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(), """
42
revision as _mod_revision,
48
from bzrlib.bundle import serializer
49
from bzrlib.recordcounter import RecordCounter
50
from bzrlib.revisiontree import RevisionTree
51
from bzrlib.store.versioned import VersionedFileStore
52
from bzrlib.testament import Testament
61
from bzrlib.decorators import needs_read_lock, needs_write_lock, only_raises
62
from bzrlib.inter import InterObject
63
from bzrlib.inventory import (
69
from bzrlib.lock import _RelockDebugMixin, LogicalLockResult
70
from bzrlib.trace import (
71
log_exception_quietly, note, mutter, mutter_callsite, warning)
74
# Old formats display a warning, but only once
75
_deprecation_warning_done = False
78
class IsInWriteGroupError(errors.InternalBzrError):
80
_fmt = "May not refresh_data of repo %(repo)s while in a write group."
82
def __init__(self, repo):
83
errors.InternalBzrError.__init__(self, repo=repo)
86
class CommitBuilder(object):
87
"""Provides an interface to build up a commit.
89
This allows describing a tree to be committed without needing to
90
know the internals of the format of the repository.
93
# all clients should supply tree roots.
94
record_root_entry = True
95
# the default CommitBuilder does not manage trees whose root is versioned.
96
_versioned_root = False
97
# this commit builder supports the record_entry_contents interface
98
supports_record_entry_contents = True
100
def __init__(self, repository, parents, config, timestamp=None,
101
timezone=None, committer=None, revprops=None,
103
"""Initiate a CommitBuilder.
105
:param repository: Repository to commit to.
106
:param parents: Revision ids of the parents of the new revision.
107
:param config: Configuration to use.
108
:param timestamp: Optional timestamp recorded for commit.
109
:param timezone: Optional timezone for timestamp.
110
:param committer: Optional committer to set for commit.
111
:param revprops: Optional dictionary of revision properties.
112
:param revision_id: Optional revision id.
114
self._config = config
116
if committer is None:
117
self._committer = self._config.username()
118
elif not isinstance(committer, unicode):
119
self._committer = committer.decode() # throw if non-ascii
121
self._committer = committer
123
self.new_inventory = Inventory(None)
124
self._new_revision_id = revision_id
125
self.parents = parents
126
self.repository = repository
129
if revprops is not None:
130
self._validate_revprops(revprops)
131
self._revprops.update(revprops)
133
if timestamp is None:
134
timestamp = time.time()
135
# Restrict resolution to 1ms
136
self._timestamp = round(timestamp, 3)
139
self._timezone = osutils.local_time_offset()
141
self._timezone = int(timezone)
143
self._generate_revision_if_needed()
144
self.__heads = graph.HeadsCache(repository.get_graph()).heads
145
self._basis_delta = []
146
# API compatibility, older code that used CommitBuilder did not call
147
# .record_delete(), which means the delta that is computed would not be
148
# valid. Callers that will call record_delete() should call
149
# .will_record_deletes() to indicate that.
150
self._recording_deletes = False
151
# memo'd check for no-op commits.
152
self._any_changes = False
154
def any_changes(self):
155
"""Return True if any entries were changed.
157
This includes merge-only changes. It is the core for the --unchanged
160
:return: True if any changes have occured.
162
return self._any_changes
164
def _validate_unicode_text(self, text, context):
165
"""Verify things like commit messages don't have bogus characters."""
167
raise ValueError('Invalid value for %s: %r' % (context, text))
169
def _validate_revprops(self, revprops):
170
for key, value in revprops.iteritems():
171
# We know that the XML serializers do not round trip '\r'
172
# correctly, so refuse to accept them
173
if not isinstance(value, basestring):
174
raise ValueError('revision property (%s) is not a valid'
175
' (unicode) string: %r' % (key, value))
176
self._validate_unicode_text(value,
177
'revision property (%s)' % (key,))
179
def _ensure_fallback_inventories(self):
180
"""Ensure that appropriate inventories are available.
182
This only applies to repositories that are stacked, and is about
183
enusring the stacking invariants. Namely, that for any revision that is
184
present, we either have all of the file content, or we have the parent
185
inventory and the delta file content.
187
if not self.repository._fallback_repositories:
189
if not self.repository._format.supports_chks:
190
raise errors.BzrError("Cannot commit directly to a stacked branch"
191
" in pre-2a formats. See "
192
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
193
# This is a stacked repo, we need to make sure we have the parent
194
# inventories for the parents.
195
parent_keys = [(p,) for p in self.parents]
196
parent_map = self.repository.inventories._index.get_parent_map(parent_keys)
197
missing_parent_keys = set([pk for pk in parent_keys
198
if pk not in parent_map])
199
fallback_repos = list(reversed(self.repository._fallback_repositories))
200
missing_keys = [('inventories', pk[0])
201
for pk in missing_parent_keys]
203
while missing_keys and fallback_repos:
204
fallback_repo = fallback_repos.pop()
205
source = fallback_repo._get_source(self.repository._format)
206
sink = self.repository._get_sink()
207
stream = source.get_stream_for_missing_keys(missing_keys)
208
missing_keys = sink.insert_stream_without_locking(stream,
209
self.repository._format)
211
raise errors.BzrError('Unable to fill in parent inventories for a'
214
def commit(self, message):
215
"""Make the actual commit.
217
:return: The revision id of the recorded revision.
219
self._validate_unicode_text(message, 'commit message')
220
rev = _mod_revision.Revision(
221
timestamp=self._timestamp,
222
timezone=self._timezone,
223
committer=self._committer,
225
inventory_sha1=self.inv_sha1,
226
revision_id=self._new_revision_id,
227
properties=self._revprops)
228
rev.parent_ids = self.parents
229
self.repository.add_revision(self._new_revision_id, rev,
230
self.new_inventory, self._config)
231
self._ensure_fallback_inventories()
232
self.repository.commit_write_group()
233
return self._new_revision_id
236
"""Abort the commit that is being built.
238
self.repository.abort_write_group()
240
def revision_tree(self):
241
"""Return the tree that was just committed.
243
After calling commit() this can be called to get a RevisionTree
244
representing the newly committed tree. This is preferred to
245
calling Repository.revision_tree() because that may require
246
deserializing the inventory, while we already have a copy in
249
if self.new_inventory is None:
250
self.new_inventory = self.repository.get_inventory(
251
self._new_revision_id)
252
return RevisionTree(self.repository, self.new_inventory,
253
self._new_revision_id)
255
def finish_inventory(self):
256
"""Tell the builder that the inventory is finished.
258
:return: The inventory id in the repository, which can be used with
259
repository.get_inventory.
261
if self.new_inventory is None:
262
# an inventory delta was accumulated without creating a new
264
basis_id = self.basis_delta_revision
265
# We ignore the 'inventory' returned by add_inventory_by_delta
266
# because self.new_inventory is used to hint to the rest of the
267
# system what code path was taken
268
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
269
basis_id, self._basis_delta, self._new_revision_id,
272
if self.new_inventory.root is None:
273
raise AssertionError('Root entry should be supplied to'
274
' record_entry_contents, as of bzr 0.10.')
275
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
276
self.new_inventory.revision_id = self._new_revision_id
277
self.inv_sha1 = self.repository.add_inventory(
278
self._new_revision_id,
282
return self._new_revision_id
284
def _gen_revision_id(self):
285
"""Return new revision-id."""
286
return generate_ids.gen_revision_id(self._committer, self._timestamp)
288
def _generate_revision_if_needed(self):
289
"""Create a revision id if None was supplied.
291
If the repository can not support user-specified revision ids
292
they should override this function and raise CannotSetRevisionId
293
if _new_revision_id is not None.
295
:raises: CannotSetRevisionId
297
if self._new_revision_id is None:
298
self._new_revision_id = self._gen_revision_id()
299
self.random_revid = True
301
self.random_revid = False
303
def _heads(self, file_id, revision_ids):
304
"""Calculate the graph heads for revision_ids in the graph of file_id.
306
This can use either a per-file graph or a global revision graph as we
307
have an identity relationship between the two graphs.
309
return self.__heads(revision_ids)
311
def _check_root(self, ie, parent_invs, tree):
312
"""Helper for record_entry_contents.
314
:param ie: An entry being added.
315
:param parent_invs: The inventories of the parent revisions of the
317
:param tree: The tree that is being committed.
319
# In this revision format, root entries have no knit or weave When
320
# serializing out to disk and back in root.revision is always
322
ie.revision = self._new_revision_id
324
def _require_root_change(self, tree):
325
"""Enforce an appropriate root object change.
327
This is called once when record_iter_changes is called, if and only if
328
the root was not in the delta calculated by record_iter_changes.
330
:param tree: The tree which is being committed.
332
if len(self.parents) == 0:
333
raise errors.RootMissing()
334
entry = entry_factory['directory'](tree.path2id(''), '',
336
entry.revision = self._new_revision_id
337
self._basis_delta.append(('', '', entry.file_id, entry))
339
def _get_delta(self, ie, basis_inv, path):
340
"""Get a delta against the basis inventory for ie."""
341
if ie.file_id not in basis_inv:
343
result = (None, path, ie.file_id, ie)
344
self._basis_delta.append(result)
346
elif ie != basis_inv[ie.file_id]:
348
# TODO: avoid tis id2path call.
349
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
350
self._basis_delta.append(result)
356
def get_basis_delta(self):
357
"""Return the complete inventory delta versus the basis inventory.
359
This has been built up with the calls to record_delete and
360
record_entry_contents. The client must have already called
361
will_record_deletes() to indicate that they will be generating a
364
:return: An inventory delta, suitable for use with apply_delta, or
365
Repository.add_inventory_by_delta, etc.
367
if not self._recording_deletes:
368
raise AssertionError("recording deletes not activated.")
369
return self._basis_delta
371
def record_delete(self, path, file_id):
372
"""Record that a delete occured against a basis tree.
374
This is an optional API - when used it adds items to the basis_delta
375
being accumulated by the commit builder. It cannot be called unless the
376
method will_record_deletes() has been called to inform the builder that
377
a delta is being supplied.
379
:param path: The path of the thing deleted.
380
:param file_id: The file id that was deleted.
382
if not self._recording_deletes:
383
raise AssertionError("recording deletes not activated.")
384
delta = (path, None, file_id, None)
385
self._basis_delta.append(delta)
386
self._any_changes = True
389
def will_record_deletes(self):
390
"""Tell the commit builder that deletes are being notified.
392
This enables the accumulation of an inventory delta; for the resulting
393
commit to be valid, deletes against the basis MUST be recorded via
394
builder.record_delete().
396
self._recording_deletes = True
398
basis_id = self.parents[0]
400
basis_id = _mod_revision.NULL_REVISION
401
self.basis_delta_revision = basis_id
403
def record_entry_contents(self, ie, parent_invs, path, tree,
405
"""Record the content of ie from tree into the commit if needed.
407
Side effect: sets ie.revision when unchanged
409
:param ie: An inventory entry present in the commit.
410
:param parent_invs: The inventories of the parent revisions of the
412
:param path: The path the entry is at in the tree.
413
:param tree: The tree which contains this entry and should be used to
415
:param content_summary: Summary data from the tree about the paths
416
content - stat, length, exec, sha/link target. This is only
417
accessed when the entry has a revision of None - that is when it is
418
a candidate to commit.
419
:return: A tuple (change_delta, version_recorded, fs_hash).
420
change_delta is an inventory_delta change for this entry against
421
the basis tree of the commit, or None if no change occured against
423
version_recorded is True if a new version of the entry has been
424
recorded. For instance, committing a merge where a file was only
425
changed on the other side will return (delta, False).
426
fs_hash is either None, or the hash details for the path (currently
427
a tuple of the contents sha1 and the statvalue returned by
428
tree.get_file_with_stat()).
430
if self.new_inventory.root is None:
431
if ie.parent_id is not None:
432
raise errors.RootMissing()
433
self._check_root(ie, parent_invs, tree)
434
if ie.revision is None:
435
kind = content_summary[0]
437
# ie is carried over from a prior commit
439
# XXX: repository specific check for nested tree support goes here - if
440
# the repo doesn't want nested trees we skip it ?
441
if (kind == 'tree-reference' and
442
not self.repository._format.supports_tree_reference):
443
# mismatch between commit builder logic and repository:
444
# this needs the entry creation pushed down into the builder.
445
raise NotImplementedError('Missing repository subtree support.')
446
self.new_inventory.add(ie)
448
# TODO: slow, take it out of the inner loop.
450
basis_inv = parent_invs[0]
452
basis_inv = Inventory(root_id=None)
454
# ie.revision is always None if the InventoryEntry is considered
455
# for committing. We may record the previous parents revision if the
456
# content is actually unchanged against a sole head.
457
if ie.revision is not None:
458
if not self._versioned_root and path == '':
459
# repositories that do not version the root set the root's
460
# revision to the new commit even when no change occurs (more
461
# specifically, they do not record a revision on the root; and
462
# the rev id is assigned to the root during deserialisation -
463
# this masks when a change may have occurred against the basis.
464
# To match this we always issue a delta, because the revision
465
# of the root will always be changing.
466
if ie.file_id in basis_inv:
467
delta = (basis_inv.id2path(ie.file_id), path,
471
delta = (None, path, ie.file_id, ie)
472
self._basis_delta.append(delta)
473
return delta, False, None
475
# we don't need to commit this, because the caller already
476
# determined that an existing revision of this file is
477
# appropriate. If it's not being considered for committing then
478
# it and all its parents to the root must be unaltered so
479
# no-change against the basis.
480
if ie.revision == self._new_revision_id:
481
raise AssertionError("Impossible situation, a skipped "
482
"inventory entry (%r) claims to be modified in this "
483
"commit (%r).", (ie, self._new_revision_id))
484
return None, False, None
485
# XXX: Friction: parent_candidates should return a list not a dict
486
# so that we don't have to walk the inventories again.
487
parent_candiate_entries = ie.parent_candidates(parent_invs)
488
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
490
for inv in parent_invs:
491
if ie.file_id in inv:
492
old_rev = inv[ie.file_id].revision
493
if old_rev in head_set:
494
heads.append(inv[ie.file_id].revision)
495
head_set.remove(inv[ie.file_id].revision)
498
# now we check to see if we need to write a new record to the
500
# We write a new entry unless there is one head to the ancestors, and
501
# the kind-derived content is unchanged.
503
# Cheapest check first: no ancestors, or more the one head in the
504
# ancestors, we write a new node.
508
# There is a single head, look it up for comparison
509
parent_entry = parent_candiate_entries[heads[0]]
510
# if the non-content specific data has changed, we'll be writing a
512
if (parent_entry.parent_id != ie.parent_id or
513
parent_entry.name != ie.name):
515
# now we need to do content specific checks:
517
# if the kind changed the content obviously has
518
if kind != parent_entry.kind:
520
# Stat cache fingerprint feedback for the caller - None as we usually
521
# don't generate one.
524
if content_summary[2] is None:
525
raise ValueError("Files must not have executable = None")
527
# We can't trust a check of the file length because of content
529
if (# if the exec bit has changed we have to store:
530
parent_entry.executable != content_summary[2]):
532
elif parent_entry.text_sha1 == content_summary[3]:
533
# all meta and content is unchanged (using a hash cache
534
# hit to check the sha)
535
ie.revision = parent_entry.revision
536
ie.text_size = parent_entry.text_size
537
ie.text_sha1 = parent_entry.text_sha1
538
ie.executable = parent_entry.executable
539
return self._get_delta(ie, basis_inv, path), False, None
541
# Either there is only a hash change(no hash cache entry,
542
# or same size content change), or there is no change on
544
# Provide the parent's hash to the store layer, so that the
545
# content is unchanged we will not store a new node.
546
nostore_sha = parent_entry.text_sha1
548
# We want to record a new node regardless of the presence or
549
# absence of a content change in the file.
551
ie.executable = content_summary[2]
552
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
554
text = file_obj.read()
558
ie.text_sha1, ie.text_size = self._add_text_to_weave(
559
ie.file_id, text, heads, nostore_sha)
560
# Let the caller know we generated a stat fingerprint.
561
fingerprint = (ie.text_sha1, stat_value)
562
except errors.ExistingContent:
563
# Turns out that the file content was unchanged, and we were
564
# only going to store a new node if it was changed. Carry over
566
ie.revision = parent_entry.revision
567
ie.text_size = parent_entry.text_size
568
ie.text_sha1 = parent_entry.text_sha1
569
ie.executable = parent_entry.executable
570
return self._get_delta(ie, basis_inv, path), False, None
571
elif kind == 'directory':
573
# all data is meta here, nothing specific to directory, so
575
ie.revision = parent_entry.revision
576
return self._get_delta(ie, basis_inv, path), False, None
577
self._add_text_to_weave(ie.file_id, '', heads, None)
578
elif kind == 'symlink':
579
current_link_target = content_summary[3]
581
# symlink target is not generic metadata, check if it has
583
if current_link_target != parent_entry.symlink_target:
586
# unchanged, carry over.
587
ie.revision = parent_entry.revision
588
ie.symlink_target = parent_entry.symlink_target
589
return self._get_delta(ie, basis_inv, path), False, None
590
ie.symlink_target = current_link_target
591
self._add_text_to_weave(ie.file_id, '', heads, None)
592
elif kind == 'tree-reference':
594
if content_summary[3] != parent_entry.reference_revision:
597
# unchanged, carry over.
598
ie.reference_revision = parent_entry.reference_revision
599
ie.revision = parent_entry.revision
600
return self._get_delta(ie, basis_inv, path), False, None
601
ie.reference_revision = content_summary[3]
602
if ie.reference_revision is None:
603
raise AssertionError("invalid content_summary for nested tree: %r"
604
% (content_summary,))
605
self._add_text_to_weave(ie.file_id, '', heads, None)
607
raise NotImplementedError('unknown kind')
608
ie.revision = self._new_revision_id
609
self._any_changes = True
610
return self._get_delta(ie, basis_inv, path), True, fingerprint
612
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
613
_entry_factory=entry_factory):
614
"""Record a new tree via iter_changes.
616
:param tree: The tree to obtain text contents from for changed objects.
617
:param basis_revision_id: The revision id of the tree the iter_changes
618
has been generated against. Currently assumed to be the same
619
as self.parents[0] - if it is not, errors may occur.
620
:param iter_changes: An iter_changes iterator with the changes to apply
621
to basis_revision_id. The iterator must not include any items with
622
a current kind of None - missing items must be either filtered out
623
or errored-on beefore record_iter_changes sees the item.
