1
# Copyright (C) 2005, 2006, 2007, 2008, 2009 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
from bzrlib.lazy_import import lazy_import
18
lazy_import(globals(), """
37
revision as _mod_revision,
43
from bzrlib.bundle import serializer
44
from bzrlib.revisiontree import RevisionTree
45
from bzrlib.store.versioned import VersionedFileStore
46
from bzrlib.testament import Testament
49
from bzrlib.decorators import needs_read_lock, needs_write_lock
50
from bzrlib.inter import InterObject
51
from bzrlib.inventory import (
57
from bzrlib import registry
58
from bzrlib.symbol_versioning import (
61
from bzrlib.trace import (
62
log_exception_quietly, note, mutter, mutter_callsite, warning)
65
# Old formats display a warning, but only once
66
_deprecation_warning_done = False
69
class CommitBuilder(object):
70
"""Provides an interface to build up a commit.
72
This allows describing a tree to be committed without needing to
73
know the internals of the format of the repository.
76
# all clients should supply tree roots.
77
record_root_entry = True
78
# the default CommitBuilder does not manage trees whose root is versioned.
79
_versioned_root = False
81
def __init__(self, repository, parents, config, timestamp=None,
82
timezone=None, committer=None, revprops=None,
84
"""Initiate a CommitBuilder.
86
:param repository: Repository to commit to.
87
:param parents: Revision ids of the parents of the new revision.
88
:param config: Configuration to use.
89
:param timestamp: Optional timestamp recorded for commit.
90
:param timezone: Optional timezone for timestamp.
91
:param committer: Optional committer to set for commit.
92
:param revprops: Optional dictionary of revision properties.
93
:param revision_id: Optional revision id.
98
self._committer = self._config.username()
100
self._committer = committer
102
self.new_inventory = Inventory(None)
103
self._new_revision_id = revision_id
104
self.parents = parents
105
self.repository = repository
108
if revprops is not None:
109
self._validate_revprops(revprops)
110
self._revprops.update(revprops)
112
if timestamp is None:
113
timestamp = time.time()
114
# Restrict resolution to 1ms
115
self._timestamp = round(timestamp, 3)
118
self._timezone = osutils.local_time_offset()
120
self._timezone = int(timezone)
122
self._generate_revision_if_needed()
123
self.__heads = graph.HeadsCache(repository.get_graph()).heads
124
self._basis_delta = []
125
# API compatibility, older code that used CommitBuilder did not call
126
# .record_delete(), which means the delta that is computed would not be
127
# valid. Callers that will call record_delete() should call
128
# .will_record_deletes() to indicate that.
129
self._recording_deletes = False
130
# memo'd check for no-op commits.
131
self._any_changes = False
133
def any_changes(self):
134
"""Return True if any entries were changed.
136
This includes merge-only changes. It is the core for the --unchanged
139
:return: True if any changes have occured.
141
return self._any_changes
143
def _validate_unicode_text(self, text, context):
144
"""Verify things like commit messages don't have bogus characters."""
146
raise ValueError('Invalid value for %s: %r' % (context, text))
148
def _validate_revprops(self, revprops):
149
for key, value in revprops.iteritems():
150
# We know that the XML serializers do not round trip '\r'
151
# correctly, so refuse to accept them
152
if not isinstance(value, basestring):
153
raise ValueError('revision property (%s) is not a valid'
154
' (unicode) string: %r' % (key, value))
155
self._validate_unicode_text(value,
156
'revision property (%s)' % (key,))
158
def commit(self, message):
159
"""Make the actual commit.
161
:return: The revision id of the recorded revision.
163
self._validate_unicode_text(message, 'commit message')
164
rev = _mod_revision.Revision(
165
timestamp=self._timestamp,
166
timezone=self._timezone,
167
committer=self._committer,
169
inventory_sha1=self.inv_sha1,
170
revision_id=self._new_revision_id,
171
properties=self._revprops)
172
rev.parent_ids = self.parents
173
self.repository.add_revision(self._new_revision_id, rev,
174
self.new_inventory, self._config)
175
self.repository.commit_write_group()
176
return self._new_revision_id
179
"""Abort the commit that is being built.
181
self.repository.abort_write_group()
183
def revision_tree(self):
184
"""Return the tree that was just committed.
186
After calling commit() this can be called to get a RevisionTree
187
representing the newly committed tree. This is preferred to
188
calling Repository.revision_tree() because that may require
189
deserializing the inventory, while we already have a copy in
192
if self.new_inventory is None:
193
self.new_inventory = self.repository.get_inventory(
194
self._new_revision_id)
195
return RevisionTree(self.repository, self.new_inventory,
196
self._new_revision_id)
198
def finish_inventory(self):
199
"""Tell the builder that the inventory is finished.
201
:return: The inventory id in the repository, which can be used with
202
repository.get_inventory.
204
if self.new_inventory is None:
205
# an inventory delta was accumulated without creating a new
207
basis_id = self.basis_delta_revision
208
self.inv_sha1 = self.repository.add_inventory_by_delta(
209
basis_id, self._basis_delta, self._new_revision_id,
212
if self.new_inventory.root is None:
213
raise AssertionError('Root entry should be supplied to'
214
' record_entry_contents, as of bzr 0.10.')
215
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
216
self.new_inventory.revision_id = self._new_revision_id
217
self.inv_sha1 = self.repository.add_inventory(
218
self._new_revision_id,
222
return self._new_revision_id
224
def _gen_revision_id(self):
225
"""Return new revision-id."""
226
return generate_ids.gen_revision_id(self._config.username(),
229
def _generate_revision_if_needed(self):
230
"""Create a revision id if None was supplied.
232
If the repository can not support user-specified revision ids
233
they should override this function and raise CannotSetRevisionId
234
if _new_revision_id is not None.
236
:raises: CannotSetRevisionId
238
if self._new_revision_id is None:
239
self._new_revision_id = self._gen_revision_id()
240
self.random_revid = True
242
self.random_revid = False
244
def _heads(self, file_id, revision_ids):
245
"""Calculate the graph heads for revision_ids in the graph of file_id.
247
This can use either a per-file graph or a global revision graph as we
248
have an identity relationship between the two graphs.
250
return self.__heads(revision_ids)
252
def _check_root(self, ie, parent_invs, tree):
253
"""Helper for record_entry_contents.
255
:param ie: An entry being added.
256
:param parent_invs: The inventories of the parent revisions of the
258
:param tree: The tree that is being committed.
260
# In this revision format, root entries have no knit or weave When
261
# serializing out to disk and back in root.revision is always
263
ie.revision = self._new_revision_id
265
def _require_root_change(self, tree):
266
"""Enforce an appropriate root object change.
268
This is called once when record_iter_changes is called, if and only if
269
the root was not in the delta calculated by record_iter_changes.
271
:param tree: The tree which is being committed.
273
# NB: if there are no parents then this method is not called, so no
274
# need to guard on parents having length.
275
entry = entry_factory['directory'](tree.path2id(''), '',
277
entry.revision = self._new_revision_id
278
self._basis_delta.append(('', '', entry.file_id, entry))
280
def _get_delta(self, ie, basis_inv, path):
281
"""Get a delta against the basis inventory for ie."""
282
if ie.file_id not in basis_inv:
284
result = (None, path, ie.file_id, ie)
285
self._basis_delta.append(result)
287
elif ie != basis_inv[ie.file_id]:
289
# TODO: avoid tis id2path call.
290
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
291
self._basis_delta.append(result)
297
def get_basis_delta(self):
298
"""Return the complete inventory delta versus the basis inventory.
300
This has been built up with the calls to record_delete and
301
record_entry_contents. The client must have already called
302
will_record_deletes() to indicate that they will be generating a
305
:return: An inventory delta, suitable for use with apply_delta, or
306
Repository.add_inventory_by_delta, etc.
308
if not self._recording_deletes:
309
raise AssertionError("recording deletes not activated.")
310
return self._basis_delta
312
def record_delete(self, path, file_id):
313
"""Record that a delete occured against a basis tree.
315
This is an optional API - when used it adds items to the basis_delta
316
being accumulated by the commit builder. It cannot be called unless the
317
method will_record_deletes() has been called to inform the builder that
318
a delta is being supplied.
320
:param path: The path of the thing deleted.
321
:param file_id: The file id that was deleted.
323
if not self._recording_deletes:
324
raise AssertionError("recording deletes not activated.")
325
delta = (path, None, file_id, None)
326
self._basis_delta.append(delta)
327
self._any_changes = True
330
def will_record_deletes(self):
331
"""Tell the commit builder that deletes are being notified.
333
This enables the accumulation of an inventory delta; for the resulting
334
commit to be valid, deletes against the basis MUST be recorded via
335
builder.record_delete().
337
self._recording_deletes = True
339
basis_id = self.parents[0]
341
basis_id = _mod_revision.NULL_REVISION
342
self.basis_delta_revision = basis_id
344
def record_entry_contents(self, ie, parent_invs, path, tree,
346
"""Record the content of ie from tree into the commit if needed.
348
Side effect: sets ie.revision when unchanged
350
:param ie: An inventory entry present in the commit.
351
:param parent_invs: The inventories of the parent revisions of the
353
:param path: The path the entry is at in the tree.
354
:param tree: The tree which contains this entry and should be used to
356
:param content_summary: Summary data from the tree about the paths
357
content - stat, length, exec, sha/link target. This is only
358
accessed when the entry has a revision of None - that is when it is
359
a candidate to commit.
360
:return: A tuple (change_delta, version_recorded, fs_hash).
361
change_delta is an inventory_delta change for this entry against
362
the basis tree of the commit, or None if no change occured against
364
version_recorded is True if a new version of the entry has been
365
recorded. For instance, committing a merge where a file was only
366
changed on the other side will return (delta, False).
367
fs_hash is either None, or the hash details for the path (currently
368
a tuple of the contents sha1 and the statvalue returned by
369
tree.get_file_with_stat()).
371
if self.new_inventory.root is None:
372
if ie.parent_id is not None:
373
raise errors.RootMissing()
374
self._check_root(ie, parent_invs, tree)
375
if ie.revision is None:
376
kind = content_summary[0]
378
# ie is carried over from a prior commit
380
# XXX: repository specific check for nested tree support goes here - if
381
# the repo doesn't want nested trees we skip it ?
382
if (kind == 'tree-reference' and
383
not self.repository._format.supports_tree_reference):
384
# mismatch between commit builder logic and repository:
385
# this needs the entry creation pushed down into the builder.
386
raise NotImplementedError('Missing repository subtree support.')
387
self.new_inventory.add(ie)
389
# TODO: slow, take it out of the inner loop.
391
basis_inv = parent_invs[0]
393
basis_inv = Inventory(root_id=None)
395
# ie.revision is always None if the InventoryEntry is considered
396
# for committing. We may record the previous parents revision if the
397
# content is actually unchanged against a sole head.
398
if ie.revision is not None:
399
if not self._versioned_root and path == '':
400
# repositories that do not version the root set the root's
401
# revision to the new commit even when no change occurs (more
402
# specifically, they do not record a revision on the root; and
403
# the rev id is assigned to the root during deserialisation -
404
# this masks when a change may have occurred against the basis.
405
# To match this we always issue a delta, because the revision
406
# of the root will always be changing.
407
if ie.file_id in basis_inv:
408
delta = (basis_inv.id2path(ie.file_id), path,
412
delta = (None, path, ie.file_id, ie)
413
self._basis_delta.append(delta)
414
return delta, False, None
416
# we don't need to commit this, because the caller already
417
# determined that an existing revision of this file is
418
# appropriate. If its not being considered for committing then
419
# it and all its parents to the root must be unaltered so
420
# no-change against the basis.
421
if ie.revision == self._new_revision_id:
422
raise AssertionError("Impossible situation, a skipped "
423
"inventory entry (%r) claims to be modified in this "
424
"commit (%r).", (ie, self._new_revision_id))
425
return None, False, None
426
# XXX: Friction: parent_candidates should return a list not a dict
427
# so that we don't have to walk the inventories again.
428
parent_candiate_entries = ie.parent_candidates(parent_invs)
429
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
431
for inv in parent_invs:
432
if ie.file_id in inv:
433
old_rev = inv[ie.file_id].revision
434
if old_rev in head_set:
435
heads.append(inv[ie.file_id].revision)
436
head_set.remove(inv[ie.file_id].revision)
439
# now we check to see if we need to write a new record to the
441
# We write a new entry unless there is one head to the ancestors, and
442
# the kind-derived content is unchanged.
444
# Cheapest check first: no ancestors, or more the one head in the
445
# ancestors, we write a new node.
449
# There is a single head, look it up for comparison
450
parent_entry = parent_candiate_entries[heads[0]]
451
# if the non-content specific data has changed, we'll be writing a
453
if (parent_entry.parent_id != ie.parent_id or
454
parent_entry.name != ie.name):
456
# now we need to do content specific checks:
458
# if the kind changed the content obviously has
459
if kind != parent_entry.kind:
461
# Stat cache fingerprint feedback for the caller - None as we usually
462
# don't generate one.
465
if content_summary[2] is None:
466
raise ValueError("Files must not have executable = None")
468
if (# if the file length changed we have to store:
469
parent_entry.text_size != content_summary[1] or
470
# if the exec bit has changed we have to store:
471
parent_entry.executable != content_summary[2]):
473
elif parent_entry.text_sha1 == content_summary[3]:
474
# all meta and content is unchanged (using a hash cache
475
# hit to check the sha)
476
ie.revision = parent_entry.revision
477
ie.text_size = parent_entry.text_size
478
ie.text_sha1 = parent_entry.text_sha1
479
ie.executable = parent_entry.executable
480
return self._get_delta(ie, basis_inv, path), False, None
482
# Either there is only a hash change(no hash cache entry,
483
# or same size content change), or there is no change on
485
# Provide the parent's hash to the store layer, so that the
486
# content is unchanged we will not store a new node.
487
nostore_sha = parent_entry.text_sha1
489
# We want to record a new node regardless of the presence or
490
# absence of a content change in the file.
492
ie.executable = content_summary[2]
493
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
495
lines = file_obj.readlines()
499
ie.text_sha1, ie.text_size = self._add_text_to_weave(
500
ie.file_id, lines, heads, nostore_sha)
501
# Let the caller know we generated a stat fingerprint.
502
fingerprint = (ie.text_sha1, stat_value)
503
except errors.ExistingContent:
504
# Turns out that the file content was unchanged, and we were
505
# only going to store a new node if it was changed. Carry over
507
ie.revision = parent_entry.revision
508
ie.text_size = parent_entry.text_size
509
ie.text_sha1 = parent_entry.text_sha1
510
ie.executable = parent_entry.executable
511
return self._get_delta(ie, basis_inv, path), False, None
512
elif kind == 'directory':
514
# all data is meta here, nothing specific to directory, so
516
ie.revision = parent_entry.revision
517
return self._get_delta(ie, basis_inv, path), False, None
519
self._add_text_to_weave(ie.file_id, lines, heads, None)
520
elif kind == 'symlink':
521
current_link_target = content_summary[3]
523
# symlink target is not generic metadata, check if it has
525
if current_link_target != parent_entry.symlink_target:
528
# unchanged, carry over.
529
ie.revision = parent_entry.revision
530
ie.symlink_target = parent_entry.symlink_target
531
return self._get_delta(ie, basis_inv, path), False, None
532
ie.symlink_target = current_link_target
534
self._add_text_to_weave(ie.file_id, lines, heads, None)
535
elif kind == 'tree-reference':
537
if content_summary[3] != parent_entry.reference_revision:
540
# unchanged, carry over.
541
ie.reference_revision = parent_entry.reference_revision
542
ie.revision = parent_entry.revision
543
return self._get_delta(ie, basis_inv, path), False, None
544
ie.reference_revision = content_summary[3]
546
self._add_text_to_weave(ie.file_id, lines, heads, None)
548
raise NotImplementedError('unknown kind')
549
ie.revision = self._new_revision_id
550
self._any_changes = True
551
return self._get_delta(ie, basis_inv, path), True, fingerprint
553
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
554
_entry_factory=entry_factory):
555
"""Record a new tree via iter_changes.
557
:param tree: The tree to obtain text contents from for changed objects.
