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(), """
40
revision as _mod_revision,
46
from bzrlib.bundle import serializer
47
from bzrlib.revisiontree import RevisionTree
48
from bzrlib.store.versioned import VersionedFileStore
49
from bzrlib.testament import Testament
52
from bzrlib.decorators import needs_read_lock, needs_write_lock
53
from bzrlib.inter import InterObject
54
from bzrlib.inventory import (
60
from bzrlib import registry
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
text = file_obj.read()
499
ie.text_sha1, ie.text_size = self._add_text_to_weave(
500
ie.file_id, text, 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
518
self._add_text_to_weave(ie.file_id, '', heads, None)
519
elif kind == 'symlink':
520
current_link_target = content_summary[3]
522
# symlink target is not generic metadata, check if it has
524
if current_link_target != parent_entry.symlink_target:
527
# unchanged, carry over.
528
ie.revision = parent_entry.revision
529
ie.symlink_target = parent_entry.symlink_target
530
return self._get_delta(ie, basis_inv, path), False, None
531
ie.symlink_target = current_link_target
532
self._add_text_to_weave(ie.file_id, '', heads, None)
533
elif kind == 'tree-reference':
535
if content_summary[3] != parent_entry.reference_revision:
538
# unchanged, carry over.
539
ie.reference_revision = parent_entry.reference_revision
540
ie.revision = parent_entry.revision
541
return self._get_delta(ie, basis_inv, path), False, None
542
ie.reference_revision = content_summary[3]
543
self._add_text_to_weave(ie.file_id, '', heads, None)
545
raise NotImplementedError('unknown kind')
546
ie.revision = self._new_revision_id
547
self._any_changes = True
548
return self._get_delta(ie, basis_inv, path), True, fingerprint
550
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
551
_entry_factory=entry_factory):
552
"""Record a new tree via iter_changes.
554
:param tree: The tree to obtain text contents from for changed objects.
555
:param basis_revision_id: The revision id of the tree the iter_changes
556
has been generated against. Currently assumed to be the same
557
as self.parents[0] - if it is not, errors may occur.
558
:param iter_changes: An iter_changes iterator with the changes to apply
559
to basis_revision_id. The iterator must not include any items with
560
a current kind of None - missing items must be either filtered out
561
or errored-on beefore record_iter_changes sees the item.
562
:param _entry_factory: Private method to bind entry_factory locally for
564
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
567
# Create an inventory delta based on deltas between all the parents and
568
# deltas between all the parent inventories. We use inventory delta's
569
# between the inventory objects because iter_changes masks
570
# last-changed-field only changes.
572
# file_id -> change map, change is fileid, paths, changed, versioneds,
573
# parents, names, kinds, executables
575
# {file_id -> revision_id -> inventory entry, for entries in parent
576
# trees that are not parents[0]
580
revtrees = list(self.repository.revision_trees(self.parents))
581
except errors.NoSuchRevision:
582
# one or more ghosts, slow path.
584
for revision_id in self.parents:
586
revtrees.append(self.repository.revision_tree(revision_id))
587
except errors.NoSuchRevision:
589
basis_revision_id = _mod_revision.NULL_REVISION
591
revtrees.append(self.repository.revision_tree(
592
_mod_revision.NULL_REVISION))
593
# The basis inventory from a repository
595
basis_inv = revtrees[0].inventory
597
basis_inv = self.repository.revision_tree(
598
_mod_revision.NULL_REVISION).inventory
599
if len(self.parents) > 0:
600
if basis_revision_id != self.parents[0] and not ghost_basis:
602
"arbitrary basis parents not yet supported with merges")
603
for revtree in revtrees[1:]:
604
for change in revtree.inventory._make_delta(basis_inv):
605
if change[1] is None:
606
# Not present in this parent.
608
if change[2] not in merged_ids:
609
if change[0] is not None:
610
basis_entry = basis_inv[change[2]]
611
merged_ids[change[2]] = [
613
basis_entry.revision,
616
parent_entries[change[2]] = {
618
basis_entry.revision:basis_entry,
620
change[3].revision:change[3],
623
merged_ids[change[2]] = [change[3].revision]
624
parent_entries[change[2]] = {change[3].revision:change[3]}
626
merged_ids[change[2]].append(change[3].revision)
627
parent_entries[change[2]][change[3].revision] = change[3]
630
# Setup the changes from the tree:
631
# changes maps file_id -> (change, [parent revision_ids])
633
for change in iter_changes:
634
# This probably looks up in basis_inv way to much.
635
if change[1][0] is not None:
636
head_candidate = [basis_inv[change[0]].revision]
639
changes[change[0]] = change, merged_ids.get(change[0],
641
unchanged_merged = set(merged_ids) - set(changes)
642
# Extend the changes dict with synthetic changes to record merges of
644
for file_id in unchanged_merged:
645
# Record a merged version of these items that did not change vs the
646
# basis. This can be either identical parallel changes, or a revert
647
# of a specific file after a merge. The recorded content will be
648
# that of the current tree (which is the same as the basis), but
649
# the per-file graph will reflect a merge.
650
# NB:XXX: We are reconstructing path information we had, this
651
# should be preserved instead.
652
# inv delta change: (file_id, (path_in_source, path_in_target),
653
# changed_content, versioned, parent, name, kind,
656
basis_entry = basis_inv[file_id]
657
except errors.NoSuchId:
658
# a change from basis->some_parents but file_id isn't in basis
659
# so was new in the merge, which means it must have changed
660
# from basis -> current, and as it hasn't the add was reverted
661
# by the user. So we discard this change.
665
(basis_inv.id2path(file_id), tree.id2path(file_id)),
667
(basis_entry.parent_id, basis_entry.parent_id),
668
(basis_entry.name, basis_entry.name),
669
(basis_entry.kind, basis_entry.kind),
670
(basis_entry.executable, basis_entry.executable))
671
changes[file_id] = (change, merged_ids[file_id])
672
# changes contains tuples with the change and a set of inventory
673
# candidates for the file.
675
# old_path, new_path, file_id, new_inventory_entry
676
seen_root = False # Is the root in the basis delta?
677
inv_delta = self._basis_delta
678
modified_rev = self._new_revision_id
679
for change, head_candidates in changes.values():
680
if change[3][1]: # versioned in target.
681
# Several things may be happening here:
682
# We may have a fork in the per-file graph
683
# - record a change with the content from tree
684
# We may have a change against < all trees
685
# - carry over the tree that hasn't changed
686
# We may have a change against all trees
687
# - record the change with the content from tree
690
entry = _entry_factory[kind](file_id, change[5][1],
692
head_set = self._heads(change[0], set(head_candidates))
695
for head_candidate in head_candidates:
696
if head_candidate in head_set:
697
heads.append(head_candidate)
698
head_set.remove(head_candidate)
701
# Could be a carry-over situation:
702
parent_entry_revs = parent_entries.get(file_id, None)
703
if parent_entry_revs:
704
parent_entry = parent_entry_revs.get(heads[0], None)
707
if parent_entry is None:
708
# The parent iter_changes was called against is the one
709
# that is the per-file head, so any change is relevant
710
# iter_changes is valid.
711
carry_over_possible = False
713
# could be a carry over situation
714
# A change against the basis may just indicate a merge,
715
# we need to check the content against the source of the
716
# merge to determine if it was changed after the merge
718
if (parent_entry.kind != entry.kind or
719
parent_entry.parent_id != entry.parent_id or
720
parent_entry.name != entry.name):
721
# Metadata common to all entries has changed
722
# against per-file parent
723
carry_over_possible = False
725
carry_over_possible = True
726
# per-type checks for changes against the parent_entry
729
# Cannot be a carry-over situation
730
carry_over_possible = False
731
# Populate the entry in the delta
733
# XXX: There is still a small race here: If someone reverts the content of a file
734
# after iter_changes examines and decides it has changed,
735
# we will unconditionally record a new version even if some
736
# other process reverts it while commit is running (with
737
# the revert happening after iter_changes did it's
740
entry.executable = True
742
entry.executable = False
743
if (carry_over_possible and
744
parent_entry.executable == entry.executable):
745
# Check the file length, content hash after reading
747
nostore_sha = parent_entry.text_sha1
750
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
752
text = file_obj.read()
756
entry.text_sha1, entry.text_size = self._add_text_to_weave(
757
file_id, text, heads, nostore_sha)
758
yield file_id, change[1][1], (entry.text_sha1, stat_value)
759
except errors.ExistingContent:
760
# No content change against a carry_over parent
761
# Perhaps this should also yield a fs hash update?
763
entry.text_size = parent_entry.text_size
764
entry.text_sha1 = parent_entry.text_sha1
765
elif kind == 'symlink':
767
entry.symlink_target = tree.get_symlink_target(file_id)
768
if (carry_over_possible and
769
parent_entry.symlink_target == entry.symlink_target):
772
self._add_text_to_weave(change[0], '', heads, None)
773
elif kind == 'directory':
774
if carry_over_possible:
777
# Nothing to set on the entry.
778
# XXX: split into the Root and nonRoot versions.
779
if change[1][1] != '' or self.repository.supports_rich_root():
780
self._add_text_to_weave(change[0], '', heads, None)
781
elif kind == 'tree-reference':
782
if not self.repository._format.supports_tree_reference:
783
# This isn't quite sane as an error, but we shouldn't
784
# ever see this code path in practice: tree's don't
785
# permit references when the repo doesn't support tree
787
raise errors.UnsupportedOperation(tree.add_reference,
789
reference_revision = tree.get_reference_revision(change[0])
790
entry.reference_revision = reference_revision
791
if (carry_over_possible and
792
parent_entry.reference_revision == reference_revision):
795
self._add_text_to_weave(change[0], '', heads, None)
797
raise AssertionError('unknown kind %r' % kind)
799
entry.revision = modified_rev
801
entry.revision = parent_entry.revision
804
new_path = change[1][1]
805
inv_delta.append((change[1][0], new_path, change[0], entry))
808
self.new_inventory = None
810
self._any_changes = True
812
# housekeeping root entry changes do not affect no-change commits.
813
self._require_root_change(tree)
814
self.basis_delta_revision = basis_revision_id
816
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
817
parent_keys = tuple([(file_id, parent) for parent in parents])
818
return self.repository.texts._add_text(
819
(file_id, self._new_revision_id), parent_keys, new_text,
820
nostore_sha=nostore_sha, random_id=self.random_revid)[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
######################################################################
853
class Repository(object):
854
"""Repository holding history for one or more branches.
856
The repository holds and retrieves historical information including
857
revisions and file history. It's normally accessed only by the Branch,
858
which views a particular line of development through that history.
860
The Repository builds on top of some byte storage facilies (the revisions,
861
signatures, inventories, texts and chk_bytes attributes) and a Transport,
862
which respectively provide byte storage and a means to access the (possibly
865
The byte storage facilities are addressed via tuples, which we refer to
866
as 'keys' throughout the code base. Revision_keys, inventory_keys and
867
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
868
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
869
byte string made up of a hash identifier and a hash value.
870
We use this interface because it allows low friction with the underlying
871
code that implements disk indices, network encoding and other parts of
874
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
875
the serialised revisions for the repository. This can be used to obtain
876
revision graph information or to access raw serialised revisions.
877
The result of trying to insert data into the repository via this store
878
is undefined: it should be considered read-only except for implementors
880
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
881
the serialised signatures for the repository. This can be used to
882
obtain access to raw serialised signatures. The result of trying to
883
insert data into the repository via this store is undefined: it should
884
be considered read-only except for implementors of repositories.
885
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
886
the serialised inventories for the repository. This can be used to
887
obtain unserialised inventories. The result of trying to insert data
888
into the repository via this store is undefined: it should be
889
considered read-only except for implementors of repositories.
890
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
891
texts of files and directories for the repository. This can be used to
892
obtain file texts or file graphs. Note that Repository.iter_file_bytes
893
is usually a better interface for accessing file texts.
894
The result of trying to insert data into the repository via this store
895
is undefined: it should be considered read-only except for implementors
897
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
898
any data the repository chooses to store or have indexed by its hash.
899
The result of trying to insert data into the repository via this store
900
is undefined: it should be considered read-only except for implementors
902
:ivar _transport: Transport for file access to repository, typically
903
pointing to .bzr/repository.
906
# What class to use for a CommitBuilder. Often its simpler to change this
907
# in a Repository class subclass rather than to override
908
# get_commit_builder.
909
_commit_builder_class = CommitBuilder
910
# The search regex used by xml based repositories to determine what things
911
# where changed in a single commit.
912
_file_ids_altered_regex = lazy_regex.lazy_compile(
913
r'file_id="(?P<file_id>[^"]+)"'
914
r'.* revision="(?P<revision_id>[^"]+)"'
917
def abort_write_group(self, suppress_errors=False):
918
"""Commit the contents accrued within the current write group.
