_signatures)
2183
yield ("revisions", None, revision_ids)
2186
def get_inventory(self, revision_id):
2187
"""Get Inventory object by revision id."""
2188
return self.iter_inventories([revision_id]).next()
2190
def iter_inventories(self, revision_ids):
2191
"""Get many inventories by revision_ids.
2193
This will buffer some or all of the texts used in constructing the
2194
inventories in memory, but will only parse a single inventory at a
2197
:param revision_ids: The expected revision ids of the inventories.
2198
:return: An iterator of inventories.
2200
if ((None in revision_ids)
2201
or (_mod_revision.NULL_REVISION in revision_ids)):
2202
raise ValueError('cannot get null revision inventory')
2203
return self._iter_inventories(revision_ids)
2205
def _iter_inventories(self, revision_ids):
2206
"""single-document based inventory iteration."""
2207
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2208
yield self.deserialise_inventory(revision_id, text)
2210
def _iter_inventory_xmls(self, revision_ids):
2211
keys = [(revision_id,) for revision_id in revision_ids]
2212
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2214
for record in stream:
2215
if record.storage_kind != 'absent':
2216
text_chunks[record.key] = record.get_bytes_as('chunked')
2218
raise errors.NoSuchRevision(self, record.key)
2220
chunks = text_chunks.pop(key)
2221
yield ''.join(chunks), key[-1]
2223
def deserialise_inventory(self, revision_id, xml):
2224
"""Transform the xml into an inventory object.
2226
:param revision_id: The expected revision id of the inventory.
2227
:param xml: A serialised inventory.
2229
result = self._serializer.read_inventory_from_string(xml, revision_id,
2230
entry_cache=self._inventory_entry_cache)
2231
if result.revision_id != revision_id:
2232
raise AssertionError('revision id mismatch %s != %s' % (
2233
result.revision_id, revision_id))
2236
def serialise_inventory(self, inv):
2237
return self._serializer.write_inventory_to_string(inv)
2239
def _serialise_inventory_to_lines(self, inv):
2240
return self._serializer.write_inventory_to_lines(inv)
2242
def get_serializer_format(self):
2243
return self._serializer.format_num
2246
def get_inventory_xml(self, revision_id):
2247
"""Get inventory XML as a file object."""
2248
texts = self._iter_inventory_xmls([revision_id])
2250
text, revision_id = texts.next()
2251
except StopIteration:
2252
raise errors.HistoryMissing(self, 'inventory', revision_id)
2256
def get_inventory_sha1(self, revision_id):
2257
"""Return the sha1 hash of the inventory entry
2259
return self.get_revision(revision_id).inventory_sha1
2261
def get_rev_id_for_revno(self, revno, known_pair):
2262
"""Return the revision id of a revno, given a later (revno, revid)
2263
pair in the same history.
2265
:return: if found (True, revid). If the available history ran out
2266
before reaching the revno, then this returns
2267
(False, (closest_revno, closest_revid)).
2269
known_revno, known_revid = known_pair
2270
partial_history = [known_revid]
2271
distance_from_known = known_revno - revno
2272
if distance_from_known < 0:
2274
'requested revno (%d) is later than given known revno (%d)'
2275
% (revno, known_revno))
2278
self, partial_history, stop_index=distance_from_known)
2279
except errors.RevisionNotPresent, err:
2280
if err.revision_id == known_revid:
2281
# The start revision (known_revid) wasn't found.
2283
# This is a stacked repository with no fallbacks, or a there's a
2284
# left-hand ghost. Either way, even though the revision named in
2285
# the error isn't in this repo, we know it's the next step in this
2286
# left-hand history.
2287
partial_history.append(err.revision_id)
2288
if len(partial_history) <= distance_from_known:
2289
# Didn't find enough history to get a revid for the revno.
2290
earliest_revno = known_revno - len(partial_history) + 1
2291
return (False, (earliest_revno, partial_history[-1]))
2292
if len(partial_history) - 1 > distance_from_known:
2293
raise AssertionError('_iter_for_revno returned too much history')
2294
return (True, partial_history[-1])
2296
def iter_reverse_revision_history(self, revision_id):
2297
"""Iterate backwards through revision ids in the lefthand history
2299
:param revision_id: The revision id to start with. All its lefthand
2300
ancestors will be traversed.
2302
graph = self.get_graph()
2303
next_id = revision_id
2305
if next_id in (None, _mod_revision.NULL_REVISION):
2308
parents = graph.get_parent_map([next_id])[next_id]
2310
raise errors.RevisionNotPresent(next_id, self)
2312
if len(parents) == 0:
2315
next_id = parents[0]
2318
def get_revision_inventory(self, revision_id):
2319
"""Return inventory of a past revision."""
2320
# TODO: Unify this with get_inventory()
2321
# bzr 0.0.6 and later imposes the constraint that the inventory_id
2322
# must be the same as its revision, so this is trivial.
2323
if revision_id is None:
2324
# This does not make sense: if there is no revision,
2325
# then it is the current tree inventory surely ?!
2326
# and thus get_root_id() is something that looks at the last
2327
# commit on the branch, and the get_root_id is an inventory check.
2328
raise NotImplementedError
2329
# return Inventory(self.get_root_id())
2331
return self.get_inventory(revision_id)
2333
def is_shared(self):
2334
"""Return True if this repository is flagged as a shared repository."""
2335
raise NotImplementedError(self.is_shared)
2338
def reconcile(self, other=None, thorough=False):
2339
"""Reconcile this repository."""
2340
from bzrlib.reconcile import RepoReconciler
2341
reconciler = RepoReconciler(self, thorough=thorough)
2342
reconciler.reconcile()
2345
def _refresh_data(self):
2346
"""Helper called from lock_* to ensure coherency with disk.
2348
The default implementation does nothing; it is however possible
2349
for repositories to maintain loaded indices across multiple locks
2350
by checking inside their implementation of this method to see
2351
whether their indices are still valid. This depends of course on
2352
the disk format being validatable in this manner. This method is
2353
also called by the refresh_data() public interface to cause a refresh
2354
to occur while in a write lock so that data inserted by a smart server
2355
push operation is visible on the client's instance of the physical
2360
def revision_tree(self, revision_id):
2361
"""Return Tree for a revision on this branch.
2363
`revision_id` may be NULL_REVISION for the empty tree revision.
2365
revision_id = _mod_revision.ensure_null(revision_id)
2366
# TODO: refactor this to use an existing revision object
2367
# so we don't need to read it in twice.
2368
if revision_id == _mod_revision.NULL_REVISION:
2369
return RevisionTree(self, Inventory(root_id=None),
2370
_mod_revision.NULL_REVISION)
2372
inv = self.get_revision_inventory(revision_id)
2373
return RevisionTree(self, inv, revision_id)
2375
def revision_trees(self, revision_ids):
2376
"""Return Trees for revisions in this repository.
2378
:param revision_ids: a sequence of revision-ids;
2379
a revision-id may not be None or 'null:'
2381
inventories = self.iter_inventories(revision_ids)
2382
for inv in inventories:
2383
yield RevisionTree(self, inv, inv.revision_id)
2385
def _filtered_revision_trees(self, revision_ids, file_ids):
2386
"""Return Tree for a revision on this branch with only some files.
2388
:param revision_ids: a sequence of revision-ids;
2389
a revision-id may not be None or 'null:'
2390
:param file_ids: if not None, the result is filtered
2391
so that only those file-ids, their parents and their
2392
children are included.
2394
inventories = self.iter_inventories(revision_ids)
2395
for inv in inventories:
2396
# Should we introduce a FilteredRevisionTree class rather
2397
# than pre-filter the inventory here?
2398
filtered_inv = inv.filter(file_ids)
2399
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2402
def get_ancestry(self, revision_id, topo_sorted=True):
2403
"""Return a list of revision-ids integrated by a revision.
2405
The first element of the list is always None, indicating the origin
2406
revision. This might change when we have history horizons, or
2407
perhaps we should have a new API.
2409
This is topologically sorted.
2411
if _mod_revision.is_null(revision_id):
2413
if not self.has_revision(revision_id):
2414
raise errors.NoSuchRevision(self, revision_id)
2415
graph = self.get_graph()
2417
search = graph._make_breadth_first_searcher([revision_id])
2420
found, ghosts = search.next_with_ghosts()
2421
except StopIteration:
2424
if _mod_revision.NULL_REVISION in keys:
2425
keys.remove(_mod_revision.NULL_REVISION)
2427
parent_map = graph.get_parent_map(keys)
2428
keys = tsort.topo_sort(parent_map)
2429
return [None] + list(keys)
2431
def pack(self, hint=None):
2432
"""Compress the data within the repository.
2434
This operation only makes sense for some repository types. For other
2435
types it should be a no-op that just returns.
2437
This stub method does not require a lock, but subclasses should use
2438
@needs_write_lock as this is a long running call its reasonable to
2439
implicitly lock for the user.
2441
:param hint: If not supplied, the whole repository is packed.
2442
If supplied, the repository may use the hint parameter as a
2443
hint for the parts of the repository to pack. A hint can be
2444
obtained from the result of commit_write_group(). Out of
2445
date hints are simply ignored, because concurrent operations
2446
can obsolete them rapidly.