558
:param basis_revision_id: The revision id of the tree the iter_changes
559
has been generated against. Currently assumed to be the same
560
as self.parents[0] - if it is not, errors may occur.
561
:param iter_changes: An iter_changes iterator with the changes to apply
562
to basis_revision_id. The iterator must not include any items with
563
a current kind of None - missing items must be either filtered out
564
or errored-on beefore record_iter_changes sees the item.
565
:param _entry_factory: Private method to bind entry_factory locally for
567
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
570
# Create an inventory delta based on deltas between all the parents and
571
# deltas between all the parent inventories. We use inventory delta's
572
# between the inventory objects because iter_changes masks
573
# last-changed-field only changes.
575
# file_id -> change map, change is fileid, paths, changed, versioneds,
576
# parents, names, kinds, executables
578
# {file_id -> revision_id -> inventory entry, for entries in parent
579
# trees that are not parents[0]
583
revtrees = list(self.repository.revision_trees(self.parents))
584
except errors.NoSuchRevision:
585
# one or more ghosts, slow path.
587
for revision_id in self.parents:
589
revtrees.append(self.repository.revision_tree(revision_id))
590
except errors.NoSuchRevision:
592
basis_revision_id = _mod_revision.NULL_REVISION
594
revtrees.append(self.repository.revision_tree(
595
_mod_revision.NULL_REVISION))
596
# The basis inventory from a repository
598
basis_inv = revtrees[0].inventory
600
basis_inv = self.repository.revision_tree(
601
_mod_revision.NULL_REVISION).inventory
602
if len(self.parents) > 0:
603
if basis_revision_id != self.parents[0] and not ghost_basis:
605
"arbitrary basis parents not yet supported with merges")
606
for revtree in revtrees[1:]:
607
for change in revtree.inventory._make_delta(basis_inv):
608
if change[1] is None:
609
# Not present in this parent.
611
if change[2] not in merged_ids:
612
if change[0] is not None:
613
merged_ids[change[2]] = [
614
basis_inv[change[2]].revision,
617
merged_ids[change[2]] = [change[3].revision]
618
parent_entries[change[2]] = {change[3].revision:change[3]}
620
merged_ids[change[2]].append(change[3].revision)
621
parent_entries[change[2]][change[3].revision] = change[3]
624
# Setup the changes from the tree:
625
# changes maps file_id -> (change, [parent revision_ids])
627
for change in iter_changes:
628
# This probably looks up in basis_inv way to much.
629
if change[1][0] is not None:
630
head_candidate = [basis_inv[change[0]].revision]
633
changes[change[0]] = change, merged_ids.get(change[0],
635
unchanged_merged = set(merged_ids) - set(changes)
636
# Extend the changes dict with synthetic changes to record merges of
638
for file_id in unchanged_merged:
639
# Record a merged version of these items that did not change vs the
640
# basis. This can be either identical parallel changes, or a revert
641
# of a specific file after a merge. The recorded content will be
642
# that of the current tree (which is the same as the basis), but
643
# the per-file graph will reflect a merge.
644
# NB:XXX: We are reconstructing path information we had, this
645
# should be preserved instead.
646
# inv delta change: (file_id, (path_in_source, path_in_target),
647
# changed_content, versioned, parent, name, kind,
650
basis_entry = basis_inv[file_id]
651
except errors.NoSuchId:
652
# a change from basis->some_parents but file_id isn't in basis
653
# so was new in the merge, which means it must have changed
654
# from basis -> current, and as it hasn't the add was reverted
655
# by the user. So we discard this change.
659
(basis_inv.id2path(file_id), tree.id2path(file_id)),
661
(basis_entry.parent_id, basis_entry.parent_id),
662
(basis_entry.name, basis_entry.name),
663
(basis_entry.kind, basis_entry.kind),
664
(basis_entry.executable, basis_entry.executable))
665
changes[file_id] = (change, merged_ids[file_id])
666
# changes contains tuples with the change and a set of inventory
667
# candidates for the file.
669
# old_path, new_path, file_id, new_inventory_entry
670
seen_root = False # Is the root in the basis delta?
671
inv_delta = self._basis_delta
672
modified_rev = self._new_revision_id
673
for change, head_candidates in changes.values():
674
if change[3][1]: # versioned in target.
675
# Several things may be happening here:
676
# We may have a fork in the per-file graph
677
# - record a change with the content from tree
678
# We may have a change against < all trees
679
# - carry over the tree that hasn't changed
680
# We may have a change against all trees
681
# - record the change with the content from tree
684
entry = _entry_factory[kind](file_id, change[5][1],
686
head_set = self._heads(change[0], set(head_candidates))
689
for head_candidate in head_candidates:
690
if head_candidate in head_set:
691
heads.append(head_candidate)
692
head_set.remove(head_candidate)
695
# Could be a carry-over situation:
696
parent_entry_revs = parent_entries.get(file_id, None)
697
if parent_entry_revs:
698
parent_entry = parent_entry_revs.get(heads[0], None)
701
if parent_entry is None:
702
# The parent iter_changes was called against is the one
703
# that is the per-file head, so any change is relevant
704
# iter_changes is valid.
705
carry_over_possible = False
707
# could be a carry over situation
708
# A change against the basis may just indicate a merge,
709
# we need to check the content against the source of the
710
# merge to determine if it was changed after the merge
712
if (parent_entry.kind != entry.kind or
713
parent_entry.parent_id != entry.parent_id or
714
parent_entry.name != entry.name):
715
# Metadata common to all entries has changed
716
# against per-file parent
717
carry_over_possible = False
719
carry_over_possible = True
720
# per-type checks for changes against the parent_entry
723
# Cannot be a carry-over situation
724
carry_over_possible = False
725
# Populate the entry in the delta
727
# XXX: There is still a small race here: If someone reverts the content of a file
728
# after iter_changes examines and decides it has changed,
729
# we will unconditionally record a new version even if some
730
# other process reverts it while commit is running (with
731
# the revert happening after iter_changes did it's
734
entry.executable = True
736
entry.executable = False
737
if (carry_over_possible and
738
parent_entry.executable == entry.executable):
739
# Check the file length, content hash after reading
741
nostore_sha = parent_entry.text_sha1
744
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
746
lines = file_obj.readlines()
750
entry.text_sha1, entry.text_size = self._add_text_to_weave(
751
file_id, lines, heads, nostore_sha)
752
yield file_id, change[1][1], (entry.text_sha1, stat_value)
753
except errors.ExistingContent:
754
# No content change against a carry_over parent
755
# Perhaps this should also yield a fs hash update?
757
entry.text_size = parent_entry.text_size
758
entry.text_sha1 = parent_entry.text_sha1
759
elif kind == 'symlink':
761
entry.symlink_target = tree.get_symlink_target(file_id)
762
if (carry_over_possible and
763
parent_entry.symlink_target == entry.symlink_target):
766
self._add_text_to_weave(change[0], [], heads, None)
767
elif kind == 'directory':
768
if carry_over_possible:
771
# Nothing to set on the entry.
772
# XXX: split into the Root and nonRoot versions.
773
if change[1][1] != '' or self.repository.supports_rich_root():
774
self._add_text_to_weave(change[0], [], heads, None)
775
elif kind == 'tree-reference':
776
if not self.repository._format.supports_tree_reference:
777
# This isn't quite sane as an error, but we shouldn't
778
# ever see this code path in practice: tree's don't
779
# permit references when the repo doesn't support tree
781
raise errors.UnsupportedOperation(tree.add_reference,
783
entry.reference_revision = \
784
tree.get_reference_revision(change[0])
785
if (carry_over_possible and
786
parent_entry.reference_revision == reference_revision):
789
self._add_text_to_weave(change[0], [], heads, None)
791
raise AssertionError('unknown kind %r' % kind)
793
entry.revision = modified_rev
795
entry.revision = parent_entry.revision
798
new_path = change[1][1]
799
inv_delta.append((change[1][0], new_path, change[0], entry))
802
self.new_inventory = None
804
self._any_changes = True
806
# housekeeping root entry changes do not affect no-change commits.
807
self._require_root_change(tree)
808
self.basis_delta_revision = basis_revision_id
810
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
811
# Note: as we read the content directly from the tree, we know its not
812
# been turned into unicode or badly split - but a broken tree
813
# implementation could give us bad output from readlines() so this is
814
# not a guarantee of safety. What would be better is always checking
815
# the content during test suite execution. RBC 20070912
816
parent_keys = tuple((file_id, parent) for parent in parents)
817
return self.repository.texts.add_lines(
818
(file_id, self._new_revision_id), parent_keys, new_lines,
819
nostore_sha=nostore_sha, random_id=self.random_revid,
820
check_content=False)[0:2]
823
class RootCommitBuilder(CommitBuilder):
824
"""This commitbuilder actually records the root id"""
826
# the root entry gets versioned properly by this builder.
827
_versioned_root = True
829
def _check_root(self, ie, parent_invs, tree):
830
"""Helper for record_entry_contents.
832
:param ie: An entry being added.
833
:param parent_invs: The inventories of the parent revisions of the
835
:param tree: The tree that is being committed.
838
def _require_root_change(self, tree):
839
"""Enforce an appropriate root object change.
841
This is called once when record_iter_changes is called, if and only if
842
the root was not in the delta calculated by record_iter_changes.
844
:param tree: The tree which is being committed.
846
# versioned roots do not change unless the tree found a change.
849
######################################################################
852
class Repository(object):
853
"""Repository holding history for one or more branches.
855
The repository holds and retrieves historical information including
856
revisions and file history. It's normally accessed only by the Branch,
857
which views a particular line of development through that history.
859
The Repository builds on top of some byte storage facilies (the revisions,
860
signatures, inventories and texts attributes) and a Transport, which
861
respectively provide byte storage and a means to access the (possibly
864
The byte storage facilities are addressed via tuples, which we refer to
865
as 'keys' throughout the code base. Revision_keys, inventory_keys and
866
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
867
(file_id, revision_id). We use this interface because it allows low
868
friction with the underlying code that implements disk indices, network
869
encoding and other parts of bzrlib.
871
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
872
the serialised revisions for the repository. This can be used to obtain
873
revision graph information or to access raw serialised revisions.
874
The result of trying to insert data into the repository via this store
875
is undefined: it should be considered read-only except for implementors
877
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
878
the serialised signatures for the repository. This can be used to
879
obtain access to raw serialised signatures. The result of trying to
880
insert data into the repository via this store is undefined: it should
881
be considered read-only except for implementors of repositories.
882
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
883
the serialised inventories for the repository. This can be used to
884
obtain unserialised inventories. The result of trying to insert data
885
into the repository via this store is undefined: it should be
886
considered read-only except for implementors of repositories.
887
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
888
texts of files and directories for the repository. This can be used to
889
obtain file texts or file graphs. Note that Repository.iter_file_bytes
890
is usually a better interface for accessing file texts.
891
The result of trying to insert data into the repository via this store
892
is undefined: it should be considered read-only except for implementors
894
:ivar _transport: Transport for file access to repository, typically
895
pointing to .bzr/repository.
898
# What class to use for a CommitBuilder. Often its simpler to change this
899
# in a Repository class subclass rather than to override
900
# get_commit_builder.
901
_commit_builder_class = CommitBuilder
902
# The search regex used by xml based repositories to determine what things
903
# where changed in a single commit.
904
_file_ids_altered_regex = lazy_regex.lazy_compile(
905
r'file_id="(?P<file_id>[^"]+)"'
906
r'.* revision="(?P<revision_id>[^"]+)"'
909
def abort_write_group(self, suppress_errors=False):
910
"""Commit the contents accrued within the current write group.
912
:param suppress_errors: if true, abort_write_group will catch and log
913
unexpected errors that happen during the abort, rather than
914
allowing them to propagate. Defaults to False.
916
:seealso: start_write_group.
918
if self._write_group is not self.get_transaction():
919
# has an unlock or relock occured ?
920
raise errors.BzrError('mismatched lock context and write group.')
922
self._abort_write_group()
923
except Exception, exc:
924
self._write_group = None
925
if not suppress_errors:
927
mutter('abort_write_group failed')
928
log_exception_quietly()
929
note('bzr: ERROR (ignored): %s', exc)
930
self._write_group = None
932
def _abort_write_group(self):
933
"""Template method for per-repository write group cleanup.
935
This is called during abort before the write group is considered to be
936
finished and should cleanup any internal state accrued during the write
937
group. There is no requirement that data handed to the repository be
938
*not* made available - this is not a rollback - but neither should any
939
attempt be made to ensure that data added is fully commited. Abort is
940
invoked when an error has occured so futher disk or network operations
941
may not be possible or may error and if possible should not be
945
def add_fallback_repository(self, repository):
946
"""Add a repository to use for looking up data not held locally.
948
:param repository: A repository.
950
if not self._format.supports_external_lookups:
951
raise errors.UnstackableRepositoryFormat(self._format, self.base)
952
self._check_fallback_repository(repository)
953
self._fallback_repositories.append(repository)
954
self.texts.add_fallback_versioned_files(repository.texts)
955
self.inventories.add_fallback_versioned_files(repository.inventories)
956
self.revisions.add_fallback_versioned_files(repository.revisions)
957
self.signatures.add_fallback_versioned_files(repository.signatures)
959
def _check_fallback_repository(self, repository):
960
"""Check that this repository can fallback to repository safely.
962
Raise an error if not.
964
:param repository: A repository to fallback to.
966
return InterRepository._assert_same_model(self, repository)
968
def add_inventory(self, revision_id, inv, parents):
969
"""Add the inventory inv to the repository as revision_id.
971
:param parents: The revision ids of the parents that revision_id
972
is known to have and are in the repository already.
974
:returns: The validator(which is a sha1 digest, though what is sha'd is
975
repository format specific) of the serialized inventory.
977
if not self.is_in_write_group():
978
raise AssertionError("%r not in write group" % (self,))
979
_mod_revision.check_not_reserved_id(revision_id)
980
if not (inv.revision_id is None or inv.revision_id == revision_id):
981
raise AssertionError(
982
"Mismatch between inventory revision"
983
" id and insertion revid (%r, %r)"
984
% (inv.revision_id, revision_id))
986
raise AssertionError()
987
inv_lines = self._serialise_inventory_to_lines(inv)
988
return self._inventory_add_lines(revision_id, parents,
989
inv_lines, check_content=False)
991
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
993
"""Add a new inventory expressed as a delta against another revision.
995
:param basis_revision_id: The inventory id the delta was created
996
against. (This does not have to be a direct parent.)
997
:param delta: The inventory delta (see Inventory.apply_delta for
999
:param new_revision_id: The revision id that the inventory is being
1001
:param parents: The revision ids of the parents that revision_id is
1002
known to have and are in the repository already. These are supplied
1003
for repositories that depend on the inventory graph for revision
1004
graph access, as well as for those that pun ancestry with delta
1007
:returns: (validator, new_inv)
1008
The validator(which is a sha1 digest, though what is sha'd is
1009
repository format specific) of the serialized inventory, and the
1010
resulting inventory.
1012
if not self.is_in_write_group():
1013
raise AssertionError("%r not in write group" % (self,))
1014
_mod_revision.check_not_reserved_id(new_revision_id)
1015
basis_tree = self.revision_tree(basis_revision_id)
1016
basis_tree.lock_read()
1018
# Note that this mutates the inventory of basis_tree, which not all
1019
# inventory implementations may support: A better idiom would be to
1020
# return a new inventory, but as there is no revision tree cache in
1021
# repository this is safe for now - RBC 20081013
1022
basis_inv = basis_tree.inventory
1023
basis_inv.apply_delta(delta)
1024
basis_inv.revision_id = new_revision_id
1025
return (self.add_inventory(new_revision_id, basis_inv, parents),
1030
def _inventory_add_lines(self, revision_id, parents, lines,
1031
check_content=True):
1032
"""Store lines in inv_vf and return the sha1 of the inventory."""
1033
parents = [(parent,) for parent in parents]
1034
return self.inventories.add_lines((revision_id,), parents, lines,
1035
check_content=check_content)[0]
1037
def add_revision(self, revision_id, rev, inv=None, config=None):
1038
"""Add rev to the revision store as revision_id.