920
:param suppress_errors: if true, abort_write_group will catch and log
921
unexpected errors that happen during the abort, rather than
922
allowing them to propagate. Defaults to False.
924
:seealso: start_write_group.
926
if self._write_group is not self.get_transaction():
927
# has an unlock or relock occured ?
930
'(suppressed) mismatched lock context and write group. %r, %r',
931
self._write_group, self.get_transaction())
933
raise errors.BzrError(
934
'mismatched lock context and write group. %r, %r' %
935
(self._write_group, self.get_transaction()))
937
self._abort_write_group()
938
except Exception, exc:
939
self._write_group = None
940
if not suppress_errors:
942
mutter('abort_write_group failed')
943
log_exception_quietly()
944
note('bzr: ERROR (ignored): %s', exc)
945
self._write_group = None
947
def _abort_write_group(self):
948
"""Template method for per-repository write group cleanup.
950
This is called during abort before the write group is considered to be
951
finished and should cleanup any internal state accrued during the write
952
group. There is no requirement that data handed to the repository be
953
*not* made available - this is not a rollback - but neither should any
954
attempt be made to ensure that data added is fully commited. Abort is
955
invoked when an error has occured so futher disk or network operations
956
may not be possible or may error and if possible should not be
960
def add_fallback_repository(self, repository):
961
"""Add a repository to use for looking up data not held locally.
963
:param repository: A repository.
965
if not self._format.supports_external_lookups:
966
raise errors.UnstackableRepositoryFormat(self._format, self.base)
968
# This repository will call fallback.unlock() when we transition to
969
# the unlocked state, so we make sure to increment the lock count
970
repository.lock_read()
971
self._check_fallback_repository(repository)
972
self._fallback_repositories.append(repository)
973
self.texts.add_fallback_versioned_files(repository.texts)
974
self.inventories.add_fallback_versioned_files(repository.inventories)
975
self.revisions.add_fallback_versioned_files(repository.revisions)
976
self.signatures.add_fallback_versioned_files(repository.signatures)
977
if self.chk_bytes is not None:
978
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
980
def _check_fallback_repository(self, repository):
981
"""Check that this repository can fallback to repository safely.
983
Raise an error if not.
985
:param repository: A repository to fallback to.
987
return InterRepository._assert_same_model(self, repository)
989
def add_inventory(self, revision_id, inv, parents):
990
"""Add the inventory inv to the repository as revision_id.
992
:param parents: The revision ids of the parents that revision_id
993
is known to have and are in the repository already.
995
:returns: The validator(which is a sha1 digest, though what is sha'd is
996
repository format specific) of the serialized inventory.
998
if not self.is_in_write_group():
999
raise AssertionError("%r not in write group" % (self,))
1000
_mod_revision.check_not_reserved_id(revision_id)
1001
if not (inv.revision_id is None or inv.revision_id == revision_id):
1002
raise AssertionError(
1003
"Mismatch between inventory revision"
1004
" id and insertion revid (%r, %r)"
1005
% (inv.revision_id, revision_id))
1006
if inv.root is None:
1007
raise AssertionError()
1008
return self._add_inventory_checked(revision_id, inv, parents)
1010
def _add_inventory_checked(self, revision_id, inv, parents):
1011
"""Add inv to the repository after checking the inputs.
1013
This function can be overridden to allow different inventory styles.
1015
:seealso: add_inventory, for the contract.
1017
inv_lines = self._serialise_inventory_to_lines(inv)
1018
return self._inventory_add_lines(revision_id, parents,
1019
inv_lines, check_content=False)
1021
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1022
parents, basis_inv=None, propagate_caches=False):
1023
"""Add a new inventory expressed as a delta against another revision.
1025
See the inventory developers documentation for the theory behind
1028
:param basis_revision_id: The inventory id the delta was created
1029
against. (This does not have to be a direct parent.)
1030
:param delta: The inventory delta (see Inventory.apply_delta for
1032
:param new_revision_id: The revision id that the inventory is being
1034
:param parents: The revision ids of the parents that revision_id is
1035
known to have and are in the repository already. These are supplied
1036
for repositories that depend on the inventory graph for revision
1037
graph access, as well as for those that pun ancestry with delta
1039
:param basis_inv: The basis inventory if it is already known,
1041
:param propagate_caches: If True, the caches for this inventory are
1042
copied to and updated for the result if possible.
1044
:returns: (validator, new_inv)
1045
The validator(which is a sha1 digest, though what is sha'd is
1046
repository format specific) of the serialized inventory, and the
1047
resulting inventory.
1049
if not self.is_in_write_group():
1050
raise AssertionError("%r not in write group" % (self,))
1051
_mod_revision.check_not_reserved_id(new_revision_id)
1052
basis_tree = self.revision_tree(basis_revision_id)
1053
basis_tree.lock_read()
1055
# Note that this mutates the inventory of basis_tree, which not all
1056
# inventory implementations may support: A better idiom would be to
1057
# return a new inventory, but as there is no revision tree cache in
1058
# repository this is safe for now - RBC 20081013
1059
if basis_inv is None:
1060
basis_inv = basis_tree.inventory
1061
basis_inv.apply_delta(delta)
1062
basis_inv.revision_id = new_revision_id
1063
return (self.add_inventory(new_revision_id, basis_inv, parents),
1068
def _inventory_add_lines(self, revision_id, parents, lines,
1069
check_content=True):
1070
"""Store lines in inv_vf and return the sha1 of the inventory."""
1071
parents = [(parent,) for parent in parents]
1072
result = self.inventories.add_lines((revision_id,), parents, lines,
1073
check_content=check_content)[0]
1074
self.inventories._access.flush()
1077
def add_revision(self, revision_id, rev, inv=None, config=None):
1078
"""Add rev to the revision store as revision_id.
1080
:param revision_id: the revision id to use.
1081
:param rev: The revision object.
1082
:param inv: The inventory for the revision. if None, it will be looked
1083
up in the inventory storer
1084
:param config: If None no digital signature will be created.
1085
If supplied its signature_needed method will be used
1086
to determine if a signature should be made.
1088
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1090
_mod_revision.check_not_reserved_id(revision_id)
1091
if config is not None and config.signature_needed():
1093
inv = self.get_inventory(revision_id)
1094
plaintext = Testament(rev, inv).as_short_text()
1095
self.store_revision_signature(
1096
gpg.GPGStrategy(config), plaintext, revision_id)
1097
# check inventory present
1098
if not self.inventories.get_parent_map([(revision_id,)]):
1100
raise errors.WeaveRevisionNotPresent(revision_id,
1103
# yes, this is not suitable for adding with ghosts.
1104
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1107
key = (revision_id,)
1108
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1109
self._add_revision(rev)
1111
def _add_revision(self, revision):
1112
text = self._serializer.write_revision_to_string(revision)
1113
key = (revision.revision_id,)
1114
parents = tuple((parent,) for parent in revision.parent_ids)
1115
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1117
def all_revision_ids(self):
1118
"""Returns a list of all the revision ids in the repository.
1120
This is conceptually deprecated because code should generally work on
1121
the graph reachable from a particular revision, and ignore any other
1122
revisions that might be present. There is no direct replacement
1125
if 'evil' in debug.debug_flags:
1126
mutter_callsite(2, "all_revision_ids is linear with history.")
1127
return self._all_revision_ids()
1129
def _all_revision_ids(self):
1130
"""Returns a list of all the revision ids in the repository.
1132
These are in as much topological order as the underlying store can
1135
raise NotImplementedError(self._all_revision_ids)
1137
def break_lock(self):
1138
"""Break a lock if one is present from another instance.
1140
Uses the ui factory to ask for confirmation if the lock may be from
1143
self.control_files.break_lock()
1146
def _eliminate_revisions_not_present(self, revision_ids):
1147
"""Check every revision id in revision_ids to see if we have it.
1149
Returns a set of the present revisions.
1152
graph = self.get_graph()
1153
parent_map = graph.get_parent_map(revision_ids)
1154
# The old API returned a list, should this actually be a set?
1155
return parent_map.keys()
1157
def _check_inventories(self, checker):
1158
"""Check the inventories found from the revision scan.
1160
This is responsible for verifying the sha1 of inventories and
1161
creating a pending_keys set that covers data referenced by inventories.
1163
bar = ui.ui_factory.nested_progress_bar()
1165
self._do_check_inventories(checker, bar)
1169
def _do_check_inventories(self, checker, bar):
1170
"""Helper for _check_inventories."""
1172
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1173
kinds = ['chk_bytes', 'texts']
1174
count = len(checker.pending_keys)
1175
bar.update("inventories", 0, 2)
1176
current_keys = checker.pending_keys
1177
checker.pending_keys = {}
1178
# Accumulate current checks.
1179
for key in current_keys:
1180
if key[0] != 'inventories' and key[0] not in kinds:
1181
checker._report_items.append('unknown key type %r' % (key,))
1182
keys[key[0]].add(key[1:])
1183
if keys['inventories']:
1184
# NB: output order *should* be roughly sorted - topo or
1185
# inverse topo depending on repository - either way decent
1186
# to just delta against. However, pre-CHK formats didn't
1187
# try to optimise inventory layout on disk. As such the
1188
# pre-CHK code path does not use inventory deltas.
1190
for record in self.inventories.check(keys=keys['inventories']):
1191
if record.storage_kind == 'absent':
1192
checker._report_items.append(
1193
'Missing inventory {%s}' % (record.key,))
1195
last_object = self._check_record('inventories', record,
1196
checker, last_object,
1197
current_keys[('inventories',) + record.key])
1198
del keys['inventories']
1201
bar.update("texts", 1)
1202
while (checker.pending_keys or keys['chk_bytes']
1204
# Something to check.
1205
current_keys = checker.pending_keys
1206
checker.pending_keys = {}
1207
# Accumulate current checks.
1208
for key in current_keys:
1209
if key[0] not in kinds:
1210
checker._report_items.append('unknown key type %r' % (key,))
1211
keys[key[0]].add(key[1:])
1212
# Check the outermost kind only - inventories || chk_bytes || texts
1216
for record in getattr(self, kind).check(keys=keys[kind]):
1217
if record.storage_kind == 'absent':
1218
checker._report_items.append(
1219
'Missing inventory {%s}' % (record.key,))
1221
last_object = self._check_record(kind, record,
1222
checker, last_object, current_keys[(kind,) + record.key])
1226
def _check_record(self, kind, record, checker, last_object, item_data):
1227
"""Check a single text from this repository."""
1228
if kind == 'inventories':
1229
rev_id = record.key[0]
1230
inv = self.deserialise_inventory(rev_id,
1231
record.get_bytes_as('fulltext'))
1232
if last_object is not None:
1233
delta = inv._make_delta(last_object)
1234
for old_path, path, file_id, ie in delta:
1237
ie.check(checker, rev_id, inv)
1239
for path, ie in inv.iter_entries():
1240
ie.check(checker, rev_id, inv)
1241
if self._format.fast_deltas:
1243
elif kind == 'chk_bytes':
1244
# No code written to check chk_bytes for this repo format.
1245
checker._report_items.append(
1246
'unsupported key type chk_bytes for %s' % (record.key,))
1247
elif kind == 'texts':
1248
self._check_text(record, checker, item_data)
1250
checker._report_items.append(
1251
'unknown key type %s for %s' % (kind, record.key))
1253
def _check_text(self, record, checker, item_data):
1254
"""Check a single text."""
1255
# Check it is extractable.
1256
# TODO: check length.
1257
if record.storage_kind == 'chunked':
1258
chunks = record.get_bytes_as(record.storage_kind)
1259
sha1 = osutils.sha_strings(chunks)
1260
length = sum(map(len, chunks))
1262
content = record.get_bytes_as('fulltext')
1263
sha1 = osutils.sha_string(content)
1264
length = len(content)
1265
if item_data and sha1 != item_data[1]:
1266
checker._report_items.append(
1267
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1268
(record.key, sha1, item_data[1], item_data[2]))
1271
def create(a_bzrdir):
1272
"""Construct the current default format repository in a_bzrdir."""
1273
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1275
def __init__(self, _format, a_bzrdir, control_files):
1276
"""instantiate a Repository.
1278
:param _format: The format of the repository on disk.
1279
:param a_bzrdir: The BzrDir of the repository.
1281
In the future we will have a single api for all stores for
1282
getting file texts, inventories and revisions, then
1283
this construct will accept instances of those things.
