1
# Copyright (C) 2005-2011 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(), """
37
revision as _mod_revision,
42
from bzrlib.bundle import serializer
43
from bzrlib.recordcounter import RecordCounter
44
from bzrlib.revisiontree import InventoryRevisionTree
45
from bzrlib.store.versioned import VersionedFileStore
46
from bzrlib.testament import Testament
55
from bzrlib.decorators import needs_read_lock, needs_write_lock, only_raises
56
from bzrlib.inter import InterObject
57
from bzrlib.inventory import (
63
from bzrlib.lock import _RelockDebugMixin, LogicalLockResult
64
from bzrlib.trace import (
65
log_exception_quietly, note, mutter, mutter_callsite, warning)
68
# Old formats display a warning, but only once
69
_deprecation_warning_done = False
72
class IsInWriteGroupError(errors.InternalBzrError):
74
_fmt = "May not refresh_data of repo %(repo)s while in a write group."
76
def __init__(self, repo):
77
errors.InternalBzrError.__init__(self, repo=repo)
80
class CommitBuilder(object):
81
"""Provides an interface to build up a commit.
83
This allows describing a tree to be committed without needing to
84
know the internals of the format of the repository.
87
# all clients should supply tree roots.
88
record_root_entry = True
89
# the default CommitBuilder does not manage trees whose root is versioned.
90
_versioned_root = False
91
# this commit builder supports the record_entry_contents interface
92
supports_record_entry_contents = True
94
def __init__(self, repository, parents, config, timestamp=None,
95
timezone=None, committer=None, revprops=None,
96
revision_id=None, lossy=False):
97
"""Initiate a CommitBuilder.
99
:param repository: Repository to commit to.
100
:param parents: Revision ids of the parents of the new revision.
101
:param timestamp: Optional timestamp recorded for commit.
102
:param timezone: Optional timezone for timestamp.
103
:param committer: Optional committer to set for commit.
104
:param revprops: Optional dictionary of revision properties.
105
:param revision_id: Optional revision id.
106
:param lossy: Whether to discard data that can not be natively
107
represented, when pushing to a foreign VCS
109
self._config = config
112
if committer is None:
113
self._committer = self._config.username()
114
elif not isinstance(committer, unicode):
115
self._committer = committer.decode() # throw if non-ascii
117
self._committer = committer
119
self.new_inventory = Inventory(None)
120
self._new_revision_id = revision_id
121
self.parents = parents
122
self.repository = repository
125
if revprops is not None:
126
self._validate_revprops(revprops)
127
self._revprops.update(revprops)
129
if timestamp is None:
130
timestamp = time.time()
131
# Restrict resolution to 1ms
132
self._timestamp = round(timestamp, 3)
135
self._timezone = osutils.local_time_offset()
137
self._timezone = int(timezone)
139
self._generate_revision_if_needed()
140
self.__heads = graph.HeadsCache(repository.get_graph()).heads
141
self._basis_delta = []
142
# API compatibility, older code that used CommitBuilder did not call
143
# .record_delete(), which means the delta that is computed would not be
144
# valid. Callers that will call record_delete() should call
145
# .will_record_deletes() to indicate that.
146
self._recording_deletes = False
147
# memo'd check for no-op commits.
148
self._any_changes = False
150
def any_changes(self):
151
"""Return True if any entries were changed.
153
This includes merge-only changes. It is the core for the --unchanged
156
:return: True if any changes have occured.
158
return self._any_changes
160
def _validate_unicode_text(self, text, context):
161
"""Verify things like commit messages don't have bogus characters."""
163
raise ValueError('Invalid value for %s: %r' % (context, text))
165
def _validate_revprops(self, revprops):
166
for key, value in revprops.iteritems():
167
# We know that the XML serializers do not round trip '\r'
168
# correctly, so refuse to accept them
169
if not isinstance(value, basestring):
170
raise ValueError('revision property (%s) is not a valid'
171
' (unicode) string: %r' % (key, value))
172
self._validate_unicode_text(value,
173
'revision property (%s)' % (key,))
175
def _ensure_fallback_inventories(self):
176
"""Ensure that appropriate inventories are available.
178
This only applies to repositories that are stacked, and is about
179
enusring the stacking invariants. Namely, that for any revision that is
180
present, we either have all of the file content, or we have the parent
181
inventory and the delta file content.
183
if not self.repository._fallback_repositories:
185
if not self.repository._format.supports_chks:
186
raise errors.BzrError("Cannot commit directly to a stacked branch"
187
" in pre-2a formats. See "
188
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
189
# This is a stacked repo, we need to make sure we have the parent
190
# inventories for the parents.
191
parent_keys = [(p,) for p in self.parents]
192
parent_map = self.repository.inventories._index.get_parent_map(parent_keys)
193
missing_parent_keys = set([pk for pk in parent_keys
194
if pk not in parent_map])
195
fallback_repos = list(reversed(self.repository._fallback_repositories))
196
missing_keys = [('inventories', pk[0])
197
for pk in missing_parent_keys]
199
while missing_keys and fallback_repos:
200
fallback_repo = fallback_repos.pop()
201
source = fallback_repo._get_source(self.repository._format)
202
sink = self.repository._get_sink()
203
stream = source.get_stream_for_missing_keys(missing_keys)
204
missing_keys = sink.insert_stream_without_locking(stream,
205
self.repository._format)
207
raise errors.BzrError('Unable to fill in parent inventories for a'
210
def commit(self, message):
211
"""Make the actual commit.
213
:return: The revision id of the recorded revision.
215
self._validate_unicode_text(message, 'commit message')
216
rev = _mod_revision.Revision(
217
timestamp=self._timestamp,
218
timezone=self._timezone,
219
committer=self._committer,
221
inventory_sha1=self.inv_sha1,
222
revision_id=self._new_revision_id,
223
properties=self._revprops)
224
rev.parent_ids = self.parents
225
self.repository.add_revision(self._new_revision_id, rev,
226
self.new_inventory, self._config)
227
self._ensure_fallback_inventories()
228
self.repository.commit_write_group()
229
return self._new_revision_id
232
"""Abort the commit that is being built.
234
self.repository.abort_write_group()
236
def revision_tree(self):
237
"""Return the tree that was just committed.
239
After calling commit() this can be called to get a
240
InventoryRevisionTree representing the newly committed tree. This is
241
preferred to calling Repository.revision_tree() because that may
242
require deserializing the inventory, while we already have a copy in
245
if self.new_inventory is None:
246
self.new_inventory = self.repository.get_inventory(
247
self._new_revision_id)
248
return InventoryRevisionTree(self.repository, self.new_inventory,
249
self._new_revision_id)
251
def finish_inventory(self):
252
"""Tell the builder that the inventory is finished.
254
:return: The inventory id in the repository, which can be used with
255
repository.get_inventory.
257
if self.new_inventory is None:
258
# an inventory delta was accumulated without creating a new
260
basis_id = self.basis_delta_revision
261
# We ignore the 'inventory' returned by add_inventory_by_delta
262
# because self.new_inventory is used to hint to the rest of the
263
# system what code path was taken
264
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
265
basis_id, self._basis_delta, self._new_revision_id,
268
if self.new_inventory.root is None:
269
raise AssertionError('Root entry should be supplied to'
270
' record_entry_contents, as of bzr 0.10.')
271
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
272
self.new_inventory.revision_id = self._new_revision_id
273
self.inv_sha1 = self.repository.add_inventory(
274
self._new_revision_id,
278
return self._new_revision_id
280
def _gen_revision_id(self):
281
"""Return new revision-id."""
282
return generate_ids.gen_revision_id(self._committer, self._timestamp)
284
def _generate_revision_if_needed(self):
285
"""Create a revision id if None was supplied.
287
If the repository can not support user-specified revision ids
288
they should override this function and raise CannotSetRevisionId
289
if _new_revision_id is not None.
291
:raises: CannotSetRevisionId
293
if self._new_revision_id is None:
294
self._new_revision_id = self._gen_revision_id()
295
self.random_revid = True
297
self.random_revid = False
299
def _heads(self, file_id, revision_ids):
300
"""Calculate the graph heads for revision_ids in the graph of file_id.
302
This can use either a per-file graph or a global revision graph as we
303
have an identity relationship between the two graphs.
305
return self.__heads(revision_ids)
307
def _check_root(self, ie, parent_invs, tree):
308
"""Helper for record_entry_contents.
310
:param ie: An entry being added.
311
:param parent_invs: The inventories of the parent revisions of the
313
:param tree: The tree that is being committed.
315
# In this revision format, root entries have no knit or weave When
316
# serializing out to disk and back in root.revision is always
318
ie.revision = self._new_revision_id
320
def _require_root_change(self, tree):
321
"""Enforce an appropriate root object change.
323
This is called once when record_iter_changes is called, if and only if
324
the root was not in the delta calculated by record_iter_changes.
326
:param tree: The tree which is being committed.
328
if len(self.parents) == 0:
329
raise errors.RootMissing()
330
entry = entry_factory['directory'](tree.path2id(''), '',
332
entry.revision = self._new_revision_id
333
self._basis_delta.append(('', '', entry.file_id, entry))
335
def _get_delta(self, ie, basis_inv, path):
336
"""Get a delta against the basis inventory for ie."""
337
if ie.file_id not in basis_inv:
339
result = (None, path, ie.file_id, ie)
340
self._basis_delta.append(result)
342
elif ie != basis_inv[ie.file_id]:
344
# TODO: avoid tis id2path call.
345
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
346
self._basis_delta.append(result)
352
def get_basis_delta(self):
353
"""Return the complete inventory delta versus the basis inventory.
355
This has been built up with the calls to record_delete and
356
record_entry_contents. The client must have already called
357
will_record_deletes() to indicate that they will be generating a
360
:return: An inventory delta, suitable for use with apply_delta, or
361
Repository.add_inventory_by_delta, etc.
363
if not self._recording_deletes:
364
raise AssertionError("recording deletes not activated.")
365
return self._basis_delta
367
def record_delete(self, path, file_id):
368
"""Record that a delete occured against a basis tree.
370
This is an optional API - when used it adds items to the basis_delta
371
being accumulated by the commit builder. It cannot be called unless the
372
method will_record_deletes() has been called to inform the builder that
373
a delta is being supplied.
375
:param path: The path of the thing deleted.
376
:param file_id: The file id that was deleted.
378
if not self._recording_deletes:
379
raise AssertionError("recording deletes not activated.")
380
delta = (path, None, file_id, None)
381
self._basis_delta.append(delta)
382
self._any_changes = True
385
def will_record_deletes(self):
386
"""Tell the commit builder that deletes are being notified.
388
This enables the accumulation of an inventory delta; for the resulting
389
commit to be valid, deletes against the basis MUST be recorded via
390
builder.record_delete().
392
self._recording_deletes = True
394
basis_id = self.parents[0]
396
basis_id = _mod_revision.NULL_REVISION
397
self.basis_delta_revision = basis_id
399
def record_entry_contents(self, ie, parent_invs, path, tree,
401
"""Record the content of ie from tree into the commit if needed.
403
Side effect: sets ie.revision when unchanged
405
:param ie: An inventory entry present in the commit.
406
:param parent_invs: The inventories of the parent revisions of the
408
:param path: The path the entry is at in the tree.
409
:param tree: The tree which contains this entry and should be used to
411
:param content_summary: Summary data from the tree about the paths
412
content - stat, length, exec, sha/link target. This is only
413
accessed when the entry has a revision of None - that is when it is
414
a candidate to commit.
415
:return: A tuple (change_delta, version_recorded, fs_hash).
416
change_delta is an inventory_delta change for this entry against
417
the basis tree of the commit, or None if no change occured against
419
version_recorded is True if a new version of the entry has been
420
recorded. For instance, committing a merge where a file was only
421
changed on the other side will return (delta, False).
422
fs_hash is either None, or the hash details for the path (currently
423
a tuple of the contents sha1 and the statvalue returned by
424
tree.get_file_with_stat()).
426
if self.new_inventory.root is None:
427
if ie.parent_id is not None:
428
raise errors.RootMissing()
429
self._check_root(ie, parent_invs, tree)
430
if ie.revision is None:
431
kind = content_summary[0]
433
# ie is carried over from a prior commit
435
# XXX: repository specific check for nested tree support goes here - if
436
# the repo doesn't want nested trees we skip it ?
437
if (kind == 'tree-reference' and
438
not self.repository._format.supports_tree_reference):
439
# mismatch between commit builder logic and repository:
440
# this needs the entry creation pushed down into the builder.
441
raise NotImplementedError('Missing repository subtree support.')
442
self.new_inventory.add(ie)
444
# TODO: slow, take it out of the inner loop.
446
basis_inv = parent_invs[0]
448
basis_inv = Inventory(root_id=None)
450
# ie.revision is always None if the InventoryEntry is considered
451
# for committing. We may record the previous parents revision if the
452
# content is actually unchanged against a sole head.
453
if ie.revision is not None:
454
if not self._versioned_root and path == '':
455
# repositories that do not version the root set the root's
456
# revision to the new commit even when no change occurs (more
457
# specifically, they do not record a revision on the root; and
458
# the rev id is assigned to the root during deserialisation -
459
# this masks when a change may have occurred against the basis.
460
# To match this we always issue a delta, because the revision
461
# of the root will always be changing.
462
if ie.file_id in basis_inv:
463
delta = (basis_inv.id2path(ie.file_id), path,
467
delta = (None, path, ie.file_id, ie)
468
self._basis_delta.append(delta)
469
return delta, False, None
471
# we don't need to commit this, because the caller already
472
# determined that an existing revision of this file is
473
# appropriate. If it's not being considered for committing then
474
# it and all its parents to the root must be unaltered so
475
# no-change against the basis.
476
if ie.revision == self._new_revision_id:
477
raise AssertionError("Impossible situation, a skipped "
478
"inventory entry (%r) claims to be modified in this "
479
"commit (%r).", (ie, self._new_revision_id))
480
return None, False, None
481
# XXX: Friction: parent_candidates should return a list not a dict
482
# so that we don't have to walk the inventories again.
483
parent_candiate_entries = ie.parent_candidates(parent_invs)
484
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
486
for inv in parent_invs:
487
if ie.file_id in inv:
488
old_rev = inv[ie.file_id].revision
489
if old_rev in head_set:
490
heads.append(inv[ie.file_id].revision)
491
head_set.remove(inv[ie.file_id].revision)
494
# now we check to see if we need to write a new record to the
496
# We write a new entry unless there is one head to the ancestors, and
497
# the kind-derived content is unchanged.
499
# Cheapest check first: no ancestors, or more the one head in the
500
# ancestors, we write a new node.
504
# There is a single head, look it up for comparison
505
parent_entry = parent_candiate_entries[heads[0]]
506
# if the non-content specific data has changed, we'll be writing a
508
if (parent_entry.parent_id != ie.parent_id or
509
parent_entry.name != ie.name):
511
# now we need to do content specific checks:
513
# if the kind changed the content obviously has
514
if kind != parent_entry.kind:
516
# Stat cache fingerprint feedback for the caller - None as we usually
517
# don't generate one.
520
if content_summary[2] is None:
521
raise ValueError("Files must not have executable = None")
523
# We can't trust a check of the file length because of content
525
if (# if the exec bit has changed we have to store:
526
parent_entry.executable != content_summary[2]):
528
elif parent_entry.text_sha1 == content_summary[3]:
529
# all meta and content is unchanged (using a hash cache
530
# hit to check the sha)
531
ie.revision = parent_entry.revision
532
ie.text_size = parent_entry.text_size
533
ie.text_sha1 = parent_entry.text_sha1
534
ie.executable = parent_entry.executable
535
return self._get_delta(ie, basis_inv, path), False, None
537
# Either there is only a hash change(no hash cache entry,
538
# or same size content change), or there is no change on
540
# Provide the parent's hash to the store layer, so that the
541
# content is unchanged we will not store a new node.
542
nostore_sha = parent_entry.text_sha1
544
# We want to record a new node regardless of the presence or
545
# absence of a content change in the file.
547
ie.executable = content_summary[2]
548
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
550
text = file_obj.read()
554
ie.text_sha1, ie.text_size = self._add_text_to_weave(
555
ie.file_id, text, heads, nostore_sha)
556
# Let the caller know we generated a stat fingerprint.
557
fingerprint = (ie.text_sha1, stat_value)
558
except errors.ExistingContent:
559
# Turns out that the file content was unchanged, and we were
560
# only going to store a new node if it was changed. Carry over
562
ie.revision = parent_entry.revision
563
ie.text_size = parent_entry.text_size
564
ie.text_sha1 = parent_entry.text_sha1
565
ie.executable = parent_entry.executable
566
return self._get_delta(ie, basis_inv, path), False, None
567
elif kind == 'directory':
569
# all data is meta here, nothing specific to directory, so
571
ie.revision = parent_entry.revision
572
return self._get_delta(ie, basis_inv, path), False, None
573
self._add_text_to_weave(ie.file_id, '', heads, None)
574
elif kind == 'symlink':
575
current_link_target = content_summary[3]
577
# symlink target is not generic metadata, check if it has
579
if current_link_target != parent_entry.symlink_target:
582
# unchanged, carry over.
583
ie.revision = parent_entry.revision
584
ie.symlink_target = parent_entry.symlink_target
585
return self._get_delta(ie, basis_inv, path), False, None
586
ie.symlink_target = current_link_target
587
self._add_text_to_weave(ie.file_id, '', heads, None)
588
elif kind == 'tree-reference':
590
if content_summary[3] != parent_entry.reference_revision:
593
# unchanged, carry over.
594
ie.reference_revision = parent_entry.reference_revision
595
ie.revision = parent_entry.revision
596
return self._get_delta(ie, basis_inv, path), False, None
597
ie.reference_revision = content_summary[3]
598
if ie.reference_revision is None:
599
raise AssertionError("invalid content_summary for nested tree: %r"
600
% (content_summary,))
601
self._add_text_to_weave(ie.file_id, '', heads, None)
603
raise NotImplementedError('unknown kind')
604
ie.revision = self._new_revision_id
605
self._any_changes = True
606
return self._get_delta(ie, basis_inv, path), True, fingerprint
608
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
609
_entry_factory=entry_factory):
610
"""Record a new tree via iter_changes.
612
:param tree: The tree to obtain text contents from for changed objects.
613
:param basis_revision_id: The revision id of the tree the iter_changes
614
has been generated against. Currently assumed to be the same
615
as self.parents[0] - if it is not, errors may occur.
616
:param iter_changes: An iter_changes iterator with the changes to apply
617
to basis_revision_id. The iterator must not include any items with
618
a current kind of None - missing items must be either filtered out
619
or errored-on beefore record_iter_changes sees the item.