1040
:param revision_id: the revision id to use.
1041
:param rev: The revision object.
1042
:param inv: The inventory for the revision. if None, it will be looked
1043
up in the inventory storer
1044
:param config: If None no digital signature will be created.
1045
If supplied its signature_needed method will be used
1046
to determine if a signature should be made.
1048
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1050
_mod_revision.check_not_reserved_id(revision_id)
1051
if config is not None and config.signature_needed():
1053
inv = self.get_inventory(revision_id)
1054
plaintext = Testament(rev, inv).as_short_text()
1055
self.store_revision_signature(
1056
gpg.GPGStrategy(config), plaintext, revision_id)
1057
# check inventory present
1058
if not self.inventories.get_parent_map([(revision_id,)]):
1060
raise errors.WeaveRevisionNotPresent(revision_id,
1063
# yes, this is not suitable for adding with ghosts.
1064
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1067
key = (revision_id,)
1068
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1069
self._add_revision(rev)
1071
def _add_revision(self, revision):
1072
text = self._serializer.write_revision_to_string(revision)
1073
key = (revision.revision_id,)
1074
parents = tuple((parent,) for parent in revision.parent_ids)
1075
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1077
def all_revision_ids(self):
1078
"""Returns a list of all the revision ids in the repository.
1080
This is conceptually deprecated because code should generally work on
1081
the graph reachable from a particular revision, and ignore any other
1082
revisions that might be present. There is no direct replacement
1085
if 'evil' in debug.debug_flags:
1086
mutter_callsite(2, "all_revision_ids is linear with history.")
1087
return self._all_revision_ids()
1089
def _all_revision_ids(self):
1090
"""Returns a list of all the revision ids in the repository.
1092
These are in as much topological order as the underlying store can
1095
raise NotImplementedError(self._all_revision_ids)
1097
def break_lock(self):
1098
"""Break a lock if one is present from another instance.
1100
Uses the ui factory to ask for confirmation if the lock may be from
1103
self.control_files.break_lock()
1106
def _eliminate_revisions_not_present(self, revision_ids):
1107
"""Check every revision id in revision_ids to see if we have it.
1109
Returns a set of the present revisions.
1112
graph = self.get_graph()
1113
parent_map = graph.get_parent_map(revision_ids)
1114
# The old API returned a list, should this actually be a set?
1115
return parent_map.keys()
1118
def create(a_bzrdir):
1119
"""Construct the current default format repository in a_bzrdir."""
1120
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1122
def __init__(self, _format, a_bzrdir, control_files):
1123
"""instantiate a Repository.
1125
:param _format: The format of the repository on disk.
1126
:param a_bzrdir: The BzrDir of the repository.
1128
In the future we will have a single api for all stores for
1129
getting file texts, inventories and revisions, then
1130
this construct will accept instances of those things.
1132
super(Repository, self).__init__()
1133
self._format = _format
1134
# the following are part of the public API for Repository:
1135
self.bzrdir = a_bzrdir
1136
self.control_files = control_files
1137
self._transport = control_files._transport
1138
self.base = self._transport.base
1140
self._reconcile_does_inventory_gc = True
1141
self._reconcile_fixes_text_parents = False
1142
self._reconcile_backsup_inventory = True
1143
# not right yet - should be more semantically clear ?
1145
# TODO: make sure to construct the right store classes, etc, depending
1146
# on whether escaping is required.
1147
self._warn_if_deprecated()
1148
self._write_group = None
1149
# Additional places to query for data.
1150
self._fallback_repositories = []
1151
# An InventoryEntry cache, used during deserialization
1152
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1155
return '%s(%r)' % (self.__class__.__name__,
1158
def has_same_location(self, other):
1159
"""Returns a boolean indicating if this repository is at the same
1160
location as another repository.
1162
This might return False even when two repository objects are accessing
1163
the same physical repository via different URLs.
1165
if self.__class__ is not other.__class__:
1167
return (self._transport.base == other._transport.base)
1169
def is_in_write_group(self):
1170
"""Return True if there is an open write group.
1172
:seealso: start_write_group.
1174
return self._write_group is not None
1176
def is_locked(self):
1177
return self.control_files.is_locked()
1179
def is_write_locked(self):
1180
"""Return True if this object is write locked."""
1181
return self.is_locked() and self.control_files._lock_mode == 'w'
1183
def lock_write(self, token=None):
1184
"""Lock this repository for writing.
1186
This causes caching within the repository obejct to start accumlating
1187
data during reads, and allows a 'write_group' to be obtained. Write
1188
groups must be used for actual data insertion.
1190
:param token: if this is already locked, then lock_write will fail
1191
unless the token matches the existing lock.
1192
:returns: a token if this instance supports tokens, otherwise None.
1193
:raises TokenLockingNotSupported: when a token is given but this
1194
instance doesn't support using token locks.
1195
:raises MismatchedToken: if the specified token doesn't match the token
1196
of the existing lock.
1197
:seealso: start_write_group.
1199
A token should be passed in if you know that you have locked the object
1200
some other way, and need to synchronise this object's state with that
1203
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1205
locked = self.is_locked()
1206
result = self.control_files.lock_write(token=token)
1207
for repo in self._fallback_repositories:
1208
# Writes don't affect fallback repos
1211
self._refresh_data()
1214
def lock_read(self):
1215
locked = self.is_locked()
1216
self.control_files.lock_read()
1217
for repo in self._fallback_repositories:
1220
self._refresh_data()
1222
def get_physical_lock_status(self):
1223
return self.control_files.get_physical_lock_status()
1225
def leave_lock_in_place(self):
1226
"""Tell this repository not to release the physical lock when this
1229
If lock_write doesn't return a token, then this method is not supported.
1231
self.control_files.leave_in_place()
1233
def dont_leave_lock_in_place(self):
1234
"""Tell this repository to release the physical lock when this
1235
object is unlocked, even if it didn't originally acquire it.
1237
If lock_write doesn't return a token, then this method is not supported.
1239
self.control_files.dont_leave_in_place()
1242
def gather_stats(self, revid=None, committers=None):
1243
"""Gather statistics from a revision id.
1245
:param revid: The revision id to gather statistics from, if None, then
1246
no revision specific statistics are gathered.
1247
:param committers: Optional parameter controlling whether to grab
1248
a count of committers from the revision specific statistics.
1249
:return: A dictionary of statistics. Currently this contains:
1250
committers: The number of committers if requested.
1251
firstrev: A tuple with timestamp, timezone for the penultimate left
1252
most ancestor of revid, if revid is not the NULL_REVISION.
1253
latestrev: A tuple with timestamp, timezone for revid, if revid is
1254
not the NULL_REVISION.
1255
revisions: The total revision count in the repository.
1256
size: An estimate disk size of the repository in bytes.
1259
if revid and committers:
1260
result['committers'] = 0
1261
if revid and revid != _mod_revision.NULL_REVISION:
1263
all_committers = set()
1264
revisions = self.get_ancestry(revid)
1265
# pop the leading None
1267
first_revision = None
1269
# ignore the revisions in the middle - just grab first and last
1270
revisions = revisions[0], revisions[-1]
1271
for revision in self.get_revisions(revisions):
1272
if not first_revision:
1273
first_revision = revision
1275
all_committers.add(revision.committer)
1276
last_revision = revision
1278
result['committers'] = len(all_committers)
1279
result['firstrev'] = (first_revision.timestamp,
1280
first_revision.timezone)
1281
result['latestrev'] = (last_revision.timestamp,
1282
last_revision.timezone)
1284
# now gather global repository information
1285
# XXX: This is available for many repos regardless of listability.
1286
if self.bzrdir.root_transport.listable():
1287
# XXX: do we want to __define len__() ?
1288
# Maybe the versionedfiles object should provide a different
1289
# method to get the number of keys.
1290
result['revisions'] = len(self.revisions.keys())
1291
# result['size'] = t
1294
def find_branches(self, using=False):
1295
"""Find branches underneath this repository.
1297
This will include branches inside other branches.
1299
:param using: If True, list only branches using this repository.
1301
if using and not self.is_shared():
1303
return [self.bzrdir.open_branch()]
1304
except errors.NotBranchError:
1306
class Evaluator(object):
1309
self.first_call = True
1311
def __call__(self, bzrdir):
1312
# On the first call, the parameter is always the bzrdir
1313
# containing the current repo.
1314
if not self.first_call:
1316
repository = bzrdir.open_repository()
1317
except errors.NoRepositoryPresent:
1320
return False, (None, repository)
1321
self.first_call = False
1323
value = (bzrdir.open_branch(), None)
1324
except errors.NotBranchError:
1325
value = (None, None)
1329
for branch, repository in bzrdir.BzrDir.find_bzrdirs(
1330
self.bzrdir.root_transport, evaluate=Evaluator()):
1331
if branch is not None:
1332
branches.append(branch)
1333
if not using and repository is not None:
1334
branches.extend(repository.find_branches())
1338
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
1339
"""Return the revision ids that other has that this does not.
1341
These are returned in topological order.
1343
revision_id: only return revision ids included by revision_id.
1345
return InterRepository.get(other, self).search_missing_revision_ids(
1346
revision_id, find_ghosts)
1350
"""Open the repository rooted at base.
1352
For instance, if the repository is at URL/.bzr/repository,
1353
Repository.open(URL) -> a Repository instance.
1355
control = bzrdir.BzrDir.open(base)
1356
return control.open_repository()
1358
def copy_content_into(self, destination, revision_id=None):
1359
"""Make a complete copy of the content in self into destination.
1361
This is a destructive operation! Do not use it on existing
1364
return InterRepository.get(self, destination).copy_content(revision_id)
1366
def commit_write_group(self):
1367
"""Commit the contents accrued within the current write group.
1369
:seealso: start_write_group.
1371
if self._write_group is not self.get_transaction():
1372
# has an unlock or relock occured ?
1373
raise errors.BzrError('mismatched lock context %r and '
1375
(self.get_transaction(), self._write_group))
1376
self._commit_write_group()
1377
self._write_group = None
1379
def _commit_write_group(self):
1380
"""Template method for per-repository write group cleanup.
1382
This is called before the write group is considered to be
1383
finished and should ensure that all data handed to the repository
1384
for writing during the write group is safely committed (to the
1385
extent possible considering file system caching etc).
1388
def suspend_write_group(self):
1389
raise errors.UnsuspendableWriteGroup(self)
1391
def refresh_data(self):
1392
"""Re-read any data needed to to synchronise with disk.
1394
This method is intended to be called after another repository instance
1395
(such as one used by a smart server) has inserted data into the
1396
repository. It may not be called during a write group, but may be
1397
called at any other time.
1399
if self.is_in_write_group():
1400
raise errors.InternalBzrError(
1401
"May not refresh_data while in a write group.")
1402
self._refresh_data()
1404
def resume_write_group(self, tokens):
1405
if not self.is_write_locked():
1406
raise errors.NotWriteLocked(self)
1407
if self._write_group:
1408
raise errors.BzrError('already in a write group')
1409
self._resume_write_group(tokens)
1410
# so we can detect unlock/relock - the write group is now entered.
1411
self._write_group = self.get_transaction()
1413
def _resume_write_group(self, tokens):
1414
raise errors.UnsuspendableWriteGroup(self)
1416
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1418
"""Fetch the content required to construct revision_id from source.
1420
If revision_id is None and fetch_spec is None, then all content is
1423
fetch() may not be used when the repository is in a write group -
1424
either finish the current write group before using fetch, or use
1425
fetch before starting the write group.
1427
:param find_ghosts: Find and copy revisions in the source that are
1428
ghosts in the target (and not reachable directly by walking out to
1429
the first-present revision in target from revision_id).
1430
:param revision_id: If specified, all the content needed for this
1431
revision ID will be copied to the target. Fetch will determine for
1432
itself which content needs to be copied.
1433
:param fetch_spec: If specified, a SearchResult or
1434
PendingAncestryResult that describes which revisions to copy. This
1435
allows copying multiple heads at once. Mutually exclusive with
1438
if fetch_spec is not None and revision_id is not None:
1439
raise AssertionError(
1440
"fetch_spec and revision_id are mutually exclusive.")
1441
if self.is_in_write_group():
1442
raise errors.InternalBzrError(
1443
"May not fetch while in a write group.")
1444
# fast path same-url fetch operations
1445
if self.has_same_location(source) and fetch_spec is None:
1446
# check that last_revision is in 'from' and then return a
1448
if (revision_id is not None and
1449
not _mod_revision.is_null(revision_id)):
1450
self.get_revision(revision_id)
1452
# if there is no specific appropriate InterRepository, this will get
1453
# the InterRepository base class, which raises an
1454
# IncompatibleRepositories when asked to fetch.
1455
inter = InterRepository.get(source, self)
1456
return inter.fetch(revision_id=revision_id, pb=pb,
1457
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1459
def create_bundle(self, target, base, fileobj, format=None):
1460
return serializer.write_bundle(self, target, base, fileobj, format)
1462
def get_commit_builder(self, branch, parents, config, timestamp=None,
1463
timezone=None, committer=None, revprops=None,
1465
"""Obtain a CommitBuilder for this repository.
1467
:param branch: Branch to commit to.
1468
:param parents: Revision ids of the parents of the new revision.
1469
:param config: Configuration to use.
1470
:param timestamp: Optional timestamp recorded for commit.
1471
:param timezone: Optional timezone for timestamp.
1472
:param committer: Optional committer to set for commit.
1473
:param revprops: Optional dictionary of revision properties.
1474
:param revision_id: Optional revision id.
1476
result = self._commit_builder_class(self, parents, config,
1477
timestamp, timezone, committer, revprops, revision_id)
1478
self.start_write_group()
1482
if (self.control_files._lock_count == 1 and
1483
self.control_files._lock_mode == 'w'):
1484
if self._write_group is not None:
1485
self.abort_write_group()
1486
self.control_files.unlock()
1487
raise errors.BzrError(
1488
'Must end write groups before releasing write locks.')
1489
self.control_files.unlock()
1490
if self.control_files._lock_count == 0:
1491
self._inventory_entry_cache.clear()
1492
for repo in self._fallback_repositories:
1496
def clone(self, a_bzrdir, revision_id=None):
1497
"""Clone this repository into a_bzrdir using the current format.
1499
Currently no check is made that the format of this repository and
1500
the bzrdir format are compatible. FIXME RBC 20060201.
1502
:return: The newly created destination repository.
1504
# TODO: deprecate after 0.16; cloning this with all its settings is
1505
# probably not very useful -- mbp 20070423
1506
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1507
self.copy_content_into(dest_repo, revision_id)
1510
def start_write_group(self):
1511
"""Start a write group in the repository.
1513
Write groups are used by repositories which do not have a 1:1 mapping
1514
between file ids and backend store to manage the insertion of data from
1515
both fetch and commit operations.
1517
A write lock is required around the start_write_group/commit_write_group
1518
for the support of lock-requiring repository formats.
1520
One can only insert data into a repository inside a write group.
1524
if not self.is_write_locked():
1525
raise errors.NotWriteLocked(self)
1526
if self._write_group:
1527
raise errors.BzrError('already in a write group')
1528
self._start_write_group()
1529
# so we can detect unlock/relock - the write group is now entered.
1530
self._write_group = self.get_transaction()
1532
def _start_write_group(self):
1533
"""Template method for per-repository write group startup.
1535
This is called before the write group is considered to be
1540
def sprout(self, to_bzrdir, revision_id=None):
1541
"""Create a descendent repository for new development.
1543
Unlike clone, this does not copy the settings of the repository.
1545
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1546
dest_repo.fetch(self, revision_id=revision_id)
1549
def _create_sprouting_repo(self, a_bzrdir, shared):
1550
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1551
# use target default format.
1552
dest_repo = a_bzrdir.create_repository()
1554
# Most control formats need the repository to be specifically
1555
# created, but on some old all-in-one formats it's not needed
1557
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1558
except errors.UninitializableFormat:
1559
dest_repo = a_bzrdir.open_repository()
1562
def _get_sink(self):
1563
"""Return a sink for streaming into this repository."""
1564
return StreamSink(self)
1566
def _get_source(self, to_format):
1567
"""Return a source for streaming from this repository."""