1285
super(Repository, self).__init__()
1286
self._format = _format
1287
# the following are part of the public API for Repository:
1288
self.bzrdir = a_bzrdir
1289
self.control_files = control_files
1290
self._transport = control_files._transport
1291
self.base = self._transport.base
1293
self._reconcile_does_inventory_gc = True
1294
self._reconcile_fixes_text_parents = False
1295
self._reconcile_backsup_inventory = True
1296
# not right yet - should be more semantically clear ?
1298
# TODO: make sure to construct the right store classes, etc, depending
1299
# on whether escaping is required.
1300
self._warn_if_deprecated()
1301
self._write_group = None
1302
# Additional places to query for data.
1303
self._fallback_repositories = []
1304
# An InventoryEntry cache, used during deserialization
1305
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1308
if self._fallback_repositories:
1309
return '%s(%r, fallback_repositories=%r)' % (
1310
self.__class__.__name__,
1312
self._fallback_repositories)
1314
return '%s(%r)' % (self.__class__.__name__,
1317
def _has_same_fallbacks(self, other_repo):
1318
"""Returns true if the repositories have the same fallbacks."""
1319
my_fb = self._fallback_repositories
1320
other_fb = other_repo._fallback_repositories
1321
if len(my_fb) != len(other_fb):
1323
for f, g in zip(my_fb, other_fb):
1324
if not f.has_same_location(g):
1328
def has_same_location(self, other):
1329
"""Returns a boolean indicating if this repository is at the same
1330
location as another repository.
1332
This might return False even when two repository objects are accessing
1333
the same physical repository via different URLs.
1335
if self.__class__ is not other.__class__:
1337
return (self._transport.base == other._transport.base)
1339
def is_in_write_group(self):
1340
"""Return True if there is an open write group.
1342
:seealso: start_write_group.
1344
return self._write_group is not None
1346
def is_locked(self):
1347
return self.control_files.is_locked()
1349
def is_write_locked(self):
1350
"""Return True if this object is write locked."""
1351
return self.is_locked() and self.control_files._lock_mode == 'w'
1353
def lock_write(self, token=None):
1354
"""Lock this repository for writing.
1356
This causes caching within the repository obejct to start accumlating
1357
data during reads, and allows a 'write_group' to be obtained. Write
1358
groups must be used for actual data insertion.
1360
:param token: if this is already locked, then lock_write will fail
1361
unless the token matches the existing lock.
1362
:returns: a token if this instance supports tokens, otherwise None.
1363
:raises TokenLockingNotSupported: when a token is given but this
1364
instance doesn't support using token locks.
1365
:raises MismatchedToken: if the specified token doesn't match the token
1366
of the existing lock.
1367
:seealso: start_write_group.
1369
A token should be passed in if you know that you have locked the object
1370
some other way, and need to synchronise this object's state with that
1373
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1375
locked = self.is_locked()
1376
result = self.control_files.lock_write(token=token)
1378
for repo in self._fallback_repositories:
1379
# Writes don't affect fallback repos
1381
self._refresh_data()
1384
def lock_read(self):
1385
locked = self.is_locked()
1386
self.control_files.lock_read()
1388
for repo in self._fallback_repositories:
1390
self._refresh_data()
1392
def get_physical_lock_status(self):
1393
return self.control_files.get_physical_lock_status()
1395
def leave_lock_in_place(self):
1396
"""Tell this repository not to release the physical lock when this
1399
If lock_write doesn't return a token, then this method is not supported.
1401
self.control_files.leave_in_place()
1403
def dont_leave_lock_in_place(self):
1404
"""Tell this repository to release the physical lock when this
1405
object is unlocked, even if it didn't originally acquire it.
1407
If lock_write doesn't return a token, then this method is not supported.
1409
self.control_files.dont_leave_in_place()
1412
def gather_stats(self, revid=None, committers=None):
1413
"""Gather statistics from a revision id.
1415
:param revid: The revision id to gather statistics from, if None, then
1416
no revision specific statistics are gathered.
1417
:param committers: Optional parameter controlling whether to grab
1418
a count of committers from the revision specific statistics.
1419
:return: A dictionary of statistics. Currently this contains:
1420
committers: The number of committers if requested.
1421
firstrev: A tuple with timestamp, timezone for the penultimate left
1422
most ancestor of revid, if revid is not the NULL_REVISION.
1423
latestrev: A tuple with timestamp, timezone for revid, if revid is
1424
not the NULL_REVISION.
1425
revisions: The total revision count in the repository.
1426
size: An estimate disk size of the repository in bytes.
1429
if revid and committers:
1430
result['committers'] = 0
1431
if revid and revid != _mod_revision.NULL_REVISION:
1433
all_committers = set()
1434
revisions = self.get_ancestry(revid)
1435
# pop the leading None
1437
first_revision = None
1439
# ignore the revisions in the middle - just grab first and last
1440
revisions = revisions[0], revisions[-1]
1441
for revision in self.get_revisions(revisions):
1442
if not first_revision:
1443
first_revision = revision
1445
all_committers.add(revision.committer)
1446
last_revision = revision
1448
result['committers'] = len(all_committers)
1449
result['firstrev'] = (first_revision.timestamp,
1450
first_revision.timezone)
1451
result['latestrev'] = (last_revision.timestamp,
1452
last_revision.timezone)
1454
# now gather global repository information
1455
# XXX: This is available for many repos regardless of listability.
1456
if self.bzrdir.root_transport.listable():
1457
# XXX: do we want to __define len__() ?
1458
# Maybe the versionedfiles object should provide a different
1459
# method to get the number of keys.
1460
result['revisions'] = len(self.revisions.keys())
1461
# result['size'] = t
1464
def find_branches(self, using=False):
1465
"""Find branches underneath this repository.
1467
This will include branches inside other branches.
1469
:param using: If True, list only branches using this repository.
1471
if using and not self.is_shared():
1473
return [self.bzrdir.open_branch()]
1474
except errors.NotBranchError:
1476
class Evaluator(object):
1479
self.first_call = True
1481
def __call__(self, bzrdir):
1482
# On the first call, the parameter is always the bzrdir
1483
# containing the current repo.
1484
if not self.first_call:
1486
repository = bzrdir.open_repository()
1487
except errors.NoRepositoryPresent:
1490
return False, (None, repository)
1491
self.first_call = False
1493
value = (bzrdir.open_branch(), None)
1494
except errors.NotBranchError:
1495
value = (None, None)
1499
for branch, repository in bzrdir.BzrDir.find_bzrdirs(
1500
self.bzrdir.root_transport, evaluate=Evaluator()):
1501
if branch is not None:
1502
branches.append(branch)
1503
if not using and repository is not None:
1504
branches.extend(repository.find_branches())
1508
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
1509
"""Return the revision ids that other has that this does not.
1511
These are returned in topological order.
1513
revision_id: only return revision ids included by revision_id.
1515
return InterRepository.get(other, self).search_missing_revision_ids(
1516
revision_id, find_ghosts)
1520
"""Open the repository rooted at base.
1522
For instance, if the repository is at URL/.bzr/repository,
1523
Repository.open(URL) -> a Repository instance.
1525
control = bzrdir.BzrDir.open(base)
1526
return control.open_repository()
1528
def copy_content_into(self, destination, revision_id=None):
1529
"""Make a complete copy of the content in self into destination.
1531
This is a destructive operation! Do not use it on existing
1534
return InterRepository.get(self, destination).copy_content(revision_id)
1536
def commit_write_group(self):
1537
"""Commit the contents accrued within the current write group.
1539
:seealso: start_write_group.
1541
:return: it may return an opaque hint that can be passed to 'pack'.
1543
if self._write_group is not self.get_transaction():
1544
# has an unlock or relock occured ?
1545
raise errors.BzrError('mismatched lock context %r and '
1547
(self.get_transaction(), self._write_group))
1548
result = self._commit_write_group()
1549
self._write_group = None
1552
def _commit_write_group(self):
1553
"""Template method for per-repository write group cleanup.
1555
This is called before the write group is considered to be
1556
finished and should ensure that all data handed to the repository
1557
for writing during the write group is safely committed (to the
1558
extent possible considering file system caching etc).
1561
def suspend_write_group(self):
1562
raise errors.UnsuspendableWriteGroup(self)
1564
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1565
"""Return the keys of missing inventory parents for revisions added in
1568
A revision is not complete if the inventory delta for that revision
1569
cannot be calculated. Therefore if the parent inventories of a
1570
revision are not present, the revision is incomplete, and e.g. cannot
1571
be streamed by a smart server. This method finds missing inventory
1572
parents for revisions added in this write group.
1574
if not self._format.supports_external_lookups:
1575
# This is only an issue for stacked repositories
1577
if not self.is_in_write_group():
1578
raise AssertionError('not in a write group')
1580
# XXX: We assume that every added revision already has its
1581
# corresponding inventory, so we only check for parent inventories that
1582
# might be missing, rather than all inventories.
1583
parents = set(self.revisions._index.get_missing_parents())
1584
parents.discard(_mod_revision.NULL_REVISION)
1585
unstacked_inventories = self.inventories._index
1586
present_inventories = unstacked_inventories.get_parent_map(
1587
key[-1:] for key in parents)
1588
parents.difference_update(present_inventories)
1589
if len(parents) == 0:
1590
# No missing parent inventories.
1592
if not check_for_missing_texts:
1593
return set(('inventories', rev_id) for (rev_id,) in parents)
1594
# Ok, now we have a list of missing inventories. But these only matter
1595
# if the inventories that reference them are missing some texts they
1596
# appear to introduce.
1597
# XXX: Texts referenced by all added inventories need to be present,
1598
# but at the moment we're only checking for texts referenced by
1599
# inventories at the graph's edge.
1600
key_deps = self.revisions._index._key_dependencies
1601
key_deps.add_keys(present_inventories)
1602
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1603
file_ids = self.fileids_altered_by_revision_ids(referrers)
1604
missing_texts = set()
1605
for file_id, version_ids in file_ids.iteritems():
1606
missing_texts.update(
1607
(file_id, version_id) for version_id in version_ids)
1608
present_texts = self.texts.get_parent_map(missing_texts)
1609
missing_texts.difference_update(present_texts)
1610
if not missing_texts:
1611
# No texts are missing, so all revisions and their deltas are
1614
# Alternatively the text versions could be returned as the missing
1615
# keys, but this is likely to be less data.
1616
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1619
def refresh_data(self):
1620
"""Re-read any data needed to to synchronise with disk.
1622
This method is intended to be called after another repository instance
1623
(such as one used by a smart server) has inserted data into the
1624
repository. It may not be called during a write group, but may be
1625
called at any other time.
1627
if self.is_in_write_group():
1628
raise errors.InternalBzrError(
1629
"May not refresh_data while in a write group.")
1630
self._refresh_data()
1632
def resume_write_group(self, tokens):
1633
if not self.is_write_locked():
1634
raise errors.NotWriteLocked(self)
1635
if self._write_group:
1636
raise errors.BzrError('already in a write group')
1637
self._resume_write_group(tokens)
1638
# so we can detect unlock/relock - the write group is now entered.
1639
self._write_group = self.get_transaction()
1641
def _resume_write_group(self, tokens):
1642
raise errors.UnsuspendableWriteGroup(self)
1644
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1646
"""Fetch the content required to construct revision_id from source.
1648
If revision_id is None and fetch_spec is None, then all content is
1651
fetch() may not be used when the repository is in a write group -
1652
either finish the current write group before using fetch, or use
1653
fetch before starting the write group.
1655
:param find_ghosts: Find and copy revisions in the source that are
1656
ghosts in the target (and not reachable directly by walking out to
1657
the first-present revision in target from revision_id).
1658
:param revision_id: If specified, all the content needed for this
1659
revision ID will be copied to the target. Fetch will determine for
1660
itself which content needs to be copied.
1661
:param fetch_spec: If specified, a SearchResult or
1662
PendingAncestryResult that describes which revisions to copy. This
1663
allows copying multiple heads at once. Mutually exclusive with
1666
if fetch_spec is not None and revision_id is not None:
1667
raise AssertionError(
1668
"fetch_spec and revision_id are mutually exclusive.")
1669
if self.is_in_write_group():
1670
raise errors.InternalBzrError(
1671
"May not fetch while in a write group.")
1672
# fast path same-url fetch operations
1673
# TODO: lift out to somewhere common with RemoteRepository
1674
# <https://bugs.edge.launchpad.net/bzr/+bug/401646>
1675
if (self.has_same_location(source)
1676
and fetch_spec is None
1677
and self._has_same_fallbacks(source)):
1678
# check that last_revision is in 'from' and then return a
1680
if (revision_id is not None and
1681
not _mod_revision.is_null(revision_id)):
1682
self.get_revision(revision_id)
1684
# if there is no specific appropriate InterRepository, this will get
1685
# the InterRepository base class, which raises an
1686
# IncompatibleRepositories when asked to fetch.
1687
inter = InterRepository.get(source, self)
1688
return inter.fetch(revision_id=revision_id, pb=pb,
1689
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1691
def create_bundle(self, target, base, fileobj, format=None):
1692
return serializer.write_bundle(self, target, base, fileobj, format)
1694
def get_commit_builder(self, branch, parents, config, timestamp=None,
1695
timezone=None, committer=None, revprops=None,
1697
"""Obtain a CommitBuilder for this repository.