620
:param _entry_factory: Private method to bind entry_factory locally for
622
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
625
# Create an inventory delta based on deltas between all the parents and
626
# deltas between all the parent inventories. We use inventory delta's
627
# between the inventory objects because iter_changes masks
628
# last-changed-field only changes.
630
# file_id -> change map, change is fileid, paths, changed, versioneds,
631
# parents, names, kinds, executables
633
# {file_id -> revision_id -> inventory entry, for entries in parent
634
# trees that are not parents[0]
638
revtrees = list(self.repository.revision_trees(self.parents))
639
except errors.NoSuchRevision:
640
# one or more ghosts, slow path.
642
for revision_id in self.parents:
644
revtrees.append(self.repository.revision_tree(revision_id))
645
except errors.NoSuchRevision:
647
basis_revision_id = _mod_revision.NULL_REVISION
649
revtrees.append(self.repository.revision_tree(
650
_mod_revision.NULL_REVISION))
651
# The basis inventory from a repository
653
basis_inv = revtrees[0].inventory
655
basis_inv = self.repository.revision_tree(
656
_mod_revision.NULL_REVISION).inventory
657
if len(self.parents) > 0:
658
if basis_revision_id != self.parents[0] and not ghost_basis:
660
"arbitrary basis parents not yet supported with merges")
661
for revtree in revtrees[1:]:
662
for change in revtree.inventory._make_delta(basis_inv):
663
if change[1] is None:
664
# Not present in this parent.
666
if change[2] not in merged_ids:
667
if change[0] is not None:
668
basis_entry = basis_inv[change[2]]
669
merged_ids[change[2]] = [
671
basis_entry.revision,
674
parent_entries[change[2]] = {
676
basis_entry.revision:basis_entry,
678
change[3].revision:change[3],
681
merged_ids[change[2]] = [change[3].revision]
682
parent_entries[change[2]] = {change[3].revision:change[3]}
684
merged_ids[change[2]].append(change[3].revision)
685
parent_entries[change[2]][change[3].revision] = change[3]
688
# Setup the changes from the tree:
689
# changes maps file_id -> (change, [parent revision_ids])
691
for change in iter_changes:
692
# This probably looks up in basis_inv way to much.
693
if change[1][0] is not None:
694
head_candidate = [basis_inv[change[0]].revision]
697
changes[change[0]] = change, merged_ids.get(change[0],
699
unchanged_merged = set(merged_ids) - set(changes)
700
# Extend the changes dict with synthetic changes to record merges of
702
for file_id in unchanged_merged:
703
# Record a merged version of these items that did not change vs the
704
# basis. This can be either identical parallel changes, or a revert
705
# of a specific file after a merge. The recorded content will be
706
# that of the current tree (which is the same as the basis), but
707
# the per-file graph will reflect a merge.
708
# NB:XXX: We are reconstructing path information we had, this
709
# should be preserved instead.
710
# inv delta change: (file_id, (path_in_source, path_in_target),
711
# changed_content, versioned, parent, name, kind,
714
basis_entry = basis_inv[file_id]
715
except errors.NoSuchId:
716
# a change from basis->some_parents but file_id isn't in basis
717
# so was new in the merge, which means it must have changed
718
# from basis -> current, and as it hasn't the add was reverted
719
# by the user. So we discard this change.
723
(basis_inv.id2path(file_id), tree.id2path(file_id)),
725
(basis_entry.parent_id, basis_entry.parent_id),
726
(basis_entry.name, basis_entry.name),
727
(basis_entry.kind, basis_entry.kind),
728
(basis_entry.executable, basis_entry.executable))
729
changes[file_id] = (change, merged_ids[file_id])
730
# changes contains tuples with the change and a set of inventory
731
# candidates for the file.
733
# old_path, new_path, file_id, new_inventory_entry
734
seen_root = False # Is the root in the basis delta?
735
inv_delta = self._basis_delta
736
modified_rev = self._new_revision_id
737
for change, head_candidates in changes.values():
738
if change[3][1]: # versioned in target.
739
# Several things may be happening here:
740
# We may have a fork in the per-file graph
741
# - record a change with the content from tree
742
# We may have a change against < all trees
743
# - carry over the tree that hasn't changed
744
# We may have a change against all trees
745
# - record the change with the content from tree
748
entry = _entry_factory[kind](file_id, change[5][1],
750
head_set = self._heads(change[0], set(head_candidates))
753
for head_candidate in head_candidates:
754
if head_candidate in head_set:
755
heads.append(head_candidate)
756
head_set.remove(head_candidate)
759
# Could be a carry-over situation:
760
parent_entry_revs = parent_entries.get(file_id, None)
761
if parent_entry_revs:
762
parent_entry = parent_entry_revs.get(heads[0], None)
765
if parent_entry is None:
766
# The parent iter_changes was called against is the one
767
# that is the per-file head, so any change is relevant
768
# iter_changes is valid.
769
carry_over_possible = False
771
# could be a carry over situation
772
# A change against the basis may just indicate a merge,
773
# we need to check the content against the source of the
774
# merge to determine if it was changed after the merge
776
if (parent_entry.kind != entry.kind or
777
parent_entry.parent_id != entry.parent_id or
778
parent_entry.name != entry.name):
779
# Metadata common to all entries has changed
780
# against per-file parent
781
carry_over_possible = False
783
carry_over_possible = True
784
# per-type checks for changes against the parent_entry
787
# Cannot be a carry-over situation
788
carry_over_possible = False
789
# Populate the entry in the delta
791
# XXX: There is still a small race here: If someone reverts the content of a file
792
# after iter_changes examines and decides it has changed,
793
# we will unconditionally record a new version even if some
794
# other process reverts it while commit is running (with
795
# the revert happening after iter_changes did its
798
entry.executable = True
800
entry.executable = False
801
if (carry_over_possible and
802
parent_entry.executable == entry.executable):
803
# Check the file length, content hash after reading
805
nostore_sha = parent_entry.text_sha1
808
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
810
text = file_obj.read()
814
entry.text_sha1, entry.text_size = self._add_text_to_weave(
815
file_id, text, heads, nostore_sha)
816
yield file_id, change[1][1], (entry.text_sha1, stat_value)
817
except errors.ExistingContent:
818
# No content change against a carry_over parent
819
# Perhaps this should also yield a fs hash update?
821
entry.text_size = parent_entry.text_size
822
entry.text_sha1 = parent_entry.text_sha1
823
elif kind == 'symlink':
825
entry.symlink_target = tree.get_symlink_target(file_id)
826
if (carry_over_possible and
827
parent_entry.symlink_target == entry.symlink_target):
830
self._add_text_to_weave(change[0], '', heads, None)
831
elif kind == 'directory':
832
if carry_over_possible:
835
# Nothing to set on the entry.
836
# XXX: split into the Root and nonRoot versions.
837
if change[1][1] != '' or self.repository.supports_rich_root():
838
self._add_text_to_weave(change[0], '', heads, None)
839
elif kind == 'tree-reference':
840
if not self.repository._format.supports_tree_reference:
841
# This isn't quite sane as an error, but we shouldn't
842
# ever see this code path in practice: tree's don't
843
# permit references when the repo doesn't support tree
845
raise errors.UnsupportedOperation(tree.add_reference,
847
reference_revision = tree.get_reference_revision(change[0])
848
entry.reference_revision = reference_revision
849
if (carry_over_possible and
850
parent_entry.reference_revision == reference_revision):
853
self._add_text_to_weave(change[0], '', heads, None)
855
raise AssertionError('unknown kind %r' % kind)
857
entry.revision = modified_rev
859
entry.revision = parent_entry.revision
862
new_path = change[1][1]
863
inv_delta.append((change[1][0], new_path, change[0], entry))
866
self.new_inventory = None
868
# This should perhaps be guarded by a check that the basis we
869
# commit against is the basis for the commit and if not do a delta
871
self._any_changes = True
873
# housekeeping root entry changes do not affect no-change commits.
874
self._require_root_change(tree)
875
self.basis_delta_revision = basis_revision_id
877
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
878
parent_keys = tuple([(file_id, parent) for parent in parents])
879
return self.repository.texts._add_text(
880
(file_id, self._new_revision_id), parent_keys, new_text,
881
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
884
class RootCommitBuilder(CommitBuilder):
885
"""This commitbuilder actually records the root id"""
887
# the root entry gets versioned properly by this builder.
888
_versioned_root = True
890
def _check_root(self, ie, parent_invs, tree):
891
"""Helper for record_entry_contents.
893
:param ie: An entry being added.
894
:param parent_invs: The inventories of the parent revisions of the
896
:param tree: The tree that is being committed.
899
def _require_root_change(self, tree):
900
"""Enforce an appropriate root object change.
902
This is called once when record_iter_changes is called, if and only if
903
the root was not in the delta calculated by record_iter_changes.
905
:param tree: The tree which is being committed.
907
# versioned roots do not change unless the tree found a change.
910
class RepositoryWriteLockResult(LogicalLockResult):
911
"""The result of write locking a repository.
913
:ivar repository_token: The token obtained from the underlying lock, or
915
:ivar unlock: A callable which will unlock the lock.
918
def __init__(self, unlock, repository_token):
919
LogicalLockResult.__init__(self, unlock)
920
self.repository_token = repository_token
923
return "RepositoryWriteLockResult(%s, %s)" % (self.repository_token,
927
######################################################################
931
class Repository(_RelockDebugMixin, controldir.ControlComponent):
932
"""Repository holding history for one or more branches.
934
The repository holds and retrieves historical information including
935
revisions and file history. It's normally accessed only by the Branch,
936
which views a particular line of development through that history.
938
The Repository builds on top of some byte storage facilies (the revisions,
939
signatures, inventories, texts and chk_bytes attributes) and a Transport,
940
which respectively provide byte storage and a means to access the (possibly
943
The byte storage facilities are addressed via tuples, which we refer to
944
as 'keys' throughout the code base. Revision_keys, inventory_keys and
945
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
946
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
947
byte string made up of a hash identifier and a hash value.
948
We use this interface because it allows low friction with the underlying
949
code that implements disk indices, network encoding and other parts of
952
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
953
the serialised revisions for the repository. This can be used to obtain
954
revision graph information or to access raw serialised revisions.
955
The result of trying to insert data into the repository via this store
956
is undefined: it should be considered read-only except for implementors
958
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
959
the serialised signatures for the repository. This can be used to
960
obtain access to raw serialised signatures. The result of trying to
961
insert data into the repository via this store is undefined: it should
962
be considered read-only except for implementors of repositories.
963
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
964
the serialised inventories for the repository. This can be used to
965
obtain unserialised inventories. The result of trying to insert data
966
into the repository via this store is undefined: it should be
967
considered read-only except for implementors of repositories.
968
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
969
texts of files and directories for the repository. This can be used to
970
obtain file texts or file graphs. Note that Repository.iter_file_bytes
971
is usually a better interface for accessing file texts.
972
The result of trying to insert data into the repository via this store
973
is undefined: it should be considered read-only except for implementors
975
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
976
any data the repository chooses to store or have indexed by its hash.
977
The result of trying to insert data into the repository via this store
978
is undefined: it should be considered read-only except for implementors
980
:ivar _transport: Transport for file access to repository, typically
981
pointing to .bzr/repository.
984
# What class to use for a CommitBuilder. Often it's simpler to change this
985
# in a Repository class subclass rather than to override
986
# get_commit_builder.
987
_commit_builder_class = CommitBuilder
989
def abort_write_group(self, suppress_errors=False):
990
"""Commit the contents accrued within the current write group.
992
:param suppress_errors: if true, abort_write_group will catch and log
993
unexpected errors that happen during the abort, rather than
994
allowing them to propagate. Defaults to False.
996
:seealso: start_write_group.
998
if self._write_group is not self.get_transaction():
999
# has an unlock or relock occured ?
1002
'(suppressed) mismatched lock context and write group. %r, %r',
1003
self._write_group, self.get_transaction())
1005
raise errors.BzrError(
1006
'mismatched lock context and write group. %r, %r' %
1007
(self._write_group, self.get_transaction()))
1009
self._abort_write_group()
1010
except Exception, exc:
1011
self._write_group = None
1012
if not suppress_errors:
1014
mutter('abort_write_group failed')
1015
log_exception_quietly()
1016
note('bzr: ERROR (ignored): %s', exc)
1017
self._write_group = None
1019
def _abort_write_group(self):
1020
"""Template method for per-repository write group cleanup.
1022
This is called during abort before the write group is considered to be
1023
finished and should cleanup any internal state accrued during the write
1024
group. There is no requirement that data handed to the repository be
1025
*not* made available - this is not a rollback - but neither should any
1026
attempt be made to ensure that data added is fully commited. Abort is
1027
invoked when an error has occured so futher disk or network operations
1028
may not be possible or may error and if possible should not be
1032
def add_fallback_repository(self, repository):
1033
"""Add a repository to use for looking up data not held locally.
1035
:param repository: A repository.
1037
if not self._format.supports_external_lookups:
1038
raise errors.UnstackableRepositoryFormat(self._format, self.base)
1039
if self.is_locked():
1040
# This repository will call fallback.unlock() when we transition to
1041
# the unlocked state, so we make sure to increment the lock count
1042
repository.lock_read()
1043
self._check_fallback_repository(repository)
1044
self._fallback_repositories.append(repository)
1045
self.texts.add_fallback_versioned_files(repository.texts)
1046
self.inventories.add_fallback_versioned_files(repository.inventories)
1047
self.revisions.add_fallback_versioned_files(repository.revisions)
1048
self.signatures.add_fallback_versioned_files(repository.signatures)
1049
if self.chk_bytes is not None:
1050
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
1052
def _check_fallback_repository(self, repository):
1053
"""Check that this repository can fallback to repository safely.
1055
Raise an error if not.
1057
:param repository: A repository to fallback to.
1059
return InterRepository._assert_same_model(self, repository)
1061
def add_inventory(self, revision_id, inv, parents):
1062
"""Add the inventory inv to the repository as revision_id.
1064
:param parents: The revision ids of the parents that revision_id
1065
is known to have and are in the repository already.
1067
:returns: The validator(which is a sha1 digest, though what is sha'd is
1068
repository format specific) of the serialized inventory.
1070
if not self.is_in_write_group():
1071
raise AssertionError("%r not in write group" % (self,))
1072
_mod_revision.check_not_reserved_id(revision_id)
1073
if not (inv.revision_id is None or inv.revision_id == revision_id):
1074
raise AssertionError(
1075
"Mismatch between inventory revision"
1076
" id and insertion revid (%r, %r)"
1077
% (inv.revision_id, revision_id))
1078
if inv.root is None:
1079
raise errors.RootMissing()
1080
return self._add_inventory_checked(revision_id, inv, parents)
1082
def _add_inventory_checked(self, revision_id, inv, parents):
1083
"""Add inv to the repository after checking the inputs.
1085
This function can be overridden to allow different inventory styles.
1087
:seealso: add_inventory, for the contract.
1089
inv_lines = self._serializer.write_inventory_to_lines(inv)
1090
return self._inventory_add_lines(revision_id, parents,
1091
inv_lines, check_content=False)
1093
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1094
parents, basis_inv=None, propagate_caches=False):
1095
"""Add a new inventory expressed as a delta against another revision.
1097
See the inventory developers documentation for the theory behind
1100
:param basis_revision_id: The inventory id the delta was created
1101
against. (This does not have to be a direct parent.)
1102
:param delta: The inventory delta (see Inventory.apply_delta for
1104
:param new_revision_id: The revision id that the inventory is being
1106
:param parents: The revision ids of the parents that revision_id is
1107
known to have and are in the repository already. These are supplied
1108
for repositories that depend on the inventory graph for revision
1109
graph access, as well as for those that pun ancestry with delta
1111
:param basis_inv: The basis inventory if it is already known,
1113
:param propagate_caches: If True, the caches for this inventory are
1114
copied to and updated for the result if possible.
1116
:returns: (validator, new_inv)
1117
The validator(which is a sha1 digest, though what is sha'd is
1118
repository format specific) of the serialized inventory, and the
1119
resulting inventory.
1121
if not self.is_in_write_group():
1122
raise AssertionError("%r not in write group" % (self,))
1123
_mod_revision.check_not_reserved_id(new_revision_id)
1124
basis_tree = self.revision_tree(basis_revision_id)
1125
basis_tree.lock_read()
1127
# Note that this mutates the inventory of basis_tree, which not all
1128
# inventory implementations may support: A better idiom would be to
1129
# return a new inventory, but as there is no revision tree cache in
1130
# repository this is safe for now - RBC 20081013
1131
if basis_inv is None:
1132
basis_inv = basis_tree.inventory
1133
basis_inv.apply_delta(delta)
1134
basis_inv.revision_id = new_revision_id
1135
return (self.add_inventory(new_revision_id, basis_inv, parents),
1140
def _inventory_add_lines(self, revision_id, parents, lines,
1141
check_content=True):
1142
"""Store lines in inv_vf and return the sha1 of the inventory."""
1143
parents = [(parent,) for parent in parents]
1144
result = self.inventories.add_lines((revision_id,), parents, lines,
1145
check_content=check_content)[0]
1146
self.inventories._access.flush()
1149
def add_revision(self, revision_id, rev, inv=None, config=None):
1150
"""Add rev to the revision store as revision_id.
1152
:param revision_id: the revision id to use.
1153
:param rev: The revision object.
1154
:param inv: The inventory for the revision. if None, it will be looked
1155
up in the inventory storer
1156
:param config: If None no digital signature will be created.
1157
If supplied its signature_needed method will be used
1158
to determine if a signature should be made.
1160
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1162
_mod_revision.check_not_reserved_id(revision_id)
1163
if config is not None and config.signature_needed():
1165
inv = self.get_inventory(revision_id)
1166
tree = InventoryRevisionTree(self, inv, revision_id)
1167
testament = Testament(rev, tree)
1168
plaintext = testament.as_short_text()
1169
self.store_revision_signature(
1170
gpg.GPGStrategy(config), plaintext, revision_id)
1171
# check inventory present
1172
if not self.inventories.get_parent_map([(revision_id,)]):
1174
raise errors.WeaveRevisionNotPresent(revision_id,
1177
# yes, this is not suitable for adding with ghosts.
1178
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1181
key = (revision_id,)
1182
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1183
self._add_revision(rev)
1185
def _add_revision(self, revision):
1186
text = self._serializer.write_revision_to_string(revision)
1187
key = (revision.revision_id,)
1188
parents = tuple((parent,) for parent in revision.parent_ids)
1189
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1191
def all_revision_ids(self):
1192
"""Returns a list of all the revision ids in the repository.
1194
This is conceptually deprecated because code should generally work on
1195
the graph reachable from a particular revision, and ignore any other
1196
revisions that might be present. There is no direct replacement
1199
if 'evil' in debug.debug_flags:
1200
mutter_callsite(2, "all_revision_ids is linear with history.")
1201
return self._all_revision_ids()
1203
def _all_revision_ids(self):
1204
"""Returns a list of all the revision ids in the repository.