1568
return StreamSource(self, to_format)
1571
def has_revision(self, revision_id):
1572
"""True if this repository has a copy of the revision."""
1573
return revision_id in self.has_revisions((revision_id,))
1576
def has_revisions(self, revision_ids):
1577
"""Probe to find out the presence of multiple revisions.
1579
:param revision_ids: An iterable of revision_ids.
1580
:return: A set of the revision_ids that were present.
1582
parent_map = self.revisions.get_parent_map(
1583
[(rev_id,) for rev_id in revision_ids])
1585
if _mod_revision.NULL_REVISION in revision_ids:
1586
result.add(_mod_revision.NULL_REVISION)
1587
result.update([key[0] for key in parent_map])
1591
def get_revision(self, revision_id):
1592
"""Return the Revision object for a named revision."""
1593
return self.get_revisions([revision_id])[0]
1596
def get_revision_reconcile(self, revision_id):
1597
"""'reconcile' helper routine that allows access to a revision always.
1599
This variant of get_revision does not cross check the weave graph
1600
against the revision one as get_revision does: but it should only
1601
be used by reconcile, or reconcile-alike commands that are correcting
1602
or testing the revision graph.
1604
return self._get_revisions([revision_id])[0]
1607
def get_revisions(self, revision_ids):
1608
"""Get many revisions at once."""
1609
return self._get_revisions(revision_ids)
1612
def _get_revisions(self, revision_ids):
1613
"""Core work logic to get many revisions without sanity checks."""
1614
for rev_id in revision_ids:
1615
if not rev_id or not isinstance(rev_id, basestring):
1616
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1617
keys = [(key,) for key in revision_ids]
1618
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1620
for record in stream:
1621
if record.storage_kind == 'absent':
1622
raise errors.NoSuchRevision(self, record.key[0])
1623
text = record.get_bytes_as('fulltext')
1624
rev = self._serializer.read_revision_from_string(text)
1625
revs[record.key[0]] = rev
1626
return [revs[revid] for revid in revision_ids]
1629
def get_revision_xml(self, revision_id):
1630
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1631
# would have already do it.
1632
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1633
rev = self.get_revision(revision_id)
1634
rev_tmp = cStringIO.StringIO()
1635
# the current serializer..
1636
self._serializer.write_revision(rev, rev_tmp)
1638
return rev_tmp.getvalue()
1640
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1641
"""Produce a generator of revision deltas.
1643
Note that the input is a sequence of REVISIONS, not revision_ids.
1644
Trees will be held in memory until the generator exits.
1645
Each delta is relative to the revision's lefthand predecessor.
1647
:param specific_fileids: if not None, the result is filtered
1648
so that only those file-ids, their parents and their
1649
children are included.
1651
# Get the revision-ids of interest
1652
required_trees = set()
1653
for revision in revisions:
1654
required_trees.add(revision.revision_id)
1655
required_trees.update(revision.parent_ids[:1])
1657
# Get the matching filtered trees. Note that it's more
1658
# efficient to pass filtered trees to changes_from() rather
1659
# than doing the filtering afterwards. changes_from() could
1660
# arguably do the filtering itself but it's path-based, not
1661
# file-id based, so filtering before or afterwards is
1663
if specific_fileids is None:
1664
trees = dict((t.get_revision_id(), t) for
1665
t in self.revision_trees(required_trees))
1667
trees = dict((t.get_revision_id(), t) for
1668
t in self._filtered_revision_trees(required_trees,
1671
# Calculate the deltas
1672
for revision in revisions:
1673
if not revision.parent_ids:
1674
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1676
old_tree = trees[revision.parent_ids[0]]
1677
yield trees[revision.revision_id].changes_from(old_tree)
1680
def get_revision_delta(self, revision_id, specific_fileids=None):
1681
"""Return the delta for one revision.
1683
The delta is relative to the left-hand predecessor of the
1686
:param specific_fileids: if not None, the result is filtered
1687
so that only those file-ids, their parents and their
1688
children are included.
1690
r = self.get_revision(revision_id)
1691
return list(self.get_deltas_for_revisions([r],
1692
specific_fileids=specific_fileids))[0]
1695
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1696
signature = gpg_strategy.sign(plaintext)
1697
self.add_signature_text(revision_id, signature)
1700
def add_signature_text(self, revision_id, signature):
1701
self.signatures.add_lines((revision_id,), (),
1702
osutils.split_lines(signature))
1704
def find_text_key_references(self):
1705
"""Find the text key references within the repository.
1707
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1708
to whether they were referred to by the inventory of the
1709
revision_id that they contain. The inventory texts from all present
1710
revision ids are assessed to generate this report.
1712
revision_keys = self.revisions.keys()
1713
w = self.inventories
1714
pb = ui.ui_factory.nested_progress_bar()
1716
return self._find_text_key_references_from_xml_inventory_lines(
1717
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1721
def _find_text_key_references_from_xml_inventory_lines(self,
1723
"""Core routine for extracting references to texts from inventories.
1725
This performs the translation of xml lines to revision ids.
1727
:param line_iterator: An iterator of lines, origin_version_id
1728
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1729
to whether they were referred to by the inventory of the
1730
revision_id that they contain. Note that if that revision_id was
1731
not part of the line_iterator's output then False will be given -
1732
even though it may actually refer to that key.
1734
if not self._serializer.support_altered_by_hack:
1735
raise AssertionError(
1736
"_find_text_key_references_from_xml_inventory_lines only "
1737
"supported for branches which store inventory as unnested xml"
1738
", not on %r" % self)
1741
# this code needs to read every new line in every inventory for the
1742
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
1743
# not present in one of those inventories is unnecessary but not
1744
# harmful because we are filtering by the revision id marker in the
1745
# inventory lines : we only select file ids altered in one of those
1746
# revisions. We don't need to see all lines in the inventory because
1747
# only those added in an inventory in rev X can contain a revision=X
1749
unescape_revid_cache = {}
1750
unescape_fileid_cache = {}
1752
# jam 20061218 In a big fetch, this handles hundreds of thousands
1753
# of lines, so it has had a lot of inlining and optimizing done.
1754
# Sorry that it is a little bit messy.
1755
# Move several functions to be local variables, since this is a long
1757
search = self._file_ids_altered_regex.search
1758
unescape = _unescape_xml
1759
setdefault = result.setdefault
1760
for line, line_key in line_iterator:
1761
match = search(line)
1764
# One call to match.group() returning multiple items is quite a
1765
# bit faster than 2 calls to match.group() each returning 1
1766
file_id, revision_id = match.group('file_id', 'revision_id')
1768
# Inlining the cache lookups helps a lot when you make 170,000
1769
# lines and 350k ids, versus 8.4 unique ids.
1770
# Using a cache helps in 2 ways:
1771
# 1) Avoids unnecessary decoding calls
1772
# 2) Re-uses cached strings, which helps in future set and
1774
# (2) is enough that removing encoding entirely along with
1775
# the cache (so we are using plain strings) results in no
1776
# performance improvement.
1778
revision_id = unescape_revid_cache[revision_id]
1780
unescaped = unescape(revision_id)
1781
unescape_revid_cache[revision_id] = unescaped
1782
revision_id = unescaped
1784
# Note that unconditionally unescaping means that we deserialise
1785
# every fileid, which for general 'pull' is not great, but we don't
1786
# really want to have some many fulltexts that this matters anyway.
1789
file_id = unescape_fileid_cache[file_id]
1791
unescaped = unescape(file_id)
1792
unescape_fileid_cache[file_id] = unescaped
1795
key = (file_id, revision_id)
1796
setdefault(key, False)
1797
if revision_id == line_key[-1]:
1801
def _inventory_xml_lines_for_keys(self, keys):
1802
"""Get a line iterator of the sort needed for findind references.
1804
Not relevant for non-xml inventory repositories.
1806
Ghosts in revision_keys are ignored.
1808
:param revision_keys: The revision keys for the inventories to inspect.
1809
:return: An iterator over (inventory line, revid) for the fulltexts of
1810
all of the xml inventories specified by revision_keys.
1812
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1813
for record in stream:
1814
if record.storage_kind != 'absent':
1815
chunks = record.get_bytes_as('chunked')
1816
revid = record.key[-1]
1817
lines = osutils.chunks_to_lines(chunks)
1821
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1823
"""Helper routine for fileids_altered_by_revision_ids.
1825
This performs the translation of xml lines to revision ids.
1827
:param line_iterator: An iterator of lines, origin_version_id
1828
:param revision_ids: The revision ids to filter for. This should be a
1829
set or other type which supports efficient __contains__ lookups, as
1830
the revision id from each parsed line will be looked up in the
1831
revision_ids filter.
1832
:return: a dictionary mapping altered file-ids to an iterable of
1833
revision_ids. Each altered file-ids has the exact revision_ids that
1834
altered it listed explicitly.
1836
seen = set(self._find_text_key_references_from_xml_inventory_lines(
1837
line_iterator).iterkeys())
1838
# Note that revision_ids are revision keys.
1839
parent_maps = self.revisions.get_parent_map(revision_ids)
1841
map(parents.update, parent_maps.itervalues())
1842
parents.difference_update(revision_ids)
1843
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
1844
self._inventory_xml_lines_for_keys(parents)))
1845
new_keys = seen - parent_seen
1847
setdefault = result.setdefault
1848
for key in new_keys:
1849
setdefault(key[0], set()).add(key[-1])
1852
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
1853
"""Find the file ids and versions affected by revisions.
1855
:param revisions: an iterable containing revision ids.
1856
:param _inv_weave: The inventory weave from this repository or None.
1857
If None, the inventory weave will be opened automatically.
1858
:return: a dictionary mapping altered file-ids to an iterable of
1859
revision_ids. Each altered file-ids has the exact revision_ids that
1860
altered it listed explicitly.
1862
selected_keys = set((revid,) for revid in revision_ids)
1863
w = _inv_weave or self.inventories
1864
pb = ui.ui_factory.nested_progress_bar()
1866
return self._find_file_ids_from_xml_inventory_lines(
1867
w.iter_lines_added_or_present_in_keys(
1868
selected_keys, pb=pb),
1873
def iter_files_bytes(self, desired_files):
1874
"""Iterate through file versions.
1876
Files will not necessarily be returned in the order they occur in
1877
desired_files. No specific order is guaranteed.
1879
Yields pairs of identifier, bytes_iterator. identifier is an opaque
1880
value supplied by the caller as part of desired_files. It should
1881
uniquely identify the file version in the caller's context. (Examples:
1882
an index number or a TreeTransform trans_id.)
1884
bytes_iterator is an iterable of bytestrings for the file. The
1885
kind of iterable and length of the bytestrings are unspecified, but for
1886
this implementation, it is a list of bytes produced by
1887
VersionedFile.get_record_stream().
1889
:param desired_files: a list of (file_id, revision_id, identifier)
1893
for file_id, revision_id, callable_data in desired_files:
1894
text_keys[(file_id, revision_id)] = callable_data
1895
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
1896
if record.storage_kind == 'absent':
1897
raise errors.RevisionNotPresent(record.key, self)
1898
yield text_keys[record.key], record.get_bytes_as('chunked')
1900
def _generate_text_key_index(self, text_key_references=None,
1902
"""Generate a new text key index for the repository.
1904
This is an expensive function that will take considerable time to run.
1906
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
1907
list of parents, also text keys. When a given key has no parents,
1908
the parents list will be [NULL_REVISION].
1910
# All revisions, to find inventory parents.
1911
if ancestors is None:
1912
graph = self.get_graph()
1913
ancestors = graph.get_parent_map(self.all_revision_ids())
1914
if text_key_references is None:
1915
text_key_references = self.find_text_key_references()
1916
pb = ui.ui_factory.nested_progress_bar()
1918
return self._do_generate_text_key_index(ancestors,
1919
text_key_references, pb)
1923
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
1924
"""Helper for _generate_text_key_index to avoid deep nesting."""
1925
revision_order = tsort.topo_sort(ancestors)
1926
invalid_keys = set()
1928
for revision_id in revision_order:
1929
revision_keys[revision_id] = set()
1930
text_count = len(text_key_references)
1931
# a cache of the text keys to allow reuse; costs a dict of all the
1932
# keys, but saves a 2-tuple for every child of a given key.
1934
for text_key, valid in text_key_references.iteritems():
1936
invalid_keys.add(text_key)
1938
revision_keys[text_key[1]].add(text_key)
1939
text_key_cache[text_key] = text_key
1940
del text_key_references
1942
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
1943
NULL_REVISION = _mod_revision.NULL_REVISION
1944
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
1945
# too small for large or very branchy trees. However, for 55K path
1946
# trees, it would be easy to use too much memory trivially. Ideally we
1947
# could gauge this by looking at available real memory etc, but this is
1948
# always a tricky proposition.
1949
inventory_cache = lru_cache.LRUCache(10)
1950
batch_size = 10 # should be ~150MB on a 55K path tree
1951
batch_count = len(revision_order) / batch_size + 1
1953
pb.update("Calculating text parents", processed_texts, text_count)
1954
for offset in xrange(batch_count):
1955
to_query = revision_order[offset * batch_size:(offset + 1) *
1959
for rev_tree in self.revision_trees(to_query):
1960
revision_id = rev_tree.get_revision_id()
1961
parent_ids = ancestors[revision_id]
1962
for text_key in revision_keys[revision_id]:
1963
pb.update("Calculating text parents", processed_texts)
1964
processed_texts += 1
1965
candidate_parents = []
1966
for parent_id in parent_ids:
1967
parent_text_key = (text_key[0], parent_id)
1969
check_parent = parent_text_key not in \
1970
revision_keys[parent_id]
1972
# the parent parent_id is a ghost:
1973
check_parent = False
1974
# truncate the derived graph against this ghost.
1975
parent_text_key = None
1977
# look at the parent commit details inventories to
1978
# determine possible candidates in the per file graph.
1981
inv = inventory_cache[parent_id]
1983
inv = self.revision_tree(parent_id).inventory
1984
inventory_cache[parent_id] = inv
1985
parent_entry = inv._byid.get(text_key[0], None)
1986
if parent_entry is not None:
1988
text_key[0], parent_entry.revision)
1990
parent_text_key = None
1991
if parent_text_key is not None:
1992
candidate_parents.append(
1993
text_key_cache[parent_text_key])
1994
parent_heads = text_graph.heads(candidate_parents)
1995
new_parents = list(parent_heads)
1996
new_parents.sort(key=lambda x:candidate_parents.index(x))
1997
if new_parents == []:
1998
new_parents = [NULL_REVISION]
1999
text_index[text_key] = new_parents
2001
for text_key in invalid_keys:
2002
text_index[text_key] = [NULL_REVISION]
2005
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2006
"""Get an iterable listing the keys of all the data introduced by a set
2009
The keys will be ordered so that the corresponding items can be safely
2010
fetched and inserted in that order.
2012
:returns: An iterable producing tuples of (knit-kind, file-id,
2013
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2014
'revisions'. file-id is None unless knit-kind is 'file'.
2016
# XXX: it's a bit weird to control the inventory weave caching in this
2017
# generator. Ideally the caching would be done in fetch.py I think. Or
2018
# maybe this generator should explicitly have the contract that it
2019
# should not be iterated until the previously yielded item has been
2021
inv_w = self.inventories
2023
# file ids that changed
2024
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2026
num_file_ids = len(file_ids)
2027
for file_id, altered_versions in file_ids.iteritems():
2028
if _files_pb is not None:
2029
_files_pb.update("fetch texts", count, num_file_ids)
2031
yield ("file", file_id, altered_versions)
2032
# We're done with the files_pb. Note that it finished by the caller,
2033
# just as it was created by the caller.
2037
yield ("inventory", None, revision_ids)
2040
# XXX: Note ATM no callers actually pay attention to this return
2041
# instead they just use the list of revision ids and ignore
2042
# missing sigs. Consider removing this work entirely
2043
revisions_with_signatures = set(self.signatures.get_parent_map(
2044
[(r,) for r in revision_ids]))
2045
revisions_with_signatures = set(
2046
[r for (r,) in revisions_with_signatures])
2047
revisions_with_signatures.intersection_update(revision_ids)
2048
yield ("signatures", None, revisions_with_signatures)
2051
yield ("revisions", None, revision_ids)
2054
def get_inventory(self, revision_id):
2055
"""Get Inventory object by revision id."""