1699
:param branch: Branch to commit to.
1700
:param parents: Revision ids of the parents of the new revision.
1701
:param config: Configuration to use.
1702
:param timestamp: Optional timestamp recorded for commit.
1703
:param timezone: Optional timezone for timestamp.
1704
:param committer: Optional committer to set for commit.
1705
:param revprops: Optional dictionary of revision properties.
1706
:param revision_id: Optional revision id.
1708
result = self._commit_builder_class(self, parents, config,
1709
timestamp, timezone, committer, revprops, revision_id)
1710
self.start_write_group()
1714
if (self.control_files._lock_count == 1 and
1715
self.control_files._lock_mode == 'w'):
1716
if self._write_group is not None:
1717
self.abort_write_group()
1718
self.control_files.unlock()
1719
raise errors.BzrError(
1720
'Must end write groups before releasing write locks.')
1721
self.control_files.unlock()
1722
if self.control_files._lock_count == 0:
1723
self._inventory_entry_cache.clear()
1724
for repo in self._fallback_repositories:
1728
def clone(self, a_bzrdir, revision_id=None):
1729
"""Clone this repository into a_bzrdir using the current format.
1731
Currently no check is made that the format of this repository and
1732
the bzrdir format are compatible. FIXME RBC 20060201.
1734
:return: The newly created destination repository.
1736
# TODO: deprecate after 0.16; cloning this with all its settings is
1737
# probably not very useful -- mbp 20070423
1738
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1739
self.copy_content_into(dest_repo, revision_id)
1742
def start_write_group(self):
1743
"""Start a write group in the repository.
1745
Write groups are used by repositories which do not have a 1:1 mapping
1746
between file ids and backend store to manage the insertion of data from
1747
both fetch and commit operations.
1749
A write lock is required around the start_write_group/commit_write_group
1750
for the support of lock-requiring repository formats.
1752
One can only insert data into a repository inside a write group.
1756
if not self.is_write_locked():
1757
raise errors.NotWriteLocked(self)
1758
if self._write_group:
1759
raise errors.BzrError('already in a write group')
1760
self._start_write_group()
1761
# so we can detect unlock/relock - the write group is now entered.
1762
self._write_group = self.get_transaction()
1764
def _start_write_group(self):
1765
"""Template method for per-repository write group startup.
1767
This is called before the write group is considered to be
1772
def sprout(self, to_bzrdir, revision_id=None):
1773
"""Create a descendent repository for new development.
1775
Unlike clone, this does not copy the settings of the repository.
1777
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1778
dest_repo.fetch(self, revision_id=revision_id)
1781
def _create_sprouting_repo(self, a_bzrdir, shared):
1782
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1783
# use target default format.
1784
dest_repo = a_bzrdir.create_repository()
1786
# Most control formats need the repository to be specifically
1787
# created, but on some old all-in-one formats it's not needed
1789
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1790
except errors.UninitializableFormat:
1791
dest_repo = a_bzrdir.open_repository()
1794
def _get_sink(self):
1795
"""Return a sink for streaming into this repository."""
1796
return StreamSink(self)
1798
def _get_source(self, to_format):
1799
"""Return a source for streaming from this repository."""
1800
return StreamSource(self, to_format)
1803
def has_revision(self, revision_id):
1804
"""True if this repository has a copy of the revision."""
1805
return revision_id in self.has_revisions((revision_id,))
1808
def has_revisions(self, revision_ids):
1809
"""Probe to find out the presence of multiple revisions.
1811
:param revision_ids: An iterable of revision_ids.
1812
:return: A set of the revision_ids that were present.
1814
parent_map = self.revisions.get_parent_map(
1815
[(rev_id,) for rev_id in revision_ids])
1817
if _mod_revision.NULL_REVISION in revision_ids:
1818
result.add(_mod_revision.NULL_REVISION)
1819
result.update([key[0] for key in parent_map])
1823
def get_revision(self, revision_id):
1824
"""Return the Revision object for a named revision."""
1825
return self.get_revisions([revision_id])[0]
1828
def get_revision_reconcile(self, revision_id):
1829
"""'reconcile' helper routine that allows access to a revision always.
1831
This variant of get_revision does not cross check the weave graph
1832
against the revision one as get_revision does: but it should only
1833
be used by reconcile, or reconcile-alike commands that are correcting
1834
or testing the revision graph.
1836
return self._get_revisions([revision_id])[0]
1839
def get_revisions(self, revision_ids):
1840
"""Get many revisions at once.
1842
Repositories that need to check data on every revision read should
1843
subclass this method.
1845
return self._get_revisions(revision_ids)
1848
def _get_revisions(self, revision_ids):
1849
"""Core work logic to get many revisions without sanity checks."""
1851
for revid, rev in self._iter_revisions(revision_ids):
1853
raise errors.NoSuchRevision(self, revid)
1855
return [revs[revid] for revid in revision_ids]
1857
def _iter_revisions(self, revision_ids):
1858
"""Iterate over revision objects.
1860
:param revision_ids: An iterable of revisions to examine. None may be
1861
passed to request all revisions known to the repository. Note that
1862
not all repositories can find unreferenced revisions; for those
1863
repositories only referenced ones will be returned.
1864
:return: An iterator of (revid, revision) tuples. Absent revisions (
1865
those asked for but not available) are returned as (revid, None).
1867
if revision_ids is None:
1868
revision_ids = self.all_revision_ids()
1870
for rev_id in revision_ids:
1871
if not rev_id or not isinstance(rev_id, basestring):
1872
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1873
keys = [(key,) for key in revision_ids]
1874
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1875
for record in stream:
1876
revid = record.key[0]
1877
if record.storage_kind == 'absent':
1880
text = record.get_bytes_as('fulltext')
1881
rev = self._serializer.read_revision_from_string(text)
1885
def get_revision_xml(self, revision_id):
1886
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1887
# would have already do it.
1888
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1889
# TODO: this can't just be replaced by:
1890
# return self._serializer.write_revision_to_string(
1891
# self.get_revision(revision_id))
1892
# as cStringIO preservers the encoding unlike write_revision_to_string
1893
# or some other call down the path.
1894
rev = self.get_revision(revision_id)
1895
rev_tmp = cStringIO.StringIO()
1896
# the current serializer..
1897
self._serializer.write_revision(rev, rev_tmp)
1899
return rev_tmp.getvalue()
1901
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1902
"""Produce a generator of revision deltas.
1904
Note that the input is a sequence of REVISIONS, not revision_ids.
1905
Trees will be held in memory until the generator exits.
1906
Each delta is relative to the revision's lefthand predecessor.
1908
:param specific_fileids: if not None, the result is filtered
1909
so that only those file-ids, their parents and their
1910
children are included.
1912
# Get the revision-ids of interest
1913
required_trees = set()
1914
for revision in revisions:
1915
required_trees.add(revision.revision_id)
1916
required_trees.update(revision.parent_ids[:1])
1918
# Get the matching filtered trees. Note that it's more
1919
# efficient to pass filtered trees to changes_from() rather
1920
# than doing the filtering afterwards. changes_from() could
1921
# arguably do the filtering itself but it's path-based, not
1922
# file-id based, so filtering before or afterwards is
1924
if specific_fileids is None:
1925
trees = dict((t.get_revision_id(), t) for
1926
t in self.revision_trees(required_trees))
1928
trees = dict((t.get_revision_id(), t) for
1929
t in self._filtered_revision_trees(required_trees,
1932
# Calculate the deltas
1933
for revision in revisions:
1934
if not revision.parent_ids:
1935
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1937
old_tree = trees[revision.parent_ids[0]]
1938
yield trees[revision.revision_id].changes_from(old_tree)
1941
def get_revision_delta(self, revision_id, specific_fileids=None):
1942
"""Return the delta for one revision.
1944
The delta is relative to the left-hand predecessor of the
1947
:param specific_fileids: if not None, the result is filtered
1948
so that only those file-ids, their parents and their
1949
children are included.
1951
r = self.get_revision(revision_id)
1952
return list(self.get_deltas_for_revisions([r],
1953
specific_fileids=specific_fileids))[0]
1956
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1957
signature = gpg_strategy.sign(plaintext)
1958
self.add_signature_text(revision_id, signature)
1961
def add_signature_text(self, revision_id, signature):
1962
self.signatures.add_lines((revision_id,), (),
1963
osutils.split_lines(signature))
1965
def find_text_key_references(self):
1966
"""Find the text key references within the repository.
1968
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1969
to whether they were referred to by the inventory of the
1970
revision_id that they contain. The inventory texts from all present
1971
revision ids are assessed to generate this report.
1973
revision_keys = self.revisions.keys()
1974
w = self.inventories
1975
pb = ui.ui_factory.nested_progress_bar()
1977
return self._find_text_key_references_from_xml_inventory_lines(
1978
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1982
def _find_text_key_references_from_xml_inventory_lines(self,
1984
"""Core routine for extracting references to texts from inventories.
1986
This performs the translation of xml lines to revision ids.
1988
:param line_iterator: An iterator of lines, origin_version_id
1989
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1990
to whether they were referred to by the inventory of the
1991
revision_id that they contain. Note that if that revision_id was
1992
not part of the line_iterator's output then False will be given -
1993
even though it may actually refer to that key.
1995
if not self._serializer.support_altered_by_hack:
1996
raise AssertionError(
1997
"_find_text_key_references_from_xml_inventory_lines only "
1998
"supported for branches which store inventory as unnested xml"
1999
", not on %r" % self)
2002
# this code needs to read every new line in every inventory for the
2003
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2004
# not present in one of those inventories is unnecessary but not
2005
# harmful because we are filtering by the revision id marker in the
2006
# inventory lines : we only select file ids altered in one of those
2007
# revisions. We don't need to see all lines in the inventory because
2008
# only those added in an inventory in rev X can contain a revision=X
2010
unescape_revid_cache = {}
2011
unescape_fileid_cache = {}
2013
# jam 20061218 In a big fetch, this handles hundreds of thousands
2014
# of lines, so it has had a lot of inlining and optimizing done.
2015
# Sorry that it is a little bit messy.
2016
# Move several functions to be local variables, since this is a long
2018
search = self._file_ids_altered_regex.search
2019
unescape = _unescape_xml
2020
setdefault = result.setdefault
2021
for line, line_key in line_iterator:
2022
match = search(line)
2025
# One call to match.group() returning multiple items is quite a
2026
# bit faster than 2 calls to match.group() each returning 1
2027
file_id, revision_id = match.group('file_id', 'revision_id')
2029
# Inlining the cache lookups helps a lot when you make 170,000
2030
# lines and 350k ids, versus 8.4 unique ids.
2031
# Using a cache helps in 2 ways:
2032
# 1) Avoids unnecessary decoding calls
2033
# 2) Re-uses cached strings, which helps in future set and
2035
# (2) is enough that removing encoding entirely along with
2036
# the cache (so we are using plain strings) results in no
2037
# performance improvement.
2039
revision_id = unescape_revid_cache[revision_id]
2041
unescaped = unescape(revision_id)
2042
unescape_revid_cache[revision_id] = unescaped
2043
revision_id = unescaped
2045
# Note that unconditionally unescaping means that we deserialise
2046
# every fileid, which for general 'pull' is not great, but we don't
2047
# really want to have some many fulltexts that this matters anyway.
2050
file_id = unescape_fileid_cache[file_id]
2052
unescaped = unescape(file_id)
2053
unescape_fileid_cache[file_id] = unescaped
2056
key = (file_id, revision_id)
2057
setdefault(key, False)
2058
if revision_id == line_key[-1]:
2062
def _inventory_xml_lines_for_keys(self, keys):
2063
"""Get a line iterator of the sort needed for findind references.
2065
Not relevant for non-xml inventory repositories.
2067
Ghosts in revision_keys are ignored.
2069
:param revision_keys: The revision keys for the inventories to inspect.
2070
:return: An iterator over (inventory line, revid) for the fulltexts of
2071
all of the xml inventories specified by revision_keys.
2073
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2074
for record in stream:
2075
if record.storage_kind != 'absent':
2076
chunks = record.get_bytes_as('chunked')
2077
revid = record.key[-1]
2078
lines = osutils.chunks_to_lines(chunks)
2082
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2084
"""Helper routine for fileids_altered_by_revision_ids.
2086
This performs the translation of xml lines to revision ids.