1206
These are in as much topological order as the underlying store can
1209
raise NotImplementedError(self._all_revision_ids)
1211
def break_lock(self):
1212
"""Break a lock if one is present from another instance.
1214
Uses the ui factory to ask for confirmation if the lock may be from
1217
self.control_files.break_lock()
1220
def _eliminate_revisions_not_present(self, revision_ids):
1221
"""Check every revision id in revision_ids to see if we have it.
1223
Returns a set of the present revisions.
1226
graph = self.get_graph()
1227
parent_map = graph.get_parent_map(revision_ids)
1228
# The old API returned a list, should this actually be a set?
1229
return parent_map.keys()
1231
def _check_inventories(self, checker):
1232
"""Check the inventories found from the revision scan.
1234
This is responsible for verifying the sha1 of inventories and
1235
creating a pending_keys set that covers data referenced by inventories.
1237
bar = ui.ui_factory.nested_progress_bar()
1239
self._do_check_inventories(checker, bar)
1243
def _do_check_inventories(self, checker, bar):
1244
"""Helper for _check_inventories."""
1246
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1247
kinds = ['chk_bytes', 'texts']
1248
count = len(checker.pending_keys)
1249
bar.update("inventories", 0, 2)
1250
current_keys = checker.pending_keys
1251
checker.pending_keys = {}
1252
# Accumulate current checks.
1253
for key in current_keys:
1254
if key[0] != 'inventories' and key[0] not in kinds:
1255
checker._report_items.append('unknown key type %r' % (key,))
1256
keys[key[0]].add(key[1:])
1257
if keys['inventories']:
1258
# NB: output order *should* be roughly sorted - topo or
1259
# inverse topo depending on repository - either way decent
1260
# to just delta against. However, pre-CHK formats didn't
1261
# try to optimise inventory layout on disk. As such the
1262
# pre-CHK code path does not use inventory deltas.
1264
for record in self.inventories.check(keys=keys['inventories']):
1265
if record.storage_kind == 'absent':
1266
checker._report_items.append(
1267
'Missing inventory {%s}' % (record.key,))
1269
last_object = self._check_record('inventories', record,
1270
checker, last_object,
1271
current_keys[('inventories',) + record.key])
1272
del keys['inventories']
1275
bar.update("texts", 1)
1276
while (checker.pending_keys or keys['chk_bytes']
1278
# Something to check.
1279
current_keys = checker.pending_keys
1280
checker.pending_keys = {}
1281
# Accumulate current checks.
1282
for key in current_keys:
1283
if key[0] not in kinds:
1284
checker._report_items.append('unknown key type %r' % (key,))
1285
keys[key[0]].add(key[1:])
1286
# Check the outermost kind only - inventories || chk_bytes || texts
1290
for record in getattr(self, kind).check(keys=keys[kind]):
1291
if record.storage_kind == 'absent':
1292
checker._report_items.append(
1293
'Missing %s {%s}' % (kind, record.key,))
1295
last_object = self._check_record(kind, record,
1296
checker, last_object, current_keys[(kind,) + record.key])
1300
def _check_record(self, kind, record, checker, last_object, item_data):
1301
"""Check a single text from this repository."""
1302
if kind == 'inventories':
1303
rev_id = record.key[0]
1304
inv = self._deserialise_inventory(rev_id,
1305
record.get_bytes_as('fulltext'))
1306
if last_object is not None:
1307
delta = inv._make_delta(last_object)
1308
for old_path, path, file_id, ie in delta:
1311
ie.check(checker, rev_id, inv)
1313
for path, ie in inv.iter_entries():
1314
ie.check(checker, rev_id, inv)
1315
if self._format.fast_deltas:
1317
elif kind == 'chk_bytes':
1318
# No code written to check chk_bytes for this repo format.
1319
checker._report_items.append(
1320
'unsupported key type chk_bytes for %s' % (record.key,))
1321
elif kind == 'texts':
1322
self._check_text(record, checker, item_data)
1324
checker._report_items.append(
1325
'unknown key type %s for %s' % (kind, record.key))
1327
def _check_text(self, record, checker, item_data):
1328
"""Check a single text."""
1329
# Check it is extractable.
1330
# TODO: check length.
1331
if record.storage_kind == 'chunked':
1332
chunks = record.get_bytes_as(record.storage_kind)
1333
sha1 = osutils.sha_strings(chunks)
1334
length = sum(map(len, chunks))
1336
content = record.get_bytes_as('fulltext')
1337
sha1 = osutils.sha_string(content)
1338
length = len(content)
1339
if item_data and sha1 != item_data[1]:
1340
checker._report_items.append(
1341
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1342
(record.key, sha1, item_data[1], item_data[2]))
1345
def create(a_bzrdir):
1346
"""Construct the current default format repository in a_bzrdir."""
1347
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1349
def __init__(self, _format, a_bzrdir, control_files):
1350
"""instantiate a Repository.
1352
:param _format: The format of the repository on disk.
1353
:param a_bzrdir: The BzrDir of the repository.
1355
# In the future we will have a single api for all stores for
1356
# getting file texts, inventories and revisions, then
1357
# this construct will accept instances of those things.
1358
super(Repository, self).__init__()
1359
self._format = _format
1360
# the following are part of the public API for Repository:
1361
self.bzrdir = a_bzrdir
1362
self.control_files = control_files
1363
self._transport = control_files._transport
1364
self.base = self._transport.base
1366
self._reconcile_does_inventory_gc = True
1367
self._reconcile_fixes_text_parents = False
1368
self._reconcile_backsup_inventory = True
1369
self._write_group = None
1370
# Additional places to query for data.
1371
self._fallback_repositories = []
1372
# An InventoryEntry cache, used during deserialization
1373
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1374
# Is it safe to return inventory entries directly from the entry cache,
1375
# rather copying them?
1376
self._safe_to_return_from_cache = False
1379
def user_transport(self):
1380
return self.bzrdir.user_transport
1383
def control_transport(self):
1384
return self._transport
1387
if self._fallback_repositories:
1388
return '%s(%r, fallback_repositories=%r)' % (
1389
self.__class__.__name__,
1391
self._fallback_repositories)
1393
return '%s(%r)' % (self.__class__.__name__,
1396
def _has_same_fallbacks(self, other_repo):
1397
"""Returns true if the repositories have the same fallbacks."""
1398
my_fb = self._fallback_repositories
1399
other_fb = other_repo._fallback_repositories
1400
if len(my_fb) != len(other_fb):
1402
for f, g in zip(my_fb, other_fb):
1403
if not f.has_same_location(g):
1407
def has_same_location(self, other):
1408
"""Returns a boolean indicating if this repository is at the same
1409
location as another repository.
1411
This might return False even when two repository objects are accessing
1412
the same physical repository via different URLs.
1414
if self.__class__ is not other.__class__:
1416
return (self._transport.base == other._transport.base)
1418
def is_in_write_group(self):
1419
"""Return True if there is an open write group.
1421
:seealso: start_write_group.
1423
return self._write_group is not None
1425
def is_locked(self):
1426
return self.control_files.is_locked()
1428
def is_write_locked(self):
1429
"""Return True if this object is write locked."""
1430
return self.is_locked() and self.control_files._lock_mode == 'w'
1432
def lock_write(self, token=None):
1433
"""Lock this repository for writing.
1435
This causes caching within the repository obejct to start accumlating
1436
data during reads, and allows a 'write_group' to be obtained. Write
1437
groups must be used for actual data insertion.
1439
A token should be passed in if you know that you have locked the object
1440
some other way, and need to synchronise this object's state with that
1443
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1445
:param token: if this is already locked, then lock_write will fail
1446
unless the token matches the existing lock.
1447
:returns: a token if this instance supports tokens, otherwise None.
1448
:raises TokenLockingNotSupported: when a token is given but this
1449
instance doesn't support using token locks.
1450
:raises MismatchedToken: if the specified token doesn't match the token
1451
of the existing lock.
1452
:seealso: start_write_group.
1453
:return: A RepositoryWriteLockResult.
1455
locked = self.is_locked()
1456
token = self.control_files.lock_write(token=token)
1458
self._warn_if_deprecated()
1459
self._note_lock('w')
1460
for repo in self._fallback_repositories:
1461
# Writes don't affect fallback repos
1463
self._refresh_data()
1464
return RepositoryWriteLockResult(self.unlock, token)
1466
def lock_read(self):
1467
"""Lock the repository for read operations.
1469
:return: An object with an unlock method which will release the lock
1472
locked = self.is_locked()
1473
self.control_files.lock_read()
1475
self._warn_if_deprecated()
1476
self._note_lock('r')
1477
for repo in self._fallback_repositories:
1479
self._refresh_data()
1480
return LogicalLockResult(self.unlock)
1482
def get_physical_lock_status(self):
1483
return self.control_files.get_physical_lock_status()
1485
def leave_lock_in_place(self):
1486
"""Tell this repository not to release the physical lock when this
1489
If lock_write doesn't return a token, then this method is not supported.
1491
self.control_files.leave_in_place()
1493
def dont_leave_lock_in_place(self):
1494
"""Tell this repository to release the physical lock when this
1495
object is unlocked, even if it didn't originally acquire it.
1497
If lock_write doesn't return a token, then this method is not supported.
1499
self.control_files.dont_leave_in_place()
1502
def gather_stats(self, revid=None, committers=None):
1503
"""Gather statistics from a revision id.
1505
:param revid: The revision id to gather statistics from, if None, then
1506
no revision specific statistics are gathered.
1507
:param committers: Optional parameter controlling whether to grab
1508
a count of committers from the revision specific statistics.
1509
:return: A dictionary of statistics. Currently this contains:
1510
committers: The number of committers if requested.
1511
firstrev: A tuple with timestamp, timezone for the penultimate left
1512
most ancestor of revid, if revid is not the NULL_REVISION.
1513
latestrev: A tuple with timestamp, timezone for revid, if revid is
1514
not the NULL_REVISION.
1515
revisions: The total revision count in the repository.
1516
size: An estimate disk size of the repository in bytes.
1519
if revid and committers:
1520
result['committers'] = 0
1521
if revid and revid != _mod_revision.NULL_REVISION:
1523
all_committers = set()
1524
revisions = self.get_ancestry(revid)
1525
# pop the leading None
1527
first_revision = None
1529
# ignore the revisions in the middle - just grab first and last
1530
revisions = revisions[0], revisions[-1]
1531
for revision in self.get_revisions(revisions):
1532
if not first_revision:
1533
first_revision = revision
1535
all_committers.add(revision.committer)
1536
last_revision = revision
1538
result['committers'] = len(all_committers)
1539
result['firstrev'] = (first_revision.timestamp,
1540
first_revision.timezone)
1541
result['latestrev'] = (last_revision.timestamp,
1542
last_revision.timezone)
1544
# now gather global repository information
1545
# XXX: This is available for many repos regardless of listability.
1546
if self.user_transport.listable():
1547
# XXX: do we want to __define len__() ?
1548
# Maybe the versionedfiles object should provide a different
1549
# method to get the number of keys.
1550
result['revisions'] = len(self.revisions.keys())
1551
# result['size'] = t
1554
def find_branches(self, using=False):
1555
"""Find branches underneath this repository.
1557
This will include branches inside other branches.
1559
:param using: If True, list only branches using this repository.
1561
if using and not self.is_shared():
1562
return self.bzrdir.list_branches()
1563
class Evaluator(object):
1566
self.first_call = True
1568
def __call__(self, bzrdir):
1569
# On the first call, the parameter is always the bzrdir
1570
# containing the current repo.
1571
if not self.first_call:
1573
repository = bzrdir.open_repository()
1574
except errors.NoRepositoryPresent:
1577
return False, ([], repository)
1578
self.first_call = False
1579
value = (bzrdir.list_branches(), None)
1583
for branches, repository in bzrdir.BzrDir.find_bzrdirs(
1584
self.user_transport, evaluate=Evaluator()):
1585
if branches is not None:
1586
ret.extend(branches)
1587
if not using and repository is not None:
1588
ret.extend(repository.find_branches())
1592
def search_missing_revision_ids(self, other,
1593
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
1594
find_ghosts=True, revision_ids=None, if_present_ids=None):
1595
"""Return the revision ids that other has that this does not.
1597
These are returned in topological order.
1599
revision_id: only return revision ids included by revision_id.
1601
if symbol_versioning.deprecated_passed(revision_id):
1602
symbol_versioning.warn(
1603
'search_missing_revision_ids(revision_id=...) was '
1604
'deprecated in 2.4. Use revision_ids=[...] instead.',
1605
DeprecationWarning, stacklevel=3)
1606
if revision_ids is not None:
1607
raise AssertionError(
1608
'revision_ids is mutually exclusive with revision_id')
1609
if revision_id is not None:
1610
revision_ids = [revision_id]
1611
return InterRepository.get(other, self).search_missing_revision_ids(
1612
find_ghosts=find_ghosts, revision_ids=revision_ids,
1613
if_present_ids=if_present_ids)
1617
"""Open the repository rooted at base.
1619
For instance, if the repository is at URL/.bzr/repository,
1620
Repository.open(URL) -> a Repository instance.
1622
control = bzrdir.BzrDir.open(base)
1623
return control.open_repository()
1625
def copy_content_into(self, destination, revision_id=None):
1626
"""Make a complete copy of the content in self into destination.
1628
This is a destructive operation! Do not use it on existing
1631
return InterRepository.get(self, destination).copy_content(revision_id)
1633
def commit_write_group(self):
1634
"""Commit the contents accrued within the current write group.
1636
:seealso: start_write_group.
1638
:return: it may return an opaque hint that can be passed to 'pack'.
1640
if self._write_group is not self.get_transaction():
1641
# has an unlock or relock occured ?
1642
raise errors.BzrError('mismatched lock context %r and '
1644
(self.get_transaction(), self._write_group))
1645
result = self._commit_write_group()
1646
self._write_group = None
1649
def _commit_write_group(self):
1650
"""Template method for per-repository write group cleanup.
1652
This is called before the write group is considered to be
1653
finished and should ensure that all data handed to the repository
1654
for writing during the write group is safely committed (to the
1655
extent possible considering file system caching etc).
1658
def suspend_write_group(self):
1659
raise errors.UnsuspendableWriteGroup(self)
1661
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1662
"""Return the keys of missing inventory parents for revisions added in
1665
A revision is not complete if the inventory delta for that revision
1666
cannot be calculated. Therefore if the parent inventories of a
1667
revision are not present, the revision is incomplete, and e.g. cannot
1668
be streamed by a smart server. This method finds missing inventory
1669
parents for revisions added in this write group.
1671
if not self._format.supports_external_lookups:
1672
# This is only an issue for stacked repositories
1674
if not self.is_in_write_group():
1675
raise AssertionError('not in a write group')
1677
# XXX: We assume that every added revision already has its
1678
# corresponding inventory, so we only check for parent inventories that
1679
# might be missing, rather than all inventories.
1680
parents = set(self.revisions._index.get_missing_parents())
1681
parents.discard(_mod_revision.NULL_REVISION)
1682
unstacked_inventories = self.inventories._index
1683
present_inventories = unstacked_inventories.get_parent_map(
1684
key[-1:] for key in parents)
1685
parents.difference_update(present_inventories)
1686
if len(parents) == 0:
1687
# No missing parent inventories.
1689
if not check_for_missing_texts:
1690
return set(('inventories', rev_id) for (rev_id,) in parents)
1691
# Ok, now we have a list of missing inventories. But these only matter
1692
# if the inventories that reference them are missing some texts they
1693
# appear to introduce.
1694
# XXX: Texts referenced by all added inventories need to be present,
1695
# but at the moment we're only checking for texts referenced by
1696
# inventories at the graph's edge.
1697
key_deps = self.revisions._index._key_dependencies
1698
key_deps.satisfy_refs_for_keys(present_inventories)
1699
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1700
file_ids = self.fileids_altered_by_revision_ids(referrers)
1701
missing_texts = set()
1702
for file_id, version_ids in file_ids.iteritems():
1703
missing_texts.update(
1704
(file_id, version_id) for version_id in version_ids)
1705
present_texts = self.texts.get_parent_map(missing_texts)
1706
missing_texts.difference_update(present_texts)
1707
if not missing_texts:
1708
# No texts are missing, so all revisions and their deltas are
1711
# Alternatively the text versions could be returned as the missing
1712
# keys, but this is likely to be less data.
1713
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1716
def refresh_data(self):
1717
"""Re-read any data needed to synchronise with disk.
1719
This method is intended to be called after another repository instance
1720
(such as one used by a smart server) has inserted data into the
1721
repository. On all repositories this will work outside of write groups.
1722
Some repository formats (pack and newer for bzrlib native formats)
1723
support refresh_data inside write groups. If called inside a write
1724
group on a repository that does not support refreshing in a write group
1725
IsInWriteGroupError will be raised.
1727
self._refresh_data()
1729
def resume_write_group(self, tokens):
1730
if not self.is_write_locked():
1731
raise errors.NotWriteLocked(self)
1732
if self._write_group:
1733
raise errors.BzrError('already in a write group')
1734
self._resume_write_group(tokens)
1735
# so we can detect unlock/relock - the write group is now entered.
1736
self._write_group = self.get_transaction()
1738
def _resume_write_group(self, tokens):
1739
raise errors.UnsuspendableWriteGroup(self)
1741
def fetch(self, source, revision_id=None, find_ghosts=False,
1743
"""Fetch the content required to construct revision_id from source.
1745
If revision_id is None and fetch_spec is None, then all content is
1748
fetch() may not be used when the repository is in a write group -
1749
either finish the current write group before using fetch, or use
1750
fetch before starting the write group.
1752
:param find_ghosts: Find and copy revisions in the source that are
1753
ghosts in the target (and not reachable directly by walking out to
1754
the first-present revision in target from revision_id).
1755
:param revision_id: If specified, all the content needed for this
1756
revision ID will be copied to the target. Fetch will determine for
1757
itself which content needs to be copied.
1758
:param fetch_spec: If specified, a SearchResult or
1759
PendingAncestryResult that describes which revisions to copy. This
1760
allows copying multiple heads at once. Mutually exclusive with
1763
if fetch_spec is not None and revision_id is not None:
1764
raise AssertionError(
1765
"fetch_spec and revision_id are mutually exclusive.")
1766
if self.is_in_write_group():
1767
raise errors.InternalBzrError(
1768
"May not fetch while in a write group.")