2056
return self.iter_inventories([revision_id]).next()
2058
def iter_inventories(self, revision_ids):
2059
"""Get many inventories by revision_ids.
2061
This will buffer some or all of the texts used in constructing the
2062
inventories in memory, but will only parse a single inventory at a
2065
:param revision_ids: The expected revision ids of the inventories.
2066
:return: An iterator of inventories.
2068
if ((None in revision_ids)
2069
or (_mod_revision.NULL_REVISION in revision_ids)):
2070
raise ValueError('cannot get null revision inventory')
2071
return self._iter_inventories(revision_ids)
2073
def _iter_inventories(self, revision_ids):
2074
"""single-document based inventory iteration."""
2075
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2076
yield self.deserialise_inventory(revision_id, text)
2078
def _iter_inventory_xmls(self, revision_ids):
2079
keys = [(revision_id,) for revision_id in revision_ids]
2080
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2082
for record in stream:
2083
if record.storage_kind != 'absent':
2084
text_chunks[record.key] = record.get_bytes_as('chunked')
2086
raise errors.NoSuchRevision(self, record.key)
2088
chunks = text_chunks.pop(key)
2089
yield ''.join(chunks), key[-1]
2091
def deserialise_inventory(self, revision_id, xml):
2092
"""Transform the xml into an inventory object.
2094
:param revision_id: The expected revision id of the inventory.
2095
:param xml: A serialised inventory.
2097
result = self._serializer.read_inventory_from_string(xml, revision_id,
2098
entry_cache=self._inventory_entry_cache)
2099
if result.revision_id != revision_id:
2100
raise AssertionError('revision id mismatch %s != %s' % (
2101
result.revision_id, revision_id))
2104
def serialise_inventory(self, inv):
2105
return self._serializer.write_inventory_to_string(inv)
2107
def _serialise_inventory_to_lines(self, inv):
2108
return self._serializer.write_inventory_to_lines(inv)
2110
def get_serializer_format(self):
2111
return self._serializer.format_num
2114
def get_inventory_xml(self, revision_id):
2115
"""Get inventory XML as a file object."""
2116
texts = self._iter_inventory_xmls([revision_id])
2118
text, revision_id = texts.next()
2119
except StopIteration:
2120
raise errors.HistoryMissing(self, 'inventory', revision_id)
2124
def get_inventory_sha1(self, revision_id):
2125
"""Return the sha1 hash of the inventory entry
2127
return self.get_revision(revision_id).inventory_sha1
2129
def iter_reverse_revision_history(self, revision_id):
2130
"""Iterate backwards through revision ids in the lefthand history
2132
:param revision_id: The revision id to start with. All its lefthand
2133
ancestors will be traversed.
2135
graph = self.get_graph()
2136
next_id = revision_id
2138
if next_id in (None, _mod_revision.NULL_REVISION):
2141
# Note: The following line may raise KeyError in the event of
2142
# truncated history. We decided not to have a try:except:raise
2143
# RevisionNotPresent here until we see a use for it, because of the
2144
# cost in an inner loop that is by its very nature O(history).
2145
# Robert Collins 20080326
2146
parents = graph.get_parent_map([next_id])[next_id]
2147
if len(parents) == 0:
2150
next_id = parents[0]
2153
def get_revision_inventory(self, revision_id):
2154
"""Return inventory of a past revision."""
2155
# TODO: Unify this with get_inventory()
2156
# bzr 0.0.6 and later imposes the constraint that the inventory_id
2157
# must be the same as its revision, so this is trivial.
2158
if revision_id is None:
2159
# This does not make sense: if there is no revision,
2160
# then it is the current tree inventory surely ?!
2161
# and thus get_root_id() is something that looks at the last
2162
# commit on the branch, and the get_root_id is an inventory check.
2163
raise NotImplementedError
2164
# return Inventory(self.get_root_id())
2166
return self.get_inventory(revision_id)
2168
def is_shared(self):
2169
"""Return True if this repository is flagged as a shared repository."""
2170
raise NotImplementedError(self.is_shared)
2173
def reconcile(self, other=None, thorough=False):
2174
"""Reconcile this repository."""
2175
from bzrlib.reconcile import RepoReconciler
2176
reconciler = RepoReconciler(self, thorough=thorough)
2177
reconciler.reconcile()
2180
def _refresh_data(self):
2181
"""Helper called from lock_* to ensure coherency with disk.
2183
The default implementation does nothing; it is however possible
2184
for repositories to maintain loaded indices across multiple locks
2185
by checking inside their implementation of this method to see
2186
whether their indices are still valid. This depends of course on
2187
the disk format being validatable in this manner. This method is
2188
also called by the refresh_data() public interface to cause a refresh
2189
to occur while in a write lock so that data inserted by a smart server
2190
push operation is visible on the client's instance of the physical
2195
def revision_tree(self, revision_id):
2196
"""Return Tree for a revision on this branch.
2198
`revision_id` may be NULL_REVISION for the empty tree revision.
2200
revision_id = _mod_revision.ensure_null(revision_id)
2201
# TODO: refactor this to use an existing revision object
2202
# so we don't need to read it in twice.
2203
if revision_id == _mod_revision.NULL_REVISION:
2204
return RevisionTree(self, Inventory(root_id=None),
2205
_mod_revision.NULL_REVISION)
2207
inv = self.get_revision_inventory(revision_id)
2208
return RevisionTree(self, inv, revision_id)
2210
def revision_trees(self, revision_ids):
2211
"""Return Trees for revisions in this repository.
2213
:param revision_ids: a sequence of revision-ids;
2214
a revision-id may not be None or 'null:'
2216
inventories = self.iter_inventories(revision_ids)
2217
for inv in inventories:
2218
yield RevisionTree(self, inv, inv.revision_id)
2220
def _filtered_revision_trees(self, revision_ids, file_ids):
2221
"""Return Tree for a revision on this branch with only some files.
2223
:param revision_ids: a sequence of revision-ids;
2224
a revision-id may not be None or 'null:'
2225
:param file_ids: if not None, the result is filtered
2226
so that only those file-ids, their parents and their
2227
children are included.
2229
inventories = self.iter_inventories(revision_ids)
2230
for inv in inventories:
2231
# Should we introduce a FilteredRevisionTree class rather
2232
# than pre-filter the inventory here?
2233
filtered_inv = inv.filter(file_ids)
2234
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2237
def get_ancestry(self, revision_id, topo_sorted=True):
2238
"""Return a list of revision-ids integrated by a revision.
2240
The first element of the list is always None, indicating the origin
2241
revision. This might change when we have history horizons, or
2242
perhaps we should have a new API.
2244
This is topologically sorted.
2246
if _mod_revision.is_null(revision_id):
2248
if not self.has_revision(revision_id):
2249
raise errors.NoSuchRevision(self, revision_id)
2250
graph = self.get_graph()
2252
search = graph._make_breadth_first_searcher([revision_id])
2255
found, ghosts = search.next_with_ghosts()
2256
except StopIteration:
2259
if _mod_revision.NULL_REVISION in keys:
2260
keys.remove(_mod_revision.NULL_REVISION)
2262
parent_map = graph.get_parent_map(keys)
2263
keys = tsort.topo_sort(parent_map)
2264
return [None] + list(keys)
2267
"""Compress the data within the repository.
2269
This operation only makes sense for some repository types. For other
2270
types it should be a no-op that just returns.
2272
This stub method does not require a lock, but subclasses should use
2273
@needs_write_lock as this is a long running call its reasonable to
2274
implicitly lock for the user.
2277
def get_transaction(self):
2278
return self.control_files.get_transaction()
2280
def get_parent_map(self, revision_ids):
2281
"""See graph._StackedParentsProvider.get_parent_map"""
2282
# revisions index works in keys; this just works in revisions
2283
# therefore wrap and unwrap
2286
for revision_id in revision_ids:
2287
if revision_id == _mod_revision.NULL_REVISION:
2288
result[revision_id] = ()
2289
elif revision_id is None:
2290
raise ValueError('get_parent_map(None) is not valid')
2292
query_keys.append((revision_id ,))
2293
for ((revision_id,), parent_keys) in \
2294
self.revisions.get_parent_map(query_keys).iteritems():
2296
result[revision_id] = tuple(parent_revid
2297
for (parent_revid,) in parent_keys)
2299
result[revision_id] = (_mod_revision.NULL_REVISION,)
2302
def _make_parents_provider(self):
2305
def get_graph(self, other_repository=None):
2306
"""Return the graph walker for this repository format"""
2307
parents_provider = self._make_parents_provider()
2308
if (other_repository is not None and
2309
not self.has_same_location(other_repository)):
2310
parents_provider = graph._StackedParentsProvider(
2311
[parents_provider, other_repository._make_parents_provider()])
2312
return graph.Graph(parents_provider)
2314
def _get_versioned_file_checker(self, text_key_references=None):
2315
"""Return an object suitable for checking versioned files.
2317
:param text_key_references: if non-None, an already built
2318
dictionary mapping text keys ((fileid, revision_id) tuples)
2319
to whether they were referred to by the inventory of the
2320
revision_id that they contain. If None, this will be
2323
return _VersionedFileChecker(self,
2324
text_key_references=text_key_references)
2326
def revision_ids_to_search_result(self, result_set):
2327
"""Convert a set of revision ids to a graph SearchResult."""
2328
result_parents = set()
2329
for parents in self.get_graph().get_parent_map(
2330
result_set).itervalues():
2331
result_parents.update(parents)
2332
included_keys = result_set.intersection(result_parents)
2333
start_keys = result_set.difference(included_keys)
2334
exclude_keys = result_parents.difference(result_set)
2335
result = graph.SearchResult(start_keys, exclude_keys,
2336
len(result_set), result_set)
2340
def set_make_working_trees(self, new_value):
2341
"""Set the policy flag for making working trees when creating branches.
2343
This only applies to branches that use this repository.
2345
The default is 'True'.
2346
:param new_value: True to restore the default, False to disable making
2349
raise NotImplementedError(self.set_make_working_trees)
2351
def make_working_trees(self):
2352
"""Returns the policy for making working trees on new branches."""
2353
raise NotImplementedError(self.make_working_trees)
2356
def sign_revision(self, revision_id, gpg_strategy):
2357
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2358
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2361
def has_signature_for_revision_id(self, revision_id):
2362
"""Query for a revision signature for revision_id in the repository."""
2363
if not self.has_revision(revision_id):
2364
raise errors.NoSuchRevision(self, revision_id)
2365
sig_present = (1 == len(
2366
self.signatures.get_parent_map([(revision_id,)])))
2370
def get_signature_text(self, revision_id):
2371
"""Return the text for a signature."""
2372
stream = self.signatures.get_record_stream([(revision_id,)],
2374
record = stream.next()
2375
if record.storage_kind == 'absent':
2376
raise errors.NoSuchRevision(self, revision_id)
2377
return record.get_bytes_as('fulltext')
2380
def check(self, revision_ids=None):
2381
"""Check consistency of all history of given revision_ids.
2383
Different repository implementations should override _check().
2385
:param revision_ids: A non-empty list of revision_ids whose ancestry
2386
will be checked. Typically the last revision_id of a branch.
2388
return self._check(revision_ids)
2390
def _check(self, revision_ids):
2391
result = check.Check(self)
2395
def _warn_if_deprecated(self):
2396
global _deprecation_warning_done
2397
if _deprecation_warning_done:
2399
_deprecation_warning_done = True
2400
warning("Format %s for %s is deprecated - please use 'bzr upgrade' to get better performance"
2401
% (self._format, self.bzrdir.transport.base))
2403
def supports_rich_root(self):
2404
return self._format.rich_root_data
2406
def _check_ascii_revisionid(self, revision_id, method):
2407
"""Private helper for ascii-only repositories."""
2408
# weave repositories refuse to store revisionids that are non-ascii.
2409
if revision_id is not None:
2410
# weaves require ascii revision ids.
2411
if isinstance(revision_id, unicode):
2413
revision_id.encode('ascii')
2414
except UnicodeEncodeError:
2415
raise errors.NonAsciiRevisionId(method, self)
2418
revision_id.decode('ascii')
2419
except UnicodeDecodeError:
2420
raise errors.NonAsciiRevisionId(method, self)
2422
def revision_graph_can_have_wrong_parents(self):
2423
"""Is it possible for this repository to have a revision graph with
2426
If True, then this repository must also implement
2427
_find_inconsistent_revision_parents so that check and reconcile can
2428
check for inconsistencies before proceeding with other checks that may
2429
depend on the revision index being consistent.
2431
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2434
# remove these delegates a while after bzr 0.15
2435
def __make_delegated(name, from_module):
2436
def _deprecated_repository_forwarder():
2437
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2438
% (name, from_module),
2441
m = __import__(from_module, globals(), locals(), [name])
2443
return getattr(m, name)
2444
except AttributeError:
2445
raise AttributeError('module %s has no name %s'
2447
globals()[name] = _deprecated_repository_forwarder
2450
'AllInOneRepository',
2451
'WeaveMetaDirRepository',
2452
'PreSplitOutRepositoryFormat',
2453
'RepositoryFormat4',
2454
'RepositoryFormat5',
2455
'RepositoryFormat6',
2456
'RepositoryFormat7',
2458
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2462
'RepositoryFormatKnit',
2463
'RepositoryFormatKnit1',
2465
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2468
def install_revision(repository, rev, revision_tree):
2469
"""Install all revision data into a repository."""
2470
install_revisions(repository, [(rev, revision_tree, None)])
2473
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2474
"""Install all revision data into a repository.
2476
Accepts an iterable of revision, tree, signature tuples. The signature
2479
repository.start_write_group()
2481
for n, (revision, revision_tree, signature) in enumerate(iterable):
2482
_install_revision(repository, revision, revision_tree, signature)
2484
pb.update('Transferring revisions', n + 1, num_revisions)
2486
repository.abort_write_group()
2489
repository.commit_write_group()
2492
def _install_revision(repository, rev, revision_tree, signature):
2493
"""Install all revision data into a repository."""
2494
present_parents = []
2496
for p_id in rev.parent_ids:
2497
if repository.has_revision(p_id):
2498
present_parents.append(p_id)
2499
parent_trees[p_id] = repository.revision_tree(p_id)
2501
parent_trees[p_id] = repository.revision_tree(
2502
_mod_revision.NULL_REVISION)
2504
inv = revision_tree.inventory
2505
entries = inv.iter_entries()
2506
# backwards compatibility hack: skip the root id.
2507
if not repository.supports_rich_root():
2508
path, root = entries.next()
2509
if root.revision != rev.revision_id:
2510
raise errors.IncompatibleRevision(repr(repository))
2512
for path, ie in entries:
2513
text_keys[(ie.file_id, ie.revision)] = ie
2514
text_parent_map = repository.texts.get_parent_map(text_keys)
2515
missing_texts = set(text_keys) - set(text_parent_map)
2516
# Add the texts that are not already present
2517
for text_key in missing_texts:
2518
ie = text_keys[text_key]
2520
# FIXME: TODO: The following loop overlaps/duplicates that done by
2521
# commit to determine parents. There is a latent/real bug here where
2522
# the parents inserted are not those commit would do - in particular
2523
# they are not filtered by heads(). RBC, AB
2524
for revision, tree in parent_trees.iteritems():
2525
if ie.file_id not in tree:
2527
parent_id = tree.inventory[ie.file_id].revision
2528
if parent_id in text_parents:
2530
text_parents.append((ie.file_id, parent_id))
2531
lines = revision_tree.get_file(ie.file_id).readlines()
2532
repository.texts.add_lines(text_key, text_parents, lines)
2534
# install the inventory
2535
repository.add_inventory(rev.revision_id, inv, present_parents)
2536
except errors.RevisionAlreadyPresent:
2538
if signature is not None:
2539
repository.add_signature_text(rev.revision_id, signature)
2540
repository.add_revision(rev.revision_id, rev, inv)
2543
class MetaDirRepository(Repository):
2544
"""Repositories in the new meta-dir layout.
2546
:ivar _transport: Transport for access to repository control files,
2547
typically pointing to .bzr/repository.