624
:param _entry_factory: Private method to bind entry_factory locally for
626
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
629
# Create an inventory delta based on deltas between all the parents and
630
# deltas between all the parent inventories. We use inventory delta's
631
# between the inventory objects because iter_changes masks
632
# last-changed-field only changes.
634
# file_id -> change map, change is fileid, paths, changed, versioneds,
635
# parents, names, kinds, executables
637
# {file_id -> revision_id -> inventory entry, for entries in parent
638
# trees that are not parents[0]
642
revtrees = list(self.repository.revision_trees(self.parents))
643
except errors.NoSuchRevision:
644
# one or more ghosts, slow path.
646
for revision_id in self.parents:
648
revtrees.append(self.repository.revision_tree(revision_id))
649
except errors.NoSuchRevision:
651
basis_revision_id = _mod_revision.NULL_REVISION
653
revtrees.append(self.repository.revision_tree(
654
_mod_revision.NULL_REVISION))
655
# The basis inventory from a repository
657
basis_inv = revtrees[0].inventory
659
basis_inv = self.repository.revision_tree(
660
_mod_revision.NULL_REVISION).inventory
661
if len(self.parents) > 0:
662
if basis_revision_id != self.parents[0] and not ghost_basis:
664
"arbitrary basis parents not yet supported with merges")
665
for revtree in revtrees[1:]:
666
for change in revtree.inventory._make_delta(basis_inv):
667
if change[1] is None:
668
# Not present in this parent.
670
if change[2] not in merged_ids:
671
if change[0] is not None:
672
basis_entry = basis_inv[change[2]]
673
merged_ids[change[2]] = [
675
basis_entry.revision,
678
parent_entries[change[2]] = {
680
basis_entry.revision:basis_entry,
682
change[3].revision:change[3],
685
merged_ids[change[2]] = [change[3].revision]
686
parent_entries[change[2]] = {change[3].revision:change[3]}
688
merged_ids[change[2]].append(change[3].revision)
689
parent_entries[change[2]][change[3].revision] = change[3]
692
# Setup the changes from the tree:
693
# changes maps file_id -> (change, [parent revision_ids])
695
for change in iter_changes:
696
# This probably looks up in basis_inv way to much.
697
if change[1][0] is not None:
698
head_candidate = [basis_inv[change[0]].revision]
701
changes[change[0]] = change, merged_ids.get(change[0],
703
unchanged_merged = set(merged_ids) - set(changes)
704
# Extend the changes dict with synthetic changes to record merges of
706
for file_id in unchanged_merged:
707
# Record a merged version of these items that did not change vs the
708
# basis. This can be either identical parallel changes, or a revert
709
# of a specific file after a merge. The recorded content will be
710
# that of the current tree (which is the same as the basis), but
711
# the per-file graph will reflect a merge.
712
# NB:XXX: We are reconstructing path information we had, this
713
# should be preserved instead.
714
# inv delta change: (file_id, (path_in_source, path_in_target),
715
# changed_content, versioned, parent, name, kind,
718
basis_entry = basis_inv[file_id]
719
except errors.NoSuchId:
720
# a change from basis->some_parents but file_id isn't in basis
721
# so was new in the merge, which means it must have changed
722
# from basis -> current, and as it hasn't the add was reverted
723
# by the user. So we discard this change.
727
(basis_inv.id2path(file_id), tree.id2path(file_id)),
729
(basis_entry.parent_id, basis_entry.parent_id),
730
(basis_entry.name, basis_entry.name),
731
(basis_entry.kind, basis_entry.kind),
732
(basis_entry.executable, basis_entry.executable))
733
changes[file_id] = (change, merged_ids[file_id])
734
# changes contains tuples with the change and a set of inventory
735
# candidates for the file.
737
# old_path, new_path, file_id, new_inventory_entry
738
seen_root = False # Is the root in the basis delta?
739
inv_delta = self._basis_delta
740
modified_rev = self._new_revision_id
741
for change, head_candidates in changes.values():
742
if change[3][1]: # versioned in target.
743
# Several things may be happening here:
744
# We may have a fork in the per-file graph
745
# - record a change with the content from tree
746
# We may have a change against < all trees
747
# - carry over the tree that hasn't changed
748
# We may have a change against all trees
749
# - record the change with the content from tree
752
entry = _entry_factory[kind](file_id, change[5][1],
754
head_set = self._heads(change[0], set(head_candidates))
757
for head_candidate in head_candidates:
758
if head_candidate in head_set:
759
heads.append(head_candidate)
760
head_set.remove(head_candidate)
763
# Could be a carry-over situation:
764
parent_entry_revs = parent_entries.get(file_id, None)
765
if parent_entry_revs:
766
parent_entry = parent_entry_revs.get(heads[0], None)
769
if parent_entry is None:
770
# The parent iter_changes was called against is the one
771
# that is the per-file head, so any change is relevant
772
# iter_changes is valid.
773
carry_over_possible = False
775
# could be a carry over situation
776
# A change against the basis may just indicate a merge,
777
# we need to check the content against the source of the
778
# merge to determine if it was changed after the merge
780
if (parent_entry.kind != entry.kind or
781
parent_entry.parent_id != entry.parent_id or
782
parent_entry.name != entry.name):
783
# Metadata common to all entries has changed
784
# against per-file parent
785
carry_over_possible = False
787
carry_over_possible = True
788
# per-type checks for changes against the parent_entry
791
# Cannot be a carry-over situation
792
carry_over_possible = False
793
# Populate the entry in the delta
795
# XXX: There is still a small race here: If someone reverts the content of a file
796
# after iter_changes examines and decides it has changed,
797
# we will unconditionally record a new version even if some
798
# other process reverts it while commit is running (with
799
# the revert happening after iter_changes did its
802
entry.executable = True
804
entry.executable = False
805
if (carry_over_possible and
806
parent_entry.executable == entry.executable):
807
# Check the file length, content hash after reading
809
nostore_sha = parent_entry.text_sha1
812
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
814
text = file_obj.read()
818
entry.text_sha1, entry.text_size = self._add_text_to_weave(
819
file_id, text, heads, nostore_sha)
820
yield file_id, change[1][1], (entry.text_sha1, stat_value)
821
except errors.ExistingContent:
822
# No content change against a carry_over parent
823
# Perhaps this should also yield a fs hash update?
825
entry.text_size = parent_entry.text_size
826
entry.text_sha1 = parent_entry.text_sha1
827
elif kind == 'symlink':
829
entry.symlink_target = tree.get_symlink_target(file_id)
830
if (carry_over_possible and
831
parent_entry.symlink_target == entry.symlink_target):
834
self._add_text_to_weave(change[0], '', heads, None)
835
elif kind == 'directory':
836
if carry_over_possible:
839
# Nothing to set on the entry.
840
# XXX: split into the Root and nonRoot versions.
841
if change[1][1] != '' or self.repository.supports_rich_root():
842
self._add_text_to_weave(change[0], '', heads, None)
843
elif kind == 'tree-reference':
844
if not self.repository._format.supports_tree_reference:
845
# This isn't quite sane as an error, but we shouldn't
846
# ever see this code path in practice: tree's don't
847
# permit references when the repo doesn't support tree
849
raise errors.UnsupportedOperation(tree.add_reference,
851
reference_revision = tree.get_reference_revision(change[0])
852
entry.reference_revision = reference_revision
853
if (carry_over_possible and
854
parent_entry.reference_revision == reference_revision):
857
self._add_text_to_weave(change[0], '', heads, None)
859
raise AssertionError('unknown kind %r' % kind)
861
entry.revision = modified_rev
863
entry.revision = parent_entry.revision
866
new_path = change[1][1]
867
inv_delta.append((change[1][0], new_path, change[0], entry))
870
self.new_inventory = None
872
# This should perhaps be guarded by a check that the basis we
873
# commit against is the basis for the commit and if not do a delta
875
self._any_changes = True
877
# housekeeping root entry changes do not affect no-change commits.
878
self._require_root_change(tree)
879
self.basis_delta_revision = basis_revision_id
881
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
882
parent_keys = tuple([(file_id, parent) for parent in parents])
883
return self.repository.texts._add_text(
884
(file_id, self._new_revision_id), parent_keys, new_text,
885
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
888
class RootCommitBuilder(CommitBuilder):
889
"""This commitbuilder actually records the root id"""
891
# the root entry gets versioned properly by this builder.
892
_versioned_root = True
894
def _check_root(self, ie, parent_invs, tree):
895
"""Helper for record_entry_contents.
897
:param ie: An entry being added.
898
:param parent_invs: The inventories of the parent revisions of the
900
:param tree: The tree that is being committed.
903
def _require_root_change(self, tree):
904
"""Enforce an appropriate root object change.
906
This is called once when record_iter_changes is called, if and only if
907
the root was not in the delta calculated by record_iter_changes.
909
:param tree: The tree which is being committed.
911
# versioned roots do not change unless the tree found a change.
914
class RepositoryWriteLockResult(LogicalLockResult):
915
"""The result of write locking a repository.
917
:ivar repository_token: The token obtained from the underlying lock, or
919
:ivar unlock: A callable which will unlock the lock.
922
def __init__(self, unlock, repository_token):
923
LogicalLockResult.__init__(self, unlock)
924
self.repository_token = repository_token
927
return "RepositoryWriteLockResult(%s, %s)" % (self.repository_token,
931
######################################################################
935
class Repository(_RelockDebugMixin, controldir.ControlComponent):
936
"""Repository holding history for one or more branches.
938
The repository holds and retrieves historical information including
939
revisions and file history. It's normally accessed only by the Branch,
940
which views a particular line of development through that history.
942
The Repository builds on top of some byte storage facilies (the revisions,
943
signatures, inventories, texts and chk_bytes attributes) and a Transport,
944
which respectively provide byte storage and a means to access the (possibly
947
The byte storage facilities are addressed via tuples, which we refer to
948
as 'keys' throughout the code base. Revision_keys, inventory_keys and
949
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
950
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
951
byte string made up of a hash identifier and a hash value.
952
We use this interface because it allows low friction with the underlying
953
code that implements disk indices, network encoding and other parts of
956
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
957
the serialised revisions for the repository. This can be used to obtain
958
revision graph information or to access raw serialised revisions.
959
The result of trying to insert data into the repository via this store
960
is undefined: it should be considered read-only except for implementors
962
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
963
the serialised signatures for the repository. This can be used to
964
obtain access to raw serialised signatures. The result of trying to
965
insert data into the repository via this store is undefined: it should
966
be considered read-only except for implementors of repositories.
967
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
968
the serialised inventories for the repository. This can be used to
969
obtain unserialised inventories. The result of trying to insert data
970
into the repository via this store is undefined: it should be
971
considered read-only except for implementors of repositories.
972
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
973
texts of files and directories for the repository. This can be used to
974
obtain file texts or file graphs. Note that Repository.iter_file_bytes
975
is usually a better interface for accessing file texts.
976
The result of trying to insert data into the repository via this store
977
is undefined: it should be considered read-only except for implementors
979
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
980
any data the repository chooses to store or have indexed by its hash.
981
The result of trying to insert data into the repository via this store
982
is undefined: it should be considered read-only except for implementors
984
:ivar _transport: Transport for file access to repository, typically
985
pointing to .bzr/repository.
988
# What class to use for a CommitBuilder. Often it's simpler to change this
989
# in a Repository class subclass rather than to override
990
# get_commit_builder.
991
_commit_builder_class = CommitBuilder
993
def abort_write_group(self, suppress_errors=False):
994
"""Commit the contents accrued within the current write group.
996
:param suppress_errors: if true, abort_write_group will catch and log
997
unexpected errors that happen during the abort, rather than
998
allowing them to propagate. Defaults to False.
1000
:seealso: start_write_group.
1002
if self._write_group is not self.get_transaction():
1003
# has an unlock or relock occured ?
1006
'(suppressed) mismatched lock context and write group. %r, %r',
1007
self._write_group, self.get_transaction())
1009
raise errors.BzrError(
1010
'mismatched lock context and write group. %r, %r' %
1011
(self._write_group, self.get_transaction()))
1013
self._abort_write_group()
1014
except Exception, exc:
1015
self._write_group = None
1016
if not suppress_errors:
1018
mutter('abort_write_group failed')
1019
log_exception_quietly()
1020
note('bzr: ERROR (ignored): %s', exc)
1021
self._write_group = None
1023
def _abort_write_group(self):
1024
"""Template method for per-repository write group cleanup.
1026
This is called during abort before the write group is considered to be
1027
finished and should cleanup any internal state accrued during the write
1028
group. There is no requirement that data handed to the repository be
1029
*not* made available - this is not a rollback - but neither should any
1030
attempt be made to ensure that data added is fully commited. Abort is
1031
invoked when an error has occured so futher disk or network operations
1032
may not be possible or may error and if possible should not be
1036
def add_fallback_repository(self, repository):
1037
"""Add a repository to use for looking up data not held locally.
1039
:param repository: A repository.
1041
if not self._format.supports_external_lookups:
1042
raise errors.UnstackableRepositoryFormat(self._format, self.base)
1043
if self.is_locked():
1044
# This repository will call fallback.unlock() when we transition to
1045
# the unlocked state, so we make sure to increment the lock count
1046
repository.lock_read()
1047
self._check_fallback_repository(repository)
1048
self._fallback_repositories.append(repository)
1049
self.texts.add_fallback_versioned_files(repository.texts)
1050
self.inventories.add_fallback_versioned_files(repository.inventories)
1051
self.revisions.add_fallback_versioned_files(repository.revisions)
1052
self.signatures.add_fallback_versioned_files(repository.signatures)
1053
if self.chk_bytes is not None:
1054
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
1056
def _check_fallback_repository(self, repository):
1057
"""Check that this repository can fallback to repository safely.
1059
Raise an error if not.
1061
:param repository: A repository to fallback to.
1063
return InterRepository._assert_same_model(self, repository)
1065
def add_inventory(self, revision_id, inv, parents):
1066
"""Add the inventory inv to the repository as revision_id.
1068
:param parents: The revision ids of the parents that revision_id
1069
is known to have and are in the repository already.
1071
:returns: The validator(which is a sha1 digest, though what is sha'd is
1072
repository format specific) of the serialized inventory.
1074
if not self.is_in_write_group():
1075
raise AssertionError("%r not in write group" % (self,))
1076
_mod_revision.check_not_reserved_id(revision_id)
1077
if not (inv.revision_id is None or inv.revision_id == revision_id):
1078
raise AssertionError(
1079
"Mismatch between inventory revision"
1080
" id and insertion revid (%r, %r)"
1081
% (inv.revision_id, revision_id))
1082
if inv.root is None:
1083
raise errors.RootMissing()
1084
return self._add_inventory_checked(revision_id, inv, parents)
1086
def _add_inventory_checked(self, revision_id, inv, parents):
1087
"""Add inv to the repository after checking the inputs.
1089
This function can be overridden to allow different inventory styles.
1091
:seealso: add_inventory, for the contract.
1093
inv_lines = self._serializer.write_inventory_to_lines(inv)
1094
return self._inventory_add_lines(revision_id, parents,
1095
inv_lines, check_content=False)
1097
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1098
parents, basis_inv=None, propagate_caches=False):
1099
"""Add a new inventory expressed as a delta against another revision.
1101
See the inventory developers documentation for the theory behind
1104
:param basis_revision_id: The inventory id the delta was created
1105
against. (This does not have to be a direct parent.)
1106
:param delta: The inventory delta (see Inventory.apply_delta for
1108
:param new_revision_id: The revision id that the inventory is being
1110
:param parents: The revision ids of the parents that revision_id is
1111
known to have and are in the repository already. These are supplied
1112
for repositories that depend on the inventory graph for revision
1113
graph access, as well as for those that pun ancestry with delta
1115
:param basis_inv: The basis inventory if it is already known,
1117
:param propagate_caches: If True, the caches for this inventory are
1118
copied to and updated for the result if possible.
1120
:returns: (validator, new_inv)
1121
The validator(which is a sha1 digest, though what is sha'd is
1122
repository format specific) of the serialized inventory, and the
1123
resulting inventory.
1125
if not self.is_in_write_group():
1126
raise AssertionError("%r not in write group" % (self,))
1127
_mod_revision.check_not_reserved_id(new_revision_id)
1128
basis_tree = self.revision_tree(basis_revision_id)
1129
basis_tree.lock_read()
1131
# Note that this mutates the inventory of basis_tree, which not all
1132
# inventory implementations may support: A better idiom would be to
1133
# return a new inventory, but as there is no revision tree cache in
1134
# repository this is safe for now - RBC 20081013
1135
if basis_inv is None:
1136
basis_inv = basis_tree.inventory
1137
basis_inv.apply_delta(delta)
1138
basis_inv.revision_id = new_revision_id
1139
return (self.add_inventory(new_revision_id, basis_inv, parents),
1144
def _inventory_add_lines(self, revision_id, parents, lines,
1145
check_content=True):
1146
"""Store lines in inv_vf and return the sha1 of the inventory."""
1147
parents = [(parent,) for parent in parents]
1148
result = self.inventories.add_lines((revision_id,), parents, lines,
1149
check_content=check_content)[0]
1150
self.inventories._access.flush()
1153
def add_revision(self, revision_id, rev, inv=None, config=None):
1154
"""Add rev to the revision store as revision_id.
1156
:param revision_id: the revision id to use.
1157
:param rev: The revision object.
1158
:param inv: The inventory for the revision. if None, it will be looked
1159
up in the inventory storer
1160
:param config: If None no digital signature will be created.
1161
If supplied its signature_needed method will be used
1162
to determine if a signature should be made.
1164
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1166
_mod_revision.check_not_reserved_id(revision_id)
1167
if config is not None and config.signature_needed():
1169
inv = self.get_inventory(revision_id)
1170
plaintext = Testament(rev, inv).as_short_text()
1171
self.store_revision_signature(
1172
gpg.GPGStrategy(config), plaintext, revision_id)
1173
# check inventory present
1174
if not self.inventories.get_parent_map([(revision_id,)]):
1176
raise errors.WeaveRevisionNotPresent(revision_id,
1179
# yes, this is not suitable for adding with ghosts.
1180
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1183
key = (revision_id,)
1184
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1185
self._add_revision(rev)
1187
def _add_revision(self, revision):
1188
text = self._serializer.write_revision_to_string(revision)
1189
key = (revision.revision_id,)
1190
parents = tuple((parent,) for parent in revision.parent_ids)
1191
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1193
def all_revision_ids(self):
1194
"""Returns a list of all the revision ids in the repository.
1196
This is conceptually deprecated because code should generally work on
1197
the graph reachable from a particular revision, and ignore any other
1198
revisions that might be present. There is no direct replacement
1201
if 'evil' in debug.debug_flags:
1202
mutter_callsite(2, "all_revision_ids is linear with history.")