1769
# fast path same-url fetch operations
1770
# TODO: lift out to somewhere common with RemoteRepository
1771
# <https://bugs.launchpad.net/bzr/+bug/401646>
1772
if (self.has_same_location(source)
1773
and fetch_spec is None
1774
and self._has_same_fallbacks(source)):
1775
# check that last_revision is in 'from' and then return a
1777
if (revision_id is not None and
1778
not _mod_revision.is_null(revision_id)):
1779
self.get_revision(revision_id)
1781
inter = InterRepository.get(source, self)
1782
return inter.fetch(revision_id=revision_id,
1783
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1785
def create_bundle(self, target, base, fileobj, format=None):
1786
return serializer.write_bundle(self, target, base, fileobj, format)
1788
def get_commit_builder(self, branch, parents, config, timestamp=None,
1789
timezone=None, committer=None, revprops=None,
1790
revision_id=None, lossy=False):
1791
"""Obtain a CommitBuilder for this repository.
1793
:param branch: Branch to commit to.
1794
:param parents: Revision ids of the parents of the new revision.
1795
:param config: Configuration to use.
1796
:param timestamp: Optional timestamp recorded for commit.
1797
:param timezone: Optional timezone for timestamp.
1798
:param committer: Optional committer to set for commit.
1799
:param revprops: Optional dictionary of revision properties.
1800
:param revision_id: Optional revision id.
1801
:param lossy: Whether to discard data that can not be natively
1802
represented, when pushing to a foreign VCS
1804
if self._fallback_repositories and not self._format.supports_chks:
1805
raise errors.BzrError("Cannot commit directly to a stacked branch"
1806
" in pre-2a formats. See "
1807
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1808
result = self._commit_builder_class(self, parents, config,
1809
timestamp, timezone, committer, revprops, revision_id,
1811
self.start_write_group()
1814
@only_raises(errors.LockNotHeld, errors.LockBroken)
1816
if (self.control_files._lock_count == 1 and
1817
self.control_files._lock_mode == 'w'):
1818
if self._write_group is not None:
1819
self.abort_write_group()
1820
self.control_files.unlock()
1821
raise errors.BzrError(
1822
'Must end write groups before releasing write locks.')
1823
self.control_files.unlock()
1824
if self.control_files._lock_count == 0:
1825
self._inventory_entry_cache.clear()
1826
for repo in self._fallback_repositories:
1830
def clone(self, a_bzrdir, revision_id=None):
1831
"""Clone this repository into a_bzrdir using the current format.
1833
Currently no check is made that the format of this repository and
1834
the bzrdir format are compatible. FIXME RBC 20060201.
1836
:return: The newly created destination repository.
1838
# TODO: deprecate after 0.16; cloning this with all its settings is
1839
# probably not very useful -- mbp 20070423
1840
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1841
self.copy_content_into(dest_repo, revision_id)
1844
def start_write_group(self):
1845
"""Start a write group in the repository.
1847
Write groups are used by repositories which do not have a 1:1 mapping
1848
between file ids and backend store to manage the insertion of data from
1849
both fetch and commit operations.
1851
A write lock is required around the start_write_group/commit_write_group
1852
for the support of lock-requiring repository formats.
1854
One can only insert data into a repository inside a write group.
1858
if not self.is_write_locked():
1859
raise errors.NotWriteLocked(self)
1860
if self._write_group:
1861
raise errors.BzrError('already in a write group')
1862
self._start_write_group()
1863
# so we can detect unlock/relock - the write group is now entered.
1864
self._write_group = self.get_transaction()
1866
def _start_write_group(self):
1867
"""Template method for per-repository write group startup.
1869
This is called before the write group is considered to be
1874
def sprout(self, to_bzrdir, revision_id=None):
1875
"""Create a descendent repository for new development.
1877
Unlike clone, this does not copy the settings of the repository.
1879
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1880
dest_repo.fetch(self, revision_id=revision_id)
1883
def _create_sprouting_repo(self, a_bzrdir, shared):
1884
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1885
# use target default format.
1886
dest_repo = a_bzrdir.create_repository()
1888
# Most control formats need the repository to be specifically
1889
# created, but on some old all-in-one formats it's not needed
1891
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1892
except errors.UninitializableFormat:
1893
dest_repo = a_bzrdir.open_repository()
1896
def _get_sink(self):
1897
"""Return a sink for streaming into this repository."""
1898
return StreamSink(self)
1900
def _get_source(self, to_format):
1901
"""Return a source for streaming from this repository."""
1902
return StreamSource(self, to_format)
1905
def has_revision(self, revision_id):
1906
"""True if this repository has a copy of the revision."""
1907
return revision_id in self.has_revisions((revision_id,))
1910
def has_revisions(self, revision_ids):
1911
"""Probe to find out the presence of multiple revisions.
1913
:param revision_ids: An iterable of revision_ids.
1914
:return: A set of the revision_ids that were present.
1916
parent_map = self.revisions.get_parent_map(
1917
[(rev_id,) for rev_id in revision_ids])
1919
if _mod_revision.NULL_REVISION in revision_ids:
1920
result.add(_mod_revision.NULL_REVISION)
1921
result.update([key[0] for key in parent_map])
1925
def get_revision(self, revision_id):
1926
"""Return the Revision object for a named revision."""
1927
return self.get_revisions([revision_id])[0]
1930
def get_revision_reconcile(self, revision_id):
1931
"""'reconcile' helper routine that allows access to a revision always.
1933
This variant of get_revision does not cross check the weave graph
1934
against the revision one as get_revision does: but it should only
1935
be used by reconcile, or reconcile-alike commands that are correcting
1936
or testing the revision graph.
1938
return self._get_revisions([revision_id])[0]
1941
def get_revisions(self, revision_ids):
1942
"""Get many revisions at once.
1944
Repositories that need to check data on every revision read should
1945
subclass this method.
1947
return self._get_revisions(revision_ids)
1950
def _get_revisions(self, revision_ids):
1951
"""Core work logic to get many revisions without sanity checks."""
1953
for revid, rev in self._iter_revisions(revision_ids):
1955
raise errors.NoSuchRevision(self, revid)
1957
return [revs[revid] for revid in revision_ids]
1959
def _iter_revisions(self, revision_ids):
1960
"""Iterate over revision objects.
1962
:param revision_ids: An iterable of revisions to examine. None may be
1963
passed to request all revisions known to the repository. Note that
1964
not all repositories can find unreferenced revisions; for those
1965
repositories only referenced ones will be returned.
1966
:return: An iterator of (revid, revision) tuples. Absent revisions (
1967
those asked for but not available) are returned as (revid, None).
1969
if revision_ids is None:
1970
revision_ids = self.all_revision_ids()
1972
for rev_id in revision_ids:
1973
if not rev_id or not isinstance(rev_id, basestring):
1974
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1975
keys = [(key,) for key in revision_ids]
1976
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1977
for record in stream:
1978
revid = record.key[0]
1979
if record.storage_kind == 'absent':
1982
text = record.get_bytes_as('fulltext')
1983
rev = self._serializer.read_revision_from_string(text)
1986
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1987
"""Produce a generator of revision deltas.
1989
Note that the input is a sequence of REVISIONS, not revision_ids.
1990
Trees will be held in memory until the generator exits.
1991
Each delta is relative to the revision's lefthand predecessor.
1993
:param specific_fileids: if not None, the result is filtered
1994
so that only those file-ids, their parents and their
1995
children are included.
1997
# Get the revision-ids of interest
1998
required_trees = set()
1999
for revision in revisions:
2000
required_trees.add(revision.revision_id)
2001
required_trees.update(revision.parent_ids[:1])
2003
# Get the matching filtered trees. Note that it's more
2004
# efficient to pass filtered trees to changes_from() rather
2005
# than doing the filtering afterwards. changes_from() could
2006
# arguably do the filtering itself but it's path-based, not
2007
# file-id based, so filtering before or afterwards is
2009
if specific_fileids is None:
2010
trees = dict((t.get_revision_id(), t) for
2011
t in self.revision_trees(required_trees))
2013
trees = dict((t.get_revision_id(), t) for
2014
t in self._filtered_revision_trees(required_trees,
2017
# Calculate the deltas
2018
for revision in revisions:
2019
if not revision.parent_ids:
2020
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
2022
old_tree = trees[revision.parent_ids[0]]
2023
yield trees[revision.revision_id].changes_from(old_tree)
2026
def get_revision_delta(self, revision_id, specific_fileids=None):
2027
"""Return the delta for one revision.
2029
The delta is relative to the left-hand predecessor of the
2032
:param specific_fileids: if not None, the result is filtered
2033
so that only those file-ids, their parents and their
2034
children are included.
2036
r = self.get_revision(revision_id)
2037
return list(self.get_deltas_for_revisions([r],
2038
specific_fileids=specific_fileids))[0]
2041
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
2042
signature = gpg_strategy.sign(plaintext)
2043
self.add_signature_text(revision_id, signature)
2046
def add_signature_text(self, revision_id, signature):
2047
self.signatures.add_lines((revision_id,), (),
2048
osutils.split_lines(signature))
2050
def find_text_key_references(self):
2051
"""Find the text key references within the repository.
2053
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2054
to whether they were referred to by the inventory of the
2055
revision_id that they contain. The inventory texts from all present
2056
revision ids are assessed to generate this report.
2058
revision_keys = self.revisions.keys()
2059
w = self.inventories
2060
pb = ui.ui_factory.nested_progress_bar()
2062
return self._serializer._find_text_key_references(
2063
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
2067
def _inventory_xml_lines_for_keys(self, keys):
2068
"""Get a line iterator of the sort needed for findind references.
2070
Not relevant for non-xml inventory repositories.
2072
Ghosts in revision_keys are ignored.
2074
:param revision_keys: The revision keys for the inventories to inspect.
2075
:return: An iterator over (inventory line, revid) for the fulltexts of
2076
all of the xml inventories specified by revision_keys.
2078
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2079
for record in stream:
2080
if record.storage_kind != 'absent':
2081
chunks = record.get_bytes_as('chunked')
2082
revid = record.key[-1]
2083
lines = osutils.chunks_to_lines(chunks)
2087
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2089
"""Helper routine for fileids_altered_by_revision_ids.
2091
This performs the translation of xml lines to revision ids.
2093
:param line_iterator: An iterator of lines, origin_version_id
2094
:param revision_keys: The revision ids to filter for. This should be a
2095
set or other type which supports efficient __contains__ lookups, as
2096
the revision key from each parsed line will be looked up in the
2097
revision_keys filter.
2098
:return: a dictionary mapping altered file-ids to an iterable of
2099
revision_ids. Each altered file-ids has the exact revision_ids that
2100
altered it listed explicitly.
2102
seen = set(self._serializer._find_text_key_references(
2103
line_iterator).iterkeys())
2104
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2105
parent_seen = set(self._serializer._find_text_key_references(
2106
self._inventory_xml_lines_for_keys(parent_keys)))
2107
new_keys = seen - parent_seen
2109
setdefault = result.setdefault
2110
for key in new_keys:
2111
setdefault(key[0], set()).add(key[-1])
2114
def _find_parent_ids_of_revisions(self, revision_ids):
2115
"""Find all parent ids that are mentioned in the revision graph.
2117
:return: set of revisions that are parents of revision_ids which are
2118
not part of revision_ids themselves
2120
parent_map = self.get_parent_map(revision_ids)
2122
map(parent_ids.update, parent_map.itervalues())
2123
parent_ids.difference_update(revision_ids)
2124
parent_ids.discard(_mod_revision.NULL_REVISION)
2127
def _find_parent_keys_of_revisions(self, revision_keys):
2128
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2130
:param revision_keys: An iterable of revision_keys.
2131
:return: The parents of all revision_keys that are not already in
2134
parent_map = self.revisions.get_parent_map(revision_keys)
2136
map(parent_keys.update, parent_map.itervalues())
2137
parent_keys.difference_update(revision_keys)
2138
parent_keys.discard(_mod_revision.NULL_REVISION)
2141
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2142
"""Find the file ids and versions affected by revisions.
2144
:param revisions: an iterable containing revision ids.
2145
:param _inv_weave: The inventory weave from this repository or None.
2146
If None, the inventory weave will be opened automatically.
2147
:return: a dictionary mapping altered file-ids to an iterable of
2148
revision_ids. Each altered file-ids has the exact revision_ids that
2149
altered it listed explicitly.
2151
selected_keys = set((revid,) for revid in revision_ids)
2152
w = _inv_weave or self.inventories
2153
return self._find_file_ids_from_xml_inventory_lines(
2154
w.iter_lines_added_or_present_in_keys(
2155
selected_keys, pb=None),
2158
def iter_files_bytes(self, desired_files):
2159
"""Iterate through file versions.
2161
Files will not necessarily be returned in the order they occur in
2162
desired_files. No specific order is guaranteed.
2164
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2165
value supplied by the caller as part of desired_files. It should
2166
uniquely identify the file version in the caller's context. (Examples:
2167
an index number or a TreeTransform trans_id.)
2169
bytes_iterator is an iterable of bytestrings for the file. The
2170
kind of iterable and length of the bytestrings are unspecified, but for
2171
this implementation, it is a list of bytes produced by
2172
VersionedFile.get_record_stream().
2174
:param desired_files: a list of (file_id, revision_id, identifier)
2178
for file_id, revision_id, callable_data in desired_files:
2179
text_keys[(file_id, revision_id)] = callable_data
2180
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2181
if record.storage_kind == 'absent':
2182
raise errors.RevisionNotPresent(record.key, self)
2183
yield text_keys[record.key], record.get_bytes_as('chunked')
2185
def _generate_text_key_index(self, text_key_references=None,
2187
"""Generate a new text key index for the repository.
2189
This is an expensive function that will take considerable time to run.
2191
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2192
list of parents, also text keys. When a given key has no parents,
2193
the parents list will be [NULL_REVISION].
2195
# All revisions, to find inventory parents.
2196
if ancestors is None:
2197
graph = self.get_graph()
2198
ancestors = graph.get_parent_map(self.all_revision_ids())
2199
if text_key_references is None:
2200
text_key_references = self.find_text_key_references()
2201
pb = ui.ui_factory.nested_progress_bar()
2203
return self._do_generate_text_key_index(ancestors,
2204
text_key_references, pb)
2208
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2209
"""Helper for _generate_text_key_index to avoid deep nesting."""
2210
revision_order = tsort.topo_sort(ancestors)
2211
invalid_keys = set()
2213
for revision_id in revision_order:
2214
revision_keys[revision_id] = set()
2215
text_count = len(text_key_references)
2216
# a cache of the text keys to allow reuse; costs a dict of all the
2217
# keys, but saves a 2-tuple for every child of a given key.
2219
for text_key, valid in text_key_references.iteritems():
2221
invalid_keys.add(text_key)
2223
revision_keys[text_key[1]].add(text_key)
2224
text_key_cache[text_key] = text_key
2225
del text_key_references
2227
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2228
NULL_REVISION = _mod_revision.NULL_REVISION
2229
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2230
# too small for large or very branchy trees. However, for 55K path
2231
# trees, it would be easy to use too much memory trivially. Ideally we
2232
# could gauge this by looking at available real memory etc, but this is
2233
# always a tricky proposition.
2234
inventory_cache = lru_cache.LRUCache(10)
2235
batch_size = 10 # should be ~150MB on a 55K path tree
2236
batch_count = len(revision_order) / batch_size + 1
2238
pb.update("Calculating text parents", processed_texts, text_count)
2239
for offset in xrange(batch_count):
2240
to_query = revision_order[offset * batch_size:(offset + 1) *
2244
for revision_id in to_query:
2245
parent_ids = ancestors[revision_id]
2246
for text_key in revision_keys[revision_id]:
2247
pb.update("Calculating text parents", processed_texts)
2248
processed_texts += 1
2249
candidate_parents = []
2250
for parent_id in parent_ids:
2251
parent_text_key = (text_key[0], parent_id)
2253
check_parent = parent_text_key not in \
2254
revision_keys[parent_id]
2256
# the parent parent_id is a ghost:
2257
check_parent = False
2258
# truncate the derived graph against this ghost.
2259
parent_text_key = None
2261
# look at the parent commit details inventories to
2262
# determine possible candidates in the per file graph.
2265
inv = inventory_cache[parent_id]
2267
inv = self.revision_tree(parent_id).inventory
2268
inventory_cache[parent_id] = inv
2270
parent_entry = inv[text_key[0]]
2271
except (KeyError, errors.NoSuchId):
2273
if parent_entry is not None:
2275
text_key[0], parent_entry.revision)
2277
parent_text_key = None
2278
if parent_text_key is not None:
2279
candidate_parents.append(
2280
text_key_cache[parent_text_key])
2281
parent_heads = text_graph.heads(candidate_parents)
2282
new_parents = list(parent_heads)
2283
new_parents.sort(key=lambda x:candidate_parents.index(x))
2284
if new_parents == []:
2285
new_parents = [NULL_REVISION]
2286
text_index[text_key] = new_parents
2288
for text_key in invalid_keys:
2289
text_index[text_key] = [NULL_REVISION]
2292
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2293
"""Get an iterable listing the keys of all the data introduced by a set
2296
The keys will be ordered so that the corresponding items can be safely
2297
fetched and inserted in that order.
2299
:returns: An iterable producing tuples of (knit-kind, file-id,
2300
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2301
'revisions'. file-id is None unless knit-kind is 'file'.
2303
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2306
for result in self._find_non_file_keys_to_fetch(revision_ids):
2309
def _find_file_keys_to_fetch(self, revision_ids, pb):
2310
# XXX: it's a bit weird to control the inventory weave caching in this
2311
# generator. Ideally the caching would be done in fetch.py I think. Or
2312
# maybe this generator should explicitly have the contract that it
2313
# should not be iterated until the previously yielded item has been
2315
inv_w = self.inventories
2317
# file ids that changed
2318
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2320
num_file_ids = len(file_ids)
2321
for file_id, altered_versions in file_ids.iteritems():
2323
pb.update("Fetch texts", count, num_file_ids)
2325
yield ("file", file_id, altered_versions)
2327
def _find_non_file_keys_to_fetch(self, revision_ids):
2329
yield ("inventory", None, revision_ids)
2332
# XXX: Note ATM no callers actually pay attention to this return
2333
# instead they just use the list of revision ids and ignore
2334
# missing sigs. Consider removing this work entirely
2335
revisions_with_signatures = set(self.signatures.get_parent_map(
2336
[(r,) for r in revision_ids]))
2337
revisions_with_signatures = set(
2338
[r for (r,) in revisions_with_signatures])
2339
revisions_with_signatures.intersection_update(revision_ids)
2340
yield ("signatures", None, revisions_with_signatures)
2343
yield ("revisions", None, revision_ids)
2346
def get_inventory(self, revision_id):
2347
"""Get Inventory object by revision id."""
2348
return self.iter_inventories([revision_id]).next()
2350
def iter_inventories(self, revision_ids, ordering=None):
2351
"""Get many inventories by revision_ids.
2353
This will buffer some or all of the texts used in constructing the
2354
inventories in memory, but will only parse a single inventory at a
2357
:param revision_ids: The expected revision ids of the inventories.
2358
:param ordering: optional ordering, e.g. 'topological'. If not
2359
specified, the order of revision_ids will be preserved (by
2360
buffering if necessary).