2550
def __init__(self, _format, a_bzrdir, control_files):
2551
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2552
self._transport = control_files._transport
2554
def is_shared(self):
2555
"""Return True if this repository is flagged as a shared repository."""
2556
return self._transport.has('shared-storage')
2559
def set_make_working_trees(self, new_value):
2560
"""Set the policy flag for making working trees when creating branches.
2562
This only applies to branches that use this repository.
2564
The default is 'True'.
2565
:param new_value: True to restore the default, False to disable making
2570
self._transport.delete('no-working-trees')
2571
except errors.NoSuchFile:
2574
self._transport.put_bytes('no-working-trees', '',
2575
mode=self.bzrdir._get_file_mode())
2577
def make_working_trees(self):
2578
"""Returns the policy for making working trees on new branches."""
2579
return not self._transport.has('no-working-trees')
2582
class MetaDirVersionedFileRepository(MetaDirRepository):
2583
"""Repositories in a meta-dir, that work via versioned file objects."""
2585
def __init__(self, _format, a_bzrdir, control_files):
2586
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2590
network_format_registry = registry.FormatRegistry()
2591
"""Registry of formats indexed by their network name.
2593
The network name for a repository format is an identifier that can be used when
2594
referring to formats with smart server operations. See
2595
RepositoryFormat.network_name() for more detail.
2599
format_registry = registry.FormatRegistry(network_format_registry)
2600
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
2602
This can contain either format instances themselves, or classes/factories that
2603
can be called to obtain one.
2607
#####################################################################
2608
# Repository Formats
2610
class RepositoryFormat(object):
2611
"""A repository format.
2613
Formats provide four things:
2614
* An initialization routine to construct repository data on disk.
2615
* a optional format string which is used when the BzrDir supports
2617
* an open routine which returns a Repository instance.
2618
* A network name for referring to the format in smart server RPC
2621
There is one and only one Format subclass for each on-disk format. But
2622
there can be one Repository subclass that is used for several different
2623
formats. The _format attribute on a Repository instance can be used to
2624
determine the disk format.
2626
Formats are placed in a registry by their format string for reference
2627
during opening. These should be subclasses of RepositoryFormat for
2630
Once a format is deprecated, just deprecate the initialize and open
2631
methods on the format class. Do not deprecate the object, as the
2632
object may be created even when a repository instnace hasn't been
2635
Common instance attributes:
2636
_matchingbzrdir - the bzrdir format that the repository format was
2637
originally written to work with. This can be used if manually
2638
constructing a bzrdir and repository, or more commonly for test suite
2642
# Set to True or False in derived classes. True indicates that the format
2643
# supports ghosts gracefully.
2644
supports_ghosts = None
2645
# Can this repository be given external locations to lookup additional
2646
# data. Set to True or False in derived classes.
2647
supports_external_lookups = None
2648
# What order should fetch operations request streams in?
2649
# The default is unordered as that is the cheapest for an origin to
2651
_fetch_order = 'unordered'
2652
# Does this repository format use deltas that can be fetched as-deltas ?
2653
# (E.g. knits, where the knit deltas can be transplanted intact.
2654
# We default to False, which will ensure that enough data to get
2655
# a full text out of any fetch stream will be grabbed.
2656
_fetch_uses_deltas = False
2657
# Should fetch trigger a reconcile after the fetch? Only needed for
2658
# some repository formats that can suffer internal inconsistencies.
2659
_fetch_reconcile = False
2660
# Does this format have < O(tree_size) delta generation. Used to hint what
2661
# code path for commit, amongst other things.
2665
return "<%s>" % self.__class__.__name__
2667
def __eq__(self, other):
2668
# format objects are generally stateless
2669
return isinstance(other, self.__class__)
2671
def __ne__(self, other):
2672
return not self == other
2675
def find_format(klass, a_bzrdir):
2676
"""Return the format for the repository object in a_bzrdir.
2678
This is used by bzr native formats that have a "format" file in
2679
the repository. Other methods may be used by different types of
2683
transport = a_bzrdir.get_repository_transport(None)
2684
format_string = transport.get("format").read()
2685
return format_registry.get(format_string)
2686
except errors.NoSuchFile:
2687
raise errors.NoRepositoryPresent(a_bzrdir)
2689
raise errors.UnknownFormatError(format=format_string,
2693
def register_format(klass, format):
2694
format_registry.register(format.get_format_string(), format)
2697
def unregister_format(klass, format):
2698
format_registry.remove(format.get_format_string())
2701
def get_default_format(klass):
2702
"""Return the current default format."""
2703
from bzrlib import bzrdir
2704
return bzrdir.format_registry.make_bzrdir('default').repository_format
2706
def get_format_string(self):
2707
"""Return the ASCII format string that identifies this format.
2709
Note that in pre format ?? repositories the format string is
2710
not permitted nor written to disk.
2712
raise NotImplementedError(self.get_format_string)
2714
def get_format_description(self):
2715
"""Return the short description for this format."""
2716
raise NotImplementedError(self.get_format_description)
2718
# TODO: this shouldn't be in the base class, it's specific to things that
2719
# use weaves or knits -- mbp 20070207
2720
def _get_versioned_file_store(self,
2725
versionedfile_class=None,
2726
versionedfile_kwargs={},
2728
if versionedfile_class is None:
2729
versionedfile_class = self._versionedfile_class
2730
weave_transport = control_files._transport.clone(name)
2731
dir_mode = control_files._dir_mode
2732
file_mode = control_files._file_mode
2733
return VersionedFileStore(weave_transport, prefixed=prefixed,
2735
file_mode=file_mode,
2736
versionedfile_class=versionedfile_class,
2737
versionedfile_kwargs=versionedfile_kwargs,
2740
def initialize(self, a_bzrdir, shared=False):
2741
"""Initialize a repository of this format in a_bzrdir.
2743
:param a_bzrdir: The bzrdir to put the new repository in it.
2744
:param shared: The repository should be initialized as a sharable one.
2745
:returns: The new repository object.
2747
This may raise UninitializableFormat if shared repository are not
2748
compatible the a_bzrdir.
2750
raise NotImplementedError(self.initialize)
2752
def is_supported(self):
2753
"""Is this format supported?
2755
Supported formats must be initializable and openable.
2756
Unsupported formats may not support initialization or committing or
2757
some other features depending on the reason for not being supported.
2761
def network_name(self):
2762
"""A simple byte string uniquely identifying this format for RPC calls.
2764
MetaDir repository formats use their disk format string to identify the
2765
repository over the wire. All in one formats such as bzr < 0.8, and
2766
foreign formats like svn/git and hg should use some marker which is
2767
unique and immutable.
2769
raise NotImplementedError(self.network_name)
2771
def check_conversion_target(self, target_format):
2772
raise NotImplementedError(self.check_conversion_target)
2774
def open(self, a_bzrdir, _found=False):
2775
"""Return an instance of this format for the bzrdir a_bzrdir.
2777
_found is a private parameter, do not use it.
2779
raise NotImplementedError(self.open)
2782
class MetaDirRepositoryFormat(RepositoryFormat):
2783
"""Common base class for the new repositories using the metadir layout."""
2785
rich_root_data = False
2786
supports_tree_reference = False
2787
supports_external_lookups = False
2790
def _matchingbzrdir(self):
2791
matching = bzrdir.BzrDirMetaFormat1()
2792
matching.repository_format = self
2796
super(MetaDirRepositoryFormat, self).__init__()
2798
def _create_control_files(self, a_bzrdir):
2799
"""Create the required files and the initial control_files object."""
2800
# FIXME: RBC 20060125 don't peek under the covers
2801
# NB: no need to escape relative paths that are url safe.
2802
repository_transport = a_bzrdir.get_repository_transport(self)
2803
control_files = lockable_files.LockableFiles(repository_transport,
2804
'lock', lockdir.LockDir)
2805
control_files.create_lock()
2806
return control_files
2808
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
2809
"""Upload the initial blank content."""
2810
control_files = self._create_control_files(a_bzrdir)
2811
control_files.lock_write()
2812
transport = control_files._transport
2814
utf8_files += [('shared-storage', '')]
2816
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
2817
for (filename, content_stream) in files:
2818
transport.put_file(filename, content_stream,
2819
mode=a_bzrdir._get_file_mode())
2820
for (filename, content_bytes) in utf8_files:
2821
transport.put_bytes_non_atomic(filename, content_bytes,
2822
mode=a_bzrdir._get_file_mode())
2824
control_files.unlock()
2826
def network_name(self):
2827
"""Metadir formats have matching disk and network format strings."""
2828
return self.get_format_string()
2831
# Pre-0.8 formats that don't have a disk format string (because they are
2832
# versioned by the matching control directory). We use the control directories
2833
# disk format string as a key for the network_name because they meet the
2834
# constraints (simple string, unique, immmutable).
2835
network_format_registry.register_lazy(
2836
"Bazaar-NG branch, format 5\n",
2837
'bzrlib.repofmt.weaverepo',
2838
'RepositoryFormat5',
2840
network_format_registry.register_lazy(
2841
"Bazaar-NG branch, format 6\n",
2842
'bzrlib.repofmt.weaverepo',
2843
'RepositoryFormat6',
2846
# formats which have no format string are not discoverable or independently
2847
# creatable on disk, so are not registered in format_registry. They're
2848
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
2849
# needed, it's constructed directly by the BzrDir. Non-native formats where
2850
# the repository is not separately opened are similar.
2852
format_registry.register_lazy(
2853
'Bazaar-NG Repository format 7',
2854
'bzrlib.repofmt.weaverepo',
2858
format_registry.register_lazy(
2859
'Bazaar-NG Knit Repository Format 1',
2860
'bzrlib.repofmt.knitrepo',
2861
'RepositoryFormatKnit1',
2864
format_registry.register_lazy(
2865
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
2866
'bzrlib.repofmt.knitrepo',
2867
'RepositoryFormatKnit3',
2870
format_registry.register_lazy(
2871
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
2872
'bzrlib.repofmt.knitrepo',
2873
'RepositoryFormatKnit4',
2876
# Pack-based formats. There is one format for pre-subtrees, and one for
2877
# post-subtrees to allow ease of testing.
2878
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
2879
format_registry.register_lazy(
2880
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
2881
'bzrlib.repofmt.pack_repo',
2882
'RepositoryFormatKnitPack1',
2884
format_registry.register_lazy(
2885
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
2886
'bzrlib.repofmt.pack_repo',
2887
'RepositoryFormatKnitPack3',
2889
format_registry.register_lazy(
2890
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
2891
'bzrlib.repofmt.pack_repo',
2892
'RepositoryFormatKnitPack4',
2894
format_registry.register_lazy(
2895
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
2896
'bzrlib.repofmt.pack_repo',
2897
'RepositoryFormatKnitPack5',
2899
format_registry.register_lazy(
2900
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
2901
'bzrlib.repofmt.pack_repo',
2902
'RepositoryFormatKnitPack5RichRoot',
2904
format_registry.register_lazy(
2905
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
2906
'bzrlib.repofmt.pack_repo',
2907
'RepositoryFormatKnitPack5RichRootBroken',
2909
format_registry.register_lazy(
2910
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
2911
'bzrlib.repofmt.pack_repo',
2912
'RepositoryFormatKnitPack6',
2914
format_registry.register_lazy(
2915
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
2916
'bzrlib.repofmt.pack_repo',
2917
'RepositoryFormatKnitPack6RichRoot',
2920
# Development formats.
2921
# 1.7->1.8 go below here
2922
format_registry.register_lazy(
2923
"Bazaar development format 2 (needs bzr.dev from before 1.8)\n",
2924
'bzrlib.repofmt.pack_repo',
2925
'RepositoryFormatPackDevelopment2',
2927
format_registry.register_lazy(
2928
("Bazaar development format 2 with subtree support "
2929
"(needs bzr.dev from before 1.8)\n"),
2930
'bzrlib.repofmt.pack_repo',
2931
'RepositoryFormatPackDevelopment2Subtree',
2935
class InterRepository(InterObject):
2936
"""This class represents operations taking place between two repositories.
2938
Its instances have methods like copy_content and fetch, and contain
2939
references to the source and target repositories these operations can be
2942
Often we will provide convenience methods on 'repository' which carry out
2943
operations with another repository - they will always forward to
2944
InterRepository.get(other).method_name(parameters).
2947
_walk_to_common_revisions_batch_size = 50
2949
"""The available optimised InterRepository types."""
2952
def copy_content(self, revision_id=None):
2953
"""Make a complete copy of the content in self into destination.
2955
This is a destructive operation! Do not use it on existing
2958
:param revision_id: Only copy the content needed to construct
2959
revision_id and its parents.
2962
self.target.set_make_working_trees(self.source.make_working_trees())
2963
except NotImplementedError:
2965
self.target.fetch(self.source, revision_id=revision_id)
2968
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
2970
"""Fetch the content required to construct revision_id.
2972
The content is copied from self.source to self.target.
2974
:param revision_id: if None all content is copied, if NULL_REVISION no
2976
:param pb: optional progress bar to use for progress reports. If not
2977
provided a default one will be created.
2980
from bzrlib.fetch import RepoFetcher
2981
f = RepoFetcher(to_repository=self.target,
2982
from_repository=self.source,
2983
last_revision=revision_id,
2984
fetch_spec=fetch_spec,
2985
pb=pb, find_ghosts=find_ghosts)
2987
def _walk_to_common_revisions(self, revision_ids):
2988
"""Walk out from revision_ids in source to revisions target has.
2990
:param revision_ids: The start point for the search.
2991
:return: A set of revision ids.
2993
target_graph = self.target.get_graph()
2994
revision_ids = frozenset(revision_ids)
2995
# Fast path for the case where all the revisions are already in the
2997
# (Although this does incur an extra round trip for the
2998
# fairly common case where the target doesn't already have the revision
3000
if set(target_graph.get_parent_map(revision_ids)) == revision_ids:
3001
return graph.SearchResult(revision_ids, set(), 0, set())
3002
missing_revs = set()
3003
source_graph = self.source.get_graph()
3004
# ensure we don't pay silly lookup costs.
3005
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3006
null_set = frozenset([_mod_revision.NULL_REVISION])
3007
searcher_exhausted = False
3011
# Iterate the searcher until we have enough next_revs
3012
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3014
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3015
next_revs.update(next_revs_part)
3016
ghosts.update(ghosts_part)
3017
except StopIteration:
3018
searcher_exhausted = True
3020
# If there are ghosts in the source graph, and the caller asked for
3021
# them, make sure that they are present in the target.
3022
# We don't care about other ghosts as we can't fetch them and
3023
# haven't been asked to.
3024
ghosts_to_check = set(revision_ids.intersection(ghosts))
3025
revs_to_get = set(next_revs).union(ghosts_to_check)
3027
have_revs = set(target_graph.get_parent_map(revs_to_get))
3028
# we always have NULL_REVISION present.
3029
have_revs = have_revs.union(null_set)
3030
# Check if the target is missing any ghosts we need.
3031
ghosts_to_check.difference_update(have_revs)
3033
# One of the caller's revision_ids is a ghost in both the
3034
# source and the target.
3035
raise errors.NoSuchRevision(
3036
self.source, ghosts_to_check.pop())
3037
missing_revs.update(next_revs - have_revs)
3038
# Because we may have walked past the original stop point, make
3039
# sure everything is stopped
3040
stop_revs = searcher.find_seen_ancestors(have_revs)
3041
searcher.stop_searching_any(stop_revs)
3042
if searcher_exhausted:
3044
return searcher.get_result()
3047
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3048
"""Return the revision ids that source has that target does not.
3050
:param revision_id: only return revision ids included by this
3052
:param find_ghosts: If True find missing revisions in deep history
3053
rather than just finding the surface difference.
3054
:return: A bzrlib.graph.SearchResult.
3056
# stop searching at found target revisions.
3057
if not find_ghosts and revision_id is not None:
3058
return self._walk_to_common_revisions([revision_id])
3059
# generic, possibly worst case, slow code path.
3060
target_ids = set(self.target.all_revision_ids())
3061
if revision_id is not None:
3062
source_ids = self.source.get_ancestry(revision_id)
3063
if source_ids[0] is not None:
3064
raise AssertionError()
3067
source_ids = self.source.all_revision_ids()
3068
result_set = set(source_ids).difference(target_ids)
3069
return self.source.revision_ids_to_search_result(result_set)
3072
def _same_model(source, target):
3073
"""True if source and target have the same data representation.