1203
return self._all_revision_ids()
1205
def _all_revision_ids(self):
1206
"""Returns a list of all the revision ids in the repository.
1208
These are in as much topological order as the underlying store can
1211
raise NotImplementedError(self._all_revision_ids)
1213
def break_lock(self):
1214
"""Break a lock if one is present from another instance.
1216
Uses the ui factory to ask for confirmation if the lock may be from
1219
self.control_files.break_lock()
1222
def _eliminate_revisions_not_present(self, revision_ids):
1223
"""Check every revision id in revision_ids to see if we have it.
1225
Returns a set of the present revisions.
1228
graph = self.get_graph()
1229
parent_map = graph.get_parent_map(revision_ids)
1230
# The old API returned a list, should this actually be a set?
1231
return parent_map.keys()
1233
def _check_inventories(self, checker):
1234
"""Check the inventories found from the revision scan.
1236
This is responsible for verifying the sha1 of inventories and
1237
creating a pending_keys set that covers data referenced by inventories.
1239
bar = ui.ui_factory.nested_progress_bar()
1241
self._do_check_inventories(checker, bar)
1245
def _do_check_inventories(self, checker, bar):
1246
"""Helper for _check_inventories."""
1248
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1249
kinds = ['chk_bytes', 'texts']
1250
count = len(checker.pending_keys)
1251
bar.update("inventories", 0, 2)
1252
current_keys = checker.pending_keys
1253
checker.pending_keys = {}
1254
# Accumulate current checks.
1255
for key in current_keys:
1256
if key[0] != 'inventories' and key[0] not in kinds:
1257
checker._report_items.append('unknown key type %r' % (key,))
1258
keys[key[0]].add(key[1:])
1259
if keys['inventories']:
1260
# NB: output order *should* be roughly sorted - topo or
1261
# inverse topo depending on repository - either way decent
1262
# to just delta against. However, pre-CHK formats didn't
1263
# try to optimise inventory layout on disk. As such the
1264
# pre-CHK code path does not use inventory deltas.
1266
for record in self.inventories.check(keys=keys['inventories']):
1267
if record.storage_kind == 'absent':
1268
checker._report_items.append(
1269
'Missing inventory {%s}' % (record.key,))
1271
last_object = self._check_record('inventories', record,
1272
checker, last_object,
1273
current_keys[('inventories',) + record.key])
1274
del keys['inventories']
1277
bar.update("texts", 1)
1278
while (checker.pending_keys or keys['chk_bytes']
1280
# Something to check.
1281
current_keys = checker.pending_keys
1282
checker.pending_keys = {}
1283
# Accumulate current checks.
1284
for key in current_keys:
1285
if key[0] not in kinds:
1286
checker._report_items.append('unknown key type %r' % (key,))
1287
keys[key[0]].add(key[1:])
1288
# Check the outermost kind only - inventories || chk_bytes || texts
1292
for record in getattr(self, kind).check(keys=keys[kind]):
1293
if record.storage_kind == 'absent':
1294
checker._report_items.append(
1295
'Missing %s {%s}' % (kind, record.key,))
1297
last_object = self._check_record(kind, record,
1298
checker, last_object, current_keys[(kind,) + record.key])
1302
def _check_record(self, kind, record, checker, last_object, item_data):
1303
"""Check a single text from this repository."""
1304
if kind == 'inventories':
1305
rev_id = record.key[0]
1306
inv = self._deserialise_inventory(rev_id,
1307
record.get_bytes_as('fulltext'))
1308
if last_object is not None:
1309
delta = inv._make_delta(last_object)
1310
for old_path, path, file_id, ie in delta:
1313
ie.check(checker, rev_id, inv)
1315
for path, ie in inv.iter_entries():
1316
ie.check(checker, rev_id, inv)
1317
if self._format.fast_deltas:
1319
elif kind == 'chk_bytes':
1320
# No code written to check chk_bytes for this repo format.
1321
checker._report_items.append(
1322
'unsupported key type chk_bytes for %s' % (record.key,))
1323
elif kind == 'texts':
1324
self._check_text(record, checker, item_data)
1326
checker._report_items.append(
1327
'unknown key type %s for %s' % (kind, record.key))
1329
def _check_text(self, record, checker, item_data):
1330
"""Check a single text."""
1331
# Check it is extractable.
1332
# TODO: check length.
1333
if record.storage_kind == 'chunked':
1334
chunks = record.get_bytes_as(record.storage_kind)
1335
sha1 = osutils.sha_strings(chunks)
1336
length = sum(map(len, chunks))
1338
content = record.get_bytes_as('fulltext')
1339
sha1 = osutils.sha_string(content)
1340
length = len(content)
1341
if item_data and sha1 != item_data[1]:
1342
checker._report_items.append(
1343
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1344
(record.key, sha1, item_data[1], item_data[2]))
1347
def create(a_bzrdir):
1348
"""Construct the current default format repository in a_bzrdir."""
1349
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1351
def __init__(self, _format, a_bzrdir, control_files):
1352
"""instantiate a Repository.
1354
:param _format: The format of the repository on disk.
1355
:param a_bzrdir: The BzrDir of the repository.
1357
# In the future we will have a single api for all stores for
1358
# getting file texts, inventories and revisions, then
1359
# this construct will accept instances of those things.
1360
super(Repository, self).__init__()
1361
self._format = _format
1362
# the following are part of the public API for Repository:
1363
self.bzrdir = a_bzrdir
1364
self.control_files = control_files
1365
self._transport = control_files._transport
1366
self.base = self._transport.base
1368
self._reconcile_does_inventory_gc = True
1369
self._reconcile_fixes_text_parents = False
1370
self._reconcile_backsup_inventory = True
1371
self._write_group = None
1372
# Additional places to query for data.
1373
self._fallback_repositories = []
1374
# An InventoryEntry cache, used during deserialization
1375
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1376
# Is it safe to return inventory entries directly from the entry cache,
1377
# rather copying them?
1378
self._safe_to_return_from_cache = False
1381
def user_transport(self):
1382
return self.bzrdir.user_transport
1385
def control_transport(self):
1386
return self._transport
1389
if self._fallback_repositories:
1390
return '%s(%r, fallback_repositories=%r)' % (
1391
self.__class__.__name__,
1393
self._fallback_repositories)
1395
return '%s(%r)' % (self.__class__.__name__,
1398
def _has_same_fallbacks(self, other_repo):
1399
"""Returns true if the repositories have the same fallbacks."""
1400
my_fb = self._fallback_repositories
1401
other_fb = other_repo._fallback_repositories
1402
if len(my_fb) != len(other_fb):
1404
for f, g in zip(my_fb, other_fb):
1405
if not f.has_same_location(g):
1409
def has_same_location(self, other):
1410
"""Returns a boolean indicating if this repository is at the same
1411
location as another repository.
1413
This might return False even when two repository objects are accessing
1414
the same physical repository via different URLs.
1416
if self.__class__ is not other.__class__:
1418
return (self._transport.base == other._transport.base)
1420
def is_in_write_group(self):
1421
"""Return True if there is an open write group.
1423
:seealso: start_write_group.
1425
return self._write_group is not None
1427
def is_locked(self):
1428
return self.control_files.is_locked()
1430
def is_write_locked(self):
1431
"""Return True if this object is write locked."""
1432
return self.is_locked() and self.control_files._lock_mode == 'w'
1434
def lock_write(self, token=None):
1435
"""Lock this repository for writing.
1437
This causes caching within the repository obejct to start accumlating
1438
data during reads, and allows a 'write_group' to be obtained. Write
1439
groups must be used for actual data insertion.
1441
A token should be passed in if you know that you have locked the object
1442
some other way, and need to synchronise this object's state with that
1445
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1447
:param token: if this is already locked, then lock_write will fail
1448
unless the token matches the existing lock.
1449
:returns: a token if this instance supports tokens, otherwise None.
1450
:raises TokenLockingNotSupported: when a token is given but this
1451
instance doesn't support using token locks.
1452
:raises MismatchedToken: if the specified token doesn't match the token
1453
of the existing lock.
1454
:seealso: start_write_group.
1455
:return: A RepositoryWriteLockResult.
1457
locked = self.is_locked()
1458
token = self.control_files.lock_write(token=token)
1460
self._warn_if_deprecated()
1461
self._note_lock('w')
1462
for repo in self._fallback_repositories:
1463
# Writes don't affect fallback repos
1465
self._refresh_data()
1466
return RepositoryWriteLockResult(self.unlock, token)
1468
def lock_read(self):
1469
"""Lock the repository for read operations.
1471
:return: An object with an unlock method which will release the lock
1474
locked = self.is_locked()
1475
self.control_files.lock_read()
1477
self._warn_if_deprecated()
1478
self._note_lock('r')
1479
for repo in self._fallback_repositories:
1481
self._refresh_data()
1482
return LogicalLockResult(self.unlock)
1484
def get_physical_lock_status(self):
1485
return self.control_files.get_physical_lock_status()
1487
def leave_lock_in_place(self):
1488
"""Tell this repository not to release the physical lock when this
1491
If lock_write doesn't return a token, then this method is not supported.
1493
self.control_files.leave_in_place()
1495
def dont_leave_lock_in_place(self):
1496
"""Tell this repository to release the physical lock when this
1497
object is unlocked, even if it didn't originally acquire it.
1499
If lock_write doesn't return a token, then this method is not supported.
1501
self.control_files.dont_leave_in_place()
1504
def gather_stats(self, revid=None, committers=None):
1505
"""Gather statistics from a revision id.
1507
:param revid: The revision id to gather statistics from, if None, then
1508
no revision specific statistics are gathered.
1509
:param committers: Optional parameter controlling whether to grab
1510
a count of committers from the revision specific statistics.
1511
:return: A dictionary of statistics. Currently this contains:
1512
committers: The number of committers if requested.
1513
firstrev: A tuple with timestamp, timezone for the penultimate left
1514
most ancestor of revid, if revid is not the NULL_REVISION.
1515
latestrev: A tuple with timestamp, timezone for revid, if revid is
1516
not the NULL_REVISION.
1517
revisions: The total revision count in the repository.
1518
size: An estimate disk size of the repository in bytes.
1521
if revid and committers:
1522
result['committers'] = 0
1523
if revid and revid != _mod_revision.NULL_REVISION:
1525
all_committers = set()
1526
revisions = self.get_ancestry(revid)
1527
# pop the leading None
1529
first_revision = None
1531
# ignore the revisions in the middle - just grab first and last
1532
revisions = revisions[0], revisions[-1]
1533
for revision in self.get_revisions(revisions):
1534
if not first_revision:
1535
first_revision = revision
1537
all_committers.add(revision.committer)
1538
last_revision = revision
1540
result['committers'] = len(all_committers)
1541
result['firstrev'] = (first_revision.timestamp,
1542
first_revision.timezone)
1543
result['latestrev'] = (last_revision.timestamp,
1544
last_revision.timezone)
1546
# now gather global repository information
1547
# XXX: This is available for many repos regardless of listability.
1548
if self.user_transport.listable():
1549
# XXX: do we want to __define len__() ?
1550
# Maybe the versionedfiles object should provide a different
1551
# method to get the number of keys.
1552
result['revisions'] = len(self.revisions.keys())
1553
# result['size'] = t
1556
def find_branches(self, using=False):
1557
"""Find branches underneath this repository.
1559
This will include branches inside other branches.
1561
:param using: If True, list only branches using this repository.
1563
if using and not self.is_shared():
1564
return self.bzrdir.list_branches()
1565
class Evaluator(object):
1568
self.first_call = True
1570
def __call__(self, bzrdir):
1571
# On the first call, the parameter is always the bzrdir
1572
# containing the current repo.
1573
if not self.first_call:
1575
repository = bzrdir.open_repository()
1576
except errors.NoRepositoryPresent:
1579
return False, ([], repository)
1580
self.first_call = False
1581
value = (bzrdir.list_branches(), None)
1585
for branches, repository in bzrdir.BzrDir.find_bzrdirs(
1586
self.user_transport, evaluate=Evaluator()):
1587
if branches is not None:
1588
ret.extend(branches)
1589
if not using and repository is not None:
1590
ret.extend(repository.find_branches())
1594
def search_missing_revision_ids(self, other,
1595
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
1596
find_ghosts=True, revision_ids=None, if_present_ids=None):
1597
"""Return the revision ids that other has that this does not.
1599
These are returned in topological order.
1601
revision_id: only return revision ids included by revision_id.
1603
if symbol_versioning.deprecated_passed(revision_id):
1604
symbol_versioning.warn(
1605
'search_missing_revision_ids(revision_id=...) was '
1606
'deprecated in 2.4. Use revision_ids=[...] instead.',
1607
DeprecationWarning, stacklevel=3)
1608
if revision_ids is not None:
1609
raise AssertionError(
1610
'revision_ids is mutually exclusive with revision_id')
1611
if revision_id is not None:
1612
revision_ids = [revision_id]
1613
return InterRepository.get(other, self).search_missing_revision_ids(
1614
find_ghosts=find_ghosts, revision_ids=revision_ids,
1615
if_present_ids=if_present_ids)
1619
"""Open the repository rooted at base.
1621
For instance, if the repository is at URL/.bzr/repository,
1622
Repository.open(URL) -> a Repository instance.
1624
control = bzrdir.BzrDir.open(base)
1625
return control.open_repository()
1627
def copy_content_into(self, destination, revision_id=None):
1628
"""Make a complete copy of the content in self into destination.
1630
This is a destructive operation! Do not use it on existing
1633
return InterRepository.get(self, destination).copy_content(revision_id)
1635
def commit_write_group(self):
1636
"""Commit the contents accrued within the current write group.
1638
:seealso: start_write_group.
1640
:return: it may return an opaque hint that can be passed to 'pack'.
1642
if self._write_group is not self.get_transaction():
1643
# has an unlock or relock occured ?
1644
raise errors.BzrError('mismatched lock context %r and '
1646
(self.get_transaction(), self._write_group))
1647
result = self._commit_write_group()
1648
self._write_group = None
1651
def _commit_write_group(self):
1652
"""Template method for per-repository write group cleanup.
1654
This is called before the write group is considered to be
1655
finished and should ensure that all data handed to the repository
1656
for writing during the write group is safely committed (to the
1657
extent possible considering file system caching etc).
1660
def suspend_write_group(self):
1661
raise errors.UnsuspendableWriteGroup(self)
1663
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1664
"""Return the keys of missing inventory parents for revisions added in
1667
A revision is not complete if the inventory delta for that revision
1668
cannot be calculated. Therefore if the parent inventories of a
1669
revision are not present, the revision is incomplete, and e.g. cannot
1670
be streamed by a smart server. This method finds missing inventory
1671
parents for revisions added in this write group.
1673
if not self._format.supports_external_lookups:
1674
# This is only an issue for stacked repositories
1676
if not self.is_in_write_group():
1677
raise AssertionError('not in a write group')
1679
# XXX: We assume that every added revision already has its
1680
# corresponding inventory, so we only check for parent inventories that
1681
# might be missing, rather than all inventories.
1682
parents = set(self.revisions._index.get_missing_parents())
1683
parents.discard(_mod_revision.NULL_REVISION)
1684
unstacked_inventories = self.inventories._index
1685
present_inventories = unstacked_inventories.get_parent_map(
1686
key[-1:] for key in parents)
1687
parents.difference_update(present_inventories)
1688
if len(parents) == 0:
1689
# No missing parent inventories.
1691
if not check_for_missing_texts:
1692
return set(('inventories', rev_id) for (rev_id,) in parents)
1693
# Ok, now we have a list of missing inventories. But these only matter
1694
# if the inventories that reference them are missing some texts they
1695
# appear to introduce.
1696
# XXX: Texts referenced by all added inventories need to be present,
1697
# but at the moment we're only checking for texts referenced by
1698
# inventories at the graph's edge.
1699
key_deps = self.revisions._index._key_dependencies
1700
key_deps.satisfy_refs_for_keys(present_inventories)
1701
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1702
file_ids = self.fileids_altered_by_revision_ids(referrers)
1703
missing_texts = set()
1704
for file_id, version_ids in file_ids.iteritems():
1705
missing_texts.update(
1706
(file_id, version_id) for version_id in version_ids)
1707
present_texts = self.texts.get_parent_map(missing_texts)
1708
missing_texts.difference_update(present_texts)
1709
if not missing_texts:
1710
# No texts are missing, so all revisions and their deltas are
1713
# Alternatively the text versions could be returned as the missing
1714
# keys, but this is likely to be less data.
1715
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1718
def refresh_data(self):
1719
"""Re-read any data needed to synchronise with disk.
1721
This method is intended to be called after another repository instance
1722
(such as one used by a smart server) has inserted data into the
1723
repository. On all repositories this will work outside of write groups.
1724
Some repository formats (pack and newer for bzrlib native formats)
1725
support refresh_data inside write groups. If called inside a write
1726
group on a repository that does not support refreshing in a write group
1727
IsInWriteGroupError will be raised.
1729
self._refresh_data()
1731
def resume_write_group(self, tokens):
1732
if not self.is_write_locked():
1733
raise errors.NotWriteLocked(self)
1734
if self._write_group:
1735
raise errors.BzrError('already in a write group')
1736
self._resume_write_group(tokens)
1737
# so we can detect unlock/relock - the write group is now entered.
1738
self._write_group = self.get_transaction()
1740
def _resume_write_group(self, tokens):
1741
raise errors.UnsuspendableWriteGroup(self)
1743
def fetch(self, source, revision_id=None, find_ghosts=False,
1745
"""Fetch the content required to construct revision_id from source.
1747
If revision_id is None and fetch_spec is None, then all content is
1750
fetch() may not be used when the repository is in a write group -
1751
either finish the current write group before using fetch, or use
1752
fetch before starting the write group.
1754
:param find_ghosts: Find and copy revisions in the source that are
1755
ghosts in the target (and not reachable directly by walking out to
1756
the first-present revision in target from revision_id).
1757
:param revision_id: If specified, all the content needed for this
1758
revision ID will be copied to the target. Fetch will determine for
1759
itself which content needs to be copied.
1760
:param fetch_spec: If specified, a SearchResult or
1761
PendingAncestryResult that describes which revisions to copy. This
1762
allows copying multiple heads at once. Mutually exclusive with
1765
if fetch_spec is not None and revision_id is not None:
1766
raise AssertionError(
1767
"fetch_spec and revision_id are mutually exclusive.")
1768
if self.is_in_write_group():
1769
raise errors.InternalBzrError(
1770
"May not fetch while in a write group.")