2361
:return: An iterator of inventories.
2363
if ((None in revision_ids)
2364
or (_mod_revision.NULL_REVISION in revision_ids)):
2365
raise ValueError('cannot get null revision inventory')
2366
return self._iter_inventories(revision_ids, ordering)
2368
def _iter_inventories(self, revision_ids, ordering):
2369
"""single-document based inventory iteration."""
2370
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2371
for text, revision_id in inv_xmls:
2372
yield self._deserialise_inventory(revision_id, text)
2374
def _iter_inventory_xmls(self, revision_ids, ordering):
2375
if ordering is None:
2376
order_as_requested = True
2377
ordering = 'unordered'
2379
order_as_requested = False
2380
keys = [(revision_id,) for revision_id in revision_ids]
2383
if order_as_requested:
2384
key_iter = iter(keys)
2385
next_key = key_iter.next()
2386
stream = self.inventories.get_record_stream(keys, ordering, True)
2388
for record in stream:
2389
if record.storage_kind != 'absent':
2390
chunks = record.get_bytes_as('chunked')
2391
if order_as_requested:
2392
text_chunks[record.key] = chunks
2394
yield ''.join(chunks), record.key[-1]
2396
raise errors.NoSuchRevision(self, record.key)
2397
if order_as_requested:
2398
# Yield as many results as we can while preserving order.
2399
while next_key in text_chunks:
2400
chunks = text_chunks.pop(next_key)
2401
yield ''.join(chunks), next_key[-1]
2403
next_key = key_iter.next()
2404
except StopIteration:
2405
# We still want to fully consume the get_record_stream,
2406
# just in case it is not actually finished at this point
2410
def _deserialise_inventory(self, revision_id, xml):
2411
"""Transform the xml into an inventory object.
2413
:param revision_id: The expected revision id of the inventory.
2414
:param xml: A serialised inventory.
2416
result = self._serializer.read_inventory_from_string(xml, revision_id,
2417
entry_cache=self._inventory_entry_cache,
2418
return_from_cache=self._safe_to_return_from_cache)
2419
if result.revision_id != revision_id:
2420
raise AssertionError('revision id mismatch %s != %s' % (
2421
result.revision_id, revision_id))
2424
def get_serializer_format(self):
2425
return self._serializer.format_num
2428
def _get_inventory_xml(self, revision_id):
2429
"""Get serialized inventory as a string."""
2430
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2432
text, revision_id = texts.next()
2433
except StopIteration:
2434
raise errors.HistoryMissing(self, 'inventory', revision_id)
2437
def get_rev_id_for_revno(self, revno, known_pair):
2438
"""Return the revision id of a revno, given a later (revno, revid)
2439
pair in the same history.
2441
:return: if found (True, revid). If the available history ran out
2442
before reaching the revno, then this returns
2443
(False, (closest_revno, closest_revid)).
2445
known_revno, known_revid = known_pair
2446
partial_history = [known_revid]
2447
distance_from_known = known_revno - revno
2448
if distance_from_known < 0:
2450
'requested revno (%d) is later than given known revno (%d)'
2451
% (revno, known_revno))
2454
self, partial_history, stop_index=distance_from_known)
2455
except errors.RevisionNotPresent, err:
2456
if err.revision_id == known_revid:
2457
# The start revision (known_revid) wasn't found.
2459
# This is a stacked repository with no fallbacks, or a there's a
2460
# left-hand ghost. Either way, even though the revision named in
2461
# the error isn't in this repo, we know it's the next step in this
2462
# left-hand history.
2463
partial_history.append(err.revision_id)
2464
if len(partial_history) <= distance_from_known:
2465
# Didn't find enough history to get a revid for the revno.
2466
earliest_revno = known_revno - len(partial_history) + 1
2467
return (False, (earliest_revno, partial_history[-1]))
2468
if len(partial_history) - 1 > distance_from_known:
2469
raise AssertionError('_iter_for_revno returned too much history')
2470
return (True, partial_history[-1])
2472
def iter_reverse_revision_history(self, revision_id):
2473
"""Iterate backwards through revision ids in the lefthand history
2475
:param revision_id: The revision id to start with. All its lefthand
2476
ancestors will be traversed.
2478
graph = self.get_graph()
2479
stop_revisions = (None, _mod_revision.NULL_REVISION)
2480
return graph.iter_lefthand_ancestry(revision_id, stop_revisions)
2482
def is_shared(self):
2483
"""Return True if this repository is flagged as a shared repository."""
2484
raise NotImplementedError(self.is_shared)
2487
def reconcile(self, other=None, thorough=False):
2488
"""Reconcile this repository."""
2489
from bzrlib.reconcile import RepoReconciler
2490
reconciler = RepoReconciler(self, thorough=thorough)
2491
reconciler.reconcile()
2494
def _refresh_data(self):
2495
"""Helper called from lock_* to ensure coherency with disk.
2497
The default implementation does nothing; it is however possible
2498
for repositories to maintain loaded indices across multiple locks
2499
by checking inside their implementation of this method to see
2500
whether their indices are still valid. This depends of course on
2501
the disk format being validatable in this manner. This method is
2502
also called by the refresh_data() public interface to cause a refresh
2503
to occur while in a write lock so that data inserted by a smart server
2504
push operation is visible on the client's instance of the physical
2509
def revision_tree(self, revision_id):
2510
"""Return Tree for a revision on this branch.
2512
`revision_id` may be NULL_REVISION for the empty tree revision.
2514
revision_id = _mod_revision.ensure_null(revision_id)
2515
# TODO: refactor this to use an existing revision object
2516
# so we don't need to read it in twice.
2517
if revision_id == _mod_revision.NULL_REVISION:
2518
return InventoryRevisionTree(self,
2519
Inventory(root_id=None), _mod_revision.NULL_REVISION)
2521
inv = self.get_inventory(revision_id)
2522
return InventoryRevisionTree(self, inv, revision_id)
2524
def revision_trees(self, revision_ids):
2525
"""Return Trees for revisions in this repository.
2527
:param revision_ids: a sequence of revision-ids;
2528
a revision-id may not be None or 'null:'
2530
inventories = self.iter_inventories(revision_ids)
2531
for inv in inventories:
2532
yield InventoryRevisionTree(self, inv, inv.revision_id)
2534
def _filtered_revision_trees(self, revision_ids, file_ids):
2535
"""Return Tree for a revision on this branch with only some files.
2537
:param revision_ids: a sequence of revision-ids;
2538
a revision-id may not be None or 'null:'
2539
:param file_ids: if not None, the result is filtered
2540
so that only those file-ids, their parents and their
2541
children are included.
2543
inventories = self.iter_inventories(revision_ids)
2544
for inv in inventories:
2545
# Should we introduce a FilteredRevisionTree class rather
2546
# than pre-filter the inventory here?
2547
filtered_inv = inv.filter(file_ids)
2548
yield InventoryRevisionTree(self, filtered_inv, filtered_inv.revision_id)
2551
def get_ancestry(self, revision_id, topo_sorted=True):
2552
"""Return a list of revision-ids integrated by a revision.
2554
The first element of the list is always None, indicating the origin
2555
revision. This might change when we have history horizons, or
2556
perhaps we should have a new API.
2558
This is topologically sorted.
2560
if _mod_revision.is_null(revision_id):
2562
if not self.has_revision(revision_id):
2563
raise errors.NoSuchRevision(self, revision_id)
2564
graph = self.get_graph()
2566
search = graph._make_breadth_first_searcher([revision_id])
2569
found, ghosts = search.next_with_ghosts()
2570
except StopIteration:
2573
if _mod_revision.NULL_REVISION in keys:
2574
keys.remove(_mod_revision.NULL_REVISION)
2576
parent_map = graph.get_parent_map(keys)
2577
keys = tsort.topo_sort(parent_map)
2578
return [None] + list(keys)
2580
def pack(self, hint=None, clean_obsolete_packs=False):
2581
"""Compress the data within the repository.
2583
This operation only makes sense for some repository types. For other
2584
types it should be a no-op that just returns.
2586
This stub method does not require a lock, but subclasses should use
2587
@needs_write_lock as this is a long running call it's reasonable to
2588
implicitly lock for the user.
2590
:param hint: If not supplied, the whole repository is packed.
2591
If supplied, the repository may use the hint parameter as a
2592
hint for the parts of the repository to pack. A hint can be
2593
obtained from the result of commit_write_group(). Out of
2594
date hints are simply ignored, because concurrent operations
2595
can obsolete them rapidly.
2597
:param clean_obsolete_packs: Clean obsolete packs immediately after
2601
def get_transaction(self):
2602
return self.control_files.get_transaction()
2604
def get_parent_map(self, revision_ids):
2605
"""See graph.StackedParentsProvider.get_parent_map"""
2606
# revisions index works in keys; this just works in revisions
2607
# therefore wrap and unwrap
2610
for revision_id in revision_ids:
2611
if revision_id == _mod_revision.NULL_REVISION:
2612
result[revision_id] = ()
2613
elif revision_id is None:
2614
raise ValueError('get_parent_map(None) is not valid')
2616
query_keys.append((revision_id ,))
2617
for ((revision_id,), parent_keys) in \
2618
self.revisions.get_parent_map(query_keys).iteritems():
2620
result[revision_id] = tuple([parent_revid
2621
for (parent_revid,) in parent_keys])
2623
result[revision_id] = (_mod_revision.NULL_REVISION,)
2626
def _make_parents_provider(self):
2630
def get_known_graph_ancestry(self, revision_ids):
2631
"""Return the known graph for a set of revision ids and their ancestors.
2633
st = static_tuple.StaticTuple
2634
revision_keys = [st(r_id).intern() for r_id in revision_ids]
2635
known_graph = self.revisions.get_known_graph_ancestry(revision_keys)
2636
return graph.GraphThunkIdsToKeys(known_graph)
2638
def get_graph(self, other_repository=None):
2639
"""Return the graph walker for this repository format"""
2640
parents_provider = self._make_parents_provider()
2641
if (other_repository is not None and
2642
not self.has_same_location(other_repository)):
2643
parents_provider = graph.StackedParentsProvider(
2644
[parents_provider, other_repository._make_parents_provider()])
2645
return graph.Graph(parents_provider)
2647
def _get_versioned_file_checker(self, text_key_references=None,
2649
"""Return an object suitable for checking versioned files.
2651
:param text_key_references: if non-None, an already built
2652
dictionary mapping text keys ((fileid, revision_id) tuples)
2653
to whether they were referred to by the inventory of the
2654
revision_id that they contain. If None, this will be
2656
:param ancestors: Optional result from
2657
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2660
return _VersionedFileChecker(self,
2661
text_key_references=text_key_references, ancestors=ancestors)
2663
def revision_ids_to_search_result(self, result_set):
2664
"""Convert a set of revision ids to a graph SearchResult."""
2665
result_parents = set()
2666
for parents in self.get_graph().get_parent_map(
2667
result_set).itervalues():
2668
result_parents.update(parents)
2669
included_keys = result_set.intersection(result_parents)
2670
start_keys = result_set.difference(included_keys)
2671
exclude_keys = result_parents.difference(result_set)
2672
result = graph.SearchResult(start_keys, exclude_keys,
2673
len(result_set), result_set)
2677
def set_make_working_trees(self, new_value):
2678
"""Set the policy flag for making working trees when creating branches.
2680
This only applies to branches that use this repository.
2682
The default is 'True'.
2683
:param new_value: True to restore the default, False to disable making
2686
raise NotImplementedError(self.set_make_working_trees)
2688
def make_working_trees(self):
2689
"""Returns the policy for making working trees on new branches."""
2690
raise NotImplementedError(self.make_working_trees)
2693
def sign_revision(self, revision_id, gpg_strategy):
2694
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2695
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2698
def has_signature_for_revision_id(self, revision_id):
2699
"""Query for a revision signature for revision_id in the repository."""
2700
if not self.has_revision(revision_id):
2701
raise errors.NoSuchRevision(self, revision_id)
2702
sig_present = (1 == len(
2703
self.signatures.get_parent_map([(revision_id,)])))
2707
def get_signature_text(self, revision_id):
2708
"""Return the text for a signature."""
2709
stream = self.signatures.get_record_stream([(revision_id,)],
2711
record = stream.next()
2712
if record.storage_kind == 'absent':
2713
raise errors.NoSuchRevision(self, revision_id)
2714
return record.get_bytes_as('fulltext')
2717
def check(self, revision_ids=None, callback_refs=None, check_repo=True):
2718
"""Check consistency of all history of given revision_ids.
2720
Different repository implementations should override _check().
2722
:param revision_ids: A non-empty list of revision_ids whose ancestry
2723
will be checked. Typically the last revision_id of a branch.
2724
:param callback_refs: A dict of check-refs to resolve and callback
2725
the check/_check method on the items listed as wanting the ref.
2727
:param check_repo: If False do not check the repository contents, just
2728
calculate the data callback_refs requires and call them back.
2730
return self._check(revision_ids, callback_refs=callback_refs,
2731
check_repo=check_repo)
2733
def _check(self, revision_ids, callback_refs, check_repo):
2734
result = check.Check(self, check_repo=check_repo)
2735
result.check(callback_refs)
2738
def _warn_if_deprecated(self, branch=None):
2739
if not self._format.is_deprecated():
2741
global _deprecation_warning_done
2742
if _deprecation_warning_done:
2746
conf = config.GlobalConfig()
2748
conf = branch.get_config()
2749
if conf.suppress_warning('format_deprecation'):
2751
warning("Format %s for %s is deprecated -"
2752
" please use 'bzr upgrade' to get better performance"
2753
% (self._format, self.bzrdir.transport.base))
2755
_deprecation_warning_done = True
2757
def supports_rich_root(self):
2758
return self._format.rich_root_data
2760
def _check_ascii_revisionid(self, revision_id, method):
2761
"""Private helper for ascii-only repositories."""
2762
# weave repositories refuse to store revisionids that are non-ascii.
2763
if revision_id is not None:
2764
# weaves require ascii revision ids.
2765
if isinstance(revision_id, unicode):
2767
revision_id.encode('ascii')
2768
except UnicodeEncodeError:
2769
raise errors.NonAsciiRevisionId(method, self)
2772
revision_id.decode('ascii')
2773
except UnicodeDecodeError:
2774
raise errors.NonAsciiRevisionId(method, self)
2776
def _find_inconsistent_revision_parents(self, revisions_iterator=None):
2777
"""Find revisions with different parent lists in the revision object
2778
and in the index graph.
2780
:param revisions_iterator: None, or an iterator of (revid,
2781
Revision-or-None). This iterator controls the revisions checked.
2782
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
2783
parents-in-revision).
2785
if not self.is_locked():
2786
raise AssertionError()
2788
if revisions_iterator is None:
2789
revisions_iterator = self._iter_revisions(None)
2790
for revid, revision in revisions_iterator:
2791
if revision is None:
2793
parent_map = vf.get_parent_map([(revid,)])
2794
parents_according_to_index = tuple(parent[-1] for parent in
2795
parent_map[(revid,)])
2796
parents_according_to_revision = tuple(revision.parent_ids)
2797
if parents_according_to_index != parents_according_to_revision:
2798
yield (revid, parents_according_to_index,
2799
parents_according_to_revision)
2801
def _check_for_inconsistent_revision_parents(self):
2802
inconsistencies = list(self._find_inconsistent_revision_parents())
2804
raise errors.BzrCheckError(
2805
"Revision knit has inconsistent parents.")
2808
def install_revision(repository, rev, revision_tree):
2809
"""Install all revision data into a repository."""
2810
install_revisions(repository, [(rev, revision_tree, None)])
2813
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2814
"""Install all revision data into a repository.
2816
Accepts an iterable of revision, tree, signature tuples. The signature
2819
repository.start_write_group()
2821
inventory_cache = lru_cache.LRUCache(10)
2822
for n, (revision, revision_tree, signature) in enumerate(iterable):
2823
_install_revision(repository, revision, revision_tree, signature,
2826
pb.update('Transferring revisions', n + 1, num_revisions)
2828
repository.abort_write_group()
2831
repository.commit_write_group()
2834
def _install_revision(repository, rev, revision_tree, signature,
2836
"""Install all revision data into a repository."""
2837
present_parents = []
2839
for p_id in rev.parent_ids:
2840
if repository.has_revision(p_id):
2841
present_parents.append(p_id)
2842
parent_trees[p_id] = repository.revision_tree(p_id)
2844
parent_trees[p_id] = repository.revision_tree(
2845
_mod_revision.NULL_REVISION)
2847
inv = revision_tree.inventory
2848
entries = inv.iter_entries()
2849
# backwards compatibility hack: skip the root id.
2850
if not repository.supports_rich_root():
2851
path, root = entries.next()
2852
if root.revision != rev.revision_id:
2853
raise errors.IncompatibleRevision(repr(repository))
2855
for path, ie in entries:
2856
text_keys[(ie.file_id, ie.revision)] = ie
2857
text_parent_map = repository.texts.get_parent_map(text_keys)
2858
missing_texts = set(text_keys) - set(text_parent_map)
2859
# Add the texts that are not already present
2860
for text_key in missing_texts:
2861
ie = text_keys[text_key]
2863
# FIXME: TODO: The following loop overlaps/duplicates that done by
2864
# commit to determine parents. There is a latent/real bug here where
2865
# the parents inserted are not those commit would do - in particular
2866
# they are not filtered by heads(). RBC, AB
2867
for revision, tree in parent_trees.iteritems():
2868
if ie.file_id not in tree:
2870
parent_id = tree.get_file_revision(ie.file_id)
2871
if parent_id in text_parents:
2873
text_parents.append((ie.file_id, parent_id))
2874
lines = revision_tree.get_file(ie.file_id).readlines()
2875
repository.texts.add_lines(text_key, text_parents, lines)
2877
# install the inventory
2878
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2879
# Cache this inventory
2880
inventory_cache[rev.revision_id] = inv
2882
basis_inv = inventory_cache[rev.parent_ids[0]]
2884
repository.add_inventory(rev.revision_id, inv, present_parents)
2886
delta = inv._make_delta(basis_inv)
2887
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2888
rev.revision_id, present_parents)
2890
repository.add_inventory(rev.revision_id, inv, present_parents)
2891
except errors.RevisionAlreadyPresent:
2893
if signature is not None:
2894
repository.add_signature_text(rev.revision_id, signature)
2895
repository.add_revision(rev.revision_id, rev, inv)
2898
class MetaDirRepository(Repository):
2899
"""Repositories in the new meta-dir layout.
2901
:ivar _transport: Transport for access to repository control files,
2902
typically pointing to .bzr/repository.
2905
def __init__(self, _format, a_bzrdir, control_files):
2906
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2907
self._transport = control_files._transport
2909
def is_shared(self):
2910
"""Return True if this repository is flagged as a shared repository."""