2449
def get_transaction(self):
2450
return self.control_files.get_transaction()
2452
def get_parent_map(self, revision_ids):
2453
"""See graph.StackedParentsProvider.get_parent_map"""
2454
# revisions index works in keys; this just works in revisions
2455
# therefore wrap and unwrap
2458
for revision_id in revision_ids:
2459
if revision_id == _mod_revision.NULL_REVISION:
2460
result[revision_id] = ()
2461
elif revision_id is None:
2462
raise ValueError('get_parent_map(None) is not valid')
2464
query_keys.append((revision_id ,))
2465
for ((revision_id,), parent_keys) in \
2466
self.revisions.get_parent_map(query_keys).iteritems():
2468
result[revision_id] = tuple(parent_revid
2469
for (parent_revid,) in parent_keys)
2471
result[revision_id] = (_mod_revision.NULL_REVISION,)
2474
def _make_parents_provider(self):
2477
def get_graph(self, other_repository=None):
2478
"""Return the graph walker for this repository format"""
2479
parents_provider = self._make_parents_provider()
2480
if (other_repository is not None and
2481
not self.has_same_location(other_repository)):
2482
parents_provider = graph.StackedParentsProvider(
2483
[parents_provider, other_repository._make_parents_provider()])
2484
return graph.Graph(parents_provider)
2486
def _get_versioned_file_checker(self, text_key_references=None):
2487
"""Return an object suitable for checking versioned files.
2489
:param text_key_references: if non-None, an already built
2490
dictionary mapping text keys ((fileid, revision_id) tuples)
2491
to whether they were referred to by the inventory of the
2492
revision_id that they contain. If None, this will be
2495
return _VersionedFileChecker(self,
2496
text_key_references=text_key_references)
2498
def revision_ids_to_search_result(self, result_set):
2499
"""Convert a set of revision ids to a graph SearchResult."""
2500
result_parents = set()
2501
for parents in self.get_graph().get_parent_map(
2502
result_set).itervalues():
2503
result_parents.update(parents)
2504
included_keys = result_set.intersection(result_parents)
2505
start_keys = result_set.difference(included_keys)
2506
exclude_keys = result_parents.difference(result_set)
2507
result = graph.SearchResult(start_keys, exclude_keys,
2508
len(result_set), result_set)
2512
def set_make_working_trees(self, new_value):
2513
"""Set the policy flag for making working trees when creating branches.
2515
This only applies to branches that use this repository.
2517
The default is 'True'.
2518
:param new_value: True to restore the default, False to disable making
2521
raise NotImplementedError(self.set_make_working_trees)
2523
def make_working_trees(self):
2524
"""Returns the policy for making working trees on new branches."""
2525
raise NotImplementedError(self.make_working_trees)
2528
def sign_revision(self, revision_id, gpg_strategy):
2529
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2530
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2533
def has_signature_for_revision_id(self, revision_id):
2534
"""Query for a revision signature for revision_id in the repository."""
2535
if not self.has_revision(revision_id):
2536
raise errors.NoSuchRevision(self, revision_id)
2537
sig_present = (1 == len(
2538
self.signatures.get_parent_map([(revision_id,)])))
2542
def get_signature_text(self, revision_id):
2543
"""Return the text for a signature."""
2544
stream = self.signatures.get_record_stream([(revision_id,)],
2546
record = stream.next()
2547
if record.storage_kind == 'absent':
2548
raise errors.NoSuchRevision(self, revision_id)
2549
return record.get_bytes_as('fulltext')
2552
def check(self, revision_ids=None):
2553
"""Check consistency of all history of given revision_ids.
2555
Different repository implementations should override _check().
2557
:param revision_ids: A non-empty list of revision_ids whose ancestry
2558
will be checked. Typically the last revision_id of a branch.
2560
return self._check(revision_ids)
2562
def _check(self, revision_ids):
2563
result = check.Check(self)
2567
def _warn_if_deprecated(self):
2568
global _deprecation_warning_done
2569
if _deprecation_warning_done:
2571
_deprecation_warning_done = True
2572
warning("Format %s for %s is deprecated - please use 'bzr upgrade' to get better performance"
2573
% (self._format, self.bzrdir.transport.base))
2575
def supports_rich_root(self):
2576
return self._format.rich_root_data
2578
def _check_ascii_revisionid(self, revision_id, method):
2579
"""Private helper for ascii-only repositories."""
2580
# weave repositories refuse to store revisionids that are non-ascii.
2581
if revision_id is not None:
2582
# weaves require ascii revision ids.
2583
if isinstance(revision_id, unicode):
2585
revision_id.encode('ascii')
2586
except UnicodeEncodeError:
2587
raise errors.NonAsciiRevisionId(method, self)
2590
revision_id.decode('ascii')
2591
except UnicodeDecodeError:
2592
raise errors.NonAsciiRevisionId(method, self)
2594
def revision_graph_can_have_wrong_parents(self):
2595
"""Is it possible for this repository to have a revision graph with
2598
If True, then this repository must also implement
2599
_find_inconsistent_revision_parents so that check and reconcile can
2600
check for inconsistencies before proceeding with other checks that may
2601
depend on the revision index being consistent.
2603
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2606
# remove these delegates a while after bzr 0.15
2607
def __make_delegated(name, from_module):
2608
def _deprecated_repository_forwarder():
2609
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2610
% (name, from_module),
2613
m = __import__(from_module, globals(), locals(), [name])
2615
return getattr(m, name)
2616
except AttributeError:
2617
raise AttributeError('module %s has no name %s'
2619
globals()[name] = _deprecated_repository_forwarder
2622
'AllInOneRepository',
2623
'WeaveMetaDirRepository',
2624
'PreSplitOutRepositoryFormat',
2625
'RepositoryFormat4',
2626
'RepositoryFormat5',
2627
'RepositoryFormat6',
2628
'RepositoryFormat7',
2630
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2634
'RepositoryFormatKnit',
2635
'RepositoryFormatKnit1',
2637
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2640
def install_revision(repository, rev, revision_tree):
2641
"""Install all revision data into a repository."""
2642
install_revisions(repository, [(rev, revision_tree, None)])
2645
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2646
"""Install all revision data into a repository.
2648
Accepts an iterable of revision, tree, signature tuples. The signature
2651
repository.start_write_group()
2653
inventory_cache = lru_cache.LRUCache(10)
2654
for n, (revision, revision_tree, signature) in enumerate(iterable):
2655
_install_revision(repository, revision, revision_tree, signature,
2658
pb.update('Transferring revisions', n + 1, num_revisions)
2660
repository.abort_write_group()
2663
repository.commit_write_group()
2666
def _install_revision(repository, rev, revision_tree, signature,
2668
"""Install all revision data into a repository."""
2669
present_parents = []
2671
for p_id in rev.parent_ids:
2672
if repository.has_revision(p_id):
2673
present_parents.append(p_id)
2674
parent_trees[p_id] = repository.revision_tree(p_id)
2676
parent_trees[p_id] = repository.revision_tree(
2677
_mod_revision.NULL_REVISION)
2679
inv = revision_tree.inventory
2680
entries = inv.iter_entries()
2681
# backwards compatibility hack: skip the root id.
2682
if not repository.supports_rich_root():
2683
path, root = entries.next()
2684
if root.revision != rev.revision_id:
2685
raise errors.IncompatibleRevision(repr(repository))
2687
for path, ie in entries:
2688
text_keys[(ie.file_id, ie.revision)] = ie
2689
text_parent_map = repository.texts.get_parent_map(text_keys)
2690
missing_texts = set(text_keys) - set(text_parent_map)
2691
# Add the texts that are not already present
2692
for text_key in missing_texts:
2693
ie = text_keys[text_key]
2695
# FIXME: TODO: The following loop overlaps/duplicates that done by
2696
# commit to determine parents. There is a latent/real bug here where
2697
# the parents inserted are not those commit would do - in particular
2698
# they are not filtered by heads(). RBC, AB
2699
for revision, tree in parent_trees.iteritems():
2700
if ie.file_id not in tree:
2702
parent_id = tree.inventory[ie.file_id].revision
2703
if parent_id in text_parents:
2705
text_parents.append((ie.file_id, parent_id))
2706
lines = revision_tree.get_file(ie.file_id).readlines()
2707
repository.texts.add_lines(text_key, text_parents, lines)
2709
# install the inventory
2710
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2711
# Cache this inventory
2712
inventory_cache[rev.revision_id] = inv
2714
basis_inv = inventory_cache[rev.parent_ids[0]]
2716
repository.add_inventory(rev.revision_id, inv, present_parents)
2718
delta = inv._make_delta(basis_inv)
2719
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2720
rev.revision_id, present_parents)
2722
repository.add_inventory(rev.revision_id, inv, present_parents)
2723
except errors.RevisionAlreadyPresent:
2725
if signature is not None:
2726
repository.add_signature_text(rev.revision_id, signature)
2727
repository.add_revision(rev.revision_id, rev, inv)
2730
class MetaDirRepository(Repository):
2731
"""Repositories in the new meta-dir layout.
2733
:ivar _transport: Transport for access to repository control files,
2734
typically pointing to .bzr/repository.
2737
def __init__(self, _format, a_bzrdir, control_files):
2738
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2739
self._transport = control_files._transport
2741
def is_shared(self):
2742
"""Return True if this repository is flagged as a shared repository."""
2743
return self._transport.has('shared-storage')
2746
def set_make_working_trees(self, new_value):
2747
"""Set the policy flag for making working trees when creating branches.