1771
# fast path same-url fetch operations
1772
# TODO: lift out to somewhere common with RemoteRepository
1773
# <https://bugs.launchpad.net/bzr/+bug/401646>
1774
if (self.has_same_location(source)
1775
and fetch_spec is None
1776
and self._has_same_fallbacks(source)):
1777
# check that last_revision is in 'from' and then return a
1779
if (revision_id is not None and
1780
not _mod_revision.is_null(revision_id)):
1781
self.get_revision(revision_id)
1783
inter = InterRepository.get(source, self)
1784
return inter.fetch(revision_id=revision_id,
1785
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1787
def create_bundle(self, target, base, fileobj, format=None):
1788
return serializer.write_bundle(self, target, base, fileobj, format)
1790
def get_commit_builder(self, branch, parents, config, timestamp=None,
1791
timezone=None, committer=None, revprops=None,
1793
"""Obtain a CommitBuilder for this repository.
1795
:param branch: Branch to commit to.
1796
:param parents: Revision ids of the parents of the new revision.
1797
:param config: Configuration to use.
1798
:param timestamp: Optional timestamp recorded for commit.
1799
:param timezone: Optional timezone for timestamp.
1800
:param committer: Optional committer to set for commit.
1801
:param revprops: Optional dictionary of revision properties.
1802
:param revision_id: Optional revision id.
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)
1810
self.start_write_group()
1813
@only_raises(errors.LockNotHeld, errors.LockBroken)
1815
if (self.control_files._lock_count == 1 and
1816
self.control_files._lock_mode == 'w'):
1817
if self._write_group is not None:
1818
self.abort_write_group()
1819
self.control_files.unlock()
1820
raise errors.BzrError(
1821
'Must end write groups before releasing write locks.')
1822
self.control_files.unlock()
1823
if self.control_files._lock_count == 0:
1824
self._inventory_entry_cache.clear()
1825
for repo in self._fallback_repositories:
1829
def clone(self, a_bzrdir, revision_id=None):
1830
"""Clone this repository into a_bzrdir using the current format.
1832
Currently no check is made that the format of this repository and
1833
the bzrdir format are compatible. FIXME RBC 20060201.
1835
:return: The newly created destination repository.
1837
# TODO: deprecate after 0.16; cloning this with all its settings is
1838
# probably not very useful -- mbp 20070423
1839
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1840
self.copy_content_into(dest_repo, revision_id)
1843
def start_write_group(self):
1844
"""Start a write group in the repository.
1846
Write groups are used by repositories which do not have a 1:1 mapping
1847
between file ids and backend store to manage the insertion of data from
1848
both fetch and commit operations.
1850
A write lock is required around the start_write_group/commit_write_group
1851
for the support of lock-requiring repository formats.
1853
One can only insert data into a repository inside a write group.
1857
if not self.is_write_locked():
1858
raise errors.NotWriteLocked(self)
1859
if self._write_group:
1860
raise errors.BzrError('already in a write group')
1861
self._start_write_group()
1862
# so we can detect unlock/relock - the write group is now entered.
1863
self._write_group = self.get_transaction()
1865
def _start_write_group(self):
1866
"""Template method for per-repository write group startup.
1868
This is called before the write group is considered to be
1873
def sprout(self, to_bzrdir, revision_id=None):
1874
"""Create a descendent repository for new development.
1876
Unlike clone, this does not copy the settings of the repository.
1878
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1879
dest_repo.fetch(self, revision_id=revision_id)
1882
def _create_sprouting_repo(self, a_bzrdir, shared):
1883
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1884
# use target default format.
1885
dest_repo = a_bzrdir.create_repository()
1887
# Most control formats need the repository to be specifically
1888
# created, but on some old all-in-one formats it's not needed
1890
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1891
except errors.UninitializableFormat:
1892
dest_repo = a_bzrdir.open_repository()
1895
def _get_sink(self):
1896
"""Return a sink for streaming into this repository."""
1897
return StreamSink(self)
1899
def _get_source(self, to_format):
1900
"""Return a source for streaming from this repository."""
1901
return StreamSource(self, to_format)
1904
def has_revision(self, revision_id):
1905
"""True if this repository has a copy of the revision."""
1906
return revision_id in self.has_revisions((revision_id,))
1909
def has_revisions(self, revision_ids):
1910
"""Probe to find out the presence of multiple revisions.
1912
:param revision_ids: An iterable of revision_ids.
1913
:return: A set of the revision_ids that were present.
1915
parent_map = self.revisions.get_parent_map(
1916
[(rev_id,) for rev_id in revision_ids])
1918
if _mod_revision.NULL_REVISION in revision_ids:
1919
result.add(_mod_revision.NULL_REVISION)
1920
result.update([key[0] for key in parent_map])
1924
def get_revision(self, revision_id):
1925
"""Return the Revision object for a named revision."""
1926
return self.get_revisions([revision_id])[0]
1929
def get_revision_reconcile(self, revision_id):
1930
"""'reconcile' helper routine that allows access to a revision always.
1932
This variant of get_revision does not cross check the weave graph
1933
against the revision one as get_revision does: but it should only
1934
be used by reconcile, or reconcile-alike commands that are correcting
1935
or testing the revision graph.
1937
return self._get_revisions([revision_id])[0]
1940
def get_revisions(self, revision_ids):
1941
"""Get many revisions at once.
1943
Repositories that need to check data on every revision read should
1944
subclass this method.
1946
return self._get_revisions(revision_ids)
1949
def _get_revisions(self, revision_ids):
1950
"""Core work logic to get many revisions without sanity checks."""
1952
for revid, rev in self._iter_revisions(revision_ids):
1954
raise errors.NoSuchRevision(self, revid)
1956
return [revs[revid] for revid in revision_ids]
1958
def _iter_revisions(self, revision_ids):
1959
"""Iterate over revision objects.
1961
:param revision_ids: An iterable of revisions to examine. None may be
1962
passed to request all revisions known to the repository. Note that
1963
not all repositories can find unreferenced revisions; for those
1964
repositories only referenced ones will be returned.
1965
:return: An iterator of (revid, revision) tuples. Absent revisions (
1966
those asked for but not available) are returned as (revid, None).
1968
if revision_ids is None:
1969
revision_ids = self.all_revision_ids()
1971
for rev_id in revision_ids:
1972
if not rev_id or not isinstance(rev_id, basestring):
1973
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1974
keys = [(key,) for key in revision_ids]
1975
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1976
for record in stream:
1977
revid = record.key[0]
1978
if record.storage_kind == 'absent':
1981
text = record.get_bytes_as('fulltext')
1982
rev = self._serializer.read_revision_from_string(text)
1985
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1986
"""Produce a generator of revision deltas.
1988
Note that the input is a sequence of REVISIONS, not revision_ids.
1989
Trees will be held in memory until the generator exits.
1990
Each delta is relative to the revision's lefthand predecessor.
1992
:param specific_fileids: if not None, the result is filtered
1993
so that only those file-ids, their parents and their
1994
children are included.
1996
# Get the revision-ids of interest
1997
required_trees = set()
1998
for revision in revisions:
1999
required_trees.add(revision.revision_id)
2000
required_trees.update(revision.parent_ids[:1])
2002
# Get the matching filtered trees. Note that it's more
2003
# efficient to pass filtered trees to changes_from() rather
2004
# than doing the filtering afterwards. changes_from() could
2005
# arguably do the filtering itself but it's path-based, not
2006
# file-id based, so filtering before or afterwards is
2008
if specific_fileids is None:
2009
trees = dict((t.get_revision_id(), t) for
2010
t in self.revision_trees(required_trees))
2012
trees = dict((t.get_revision_id(), t) for
2013
t in self._filtered_revision_trees(required_trees,
2016
# Calculate the deltas
2017
for revision in revisions:
2018
if not revision.parent_ids:
2019
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
2021
old_tree = trees[revision.parent_ids[0]]
2022
yield trees[revision.revision_id].changes_from(old_tree)
2025
def get_revision_delta(self, revision_id, specific_fileids=None):
2026
"""Return the delta for one revision.
2028
The delta is relative to the left-hand predecessor of the
2031
:param specific_fileids: if not None, the result is filtered
2032
so that only those file-ids, their parents and their
2033
children are included.
2035
r = self.get_revision(revision_id)
2036
return list(self.get_deltas_for_revisions([r],
2037
specific_fileids=specific_fileids))[0]
2040
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
2041
signature = gpg_strategy.sign(plaintext)
2042
self.add_signature_text(revision_id, signature)
2045
def add_signature_text(self, revision_id, signature):
2046
self.signatures.add_lines((revision_id,), (),
2047
osutils.split_lines(signature))
2049
def find_text_key_references(self):
2050
"""Find the text key references within the repository.
2052
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2053
to whether they were referred to by the inventory of the
2054
revision_id that they contain. The inventory texts from all present
2055
revision ids are assessed to generate this report.
2057
revision_keys = self.revisions.keys()
2058
w = self.inventories
2059
pb = ui.ui_factory.nested_progress_bar()
2061
return self._serializer._find_text_key_references(
2062
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
2066
def _inventory_xml_lines_for_keys(self, keys):
2067
"""Get a line iterator of the sort needed for findind references.
2069
Not relevant for non-xml inventory repositories.
2071
Ghosts in revision_keys are ignored.
2073
:param revision_keys: The revision keys for the inventories to inspect.
2074
:return: An iterator over (inventory line, revid) for the fulltexts of
2075
all of the xml inventories specified by revision_keys.
2077
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2078
for record in stream:
2079
if record.storage_kind != 'absent':
2080
chunks = record.get_bytes_as('chunked')
2081
revid = record.key[-1]
2082
lines = osutils.chunks_to_lines(chunks)
2086
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2088
"""Helper routine for fileids_altered_by_revision_ids.
2090
This performs the translation of xml lines to revision ids.
2092
:param line_iterator: An iterator of lines, origin_version_id
2093
:param revision_keys: The revision ids to filter for. This should be a
2094
set or other type which supports efficient __contains__ lookups, as
2095
the revision key from each parsed line will be looked up in the
2096
revision_keys filter.
2097
:return: a dictionary mapping altered file-ids to an iterable of
2098
revision_ids. Each altered file-ids has the exact revision_ids that
2099
altered it listed explicitly.
2101
seen = set(self._serializer._find_text_key_references(
2102
line_iterator).iterkeys())
2103
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2104
parent_seen = set(self._serializer._find_text_key_references(
2105
self._inventory_xml_lines_for_keys(parent_keys)))
2106
new_keys = seen - parent_seen
2108
setdefault = result.setdefault
2109
for key in new_keys:
2110
setdefault(key[0], set()).add(key[-1])
2113
def _find_parent_ids_of_revisions(self, revision_ids):
2114
"""Find all parent ids that are mentioned in the revision graph.
2116
:return: set of revisions that are parents of revision_ids which are
2117
not part of revision_ids themselves
2119
parent_map = self.get_parent_map(revision_ids)
2121
map(parent_ids.update, parent_map.itervalues())
2122
parent_ids.difference_update(revision_ids)
2123
parent_ids.discard(_mod_revision.NULL_REVISION)
2126
def _find_parent_keys_of_revisions(self, revision_keys):
2127
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2129
:param revision_keys: An iterable of revision_keys.
2130
:return: The parents of all revision_keys that are not already in
2133
parent_map = self.revisions.get_parent_map(revision_keys)
2135
map(parent_keys.update, parent_map.itervalues())
2136
parent_keys.difference_update(revision_keys)
2137
parent_keys.discard(_mod_revision.NULL_REVISION)
2140
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2141
"""Find the file ids and versions affected by revisions.
2143
:param revisions: an iterable containing revision ids.
2144
:param _inv_weave: The inventory weave from this repository or None.
2145
If None, the inventory weave will be opened automatically.
2146
:return: a dictionary mapping altered file-ids to an iterable of
2147
revision_ids. Each altered file-ids has the exact revision_ids that
2148
altered it listed explicitly.
2150
selected_keys = set((revid,) for revid in revision_ids)
2151
w = _inv_weave or self.inventories
2152
return self._find_file_ids_from_xml_inventory_lines(
2153
w.iter_lines_added_or_present_in_keys(
2154
selected_keys, pb=None),
2157
def iter_files_bytes(self, desired_files):
2158
"""Iterate through file versions.
2160
Files will not necessarily be returned in the order they occur in
2161
desired_files. No specific order is guaranteed.
2163
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2164
value supplied by the caller as part of desired_files. It should
2165
uniquely identify the file version in the caller's context. (Examples:
2166
an index number or a TreeTransform trans_id.)
2168
bytes_iterator is an iterable of bytestrings for the file. The
2169
kind of iterable and length of the bytestrings are unspecified, but for
2170
this implementation, it is a list of bytes produced by
2171
VersionedFile.get_record_stream().
2173
:param desired_files: a list of (file_id, revision_id, identifier)
2177
for file_id, revision_id, callable_data in desired_files:
2178
text_keys[(file_id, revision_id)] = callable_data
2179
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2180
if record.storage_kind == 'absent':
2181
raise errors.RevisionNotPresent(record.key, self)
2182
yield text_keys[record.key], record.get_bytes_as('chunked')
2184
def _generate_text_key_index(self, text_key_references=None,
2186
"""Generate a new text key index for the repository.
2188
This is an expensive function that will take considerable time to run.
2190
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2191
list of parents, also text keys. When a given key has no parents,
2192
the parents list will be [NULL_REVISION].
2194
# All revisions, to find inventory parents.
2195
if ancestors is None:
2196
graph = self.get_graph()
2197
ancestors = graph.get_parent_map(self.all_revision_ids())
2198
if text_key_references is None:
2199
text_key_references = self.find_text_key_references()
2200
pb = ui.ui_factory.nested_progress_bar()
2202
return self._do_generate_text_key_index(ancestors,
2203
text_key_references, pb)
2207
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2208
"""Helper for _generate_text_key_index to avoid deep nesting."""
2209
revision_order = tsort.topo_sort(ancestors)
2210
invalid_keys = set()
2212
for revision_id in revision_order:
2213
revision_keys[revision_id] = set()
2214
text_count = len(text_key_references)
2215
# a cache of the text keys to allow reuse; costs a dict of all the
2216
# keys, but saves a 2-tuple for every child of a given key.
2218
for text_key, valid in text_key_references.iteritems():
2220
invalid_keys.add(text_key)
2222
revision_keys[text_key[1]].add(text_key)
2223
text_key_cache[text_key] = text_key
2224
del text_key_references
2226
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2227
NULL_REVISION = _mod_revision.NULL_REVISION
2228
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2229
# too small for large or very branchy trees. However, for 55K path
2230
# trees, it would be easy to use too much memory trivially. Ideally we
2231
# could gauge this by looking at available real memory etc, but this is
2232
# always a tricky proposition.
2233
inventory_cache = lru_cache.LRUCache(10)
2234
batch_size = 10 # should be ~150MB on a 55K path tree
2235
batch_count = len(revision_order) / batch_size + 1
2237
pb.update("Calculating text parents", processed_texts, text_count)
2238
for offset in xrange(batch_count):
2239
to_query = revision_order[offset * batch_size:(offset + 1) *
2243
for revision_id in to_query:
2244
parent_ids = ancestors[revision_id]
2245
for text_key in revision_keys[revision_id]:
2246
pb.update("Calculating text parents", processed_texts)
2247
processed_texts += 1
2248
candidate_parents = []
2249
for parent_id in parent_ids:
2250
parent_text_key = (text_key[0], parent_id)
2252
check_parent = parent_text_key not in \
2253
revision_keys[parent_id]
2255
# the parent parent_id is a ghost:
2256
check_parent = False
2257
# truncate the derived graph against this ghost.
2258
parent_text_key = None
2260
# look at the parent commit details inventories to
2261
# determine possible candidates in the per file graph.
2264
inv = inventory_cache[parent_id]
2266
inv = self.revision_tree(parent_id).inventory
2267
inventory_cache[parent_id] = inv
2269
parent_entry = inv[text_key[0]]
2270
except (KeyError, errors.NoSuchId):
2272
if parent_entry is not None:
2274
text_key[0], parent_entry.revision)
2276
parent_text_key = None
2277
if parent_text_key is not None:
2278
candidate_parents.append(
2279
text_key_cache[parent_text_key])
2280
parent_heads = text_graph.heads(candidate_parents)
2281
new_parents = list(parent_heads)
2282
new_parents.sort(key=lambda x:candidate_parents.index(x))
2283
if new_parents == []:
2284
new_parents = [NULL_REVISION]
2285
text_index[text_key] = new_parents
2287
for text_key in invalid_keys:
2288
text_index[text_key] = [NULL_REVISION]
2291
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2292
"""Get an iterable listing the keys of all the data introduced by a set
2295
The keys will be ordered so that the corresponding items can be safely
2296
fetched and inserted in that order.
2298
:returns: An iterable producing tuples of (knit-kind, file-id,
2299
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2300
'revisions'. file-id is None unless knit-kind is 'file'.
2302
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2305
for result in self._find_non_file_keys_to_fetch(revision_ids):
2308
def _find_file_keys_to_fetch(self, revision_ids, pb):
2309
# XXX: it's a bit weird to control the inventory weave caching in this
2310
# generator. Ideally the caching would be done in fetch.py I think. Or
2311
# maybe this generator should explicitly have the contract that it
2312
# should not be iterated until the previously yielded item has been
2314
inv_w = self.inventories
2316
# file ids that changed
2317
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2319
num_file_ids = len(file_ids)
2320
for file_id, altered_versions in file_ids.iteritems():
2322
pb.update("Fetch texts", count, num_file_ids)
2324
yield ("file", file_id, altered_versions)
2326
def _find_non_file_keys_to_fetch(self, revision_ids):
2328
yield ("inventory", None, revision_ids)
2331
# XXX: Note ATM no callers actually pay attention to this return
2332
# instead they just use the list of revision ids and ignore
2333
# missing sigs. Consider removing this work entirely
2334
revisions_with_signatures = set(self.signatures.get_parent_map(
2335
[(r,) for r in revision_ids]))
2336
revisions_with_signatures = set(
2337
[r for (r,) in revisions_with_signatures])
2338
revisions_with_signatures.intersection_update(revision_ids)
2339
yield ("signatures", None, revisions_with_signatures)
2342
yield ("revisions", None, revision_ids)
2345
def get_inventory(self, revision_id):
2346
"""Get Inventory object by revision id."""
2347
return self.iter_inventories([revision_id]).next()
2349
def iter_inventories(self, revision_ids, ordering=None):
2350
"""Get many inventories by revision_ids.