2088
:param line_iterator: An iterator of lines, origin_version_id
2089
:param revision_keys: The revision ids to filter for. This should be a
2090
set or other type which supports efficient __contains__ lookups, as
2091
the revision key from each parsed line will be looked up in the
2092
revision_keys filter.
2093
:return: a dictionary mapping altered file-ids to an iterable of
2094
revision_ids. Each altered file-ids has the exact revision_ids that
2095
altered it listed explicitly.
2097
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2098
line_iterator).iterkeys())
2099
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2100
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2101
self._inventory_xml_lines_for_keys(parent_keys)))
2102
new_keys = seen - parent_seen
2104
setdefault = result.setdefault
2105
for key in new_keys:
2106
setdefault(key[0], set()).add(key[-1])
2109
def _find_parent_ids_of_revisions(self, revision_ids):
2110
"""Find all parent ids that are mentioned in the revision graph.
2112
:return: set of revisions that are parents of revision_ids which are
2113
not part of revision_ids themselves
2115
parent_map = self.get_parent_map(revision_ids)
2117
map(parent_ids.update, parent_map.itervalues())
2118
parent_ids.difference_update(revision_ids)
2119
parent_ids.discard(_mod_revision.NULL_REVISION)
2122
def _find_parent_keys_of_revisions(self, revision_keys):
2123
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2125
:param revision_keys: An iterable of revision_keys.
2126
:return: The parents of all revision_keys that are not already in
2129
parent_map = self.revisions.get_parent_map(revision_keys)
2131
map(parent_keys.update, parent_map.itervalues())
2132
parent_keys.difference_update(revision_keys)
2133
parent_keys.discard(_mod_revision.NULL_REVISION)
2136
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2137
"""Find the file ids and versions affected by revisions.
2139
:param revisions: an iterable containing revision ids.
2140
:param _inv_weave: The inventory weave from this repository or None.
2141
If None, the inventory weave will be opened automatically.
2142
:return: a dictionary mapping altered file-ids to an iterable of
2143
revision_ids. Each altered file-ids has the exact revision_ids that
2144
altered it listed explicitly.
2146
selected_keys = set((revid,) for revid in revision_ids)
2147
w = _inv_weave or self.inventories
2148
pb = ui.ui_factory.nested_progress_bar()
2150
return self._find_file_ids_from_xml_inventory_lines(
2151
w.iter_lines_added_or_present_in_keys(
2152
selected_keys, pb=pb),
2157
def iter_files_bytes(self, desired_files):
2158
"""Iterate through file versions.
2160
Files will not necessarily be returned in the order they occur in
2161
desired_files. No specific order is guaranteed.
2163
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2164
value supplied by the caller as part of desired_files. It should
2165
uniquely identify the file version in the caller's context. (Examples:
2166
an index number or a TreeTransform trans_id.)
2168
bytes_iterator is an iterable of bytestrings for the file. The
2169
kind of iterable and length of the bytestrings are unspecified, but for
2170
this implementation, it is a list of bytes produced by
2171
VersionedFile.get_record_stream().
2173
:param desired_files: a list of (file_id, revision_id, identifier)
2177
for file_id, revision_id, callable_data in desired_files:
2178
text_keys[(file_id, revision_id)] = callable_data
2179
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2180
if record.storage_kind == 'absent':
2181
raise errors.RevisionNotPresent(record.key, self)
2182
yield text_keys[record.key], record.get_bytes_as('chunked')
2184
def _generate_text_key_index(self, text_key_references=None,
2186
"""Generate a new text key index for the repository.
2188
This is an expensive function that will take considerable time to run.
2190
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2191
list of parents, also text keys. When a given key has no parents,
2192
the parents list will be [NULL_REVISION].
2194
# All revisions, to find inventory parents.
2195
if ancestors is None:
2196
graph = self.get_graph()
2197
ancestors = graph.get_parent_map(self.all_revision_ids())
2198
if text_key_references is None:
2199
text_key_references = self.find_text_key_references()
2200
pb = ui.ui_factory.nested_progress_bar()
2202
return self._do_generate_text_key_index(ancestors,
2203
text_key_references, pb)
2207
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2208
"""Helper for _generate_text_key_index to avoid deep nesting."""
2209
revision_order = tsort.topo_sort(ancestors)
2210
invalid_keys = set()
2212
for revision_id in revision_order:
2213
revision_keys[revision_id] = set()
2214
text_count = len(text_key_references)
2215
# a cache of the text keys to allow reuse; costs a dict of all the
2216
# keys, but saves a 2-tuple for every child of a given key.
2218
for text_key, valid in text_key_references.iteritems():
2220
invalid_keys.add(text_key)
2222
revision_keys[text_key[1]].add(text_key)
2223
text_key_cache[text_key] = text_key
2224
del text_key_references
2226
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2227
NULL_REVISION = _mod_revision.NULL_REVISION
2228
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2229
# too small for large or very branchy trees. However, for 55K path
2230
# trees, it would be easy to use too much memory trivially. Ideally we
2231
# could gauge this by looking at available real memory etc, but this is
2232
# always a tricky proposition.
2233
inventory_cache = lru_cache.LRUCache(10)
2234
batch_size = 10 # should be ~150MB on a 55K path tree
2235
batch_count = len(revision_order) / batch_size + 1
2237
pb.update("Calculating text parents", processed_texts, text_count)
2238
for offset in xrange(batch_count):
2239
to_query = revision_order[offset * batch_size:(offset + 1) *
2243
for revision_id in to_query:
2244
parent_ids = ancestors[revision_id]
2245
for text_key in revision_keys[revision_id]:
2246
pb.update("Calculating text parents", processed_texts)
2247
processed_texts += 1
2248
candidate_parents = []
2249
for parent_id in parent_ids:
2250
parent_text_key = (text_key[0], parent_id)
2252
check_parent = parent_text_key not in \
2253
revision_keys[parent_id]
2255
# the parent parent_id is a ghost:
2256
check_parent = False
2257
# truncate the derived graph against this ghost.
2258
parent_text_key = None
2260
# look at the parent commit details inventories to
2261
# determine possible candidates in the per file graph.
2264
inv = inventory_cache[parent_id]
2266
inv = self.revision_tree(parent_id).inventory
2267
inventory_cache[parent_id] = inv
2269
parent_entry = inv[text_key[0]]
2270
except (KeyError, errors.NoSuchId):
2272
if parent_entry is not None:
2274
text_key[0], parent_entry.revision)
2276
parent_text_key = None
2277
if parent_text_key is not None:
2278
candidate_parents.append(
2279
text_key_cache[parent_text_key])
2280
parent_heads = text_graph.heads(candidate_parents)
2281
new_parents = list(parent_heads)
2282
new_parents.sort(key=lambda x:candidate_parents.index(x))
2283
if new_parents == []:
2284
new_parents = [NULL_REVISION]
2285
text_index[text_key] = new_parents
2287
for text_key in invalid_keys:
2288
text_index[text_key] = [NULL_REVISION]
2291
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2292
"""Get an iterable listing the keys of all the data introduced by a set
2295
The keys will be ordered so that the corresponding items can be safely
2296
fetched and inserted in that order.
2298
:returns: An iterable producing tuples of (knit-kind, file-id,
2299
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2300
'revisions'. file-id is None unless knit-kind is 'file'.
2302
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2305
for result in self._find_non_file_keys_to_fetch(revision_ids):
2308
def _find_file_keys_to_fetch(self, revision_ids, pb):
2309
# XXX: it's a bit weird to control the inventory weave caching in this
2310
# generator. Ideally the caching would be done in fetch.py I think. Or
2311
# maybe this generator should explicitly have the contract that it
2312
# should not be iterated until the previously yielded item has been
2314
inv_w = self.inventories
2316
# file ids that changed
2317
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2319
num_file_ids = len(file_ids)
2320
for file_id, altered_versions in file_ids.iteritems():
2322
pb.update("fetch texts", count, num_file_ids)
2324
yield ("file", file_id, altered_versions)
2326
def _find_non_file_keys_to_fetch(self, revision_ids):
2328
yield ("inventory", None, revision_ids)
2331
# XXX: Note ATM no callers actually pay attention to this return
2332
# instead they just use the list of revision ids and ignore
2333
# missing sigs. Consider removing this work entirely
2334
revisions_with_signatures = set(self.signatures.get_parent_map(
2335
[(r,) for r in revision_ids]))
2336
revisions_with_signatures = set(
2337
[r for (r,) in revisions_with_signatures])
2338
revisions_with_signatures.intersection_update(revision_ids)
2339
yield ("signatures", None, revisions_with_signatures)
2342
yield ("revisions", None, revision_ids)
2345
def get_inventory(self, revision_id):
2346
"""Get Inventory object by revision id."""
2347
return self.iter_inventories([revision_id]).next()
2349
def iter_inventories(self, revision_ids, ordering=None):
2350
"""Get many inventories by revision_ids.
2352
This will buffer some or all of the texts used in constructing the
2353
inventories in memory, but will only parse a single inventory at a
2356
:param revision_ids: The expected revision ids of the inventories.
2357
:param ordering: optional ordering, e.g. 'topological'. If not
2358
specified, the order of revision_ids will be preserved (by
2359
buffering if necessary).
2360
:return: An iterator of inventories.
2362
if ((None in revision_ids)
2363
or (_mod_revision.NULL_REVISION in revision_ids)):
2364
raise ValueError('cannot get null revision inventory')
2365
return self._iter_inventories(revision_ids, ordering)
2367
def _iter_inventories(self, revision_ids, ordering):
2368
"""single-document based inventory iteration."""
2369
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2370
for text, revision_id in inv_xmls:
2371
yield self.deserialise_inventory(revision_id, text)
2373
def _iter_inventory_xmls(self, revision_ids, ordering):
2374
if ordering is None:
2375
order_as_requested = True
2376
ordering = 'unordered'
2378
order_as_requested = False
2379
keys = [(revision_id,) for revision_id in revision_ids]
2382
if order_as_requested:
2383
key_iter = iter(keys)
2384
next_key = key_iter.next()
2385
stream = self.inventories.get_record_stream(keys, ordering, True)
2387
for record in stream:
2388
if record.storage_kind != 'absent':
2389
chunks = record.get_bytes_as('chunked')
2390
if order_as_requested:
2391
text_chunks[record.key] = chunks
2393
yield ''.join(chunks), record.key[-1]
2395
raise errors.NoSuchRevision(self, record.key)
2396
if order_as_requested:
2397
# Yield as many results as we can while preserving order.
2398
while next_key in text_chunks:
2399
chunks = text_chunks.pop(next_key)
2400
yield ''.join(chunks), next_key[-1]
2402
next_key = key_iter.next()
2403
except StopIteration:
2404
# We still want to fully consume the get_record_stream,
2405
# just in case it is not actually finished at this point
2409
def deserialise_inventory(self, revision_id, xml):
2410
"""Transform the xml into an inventory object.
2412
:param revision_id: The expected revision id of the inventory.
2413
:param xml: A serialised inventory.
2415
result = self._serializer.read_inventory_from_string(xml, revision_id,
2416
entry_cache=self._inventory_entry_cache)
2417
if result.revision_id != revision_id:
2418
raise AssertionError('revision id mismatch %s != %s' % (
2419
result.revision_id, revision_id))
2422
def serialise_inventory(self, inv):
2423
return self._serializer.write_inventory_to_string(inv)
2425
def _serialise_inventory_to_lines(self, inv):
2426
return self._serializer.write_inventory_to_lines(inv)
2428
def get_serializer_format(self):
2429
return self._serializer.format_num
2432
def get_inventory_xml(self, revision_id):
2433
"""Get inventory XML as a file object."""
2434
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2436
text, revision_id = texts.next()
2437
except StopIteration:
2438
raise errors.HistoryMissing(self, 'inventory', revision_id)
2442
def get_inventory_sha1(self, revision_id):
2443
"""Return the sha1 hash of the inventory entry
2445
return self.get_revision(revision_id).inventory_sha1
2447
def get_rev_id_for_revno(self, revno, known_pair):
2448
"""Return the revision id of a revno, given a later (revno, revid)
2449
pair in the same history.
2451
:return: if found (True, revid). If the available history ran out
2452
before reaching the revno, then this returns
2453
(False, (closest_revno, closest_revid)).