2911
return self._transport.has('shared-storage')
2914
def set_make_working_trees(self, new_value):
2915
"""Set the policy flag for making working trees when creating branches.
2917
This only applies to branches that use this repository.
2919
The default is 'True'.
2920
:param new_value: True to restore the default, False to disable making
2925
self._transport.delete('no-working-trees')
2926
except errors.NoSuchFile:
2929
self._transport.put_bytes('no-working-trees', '',
2930
mode=self.bzrdir._get_file_mode())
2932
def make_working_trees(self):
2933
"""Returns the policy for making working trees on new branches."""
2934
return not self._transport.has('no-working-trees')
2937
class MetaDirVersionedFileRepository(MetaDirRepository):
2938
"""Repositories in a meta-dir, that work via versioned file objects."""
2940
def __init__(self, _format, a_bzrdir, control_files):
2941
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2945
class RepositoryFormatRegistry(controldir.ControlComponentFormatRegistry):
2946
"""Repository format registry."""
2948
def get_default(self):
2949
"""Return the current default format."""
2950
from bzrlib import bzrdir
2951
return bzrdir.format_registry.make_bzrdir('default').repository_format
2954
network_format_registry = registry.FormatRegistry()
2955
"""Registry of formats indexed by their network name.
2957
The network name for a repository format is an identifier that can be used when
2958
referring to formats with smart server operations. See
2959
RepositoryFormat.network_name() for more detail.
2963
format_registry = RepositoryFormatRegistry(network_format_registry)
2964
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
2966
This can contain either format instances themselves, or classes/factories that
2967
can be called to obtain one.
2971
#####################################################################
2972
# Repository Formats
2974
class RepositoryFormat(controldir.ControlComponentFormat):
2975
"""A repository format.
2977
Formats provide four things:
2978
* An initialization routine to construct repository data on disk.
2979
* a optional format string which is used when the BzrDir supports
2981
* an open routine which returns a Repository instance.
2982
* A network name for referring to the format in smart server RPC
2985
There is one and only one Format subclass for each on-disk format. But
2986
there can be one Repository subclass that is used for several different
2987
formats. The _format attribute on a Repository instance can be used to
2988
determine the disk format.
2990
Formats are placed in a registry by their format string for reference
2991
during opening. These should be subclasses of RepositoryFormat for
2994
Once a format is deprecated, just deprecate the initialize and open
2995
methods on the format class. Do not deprecate the object, as the
2996
object may be created even when a repository instance hasn't been
2999
Common instance attributes:
3000
_matchingbzrdir - the bzrdir format that the repository format was
3001
originally written to work with. This can be used if manually
3002
constructing a bzrdir and repository, or more commonly for test suite
3006
# Set to True or False in derived classes. True indicates that the format
3007
# supports ghosts gracefully.
3008
supports_ghosts = None
3009
# Can this repository be given external locations to lookup additional
3010
# data. Set to True or False in derived classes.
3011
supports_external_lookups = None
3012
# Does this format support CHK bytestring lookups. Set to True or False in
3014
supports_chks = None
3015
# Should commit add an inventory, or an inventory delta to the repository.
3016
_commit_inv_deltas = True
3017
# What order should fetch operations request streams in?
3018
# The default is unordered as that is the cheapest for an origin to
3020
_fetch_order = 'unordered'
3021
# Does this repository format use deltas that can be fetched as-deltas ?
3022
# (E.g. knits, where the knit deltas can be transplanted intact.
3023
# We default to False, which will ensure that enough data to get
3024
# a full text out of any fetch stream will be grabbed.
3025
_fetch_uses_deltas = False
3026
# Should fetch trigger a reconcile after the fetch? Only needed for
3027
# some repository formats that can suffer internal inconsistencies.
3028
_fetch_reconcile = False
3029
# Does this format have < O(tree_size) delta generation. Used to hint what
3030
# code path for commit, amongst other things.
3032
# Does doing a pack operation compress data? Useful for the pack UI command
3033
# (so if there is one pack, the operation can still proceed because it may
3034
# help), and for fetching when data won't have come from the same
3036
pack_compresses = False
3037
# Does the repository inventory storage understand references to trees?
3038
supports_tree_reference = None
3039
# Is the format experimental ?
3040
experimental = False
3041
# Does this repository format escape funky characters, or does it create files with
3042
# similar names as the versioned files in its contents on disk ?
3043
supports_funky_characters = None
3044
# Does this repository format support leaving locks?
3045
supports_leaving_lock = None
3046
# Does this format support the full VersionedFiles interface?
3047
supports_full_versioned_files = None
3048
# Does this format support signing revision signatures?
3049
supports_revision_signatures = True
3050
# Can the revision graph have incorrect parents?
3051
revision_graph_can_have_wrong_parents = None
3054
return "%s()" % self.__class__.__name__
3056
def __eq__(self, other):
3057
# format objects are generally stateless
3058
return isinstance(other, self.__class__)
3060
def __ne__(self, other):
3061
return not self == other
3064
def find_format(klass, a_bzrdir):
3065
"""Return the format for the repository object in a_bzrdir.
3067
This is used by bzr native formats that have a "format" file in
3068
the repository. Other methods may be used by different types of
3072
transport = a_bzrdir.get_repository_transport(None)
3073
format_string = transport.get_bytes("format")
3074
return format_registry.get(format_string)
3075
except errors.NoSuchFile:
3076
raise errors.NoRepositoryPresent(a_bzrdir)
3078
raise errors.UnknownFormatError(format=format_string,
3082
@symbol_versioning.deprecated_method(symbol_versioning.deprecated_in((2, 4, 0)))
3083
def register_format(klass, format):
3084
format_registry.register(format)
3087
@symbol_versioning.deprecated_method(symbol_versioning.deprecated_in((2, 4, 0)))
3088
def unregister_format(klass, format):
3089
format_registry.remove(format)
3092
@symbol_versioning.deprecated_method(symbol_versioning.deprecated_in((2, 4, 0)))
3093
def get_default_format(klass):
3094
"""Return the current default format."""
3095
return format_registry.get_default()
3097
def get_format_string(self):
3098
"""Return the ASCII format string that identifies this format.
3100
Note that in pre format ?? repositories the format string is
3101
not permitted nor written to disk.
3103
raise NotImplementedError(self.get_format_string)
3105
def get_format_description(self):
3106
"""Return the short description for this format."""
3107
raise NotImplementedError(self.get_format_description)
3109
# TODO: this shouldn't be in the base class, it's specific to things that
3110
# use weaves or knits -- mbp 20070207
3111
def _get_versioned_file_store(self,
3116
versionedfile_class=None,
3117
versionedfile_kwargs={},
3119
if versionedfile_class is None:
3120
versionedfile_class = self._versionedfile_class
3121
weave_transport = control_files._transport.clone(name)
3122
dir_mode = control_files._dir_mode
3123
file_mode = control_files._file_mode
3124
return VersionedFileStore(weave_transport, prefixed=prefixed,
3126
file_mode=file_mode,
3127
versionedfile_class=versionedfile_class,
3128
versionedfile_kwargs=versionedfile_kwargs,
3131
def initialize(self, a_bzrdir, shared=False):
3132
"""Initialize a repository of this format in a_bzrdir.
3134
:param a_bzrdir: The bzrdir to put the new repository in it.
3135
:param shared: The repository should be initialized as a sharable one.
3136
:returns: The new repository object.
3138
This may raise UninitializableFormat if shared repository are not
3139
compatible the a_bzrdir.
3141
raise NotImplementedError(self.initialize)
3143
def is_supported(self):
3144
"""Is this format supported?
3146
Supported formats must be initializable and openable.
3147
Unsupported formats may not support initialization or committing or
3148
some other features depending on the reason for not being supported.
3152
def is_deprecated(self):
3153
"""Is this format deprecated?
3155
Deprecated formats may trigger a user-visible warning recommending
3156
the user to upgrade. They are still fully supported.
3160
def network_name(self):
3161
"""A simple byte string uniquely identifying this format for RPC calls.
3163
MetaDir repository formats use their disk format string to identify the
3164
repository over the wire. All in one formats such as bzr < 0.8, and
3165
foreign formats like svn/git and hg should use some marker which is
3166
unique and immutable.
3168
raise NotImplementedError(self.network_name)
3170
def check_conversion_target(self, target_format):
3171
if self.rich_root_data and not target_format.rich_root_data:
3172
raise errors.BadConversionTarget(
3173
'Does not support rich root data.', target_format,
3175
if (self.supports_tree_reference and
3176
not getattr(target_format, 'supports_tree_reference', False)):
3177
raise errors.BadConversionTarget(
3178
'Does not support nested trees', target_format,
3181
def open(self, a_bzrdir, _found=False):
3182
"""Return an instance of this format for the bzrdir a_bzrdir.
3184
_found is a private parameter, do not use it.
3186
raise NotImplementedError(self.open)
3188
def _run_post_repo_init_hooks(self, repository, a_bzrdir, shared):
3189
from bzrlib.bzrdir import BzrDir, RepoInitHookParams
3190
hooks = BzrDir.hooks['post_repo_init']
3193
params = RepoInitHookParams(repository, self, a_bzrdir, shared)
3198
class MetaDirRepositoryFormat(RepositoryFormat):
3199
"""Common base class for the new repositories using the metadir layout."""
3201
rich_root_data = False
3202
supports_tree_reference = False
3203
supports_external_lookups = False
3204
supports_leaving_lock = True
3207
def _matchingbzrdir(self):
3208
matching = bzrdir.BzrDirMetaFormat1()
3209
matching.repository_format = self
3213
super(MetaDirRepositoryFormat, self).__init__()
3215
def _create_control_files(self, a_bzrdir):
3216
"""Create the required files and the initial control_files object."""
3217
# FIXME: RBC 20060125 don't peek under the covers
3218
# NB: no need to escape relative paths that are url safe.
3219
repository_transport = a_bzrdir.get_repository_transport(self)
3220
control_files = lockable_files.LockableFiles(repository_transport,
3221
'lock', lockdir.LockDir)
3222
control_files.create_lock()
3223
return control_files
3225
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
3226
"""Upload the initial blank content."""
3227
control_files = self._create_control_files(a_bzrdir)
3228
control_files.lock_write()
3229
transport = control_files._transport
3231
utf8_files += [('shared-storage', '')]
3233
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
3234
for (filename, content_stream) in files:
3235
transport.put_file(filename, content_stream,
3236
mode=a_bzrdir._get_file_mode())
3237
for (filename, content_bytes) in utf8_files:
3238
transport.put_bytes_non_atomic(filename, content_bytes,
3239
mode=a_bzrdir._get_file_mode())
3241
control_files.unlock()
3243
def network_name(self):
3244
"""Metadir formats have matching disk and network format strings."""
3245
return self.get_format_string()
3248
# formats which have no format string are not discoverable or independently
3249
# creatable on disk, so are not registered in format_registry. They're
3250
# all in bzrlib.repofmt.knitreponow. When an instance of one of these is
3251
# needed, it's constructed directly by the BzrDir. Non-native formats where
3252
# the repository is not separately opened are similar.
3254
format_registry.register_lazy(
3255
'Bazaar-NG Knit Repository Format 1',
3256
'bzrlib.repofmt.knitrepo',
3257
'RepositoryFormatKnit1',
3260
format_registry.register_lazy(
3261
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
3262
'bzrlib.repofmt.knitrepo',
3263
'RepositoryFormatKnit3',
3266
format_registry.register_lazy(
3267
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
3268
'bzrlib.repofmt.knitrepo',
3269
'RepositoryFormatKnit4',
3272
# Pack-based formats. There is one format for pre-subtrees, and one for
3273
# post-subtrees to allow ease of testing.
3274
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
3275
format_registry.register_lazy(
3276
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
3277
'bzrlib.repofmt.knitpack_repo',
3278
'RepositoryFormatKnitPack1',
3280
format_registry.register_lazy(
3281
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
3282
'bzrlib.repofmt.knitpack_repo',
3283
'RepositoryFormatKnitPack3',
3285
format_registry.register_lazy(
3286
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
3287
'bzrlib.repofmt.knitpack_repo',
3288
'RepositoryFormatKnitPack4',
3290
format_registry.register_lazy(
3291
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
3292
'bzrlib.repofmt.knitpack_repo',
3293
'RepositoryFormatKnitPack5',
3295
format_registry.register_lazy(
3296
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
3297
'bzrlib.repofmt.knitpack_repo',
3298
'RepositoryFormatKnitPack5RichRoot',
3300
format_registry.register_lazy(
3301
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3302
'bzrlib.repofmt.knitpack_repo',
3303
'RepositoryFormatKnitPack5RichRootBroken',
3305
format_registry.register_lazy(
3306
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3307
'bzrlib.repofmt.knitpack_repo',
3308
'RepositoryFormatKnitPack6',
3310
format_registry.register_lazy(
3311
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3312
'bzrlib.repofmt.knitpack_repo',
3313
'RepositoryFormatKnitPack6RichRoot',
3315
format_registry.register_lazy(
3316
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3317
'bzrlib.repofmt.groupcompress_repo',
3318
'RepositoryFormat2a',
3321
# Development formats.
3322
# Check their docstrings to see if/when they are obsolete.
3323
format_registry.register_lazy(
3324
("Bazaar development format 2 with subtree support "
3325
"(needs bzr.dev from before 1.8)\n"),
3326
'bzrlib.repofmt.knitpack_repo',
3327
'RepositoryFormatPackDevelopment2Subtree',
3329
format_registry.register_lazy(
3330
'Bazaar development format 8\n',
3331
'bzrlib.repofmt.groupcompress_repo',
3332
'RepositoryFormat2aSubtree',
3336
class InterRepository(InterObject):
3337
"""This class represents operations taking place between two repositories.
3339
Its instances have methods like copy_content and fetch, and contain
3340
references to the source and target repositories these operations can be
3343
Often we will provide convenience methods on 'repository' which carry out
3344
operations with another repository - they will always forward to
3345
InterRepository.get(other).method_name(parameters).
3348
_walk_to_common_revisions_batch_size = 50
3350
"""The available optimised InterRepository types."""
3353
def copy_content(self, revision_id=None):
3354
"""Make a complete copy of the content in self into destination.
3356
This is a destructive operation! Do not use it on existing
3359
:param revision_id: Only copy the content needed to construct
3360
revision_id and its parents.
3363
self.target.set_make_working_trees(self.source.make_working_trees())
3364
except NotImplementedError:
3366
self.target.fetch(self.source, revision_id=revision_id)
3369
def fetch(self, revision_id=None, find_ghosts=False,
3371
"""Fetch the content required to construct revision_id.
3373
The content is copied from self.source to self.target.
3375
:param revision_id: if None all content is copied, if NULL_REVISION no
3379
ui.ui_factory.warn_experimental_format_fetch(self)
3380
from bzrlib.fetch import RepoFetcher
3381
# See <https://launchpad.net/bugs/456077> asking for a warning here
3382
if self.source._format.network_name() != self.target._format.network_name():
3383
ui.ui_factory.show_user_warning('cross_format_fetch',
3384
from_format=self.source._format,
3385
to_format=self.target._format)
3386
f = RepoFetcher(to_repository=self.target,
3387
from_repository=self.source,
3388
last_revision=revision_id,
3389
fetch_spec=fetch_spec,
3390
find_ghosts=find_ghosts)
3392
def _walk_to_common_revisions(self, revision_ids, if_present_ids=None):
3393
"""Walk out from revision_ids in source to revisions target has.
3395
:param revision_ids: The start point for the search.
3396
:return: A set of revision ids.
3398
target_graph = self.target.get_graph()
3399
revision_ids = frozenset(revision_ids)
3401
all_wanted_revs = revision_ids.union(if_present_ids)
3403
all_wanted_revs = revision_ids
3404
missing_revs = set()
3405
source_graph = self.source.get_graph()
3406
# ensure we don't pay silly lookup costs.
3407
searcher = source_graph._make_breadth_first_searcher(all_wanted_revs)
3408
null_set = frozenset([_mod_revision.NULL_REVISION])
3409
searcher_exhausted = False
3413
# Iterate the searcher until we have enough next_revs
3414
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3416
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3417
next_revs.update(next_revs_part)
3418
ghosts.update(ghosts_part)
3419
except StopIteration:
3420
searcher_exhausted = True
3422
# If there are ghosts in the source graph, and the caller asked for
3423
# them, make sure that they are present in the target.
3424
# We don't care about other ghosts as we can't fetch them and
3425
# haven't been asked to.
3426
ghosts_to_check = set(revision_ids.intersection(ghosts))
3427
revs_to_get = set(next_revs).union(ghosts_to_check)
3429
have_revs = set(target_graph.get_parent_map(revs_to_get))
3430
# we always have NULL_REVISION present.
3431
have_revs = have_revs.union(null_set)
3432
# Check if the target is missing any ghosts we need.
3433
ghosts_to_check.difference_update(have_revs)
3435
# One of the caller's revision_ids is a ghost in both the
3436
# source and the target.
3437
raise errors.NoSuchRevision(
3438
self.source, ghosts_to_check.pop())
3439
missing_revs.update(next_revs - have_revs)
3440
# Because we may have walked past the original stop point, make
3441
# sure everything is stopped
3442
stop_revs = searcher.find_seen_ancestors(have_revs)
3443
searcher.stop_searching_any(stop_revs)
3444
if searcher_exhausted:
3446
return searcher.get_result()
3449
def search_missing_revision_ids(self,
3450
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
3451
find_ghosts=True, revision_ids=None, if_present_ids=None):
3452
"""Return the revision ids that source has that target does not.
3454
:param revision_id: only return revision ids included by this
3456
:param revision_ids: return revision ids included by these
3457
revision_ids. NoSuchRevision will be raised if any of these
3458
revisions are not present.
3459
:param if_present_ids: like revision_ids, but will not cause
3460
NoSuchRevision if any of these are absent, instead they will simply
3461
not be in the result. This is useful for e.g. finding revisions
3462
to fetch for tags, which may reference absent revisions.
3463
:param find_ghosts: If True find missing revisions in deep history
3464
rather than just finding the surface difference.
3465
:return: A bzrlib.graph.SearchResult.
3467
if symbol_versioning.deprecated_passed(revision_id):
3468
symbol_versioning.warn(
3469
'search_missing_revision_ids(revision_id=...) was '
3470
'deprecated in 2.4. Use revision_ids=[...] instead.',
3471
DeprecationWarning, stacklevel=2)
3472
if revision_ids is not None:
3473
raise AssertionError(
3474
'revision_ids is mutually exclusive with revision_id')
3475
if revision_id is not None:
3476
revision_ids = [revision_id]
3478
# stop searching at found target revisions.
3479
if not find_ghosts and (revision_ids is not None or if_present_ids is
3481
return self._walk_to_common_revisions(revision_ids,
3482
if_present_ids=if_present_ids)
3483
# generic, possibly worst case, slow code path.
3484
target_ids = set(self.target.all_revision_ids())
3485
source_ids = self._present_source_revisions_for(
3486
revision_ids, if_present_ids)
3487
result_set = set(source_ids).difference(target_ids)
3488
return self.source.revision_ids_to_search_result(result_set)
3490
def _present_source_revisions_for(self, revision_ids, if_present_ids=None):
3491
"""Returns set of all revisions in ancestry of revision_ids present in
3494
:param revision_ids: if None, all revisions in source are returned.