2352
This will buffer some or all of the texts used in constructing the
2353
inventories in memory, but will only parse a single inventory at a
2356
:param revision_ids: The expected revision ids of the inventories.
2357
:param ordering: optional ordering, e.g. 'topological'. If not
2358
specified, the order of revision_ids will be preserved (by
2359
buffering if necessary).
2360
:return: An iterator of inventories.
2362
if ((None in revision_ids)
2363
or (_mod_revision.NULL_REVISION in revision_ids)):
2364
raise ValueError('cannot get null revision inventory')
2365
return self._iter_inventories(revision_ids, ordering)
2367
def _iter_inventories(self, revision_ids, ordering):
2368
"""single-document based inventory iteration."""
2369
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2370
for text, revision_id in inv_xmls:
2371
yield self._deserialise_inventory(revision_id, text)
2373
def _iter_inventory_xmls(self, revision_ids, ordering):
2374
if ordering is None:
2375
order_as_requested = True
2376
ordering = 'unordered'
2378
order_as_requested = False
2379
keys = [(revision_id,) for revision_id in revision_ids]
2382
if order_as_requested:
2383
key_iter = iter(keys)
2384
next_key = key_iter.next()
2385
stream = self.inventories.get_record_stream(keys, ordering, True)
2387
for record in stream:
2388
if record.storage_kind != 'absent':
2389
chunks = record.get_bytes_as('chunked')
2390
if order_as_requested:
2391
text_chunks[record.key] = chunks
2393
yield ''.join(chunks), record.key[-1]
2395
raise errors.NoSuchRevision(self, record.key)
2396
if order_as_requested:
2397
# Yield as many results as we can while preserving order.
2398
while next_key in text_chunks:
2399
chunks = text_chunks.pop(next_key)
2400
yield ''.join(chunks), next_key[-1]
2402
next_key = key_iter.next()
2403
except StopIteration:
2404
# We still want to fully consume the get_record_stream,
2405
# just in case it is not actually finished at this point
2409
def _deserialise_inventory(self, revision_id, xml):
2410
"""Transform the xml into an inventory object.
2412
:param revision_id: The expected revision id of the inventory.
2413
:param xml: A serialised inventory.
2415
result = self._serializer.read_inventory_from_string(xml, revision_id,
2416
entry_cache=self._inventory_entry_cache,
2417
return_from_cache=self._safe_to_return_from_cache)
2418
if result.revision_id != revision_id:
2419
raise AssertionError('revision id mismatch %s != %s' % (
2420
result.revision_id, revision_id))
2423
def get_serializer_format(self):
2424
return self._serializer.format_num
2427
def _get_inventory_xml(self, revision_id):
2428
"""Get serialized inventory as a string."""
2429
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2431
text, revision_id = texts.next()
2432
except StopIteration:
2433
raise errors.HistoryMissing(self, 'inventory', revision_id)
2436
def get_rev_id_for_revno(self, revno, known_pair):
2437
"""Return the revision id of a revno, given a later (revno, revid)
2438
pair in the same history.
2440
:return: if found (True, revid). If the available history ran out
2441
before reaching the revno, then this returns
2442
(False, (closest_revno, closest_revid)).
2444
known_revno, known_revid = known_pair
2445
partial_history = [known_revid]
2446
distance_from_known = known_revno - revno
2447
if distance_from_known < 0:
2449
'requested revno (%d) is later than given known revno (%d)'
2450
% (revno, known_revno))
2453
self, partial_history, stop_index=distance_from_known)
2454
except errors.RevisionNotPresent, err:
2455
if err.revision_id == known_revid:
2456
# The start revision (known_revid) wasn't found.
2458
# This is a stacked repository with no fallbacks, or a there's a
2459
# left-hand ghost. Either way, even though the revision named in
2460
# the error isn't in this repo, we know it's the next step in this
2461
# left-hand history.
2462
partial_history.append(err.revision_id)
2463
if len(partial_history) <= distance_from_known:
2464
# Didn't find enough history to get a revid for the revno.
2465
earliest_revno = known_revno - len(partial_history) + 1
2466
return (False, (earliest_revno, partial_history[-1]))
2467
if len(partial_history) - 1 > distance_from_known:
2468
raise AssertionError('_iter_for_revno returned too much history')
2469
return (True, partial_history[-1])
2471
def iter_reverse_revision_history(self, revision_id):
2472
"""Iterate backwards through revision ids in the lefthand history
2474
:param revision_id: The revision id to start with. All its lefthand
2475
ancestors will be traversed.
2477
graph = self.get_graph()
2478
stop_revisions = (None, _mod_revision.NULL_REVISION)
2479
return graph.iter_lefthand_ancestry(revision_id, stop_revisions)
2481
def is_shared(self):
2482
"""Return True if this repository is flagged as a shared repository."""
2483
raise NotImplementedError(self.is_shared)
2486
def reconcile(self, other=None, thorough=False):
2487
"""Reconcile this repository."""
2488
from bzrlib.reconcile import RepoReconciler
2489
reconciler = RepoReconciler(self, thorough=thorough)
2490
reconciler.reconcile()
2493
def _refresh_data(self):
2494
"""Helper called from lock_* to ensure coherency with disk.
2496
The default implementation does nothing; it is however possible
2497
for repositories to maintain loaded indices across multiple locks
2498
by checking inside their implementation of this method to see
2499
whether their indices are still valid. This depends of course on
2500
the disk format being validatable in this manner. This method is
2501
also called by the refresh_data() public interface to cause a refresh
2502
to occur while in a write lock so that data inserted by a smart server
2503
push operation is visible on the client's instance of the physical
2508
def revision_tree(self, revision_id):
2509
"""Return Tree for a revision on this branch.
2511
`revision_id` may be NULL_REVISION for the empty tree revision.
2513
revision_id = _mod_revision.ensure_null(revision_id)
2514
# TODO: refactor this to use an existing revision object
2515
# so we don't need to read it in twice.
2516
if revision_id == _mod_revision.NULL_REVISION:
2517
return RevisionTree(self, Inventory(root_id=None),
2518
_mod_revision.NULL_REVISION)
2520
inv = self.get_inventory(revision_id)
2521
return RevisionTree(self, inv, revision_id)
2523
def revision_trees(self, revision_ids):
2524
"""Return Trees for revisions in this repository.
2526
:param revision_ids: a sequence of revision-ids;
2527
a revision-id may not be None or 'null:'
2529
inventories = self.iter_inventories(revision_ids)
2530
for inv in inventories:
2531
yield RevisionTree(self, inv, inv.revision_id)
2533
def _filtered_revision_trees(self, revision_ids, file_ids):
2534
"""Return Tree for a revision on this branch with only some files.
2536
:param revision_ids: a sequence of revision-ids;
2537
a revision-id may not be None or 'null:'
2538
:param file_ids: if not None, the result is filtered
2539
so that only those file-ids, their parents and their
2540
children are included.
2542
inventories = self.iter_inventories(revision_ids)
2543
for inv in inventories:
2544
# Should we introduce a FilteredRevisionTree class rather
2545
# than pre-filter the inventory here?
2546
filtered_inv = inv.filter(file_ids)
2547
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2550
def get_ancestry(self, revision_id, topo_sorted=True):
2551
"""Return a list of revision-ids integrated by a revision.
2553
The first element of the list is always None, indicating the origin
2554
revision. This might change when we have history horizons, or
2555
perhaps we should have a new API.
2557
This is topologically sorted.
2559
if _mod_revision.is_null(revision_id):
2561
if not self.has_revision(revision_id):
2562
raise errors.NoSuchRevision(self, revision_id)
2563
graph = self.get_graph()
2565
search = graph._make_breadth_first_searcher([revision_id])
2568
found, ghosts = search.next_with_ghosts()
2569
except StopIteration:
2572
if _mod_revision.NULL_REVISION in keys:
2573
keys.remove(_mod_revision.NULL_REVISION)
2575
parent_map = graph.get_parent_map(keys)
2576
keys = tsort.topo_sort(parent_map)
2577
return [None] + list(keys)
2579
def pack(self, hint=None, clean_obsolete_packs=False):
2580
"""Compress the data within the repository.
2582
This operation only makes sense for some repository types. For other
2583
types it should be a no-op that just returns.
2585
This stub method does not require a lock, but subclasses should use
2586
@needs_write_lock as this is a long running call it's reasonable to
2587
implicitly lock for the user.
2589
:param hint: If not supplied, the whole repository is packed.
2590
If supplied, the repository may use the hint parameter as a
2591
hint for the parts of the repository to pack. A hint can be
2592
obtained from the result of commit_write_group(). Out of
2593
date hints are simply ignored, because concurrent operations
2594
can obsolete them rapidly.
2596
:param clean_obsolete_packs: Clean obsolete packs immediately after
2600
def get_transaction(self):
2601
return self.control_files.get_transaction()
2603
def get_parent_map(self, revision_ids):
2604
"""See graph.StackedParentsProvider.get_parent_map"""
2605
# revisions index works in keys; this just works in revisions
2606
# therefore wrap and unwrap
2609
for revision_id in revision_ids:
2610
if revision_id == _mod_revision.NULL_REVISION:
2611
result[revision_id] = ()
2612
elif revision_id is None:
2613
raise ValueError('get_parent_map(None) is not valid')
2615
query_keys.append((revision_id ,))
2616
for ((revision_id,), parent_keys) in \
2617
self.revisions.get_parent_map(query_keys).iteritems():
2619
result[revision_id] = tuple([parent_revid
2620
for (parent_revid,) in parent_keys])
2622
result[revision_id] = (_mod_revision.NULL_REVISION,)
2625
def _make_parents_provider(self):
2629
def get_known_graph_ancestry(self, revision_ids):
2630
"""Return the known graph for a set of revision ids and their ancestors.
2632
st = static_tuple.StaticTuple
2633
revision_keys = [st(r_id).intern() for r_id in revision_ids]
2634
known_graph = self.revisions.get_known_graph_ancestry(revision_keys)
2635
return graph.GraphThunkIdsToKeys(known_graph)
2637
def get_graph(self, other_repository=None):
2638
"""Return the graph walker for this repository format"""
2639
parents_provider = self._make_parents_provider()
2640
if (other_repository is not None and
2641
not self.has_same_location(other_repository)):
2642
parents_provider = graph.StackedParentsProvider(
2643
[parents_provider, other_repository._make_parents_provider()])
2644
return graph.Graph(parents_provider)
2646
def _get_versioned_file_checker(self, text_key_references=None,
2648
"""Return an object suitable for checking versioned files.
2650
:param text_key_references: if non-None, an already built
2651
dictionary mapping text keys ((fileid, revision_id) tuples)
2652
to whether they were referred to by the inventory of the
2653
revision_id that they contain. If None, this will be
2655
:param ancestors: Optional result from
2656
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2659
return _VersionedFileChecker(self,
2660
text_key_references=text_key_references, ancestors=ancestors)
2662
def revision_ids_to_search_result(self, result_set):
2663
"""Convert a set of revision ids to a graph SearchResult."""
2664
result_parents = set()
2665
for parents in self.get_graph().get_parent_map(
2666
result_set).itervalues():
2667
result_parents.update(parents)
2668
included_keys = result_set.intersection(result_parents)
2669
start_keys = result_set.difference(included_keys)
2670
exclude_keys = result_parents.difference(result_set)
2671
result = graph.SearchResult(start_keys, exclude_keys,
2672
len(result_set), result_set)
2676
def set_make_working_trees(self, new_value):
2677
"""Set the policy flag for making working trees when creating branches.
2679
This only applies to branches that use this repository.
2681
The default is 'True'.
2682
:param new_value: True to restore the default, False to disable making
2685
raise NotImplementedError(self.set_make_working_trees)
2687
def make_working_trees(self):
2688
"""Returns the policy for making working trees on new branches."""
2689
raise NotImplementedError(self.make_working_trees)
2692
def sign_revision(self, revision_id, gpg_strategy):
2693
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2694
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2697
def has_signature_for_revision_id(self, revision_id):
2698
"""Query for a revision signature for revision_id in the repository."""
2699
if not self.has_revision(revision_id):
2700
raise errors.NoSuchRevision(self, revision_id)
2701
sig_present = (1 == len(
2702
self.signatures.get_parent_map([(revision_id,)])))
2706
def get_signature_text(self, revision_id):
2707
"""Return the text for a signature."""
2708
stream = self.signatures.get_record_stream([(revision_id,)],
2710
record = stream.next()
2711
if record.storage_kind == 'absent':
2712
raise errors.NoSuchRevision(self, revision_id)
2713
return record.get_bytes_as('fulltext')
2716
def check(self, revision_ids=None, callback_refs=None, check_repo=True):
2717
"""Check consistency of all history of given revision_ids.
2719
Different repository implementations should override _check().
2721
:param revision_ids: A non-empty list of revision_ids whose ancestry
2722
will be checked. Typically the last revision_id of a branch.
2723
:param callback_refs: A dict of check-refs to resolve and callback
2724
the check/_check method on the items listed as wanting the ref.
2726
:param check_repo: If False do not check the repository contents, just
2727
calculate the data callback_refs requires and call them back.
2729
return self._check(revision_ids, callback_refs=callback_refs,
2730
check_repo=check_repo)
2732
def _check(self, revision_ids, callback_refs, check_repo):
2733
result = check.Check(self, check_repo=check_repo)
2734
result.check(callback_refs)
2737
def _warn_if_deprecated(self, branch=None):
2738
if not self._format.is_deprecated():
2740
global _deprecation_warning_done
2741
if _deprecation_warning_done:
2745
conf = config.GlobalConfig()
2747
conf = branch.get_config()
2748
if conf.suppress_warning('format_deprecation'):
2750
warning("Format %s for %s is deprecated -"
2751
" please use 'bzr upgrade' to get better performance"
2752
% (self._format, self.bzrdir.transport.base))
2754
_deprecation_warning_done = True
2756
def supports_rich_root(self):
2757
return self._format.rich_root_data
2759
def _check_ascii_revisionid(self, revision_id, method):
2760
"""Private helper for ascii-only repositories."""
2761
# weave repositories refuse to store revisionids that are non-ascii.
2762
if revision_id is not None:
2763
# weaves require ascii revision ids.
2764
if isinstance(revision_id, unicode):
2766
revision_id.encode('ascii')
2767
except UnicodeEncodeError:
2768
raise errors.NonAsciiRevisionId(method, self)
2771
revision_id.decode('ascii')
2772
except UnicodeDecodeError:
2773
raise errors.NonAsciiRevisionId(method, self)
2775
def revision_graph_can_have_wrong_parents(self):
2776
"""Is it possible for this repository to have a revision graph with
2779
If True, then this repository must also implement
2780
_find_inconsistent_revision_parents so that check and reconcile can
2781
check for inconsistencies before proceeding with other checks that may
2782
depend on the revision index being consistent.
2784
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2787
def install_revision(repository, rev, revision_tree):
2788
"""Install all revision data into a repository."""
2789
install_revisions(repository, [(rev, revision_tree, None)])
2792
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2793
"""Install all revision data into a repository.
2795
Accepts an iterable of revision, tree, signature tuples. The signature
2798
repository.start_write_group()
2800
inventory_cache = lru_cache.LRUCache(10)
2801
for n, (revision, revision_tree, signature) in enumerate(iterable):
2802
_install_revision(repository, revision, revision_tree, signature,
2805
pb.update('Transferring revisions', n + 1, num_revisions)
2807
repository.abort_write_group()
2810
repository.commit_write_group()
2813
def _install_revision(repository, rev, revision_tree, signature,
2815
"""Install all revision data into a repository."""
2816
present_parents = []
2818
for p_id in rev.parent_ids:
2819
if repository.has_revision(p_id):
2820
present_parents.append(p_id)
2821
parent_trees[p_id] = repository.revision_tree(p_id)
2823
parent_trees[p_id] = repository.revision_tree(
2824
_mod_revision.NULL_REVISION)
2826
inv = revision_tree.inventory
2827
entries = inv.iter_entries()
2828
# backwards compatibility hack: skip the root id.
2829
if not repository.supports_rich_root():
2830
path, root = entries.next()
2831
if root.revision != rev.revision_id:
2832
raise errors.IncompatibleRevision(repr(repository))
2834
for path, ie in entries:
2835
text_keys[(ie.file_id, ie.revision)] = ie
2836
text_parent_map = repository.texts.get_parent_map(text_keys)
2837
missing_texts = set(text_keys) - set(text_parent_map)
2838
# Add the texts that are not already present
2839
for text_key in missing_texts:
2840
ie = text_keys[text_key]
2842
# FIXME: TODO: The following loop overlaps/duplicates that done by
2843
# commit to determine parents. There is a latent/real bug here where
2844
# the parents inserted are not those commit would do - in particular
2845
# they are not filtered by heads(). RBC, AB
2846
for revision, tree in parent_trees.iteritems():
2847
if ie.file_id not in tree:
2849
parent_id = tree.inventory[ie.file_id].revision
2850
if parent_id in text_parents:
2852
text_parents.append((ie.file_id, parent_id))
2853
lines = revision_tree.get_file(ie.file_id).readlines()
2854
repository.texts.add_lines(text_key, text_parents, lines)
2856
# install the inventory
2857
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2858
# Cache this inventory
2859
inventory_cache[rev.revision_id] = inv
2861
basis_inv = inventory_cache[rev.parent_ids[0]]
2863
repository.add_inventory(rev.revision_id, inv, present_parents)
2865
delta = inv._make_delta(basis_inv)
2866
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2867
rev.revision_id, present_parents)
2869
repository.add_inventory(rev.revision_id, inv, present_parents)
2870
except errors.RevisionAlreadyPresent:
2872
if signature is not None:
2873
repository.add_signature_text(rev.revision_id, signature)
2874
repository.add_revision(rev.revision_id, rev, inv)
2877
class MetaDirRepository(Repository):
2878
"""Repositories in the new meta-dir layout.
2880
:ivar _transport: Transport for access to repository control files,
2881
typically pointing to .bzr/repository.
2884
def __init__(self, _format, a_bzrdir, control_files):
2885
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2886
self._transport = control_files._transport
2888
def is_shared(self):
2889
"""Return True if this repository is flagged as a shared repository."""
2890
return self._transport.has('shared-storage')
2893
def set_make_working_trees(self, new_value):
2894
"""Set the policy flag for making working trees when creating branches.
2896
This only applies to branches that use this repository.
2898
The default is 'True'.
2899
:param new_value: True to restore the default, False to disable making
2904
self._transport.delete('no-working-trees')
2905
except errors.NoSuchFile:
2908
self._transport.put_bytes('no-working-trees', '',
2909
mode=self.bzrdir._get_file_mode())
2911
def make_working_trees(self):
2912
"""Returns the policy for making working trees on new branches."""