2455
known_revno, known_revid = known_pair
2456
partial_history = [known_revid]
2457
distance_from_known = known_revno - revno
2458
if distance_from_known < 0:
2460
'requested revno (%d) is later than given known revno (%d)'
2461
% (revno, known_revno))
2464
self, partial_history, stop_index=distance_from_known)
2465
except errors.RevisionNotPresent, err:
2466
if err.revision_id == known_revid:
2467
# The start revision (known_revid) wasn't found.
2469
# This is a stacked repository with no fallbacks, or a there's a
2470
# left-hand ghost. Either way, even though the revision named in
2471
# the error isn't in this repo, we know it's the next step in this
2472
# left-hand history.
2473
partial_history.append(err.revision_id)
2474
if len(partial_history) <= distance_from_known:
2475
# Didn't find enough history to get a revid for the revno.
2476
earliest_revno = known_revno - len(partial_history) + 1
2477
return (False, (earliest_revno, partial_history[-1]))
2478
if len(partial_history) - 1 > distance_from_known:
2479
raise AssertionError('_iter_for_revno returned too much history')
2480
return (True, partial_history[-1])
2482
def iter_reverse_revision_history(self, revision_id):
2483
"""Iterate backwards through revision ids in the lefthand history
2485
:param revision_id: The revision id to start with. All its lefthand
2486
ancestors will be traversed.
2488
graph = self.get_graph()
2489
next_id = revision_id
2491
if next_id in (None, _mod_revision.NULL_REVISION):
2494
parents = graph.get_parent_map([next_id])[next_id]
2496
raise errors.RevisionNotPresent(next_id, self)
2498
if len(parents) == 0:
2501
next_id = parents[0]
2504
def get_revision_inventory(self, revision_id):
2505
"""Return inventory of a past revision."""
2506
# TODO: Unify this with get_inventory()
2507
# bzr 0.0.6 and later imposes the constraint that the inventory_id
2508
# must be the same as its revision, so this is trivial.
2509
if revision_id is None:
2510
# This does not make sense: if there is no revision,
2511
# then it is the current tree inventory surely ?!
2512
# and thus get_root_id() is something that looks at the last
2513
# commit on the branch, and the get_root_id is an inventory check.
2514
raise NotImplementedError
2515
# return Inventory(self.get_root_id())
2517
return self.get_inventory(revision_id)
2519
def is_shared(self):
2520
"""Return True if this repository is flagged as a shared repository."""
2521
raise NotImplementedError(self.is_shared)
2524
def reconcile(self, other=None, thorough=False):
2525
"""Reconcile this repository."""
2526
from bzrlib.reconcile import RepoReconciler
2527
reconciler = RepoReconciler(self, thorough=thorough)
2528
reconciler.reconcile()
2531
def _refresh_data(self):
2532
"""Helper called from lock_* to ensure coherency with disk.
2534
The default implementation does nothing; it is however possible
2535
for repositories to maintain loaded indices across multiple locks
2536
by checking inside their implementation of this method to see
2537
whether their indices are still valid. This depends of course on
2538
the disk format being validatable in this manner. This method is
2539
also called by the refresh_data() public interface to cause a refresh
2540
to occur while in a write lock so that data inserted by a smart server
2541
push operation is visible on the client's instance of the physical
2546
def revision_tree(self, revision_id):
2547
"""Return Tree for a revision on this branch.
2549
`revision_id` may be NULL_REVISION for the empty tree revision.
2551
revision_id = _mod_revision.ensure_null(revision_id)
2552
# TODO: refactor this to use an existing revision object
2553
# so we don't need to read it in twice.
2554
if revision_id == _mod_revision.NULL_REVISION:
2555
return RevisionTree(self, Inventory(root_id=None),
2556
_mod_revision.NULL_REVISION)
2558
inv = self.get_revision_inventory(revision_id)
2559
return RevisionTree(self, inv, revision_id)
2561
def revision_trees(self, revision_ids):
2562
"""Return Trees for revisions in this repository.
2564
:param revision_ids: a sequence of revision-ids;
2565
a revision-id may not be None or 'null:'
2567
inventories = self.iter_inventories(revision_ids)
2568
for inv in inventories:
2569
yield RevisionTree(self, inv, inv.revision_id)
2571
def _filtered_revision_trees(self, revision_ids, file_ids):
2572
"""Return Tree for a revision on this branch with only some files.
2574
:param revision_ids: a sequence of revision-ids;
2575
a revision-id may not be None or 'null:'
2576
:param file_ids: if not None, the result is filtered
2577
so that only those file-ids, their parents and their
2578
children are included.
2580
inventories = self.iter_inventories(revision_ids)
2581
for inv in inventories:
2582
# Should we introduce a FilteredRevisionTree class rather
2583
# than pre-filter the inventory here?
2584
filtered_inv = inv.filter(file_ids)
2585
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2588
def get_ancestry(self, revision_id, topo_sorted=True):
2589
"""Return a list of revision-ids integrated by a revision.
2591
The first element of the list is always None, indicating the origin
2592
revision. This might change when we have history horizons, or
2593
perhaps we should have a new API.
2595
This is topologically sorted.
2597
if _mod_revision.is_null(revision_id):
2599
if not self.has_revision(revision_id):
2600
raise errors.NoSuchRevision(self, revision_id)
2601
graph = self.get_graph()
2603
search = graph._make_breadth_first_searcher([revision_id])
2606
found, ghosts = search.next_with_ghosts()
2607
except StopIteration:
2610
if _mod_revision.NULL_REVISION in keys:
2611
keys.remove(_mod_revision.NULL_REVISION)
2613
parent_map = graph.get_parent_map(keys)
2614
keys = tsort.topo_sort(parent_map)
2615
return [None] + list(keys)
2617
def pack(self, hint=None):
2618
"""Compress the data within the repository.
2620
This operation only makes sense for some repository types. For other
2621
types it should be a no-op that just returns.
2623
This stub method does not require a lock, but subclasses should use
2624
@needs_write_lock as this is a long running call its reasonable to
2625
implicitly lock for the user.
2627
:param hint: If not supplied, the whole repository is packed.
2628
If supplied, the repository may use the hint parameter as a
2629
hint for the parts of the repository to pack. A hint can be
2630
obtained from the result of commit_write_group(). Out of
2631
date hints are simply ignored, because concurrent operations
2632
can obsolete them rapidly.
2635
def get_transaction(self):
2636
return self.control_files.get_transaction()
2638
def get_parent_map(self, revision_ids):
2639
"""See graph.StackedParentsProvider.get_parent_map"""
2640
# revisions index works in keys; this just works in revisions
2641
# therefore wrap and unwrap
2644
for revision_id in revision_ids:
2645
if revision_id == _mod_revision.NULL_REVISION:
2646
result[revision_id] = ()
2647
elif revision_id is None:
2648
raise ValueError('get_parent_map(None) is not valid')
2650
query_keys.append((revision_id ,))
2651
for ((revision_id,), parent_keys) in \
2652
self.revisions.get_parent_map(query_keys).iteritems():
2654
result[revision_id] = tuple(parent_revid
2655
for (parent_revid,) in parent_keys)
2657
result[revision_id] = (_mod_revision.NULL_REVISION,)
2660
def _make_parents_provider(self):
2663
def get_graph(self, other_repository=None):
2664
"""Return the graph walker for this repository format"""
2665
parents_provider = self._make_parents_provider()
2666
if (other_repository is not None and
2667
not self.has_same_location(other_repository)):
2668
parents_provider = graph.StackedParentsProvider(
2669
[parents_provider, other_repository._make_parents_provider()])
2670
return graph.Graph(parents_provider)
2672
def _get_versioned_file_checker(self, text_key_references=None,
2674
"""Return an object suitable for checking versioned files.
2676
:param text_key_references: if non-None, an already built
2677
dictionary mapping text keys ((fileid, revision_id) tuples)
2678
to whether they were referred to by the inventory of the
2679
revision_id that they contain. If None, this will be
2681
:param ancestors: Optional result from
2682
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2685
return _VersionedFileChecker(self,
2686
text_key_references=text_key_references, ancestors=ancestors)
2688
def revision_ids_to_search_result(self, result_set):
2689
"""Convert a set of revision ids to a graph SearchResult."""
2690
result_parents = set()
2691
for parents in self.get_graph().get_parent_map(
2692
result_set).itervalues():
2693
result_parents.update(parents)
2694
included_keys = result_set.intersection(result_parents)
2695
start_keys = result_set.difference(included_keys)
2696
exclude_keys = result_parents.difference(result_set)
2697
result = graph.SearchResult(start_keys, exclude_keys,
2698
len(result_set), result_set)
2702
def set_make_working_trees(self, new_value):
2703
"""Set the policy flag for making working trees when creating branches.
2705
This only applies to branches that use this repository.
2707
The default is 'True'.
2708
:param new_value: True to restore the default, False to disable making
2711
raise NotImplementedError(self.set_make_working_trees)
2713
def make_working_trees(self):
2714
"""Returns the policy for making working trees on new branches."""
2715
raise NotImplementedError(self.make_working_trees)
2718
def sign_revision(self, revision_id, gpg_strategy):
2719
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2720
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2723
def has_signature_for_revision_id(self, revision_id):
2724
"""Query for a revision signature for revision_id in the repository."""
2725
if not self.has_revision(revision_id):
2726
raise errors.NoSuchRevision(self, revision_id)
2727
sig_present = (1 == len(
2728
self.signatures.get_parent_map([(revision_id,)])))
2732
def get_signature_text(self, revision_id):
2733
"""Return the text for a signature."""
2734
stream = self.signatures.get_record_stream([(revision_id,)],
2736
record = stream.next()
2737
if record.storage_kind == 'absent':
2738
raise errors.NoSuchRevision(self, revision_id)
2739
return record.get_bytes_as('fulltext')
2742
def check(self, revision_ids=None, callback_refs=None, check_repo=True):
2743
"""Check consistency of all history of given revision_ids.
2745
Different repository implementations should override _check().
2747
:param revision_ids: A non-empty list of revision_ids whose ancestry
2748
will be checked. Typically the last revision_id of a branch.
2749
:param callback_refs: A dict of check-refs to resolve and callback
2750
the check/_check method on the items listed as wanting the ref.
2752
:param check_repo: If False do not check the repository contents, just
2753
calculate the data callback_refs requires and call them back.
2755
return self._check(revision_ids, callback_refs=callback_refs,
2756
check_repo=check_repo)
2758
def _check(self, revision_ids, callback_refs, check_repo):
2759
result = check.Check(self, check_repo=check_repo)
2760
result.check(callback_refs)
2763
def _warn_if_deprecated(self):
2764
global _deprecation_warning_done
2765
if _deprecation_warning_done:
2767
_deprecation_warning_done = True
2768
warning("Format %s for %s is deprecated - please use 'bzr upgrade' to get better performance"
2769
% (self._format, self.bzrdir.transport.base))
2771
def supports_rich_root(self):
2772
return self._format.rich_root_data
2774
def _check_ascii_revisionid(self, revision_id, method):
2775
"""Private helper for ascii-only repositories."""
2776
# weave repositories refuse to store revisionids that are non-ascii.
2777
if revision_id is not None:
2778
# weaves require ascii revision ids.
2779
if isinstance(revision_id, unicode):
2781
revision_id.encode('ascii')
2782
except UnicodeEncodeError:
2783
raise errors.NonAsciiRevisionId(method, self)
2786
revision_id.decode('ascii')
2787
except UnicodeDecodeError:
2788
raise errors.NonAsciiRevisionId(method, self)
2790
def revision_graph_can_have_wrong_parents(self):
2791
"""Is it possible for this repository to have a revision graph with
2794
If True, then this repository must also implement
2795
_find_inconsistent_revision_parents so that check and reconcile can
2796
check for inconsistencies before proceeding with other checks that may
2797
depend on the revision index being consistent.
2799
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2802
# remove these delegates a while after bzr 0.15
2803
def __make_delegated(name, from_module):
2804
def _deprecated_repository_forwarder():
2805
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2806
% (name, from_module),
2809
m = __import__(from_module, globals(), locals(), [name])
2811
return getattr(m, name)
2812
except AttributeError:
2813
raise AttributeError('module %s has no name %s'
2815
globals()[name] = _deprecated_repository_forwarder
2818
'AllInOneRepository',
2819
'WeaveMetaDirRepository',
2820
'PreSplitOutRepositoryFormat',
2821
'RepositoryFormat4',
2822
'RepositoryFormat5',
2823
'RepositoryFormat6',
2824
'RepositoryFormat7',
2826
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2830
'RepositoryFormatKnit',
2831
'RepositoryFormatKnit1',
2833
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2836
def install_revision(repository, rev, revision_tree):
2837
"""Install all revision data into a repository."""
2838
install_revisions(repository, [(rev, revision_tree, None)])
2841
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2842
"""Install all revision data into a repository.
2844
Accepts an iterable of revision, tree, signature tuples. The signature
2847
repository.start_write_group()
2849
inventory_cache = lru_cache.LRUCache(10)
2850
for n, (revision, revision_tree, signature) in enumerate(iterable):
2851
_install_revision(repository, revision, revision_tree, signature,
2854
pb.update('Transferring revisions', n + 1, num_revisions)
2856
repository.abort_write_group()
2859
repository.commit_write_group()
2862
def _install_revision(repository, rev, revision_tree, signature,
2864
"""Install all revision data into a repository."""