3495
:param if_present_ids: like revision_ids, but if any/all of these are
3496
absent no error is raised.
3498
if revision_ids is not None or if_present_ids is not None:
3499
# First, ensure all specified revisions exist. Callers expect
3500
# NoSuchRevision when they pass absent revision_ids here.
3501
if revision_ids is None:
3502
revision_ids = set()
3503
if if_present_ids is None:
3504
if_present_ids = set()
3505
revision_ids = set(revision_ids)
3506
if_present_ids = set(if_present_ids)
3507
all_wanted_ids = revision_ids.union(if_present_ids)
3508
graph = self.source.get_graph()
3509
present_revs = set(graph.get_parent_map(all_wanted_ids))
3510
missing = revision_ids.difference(present_revs)
3512
raise errors.NoSuchRevision(self.source, missing.pop())
3513
found_ids = all_wanted_ids.intersection(present_revs)
3514
source_ids = [rev_id for (rev_id, parents) in
3515
graph.iter_ancestry(found_ids)
3516
if rev_id != _mod_revision.NULL_REVISION
3517
and parents is not None]
3519
source_ids = self.source.all_revision_ids()
3520
return set(source_ids)
3523
def _same_model(source, target):
3524
"""True if source and target have the same data representation.
3526
Note: this is always called on the base class; overriding it in a
3527
subclass will have no effect.
3530
InterRepository._assert_same_model(source, target)
3532
except errors.IncompatibleRepositories, e:
3536
def _assert_same_model(source, target):
3537
"""Raise an exception if two repositories do not use the same model.
3539
if source.supports_rich_root() != target.supports_rich_root():
3540
raise errors.IncompatibleRepositories(source, target,
3541
"different rich-root support")
3542
if source._serializer != target._serializer:
3543
raise errors.IncompatibleRepositories(source, target,
3544
"different serializers")
3547
class InterSameDataRepository(InterRepository):
3548
"""Code for converting between repositories that represent the same data.
3550
Data format and model must match for this to work.
3554
def _get_repo_format_to_test(self):
3555
"""Repository format for testing with.
3557
InterSameData can pull from subtree to subtree and from non-subtree to
3558
non-subtree, so we test this with the richest repository format.
3560
from bzrlib.repofmt import knitrepo
3561
return knitrepo.RepositoryFormatKnit3()
3564
def is_compatible(source, target):
3565
return InterRepository._same_model(source, target)
3568
class InterDifferingSerializer(InterRepository):
3571
def _get_repo_format_to_test(self):
3575
def is_compatible(source, target):
3576
"""Be compatible with Knit2 source and Knit3 target"""
3577
# This is redundant with format.check_conversion_target(), however that
3578
# raises an exception, and we just want to say "False" as in we won't
3579
# support converting between these formats.
3580
if 'IDS_never' in debug.debug_flags:
3582
if source.supports_rich_root() and not target.supports_rich_root():
3584
if (source._format.supports_tree_reference
3585
and not target._format.supports_tree_reference):
3587
if target._fallback_repositories and target._format.supports_chks:
3588
# IDS doesn't know how to copy CHKs for the parent inventories it
3589
# adds to stacked repos.
3591
if 'IDS_always' in debug.debug_flags:
3593
# Only use this code path for local source and target. IDS does far
3594
# too much IO (both bandwidth and roundtrips) over a network.
3595
if not source.bzrdir.transport.base.startswith('file:///'):
3597
if not target.bzrdir.transport.base.startswith('file:///'):
3601
def _get_trees(self, revision_ids, cache):
3603
for rev_id in revision_ids:
3605
possible_trees.append((rev_id, cache[rev_id]))
3607
# Not cached, but inventory might be present anyway.
3609
tree = self.source.revision_tree(rev_id)
3610
except errors.NoSuchRevision:
3611
# Nope, parent is ghost.
3614
cache[rev_id] = tree
3615
possible_trees.append((rev_id, tree))
3616
return possible_trees
3618
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3619
"""Get the best delta and base for this revision.
3621
:return: (basis_id, delta)
3624
# Generate deltas against each tree, to find the shortest.
3625
texts_possibly_new_in_tree = set()
3626
for basis_id, basis_tree in possible_trees:
3627
delta = tree.inventory._make_delta(basis_tree.inventory)
3628
for old_path, new_path, file_id, new_entry in delta:
3629
if new_path is None:
3630
# This file_id isn't present in the new rev, so we don't
3634
# Rich roots are handled elsewhere...
3636
kind = new_entry.kind
3637
if kind != 'directory' and kind != 'file':
3638
# No text record associated with this inventory entry.
3640
# This is a directory or file that has changed somehow.
3641
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3642
deltas.append((len(delta), basis_id, delta))
3644
return deltas[0][1:]
3646
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3647
"""Find all parent revisions that are absent, but for which the
3648
inventory is present, and copy those inventories.
3650
This is necessary to preserve correctness when the source is stacked
3651
without fallbacks configured. (Note that in cases like upgrade the
3652
source may be not have _fallback_repositories even though it is
3656
for parents in parent_map.values():
3657
parent_revs.update(parents)
3658
present_parents = self.source.get_parent_map(parent_revs)
3659
absent_parents = set(parent_revs).difference(present_parents)
3660
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3661
(rev_id,) for rev_id in absent_parents)
3662
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3663
for parent_tree in self.source.revision_trees(parent_inv_ids):
3664
current_revision_id = parent_tree.get_revision_id()
3665
parents_parents_keys = parent_invs_keys_for_stacking[
3666
(current_revision_id,)]
3667
parents_parents = [key[-1] for key in parents_parents_keys]
3668
basis_id = _mod_revision.NULL_REVISION
3669
basis_tree = self.source.revision_tree(basis_id)
3670
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3671
self.target.add_inventory_by_delta(
3672
basis_id, delta, current_revision_id, parents_parents)
3673
cache[current_revision_id] = parent_tree
3675
def _fetch_batch(self, revision_ids, basis_id, cache):
3676
"""Fetch across a few revisions.
3678
:param revision_ids: The revisions to copy
3679
:param basis_id: The revision_id of a tree that must be in cache, used
3680
as a basis for delta when no other base is available
3681
:param cache: A cache of RevisionTrees that we can use.
3682
:return: The revision_id of the last converted tree. The RevisionTree
3683
for it will be in cache
3685
# Walk though all revisions; get inventory deltas, copy referenced
3686
# texts that delta references, insert the delta, revision and
3688
root_keys_to_create = set()
3691
pending_revisions = []
3692
parent_map = self.source.get_parent_map(revision_ids)
3693
self._fetch_parent_invs_for_stacking(parent_map, cache)
3694
self.source._safe_to_return_from_cache = True
3695
for tree in self.source.revision_trees(revision_ids):
3696
# Find a inventory delta for this revision.
3697
# Find text entries that need to be copied, too.
3698
current_revision_id = tree.get_revision_id()
3699
parent_ids = parent_map.get(current_revision_id, ())
3700
parent_trees = self._get_trees(parent_ids, cache)
3701
possible_trees = list(parent_trees)
3702
if len(possible_trees) == 0:
3703
# There either aren't any parents, or the parents are ghosts,
3704
# so just use the last converted tree.
3705
possible_trees.append((basis_id, cache[basis_id]))
3706
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3708
revision = self.source.get_revision(current_revision_id)
3709
pending_deltas.append((basis_id, delta,
3710
current_revision_id, revision.parent_ids))
3711
if self._converting_to_rich_root:
3712
self._revision_id_to_root_id[current_revision_id] = \
3714
# Determine which texts are in present in this revision but not in
3715
# any of the available parents.
3716
texts_possibly_new_in_tree = set()
3717
for old_path, new_path, file_id, entry in delta:
3718
if new_path is None:
3719
# This file_id isn't present in the new rev
3723
if not self.target.supports_rich_root():
3724
# The target doesn't support rich root, so we don't
3727
if self._converting_to_rich_root:
3728
# This can't be copied normally, we have to insert
3730
root_keys_to_create.add((file_id, entry.revision))
3733
texts_possibly_new_in_tree.add((file_id, entry.revision))
3734
for basis_id, basis_tree in possible_trees:
3735
basis_inv = basis_tree.inventory
3736
for file_key in list(texts_possibly_new_in_tree):
3737
file_id, file_revision = file_key
3739
entry = basis_inv[file_id]
3740
except errors.NoSuchId:
3742
if entry.revision == file_revision:
3743
texts_possibly_new_in_tree.remove(file_key)
3744
text_keys.update(texts_possibly_new_in_tree)
3745
pending_revisions.append(revision)
3746
cache[current_revision_id] = tree
3747
basis_id = current_revision_id
3748
self.source._safe_to_return_from_cache = False
3750
from_texts = self.source.texts
3751
to_texts = self.target.texts
3752
if root_keys_to_create:
3753
root_stream = _mod_fetch._new_root_data_stream(
3754
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3756
to_texts.insert_record_stream(root_stream)
3757
to_texts.insert_record_stream(from_texts.get_record_stream(
3758
text_keys, self.target._format._fetch_order,
3759
not self.target._format._fetch_uses_deltas))
3760
# insert inventory deltas
3761
for delta in pending_deltas:
3762
self.target.add_inventory_by_delta(*delta)
3763
if self.target._fallback_repositories:
3764
# Make sure this stacked repository has all the parent inventories
3765
# for the new revisions that we are about to insert. We do this
3766
# before adding the revisions so that no revision is added until
3767
# all the inventories it may depend on are added.
3768
# Note that this is overzealous, as we may have fetched these in an
3771
revision_ids = set()
3772
for revision in pending_revisions:
3773
revision_ids.add(revision.revision_id)
3774
parent_ids.update(revision.parent_ids)
3775
parent_ids.difference_update(revision_ids)
3776
parent_ids.discard(_mod_revision.NULL_REVISION)
3777
parent_map = self.source.get_parent_map(parent_ids)
3778
# we iterate over parent_map and not parent_ids because we don't
3779
# want to try copying any revision which is a ghost
3780
for parent_tree in self.source.revision_trees(parent_map):
3781
current_revision_id = parent_tree.get_revision_id()
3782
parents_parents = parent_map[current_revision_id]
3783
possible_trees = self._get_trees(parents_parents, cache)
3784
if len(possible_trees) == 0:
3785
# There either aren't any parents, or the parents are
3786
# ghosts, so just use the last converted tree.
3787
possible_trees.append((basis_id, cache[basis_id]))
3788
basis_id, delta = self._get_delta_for_revision(parent_tree,
3789
parents_parents, possible_trees)
3790
self.target.add_inventory_by_delta(
3791
basis_id, delta, current_revision_id, parents_parents)
3792
# insert signatures and revisions
3793
for revision in pending_revisions:
3795
signature = self.source.get_signature_text(
3796
revision.revision_id)
3797
self.target.add_signature_text(revision.revision_id,
3799
except errors.NoSuchRevision:
3801
self.target.add_revision(revision.revision_id, revision)
3804
def _fetch_all_revisions(self, revision_ids, pb):
3805
"""Fetch everything for the list of revisions.
3807
:param revision_ids: The list of revisions to fetch. Must be in
3809
:param pb: A ProgressTask
3812
basis_id, basis_tree = self._get_basis(revision_ids[0])
3814
cache = lru_cache.LRUCache(100)
3815
cache[basis_id] = basis_tree
3816
del basis_tree # We don't want to hang on to it here
3820
for offset in range(0, len(revision_ids), batch_size):
3821
self.target.start_write_group()
3823
pb.update('Transferring revisions', offset,
3825
batch = revision_ids[offset:offset+batch_size]
3826
basis_id = self._fetch_batch(batch, basis_id, cache)
3828
self.source._safe_to_return_from_cache = False
3829
self.target.abort_write_group()
3832
hint = self.target.commit_write_group()
3835
if hints and self.target._format.pack_compresses:
3836
self.target.pack(hint=hints)
3837
pb.update('Transferring revisions', len(revision_ids),
3841
def fetch(self, revision_id=None, find_ghosts=False,
3843
"""See InterRepository.fetch()."""
3844
if fetch_spec is not None:
3845
revision_ids = fetch_spec.get_keys()
3848
ui.ui_factory.warn_experimental_format_fetch(self)
3849
if (not self.source.supports_rich_root()
3850
and self.target.supports_rich_root()):
3851
self._converting_to_rich_root = True
3852
self._revision_id_to_root_id = {}
3854
self._converting_to_rich_root = False
3855
# See <https://launchpad.net/bugs/456077> asking for a warning here
3856
if self.source._format.network_name() != self.target._format.network_name():
3857
ui.ui_factory.show_user_warning('cross_format_fetch',
3858
from_format=self.source._format,
3859
to_format=self.target._format)
3860
if revision_ids is None:
3862
search_revision_ids = [revision_id]
3864
search_revision_ids = None
3865
revision_ids = self.target.search_missing_revision_ids(self.source,
3866
revision_ids=search_revision_ids,
3867
find_ghosts=find_ghosts).get_keys()
3868
if not revision_ids:
3870
revision_ids = tsort.topo_sort(
3871
self.source.get_graph().get_parent_map(revision_ids))
3872
if not revision_ids:
3874
# Walk though all revisions; get inventory deltas, copy referenced
3875
# texts that delta references, insert the delta, revision and
3877
pb = ui.ui_factory.nested_progress_bar()
3879
self._fetch_all_revisions(revision_ids, pb)
3882
return len(revision_ids), 0
3884
def _get_basis(self, first_revision_id):
3885
"""Get a revision and tree which exists in the target.
3887
This assumes that first_revision_id is selected for transmission
3888
because all other ancestors are already present. If we can't find an
3889
ancestor we fall back to NULL_REVISION since we know that is safe.
3891
:return: (basis_id, basis_tree)
3893
first_rev = self.source.get_revision(first_revision_id)
3895
basis_id = first_rev.parent_ids[0]
3896
# only valid as a basis if the target has it
3897
self.target.get_revision(basis_id)
3898
# Try to get a basis tree - if it's a ghost it will hit the
3899
# NoSuchRevision case.
3900
basis_tree = self.source.revision_tree(basis_id)
3901
except (IndexError, errors.NoSuchRevision):
3902
basis_id = _mod_revision.NULL_REVISION
3903
basis_tree = self.source.revision_tree(basis_id)
3904
return basis_id, basis_tree
3907
InterRepository.register_optimiser(InterDifferingSerializer)
3908
InterRepository.register_optimiser(InterSameDataRepository)
3911
class CopyConverter(object):
3912
"""A repository conversion tool which just performs a copy of the content.
3914
This is slow but quite reliable.
3917
def __init__(self, target_format):
3918
"""Create a CopyConverter.
3920
:param target_format: The format the resulting repository should be.
3922
self.target_format = target_format
3924
def convert(self, repo, pb):
3925
"""Perform the conversion of to_convert, giving feedback via pb.
3927
:param to_convert: The disk object to convert.
3928
:param pb: a progress bar to use for progress information.
3930
pb = ui.ui_factory.nested_progress_bar()
3933
# this is only useful with metadir layouts - separated repo content.
3934
# trigger an assertion if not such
3935
repo._format.get_format_string()
3936
self.repo_dir = repo.bzrdir
3937
pb.update('Moving repository to repository.backup')
3938
self.repo_dir.transport.move('repository', 'repository.backup')
3939
backup_transport = self.repo_dir.transport.clone('repository.backup')
3940
repo._format.check_conversion_target(self.target_format)
3941
self.source_repo = repo._format.open(self.repo_dir,
3943
_override_transport=backup_transport)
3944
pb.update('Creating new repository')
3945
converted = self.target_format.initialize(self.repo_dir,
3946
self.source_repo.is_shared())
3947
converted.lock_write()
3949
pb.update('Copying content')
3950
self.source_repo.copy_content_into(converted)
3953
pb.update('Deleting old repository content')
3954
self.repo_dir.transport.delete_tree('repository.backup')
3955
ui.ui_factory.note('repository converted')
3959
class _VersionedFileChecker(object):
3961
def __init__(self, repository, text_key_references=None, ancestors=None):
3962
self.repository = repository
3963
self.text_index = self.repository._generate_text_key_index(
3964
text_key_references=text_key_references, ancestors=ancestors)
3966
def calculate_file_version_parents(self, text_key):
3967
"""Calculate the correct parents for a file version according to
3970
parent_keys = self.text_index[text_key]
3971
if parent_keys == [_mod_revision.NULL_REVISION]:
3973
return tuple(parent_keys)
3975
def check_file_version_parents(self, texts, progress_bar=None):
3976
"""Check the parents stored in a versioned file are correct.
3978
It also detects file versions that are not referenced by their
3979
corresponding revision's inventory.
3981
:returns: A tuple of (wrong_parents, dangling_file_versions).
3982
wrong_parents is a dict mapping {revision_id: (stored_parents,
3983
correct_parents)} for each revision_id where the stored parents
3984
are not correct. dangling_file_versions is a set of (file_id,
3985
revision_id) tuples for versions that are present in this versioned
3986
file, but not used by the corresponding inventory.
3988
local_progress = None
3989
if progress_bar is None:
3990
local_progress = ui.ui_factory.nested_progress_bar()
3991
progress_bar = local_progress
3993
return self._check_file_version_parents(texts, progress_bar)
3996
local_progress.finished()
3998
def _check_file_version_parents(self, texts, progress_bar):
3999
"""See check_file_version_parents."""
4001
self.file_ids = set([file_id for file_id, _ in
4002
self.text_index.iterkeys()])
4003
# text keys is now grouped by file_id
4004
n_versions = len(self.text_index)
4005
progress_bar.update('loading text store', 0, n_versions)
4006
parent_map = self.repository.texts.get_parent_map(self.text_index)
4007
# On unlistable transports this could well be empty/error...
4008
text_keys = self.repository.texts.keys()
4009
unused_keys = frozenset(text_keys) - set(self.text_index)
4010
for num, key in enumerate(self.text_index.iterkeys()):
4011
progress_bar.update('checking text graph', num, n_versions)
4012
correct_parents = self.calculate_file_version_parents(key)
4014
knit_parents = parent_map[key]
4015
except errors.RevisionNotPresent:
4018
if correct_parents != knit_parents:
4019
wrong_parents[key] = (knit_parents, correct_parents)
4020
return wrong_parents, unused_keys
4023
def _strip_NULL_ghosts(revision_graph):
4024
"""Also don't use this. more compatibility code for unmigrated clients."""