2913
return not self._transport.has('no-working-trees')
2916
class MetaDirVersionedFileRepository(MetaDirRepository):
2917
"""Repositories in a meta-dir, that work via versioned file objects."""
2919
def __init__(self, _format, a_bzrdir, control_files):
2920
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2924
class RepositoryFormatRegistry(controldir.ControlComponentFormatRegistry):
2925
"""Repository format registry."""
2927
def get_default(self):
2928
"""Return the current default format."""
2929
from bzrlib import bzrdir
2930
return bzrdir.format_registry.make_bzrdir('default').repository_format
2933
network_format_registry = registry.FormatRegistry()
2934
"""Registry of formats indexed by their network name.
2936
The network name for a repository format is an identifier that can be used when
2937
referring to formats with smart server operations. See
2938
RepositoryFormat.network_name() for more detail.
2942
format_registry = RepositoryFormatRegistry(network_format_registry)
2943
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
2945
This can contain either format instances themselves, or classes/factories that
2946
can be called to obtain one.
2950
#####################################################################
2951
# Repository Formats
2953
class RepositoryFormat(controldir.ControlComponentFormat):
2954
"""A repository format.
2956
Formats provide four things:
2957
* An initialization routine to construct repository data on disk.
2958
* a optional format string which is used when the BzrDir supports
2960
* an open routine which returns a Repository instance.
2961
* A network name for referring to the format in smart server RPC
2964
There is one and only one Format subclass for each on-disk format. But
2965
there can be one Repository subclass that is used for several different
2966
formats. The _format attribute on a Repository instance can be used to
2967
determine the disk format.
2969
Formats are placed in a registry by their format string for reference
2970
during opening. These should be subclasses of RepositoryFormat for
2973
Once a format is deprecated, just deprecate the initialize and open
2974
methods on the format class. Do not deprecate the object, as the
2975
object may be created even when a repository instance hasn't been
2978
Common instance attributes:
2979
_matchingbzrdir - the bzrdir format that the repository format was
2980
originally written to work with. This can be used if manually
2981
constructing a bzrdir and repository, or more commonly for test suite
2985
# Set to True or False in derived classes. True indicates that the format
2986
# supports ghosts gracefully.
2987
supports_ghosts = None
2988
# Can this repository be given external locations to lookup additional
2989
# data. Set to True or False in derived classes.
2990
supports_external_lookups = None
2991
# Does this format support CHK bytestring lookups. Set to True or False in
2993
supports_chks = None
2994
# Should commit add an inventory, or an inventory delta to the repository.
2995
_commit_inv_deltas = True
2996
# What order should fetch operations request streams in?
2997
# The default is unordered as that is the cheapest for an origin to
2999
_fetch_order = 'unordered'
3000
# Does this repository format use deltas that can be fetched as-deltas ?
3001
# (E.g. knits, where the knit deltas can be transplanted intact.
3002
# We default to False, which will ensure that enough data to get
3003
# a full text out of any fetch stream will be grabbed.
3004
_fetch_uses_deltas = False
3005
# Should fetch trigger a reconcile after the fetch? Only needed for
3006
# some repository formats that can suffer internal inconsistencies.
3007
_fetch_reconcile = False
3008
# Does this format have < O(tree_size) delta generation. Used to hint what
3009
# code path for commit, amongst other things.
3011
# Does doing a pack operation compress data? Useful for the pack UI command
3012
# (so if there is one pack, the operation can still proceed because it may
3013
# help), and for fetching when data won't have come from the same
3015
pack_compresses = False
3016
# Does the repository inventory storage understand references to trees?
3017
supports_tree_reference = None
3018
# Is the format experimental ?
3019
experimental = False
3020
# Does this repository format escape funky characters, or does it create files with
3021
# similar names as the versioned files in its contents on disk ?
3022
supports_funky_characters = None
3023
# Does this repository format support leaving locks?
3024
supports_leaving_lock = None
3025
# Does this format support the full VersionedFiles interface?
3026
supports_full_versioned_files = None
3029
return "%s()" % self.__class__.__name__
3031
def __eq__(self, other):
3032
# format objects are generally stateless
3033
return isinstance(other, self.__class__)
3035
def __ne__(self, other):
3036
return not self == other
3039
def find_format(klass, a_bzrdir):
3040
"""Return the format for the repository object in a_bzrdir.
3042
This is used by bzr native formats that have a "format" file in
3043
the repository. Other methods may be used by different types of
3047
transport = a_bzrdir.get_repository_transport(None)
3048
format_string = transport.get_bytes("format")
3049
return format_registry.get(format_string)
3050
except errors.NoSuchFile:
3051
raise errors.NoRepositoryPresent(a_bzrdir)
3053
raise errors.UnknownFormatError(format=format_string,
3057
@symbol_versioning.deprecated_method(symbol_versioning.deprecated_in((2, 4, 0)))
3058
def register_format(klass, format):
3059
format_registry.register(format)
3062
@symbol_versioning.deprecated_method(symbol_versioning.deprecated_in((2, 4, 0)))
3063
def unregister_format(klass, format):
3064
format_registry.remove(format)
3067
@symbol_versioning.deprecated_method(symbol_versioning.deprecated_in((2, 4, 0)))
3068
def get_default_format(klass):
3069
"""Return the current default format."""
3070
return format_registry.get_default()
3072
def get_format_string(self):
3073
"""Return the ASCII format string that identifies this format.
3075
Note that in pre format ?? repositories the format string is
3076
not permitted nor written to disk.
3078
raise NotImplementedError(self.get_format_string)
3080
def get_format_description(self):
3081
"""Return the short description for this format."""
3082
raise NotImplementedError(self.get_format_description)
3084
# TODO: this shouldn't be in the base class, it's specific to things that
3085
# use weaves or knits -- mbp 20070207
3086
def _get_versioned_file_store(self,
3091
versionedfile_class=None,
3092
versionedfile_kwargs={},
3094
if versionedfile_class is None:
3095
versionedfile_class = self._versionedfile_class
3096
weave_transport = control_files._transport.clone(name)
3097
dir_mode = control_files._dir_mode
3098
file_mode = control_files._file_mode
3099
return VersionedFileStore(weave_transport, prefixed=prefixed,
3101
file_mode=file_mode,
3102
versionedfile_class=versionedfile_class,
3103
versionedfile_kwargs=versionedfile_kwargs,
3106
def initialize(self, a_bzrdir, shared=False):
3107
"""Initialize a repository of this format in a_bzrdir.
3109
:param a_bzrdir: The bzrdir to put the new repository in it.
3110
:param shared: The repository should be initialized as a sharable one.
3111
:returns: The new repository object.
3113
This may raise UninitializableFormat if shared repository are not
3114
compatible the a_bzrdir.
3116
raise NotImplementedError(self.initialize)
3118
def is_supported(self):
3119
"""Is this format supported?
3121
Supported formats must be initializable and openable.
3122
Unsupported formats may not support initialization or committing or
3123
some other features depending on the reason for not being supported.
3127
def is_deprecated(self):
3128
"""Is this format deprecated?
3130
Deprecated formats may trigger a user-visible warning recommending
3131
the user to upgrade. They are still fully supported.
3135
def network_name(self):
3136
"""A simple byte string uniquely identifying this format for RPC calls.
3138
MetaDir repository formats use their disk format string to identify the
3139
repository over the wire. All in one formats such as bzr < 0.8, and
3140
foreign formats like svn/git and hg should use some marker which is
3141
unique and immutable.
3143
raise NotImplementedError(self.network_name)
3145
def check_conversion_target(self, target_format):
3146
if self.rich_root_data and not target_format.rich_root_data:
3147
raise errors.BadConversionTarget(
3148
'Does not support rich root data.', target_format,
3150
if (self.supports_tree_reference and
3151
not getattr(target_format, 'supports_tree_reference', False)):
3152
raise errors.BadConversionTarget(
3153
'Does not support nested trees', target_format,
3156
def open(self, a_bzrdir, _found=False):
3157
"""Return an instance of this format for the bzrdir a_bzrdir.
3159
_found is a private parameter, do not use it.
3161
raise NotImplementedError(self.open)
3163
def _run_post_repo_init_hooks(self, repository, a_bzrdir, shared):
3164
from bzrlib.bzrdir import BzrDir, RepoInitHookParams
3165
hooks = BzrDir.hooks['post_repo_init']
3168
params = RepoInitHookParams(repository, self, a_bzrdir, shared)
3173
class MetaDirRepositoryFormat(RepositoryFormat):
3174
"""Common base class for the new repositories using the metadir layout."""
3176
rich_root_data = False
3177
supports_tree_reference = False
3178
supports_external_lookups = False
3179
supports_leaving_lock = True
3182
def _matchingbzrdir(self):
3183
matching = bzrdir.BzrDirMetaFormat1()
3184
matching.repository_format = self
3188
super(MetaDirRepositoryFormat, self).__init__()
3190
def _create_control_files(self, a_bzrdir):
3191
"""Create the required files and the initial control_files object."""
3192
# FIXME: RBC 20060125 don't peek under the covers
3193
# NB: no need to escape relative paths that are url safe.
3194
repository_transport = a_bzrdir.get_repository_transport(self)
3195
control_files = lockable_files.LockableFiles(repository_transport,
3196
'lock', lockdir.LockDir)
3197
control_files.create_lock()
3198
return control_files
3200
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
3201
"""Upload the initial blank content."""
3202
control_files = self._create_control_files(a_bzrdir)
3203
control_files.lock_write()
3204
transport = control_files._transport
3206
utf8_files += [('shared-storage', '')]
3208
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
3209
for (filename, content_stream) in files:
3210
transport.put_file(filename, content_stream,
3211
mode=a_bzrdir._get_file_mode())
3212
for (filename, content_bytes) in utf8_files:
3213
transport.put_bytes_non_atomic(filename, content_bytes,
3214
mode=a_bzrdir._get_file_mode())
3216
control_files.unlock()
3218
def network_name(self):
3219
"""Metadir formats have matching disk and network format strings."""
3220
return self.get_format_string()
3223
# formats which have no format string are not discoverable or independently
3224
# creatable on disk, so are not registered in format_registry. They're
3225
# all in bzrlib.repofmt.knitreponow. When an instance of one of these is
3226
# needed, it's constructed directly by the BzrDir. Non-native formats where
3227
# the repository is not separately opened are similar.
3229
format_registry.register_lazy(
3230
'Bazaar-NG Knit Repository Format 1',
3231
'bzrlib.repofmt.knitrepo',
3232
'RepositoryFormatKnit1',
3235
format_registry.register_lazy(
3236
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
3237
'bzrlib.repofmt.knitrepo',
3238
'RepositoryFormatKnit3',
3241
format_registry.register_lazy(
3242
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
3243
'bzrlib.repofmt.knitrepo',
3244
'RepositoryFormatKnit4',
3247
# Pack-based formats. There is one format for pre-subtrees, and one for
3248
# post-subtrees to allow ease of testing.
3249
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
3250
format_registry.register_lazy(
3251
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
3252
'bzrlib.repofmt.pack_repo',
3253
'RepositoryFormatKnitPack1',
3255
format_registry.register_lazy(
3256
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
3257
'bzrlib.repofmt.pack_repo',
3258
'RepositoryFormatKnitPack3',
3260
format_registry.register_lazy(
3261
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
3262
'bzrlib.repofmt.pack_repo',
3263
'RepositoryFormatKnitPack4',
3265
format_registry.register_lazy(
3266
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
3267
'bzrlib.repofmt.pack_repo',
3268
'RepositoryFormatKnitPack5',
3270
format_registry.register_lazy(
3271
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
3272
'bzrlib.repofmt.pack_repo',
3273
'RepositoryFormatKnitPack5RichRoot',
3275
format_registry.register_lazy(
3276
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3277
'bzrlib.repofmt.pack_repo',
3278
'RepositoryFormatKnitPack5RichRootBroken',
3280
format_registry.register_lazy(
3281
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3282
'bzrlib.repofmt.pack_repo',
3283
'RepositoryFormatKnitPack6',
3285
format_registry.register_lazy(
3286
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3287
'bzrlib.repofmt.pack_repo',
3288
'RepositoryFormatKnitPack6RichRoot',
3290
format_registry.register_lazy(
3291
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3292
'bzrlib.repofmt.groupcompress_repo',
3293
'RepositoryFormat2a',
3296
# Development formats.
3297
# Check their docstrings to see if/when they are obsolete.
3298
format_registry.register_lazy(
3299
("Bazaar development format 2 with subtree support "
3300
"(needs bzr.dev from before 1.8)\n"),
3301
'bzrlib.repofmt.pack_repo',
3302
'RepositoryFormatPackDevelopment2Subtree',
3304
format_registry.register_lazy(
3305
'Bazaar development format 8\n',
3306
'bzrlib.repofmt.groupcompress_repo',
3307
'RepositoryFormat2aSubtree',
3311
class InterRepository(InterObject):
3312
"""This class represents operations taking place between two repositories.
3314
Its instances have methods like copy_content and fetch, and contain
3315
references to the source and target repositories these operations can be
3318
Often we will provide convenience methods on 'repository' which carry out
3319
operations with another repository - they will always forward to
3320
InterRepository.get(other).method_name(parameters).
3323
_walk_to_common_revisions_batch_size = 50
3325
"""The available optimised InterRepository types."""
3328
def copy_content(self, revision_id=None):
3329
"""Make a complete copy of the content in self into destination.
3331
This is a destructive operation! Do not use it on existing
3334
:param revision_id: Only copy the content needed to construct
3335
revision_id and its parents.
3338
self.target.set_make_working_trees(self.source.make_working_trees())
3339
except NotImplementedError:
3341
self.target.fetch(self.source, revision_id=revision_id)
3344
def fetch(self, revision_id=None, find_ghosts=False,
3346
"""Fetch the content required to construct revision_id.
3348
The content is copied from self.source to self.target.
3350
:param revision_id: if None all content is copied, if NULL_REVISION no
3354
ui.ui_factory.warn_experimental_format_fetch(self)
3355
from bzrlib.fetch import RepoFetcher
3356
# See <https://launchpad.net/bugs/456077> asking for a warning here
3357
if self.source._format.network_name() != self.target._format.network_name():
3358
ui.ui_factory.show_user_warning('cross_format_fetch',
3359
from_format=self.source._format,
3360
to_format=self.target._format)
3361
f = RepoFetcher(to_repository=self.target,
3362
from_repository=self.source,
3363
last_revision=revision_id,
3364
fetch_spec=fetch_spec,
3365
find_ghosts=find_ghosts)
3367
def _walk_to_common_revisions(self, revision_ids, if_present_ids=None):
3368
"""Walk out from revision_ids in source to revisions target has.
3370
:param revision_ids: The start point for the search.
3371
:return: A set of revision ids.
3373
target_graph = self.target.get_graph()
3374
revision_ids = frozenset(revision_ids)
3376
all_wanted_revs = revision_ids.union(if_present_ids)
3378
all_wanted_revs = revision_ids
3379
missing_revs = set()
3380
source_graph = self.source.get_graph()
3381
# ensure we don't pay silly lookup costs.
3382
searcher = source_graph._make_breadth_first_searcher(all_wanted_revs)
3383
null_set = frozenset([_mod_revision.NULL_REVISION])
3384
searcher_exhausted = False
3388
# Iterate the searcher until we have enough next_revs
3389
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3391
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3392
next_revs.update(next_revs_part)
3393
ghosts.update(ghosts_part)
3394
except StopIteration:
3395
searcher_exhausted = True
3397
# If there are ghosts in the source graph, and the caller asked for
3398
# them, make sure that they are present in the target.
3399
# We don't care about other ghosts as we can't fetch them and
3400
# haven't been asked to.
3401
ghosts_to_check = set(revision_ids.intersection(ghosts))
3402
revs_to_get = set(next_revs).union(ghosts_to_check)
3404
have_revs = set(target_graph.get_parent_map(revs_to_get))
3405
# we always have NULL_REVISION present.
3406
have_revs = have_revs.union(null_set)
3407
# Check if the target is missing any ghosts we need.
3408
ghosts_to_check.difference_update(have_revs)
3410
# One of the caller's revision_ids is a ghost in both the
3411
# source and the target.
3412
raise errors.NoSuchRevision(
3413
self.source, ghosts_to_check.pop())
3414
missing_revs.update(next_revs - have_revs)
3415
# Because we may have walked past the original stop point, make
3416
# sure everything is stopped
3417
stop_revs = searcher.find_seen_ancestors(have_revs)
3418
searcher.stop_searching_any(stop_revs)
3419
if searcher_exhausted:
3421
return searcher.get_result()
3424
def search_missing_revision_ids(self,
3425
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
3426
find_ghosts=True, revision_ids=None, if_present_ids=None):
3427
"""Return the revision ids that source has that target does not.
3429
:param revision_id: only return revision ids included by this
3431
:param revision_ids: return revision ids included by these
3432
revision_ids. NoSuchRevision will be raised if any of these
3433
revisions are not present.
3434
:param if_present_ids: like revision_ids, but will not cause
3435
NoSuchRevision if any of these are absent, instead they will simply
3436
not be in the result. This is useful for e.g. finding revisions
3437
to fetch for tags, which may reference absent revisions.
3438
:param find_ghosts: If True find missing revisions in deep history
3439
rather than just finding the surface difference.
3440
:return: A bzrlib.graph.SearchResult.
3442
if symbol_versioning.deprecated_passed(revision_id):
3443
symbol_versioning.warn(
3444
'search_missing_revision_ids(revision_id=...) was '
3445
'deprecated in 2.4. Use revision_ids=[...] instead.',
3446
DeprecationWarning, stacklevel=2)
3447
if revision_ids is not None:
3448
raise AssertionError(
3449
'revision_ids is mutually exclusive with revision_id')
3450
if revision_id is not None:
3451
revision_ids = [revision_id]
3453
# stop searching at found target revisions.
3454
if not find_ghosts and (revision_ids is not None or if_present_ids is
3456
return self._walk_to_common_revisions(revision_ids,
3457
if_present_ids=if_present_ids)
3458
# generic, possibly worst case, slow code path.
3459
target_ids = set(self.target.all_revision_ids())
3460
source_ids = self._present_source_revisions_for(
3461
revision_ids, if_present_ids)
3462
result_set = set(source_ids).difference(target_ids)
3463
return self.source.revision_ids_to_search_result(result_set)
3465
def _present_source_revisions_for(self, revision_ids, if_present_ids=None):
3466
"""Returns set of all revisions in ancestry of revision_ids present in
3469
:param revision_ids: if None, all revisions in source are returned.