2865
present_parents = []
2867
for p_id in rev.parent_ids:
2868
if repository.has_revision(p_id):
2869
present_parents.append(p_id)
2870
parent_trees[p_id] = repository.revision_tree(p_id)
2872
parent_trees[p_id] = repository.revision_tree(
2873
_mod_revision.NULL_REVISION)
2875
inv = revision_tree.inventory
2876
entries = inv.iter_entries()
2877
# backwards compatibility hack: skip the root id.
2878
if not repository.supports_rich_root():
2879
path, root = entries.next()
2880
if root.revision != rev.revision_id:
2881
raise errors.IncompatibleRevision(repr(repository))
2883
for path, ie in entries:
2884
text_keys[(ie.file_id, ie.revision)] = ie
2885
text_parent_map = repository.texts.get_parent_map(text_keys)
2886
missing_texts = set(text_keys) - set(text_parent_map)
2887
# Add the texts that are not already present
2888
for text_key in missing_texts:
2889
ie = text_keys[text_key]
2891
# FIXME: TODO: The following loop overlaps/duplicates that done by
2892
# commit to determine parents. There is a latent/real bug here where
2893
# the parents inserted are not those commit would do - in particular
2894
# they are not filtered by heads(). RBC, AB
2895
for revision, tree in parent_trees.iteritems():
2896
if ie.file_id not in tree:
2898
parent_id = tree.inventory[ie.file_id].revision
2899
if parent_id in text_parents:
2901
text_parents.append((ie.file_id, parent_id))
2902
lines = revision_tree.get_file(ie.file_id).readlines()
2903
repository.texts.add_lines(text_key, text_parents, lines)
2905
# install the inventory
2906
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2907
# Cache this inventory
2908
inventory_cache[rev.revision_id] = inv
2910
basis_inv = inventory_cache[rev.parent_ids[0]]
2912
repository.add_inventory(rev.revision_id, inv, present_parents)
2914
delta = inv._make_delta(basis_inv)
2915
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2916
rev.revision_id, present_parents)
2918
repository.add_inventory(rev.revision_id, inv, present_parents)
2919
except errors.RevisionAlreadyPresent:
2921
if signature is not None:
2922
repository.add_signature_text(rev.revision_id, signature)
2923
repository.add_revision(rev.revision_id, rev, inv)
2926
class MetaDirRepository(Repository):
2927
"""Repositories in the new meta-dir layout.
2929
:ivar _transport: Transport for access to repository control files,
2930
typically pointing to .bzr/repository.
2933
def __init__(self, _format, a_bzrdir, control_files):
2934
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2935
self._transport = control_files._transport
2937
def is_shared(self):
2938
"""Return True if this repository is flagged as a shared repository."""
2939
return self._transport.has('shared-storage')
2942
def set_make_working_trees(self, new_value):
2943
"""Set the policy flag for making working trees when creating branches.
2945
This only applies to branches that use this repository.
2947
The default is 'True'.
2948
:param new_value: True to restore the default, False to disable making
2953
self._transport.delete('no-working-trees')
2954
except errors.NoSuchFile:
2957
self._transport.put_bytes('no-working-trees', '',
2958
mode=self.bzrdir._get_file_mode())
2960
def make_working_trees(self):
2961
"""Returns the policy for making working trees on new branches."""
2962
return not self._transport.has('no-working-trees')
2965
class MetaDirVersionedFileRepository(MetaDirRepository):
2966
"""Repositories in a meta-dir, that work via versioned file objects."""
2968
def __init__(self, _format, a_bzrdir, control_files):
2969
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2973
network_format_registry = registry.FormatRegistry()
2974
"""Registry of formats indexed by their network name.
2976
The network name for a repository format is an identifier that can be used when
2977
referring to formats with smart server operations. See
2978
RepositoryFormat.network_name() for more detail.
2982
format_registry = registry.FormatRegistry(network_format_registry)
2983
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
2985
This can contain either format instances themselves, or classes/factories that
2986
can be called to obtain one.
2990
#####################################################################
2991
# Repository Formats
2993
class RepositoryFormat(object):
2994
"""A repository format.
2996
Formats provide four things:
2997
* An initialization routine to construct repository data on disk.
2998
* a optional format string which is used when the BzrDir supports
3000
* an open routine which returns a Repository instance.
3001
* A network name for referring to the format in smart server RPC
3004
There is one and only one Format subclass for each on-disk format. But
3005
there can be one Repository subclass that is used for several different
3006
formats. The _format attribute on a Repository instance can be used to
3007
determine the disk format.
3009
Formats are placed in a registry by their format string for reference
3010
during opening. These should be subclasses of RepositoryFormat for
3013
Once a format is deprecated, just deprecate the initialize and open
3014
methods on the format class. Do not deprecate the object, as the
3015
object may be created even when a repository instance hasn't been
3018
Common instance attributes:
3019
_matchingbzrdir - the bzrdir format that the repository format was
3020
originally written to work with. This can be used if manually
3021
constructing a bzrdir and repository, or more commonly for test suite
3025
# Set to True or False in derived classes. True indicates that the format
3026
# supports ghosts gracefully.
3027
supports_ghosts = None
3028
# Can this repository be given external locations to lookup additional
3029
# data. Set to True or False in derived classes.
3030
supports_external_lookups = None
3031
# Does this format support CHK bytestring lookups. Set to True or False in
3033
supports_chks = None
3034
# Should commit add an inventory, or an inventory delta to the repository.
3035
_commit_inv_deltas = True
3036
# What order should fetch operations request streams in?
3037
# The default is unordered as that is the cheapest for an origin to
3039
_fetch_order = 'unordered'
3040
# Does this repository format use deltas that can be fetched as-deltas ?
3041
# (E.g. knits, where the knit deltas can be transplanted intact.
3042
# We default to False, which will ensure that enough data to get
3043
# a full text out of any fetch stream will be grabbed.
3044
_fetch_uses_deltas = False
3045
# Should fetch trigger a reconcile after the fetch? Only needed for
3046
# some repository formats that can suffer internal inconsistencies.
3047
_fetch_reconcile = False
3048
# Does this format have < O(tree_size) delta generation. Used to hint what
3049
# code path for commit, amongst other things.
3051
# Does doing a pack operation compress data? Useful for the pack UI command
3052
# (so if there is one pack, the operation can still proceed because it may
3053
# help), and for fetching when data won't have come from the same
3055
pack_compresses = False
3058
return "<%s>" % self.__class__.__name__
3060
def __eq__(self, other):
3061
# format objects are generally stateless
3062
return isinstance(other, self.__class__)
3064
def __ne__(self, other):
3065
return not self == other
3068
def find_format(klass, a_bzrdir):
3069
"""Return the format for the repository object in a_bzrdir.
3071
This is used by bzr native formats that have a "format" file in
3072
the repository. Other methods may be used by different types of
3076
transport = a_bzrdir.get_repository_transport(None)
3077
format_string = transport.get("format").read()
3078
return format_registry.get(format_string)
3079
except errors.NoSuchFile:
3080
raise errors.NoRepositoryPresent(a_bzrdir)
3082
raise errors.UnknownFormatError(format=format_string,
3086
def register_format(klass, format):
3087
format_registry.register(format.get_format_string(), format)
3090
def unregister_format(klass, format):
3091
format_registry.remove(format.get_format_string())
3094
def get_default_format(klass):
3095
"""Return the current default format."""
3096
from bzrlib import bzrdir
3097
return bzrdir.format_registry.make_bzrdir('default').repository_format
3099
def get_format_string(self):
3100
"""Return the ASCII format string that identifies this format.
3102
Note that in pre format ?? repositories the format string is
3103
not permitted nor written to disk.
3105
raise NotImplementedError(self.get_format_string)
3107
def get_format_description(self):
3108
"""Return the short description for this format."""
3109
raise NotImplementedError(self.get_format_description)
3111
# TODO: this shouldn't be in the base class, it's specific to things that
3112
# use weaves or knits -- mbp 20070207
3113
def _get_versioned_file_store(self,
3118
versionedfile_class=None,
3119
versionedfile_kwargs={},
3121
if versionedfile_class is None:
3122
versionedfile_class = self._versionedfile_class
3123
weave_transport = control_files._transport.clone(name)
3124
dir_mode = control_files._dir_mode
3125
file_mode = control_files._file_mode
3126
return VersionedFileStore(weave_transport, prefixed=prefixed,
3128
file_mode=file_mode,
3129
versionedfile_class=versionedfile_class,
3130
versionedfile_kwargs=versionedfile_kwargs,
3133
def initialize(self, a_bzrdir, shared=False):
3134
"""Initialize a repository of this format in a_bzrdir.
3136
:param a_bzrdir: The bzrdir to put the new repository in it.
3137
:param shared: The repository should be initialized as a sharable one.
3138
:returns: The new repository object.
3140
This may raise UninitializableFormat if shared repository are not
3141
compatible the a_bzrdir.
3143
raise NotImplementedError(self.initialize)
3145
def is_supported(self):
3146
"""Is this format supported?
3148
Supported formats must be initializable and openable.
3149
Unsupported formats may not support initialization or committing or
3150
some other features depending on the reason for not being supported.
3154
def network_name(self):
3155
"""A simple byte string uniquely identifying this format for RPC calls.
3157
MetaDir repository formats use their disk format string to identify the
3158
repository over the wire. All in one formats such as bzr < 0.8, and
3159
foreign formats like svn/git and hg should use some marker which is
3160
unique and immutable.
3162
raise NotImplementedError(self.network_name)
3164
def check_conversion_target(self, target_format):
3165
raise NotImplementedError(self.check_conversion_target)
3167
def open(self, a_bzrdir, _found=False):
3168
"""Return an instance of this format for the bzrdir a_bzrdir.
3170
_found is a private parameter, do not use it.
3172
raise NotImplementedError(self.open)
3175
class MetaDirRepositoryFormat(RepositoryFormat):
3176
"""Common base class for the new repositories using the metadir layout."""
3178
rich_root_data = False
3179
supports_tree_reference = False
3180
supports_external_lookups = False
3183
def _matchingbzrdir(self):
3184
matching = bzrdir.BzrDirMetaFormat1()
3185
matching.repository_format = self
3189
super(MetaDirRepositoryFormat, self).__init__()
3191
def _create_control_files(self, a_bzrdir):
3192
"""Create the required files and the initial control_files object."""
3193
# FIXME: RBC 20060125 don't peek under the covers
3194
# NB: no need to escape relative paths that are url safe.
3195
repository_transport = a_bzrdir.get_repository_transport(self)
3196
control_files = lockable_files.LockableFiles(repository_transport,
3197
'lock', lockdir.LockDir)
3198
control_files.create_lock()
3199
return control_files
3201
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
3202
"""Upload the initial blank content."""
3203
control_files = self._create_control_files(a_bzrdir)
3204
control_files.lock_write()
3205
transport = control_files._transport
3207
utf8_files += [('shared-storage', '')]
3209
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
3210
for (filename, content_stream) in files:
3211
transport.put_file(filename, content_stream,
3212
mode=a_bzrdir._get_file_mode())
3213
for (filename, content_bytes) in utf8_files:
3214
transport.put_bytes_non_atomic(filename, content_bytes,
3215
mode=a_bzrdir._get_file_mode())
3217
control_files.unlock()
3219
def network_name(self):
3220
"""Metadir formats have matching disk and network format strings."""