2749
This only applies to branches that use this repository.
2751
The default is 'True'.
2752
:param new_value: True to restore the default, False to disable making
2757
self._transport.delete('no-working-trees')
2758
except errors.NoSuchFile:
2761
self._transport.put_bytes('no-working-trees', '',
2762
mode=self.bzrdir._get_file_mode())
2764
def make_working_trees(self):
2765
"""Returns the policy for making working trees on new branches."""
2766
return not self._transport.has('no-working-trees')
2769
class MetaDirVersionedFileRepository(MetaDirRepository):
2770
"""Repositories in a meta-dir, that work via versioned file objects."""
2772
def __init__(self, _format, a_bzrdir, control_files):
2773
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2777
network_format_registry = registry.FormatRegistry()
2778
"""Registry of formats indexed by their network name.
2780
The network name for a repository format is an identifier that can be used when
2781
referring to formats with smart server operations. See
2782
RepositoryFormat.network_name() for more detail.
2786
format_registry = registry.FormatRegistry(network_format_registry)
2787
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
2789
This can contain either format instances themselves, or classes/factories that
2790
can be called to obtain one.
2794
#####################################################################
2795
# Repository Formats
2797
class RepositoryFormat(object):
2798
"""A repository format.
2800
Formats provide four things:
2801
* An initialization routine to construct repository data on disk.
2802
* a optional format string which is used when the BzrDir supports
2804
* an open routine which returns a Repository instance.
2805
* A network name for referring to the format in smart server RPC
2808
There is one and only one Format subclass for each on-disk format. But
2809
there can be one Repository subclass that is used for several different
2810
formats. The _format attribute on a Repository instance can be used to
2811
determine the disk format.
2813
Formats are placed in a registry by their format string for reference
2814
during opening. These should be subclasses of RepositoryFormat for
2817
Once a format is deprecated, just deprecate the initialize and open
2818
methods on the format class. Do not deprecate the object, as the
2819
object may be created even when a repository instance hasn't been
2822
Common instance attributes:
2823
_matchingbzrdir - the bzrdir format that the repository format was
2824
originally written to work with. This can be used if manually
2825
constructing a bzrdir and repository, or more commonly for test suite
2829
# Set to True or False in derived classes. True indicates that the format
2830
# supports ghosts gracefully.
2831
supports_ghosts = None
2832
# Can this repository be given external locations to lookup additional
2833
# data. Set to True or False in derived classes.
2834
supports_external_lookups = None
2835
# Does this format support CHK bytestring lookups. Set to True or False in
2837
supports_chks = None
2838
# Should commit add an inventory, or an inventory delta to the repository.
2839
_commit_inv_deltas = True
2840
# What order should fetch operations request streams in?
2841
# The default is unordered as that is the cheapest for an origin to
2843
_fetch_order = 'unordered'
2844
# Does this repository format use deltas that can be fetched as-deltas ?
2845
# (E.g. knits, where the knit deltas can be transplanted intact.
2846
# We default to False, which will ensure that enough data to get
2847
# a full text out of any fetch stream will be grabbed.
2848
_fetch_uses_deltas = False
2849
# Should fetch trigger a reconcile after the fetch? Only needed for
2850
# some repository formats that can suffer internal inconsistencies.
2851
_fetch_reconcile = False
2852
# Does this format have < O(tree_size) delta generation. Used to hint what
2853
# code path for commit, amongst other things.
2855
# Does doing a pack operation compress data? Useful for the pack UI command
2856
# (so if there is one pack, the operation can still proceed because it may
2857
# help), and for fetching when data won't have come from the same
2859
pack_compresses = False
2862
return "<%s>" % self.__class__.__name__
2864
def __eq__(self, other):
2865
# format objects are generally stateless
2866
return isinstance(other, self.__class__)
2868
def __ne__(self, other):
2869
return not self == other
2872
def find_format(klass, a_bzrdir):
2873
"""Return the format for the repository object in a_bzrdir.
2875
This is used by bzr native formats that have a "format" file in
2876
the repository. Other methods may be used by different types of
2880
transport = a_bzrdir.get_repository_transport(None)
2881
format_string = transport.get("format").read()
2882
return format_registry.get(format_string)
2883
except errors.NoSuchFile:
2884
raise errors.NoRepositoryPresent(a_bzrdir)
2886
raise errors.UnknownFormatError(format=format_string,
2890
def register_format(klass, format):
2891
format_registry.register(format.get_format_string(), format)
2894
def unregister_format(klass, format):
2895
format_registry.remove(format.get_format_string())
2898
def get_default_format(klass):
2899
"""Return the current default format."""
2900
from bzrlib import bzrdir
2901
return bzrdir.format_registry.make_bzrdir('default').repository_format
2903
def get_format_string(self):
2904
"""Return the ASCII format string that identifies this format.
2906
Note that in pre format ?? repositories the format string is
2907
not permitted nor written to disk.
2909
raise NotImplementedError(self.get_format_string)
2911
def get_format_description(self):
2912
"""Return the short description for this format."""
2913
raise NotImplementedError(self.get_format_description)
2915
# TODO: this shouldn't be in the base class, it's specific to things that
2916
# use weaves or knits -- mbp 20070207
2917
def _get_versioned_file_store(self,
2922
versionedfile_class=None,
2923
versionedfile_kwargs={},
2925
if versionedfile_class is None:
2926
versionedfile_class = self._versionedfile_class
2927
weave_transport = control_files._transport.clone(name)
2928
dir_mode = control_files._dir_mode
2929
file_mode = control_files._file_mode
2930
return VersionedFileStore(weave_transport, prefixed=prefixed,
2932
file_mode=file_mode,
2933
versionedfile_class=versionedfile_class,
2934
versionedfile_kwargs=versionedfile_kwargs,
2937
def initialize(self, a_bzrdir, shared=False):
2938
"""Initialize a repository of this format in a_bzrdir.
2940
:param a_bzrdir: The bzrdir to put the new repository in it.
2941
:param shared: The repository should be initialized as a sharable one.
2942
:returns: The new repository object.
2944
This may raise UninitializableFormat if shared repository are not
2945
compatible the a_bzrdir.
2947
raise NotImplementedError(self.initialize)
2949
def is_supported(self):
2950
"""Is this format supported?
2952
Supported formats must be initializable and openable.
2953
Unsupported formats may not support initialization or committing or
2954
some other features depending on the reason for not being supported.
2958
def network_name(self):
2959
"""A simple byte string uniquely identifying this format for RPC calls.
2961
MetaDir repository formats use their disk format string to identify the
2962
repository over the wire. All in one formats such as bzr < 0.8, and
2963
foreign formats like svn/git and hg should use some marker which is
2964
unique and immutable.
2966
raise NotImplementedError(self.network_name)
2968
def check_conversion_target(self, target_format):
2969
raise NotImplementedError(self.check_conversion_target)
2971
def open(self, a_bzrdir, _found=False):
2972
"""Return an instance of this format for the bzrdir a_bzrdir.
2974
_found is a private parameter, do not use it.
2976
raise NotImplementedError(self.open)
2979
class MetaDirRepositoryFormat(RepositoryFormat):
2980
"""Common base class for the new repositories using the metadir layout."""
2982
rich_root_data = False
2983
supports_tree_reference = False
2984
supports_external_lookups = False
2987
def _matchingbzrdir(self):
2988
matching = bzrdir.BzrDirMetaFormat1()
2989
matching.repository_format = self
2993
super(MetaDirRepositoryFormat, self).__init__()
2995
def _create_control_files(self, a_bzrdir):
2996
"""Create the required files and the initial control_files object."""
2997
# FIXME: RBC 20060125 don't peek under the covers
2998
# NB: no need to escape relative paths that are url safe.
2999
repository_transport = a_bzrdir.get_repository_transport(self)
3000
control_files = lockable_files.LockableFiles(repository_transport,
3001
'lock', lockdir.LockDir)
3002
control_files.create_lock()
3003
return control_files
3005
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
3006
"""Upload the initial blank content."""
3007
control_files = self._create_control_files(a_bzrdir)
3008
control_files.lock_write()
3009
transport = control_files._transport
3011
utf8_files += [('shared-storage', '')]
3013
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
3014
for (filename, content_stream) in files:
3015
transport.put_file(filename, content_stream,
3016
mode=a_bzrdir._get_file_mode())
3017
for (filename, content_bytes) in utf8_files:
3018
transport.put_bytes_non_atomic(filename, content_bytes,
3019
mode=a_bzrdir._get_file_mode())
3021
control_files.unlock()
3023
def network_name(self):
3024
"""Metadir formats have matching disk and network format strings."""