3470
:param if_present_ids: like revision_ids, but if any/all of these are
3471
absent no error is raised.
3473
if revision_ids is not None or if_present_ids is not None:
3474
# First, ensure all specified revisions exist. Callers expect
3475
# NoSuchRevision when they pass absent revision_ids here.
3476
if revision_ids is None:
3477
revision_ids = set()
3478
if if_present_ids is None:
3479
if_present_ids = set()
3480
revision_ids = set(revision_ids)
3481
if_present_ids = set(if_present_ids)
3482
all_wanted_ids = revision_ids.union(if_present_ids)
3483
graph = self.source.get_graph()
3484
present_revs = set(graph.get_parent_map(all_wanted_ids))
3485
missing = revision_ids.difference(present_revs)
3487
raise errors.NoSuchRevision(self.source, missing.pop())
3488
found_ids = all_wanted_ids.intersection(present_revs)
3489
source_ids = [rev_id for (rev_id, parents) in
3490
graph.iter_ancestry(found_ids)
3491
if rev_id != _mod_revision.NULL_REVISION
3492
and parents is not None]
3494
source_ids = self.source.all_revision_ids()
3495
return set(source_ids)
3498
def _same_model(source, target):
3499
"""True if source and target have the same data representation.
3501
Note: this is always called on the base class; overriding it in a
3502
subclass will have no effect.
3505
InterRepository._assert_same_model(source, target)
3507
except errors.IncompatibleRepositories, e:
3511
def _assert_same_model(source, target):
3512
"""Raise an exception if two repositories do not use the same model.
3514
if source.supports_rich_root() != target.supports_rich_root():
3515
raise errors.IncompatibleRepositories(source, target,
3516
"different rich-root support")
3517
if source._serializer != target._serializer:
3518
raise errors.IncompatibleRepositories(source, target,
3519
"different serializers")
3522
class InterSameDataRepository(InterRepository):
3523
"""Code for converting between repositories that represent the same data.
3525
Data format and model must match for this to work.
3529
def _get_repo_format_to_test(self):
3530
"""Repository format for testing with.
3532
InterSameData can pull from subtree to subtree and from non-subtree to
3533
non-subtree, so we test this with the richest repository format.
3535
from bzrlib.repofmt import knitrepo
3536
return knitrepo.RepositoryFormatKnit3()
3539
def is_compatible(source, target):
3540
return InterRepository._same_model(source, target)
3543
class InterDifferingSerializer(InterRepository):
3546
def _get_repo_format_to_test(self):
3550
def is_compatible(source, target):
3551
"""Be compatible with Knit2 source and Knit3 target"""
3552
# This is redundant with format.check_conversion_target(), however that
3553
# raises an exception, and we just want to say "False" as in we won't
3554
# support converting between these formats.
3555
if 'IDS_never' in debug.debug_flags:
3557
if source.supports_rich_root() and not target.supports_rich_root():
3559
if (source._format.supports_tree_reference
3560
and not target._format.supports_tree_reference):
3562
if target._fallback_repositories and target._format.supports_chks:
3563
# IDS doesn't know how to copy CHKs for the parent inventories it
3564
# adds to stacked repos.
3566
if 'IDS_always' in debug.debug_flags:
3568
# Only use this code path for local source and target. IDS does far
3569
# too much IO (both bandwidth and roundtrips) over a network.
3570
if not source.bzrdir.transport.base.startswith('file:///'):
3572
if not target.bzrdir.transport.base.startswith('file:///'):
3576
def _get_trees(self, revision_ids, cache):
3578
for rev_id in revision_ids:
3580
possible_trees.append((rev_id, cache[rev_id]))
3582
# Not cached, but inventory might be present anyway.
3584
tree = self.source.revision_tree(rev_id)
3585
except errors.NoSuchRevision:
3586
# Nope, parent is ghost.
3589
cache[rev_id] = tree
3590
possible_trees.append((rev_id, tree))
3591
return possible_trees
3593
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3594
"""Get the best delta and base for this revision.
3596
:return: (basis_id, delta)
3599
# Generate deltas against each tree, to find the shortest.
3600
texts_possibly_new_in_tree = set()
3601
for basis_id, basis_tree in possible_trees:
3602
delta = tree.inventory._make_delta(basis_tree.inventory)
3603
for old_path, new_path, file_id, new_entry in delta:
3604
if new_path is None:
3605
# This file_id isn't present in the new rev, so we don't
3609
# Rich roots are handled elsewhere...
3611
kind = new_entry.kind
3612
if kind != 'directory' and kind != 'file':
3613
# No text record associated with this inventory entry.
3615
# This is a directory or file that has changed somehow.
3616
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3617
deltas.append((len(delta), basis_id, delta))
3619
return deltas[0][1:]
3621
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3622
"""Find all parent revisions that are absent, but for which the
3623
inventory is present, and copy those inventories.
3625
This is necessary to preserve correctness when the source is stacked
3626
without fallbacks configured. (Note that in cases like upgrade the
3627
source may be not have _fallback_repositories even though it is
3631
for parents in parent_map.values():
3632
parent_revs.update(parents)
3633
present_parents = self.source.get_parent_map(parent_revs)
3634
absent_parents = set(parent_revs).difference(present_parents)
3635
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3636
(rev_id,) for rev_id in absent_parents)
3637
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3638
for parent_tree in self.source.revision_trees(parent_inv_ids):
3639
current_revision_id = parent_tree.get_revision_id()
3640
parents_parents_keys = parent_invs_keys_for_stacking[
3641
(current_revision_id,)]
3642
parents_parents = [key[-1] for key in parents_parents_keys]
3643
basis_id = _mod_revision.NULL_REVISION
3644
basis_tree = self.source.revision_tree(basis_id)
3645
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3646
self.target.add_inventory_by_delta(
3647
basis_id, delta, current_revision_id, parents_parents)
3648
cache[current_revision_id] = parent_tree
3650
def _fetch_batch(self, revision_ids, basis_id, cache):
3651
"""Fetch across a few revisions.
3653
:param revision_ids: The revisions to copy
3654
:param basis_id: The revision_id of a tree that must be in cache, used
3655
as a basis for delta when no other base is available
3656
:param cache: A cache of RevisionTrees that we can use.
3657
:return: The revision_id of the last converted tree. The RevisionTree
3658
for it will be in cache
3660
# Walk though all revisions; get inventory deltas, copy referenced
3661
# texts that delta references, insert the delta, revision and
3663
root_keys_to_create = set()
3666
pending_revisions = []
3667
parent_map = self.source.get_parent_map(revision_ids)
3668
self._fetch_parent_invs_for_stacking(parent_map, cache)
3669
self.source._safe_to_return_from_cache = True
3670
for tree in self.source.revision_trees(revision_ids):
3671
# Find a inventory delta for this revision.
3672
# Find text entries that need to be copied, too.
3673
current_revision_id = tree.get_revision_id()
3674
parent_ids = parent_map.get(current_revision_id, ())
3675
parent_trees = self._get_trees(parent_ids, cache)
3676
possible_trees = list(parent_trees)
3677
if len(possible_trees) == 0:
3678
# There either aren't any parents, or the parents are ghosts,
3679
# so just use the last converted tree.
3680
possible_trees.append((basis_id, cache[basis_id]))
3681
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3683
revision = self.source.get_revision(current_revision_id)
3684
pending_deltas.append((basis_id, delta,
3685
current_revision_id, revision.parent_ids))
3686
if self._converting_to_rich_root:
3687
self._revision_id_to_root_id[current_revision_id] = \
3689
# Determine which texts are in present in this revision but not in
3690
# any of the available parents.
3691
texts_possibly_new_in_tree = set()
3692
for old_path, new_path, file_id, entry in delta:
3693
if new_path is None:
3694
# This file_id isn't present in the new rev
3698
if not self.target.supports_rich_root():
3699
# The target doesn't support rich root, so we don't
3702
if self._converting_to_rich_root:
3703
# This can't be copied normally, we have to insert
3705
root_keys_to_create.add((file_id, entry.revision))
3708
texts_possibly_new_in_tree.add((file_id, entry.revision))
3709
for basis_id, basis_tree in possible_trees:
3710
basis_inv = basis_tree.inventory
3711
for file_key in list(texts_possibly_new_in_tree):
3712
file_id, file_revision = file_key
3714
entry = basis_inv[file_id]
3715
except errors.NoSuchId:
3717
if entry.revision == file_revision:
3718
texts_possibly_new_in_tree.remove(file_key)
3719
text_keys.update(texts_possibly_new_in_tree)
3720
pending_revisions.append(revision)
3721
cache[current_revision_id] = tree
3722
basis_id = current_revision_id
3723
self.source._safe_to_return_from_cache = False
3725
from_texts = self.source.texts
3726
to_texts = self.target.texts
3727
if root_keys_to_create:
3728
root_stream = _mod_fetch._new_root_data_stream(
3729
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3731
to_texts.insert_record_stream(root_stream)
3732
to_texts.insert_record_stream(from_texts.get_record_stream(
3733
text_keys, self.target._format._fetch_order,
3734
not self.target._format._fetch_uses_deltas))
3735
# insert inventory deltas
3736
for delta in pending_deltas:
3737
self.target.add_inventory_by_delta(*delta)
3738
if self.target._fallback_repositories:
3739
# Make sure this stacked repository has all the parent inventories
3740
# for the new revisions that we are about to insert. We do this
3741
# before adding the revisions so that no revision is added until
3742
# all the inventories it may depend on are added.
3743
# Note that this is overzealous, as we may have fetched these in an
3746
revision_ids = set()
3747
for revision in pending_revisions:
3748
revision_ids.add(revision.revision_id)
3749
parent_ids.update(revision.parent_ids)
3750
parent_ids.difference_update(revision_ids)
3751
parent_ids.discard(_mod_revision.NULL_REVISION)
3752
parent_map = self.source.get_parent_map(parent_ids)
3753
# we iterate over parent_map and not parent_ids because we don't
3754
# want to try copying any revision which is a ghost
3755
for parent_tree in self.source.revision_trees(parent_map):
3756
current_revision_id = parent_tree.get_revision_id()
3757
parents_parents = parent_map[current_revision_id]
3758
possible_trees = self._get_trees(parents_parents, cache)
3759
if len(possible_trees) == 0:
3760
# There either aren't any parents, or the parents are
3761
# ghosts, so just use the last converted tree.
3762
possible_trees.append((basis_id, cache[basis_id]))
3763
basis_id, delta = self._get_delta_for_revision(parent_tree,
3764
parents_parents, possible_trees)
3765
self.target.add_inventory_by_delta(
3766
basis_id, delta, current_revision_id, parents_parents)
3767
# insert signatures and revisions
3768
for revision in pending_revisions:
3770
signature = self.source.get_signature_text(
3771
revision.revision_id)
3772
self.target.add_signature_text(revision.revision_id,
3774
except errors.NoSuchRevision:
3776
self.target.add_revision(revision.revision_id, revision)
3779
def _fetch_all_revisions(self, revision_ids, pb):
3780
"""Fetch everything for the list of revisions.
3782
:param revision_ids: The list of revisions to fetch. Must be in
3784
:param pb: A ProgressTask
3787
basis_id, basis_tree = self._get_basis(revision_ids[0])
3789
cache = lru_cache.LRUCache(100)
3790
cache[basis_id] = basis_tree
3791
del basis_tree # We don't want to hang on to it here
3795
for offset in range(0, len(revision_ids), batch_size):
3796
self.target.start_write_group()
3798
pb.update('Transferring revisions', offset,
3800
batch = revision_ids[offset:offset+batch_size]
3801
basis_id = self._fetch_batch(batch, basis_id, cache)
3803
self.source._safe_to_return_from_cache = False
3804
self.target.abort_write_group()
3807
hint = self.target.commit_write_group()
3810
if hints and self.target._format.pack_compresses:
3811
self.target.pack(hint=hints)
3812
pb.update('Transferring revisions', len(revision_ids),
3816
def fetch(self, revision_id=None, find_ghosts=False,
3818
"""See InterRepository.fetch()."""
3819
if fetch_spec is not None:
3820
revision_ids = fetch_spec.get_keys()
3823
ui.ui_factory.warn_experimental_format_fetch(self)
3824
if (not self.source.supports_rich_root()
3825
and self.target.supports_rich_root()):
3826
self._converting_to_rich_root = True
3827
self._revision_id_to_root_id = {}
3829
self._converting_to_rich_root = False
3830
# See <https://launchpad.net/bugs/456077> asking for a warning here
3831
if self.source._format.network_name() != self.target._format.network_name():
3832
ui.ui_factory.show_user_warning('cross_format_fetch',
3833
from_format=self.source._format,
3834
to_format=self.target._format)
3835
if revision_ids is None:
3837
search_revision_ids = [revision_id]
3839
search_revision_ids = None
3840
revision_ids = self.target.search_missing_revision_ids(self.source,
3841
revision_ids=search_revision_ids,
3842
find_ghosts=find_ghosts).get_keys()
3843
if not revision_ids:
3845
revision_ids = tsort.topo_sort(
3846
self.source.get_graph().get_parent_map(revision_ids))
3847
if not revision_ids:
3849
# Walk though all revisions; get inventory deltas, copy referenced
3850
# texts that delta references, insert the delta, revision and
3852
pb = ui.ui_factory.nested_progress_bar()
3854
self._fetch_all_revisions(revision_ids, pb)
3857
return len(revision_ids), 0
3859
def _get_basis(self, first_revision_id):
3860
"""Get a revision and tree which exists in the target.
3862
This assumes that first_revision_id is selected for transmission
3863
because all other ancestors are already present. If we can't find an
3864
ancestor we fall back to NULL_REVISION since we know that is safe.
3866
:return: (basis_id, basis_tree)
3868
first_rev = self.source.get_revision(first_revision_id)
3870
basis_id = first_rev.parent_ids[0]
3871
# only valid as a basis if the target has it
3872
self.target.get_revision(basis_id)
3873
# Try to get a basis tree - if it's a ghost it will hit the
3874
# NoSuchRevision case.
3875
basis_tree = self.source.revision_tree(basis_id)
3876
except (IndexError, errors.NoSuchRevision):
3877
basis_id = _mod_revision.NULL_REVISION
3878
basis_tree = self.source.revision_tree(basis_id)
3879
return basis_id, basis_tree
3882
InterRepository.register_optimiser(InterDifferingSerializer)
3883
InterRepository.register_optimiser(InterSameDataRepository)
3886
class CopyConverter(object):
3887
"""A repository conversion tool which just performs a copy of the content.
3889
This is slow but quite reliable.
3892
def __init__(self, target_format):
3893
"""Create a CopyConverter.
3895
:param target_format: The format the resulting repository should be.
3897
self.target_format = target_format
3899
def convert(self, repo, pb):
3900
"""Perform the conversion of to_convert, giving feedback via pb.
3902
:param to_convert: The disk object to convert.
3903
:param pb: a progress bar to use for progress information.
3905
pb = ui.ui_factory.nested_progress_bar()
3908
# this is only useful with metadir layouts - separated repo content.
3909
# trigger an assertion if not such
3910
repo._format.get_format_string()
3911
self.repo_dir = repo.bzrdir
3912
pb.update('Moving repository to repository.backup')
3913
self.repo_dir.transport.move('repository', 'repository.backup')
3914
backup_transport = self.repo_dir.transport.clone('repository.backup')
3915
repo._format.check_conversion_target(self.target_format)
3916
self.source_repo = repo._format.open(self.repo_dir,
3918
_override_transport=backup_transport)
3919
pb.update('Creating new repository')
3920
converted = self.target_format.initialize(self.repo_dir,
3921
self.source_repo.is_shared())
3922
converted.lock_write()
3924
pb.update('Copying content')
3925
self.source_repo.copy_content_into(converted)
3928
pb.update('Deleting old repository content')
3929
self.repo_dir.transport.delete_tree('repository.backup')
3930
ui.ui_factory.note('repository converted')
3934
class _VersionedFileChecker(object):
3936
def __init__(self, repository, text_key_references=None, ancestors=None):
3937
self.repository = repository
3938
self.text_index = self.repository._generate_text_key_index(
3939
text_key_references=text_key_references, ancestors=ancestors)
3941
def calculate_file_version_parents(self, text_key):
3942
"""Calculate the correct parents for a file version according to
3945
parent_keys = self.text_index[text_key]
3946
if parent_keys == [_mod_revision.NULL_REVISION]:
3948
return tuple(parent_keys)
3950
def check_file_version_parents(self, texts, progress_bar=None):
3951
"""Check the parents stored in a versioned file are correct.
3953
It also detects file versions that are not referenced by their
3954
corresponding revision's inventory.
3956
:returns: A tuple of (wrong_parents, dangling_file_versions).
3957
wrong_parents is a dict mapping {revision_id: (stored_parents,
3958
correct_parents)} for each revision_id where the stored parents
3959
are not correct. dangling_file_versions is a set of (file_id,
3960
revision_id) tuples for versions that are present in this versioned
3961
file, but not used by the corresponding inventory.
3963
local_progress = None
3964
if progress_bar is None:
3965
local_progress = ui.ui_factory.nested_progress_bar()
3966
progress_bar = local_progress
3968
return self._check_file_version_parents(texts, progress_bar)
3971
local_progress.finished()
3973
def _check_file_version_parents(self, texts, progress_bar):
3974
"""See check_file_version_parents."""
3976
self.file_ids = set([file_id for file_id, _ in
3977
self.text_index.iterkeys()])
3978
# text keys is now grouped by file_id
3979
n_versions = len(self.text_index)
3980
progress_bar.update('loading text store', 0, n_versions)
3981
parent_map = self.repository.texts.get_parent_map(self.text_index)
3982
# On unlistable transports this could well be empty/error...
3983
text_keys = self.repository.texts.keys()
3984
unused_keys = frozenset(text_keys) - set(self.text_index)
3985
for num, key in enumerate(self.text_index.iterkeys()):
3986
progress_bar.update('checking text graph', num, n_versions)
3987
correct_parents = self.calculate_file_version_parents(key)
3989
knit_parents = parent_map[key]
3990
except errors.RevisionNotPresent:
3993
if correct_parents != knit_parents:
3994
wrong_parents[key] = (knit_parents, correct_parents)
3995
return wrong_parents, unused_keys
3998
def _strip_NULL_ghosts(revision_graph):
3999
"""Also don't use this. more compatibility code for unmigrated clients."""