3075
Note: this is always called on the base class; overriding it in a
3076
subclass will have no effect.
3079
InterRepository._assert_same_model(source, target)
3081
except errors.IncompatibleRepositories, e:
3085
def _assert_same_model(source, target):
3086
"""Raise an exception if two repositories do not use the same model.
3088
if source.supports_rich_root() != target.supports_rich_root():
3089
raise errors.IncompatibleRepositories(source, target,
3090
"different rich-root support")
3091
if source._serializer != target._serializer:
3092
raise errors.IncompatibleRepositories(source, target,
3093
"different serializers")
3096
class InterSameDataRepository(InterRepository):
3097
"""Code for converting between repositories that represent the same data.
3099
Data format and model must match for this to work.
3103
def _get_repo_format_to_test(self):
3104
"""Repository format for testing with.
3106
InterSameData can pull from subtree to subtree and from non-subtree to
3107
non-subtree, so we test this with the richest repository format.
3109
from bzrlib.repofmt import knitrepo
3110
return knitrepo.RepositoryFormatKnit3()
3113
def is_compatible(source, target):
3114
return InterRepository._same_model(source, target)
3117
class InterWeaveRepo(InterSameDataRepository):
3118
"""Optimised code paths between Weave based repositories.
3120
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3121
implemented lazy inter-object optimisation.
3125
def _get_repo_format_to_test(self):
3126
from bzrlib.repofmt import weaverepo
3127
return weaverepo.RepositoryFormat7()
3130
def is_compatible(source, target):
3131
"""Be compatible with known Weave formats.
3133
We don't test for the stores being of specific types because that
3134
could lead to confusing results, and there is no need to be
3137
from bzrlib.repofmt.weaverepo import (
3143
return (isinstance(source._format, (RepositoryFormat5,
3145
RepositoryFormat7)) and
3146
isinstance(target._format, (RepositoryFormat5,
3148
RepositoryFormat7)))
3149
except AttributeError:
3153
def copy_content(self, revision_id=None):
3154
"""See InterRepository.copy_content()."""
3155
# weave specific optimised path:
3157
self.target.set_make_working_trees(self.source.make_working_trees())
3158
except (errors.RepositoryUpgradeRequired, NotImplemented):
3160
# FIXME do not peek!
3161
if self.source._transport.listable():
3162
pb = ui.ui_factory.nested_progress_bar()
3164
self.target.texts.insert_record_stream(
3165
self.source.texts.get_record_stream(
3166
self.source.texts.keys(), 'topological', False))
3167
pb.update('copying inventory', 0, 1)
3168
self.target.inventories.insert_record_stream(
3169
self.source.inventories.get_record_stream(
3170
self.source.inventories.keys(), 'topological', False))
3171
self.target.signatures.insert_record_stream(
3172
self.source.signatures.get_record_stream(
3173
self.source.signatures.keys(),
3175
self.target.revisions.insert_record_stream(
3176
self.source.revisions.get_record_stream(
3177
self.source.revisions.keys(),
3178
'topological', True))
3182
self.target.fetch(self.source, revision_id=revision_id)
3185
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3186
"""See InterRepository.missing_revision_ids()."""
3187
# we want all revisions to satisfy revision_id in source.
3188
# but we don't want to stat every file here and there.
3189
# we want then, all revisions other needs to satisfy revision_id
3190
# checked, but not those that we have locally.
3191
# so the first thing is to get a subset of the revisions to
3192
# satisfy revision_id in source, and then eliminate those that
3193
# we do already have.
3194
# this is slow on high latency connection to self, but as as this
3195
# disk format scales terribly for push anyway due to rewriting
3196
# inventory.weave, this is considered acceptable.
3198
if revision_id is not None:
3199
source_ids = self.source.get_ancestry(revision_id)
3200
if source_ids[0] is not None:
3201
raise AssertionError()
3204
source_ids = self.source._all_possible_ids()
3205
source_ids_set = set(source_ids)
3206
# source_ids is the worst possible case we may need to pull.
3207
# now we want to filter source_ids against what we actually
3208
# have in target, but don't try to check for existence where we know
3209
# we do not have a revision as that would be pointless.
3210
target_ids = set(self.target._all_possible_ids())
3211
possibly_present_revisions = target_ids.intersection(source_ids_set)
3212
actually_present_revisions = set(
3213
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3214
required_revisions = source_ids_set.difference(actually_present_revisions)
3215
if revision_id is not None:
3216
# we used get_ancestry to determine source_ids then we are assured all
3217
# revisions referenced are present as they are installed in topological order.
3218
# and the tip revision was validated by get_ancestry.
3219
result_set = required_revisions
3221
# if we just grabbed the possibly available ids, then
3222
# we only have an estimate of whats available and need to validate
3223
# that against the revision records.
3225
self.source._eliminate_revisions_not_present(required_revisions))
3226
return self.source.revision_ids_to_search_result(result_set)
3229
class InterKnitRepo(InterSameDataRepository):
3230
"""Optimised code paths between Knit based repositories."""
3233
def _get_repo_format_to_test(self):
3234
from bzrlib.repofmt import knitrepo
3235
return knitrepo.RepositoryFormatKnit1()
3238
def is_compatible(source, target):
3239
"""Be compatible with known Knit formats.
3241
We don't test for the stores being of specific types because that
3242
could lead to confusing results, and there is no need to be
3245
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3247
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3248
isinstance(target._format, RepositoryFormatKnit))
3249
except AttributeError:
3251
return are_knits and InterRepository._same_model(source, target)
3254
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3255
"""See InterRepository.missing_revision_ids()."""
3256
if revision_id is not None:
3257
source_ids = self.source.get_ancestry(revision_id)
3258
if source_ids[0] is not None:
3259
raise AssertionError()
3262
source_ids = self.source.all_revision_ids()
3263
source_ids_set = set(source_ids)
3264
# source_ids is the worst possible case we may need to pull.
3265
# now we want to filter source_ids against what we actually
3266
# have in target, but don't try to check for existence where we know
3267
# we do not have a revision as that would be pointless.
3268
target_ids = set(self.target.all_revision_ids())
3269
possibly_present_revisions = target_ids.intersection(source_ids_set)
3270
actually_present_revisions = set(
3271
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3272
required_revisions = source_ids_set.difference(actually_present_revisions)
3273
if revision_id is not None:
3274
# we used get_ancestry to determine source_ids then we are assured all
3275
# revisions referenced are present as they are installed in topological order.
3276
# and the tip revision was validated by get_ancestry.
3277
result_set = required_revisions
3279
# if we just grabbed the possibly available ids, then
3280
# we only have an estimate of whats available and need to validate
3281
# that against the revision records.
3283
self.source._eliminate_revisions_not_present(required_revisions))
3284
return self.source.revision_ids_to_search_result(result_set)
3287
class InterPackRepo(InterSameDataRepository):
3288
"""Optimised code paths between Pack based repositories."""
3291
def _get_repo_format_to_test(self):
3292
from bzrlib.repofmt import pack_repo
3293
return pack_repo.RepositoryFormatKnitPack1()
3296
def is_compatible(source, target):
3297
"""Be compatible with known Pack formats.
3299
We don't test for the stores being of specific types because that
3300
could lead to confusing results, and there is no need to be
3303
from bzrlib.repofmt.pack_repo import RepositoryFormatPack
3305
are_packs = (isinstance(source._format, RepositoryFormatPack) and
3306
isinstance(target._format, RepositoryFormatPack))
3307
except AttributeError:
3309
return are_packs and InterRepository._same_model(source, target)
3312
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3314
"""See InterRepository.fetch()."""
3315
if (len(self.source._fallback_repositories) > 0 or
3316
len(self.target._fallback_repositories) > 0):
3317
# The pack layer is not aware of fallback repositories, so when
3318
# fetching from a stacked repository or into a stacked repository
3319
# we use the generic fetch logic which uses the VersionedFiles
3320
# attributes on repository.
3321
from bzrlib.fetch import RepoFetcher
3322
fetcher = RepoFetcher(self.target, self.source, revision_id,
3323
pb, find_ghosts, fetch_spec=fetch_spec)
3324
if fetch_spec is not None:
3325
if len(list(fetch_spec.heads)) != 1:
3326
raise AssertionError(
3327
"InterPackRepo.fetch doesn't support "
3328
"fetching multiple heads yet.")
3329
revision_id = list(fetch_spec.heads)[0]
3331
if revision_id is None:
3333
# everything to do - use pack logic
3334
# to fetch from all packs to one without
3335
# inventory parsing etc, IFF nothing to be copied is in the target.
3337
source_revision_ids = frozenset(self.source.all_revision_ids())
3338
revision_ids = source_revision_ids - \
3339
frozenset(self.target.get_parent_map(source_revision_ids))
3340
revision_keys = [(revid,) for revid in revision_ids]
3341
index = self.target._pack_collection.revision_index.combined_index
3342
present_revision_ids = set(item[1][0] for item in
3343
index.iter_entries(revision_keys))
3344
revision_ids = set(revision_ids) - present_revision_ids
3345
# implementing the TODO will involve:
3346
# - detecting when all of a pack is selected
3347
# - avoiding as much as possible pre-selection, so the
3348
# more-core routines such as create_pack_from_packs can filter in
3349
# a just-in-time fashion. (though having a HEADS list on a
3350
# repository might make this a lot easier, because we could
3351
# sensibly detect 'new revisions' without doing a full index scan.
3352
elif _mod_revision.is_null(revision_id):
3357
revision_ids = self.search_missing_revision_ids(revision_id,
3358
find_ghosts=find_ghosts).get_keys()
3359
except errors.NoSuchRevision:
3360
raise errors.InstallFailed([revision_id])
3361
if len(revision_ids) == 0:
3363
return self._pack(self.source, self.target, revision_ids)
3365
def _pack(self, source, target, revision_ids):
3366
from bzrlib.repofmt.pack_repo import Packer
3367
packs = source._pack_collection.all_packs()
3368
pack = Packer(self.target._pack_collection, packs, '.fetch',
3369
revision_ids).pack()
3370
if pack is not None:
3371
self.target._pack_collection._save_pack_names()
3372
copied_revs = pack.get_revision_count()
3373
# Trigger an autopack. This may duplicate effort as we've just done
3374
# a pack creation, but for now it is simpler to think about as
3375
# 'upload data, then repack if needed'.
3376
self.target._pack_collection.autopack()
3377
return (copied_revs, [])
3382
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3383
"""See InterRepository.missing_revision_ids().
3385
:param find_ghosts: Find ghosts throughout the ancestry of
3388
if not find_ghosts and revision_id is not None:
3389
return self._walk_to_common_revisions([revision_id])
3390
elif revision_id is not None:
3391
# Find ghosts: search for revisions pointing from one repository to
3392
# the other, and vice versa, anywhere in the history of revision_id.
3393
graph = self.target.get_graph(other_repository=self.source)
3394
searcher = graph._make_breadth_first_searcher([revision_id])
3398
next_revs, ghosts = searcher.next_with_ghosts()
3399
except StopIteration:
3401
if revision_id in ghosts:
3402
raise errors.NoSuchRevision(self.source, revision_id)
3403
found_ids.update(next_revs)
3404
found_ids.update(ghosts)
3405
found_ids = frozenset(found_ids)
3406
# Double query here: should be able to avoid this by changing the
3407
# graph api further.
3408
result_set = found_ids - frozenset(
3409
self.target.get_parent_map(found_ids))
3411
source_ids = self.source.all_revision_ids()
3412
# source_ids is the worst possible case we may need to pull.
3413
# now we want to filter source_ids against what we actually
3414
# have in target, but don't try to check for existence where we know
3415
# we do not have a revision as that would be pointless.
3416
target_ids = set(self.target.all_revision_ids())
3417
result_set = set(source_ids).difference(target_ids)
3418
return self.source.revision_ids_to_search_result(result_set)
3421
class InterDifferingSerializer(InterKnitRepo):
3424
def _get_repo_format_to_test(self):
3428
def is_compatible(source, target):
3429
"""Be compatible with Knit2 source and Knit3 target"""
3430
if source.supports_rich_root() != target.supports_rich_root():
3432
# Ideally, we'd support fetching if the source had no tree references
3433
# even if it supported them...
3434
if (getattr(source, '_format.supports_tree_reference', False) and
3435
not getattr(target, '_format.supports_tree_reference', False)):
3439
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3440
"""Get the best delta and base for this revision.
3442
:return: (basis_id, delta)
3444
possible_trees = [(parent_id, cache[parent_id])
3445
for parent_id in parent_ids
3446
if parent_id in cache]
3447
if len(possible_trees) == 0:
3448
# There either aren't any parents, or the parents aren't in the
3449
# cache, so just use the last converted tree
3450
possible_trees.append((basis_id, cache[basis_id]))
3452
for basis_id, basis_tree in possible_trees:
3453
delta = tree.inventory._make_delta(basis_tree.inventory)
3454
deltas.append((len(delta), basis_id, delta))
3456
return deltas[0][1:]
3458
def _fetch_batch(self, revision_ids, basis_id, cache):
3459
"""Fetch across a few revisions.
3461
:param revision_ids: The revisions to copy
3462
:param basis_id: The revision_id of a tree that must be in cache, used
3463
as a basis for delta when no other base is available
3464
:param cache: A cache of RevisionTrees that we can use.
3465
:return: The revision_id of the last converted tree. The RevisionTree
3466
for it will be in cache
3468
# Walk though all revisions; get inventory deltas, copy referenced
3469
# texts that delta references, insert the delta, revision and
3473
pending_revisions = []
3474
parent_map = self.source.get_parent_map(revision_ids)
3475
for tree in self.source.revision_trees(revision_ids):
3476
current_revision_id = tree.get_revision_id()
3477
parent_ids = parent_map.get(current_revision_id, ())
3478
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3480
# Find text entries that need to be copied
3481
for old_path, new_path, file_id, entry in delta:
3482
if new_path is not None:
3483
if not (new_path or self.target.supports_rich_root()):
3484
# We don't copy the text for the root node unless the
3485
# target supports_rich_root.
3487
text_keys.add((file_id, entry.revision))
3488
revision = self.source.get_revision(current_revision_id)
3489
pending_deltas.append((basis_id, delta,
3490
current_revision_id, revision.parent_ids))
3491
pending_revisions.append(revision)
3492
cache[current_revision_id] = tree
3493
basis_id = current_revision_id
3495
from_texts = self.source.texts
3496
to_texts = self.target.texts
3497
to_texts.insert_record_stream(from_texts.get_record_stream(
3498
text_keys, self.target._format._fetch_order,
3499
not self.target._format._fetch_uses_deltas))
3501
for delta in pending_deltas:
3502
self.target.add_inventory_by_delta(*delta)
3503
# insert signatures and revisions
3504
for revision in pending_revisions:
3506
signature = self.source.get_signature_text(
3507
revision.revision_id)
3508
self.target.add_signature_text(revision.revision_id,
3510
except errors.NoSuchRevision:
3512
self.target.add_revision(revision.revision_id, revision)
3515
def _fetch_all_revisions(self, revision_ids, pb):
3516
"""Fetch everything for the list of revisions.
3518
:param revision_ids: The list of revisions to fetch. Must be in
3520
:param pb: A ProgressBar
3523
basis_id, basis_tree = self._get_basis(revision_ids[0])
3525
cache = lru_cache.LRUCache(100)
3526
cache[basis_id] = basis_tree
3527
del basis_tree # We don't want to hang on to it here
3528
for offset in range(0, len(revision_ids), batch_size):
3529
self.target.start_write_group()
3531
pb.update('Transferring revisions', offset,
3533
batch = revision_ids[offset:offset+batch_size]
3534
basis_id = self._fetch_batch(batch, basis_id, cache)
3536
self.target.abort_write_group()
3539
self.target.commit_write_group()
3540
pb.update('Transferring revisions', len(revision_ids),
3544
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3546
"""See InterRepository.fetch()."""