3025
return self.get_format_string()
3028
# Pre-0.8 formats that don't have a disk format string (because they are
3029
# versioned by the matching control directory). We use the control directories
3030
# disk format string as a key for the network_name because they meet the
3031
# constraints (simple string, unique, immutable).
3032
network_format_registry.register_lazy(
3033
"Bazaar-NG branch, format 5\n",
3034
'bzrlib.repofmt.weaverepo',
3035
'RepositoryFormat5',
3037
network_format_registry.register_lazy(
3038
"Bazaar-NG branch, format 6\n",
3039
'bzrlib.repofmt.weaverepo',
3040
'RepositoryFormat6',
3043
# formats which have no format string are not discoverable or independently
3044
# creatable on disk, so are not registered in format_registry. They're
3045
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
3046
# needed, it's constructed directly by the BzrDir. Non-native formats where
3047
# the repository is not separately opened are similar.
3049
format_registry.register_lazy(
3050
'Bazaar-NG Repository format 7',
3051
'bzrlib.repofmt.weaverepo',
3055
format_registry.register_lazy(
3056
'Bazaar-NG Knit Repository Format 1',
3057
'bzrlib.repofmt.knitrepo',
3058
'RepositoryFormatKnit1',
3061
format_registry.register_lazy(
3062
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
3063
'bzrlib.repofmt.knitrepo',
3064
'RepositoryFormatKnit3',
3067
format_registry.register_lazy(
3068
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
3069
'bzrlib.repofmt.knitrepo',
3070
'RepositoryFormatKnit4',
3073
# Pack-based formats. There is one format for pre-subtrees, and one for
3074
# post-subtrees to allow ease of testing.
3075
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
3076
format_registry.register_lazy(
3077
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
3078
'bzrlib.repofmt.pack_repo',
3079
'RepositoryFormatKnitPack1',
3081
format_registry.register_lazy(
3082
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
3083
'bzrlib.repofmt.pack_repo',
3084
'RepositoryFormatKnitPack3',
3086
format_registry.register_lazy(
3087
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
3088
'bzrlib.repofmt.pack_repo',
3089
'RepositoryFormatKnitPack4',
3091
format_registry.register_lazy(
3092
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
3093
'bzrlib.repofmt.pack_repo',
3094
'RepositoryFormatKnitPack5',
3096
format_registry.register_lazy(
3097
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
3098
'bzrlib.repofmt.pack_repo',
3099
'RepositoryFormatKnitPack5RichRoot',
3101
format_registry.register_lazy(
3102
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3103
'bzrlib.repofmt.pack_repo',
3104
'RepositoryFormatKnitPack5RichRootBroken',
3106
format_registry.register_lazy(
3107
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3108
'bzrlib.repofmt.pack_repo',
3109
'RepositoryFormatKnitPack6',
3111
format_registry.register_lazy(
3112
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3113
'bzrlib.repofmt.pack_repo',
3114
'RepositoryFormatKnitPack6RichRoot',
3117
# Development formats.
3118
# Obsolete but kept pending a CHK based subtree format.
3119
format_registry.register_lazy(
3120
("Bazaar development format 2 with subtree support "
3121
"(needs bzr.dev from before 1.8)\n"),
3122
'bzrlib.repofmt.pack_repo',
3123
'RepositoryFormatPackDevelopment2Subtree',
3126
# 1.14->1.16 go below here
3127
format_registry.register_lazy(
3128
'Bazaar development format - group compression and chk inventory'
3129
' (needs bzr.dev from 1.14)\n',
3130
'bzrlib.repofmt.groupcompress_repo',
3131
'RepositoryFormatCHK1',
3134
format_registry.register_lazy(
3135
'Bazaar development format - chk repository with bencode revision '
3136
'serialization (needs bzr.dev from 1.16)\n',
3137
'bzrlib.repofmt.groupcompress_repo',
3138
'RepositoryFormatCHK2',
3140
format_registry.register_lazy(
3141
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3142
'bzrlib.repofmt.groupcompress_repo',
3143
'RepositoryFormat2a',
3147
class InterRepository(InterObject):
3148
"""This class represents operations taking place between two repositories.
3150
Its instances have methods like copy_content and fetch, and contain
3151
references to the source and target repositories these operations can be
3154
Often we will provide convenience methods on 'repository' which carry out
3155
operations with another repository - they will always forward to
3156
InterRepository.get(other).method_name(parameters).
3159
_walk_to_common_revisions_batch_size = 50
3161
"""The available optimised InterRepository types."""
3164
def copy_content(self, revision_id=None):
3165
"""Make a complete copy of the content in self into destination.
3167
This is a destructive operation! Do not use it on existing
3170
:param revision_id: Only copy the content needed to construct
3171
revision_id and its parents.
3174
self.target.set_make_working_trees(self.source.make_working_trees())
3175
except NotImplementedError:
3177
self.target.fetch(self.source, revision_id=revision_id)
3180
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3182
"""Fetch the content required to construct revision_id.
3184
The content is copied from self.source to self.target.
3186
:param revision_id: if None all content is copied, if NULL_REVISION no
3188
:param pb: optional progress bar to use for progress reports. If not
3189
provided a default one will be created.
3192
from bzrlib.fetch import RepoFetcher
3193
f = RepoFetcher(to_repository=self.target,
3194
from_repository=self.source,
3195
last_revision=revision_id,
3196
fetch_spec=fetch_spec,
3197
pb=pb, find_ghosts=find_ghosts)
3199
def _walk_to_common_revisions(self, revision_ids):
3200
"""Walk out from revision_ids in source to revisions target has.
3202
:param revision_ids: The start point for the search.
3203
:return: A set of revision ids.
3205
target_graph = self.target.get_graph()
3206
revision_ids = frozenset(revision_ids)
3207
missing_revs = set()
3208
source_graph = self.source.get_graph()
3209
# ensure we don't pay silly lookup costs.
3210
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3211
null_set = frozenset([_mod_revision.NULL_REVISION])
3212
searcher_exhausted = False
3216
# Iterate the searcher until we have enough next_revs
3217
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3219
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3220
next_revs.update(next_revs_part)
3221
ghosts.update(ghosts_part)
3222
except StopIteration:
3223
searcher_exhausted = True
3225
# If there are ghosts in the source graph, and the caller asked for
3226
# them, make sure that they are present in the target.
3227
# We don't care about other ghosts as we can't fetch them and
3228
# haven't been asked to.
3229
ghosts_to_check = set(revision_ids.intersection(ghosts))
3230
revs_to_get = set(next_revs).union(ghosts_to_check)
3232
have_revs = set(target_graph.get_parent_map(revs_to_get))
3233
# we always have NULL_REVISION present.
3234
have_revs = have_revs.union(null_set)
3235
# Check if the target is missing any ghosts we need.
3236
ghosts_to_check.difference_update(have_revs)
3238
# One of the caller's revision_ids is a ghost in both the
3239
# source and the target.
3240
raise errors.NoSuchRevision(
3241
self.source, ghosts_to_check.pop())
3242
missing_revs.update(next_revs - have_revs)
3243
# Because we may have walked past the original stop point, make
3244
# sure everything is stopped
3245
stop_revs = searcher.find_seen_ancestors(have_revs)
3246
searcher.stop_searching_any(stop_revs)
3247
if searcher_exhausted:
3249
return searcher.get_result()
3252
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3253
"""Return the revision ids that source has that target does not.
3255
:param revision_id: only return revision ids included by this
3257
:param find_ghosts: If True find missing revisions in deep history
3258
rather than just finding the surface difference.
3259
:return: A bzrlib.graph.SearchResult.
3261
# stop searching at found target revisions.
3262
if not find_ghosts and revision_id is not None:
3263
return self._walk_to_common_revisions([revision_id])
3264
# generic, possibly worst case, slow code path.
3265
target_ids = set(self.target.all_revision_ids())
3266
if revision_id is not None:
3267
source_ids = self.source.get_ancestry(revision_id)
3268
if source_ids[0] is not None:
3269
raise AssertionError()
3272
source_ids = self.source.all_revision_ids()
3273
result_set = set(source_ids).difference(target_ids)
3274
return self.source.revision_ids_to_search_result(result_set)
3277
def _same_model(source, target):
3278
"""True if source and target have the same data representation.
3280
Note: this is always called on the base class; overriding it in a
3281
subclass will have no effect.
3284
InterRepository._assert_same_model(source, target)
3286
except errors.IncompatibleRepositories, e:
3290
def _assert_same_model(source, target):
3291
"""Raise an exception if two repositories do not use the same model.
3293
if source.supports_rich_root() != target.supports_rich_root():
3294
raise errors.IncompatibleRepositories(source, target,
3295
"different rich-root support")
3296
if source._serializer != target._serializer:
3297
raise errors.IncompatibleRepositories(source, target,
3298
"different serializers")
3301
class InterSameDataRepository(InterRepository):
3302
"""Code for converting between repositories that represent the same data.
3304
Data format and model must match for this to work.
3308
def _get_repo_format_to_test(self):
3309
"""Repository format for testing with.
3311
InterSameData can pull from subtree to subtree and from non-subtree to
3312
non-subtree, so we test this with the richest repository format.
3314
from bzrlib.repofmt import knitrepo
3315
return knitrepo.RepositoryFormatKnit3()
3318
def is_compatible(source, target):
3319
return InterRepository._same_model(source, target)
3322
class InterWeaveRepo(InterSameDataRepository):
3323
"""Optimised code paths between Weave based repositories.
3325
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3326
implemented lazy inter-object optimisation.
3330
def _get_repo_format_to_test(self):
3331
from bzrlib.repofmt import weaverepo
3332
return weaverepo.RepositoryFormat7()
3335
def is_compatible(source, target):
3336
"""Be compatible with known Weave formats.
3338
We don't test for the stores being of specific types because that
3339
could lead to confusing results, and there is no need to be
3342
from bzrlib.repofmt.weaverepo import (
3348
return (isinstance(source._format, (RepositoryFormat5,
3350
RepositoryFormat7)) and
3351
isinstance(target._format, (RepositoryFormat5,
3353
RepositoryFormat7)))
3354
except AttributeError:
3358
def copy_content(self, revision_id=None):
3359
"""See InterRepository.copy_content()."""