4025
# Filter ghosts, and null:
4026
if _mod_revision.NULL_REVISION in revision_graph:
4027
del revision_graph[_mod_revision.NULL_REVISION]
4028
for key, parents in revision_graph.items():
4029
revision_graph[key] = tuple(parent for parent in parents if parent
4031
return revision_graph
4034
class StreamSink(object):
4035
"""An object that can insert a stream into a repository.
4037
This interface handles the complexity of reserialising inventories and
4038
revisions from different formats, and allows unidirectional insertion into
4039
stacked repositories without looking for the missing basis parents
4043
def __init__(self, target_repo):
4044
self.target_repo = target_repo
4046
def insert_stream(self, stream, src_format, resume_tokens):
4047
"""Insert a stream's content into the target repository.
4049
:param src_format: a bzr repository format.
4051
:return: a list of resume tokens and an iterable of keys additional
4052
items required before the insertion can be completed.
4054
self.target_repo.lock_write()
4057
self.target_repo.resume_write_group(resume_tokens)
4060
self.target_repo.start_write_group()
4063
# locked_insert_stream performs a commit|suspend.
4064
missing_keys = self.insert_stream_without_locking(stream,
4065
src_format, is_resume)
4067
# suspend the write group and tell the caller what we is
4068
# missing. We know we can suspend or else we would not have
4069
# entered this code path. (All repositories that can handle
4070
# missing keys can handle suspending a write group).
4071
write_group_tokens = self.target_repo.suspend_write_group()
4072
return write_group_tokens, missing_keys
4073
hint = self.target_repo.commit_write_group()
4074
to_serializer = self.target_repo._format._serializer
4075
src_serializer = src_format._serializer
4076
if (to_serializer != src_serializer and
4077
self.target_repo._format.pack_compresses):
4078
self.target_repo.pack(hint=hint)
4081
self.target_repo.abort_write_group(suppress_errors=True)
4084
self.target_repo.unlock()
4086
def insert_stream_without_locking(self, stream, src_format,
4088
"""Insert a stream's content into the target repository.
4090
This assumes that you already have a locked repository and an active
4093
:param src_format: a bzr repository format.
4094
:param is_resume: Passed down to get_missing_parent_inventories to
4095
indicate if we should be checking for missing texts at the same
4098
:return: A set of keys that are missing.
4100
if not self.target_repo.is_write_locked():
4101
raise errors.ObjectNotLocked(self)
4102
if not self.target_repo.is_in_write_group():
4103
raise errors.BzrError('you must already be in a write group')
4104
to_serializer = self.target_repo._format._serializer
4105
src_serializer = src_format._serializer
4107
if to_serializer == src_serializer:
4108
# If serializers match and the target is a pack repository, set the
4109
# write cache size on the new pack. This avoids poor performance
4110
# on transports where append is unbuffered (such as
4111
# RemoteTransport). This is safe to do because nothing should read
4112
# back from the target repository while a stream with matching
4113
# serialization is being inserted.
4114
# The exception is that a delta record from the source that should
4115
# be a fulltext may need to be expanded by the target (see
4116
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4117
# explicitly flush any buffered writes first in that rare case.
4119
new_pack = self.target_repo._pack_collection._new_pack
4120
except AttributeError:
4121
# Not a pack repository
4124
new_pack.set_write_cache_size(1024*1024)
4125
for substream_type, substream in stream:
4126
if 'stream' in debug.debug_flags:
4127
mutter('inserting substream: %s', substream_type)
4128
if substream_type == 'texts':
4129
self.target_repo.texts.insert_record_stream(substream)
4130
elif substream_type == 'inventories':
4131
if src_serializer == to_serializer:
4132
self.target_repo.inventories.insert_record_stream(
4135
self._extract_and_insert_inventories(
4136
substream, src_serializer)
4137
elif substream_type == 'inventory-deltas':
4138
self._extract_and_insert_inventory_deltas(
4139
substream, src_serializer)
4140
elif substream_type == 'chk_bytes':
4141
# XXX: This doesn't support conversions, as it assumes the
4142
# conversion was done in the fetch code.
4143
self.target_repo.chk_bytes.insert_record_stream(substream)
4144
elif substream_type == 'revisions':
4145
# This may fallback to extract-and-insert more often than
4146
# required if the serializers are different only in terms of
4148
if src_serializer == to_serializer:
4149
self.target_repo.revisions.insert_record_stream(substream)
4151
self._extract_and_insert_revisions(substream,
4153
elif substream_type == 'signatures':
4154
self.target_repo.signatures.insert_record_stream(substream)
4156
raise AssertionError('kaboom! %s' % (substream_type,))
4157
# Done inserting data, and the missing_keys calculations will try to
4158
# read back from the inserted data, so flush the writes to the new pack
4159
# (if this is pack format).
4160
if new_pack is not None:
4161
new_pack._write_data('', flush=True)
4162
# Find all the new revisions (including ones from resume_tokens)
4163
missing_keys = self.target_repo.get_missing_parent_inventories(
4164
check_for_missing_texts=is_resume)
4166
for prefix, versioned_file in (
4167
('texts', self.target_repo.texts),
4168
('inventories', self.target_repo.inventories),
4169
('revisions', self.target_repo.revisions),
4170
('signatures', self.target_repo.signatures),
4171
('chk_bytes', self.target_repo.chk_bytes),
4173
if versioned_file is None:
4175
# TODO: key is often going to be a StaticTuple object
4176
# I don't believe we can define a method by which
4177
# (prefix,) + StaticTuple will work, though we could
4178
# define a StaticTuple.sq_concat that would allow you to
4179
# pass in either a tuple or a StaticTuple as the second
4180
# object, so instead we could have:
4181
# StaticTuple(prefix) + key here...
4182
missing_keys.update((prefix,) + key for key in
4183
versioned_file.get_missing_compression_parent_keys())
4184
except NotImplementedError:
4185
# cannot even attempt suspending, and missing would have failed
4186
# during stream insertion.
4187
missing_keys = set()
4190
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4191
target_rich_root = self.target_repo._format.rich_root_data
4192
target_tree_refs = self.target_repo._format.supports_tree_reference
4193
for record in substream:
4194
# Insert the delta directly
4195
inventory_delta_bytes = record.get_bytes_as('fulltext')
4196
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4198
parse_result = deserialiser.parse_text_bytes(
4199
inventory_delta_bytes)
4200
except inventory_delta.IncompatibleInventoryDelta, err:
4201
mutter("Incompatible delta: %s", err.msg)
4202
raise errors.IncompatibleRevision(self.target_repo._format)
4203
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4204
revision_id = new_id
4205
parents = [key[0] for key in record.parents]
4206
self.target_repo.add_inventory_by_delta(
4207
basis_id, inv_delta, revision_id, parents)
4209
def _extract_and_insert_inventories(self, substream, serializer,
4211
"""Generate a new inventory versionedfile in target, converting data.
4213
The inventory is retrieved from the source, (deserializing it), and
4214
stored in the target (reserializing it in a different format).
4216
target_rich_root = self.target_repo._format.rich_root_data
4217
target_tree_refs = self.target_repo._format.supports_tree_reference
4218
for record in substream:
4219
# It's not a delta, so it must be a fulltext in the source
4220
# serializer's format.
4221
bytes = record.get_bytes_as('fulltext')
4222
revision_id = record.key[0]
4223
inv = serializer.read_inventory_from_string(bytes, revision_id)
4224
parents = [key[0] for key in record.parents]
4225
self.target_repo.add_inventory(revision_id, inv, parents)
4226
# No need to keep holding this full inv in memory when the rest of
4227
# the substream is likely to be all deltas.
4230
def _extract_and_insert_revisions(self, substream, serializer):
4231
for record in substream:
4232
bytes = record.get_bytes_as('fulltext')
4233
revision_id = record.key[0]
4234
rev = serializer.read_revision_from_string(bytes)
4235
if rev.revision_id != revision_id:
4236
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4237
self.target_repo.add_revision(revision_id, rev)
4240
if self.target_repo._format._fetch_reconcile:
4241
self.target_repo.reconcile()
4244
class StreamSource(object):
4245
"""A source of a stream for fetching between repositories."""
4247
def __init__(self, from_repository, to_format):
4248
"""Create a StreamSource streaming from from_repository."""
4249
self.from_repository = from_repository
4250
self.to_format = to_format
4251
self._record_counter = RecordCounter()
4253
def delta_on_metadata(self):
4254
"""Return True if delta's are permitted on metadata streams.
4256
That is on revisions and signatures.
4258
src_serializer = self.from_repository._format._serializer
4259
target_serializer = self.to_format._serializer
4260
return (self.to_format._fetch_uses_deltas and
4261
src_serializer == target_serializer)
4263
def _fetch_revision_texts(self, revs):
4264
# fetch signatures first and then the revision texts
4265
# may need to be a InterRevisionStore call here.
4266
from_sf = self.from_repository.signatures
4267
# A missing signature is just skipped.
4268
keys = [(rev_id,) for rev_id in revs]
4269
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4271
self.to_format._fetch_order,
4272
not self.to_format._fetch_uses_deltas))
4273
# If a revision has a delta, this is actually expanded inside the
4274
# insert_record_stream code now, which is an alternate fix for
4276
from_rf = self.from_repository.revisions
4277
revisions = from_rf.get_record_stream(
4279
self.to_format._fetch_order,
4280
not self.delta_on_metadata())
4281
return [('signatures', signatures), ('revisions', revisions)]
4283
def _generate_root_texts(self, revs):
4284
"""This will be called by get_stream between fetching weave texts and
4285
fetching the inventory weave.
4287
if self._rich_root_upgrade():
4288
return _mod_fetch.Inter1and2Helper(
4289
self.from_repository).generate_root_texts(revs)
4293
def get_stream(self, search):
4295
revs = search.get_keys()
4296
graph = self.from_repository.get_graph()
4297
revs = tsort.topo_sort(graph.get_parent_map(revs))
4298
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4300
for knit_kind, file_id, revisions in data_to_fetch:
4301
if knit_kind != phase:
4303
# Make a new progress bar for this phase
4304
if knit_kind == "file":
4305
# Accumulate file texts
4306
text_keys.extend([(file_id, revision) for revision in
4308
elif knit_kind == "inventory":
4309
# Now copy the file texts.
4310
from_texts = self.from_repository.texts
4311
yield ('texts', from_texts.get_record_stream(
4312
text_keys, self.to_format._fetch_order,
4313
not self.to_format._fetch_uses_deltas))
4314
# Cause an error if a text occurs after we have done the
4317
# Before we process the inventory we generate the root
4318
# texts (if necessary) so that the inventories references
4320
for _ in self._generate_root_texts(revs):
4322
# we fetch only the referenced inventories because we do not
4323
# know for unselected inventories whether all their required
4324
# texts are present in the other repository - it could be
4326
for info in self._get_inventory_stream(revs):
4328
elif knit_kind == "signatures":
4329
# Nothing to do here; this will be taken care of when
4330
# _fetch_revision_texts happens.
4332
elif knit_kind == "revisions":
4333
for record in self._fetch_revision_texts(revs):
4336
raise AssertionError("Unknown knit kind %r" % knit_kind)
4338
def get_stream_for_missing_keys(self, missing_keys):
4339
# missing keys can only occur when we are byte copying and not
4340
# translating (because translation means we don't send
4341
# unreconstructable deltas ever).
4343
keys['texts'] = set()
4344
keys['revisions'] = set()
4345
keys['inventories'] = set()
4346
keys['chk_bytes'] = set()
4347
keys['signatures'] = set()
4348
for key in missing_keys:
4349
keys[key[0]].add(key[1:])
4350
if len(keys['revisions']):
4351
# If we allowed copying revisions at this point, we could end up
4352
# copying a revision without copying its required texts: a
4353
# violation of the requirements for repository integrity.
4354
raise AssertionError(
4355
'cannot copy revisions to fill in missing deltas %s' % (
4356
keys['revisions'],))
4357
for substream_kind, keys in keys.iteritems():
4358
vf = getattr(self.from_repository, substream_kind)
4359
if vf is None and keys:
4360
raise AssertionError(
4361
"cannot fill in keys for a versioned file we don't"
4362
" have: %s needs %s" % (substream_kind, keys))
4364
# No need to stream something we don't have
4366
if substream_kind == 'inventories':
4367
# Some missing keys are genuinely ghosts, filter those out.
4368
present = self.from_repository.inventories.get_parent_map(keys)
4369
revs = [key[0] for key in present]
4370
# Get the inventory stream more-or-less as we do for the
4371
# original stream; there's no reason to assume that records
4372
# direct from the source will be suitable for the sink. (Think
4373
# e.g. 2a -> 1.9-rich-root).
4374
for info in self._get_inventory_stream(revs, missing=True):
4378
# Ask for full texts always so that we don't need more round trips
4379
# after this stream.
4380
# Some of the missing keys are genuinely ghosts, so filter absent
4381
# records. The Sink is responsible for doing another check to
4382
# ensure that ghosts don't introduce missing data for future
4384
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4385
self.to_format._fetch_order, True))
4386
yield substream_kind, stream
4388
def inventory_fetch_order(self):
4389
if self._rich_root_upgrade():
4390
return 'topological'
4392
return self.to_format._fetch_order
4394
def _rich_root_upgrade(self):
4395
return (not self.from_repository._format.rich_root_data and
4396
self.to_format.rich_root_data)
4398
def _get_inventory_stream(self, revision_ids, missing=False):
4399
from_format = self.from_repository._format
4400
if (from_format.supports_chks and self.to_format.supports_chks and
4401
from_format.network_name() == self.to_format.network_name()):
4402
raise AssertionError(
4403
"this case should be handled by GroupCHKStreamSource")
4404
elif 'forceinvdeltas' in debug.debug_flags:
4405
return self._get_convertable_inventory_stream(revision_ids,
4406
delta_versus_null=missing)
4407
elif from_format.network_name() == self.to_format.network_name():
4409
return self._get_simple_inventory_stream(revision_ids,
4411
elif (not from_format.supports_chks and not self.to_format.supports_chks
4412
and from_format._serializer == self.to_format._serializer):
4413
# Essentially the same format.
4414
return self._get_simple_inventory_stream(revision_ids,
4417
# Any time we switch serializations, we want to use an
4418
# inventory-delta based approach.
4419
return self._get_convertable_inventory_stream(revision_ids,
4420
delta_versus_null=missing)
4422
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4423
# NB: This currently reopens the inventory weave in source;
4424
# using a single stream interface instead would avoid this.
4425
from_weave = self.from_repository.inventories
4427
delta_closure = True
4429
delta_closure = not self.delta_on_metadata()
4430
yield ('inventories', from_weave.get_record_stream(
4431
[(rev_id,) for rev_id in revision_ids],
4432
self.inventory_fetch_order(), delta_closure))
4434
def _get_convertable_inventory_stream(self, revision_ids,
4435
delta_versus_null=False):
4436
# The two formats are sufficiently different that there is no fast
4437
# path, so we need to send just inventorydeltas, which any
4438
# sufficiently modern client can insert into any repository.
4439
# The StreamSink code expects to be able to
4440
# convert on the target, so we need to put bytes-on-the-wire that can
4441
# be converted. That means inventory deltas (if the remote is <1.19,
4442
# RemoteStreamSink will fallback to VFS to insert the deltas).
4443
yield ('inventory-deltas',
4444
self._stream_invs_as_deltas(revision_ids,
4445
delta_versus_null=delta_versus_null))
4447
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4448
"""Return a stream of inventory-deltas for the given rev ids.
4450
:param revision_ids: The list of inventories to transmit
4451
:param delta_versus_null: Don't try to find a minimal delta for this
4452
entry, instead compute the delta versus the NULL_REVISION. This
4453
effectively streams a complete inventory. Used for stuff like
4454
filling in missing parents, etc.
4456
from_repo = self.from_repository
4457
revision_keys = [(rev_id,) for rev_id in revision_ids]
4458
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4459
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4461
inventories = self.from_repository.iter_inventories(
4462
revision_ids, 'topological')
4463
format = from_repo._format
4464
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4465
inventory_cache = lru_cache.LRUCache(50)
4466
null_inventory = from_repo.revision_tree(
4467
_mod_revision.NULL_REVISION).inventory
4468
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4469
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4470
# repo back into a non-rich-root repo ought to be allowed)
4471
serializer = inventory_delta.InventoryDeltaSerializer(
4472
versioned_root=format.rich_root_data,
4473
tree_references=format.supports_tree_reference)
4474
for inv in inventories:
4475
key = (inv.revision_id,)
4476
parent_keys = parent_map.get(key, ())
4478
if not delta_versus_null and parent_keys:
4479
# The caller did not ask for complete inventories and we have
4480
# some parents that we can delta against. Make a delta against
4481
# each parent so that we can find the smallest.
4482
parent_ids = [parent_key[0] for parent_key in parent_keys]
4483
for parent_id in parent_ids:
4484
if parent_id not in invs_sent_so_far:
4485
# We don't know that the remote side has this basis, so
4488
if parent_id == _mod_revision.NULL_REVISION:
4489
parent_inv = null_inventory
4491
parent_inv = inventory_cache.get(parent_id, None)
4492
if parent_inv is None:
4493
parent_inv = from_repo.get_inventory(parent_id)
4494
candidate_delta = inv._make_delta(parent_inv)
4495
if (delta is None or
4496
len(delta) > len(candidate_delta)):
4497
delta = candidate_delta
4498
basis_id = parent_id
4500
# Either none of the parents ended up being suitable, or we
4501
# were asked to delta against NULL
4502
basis_id = _mod_revision.NULL_REVISION
4503
delta = inv._make_delta(null_inventory)
4504
invs_sent_so_far.add(inv.revision_id)
4505
inventory_cache[inv.revision_id] = inv
4506
delta_serialized = ''.join(
4507
serializer.delta_to_lines(basis_id, key[-1], delta))
4508
yield versionedfile.FulltextContentFactory(
4509
key, parent_keys, None, delta_serialized)
4512
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4513
stop_revision=None):
4514
"""Extend the partial history to include a given index
4516
If a stop_index is supplied, stop when that index has been reached.
4517
If a stop_revision is supplied, stop when that revision is
4518
encountered. Otherwise, stop when the beginning of history is
4521
:param stop_index: The index which should be present. When it is
4522
present, history extension will stop.
4523
:param stop_revision: The revision id which should be present. When
4524
it is encountered, history extension will stop.
4526
start_revision = partial_history_cache[-1]
4527
iterator = repo.iter_reverse_revision_history(start_revision)
4529
#skip the last revision in the list
4532
if (stop_index is not None and
4533
len(partial_history_cache) > stop_index):
4535
if partial_history_cache[-1] == stop_revision:
4537
revision_id = iterator.next()
4538
partial_history_cache.append(revision_id)
4539
except StopIteration:
added
removed
bzrlib/repository.py