3547
if fetch_spec is not None:
3548
raise AssertionError("Not implemented yet...")
3549
revision_ids = self.target.search_missing_revision_ids(self.source,
3550
revision_id, find_ghosts=find_ghosts).get_keys()
3551
if not revision_ids:
3553
revision_ids = tsort.topo_sort(
3554
self.source.get_graph().get_parent_map(revision_ids))
3556
my_pb = ui.ui_factory.nested_progress_bar()
3559
symbol_versioning.warn(
3560
symbol_versioning.deprecated_in((1, 14, 0))
3561
% "pb parameter to fetch()")
3564
self._fetch_all_revisions(revision_ids, pb)
3566
if my_pb is not None:
3568
return len(revision_ids), 0
3570
def _get_basis(self, first_revision_id):
3571
"""Get a revision and tree which exists in the target.
3573
This assumes that first_revision_id is selected for transmission
3574
because all other ancestors are already present. If we can't find an
3575
ancestor we fall back to NULL_REVISION since we know that is safe.
3577
:return: (basis_id, basis_tree)
3579
first_rev = self.source.get_revision(first_revision_id)
3581
basis_id = first_rev.parent_ids[0]
3582
# only valid as a basis if the target has it
3583
self.target.get_revision(basis_id)
3584
# Try to get a basis tree - if its a ghost it will hit the
3585
# NoSuchRevision case.
3586
basis_tree = self.source.revision_tree(basis_id)
3587
except (IndexError, errors.NoSuchRevision):
3588
basis_id = _mod_revision.NULL_REVISION
3589
basis_tree = self.source.revision_tree(basis_id)
3590
return basis_id, basis_tree
3593
InterRepository.register_optimiser(InterDifferingSerializer)
3594
InterRepository.register_optimiser(InterSameDataRepository)
3595
InterRepository.register_optimiser(InterWeaveRepo)
3596
InterRepository.register_optimiser(InterKnitRepo)
3597
InterRepository.register_optimiser(InterPackRepo)
3600
class CopyConverter(object):
3601
"""A repository conversion tool which just performs a copy of the content.
3603
This is slow but quite reliable.
3606
def __init__(self, target_format):
3607
"""Create a CopyConverter.
3609
:param target_format: The format the resulting repository should be.
3611
self.target_format = target_format
3613
def convert(self, repo, pb):
3614
"""Perform the conversion of to_convert, giving feedback via pb.
3616
:param to_convert: The disk object to convert.
3617
:param pb: a progress bar to use for progress information.
3622
# this is only useful with metadir layouts - separated repo content.
3623
# trigger an assertion if not such
3624
repo._format.get_format_string()
3625
self.repo_dir = repo.bzrdir
3626
self.step('Moving repository to repository.backup')
3627
self.repo_dir.transport.move('repository', 'repository.backup')
3628
backup_transport = self.repo_dir.transport.clone('repository.backup')
3629
repo._format.check_conversion_target(self.target_format)
3630
self.source_repo = repo._format.open(self.repo_dir,
3632
_override_transport=backup_transport)
3633
self.step('Creating new repository')
3634
converted = self.target_format.initialize(self.repo_dir,
3635
self.source_repo.is_shared())
3636
converted.lock_write()
3638
self.step('Copying content into repository.')
3639
self.source_repo.copy_content_into(converted)
3642
self.step('Deleting old repository content.')
3643
self.repo_dir.transport.delete_tree('repository.backup')
3644
self.pb.note('repository converted')
3646
def step(self, message):
3647
"""Update the pb by a step."""
3649
self.pb.update(message, self.count, self.total)
3661
def _unescaper(match, _map=_unescape_map):
3662
code = match.group(1)
3666
if not code.startswith('#'):
3668
return unichr(int(code[1:])).encode('utf8')
3674
def _unescape_xml(data):
3675
"""Unescape predefined XML entities in a string of data."""
3677
if _unescape_re is None:
3678
_unescape_re = re.compile('\&([^;]*);')
3679
return _unescape_re.sub(_unescaper, data)
3682
class _VersionedFileChecker(object):
3684
def __init__(self, repository, text_key_references=None):
3685
self.repository = repository
3686
self.text_index = self.repository._generate_text_key_index(
3687
text_key_references=text_key_references)
3689
def calculate_file_version_parents(self, text_key):
3690
"""Calculate the correct parents for a file version according to
3693
parent_keys = self.text_index[text_key]
3694
if parent_keys == [_mod_revision.NULL_REVISION]:
3696
return tuple(parent_keys)
3698
def check_file_version_parents(self, texts, progress_bar=None):
3699
"""Check the parents stored in a versioned file are correct.
3701
It also detects file versions that are not referenced by their
3702
corresponding revision's inventory.
3704
:returns: A tuple of (wrong_parents, dangling_file_versions).
3705
wrong_parents is a dict mapping {revision_id: (stored_parents,
3706
correct_parents)} for each revision_id where the stored parents
3707
are not correct. dangling_file_versions is a set of (file_id,
3708
revision_id) tuples for versions that are present in this versioned
3709
file, but not used by the corresponding inventory.
3712
self.file_ids = set([file_id for file_id, _ in
3713
self.text_index.iterkeys()])
3714
# text keys is now grouped by file_id
3715
n_weaves = len(self.file_ids)
3716
files_in_revisions = {}
3717
revisions_of_files = {}
3718
n_versions = len(self.text_index)
3719
progress_bar.update('loading text store', 0, n_versions)
3720
parent_map = self.repository.texts.get_parent_map(self.text_index)
3721
# On unlistable transports this could well be empty/error...
3722
text_keys = self.repository.texts.keys()
3723
unused_keys = frozenset(text_keys) - set(self.text_index)
3724
for num, key in enumerate(self.text_index.iterkeys()):
3725
if progress_bar is not None:
3726
progress_bar.update('checking text graph', num, n_versions)
3727
correct_parents = self.calculate_file_version_parents(key)
3729
knit_parents = parent_map[key]
3730
except errors.RevisionNotPresent:
3733
if correct_parents != knit_parents:
3734
wrong_parents[key] = (knit_parents, correct_parents)
3735
return wrong_parents, unused_keys
3738
def _old_get_graph(repository, revision_id):
3739
"""DO NOT USE. That is all. I'm serious."""
3740
graph = repository.get_graph()
3741
revision_graph = dict(((key, value) for key, value in
3742
graph.iter_ancestry([revision_id]) if value is not None))
3743
return _strip_NULL_ghosts(revision_graph)
3746
def _strip_NULL_ghosts(revision_graph):
3747
"""Also don't use this. more compatibility code for unmigrated clients."""
3748
# Filter ghosts, and null:
3749
if _mod_revision.NULL_REVISION in revision_graph:
3750
del revision_graph[_mod_revision.NULL_REVISION]
3751
for key, parents in revision_graph.items():
3752
revision_graph[key] = tuple(parent for parent in parents if parent
3754
return revision_graph
3757
class StreamSink(object):
3758
"""An object that can insert a stream into a repository.
3760
This interface handles the complexity of reserialising inventories and
3761
revisions from different formats, and allows unidirectional insertion into
3762
stacked repositories without looking for the missing basis parents
3766
def __init__(self, target_repo):
3767
self.target_repo = target_repo
3769
def insert_stream(self, stream, src_format, resume_tokens):
3770
"""Insert a stream's content into the target repository.
3772
:param src_format: a bzr repository format.
3774
:return: a list of resume tokens and an iterable of keys additional
3775
items required before the insertion can be completed.
3777
self.target_repo.lock_write()
3780
self.target_repo.resume_write_group(resume_tokens)
3782
self.target_repo.start_write_group()
3784
# locked_insert_stream performs a commit|suspend.
3785
return self._locked_insert_stream(stream, src_format)
3787
self.target_repo.abort_write_group(suppress_errors=True)
3790
self.target_repo.unlock()
3792
def _locked_insert_stream(self, stream, src_format):
3793
to_serializer = self.target_repo._format._serializer
3794
src_serializer = src_format._serializer
3795
if to_serializer == src_serializer:
3796
# If serializers match and the target is a pack repository, set the
3797
# write cache size on the new pack. This avoids poor performance
3798
# on transports where append is unbuffered (such as
3799
# RemoteTransport). This is safe to do because nothing should read
3800
# back from the target repository while a stream with matching
3801
# serialization is being inserted.
3802
# The exception is that a delta record from the source that should
3803
# be a fulltext may need to be expanded by the target (see
3804
# test_fetch_revisions_with_deltas_into_pack); but we take care to
3805
# explicitly flush any buffered writes first in that rare case.
3807
new_pack = self.target_repo._pack_collection._new_pack
3808
except AttributeError:
3809
# Not a pack repository
3812
new_pack.set_write_cache_size(1024*1024)
3813
for substream_type, substream in stream:
3814
if substream_type == 'texts':
3815
self.target_repo.texts.insert_record_stream(substream)
3816
elif substream_type == 'inventories':
3817
if src_serializer == to_serializer:
3818
self.target_repo.inventories.insert_record_stream(
3821
self._extract_and_insert_inventories(
3822
substream, src_serializer)
3823
elif substream_type == 'revisions':
3824
# This may fallback to extract-and-insert more often than
3825
# required if the serializers are different only in terms of
3827
if src_serializer == to_serializer:
3828
self.target_repo.revisions.insert_record_stream(
3831
self._extract_and_insert_revisions(substream,
3833
elif substream_type == 'signatures':
3834
self.target_repo.signatures.insert_record_stream(substream)
3836
raise AssertionError('kaboom! %s' % (substream_type,))
3838
missing_keys = set()
3839
for prefix, versioned_file in (
3840
('texts', self.target_repo.texts),
3841
('inventories', self.target_repo.inventories),
3842
('revisions', self.target_repo.revisions),
3843
('signatures', self.target_repo.signatures),
3845
missing_keys.update((prefix,) + key for key in
3846
versioned_file.get_missing_compression_parent_keys())
3847
except NotImplementedError:
3848
# cannot even attempt suspending, and missing would have failed
3849
# during stream insertion.
3850
missing_keys = set()
3853
# suspend the write group and tell the caller what we is
3854
# missing. We know we can suspend or else we would not have
3855
# entered this code path. (All repositories that can handle
3856
# missing keys can handle suspending a write group).
3857
write_group_tokens = self.target_repo.suspend_write_group()
3858
return write_group_tokens, missing_keys
3859
self.target_repo.commit_write_group()
3862
def _extract_and_insert_inventories(self, substream, serializer):
3863
"""Generate a new inventory versionedfile in target, converting data.
3865
The inventory is retrieved from the source, (deserializing it), and
3866
stored in the target (reserializing it in a different format).
3868
for record in substream:
3869
bytes = record.get_bytes_as('fulltext')
3870
revision_id = record.key[0]
3871
inv = serializer.read_inventory_from_string(bytes, revision_id)
3872
parents = [key[0] for key in record.parents]
3873
self.target_repo.add_inventory(revision_id, inv, parents)
3875
def _extract_and_insert_revisions(self, substream, serializer):
3876
for record in substream:
3877
bytes = record.get_bytes_as('fulltext')
3878
revision_id = record.key[0]
3879
rev = serializer.read_revision_from_string(bytes)
3880
if rev.revision_id != revision_id:
3881
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
3882
self.target_repo.add_revision(revision_id, rev)
3885
if self.target_repo._format._fetch_reconcile:
3886
self.target_repo.reconcile()
3889
class StreamSource(object):
3890
"""A source of a stream for fetching between repositories."""
3892
def __init__(self, from_repository, to_format):
3893
"""Create a StreamSource streaming from from_repository."""
3894
self.from_repository = from_repository
3895
self.to_format = to_format
3897
def delta_on_metadata(self):
3898
"""Return True if delta's are permitted on metadata streams.
3900
That is on revisions and signatures.
3902
src_serializer = self.from_repository._format._serializer
3903
target_serializer = self.to_format._serializer
3904
return (self.to_format._fetch_uses_deltas and
3905
src_serializer == target_serializer)
3907
def _fetch_revision_texts(self, revs):
3908
# fetch signatures first and then the revision texts
3909
# may need to be a InterRevisionStore call here.
3910
from_sf = self.from_repository.signatures
3911
# A missing signature is just skipped.
3912
keys = [(rev_id,) for rev_id in revs]
3913
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
3915
self.to_format._fetch_order,
3916
not self.to_format._fetch_uses_deltas))
3917
# If a revision has a delta, this is actually expanded inside the
3918
# insert_record_stream code now, which is an alternate fix for
3920
from_rf = self.from_repository.revisions
3921
revisions = from_rf.get_record_stream(
3923
self.to_format._fetch_order,
3924
not self.delta_on_metadata())
3925
return [('signatures', signatures), ('revisions', revisions)]
3927
def _generate_root_texts(self, revs):
3928
"""This will be called by __fetch between fetching weave texts and
3929
fetching the inventory weave.
3931
Subclasses should override this if they need to generate root texts
3932
after fetching weave texts.
3934
if self._rich_root_upgrade():
3936
return bzrlib.fetch.Inter1and2Helper(
3937
self.from_repository).generate_root_texts(revs)
3941
def get_stream(self, search):
3943
revs = search.get_keys()
3944
graph = self.from_repository.get_graph()
3945
revs = list(graph.iter_topo_order(revs))
3946
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
3948
for knit_kind, file_id, revisions in data_to_fetch:
3949
if knit_kind != phase:
3951
# Make a new progress bar for this phase
3952
if knit_kind == "file":
3953
# Accumulate file texts
3954
text_keys.extend([(file_id, revision) for revision in
3956
elif knit_kind == "inventory":
3957
# Now copy the file texts.
3958
from_texts = self.from_repository.texts
3959
yield ('texts', from_texts.get_record_stream(
3960
text_keys, self.to_format._fetch_order,
3961
not self.to_format._fetch_uses_deltas))
3962
# Cause an error if a text occurs after we have done the
3965
# Before we process the inventory we generate the root
3966
# texts (if necessary) so that the inventories references
3968
for _ in self._generate_root_texts(revs):
3970
# NB: This currently reopens the inventory weave in source;
3971
# using a single stream interface instead would avoid this.
3972
from_weave = self.from_repository.inventories
3973
# we fetch only the referenced inventories because we do not
3974
# know for unselected inventories whether all their required
3975
# texts are present in the other repository - it could be
3977
yield ('inventories', from_weave.get_record_stream(
3978
[(rev_id,) for rev_id in revs],
3979
self.inventory_fetch_order(),
3980
not self.delta_on_metadata()))
3981
elif knit_kind == "signatures":
3982
# Nothing to do here; this will be taken care of when
3983
# _fetch_revision_texts happens.
3985
elif knit_kind == "revisions":
3986
for record in self._fetch_revision_texts(revs):
3989
raise AssertionError("Unknown knit kind %r" % knit_kind)
3991
def get_stream_for_missing_keys(self, missing_keys):
3992
# missing keys can only occur when we are byte copying and not
3993
# translating (because translation means we don't send
3994
# unreconstructable deltas ever).
3996
keys['texts'] = set()
3997
keys['revisions'] = set()
3998
keys['inventories'] = set()
3999
keys['signatures'] = set()
4000
for key in missing_keys:
4001
keys[key[0]].add(key[1:])
4002
if len(keys['revisions']):
4003
# If we allowed copying revisions at this point, we could end up
4004
# copying a revision without copying its required texts: a
4005
# violation of the requirements for repository integrity.
4006
raise AssertionError(
4007
'cannot copy revisions to fill in missing deltas %s' % (
4008
keys['revisions'],))
4009
for substream_kind, keys in keys.iteritems():
4010
vf = getattr(self.from_repository, substream_kind)
4011
# Ask for full texts always so that we don't need more round trips
4012
# after this stream.
4013
stream = vf.get_record_stream(keys,
4014
self.to_format._fetch_order, True)
4015
yield substream_kind, stream
4017
def inventory_fetch_order(self):
4018
if self._rich_root_upgrade():
4019
return 'topological'
4021
return self.to_format._fetch_order
4023
def _rich_root_upgrade(self):
4024
return (not self.from_repository._format.rich_root_data and
4025
self.to_format.rich_root_data)
added
removed
bzrlib/repository.py