3360
# weave specific optimised path:
3362
self.target.set_make_working_trees(self.source.make_working_trees())
3363
except (errors.RepositoryUpgradeRequired, NotImplemented):
3365
# FIXME do not peek!
3366
if self.source._transport.listable():
3367
pb = ui.ui_factory.nested_progress_bar()
3369
self.target.texts.insert_record_stream(
3370
self.source.texts.get_record_stream(
3371
self.source.texts.keys(), 'topological', False))
3372
pb.update('copying inventory', 0, 1)
3373
self.target.inventories.insert_record_stream(
3374
self.source.inventories.get_record_stream(
3375
self.source.inventories.keys(), 'topological', False))
3376
self.target.signatures.insert_record_stream(
3377
self.source.signatures.get_record_stream(
3378
self.source.signatures.keys(),
3380
self.target.revisions.insert_record_stream(
3381
self.source.revisions.get_record_stream(
3382
self.source.revisions.keys(),
3383
'topological', True))
3387
self.target.fetch(self.source, revision_id=revision_id)
3390
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3391
"""See InterRepository.missing_revision_ids()."""
3392
# we want all revisions to satisfy revision_id in source.
3393
# but we don't want to stat every file here and there.
3394
# we want then, all revisions other needs to satisfy revision_id
3395
# checked, but not those that we have locally.
3396
# so the first thing is to get a subset of the revisions to
3397
# satisfy revision_id in source, and then eliminate those that
3398
# we do already have.
3399
# this is slow on high latency connection to self, but as this
3400
# disk format scales terribly for push anyway due to rewriting
3401
# inventory.weave, this is considered acceptable.
3403
if revision_id is not None:
3404
source_ids = self.source.get_ancestry(revision_id)
3405
if source_ids[0] is not None:
3406
raise AssertionError()
3409
source_ids = self.source._all_possible_ids()
3410
source_ids_set = set(source_ids)
3411
# source_ids is the worst possible case we may need to pull.
3412
# now we want to filter source_ids against what we actually
3413
# have in target, but don't try to check for existence where we know
3414
# we do not have a revision as that would be pointless.
3415
target_ids = set(self.target._all_possible_ids())
3416
possibly_present_revisions = target_ids.intersection(source_ids_set)
3417
actually_present_revisions = set(
3418
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3419
required_revisions = source_ids_set.difference(actually_present_revisions)
3420
if revision_id is not None:
3421
# we used get_ancestry to determine source_ids then we are assured all
3422
# revisions referenced are present as they are installed in topological order.
3423
# and the tip revision was validated by get_ancestry.
3424
result_set = required_revisions
3426
# if we just grabbed the possibly available ids, then
3427
# we only have an estimate of whats available and need to validate
3428
# that against the revision records.
3430
self.source._eliminate_revisions_not_present(required_revisions))
3431
return self.source.revision_ids_to_search_result(result_set)
3434
class InterKnitRepo(InterSameDataRepository):
3435
"""Optimised code paths between Knit based repositories."""
3438
def _get_repo_format_to_test(self):
3439
from bzrlib.repofmt import knitrepo
3440
return knitrepo.RepositoryFormatKnit1()
3443
def is_compatible(source, target):
3444
"""Be compatible with known Knit formats.
3446
We don't test for the stores being of specific types because that
3447
could lead to confusing results, and there is no need to be
3450
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3452
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3453
isinstance(target._format, RepositoryFormatKnit))
3454
except AttributeError:
3456
return are_knits and InterRepository._same_model(source, target)
3459
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3460
"""See InterRepository.missing_revision_ids()."""
3461
if revision_id is not None:
3462
source_ids = self.source.get_ancestry(revision_id)
3463
if source_ids[0] is not None:
3464
raise AssertionError()
3467
source_ids = self.source.all_revision_ids()
3468
source_ids_set = set(source_ids)
3469
# source_ids is the worst possible case we may need to pull.
3470
# now we want to filter source_ids against what we actually
3471
# have in target, but don't try to check for existence where we know
3472
# we do not have a revision as that would be pointless.
3473
target_ids = set(self.target.all_revision_ids())
3474
possibly_present_revisions = target_ids.intersection(source_ids_set)
3475
actually_present_revisions = set(
3476
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3477
required_revisions = source_ids_set.difference(actually_present_revisions)
3478
if revision_id is not None:
3479
# we used get_ancestry to determine source_ids then we are assured all
3480
# revisions referenced are present as they are installed in topological order.
3481
# and the tip revision was validated by get_ancestry.
3482
result_set = required_revisions
3484
# if we just grabbed the possibly available ids, then
3485
# we only have an estimate of whats available and need to validate
3486
# that against the revision records.
3488
self.source._eliminate_revisions_not_present(required_revisions))
3489
return self.source.revision_ids_to_search_result(result_set)
3492
class InterDifferingSerializer(InterRepository):
3495
def _get_repo_format_to_test(self):
3499
def is_compatible(source, target):
3500
"""Be compatible with Knit2 source and Knit3 target"""
3501
# This is redundant with format.check_conversion_target(), however that
3502
# raises an exception, and we just want to say "False" as in we won't
3503
# support converting between these formats.
3504
if source.supports_rich_root() and not target.supports_rich_root():
3506
if (source._format.supports_tree_reference
3507
and not target._format.supports_tree_reference):
3511
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3512
"""Get the best delta and base for this revision.
3514
:return: (basis_id, delta)
3516
possible_trees = [(parent_id, cache[parent_id])
3517
for parent_id in parent_ids
3518
if parent_id in cache]
3519
if len(possible_trees) == 0:
3520
# There either aren't any parents, or the parents aren't in the
3521
# cache, so just use the last converted tree
3522
possible_trees.append((basis_id, cache[basis_id]))
3524
for basis_id, basis_tree in possible_trees:
3525
delta = tree.inventory._make_delta(basis_tree.inventory)
3526
deltas.append((len(delta), basis_id, delta))
3528
return deltas[0][1:]
3530
def _get_parent_keys(self, root_key, parent_map):
3531
"""Get the parent keys for a given root id."""
3532
root_id, rev_id = root_key
3533
# Include direct parents of the revision, but only if they used
3534
# the same root_id and are heads.
3536
for parent_id in parent_map[rev_id]:
3537
if parent_id == _mod_revision.NULL_REVISION:
3539
if parent_id not in self._revision_id_to_root_id:
3540
# We probably didn't read this revision, go spend the
3541
# extra effort to actually check
3543
tree = self.source.revision_tree(parent_id)
3544
except errors.NoSuchRevision:
3545
# Ghost, fill out _revision_id_to_root_id in case we
3546
# encounter this again.
3547
# But set parent_root_id to None since we don't really know
3548
parent_root_id = None
3550
parent_root_id = tree.get_root_id()
3551
self._revision_id_to_root_id[parent_id] = None
3553
parent_root_id = self._revision_id_to_root_id[parent_id]
3554
if root_id == parent_root_id:
3555
# With stacking we _might_ want to refer to a non-local
3556
# revision, but this code path only applies when we have the
3557
# full content available, so ghosts really are ghosts, not just
3558
# the edge of local data.
3559
parent_keys.append((parent_id,))
3561
# root_id may be in the parent anyway.
3563
tree = self.source.revision_tree(parent_id)
3564
except errors.NoSuchRevision:
3565
# ghost, can't refer to it.
3569
parent_keys.append((tree.inventory[root_id].revision,))
3570
except errors.NoSuchId:
3573
g = graph.Graph(self.source.revisions)
3574
heads = g.heads(parent_keys)
3576
for key in parent_keys:
3577
if key in heads and key not in selected_keys:
3578
selected_keys.append(key)
3579
return tuple([(root_id,)+ key for key in selected_keys])
3581
def _new_root_data_stream(self, root_keys_to_create, parent_map):
3582
for root_key in root_keys_to_create:
3583
parent_keys = self._get_parent_keys(root_key, parent_map)
3584
yield versionedfile.FulltextContentFactory(root_key,
3585
parent_keys, None, '')
3587
def _fetch_batch(self, revision_ids, basis_id, cache):
3588
"""Fetch across a few revisions.
3590
:param revision_ids: The revisions to copy
3591
:param basis_id: The revision_id of a tree that must be in cache, used
3592
as a basis for delta when no other base is available
3593
:param cache: A cache of RevisionTrees that we can use.