4000
# Filter ghosts, and null:
4001
if _mod_revision.NULL_REVISION in revision_graph:
4002
del revision_graph[_mod_revision.NULL_REVISION]
4003
for key, parents in revision_graph.items():
4004
revision_graph[key] = tuple(parent for parent in parents if parent
4006
return revision_graph
4009
class StreamSink(object):
4010
"""An object that can insert a stream into a repository.
4012
This interface handles the complexity of reserialising inventories and
4013
revisions from different formats, and allows unidirectional insertion into
4014
stacked repositories without looking for the missing basis parents
4018
def __init__(self, target_repo):
4019
self.target_repo = target_repo
4021
def insert_stream(self, stream, src_format, resume_tokens):
4022
"""Insert a stream's content into the target repository.
4024
:param src_format: a bzr repository format.
4026
:return: a list of resume tokens and an iterable of keys additional
4027
items required before the insertion can be completed.
4029
self.target_repo.lock_write()
4032
self.target_repo.resume_write_group(resume_tokens)
4035
self.target_repo.start_write_group()
4038
# locked_insert_stream performs a commit|suspend.
4039
missing_keys = self.insert_stream_without_locking(stream,
4040
src_format, is_resume)
4042
# suspend the write group and tell the caller what we is
4043
# missing. We know we can suspend or else we would not have
4044
# entered this code path. (All repositories that can handle
4045
# missing keys can handle suspending a write group).
4046
write_group_tokens = self.target_repo.suspend_write_group()
4047
return write_group_tokens, missing_keys
4048
hint = self.target_repo.commit_write_group()
4049
to_serializer = self.target_repo._format._serializer
4050
src_serializer = src_format._serializer
4051
if (to_serializer != src_serializer and
4052
self.target_repo._format.pack_compresses):
4053
self.target_repo.pack(hint=hint)
4056
self.target_repo.abort_write_group(suppress_errors=True)
4059
self.target_repo.unlock()
4061
def insert_stream_without_locking(self, stream, src_format,
4063
"""Insert a stream's content into the target repository.
4065
This assumes that you already have a locked repository and an active
4068
:param src_format: a bzr repository format.
4069
:param is_resume: Passed down to get_missing_parent_inventories to
4070
indicate if we should be checking for missing texts at the same
4073
:return: A set of keys that are missing.
4075
if not self.target_repo.is_write_locked():
4076
raise errors.ObjectNotLocked(self)
4077
if not self.target_repo.is_in_write_group():
4078
raise errors.BzrError('you must already be in a write group')
4079
to_serializer = self.target_repo._format._serializer
4080
src_serializer = src_format._serializer
4082
if to_serializer == src_serializer:
4083
# If serializers match and the target is a pack repository, set the
4084
# write cache size on the new pack. This avoids poor performance
4085
# on transports where append is unbuffered (such as
4086
# RemoteTransport). This is safe to do because nothing should read
4087
# back from the target repository while a stream with matching
4088
# serialization is being inserted.
4089
# The exception is that a delta record from the source that should
4090
# be a fulltext may need to be expanded by the target (see
4091
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4092
# explicitly flush any buffered writes first in that rare case.
4094
new_pack = self.target_repo._pack_collection._new_pack
4095
except AttributeError:
4096
# Not a pack repository
4099
new_pack.set_write_cache_size(1024*1024)
4100
for substream_type, substream in stream:
4101
if 'stream' in debug.debug_flags:
4102
mutter('inserting substream: %s', substream_type)
4103
if substream_type == 'texts':
4104
self.target_repo.texts.insert_record_stream(substream)
4105
elif substream_type == 'inventories':
4106
if src_serializer == to_serializer:
4107
self.target_repo.inventories.insert_record_stream(
4110
self._extract_and_insert_inventories(
4111
substream, src_serializer)
4112
elif substream_type == 'inventory-deltas':
4113
self._extract_and_insert_inventory_deltas(
4114
substream, src_serializer)
4115
elif substream_type == 'chk_bytes':
4116
# XXX: This doesn't support conversions, as it assumes the
4117
# conversion was done in the fetch code.
4118
self.target_repo.chk_bytes.insert_record_stream(substream)
4119
elif substream_type == 'revisions':
4120
# This may fallback to extract-and-insert more often than
4121
# required if the serializers are different only in terms of
4123
if src_serializer == to_serializer:
4124
self.target_repo.revisions.insert_record_stream(substream)
4126
self._extract_and_insert_revisions(substream,
4128
elif substream_type == 'signatures':
4129
self.target_repo.signatures.insert_record_stream(substream)
4131
raise AssertionError('kaboom! %s' % (substream_type,))
4132
# Done inserting data, and the missing_keys calculations will try to
4133
# read back from the inserted data, so flush the writes to the new pack
4134
# (if this is pack format).
4135
if new_pack is not None:
4136
new_pack._write_data('', flush=True)
4137
# Find all the new revisions (including ones from resume_tokens)
4138
missing_keys = self.target_repo.get_missing_parent_inventories(
4139
check_for_missing_texts=is_resume)
4141
for prefix, versioned_file in (
4142
('texts', self.target_repo.texts),
4143
('inventories', self.target_repo.inventories),
4144
('revisions', self.target_repo.revisions),
4145
('signatures', self.target_repo.signatures),
4146
('chk_bytes', self.target_repo.chk_bytes),
4148
if versioned_file is None:
4150
# TODO: key is often going to be a StaticTuple object
4151
# I don't believe we can define a method by which
4152
# (prefix,) + StaticTuple will work, though we could
4153
# define a StaticTuple.sq_concat that would allow you to
4154
# pass in either a tuple or a StaticTuple as the second
4155
# object, so instead we could have:
4156
# StaticTuple(prefix) + key here...
4157
missing_keys.update((prefix,) + key for key in
4158
versioned_file.get_missing_compression_parent_keys())
4159
except NotImplementedError:
4160
# cannot even attempt suspending, and missing would have failed
4161
# during stream insertion.
4162
missing_keys = set()
4165
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4166
target_rich_root = self.target_repo._format.rich_root_data
4167
target_tree_refs = self.target_repo._format.supports_tree_reference
4168
for record in substream:
4169
# Insert the delta directly
4170
inventory_delta_bytes = record.get_bytes_as('fulltext')
4171
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4173
parse_result = deserialiser.parse_text_bytes(
4174
inventory_delta_bytes)
4175
except inventory_delta.IncompatibleInventoryDelta, err:
4176
trace.mutter("Incompatible delta: %s", err.msg)
4177
raise errors.IncompatibleRevision(self.target_repo._format)
4178
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4179
revision_id = new_id
4180
parents = [key[0] for key in record.parents]
4181
self.target_repo.add_inventory_by_delta(
4182
basis_id, inv_delta, revision_id, parents)
4184
def _extract_and_insert_inventories(self, substream, serializer,
4186
"""Generate a new inventory versionedfile in target, converting data.
4188
The inventory is retrieved from the source, (deserializing it), and
4189
stored in the target (reserializing it in a different format).
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
# It's not a delta, so it must be a fulltext in the source
4195
# serializer's format.
4196
bytes = record.get_bytes_as('fulltext')
4197
revision_id = record.key[0]
4198
inv = serializer.read_inventory_from_string(bytes, revision_id)
4199
parents = [key[0] for key in record.parents]
4200
self.target_repo.add_inventory(revision_id, inv, parents)
4201
# No need to keep holding this full inv in memory when the rest of
4202
# the substream is likely to be all deltas.
4205
def _extract_and_insert_revisions(self, substream, serializer):
4206
for record in substream:
4207
bytes = record.get_bytes_as('fulltext')
4208
revision_id = record.key[0]
4209
rev = serializer.read_revision_from_string(bytes)
4210
if rev.revision_id != revision_id:
4211
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4212
self.target_repo.add_revision(revision_id, rev)
4215
if self.target_repo._format._fetch_reconcile:
4216
self.target_repo.reconcile()
4219
class StreamSource(object):
4220
"""A source of a stream for fetching between repositories."""
4222
def __init__(self, from_repository, to_format):
4223
"""Create a StreamSource streaming from from_repository."""
4224
self.from_repository = from_repository
4225
self.to_format = to_format
4226
self._record_counter = RecordCounter()
4228
def delta_on_metadata(self):
4229
"""Return True if delta's are permitted on metadata streams.
4231
That is on revisions and signatures.
4233
src_serializer = self.from_repository._format._serializer
4234
target_serializer = self.to_format._serializer
4235
return (self.to_format._fetch_uses_deltas and
4236
src_serializer == target_serializer)
4238
def _fetch_revision_texts(self, revs):
4239
# fetch signatures first and then the revision texts
4240
# may need to be a InterRevisionStore call here.
4241
from_sf = self.from_repository.signatures
4242
# A missing signature is just skipped.
4243
keys = [(rev_id,) for rev_id in revs]
4244
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4246
self.to_format._fetch_order,
4247
not self.to_format._fetch_uses_deltas))
4248
# If a revision has a delta, this is actually expanded inside the
4249
# insert_record_stream code now, which is an alternate fix for
4251
from_rf = self.from_repository.revisions
4252
revisions = from_rf.get_record_stream(
4254
self.to_format._fetch_order,
4255
not self.delta_on_metadata())
4256
return [('signatures', signatures), ('revisions', revisions)]
4258
def _generate_root_texts(self, revs):
4259
"""This will be called by get_stream between fetching weave texts and
4260
fetching the inventory weave.
4262
if self._rich_root_upgrade():
4263
return _mod_fetch.Inter1and2Helper(
4264
self.from_repository).generate_root_texts(revs)
4268
def get_stream(self, search):
4270
revs = search.get_keys()
4271
graph = self.from_repository.get_graph()
4272
revs = tsort.topo_sort(graph.get_parent_map(revs))
4273
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4275
for knit_kind, file_id, revisions in data_to_fetch:
4276
if knit_kind != phase:
4278
# Make a new progress bar for this phase
4279
if knit_kind == "file":
4280
# Accumulate file texts
4281
text_keys.extend([(file_id, revision) for revision in
4283
elif knit_kind == "inventory":
4284
# Now copy the file texts.
4285
from_texts = self.from_repository.texts
4286
yield ('texts', from_texts.get_record_stream(
4287
text_keys, self.to_format._fetch_order,
4288
not self.to_format._fetch_uses_deltas))
4289
# Cause an error if a text occurs after we have done the
4292
# Before we process the inventory we generate the root
4293
# texts (if necessary) so that the inventories references
4295
for _ in self._generate_root_texts(revs):
4297
# we fetch only the referenced inventories because we do not
4298
# know for unselected inventories whether all their required
4299
# texts are present in the other repository - it could be
4301
for info in self._get_inventory_stream(revs):
4303
elif knit_kind == "signatures":
4304
# Nothing to do here; this will be taken care of when
4305
# _fetch_revision_texts happens.
4307
elif knit_kind == "revisions":
4308
for record in self._fetch_revision_texts(revs):
4311
raise AssertionError("Unknown knit kind %r" % knit_kind)
4313
def get_stream_for_missing_keys(self, missing_keys):
4314
# missing keys can only occur when we are byte copying and not
4315
# translating (because translation means we don't send
4316
# unreconstructable deltas ever).
4318
keys['texts'] = set()
4319
keys['revisions'] = set()
4320
keys['inventories'] = set()
4321
keys['chk_bytes'] = set()
4322
keys['signatures'] = set()
4323
for key in missing_keys:
4324
keys[key[0]].add(key[1:])
4325
if len(keys['revisions']):
4326
# If we allowed copying revisions at this point, we could end up
4327
# copying a revision without copying its required texts: a
4328
# violation of the requirements for repository integrity.
4329
raise AssertionError(
4330
'cannot copy revisions to fill in missing deltas %s' % (
4331
keys['revisions'],))
4332
for substream_kind, keys in keys.iteritems():
4333
vf = getattr(self.from_repository, substream_kind)
4334
if vf is None and keys:
4335
raise AssertionError(
4336
"cannot fill in keys for a versioned file we don't"
4337
" have: %s needs %s" % (substream_kind, keys))
4339
# No need to stream something we don't have
4341
if substream_kind == 'inventories':
4342
# Some missing keys are genuinely ghosts, filter those out.
4343
present = self.from_repository.inventories.get_parent_map(keys)
4344
revs = [key[0] for key in present]
4345
# Get the inventory stream more-or-less as we do for the
4346
# original stream; there's no reason to assume that records
4347
# direct from the source will be suitable for the sink. (Think
4348
# e.g. 2a -> 1.9-rich-root).
4349
for info in self._get_inventory_stream(revs, missing=True):
4353
# Ask for full texts always so that we don't need more round trips
4354
# after this stream.
4355
# Some of the missing keys are genuinely ghosts, so filter absent
4356
# records. The Sink is responsible for doing another check to
4357
# ensure that ghosts don't introduce missing data for future
4359
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4360
self.to_format._fetch_order, True))
4361
yield substream_kind, stream
4363
def inventory_fetch_order(self):
4364
if self._rich_root_upgrade():
4365
return 'topological'
4367
return self.to_format._fetch_order
4369
def _rich_root_upgrade(self):
4370
return (not self.from_repository._format.rich_root_data and
4371
self.to_format.rich_root_data)
4373
def _get_inventory_stream(self, revision_ids, missing=False):
4374
from_format = self.from_repository._format
4375
if (from_format.supports_chks and self.to_format.supports_chks and
4376
from_format.network_name() == self.to_format.network_name()):
4377
raise AssertionError(
4378
"this case should be handled by GroupCHKStreamSource")
4379
elif 'forceinvdeltas' in debug.debug_flags:
4380
return self._get_convertable_inventory_stream(revision_ids,
4381
delta_versus_null=missing)
4382
elif from_format.network_name() == self.to_format.network_name():
4384
return self._get_simple_inventory_stream(revision_ids,
4386
elif (not from_format.supports_chks and not self.to_format.supports_chks
4387
and from_format._serializer == self.to_format._serializer):
4388
# Essentially the same format.
4389
return self._get_simple_inventory_stream(revision_ids,
4392
# Any time we switch serializations, we want to use an
4393
# inventory-delta based approach.
4394
return self._get_convertable_inventory_stream(revision_ids,
4395
delta_versus_null=missing)
4397
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4398
# NB: This currently reopens the inventory weave in source;
4399
# using a single stream interface instead would avoid this.
4400
from_weave = self.from_repository.inventories
4402
delta_closure = True
4404
delta_closure = not self.delta_on_metadata()
4405
yield ('inventories', from_weave.get_record_stream(
4406
[(rev_id,) for rev_id in revision_ids],
4407
self.inventory_fetch_order(), delta_closure))
4409
def _get_convertable_inventory_stream(self, revision_ids,
4410
delta_versus_null=False):
4411
# The two formats are sufficiently different that there is no fast
4412
# path, so we need to send just inventorydeltas, which any
4413
# sufficiently modern client can insert into any repository.
4414
# The StreamSink code expects to be able to
4415
# convert on the target, so we need to put bytes-on-the-wire that can
4416
# be converted. That means inventory deltas (if the remote is <1.19,
4417
# RemoteStreamSink will fallback to VFS to insert the deltas).
4418
yield ('inventory-deltas',
4419
self._stream_invs_as_deltas(revision_ids,
4420
delta_versus_null=delta_versus_null))
4422
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4423
"""Return a stream of inventory-deltas for the given rev ids.
4425
:param revision_ids: The list of inventories to transmit
4426
:param delta_versus_null: Don't try to find a minimal delta for this
4427
entry, instead compute the delta versus the NULL_REVISION. This
4428
effectively streams a complete inventory. Used for stuff like
4429
filling in missing parents, etc.
4431
from_repo = self.from_repository
4432
revision_keys = [(rev_id,) for rev_id in revision_ids]
4433
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4434
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4436
inventories = self.from_repository.iter_inventories(
4437
revision_ids, 'topological')
4438
format = from_repo._format
4439
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4440
inventory_cache = lru_cache.LRUCache(50)
4441
null_inventory = from_repo.revision_tree(
4442
_mod_revision.NULL_REVISION).inventory
4443
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4444
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4445
# repo back into a non-rich-root repo ought to be allowed)
4446
serializer = inventory_delta.InventoryDeltaSerializer(
4447
versioned_root=format.rich_root_data,
4448
tree_references=format.supports_tree_reference)
4449
for inv in inventories:
4450
key = (inv.revision_id,)
4451
parent_keys = parent_map.get(key, ())
4453
if not delta_versus_null and parent_keys:
4454
# The caller did not ask for complete inventories and we have
4455
# some parents that we can delta against. Make a delta against
4456
# each parent so that we can find the smallest.
4457
parent_ids = [parent_key[0] for parent_key in parent_keys]
4458
for parent_id in parent_ids:
4459
if parent_id not in invs_sent_so_far:
4460
# We don't know that the remote side has this basis, so
4463
if parent_id == _mod_revision.NULL_REVISION:
4464
parent_inv = null_inventory
4466
parent_inv = inventory_cache.get(parent_id, None)
4467
if parent_inv is None:
4468
parent_inv = from_repo.get_inventory(parent_id)
4469
candidate_delta = inv._make_delta(parent_inv)
4470
if (delta is None or
4471
len(delta) > len(candidate_delta)):
4472
delta = candidate_delta
4473
basis_id = parent_id
4475
# Either none of the parents ended up being suitable, or we
4476
# were asked to delta against NULL
4477
basis_id = _mod_revision.NULL_REVISION
4478
delta = inv._make_delta(null_inventory)
4479
invs_sent_so_far.add(inv.revision_id)
4480
inventory_cache[inv.revision_id] = inv
4481
delta_serialized = ''.join(
4482
serializer.delta_to_lines(basis_id, key[-1], delta))
4483
yield versionedfile.FulltextContentFactory(
4484
key, parent_keys, None, delta_serialized)
4487
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4488
stop_revision=None):
4489
"""Extend the partial history to include a given index
4491
If a stop_index is supplied, stop when that index has been reached.
4492
If a stop_revision is supplied, stop when that revision is
4493
encountered. Otherwise, stop when the beginning of history is
4496
:param stop_index: The index which should be present. When it is
4497
present, history extension will stop.
4498
:param stop_revision: The revision id which should be present. When
4499
it is encountered, history extension will stop.
4501
start_revision = partial_history_cache[-1]
4502
iterator = repo.iter_reverse_revision_history(start_revision)
4504
#skip the last revision in the list
4507
if (stop_index is not None and
4508
len(partial_history_cache) > stop_index):
4510
if partial_history_cache[-1] == stop_revision:
4512
revision_id = iterator.next()
4513
partial_history_cache.append(revision_id)
4514
except StopIteration:
added
removed
bzrlib/repository.py