3594
:return: The revision_id of the last converted tree. The RevisionTree
3595
for it will be in cache
3597
# Walk though all revisions; get inventory deltas, copy referenced
3598
# texts that delta references, insert the delta, revision and
3600
root_keys_to_create = set()
3603
pending_revisions = []
3604
parent_map = self.source.get_parent_map(revision_ids)
3605
for tree in self.source.revision_trees(revision_ids):
3606
current_revision_id = tree.get_revision_id()
3607
parent_ids = parent_map.get(current_revision_id, ())
3608
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3610
if self._converting_to_rich_root:
3611
self._revision_id_to_root_id[current_revision_id] = \
3613
# Find text entries that need to be copied
3614
for old_path, new_path, file_id, entry in delta:
3615
if new_path is not None:
3618
if not self.target.supports_rich_root():
3619
# The target doesn't support rich root, so we don't
3622
if self._converting_to_rich_root:
3623
# This can't be copied normally, we have to insert
3625
root_keys_to_create.add((file_id, entry.revision))
3627
text_keys.add((file_id, entry.revision))
3628
revision = self.source.get_revision(current_revision_id)
3629
pending_deltas.append((basis_id, delta,
3630
current_revision_id, revision.parent_ids))
3631
pending_revisions.append(revision)
3632
cache[current_revision_id] = tree
3633
basis_id = current_revision_id
3635
from_texts = self.source.texts
3636
to_texts = self.target.texts
3637
if root_keys_to_create:
3638
root_stream = self._new_root_data_stream(root_keys_to_create,
3640
to_texts.insert_record_stream(root_stream)
3641
to_texts.insert_record_stream(from_texts.get_record_stream(
3642
text_keys, self.target._format._fetch_order,
3643
not self.target._format._fetch_uses_deltas))
3644
# insert inventory deltas
3645
for delta in pending_deltas:
3646
self.target.add_inventory_by_delta(*delta)
3647
if self.target._fallback_repositories:
3648
# Make sure this stacked repository has all the parent inventories
3649
# for the new revisions that we are about to insert. We do this
3650
# before adding the revisions so that no revision is added until
3651
# all the inventories it may depend on are added.
3653
revision_ids = set()
3654
for revision in pending_revisions:
3655
revision_ids.add(revision.revision_id)
3656
parent_ids.update(revision.parent_ids)
3657
parent_ids.difference_update(revision_ids)
3658
parent_ids.discard(_mod_revision.NULL_REVISION)
3659
parent_map = self.source.get_parent_map(parent_ids)
3660
for parent_tree in self.source.revision_trees(parent_ids):
3661
basis_id, delta = self._get_delta_for_revision(tree, parent_ids, basis_id, cache)
3662
current_revision_id = parent_tree.get_revision_id()
3663
parents_parents = parent_map[current_revision_id]
3664
self.target.add_inventory_by_delta(
3665
basis_id, delta, current_revision_id, parents_parents)
3666
# insert signatures and revisions
3667
for revision in pending_revisions:
3669
signature = self.source.get_signature_text(
3670
revision.revision_id)
3671
self.target.add_signature_text(revision.revision_id,
3673
except errors.NoSuchRevision:
3675
self.target.add_revision(revision.revision_id, revision)
3678
def _fetch_all_revisions(self, revision_ids, pb):
3679
"""Fetch everything for the list of revisions.
3681
:param revision_ids: The list of revisions to fetch. Must be in
3683
:param pb: A ProgressBar
3686
basis_id, basis_tree = self._get_basis(revision_ids[0])
3688
cache = lru_cache.LRUCache(100)
3689
cache[basis_id] = basis_tree
3690
del basis_tree # We don't want to hang on to it here
3691
for offset in range(0, len(revision_ids), batch_size):
3692
self.target.start_write_group()
3694
pb.update('Transferring revisions', offset,
3696
batch = revision_ids[offset:offset+batch_size]
3697
basis_id = self._fetch_batch(batch, basis_id, cache)
3699
self.target.abort_write_group()
3702
self.target.commit_write_group()
3703
pb.update('Transferring revisions', len(revision_ids),
3707
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3709
"""See InterRepository.fetch()."""
3710
if fetch_spec is not None:
3711
raise AssertionError("Not implemented yet...")
3712
if (not self.source.supports_rich_root()
3713
and self.target.supports_rich_root()):
3714
self._converting_to_rich_root = True
3715
self._revision_id_to_root_id = {}
3717
self._converting_to_rich_root = False
3718
revision_ids = self.target.search_missing_revision_ids(self.source,
3719
revision_id, find_ghosts=find_ghosts).get_keys()
3720
if not revision_ids:
3722
revision_ids = tsort.topo_sort(
3723
self.source.get_graph().get_parent_map(revision_ids))
3724
if not revision_ids:
3726
# Walk though all revisions; get inventory deltas, copy referenced
3727
# texts that delta references, insert the delta, revision and
3729
first_rev = self.source.get_revision(revision_ids[0])
3731
my_pb = ui.ui_factory.nested_progress_bar()
3734
symbol_versioning.warn(
3735
symbol_versioning.deprecated_in((1, 14, 0))
3736
% "pb parameter to fetch()")
3739
self._fetch_all_revisions(revision_ids, pb)
3741
if my_pb is not None:
3743
return len(revision_ids), 0
3745
def _get_basis(self, first_revision_id):
3746
"""Get a revision and tree which exists in the target.
3748
This assumes that first_revision_id is selected for transmission
3749
because all other ancestors are already present. If we can't find an
3750
ancestor we fall back to NULL_REVISION since we know that is safe.
3752
:return: (basis_id, basis_tree)
3754
first_rev = self.source.get_revision(first_revision_id)
3756
basis_id = first_rev.parent_ids[0]
3757
# only valid as a basis if the target has it
3758
self.target.get_revision(basis_id)
3759
# Try to get a basis tree - if its a ghost it will hit the
3760
# NoSuchRevision case.
3761
basis_tree = self.source.revision_tree(basis_id)
3762
except (IndexError, errors.NoSuchRevision):
3763
basis_id = _mod_revision.NULL_REVISION
3764
basis_tree = self.source.revision_tree(basis_id)
3765
return basis_id, basis_tree
3768
InterRepository.register_optimiser(InterDifferingSerializer)
3769
InterRepository.register_optimiser(InterSameDataRepository)
3770
InterRepository.register_optimiser(InterWeaveRepo)
3771
InterRepository.register_optimiser(InterKnitRepo)
3774
class CopyConverter(object):
3775
"""A repository conversion tool which just performs a copy of the content.
3777
This is slow but quite reliable.
3780
def __init__(self, target_format):
3781
"""Create a CopyConverter.
3783
:param target_format: The format the resulting repository should be.
3785
self.target_format = target_format
3787
def convert(self, repo, pb):
3788
"""Perform the conversion of to_convert, giving feedback via pb.
3790
:param to_convert: The disk object to convert.
3791
:param pb: a progress bar to use for progress information.
3796
# this is only useful with metadir layouts - separated repo content.
3797
# trigger an assertion if not such
3798
repo._format.get_format_string()
3799
self.repo_dir = repo.bzrdir
3800
self.step('Moving repository to repository.backup')
3801
self.repo_dir.transport.move('repository', 'repository.backup')
3802
backup_transport = self.repo_dir.transport.clone('repository.backup')
3803
repo._format.check_conversion_target(self.target_format)
3804
self.source_repo = repo._format.open(self.repo_dir,
3806
_override_transport=backup_transport)
3807
self.step('Creating new repository')
3808
converted = self.target_format.initialize(self.repo_dir,
3809
self.source_repo.is_shared())
3810
converted.lock_write()
3812
self.step('Copying content into repository.')
3813
self.source_repo.copy_content_into(converted)
3816
self.step('Deleting old repository content.')
3817
self.repo_dir.transport.delete_tree('repository.backup')
3818
self.pb.note('repository converted')
3820
def step(self, message):
3821
"""Update the pb by a step."""
3823
self.pb.update(message, self.count, self.total)
3835
def _unescaper(match, _map=_unescape_map):
3836
code = match.group(1)
3840
if not code.startswith('#'):
3842
return unichr(int(code[1:])).encode('utf8')
3848
def _unescape_xml(data):
3849
"""Unescape predefined XML entities in a string of data."""
3851
if _unescape_re is None:
3852
_unescape_re = re.compile('\&([^;]*);')
3853
return _unescape_re.sub(_unescaper, data)
3856
class _VersionedFileChecker(object):
3858
def __init__(self, repository, text_key_references=None):
3859
self.repository = repository
3860
self.text_index = self.repository._generate_text_key_index(
3861
text_key_references=text_key_references)
3863
def calculate_file_version_parents(self, text_key):
3864
"""Calculate the correct parents for a file version according to
3867
parent_keys = self.text_index[text_key]
3868
if parent_keys == [_mod_revision.NULL_REVISION]:
3870
return tuple(parent_keys)
3872
def check_file_version_parents(self, texts, progress_bar=None):
3873
"""Check the parents stored in a versioned file are correct.
3875
It also detects file versions that are not referenced by their
3876
corresponding revision's inventory.
3878
:returns: A tuple of (wrong_parents, dangling_file_versions).
3879
wrong_parents is a dict mapping {revision_id: (stored_parents,
3880
correct_parents)} for each revision_id where the stored parents
3881
are not correct. dangling_file_versions is a set of (file_id,
3882
revision_id) tuples for versions that are present in this versioned
3883
file, but not used by the corresponding inventory.
3886
self.file_ids = set([file_id for file_id, _ in
3887
self.text_index.iterkeys()])
3888
# text keys is now grouped by file_id
3889
n_weaves = len(self.file_ids)
3890
files_in_revisions = {}
3891
revisions_of_files = {}
3892
n_versions = len(self.text_index)
3893
progress_bar.update('loading text store', 0, n_versions)
3894
parent_map = self.repository.texts.get_parent_map(self.text_index)
3895
# On unlistable transports this could well be empty/error...
3896
text_keys = self.repository.texts.keys()
3897
unused_keys = frozenset(text_keys) - set(self.text_index)
3898
for num, key in enumerate(self.text_index.iterkeys()):
3899
if progress_bar is not None:
3900
progress_bar.update('checking text graph', num, n_versions)
3901
correct_parents = self.calculate_file_version_parents(key)
3903
knit_parents = parent_map[key]
3904
except errors.RevisionNotPresent:
3907
if correct_parents != knit_parents:
3908
wrong_parents[key] = (knit_parents, correct_parents)
3909
return wrong_parents, unused_keys
3912
def _old_get_graph(repository, revision_id):
3913
"""DO NOT USE. That is all. I'm serious."""
3914
graph = repository.get_graph()
3915
revision_graph = dict(((key, value) for key, value in
3916
graph.iter_ancestry([revision_id]) if value is not None))
3917
return _strip_NULL_ghosts(revision_graph)
3920
def _strip_NULL_ghosts(revision_graph):
3921
"""Also don't use this. more compatibility code for unmigrated clients."""
3922
# Filter ghosts, and null:
3923
if _mod_revision.NULL_REVISION in revision_graph:
3924
del revision_graph[_mod_revision.NULL_REVISION]
3925
for key, parents in revision_graph.items():
3926
revision_graph[key] = tuple(parent for parent in parents if parent
3928
return revision_graph
3931
class StreamSink(object):
3932
"""An object that can insert a stream into a repository.
3934
This interface handles the complexity of reserialising inventories and
3935
revisions from different formats, and allows unidirectional insertion into
3936
stacked repositories without looking for the missing basis parents
3940
def __init__(self, target_repo):
3941
self.target_repo = target_repo
3943
def insert_stream(self, stream, src_format, resume_tokens):
3944
"""Insert a stream's content into the target repository.
3946
:param src_format: a bzr repository format.
3948
:return: a list of resume tokens and an iterable of keys additional
3949
items required before the insertion can be completed.
3951
self.target_repo.lock_write()
3954
self.target_repo.resume_write_group(resume_tokens)
3957
self.target_repo.start_write_group()
3960
# locked_insert_stream performs a commit|suspend.
3961
return self._locked_insert_stream(stream, src_format, is_resume)
3963
self.target_repo.abort_write_group(suppress_errors=True)
3966
self.target_repo.unlock()
3968
def _locked_insert_stream(self, stream, src_format, is_resume):
3969
to_serializer = self.target_repo._format._serializer
3970
src_serializer = src_format._serializer
3972
if to_serializer == src_serializer:
3973
# If serializers match and the target is a pack repository, set the
3974
# write cache size on the new pack. This avoids poor performance
3975
# on transports where append is unbuffered (such as
3976
# RemoteTransport). This is safe to do because nothing should read
3977
# back from the target repository while a stream with matching
3978
# serialization is being inserted.
3979
# The exception is that a delta record from the source that should
3980
# be a fulltext may need to be expanded by the target (see
3981
# test_fetch_revisions_with_deltas_into_pack); but we take care to
3982
# explicitly flush any buffered writes first in that rare case.
3984
new_pack = self.target_repo._pack_collection._new_pack
3985
except AttributeError:
3986
# Not a pack repository
3989
new_pack.set_write_cache_size(1024*1024)
3990
for substream_type, substream in stream:
3991
if substream_type == 'texts':
3992
self.target_repo.texts.insert_record_stream(substream)
3993
elif substream_type == 'inventories':
3994
if src_serializer == to_serializer:
3995
self.target_repo.inventories.insert_record_stream(
3998
self._extract_and_insert_inventories(
3999
substream, src_serializer)
4000
elif substream_type == 'chk_bytes':
4001
# XXX: This doesn't support conversions, as it assumes the
4002
# conversion was done in the fetch code.
4003
self.target_repo.chk_bytes.insert_record_stream(substream)
4004
elif substream_type == 'revisions':
4005
# This may fallback to extract-and-insert more often than
4006
# required if the serializers are different only in terms of
4008
if src_serializer == to_serializer:
4009
self.target_repo.revisions.insert_record_stream(
4012
self._extract_and_insert_revisions(substream,
4014
elif substream_type == 'signatures':
4015
self.target_repo.signatures.insert_record_stream(substream)
4017
raise AssertionError('kaboom! %s' % (substream_type,))
4018
# Done inserting data, and the missing_keys calculations will try to
4019
# read back from the inserted data, so flush the writes to the new pack
4020
# (if this is pack format).
4021
if new_pack is not None:
4022
new_pack._write_data('', flush=True)
4023
# Find all the new revisions (including ones from resume_tokens)
4024
missing_keys = self.target_repo.get_missing_parent_inventories(
4025
check_for_missing_texts=is_resume)
4027
for prefix, versioned_file in (
4028
('texts', self.target_repo.texts),
4029
('inventories', self.target_repo.inventories),
4030
('revisions', self.target_repo.revisions),
4031
('signatures', self.target_repo.signatures),
4032
('chk_bytes', self.target_repo.chk_bytes),
4034
if versioned_file is None:
4036
missing_keys.update((prefix,) + key for key in
4037
versioned_file.get_missing_compression_parent_keys())
4038
except NotImplementedError:
4039
# cannot even attempt suspending, and missing would have failed
4040
# during stream insertion.
4041
missing_keys = set()
4044
# suspend the write group and tell the caller what we is
4045
# missing. We know we can suspend or else we would not have
4046
# entered this code path. (All repositories that can handle
4047
# missing keys can handle suspending a write group).
4048
write_group_tokens = self.target_repo.suspend_write_group()
4049
return write_group_tokens, missing_keys
4050
self.target_repo.commit_write_group()
4053
def _extract_and_insert_inventories(self, substream, serializer):
4054
"""Generate a new inventory versionedfile in target, converting data.
4056
The inventory is retrieved from the source, (deserializing it), and
4057
stored in the target (reserializing it in a different format).
4059
for record in substream:
4060
bytes = record.get_bytes_as('fulltext')
4061
revision_id = record.key[0]
4062
inv = serializer.read_inventory_from_string(bytes, revision_id)
4063
parents = [key[0] for key in record.parents]
4064
self.target_repo.add_inventory(revision_id, inv, parents)
4066
def _extract_and_insert_revisions(self, substream, serializer):
4067
for record in substream:
4068
bytes = record.get_bytes_as('fulltext')
4069
revision_id = record.key[0]
4070
rev = serializer.read_revision_from_string(bytes)
4071
if rev.revision_id != revision_id:
4072
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4073
self.target_repo.add_revision(revision_id, rev)
4076
if self.target_repo._format._fetch_reconcile:
4077
self.target_repo.reconcile()
4080
class StreamSource(object):
4081
"""A source of a stream for fetching between repositories."""
4083
def __init__(self, from_repository, to_format):
4084
"""Create a StreamSource streaming from from_repository."""
4085
self.from_repository = from_repository
4086
self.to_format = to_format
4088
def delta_on_metadata(self):
4089
"""Return True if delta's are permitted on metadata streams.
4091
That is on revisions and signatures.
4093
src_serializer = self.from_repository._format._serializer
4094
target_serializer = self.to_format._serializer
4095
return (self.to_format._fetch_uses_deltas and
4096
src_serializer == target_serializer)
4098
def _fetch_revision_texts(self, revs):
4099
# fetch signatures first and then the revision texts
4100
# may need to be a InterRevisionStore call here.
4101
from_sf = self.from_repository.signatures
4102
# A missing signature is just skipped.
4103
keys = [(rev_id,) for rev_id in revs]
4104
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4106
self.to_format._fetch_order,
4107
not self.to_format._fetch_uses_deltas))
4108
# If a revision has a delta, this is actually expanded inside the
4109
# insert_record_stream code now, which is an alternate fix for
4111
from_rf = self.from_repository.revisions
4112
revisions = from_rf.get_record_stream(
4114
self.to_format._fetch_order,
4115
not self.delta_on_metadata())
4116
return [('signatures', signatures), ('revisions', revisions)]
4118
def _generate_root_texts(self, revs):
4119
"""This will be called by __fetch between fetching weave texts and
4120
fetching the inventory weave.
4122
Subclasses should override this if they need to generate root texts
4123
after fetching weave texts.
4125
if self._rich_root_upgrade():
4127
return bzrlib.fetch.Inter1and2Helper(
4128
self.from_repository).generate_root_texts(revs)
4132
def get_stream(self, search):
4134
revs = search.get_keys()
4135
graph = self.from_repository.get_graph()
4136
revs = list(graph.iter_topo_order(revs))
4137
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4139
for knit_kind, file_id, revisions in data_to_fetch:
4140
if knit_kind != phase:
4142
# Make a new progress bar for this phase
4143
if knit_kind == "file":
4144
# Accumulate file texts
4145
text_keys.extend([(file_id, revision) for revision in
4147
elif knit_kind == "inventory":
4148
# Now copy the file texts.
4149
from_texts = self.from_repository.texts
4150
yield ('texts', from_texts.get_record_stream(
4151
text_keys, self.to_format._fetch_order,
4152
not self.to_format._fetch_uses_deltas))
4153
# Cause an error if a text occurs after we have done the
4156
# Before we process the inventory we generate the root
4157
# texts (if necessary) so that the inventories references
4159
for _ in self._generate_root_texts(revs):
4161
# NB: This currently reopens the inventory weave in source;
4162
# using a single stream interface instead would avoid this.
4163
from_weave = self.from_repository.inventories
4164
# we fetch only the referenced inventories because we do not
4165
# know for unselected inventories whether all their required
4166
# texts are present in the other repository - it could be
4168
for info in self._get_inventory_stream(revs):
4170
elif knit_kind == "signatures":
4171
# Nothing to do here; this will be taken care of when
4172
# _fetch_revision_texts happens.
4174
elif knit_kind == "revisions":
4175
for record in self._fetch_revision_texts(revs):
4178
raise AssertionError("Unknown knit kind %r" % knit_kind)
4180
def get_stream_for_missing_keys(self, missing_keys):
4181
# missing keys can only occur when we are byte copying and not
4182
# translating (because translation means we don't send
4183
# unreconstructable deltas ever).
4185
keys['texts'] = set()
4186
keys['revisions'] = set()
4187
keys['inventories'] = set()
4188
keys['chk_bytes'] = set()
4189
keys['signatures'] = set()
4190
for key in missing_keys:
4191
keys[key[0]].add(key[1:])
4192
if len(keys['revisions']):
4193
# If we allowed copying revisions at this point, we could end up
4194
# copying a revision without copying its required texts: a
4195
# violation of the requirements for repository integrity.
4196
raise AssertionError(
4197
'cannot copy revisions to fill in missing deltas %s' % (
4198
keys['revisions'],))
4199
for substream_kind, keys in keys.iteritems():
4200
vf = getattr(self.from_repository, substream_kind)
4201
if vf is None and keys:
4202
raise AssertionError(
4203
"cannot fill in keys for a versioned file we don't"
4204
" have: %s needs %s" % (substream_kind, keys))
4206
# No need to stream something we don't have
4208
# Ask for full texts always so that we don't need more round trips
4209
# after this stream.
4210
# Some of the missing keys are genuinely ghosts, so filter absent
4211
# records. The Sink is responsible for doing another check to
4212
# ensure that ghosts don't introduce missing data for future
4214
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4215
self.to_format._fetch_order, True))
4216
yield substream_kind, stream
4218
def inventory_fetch_order(self):
4219
if self._rich_root_upgrade():
4220
return 'topological'
4222
return self.to_format._fetch_order
4224
def _rich_root_upgrade(self):
4225
return (not self.from_repository._format.rich_root_data and
4226
self.to_format.rich_root_data)
4228
def _get_inventory_stream(self, revision_ids):
4229
from_format = self.from_repository._format
4230
if (from_format.supports_chks and self.to_format.supports_chks
4231
and (from_format._serializer == self.to_format._serializer)):
4232
# Both sides support chks, and they use the same serializer, so it
4233
# is safe to transmit the chk pages and inventory pages across
4235
return self._get_chk_inventory_stream(revision_ids)
4236
elif (not from_format.supports_chks):
4237
# Source repository doesn't support chks. So we can transmit the
4238
# inventories 'as-is' and either they are just accepted on the
4239
# target, or the Sink will properly convert it.
4240
return self._get_simple_inventory_stream(revision_ids)
4242
# XXX: Hack to make not-chk->chk fetch: copy the inventories as
4243
# inventories. Note that this should probably be done somehow
4244
# as part of bzrlib.repository.StreamSink. Except JAM couldn't
4245
# figure out how a non-chk repository could possibly handle
4246
# deserializing an inventory stream from a chk repo, as it
4247
# doesn't have a way to understand individual pages.
4248
return self._get_convertable_inventory_stream(revision_ids)
4250
def _get_simple_inventory_stream(self, revision_ids):
4251
from_weave = self.from_repository.inventories
4252
yield ('inventories', from_weave.get_record_stream(
4253
[(rev_id,) for rev_id in revision_ids],
4254
self.inventory_fetch_order(),
4255
not self.delta_on_metadata()))
4257
def _get_chk_inventory_stream(self, revision_ids):
4258
"""Fetch the inventory texts, along with the associated chk maps."""
4259
# We want an inventory outside of the search set, so that we can filter
4260
# out uninteresting chk pages. For now we use
4261
# _find_revision_outside_set, but if we had a Search with cut_revs, we
4262
# could use that instead.
4263
start_rev_id = self.from_repository._find_revision_outside_set(
4265
start_rev_key = (start_rev_id,)
4266
inv_keys_to_fetch = [(rev_id,) for rev_id in revision_ids]
4267
if start_rev_id != _mod_revision.NULL_REVISION:
4268
inv_keys_to_fetch.append((start_rev_id,))
4269
# Any repo that supports chk_bytes must also support out-of-order
4270
# insertion. At least, that is how we expect it to work
4271
# We use get_record_stream instead of iter_inventories because we want
4272
# to be able to insert the stream as well. We could instead fetch
4273
# allowing deltas, and then iter_inventories, but we don't know whether
4274
# source or target is more 'local' anway.
4275
inv_stream = self.from_repository.inventories.get_record_stream(
4276
inv_keys_to_fetch, 'unordered',
4277
True) # We need them as full-texts so we can find their references
4278
uninteresting_chk_roots = set()
4279
interesting_chk_roots = set()
4280
def filter_inv_stream(inv_stream):
4281
for idx, record in enumerate(inv_stream):
4282
### child_pb.update('fetch inv', idx, len(inv_keys_to_fetch))
4283
bytes = record.get_bytes_as('fulltext')
4284
chk_inv = inventory.CHKInventory.deserialise(
4285
self.from_repository.chk_bytes, bytes, record.key)
4286
if record.key == start_rev_key:
4287
uninteresting_chk_roots.add(chk_inv.id_to_entry.key())
4288
p_id_map = chk_inv.parent_id_basename_to_file_id
4289
if p_id_map is not None:
4290
uninteresting_chk_roots.add(p_id_map.key())
4293
interesting_chk_roots.add(chk_inv.id_to_entry.key())
4294
p_id_map = chk_inv.parent_id_basename_to_file_id
4295
if p_id_map is not None:
4296
interesting_chk_roots.add(p_id_map.key())
4297
### pb.update('fetch inventory', 0, 2)
4298
yield ('inventories', filter_inv_stream(inv_stream))
4299
# Now that we have worked out all of the interesting root nodes, grab
4300
# all of the interesting pages and insert them
4301
### pb.update('fetch inventory', 1, 2)
4302
interesting = chk_map.iter_interesting_nodes(
4303
self.from_repository.chk_bytes, interesting_chk_roots,
4304
uninteresting_chk_roots)
4305
def to_stream_adapter():
4306
"""Adapt the iter_interesting_nodes result to a single stream.
4308
iter_interesting_nodes returns records as it processes them, along
4309
with keys. However, we only want to return the records themselves.
4311
for record, items in interesting:
4312
if record is not None:
4314
# XXX: We could instead call get_record_stream(records.keys())
4315
# ATM, this will always insert the records as fulltexts, and
4316
# requires that you can hang on to records once you have gone
4317
# on to the next one. Further, it causes the target to
4318
# recompress the data. Testing shows it to be faster than
4319
# requesting the records again, though.
4320
yield ('chk_bytes', to_stream_adapter())
4321
### pb.update('fetch inventory', 2, 2)
4323
def _get_convertable_inventory_stream(self, revision_ids):
4324
# XXX: One of source or target is using chks, and they don't have
4325
# compatible serializations. The StreamSink code expects to be
4326
# able to convert on the target, so we need to put
4327
# bytes-on-the-wire that can be converted
4328
yield ('inventories', self._stream_invs_as_fulltexts(revision_ids))
4330
def _stream_invs_as_fulltexts(self, revision_ids):
4331
from_repo = self.from_repository
4332
from_serializer = from_repo._format._serializer
4333
revision_keys = [(rev_id,) for rev_id in revision_ids]
4334
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4335
for inv in self.from_repository.iter_inventories(revision_ids):
4336
# XXX: This is a bit hackish, but it works. Basically,
4337
# CHKSerializer 'accidentally' supports
4338
# read/write_inventory_to_string, even though that is never
4339
# the format that is stored on disk. It *does* give us a
4340
# single string representation for an inventory, so live with
4342
# This would be far better if we had a 'serialized inventory
4343
# delta' form. Then we could use 'inventory._make_delta', and
4344
# transmit that. This would both be faster to generate, and
4345
# result in fewer bytes-on-the-wire.
4346
as_bytes = from_serializer.write_inventory_to_string(inv)
4347
key = (inv.revision_id,)
4348
parent_keys = parent_map.get(key, ())
4349
yield versionedfile.FulltextContentFactory(
4350
key, parent_keys, None, as_bytes)
4353
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4354
stop_revision=None):
4355
"""Extend the partial history to include a given index
4357
If a stop_index is supplied, stop when that index has been reached.
4358
If a stop_revision is supplied, stop when that revision is
4359
encountered. Otherwise, stop when the beginning of history is
4362
:param stop_index: The index which should be present. When it is
4363
present, history extension will stop.
4364
:param stop_revision: The revision id which should be present. When
4365
it is encountered, history extension will stop.
4367
start_revision = partial_history_cache[-1]
4368
iterator = repo.iter_reverse_revision_history(start_revision)
4370
#skip the last revision in the list
4373
if (stop_index is not None and
4374
len(partial_history_cache) > stop_index):
4376
if partial_history_cache[-1] == stop_revision:
4378
revision_id = iterator.next()
4379
partial_history_cache.append(revision_id)
4380
except StopIteration:
added
removed
bzrlib/repository.py