3221
return self.get_format_string()
3224
# Pre-0.8 formats that don't have a disk format string (because they are
3225
# versioned by the matching control directory). We use the control directories
3226
# disk format string as a key for the network_name because they meet the
3227
# constraints (simple string, unique, immutable).
3228
network_format_registry.register_lazy(
3229
"Bazaar-NG branch, format 5\n",
3230
'bzrlib.repofmt.weaverepo',
3231
'RepositoryFormat5',
3233
network_format_registry.register_lazy(
3234
"Bazaar-NG branch, format 6\n",
3235
'bzrlib.repofmt.weaverepo',
3236
'RepositoryFormat6',
3239
# formats which have no format string are not discoverable or independently
3240
# creatable on disk, so are not registered in format_registry. They're
3241
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
3242
# needed, it's constructed directly by the BzrDir. Non-native formats where
3243
# the repository is not separately opened are similar.
3245
format_registry.register_lazy(
3246
'Bazaar-NG Repository format 7',
3247
'bzrlib.repofmt.weaverepo',
3251
format_registry.register_lazy(
3252
'Bazaar-NG Knit Repository Format 1',
3253
'bzrlib.repofmt.knitrepo',
3254
'RepositoryFormatKnit1',
3257
format_registry.register_lazy(
3258
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
3259
'bzrlib.repofmt.knitrepo',
3260
'RepositoryFormatKnit3',
3263
format_registry.register_lazy(
3264
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
3265
'bzrlib.repofmt.knitrepo',
3266
'RepositoryFormatKnit4',
3269
# Pack-based formats. There is one format for pre-subtrees, and one for
3270
# post-subtrees to allow ease of testing.
3271
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
3272
format_registry.register_lazy(
3273
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
3274
'bzrlib.repofmt.pack_repo',
3275
'RepositoryFormatKnitPack1',
3277
format_registry.register_lazy(
3278
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
3279
'bzrlib.repofmt.pack_repo',
3280
'RepositoryFormatKnitPack3',
3282
format_registry.register_lazy(
3283
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
3284
'bzrlib.repofmt.pack_repo',
3285
'RepositoryFormatKnitPack4',
3287
format_registry.register_lazy(
3288
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
3289
'bzrlib.repofmt.pack_repo',
3290
'RepositoryFormatKnitPack5',
3292
format_registry.register_lazy(
3293
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
3294
'bzrlib.repofmt.pack_repo',
3295
'RepositoryFormatKnitPack5RichRoot',
3297
format_registry.register_lazy(
3298
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3299
'bzrlib.repofmt.pack_repo',
3300
'RepositoryFormatKnitPack5RichRootBroken',
3302
format_registry.register_lazy(
3303
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3304
'bzrlib.repofmt.pack_repo',
3305
'RepositoryFormatKnitPack6',
3307
format_registry.register_lazy(
3308
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3309
'bzrlib.repofmt.pack_repo',
3310
'RepositoryFormatKnitPack6RichRoot',
3313
# Development formats.
3314
# Obsolete but kept pending a CHK based subtree format.
3315
format_registry.register_lazy(
3316
("Bazaar development format 2 with subtree support "
3317
"(needs bzr.dev from before 1.8)\n"),
3318
'bzrlib.repofmt.pack_repo',
3319
'RepositoryFormatPackDevelopment2Subtree',
3322
# 1.14->1.16 go below here
3323
format_registry.register_lazy(
3324
'Bazaar development format - group compression and chk inventory'
3325
' (needs bzr.dev from 1.14)\n',
3326
'bzrlib.repofmt.groupcompress_repo',
3327
'RepositoryFormatCHK1',
3330
format_registry.register_lazy(
3331
'Bazaar development format - chk repository with bencode revision '
3332
'serialization (needs bzr.dev from 1.16)\n',
3333
'bzrlib.repofmt.groupcompress_repo',
3334
'RepositoryFormatCHK2',
3336
format_registry.register_lazy(
3337
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3338
'bzrlib.repofmt.groupcompress_repo',
3339
'RepositoryFormat2a',
3343
class InterRepository(InterObject):
3344
"""This class represents operations taking place between two repositories.
3346
Its instances have methods like copy_content and fetch, and contain
3347
references to the source and target repositories these operations can be
3350
Often we will provide convenience methods on 'repository' which carry out
3351
operations with another repository - they will always forward to
3352
InterRepository.get(other).method_name(parameters).
3355
_walk_to_common_revisions_batch_size = 50
3357
"""The available optimised InterRepository types."""
3360
def copy_content(self, revision_id=None):
3361
"""Make a complete copy of the content in self into destination.
3363
This is a destructive operation! Do not use it on existing
3366
:param revision_id: Only copy the content needed to construct
3367
revision_id and its parents.
3370
self.target.set_make_working_trees(self.source.make_working_trees())
3371
except NotImplementedError:
3373
self.target.fetch(self.source, revision_id=revision_id)
3376
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3378
"""Fetch the content required to construct revision_id.
3380
The content is copied from self.source to self.target.
3382
:param revision_id: if None all content is copied, if NULL_REVISION no
3384
:param pb: optional progress bar to use for progress reports. If not
3385
provided a default one will be created.
3388
from bzrlib.fetch import RepoFetcher
3389
f = RepoFetcher(to_repository=self.target,
3390
from_repository=self.source,
3391
last_revision=revision_id,
3392
fetch_spec=fetch_spec,
3393
pb=pb, find_ghosts=find_ghosts)
3395
def _walk_to_common_revisions(self, revision_ids):
3396
"""Walk out from revision_ids in source to revisions target has.
3398
:param revision_ids: The start point for the search.
3399
:return: A set of revision ids.
3401
target_graph = self.target.get_graph()
3402
revision_ids = frozenset(revision_ids)
3403
missing_revs = set()
3404
source_graph = self.source.get_graph()
3405
# ensure we don't pay silly lookup costs.
3406
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3407
null_set = frozenset([_mod_revision.NULL_REVISION])
3408
searcher_exhausted = False
3412
# Iterate the searcher until we have enough next_revs
3413
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3415
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3416
next_revs.update(next_revs_part)
3417
ghosts.update(ghosts_part)
3418
except StopIteration:
3419
searcher_exhausted = True
3421
# If there are ghosts in the source graph, and the caller asked for
3422
# them, make sure that they are present in the target.
3423
# We don't care about other ghosts as we can't fetch them and
3424
# haven't been asked to.
3425
ghosts_to_check = set(revision_ids.intersection(ghosts))
3426
revs_to_get = set(next_revs).union(ghosts_to_check)
3428
have_revs = set(target_graph.get_parent_map(revs_to_get))
3429
# we always have NULL_REVISION present.
3430
have_revs = have_revs.union(null_set)
3431
# Check if the target is missing any ghosts we need.
3432
ghosts_to_check.difference_update(have_revs)
3434
# One of the caller's revision_ids is a ghost in both the
3435
# source and the target.
3436
raise errors.NoSuchRevision(
3437
self.source, ghosts_to_check.pop())
3438
missing_revs.update(next_revs - have_revs)
3439
# Because we may have walked past the original stop point, make
3440
# sure everything is stopped
3441
stop_revs = searcher.find_seen_ancestors(have_revs)
3442
searcher.stop_searching_any(stop_revs)
3443
if searcher_exhausted:
3445
return searcher.get_result()
3448
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3449
"""Return the revision ids that source has that target does not.
3451
:param revision_id: only return revision ids included by this
3453
:param find_ghosts: If True find missing revisions in deep history
3454
rather than just finding the surface difference.
3455
:return: A bzrlib.graph.SearchResult.
3457
# stop searching at found target revisions.
3458
if not find_ghosts and revision_id is not None:
3459
return self._walk_to_common_revisions([revision_id])
3460
# generic, possibly worst case, slow code path.
3461
target_ids = set(self.target.all_revision_ids())
3462
if revision_id is not None:
3463
source_ids = self.source.get_ancestry(revision_id)
3464
if source_ids[0] is not None:
3465
raise AssertionError()
3468
source_ids = self.source.all_revision_ids()
3469
result_set = set(source_ids).difference(target_ids)
3470
return self.source.revision_ids_to_search_result(result_set)
3473
def _same_model(source, target):
3474
"""True if source and target have the same data representation.
3476
Note: this is always called on the base class; overriding it in a
3477
subclass will have no effect.
3480
InterRepository._assert_same_model(source, target)
3482
except errors.IncompatibleRepositories, e:
3486
def _assert_same_model(source, target):
3487
"""Raise an exception if two repositories do not use the same model.
3489
if source.supports_rich_root() != target.supports_rich_root():
3490
raise errors.IncompatibleRepositories(source, target,
3491
"different rich-root support")
3492
if source._serializer != target._serializer:
3493
raise errors.IncompatibleRepositories(source, target,
3494
"different serializers")
3497
class InterSameDataRepository(InterRepository):
3498
"""Code for converting between repositories that represent the same data.
3500
Data format and model must match for this to work.
3504
def _get_repo_format_to_test(self):
3505
"""Repository format for testing with.
3507
InterSameData can pull from subtree to subtree and from non-subtree to
3508
non-subtree, so we test this with the richest repository format.
3510
from bzrlib.repofmt import knitrepo
3511
return knitrepo.RepositoryFormatKnit3()
3514
def is_compatible(source, target):
3515
return InterRepository._same_model(source, target)
3518
class InterWeaveRepo(InterSameDataRepository):
3519
"""Optimised code paths between Weave based repositories.
3521
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3522
implemented lazy inter-object optimisation.
3526
def _get_repo_format_to_test(self):
3527
from bzrlib.repofmt import weaverepo
3528
return weaverepo.RepositoryFormat7()
3531
def is_compatible(source, target):
3532
"""Be compatible with known Weave formats.
3534
We don't test for the stores being of specific types because that
3535
could lead to confusing results, and there is no need to be
3538
from bzrlib.repofmt.weaverepo import (
3544
return (isinstance(source._format, (RepositoryFormat5,
3546
RepositoryFormat7)) and
3547
isinstance(target._format, (RepositoryFormat5,
3549
RepositoryFormat7)))
3550
except AttributeError:
3554
def copy_content(self, revision_id=None):
3555
"""See InterRepository.copy_content()."""
3556
# weave specific optimised path:
3558
self.target.set_make_working_trees(self.source.make_working_trees())
3559
except (errors.RepositoryUpgradeRequired, NotImplemented):
3561
# FIXME do not peek!
3562
if self.source._transport.listable():
3563
pb = ui.ui_factory.nested_progress_bar()
3565
self.target.texts.insert_record_stream(
3566
self.source.texts.get_record_stream(
3567
self.source.texts.keys(), 'topological', False))
3568
pb.update('copying inventory', 0, 1)
3569
self.target.inventories.insert_record_stream(
3570
self.source.inventories.get_record_stream(
3571
self.source.inventories.keys(), 'topological', False))
3572
self.target.signatures.insert_record_stream(
3573
self.source.signatures.get_record_stream(
3574
self.source.signatures.keys(),
3576
self.target.revisions.insert_record_stream(
3577
self.source.revisions.get_record_stream(
3578
self.source.revisions.keys(),
3579
'topological', True))
3583
self.target.fetch(self.source, revision_id=revision_id)
3586
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3587
"""See InterRepository.missing_revision_ids()."""
3588
# we want all revisions to satisfy revision_id in source.
3589
# but we don't want to stat every file here and there.
3590
# we want then, all revisions other needs to satisfy revision_id
3591
# checked, but not those that we have locally.
3592
# so the first thing is to get a subset of the revisions to
3593
# satisfy revision_id in source, and then eliminate those that
3594
# we do already have.
3595
# this is slow on high latency connection to self, but as this
3596
# disk format scales terribly for push anyway due to rewriting
3597
# inventory.weave, this is considered acceptable.
3599
if revision_id is not None:
3600
source_ids = self.source.get_ancestry(revision_id)
3601
if source_ids[0] is not None:
3602
raise AssertionError()
3605
source_ids = self.source._all_possible_ids()
3606
source_ids_set = set(source_ids)
3607
# source_ids is the worst possible case we may need to pull.
3608
# now we want to filter source_ids against what we actually
3609
# have in target, but don't try to check for existence where we know
3610
# we do not have a revision as that would be pointless.
3611
target_ids = set(self.target._all_possible_ids())
3612
possibly_present_revisions = target_ids.intersection(source_ids_set)
3613
actually_present_revisions = set(
3614
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3615
required_revisions = source_ids_set.difference(actually_present_revisions)
3616
if revision_id is not None:
3617
# we used get_ancestry to determine source_ids then we are assured all
3618
# revisions referenced are present as they are installed in topological order.
3619
# and the tip revision was validated by get_ancestry.
3620
result_set = required_revisions
3622
# if we just grabbed the possibly available ids, then
3623
# we only have an estimate of whats available and need to validate
3624
# that against the revision records.
3626
self.source._eliminate_revisions_not_present(required_revisions))
3627
return self.source.revision_ids_to_search_result(result_set)
3630
class InterKnitRepo(InterSameDataRepository):
3631
"""Optimised code paths between Knit based repositories."""
3634
def _get_repo_format_to_test(self):
3635
from bzrlib.repofmt import knitrepo
3636
return knitrepo.RepositoryFormatKnit1()
3639
def is_compatible(source, target):
3640
"""Be compatible with known Knit formats.
3642
We don't test for the stores being of specific types because that
3643
could lead to confusing results, and there is no need to be
3646
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3648
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3649
isinstance(target._format, RepositoryFormatKnit))
3650
except AttributeError:
3652
return are_knits and InterRepository._same_model(source, target)
3655
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3656
"""See InterRepository.missing_revision_ids()."""
3657
if revision_id is not None:
3658
source_ids = self.source.get_ancestry(revision_id)
3659
if source_ids[0] is not None:
3660
raise AssertionError()
3663
source_ids = self.source.all_revision_ids()
3664
source_ids_set = set(source_ids)
3665
# source_ids is the worst possible case we may need to pull.
3666
# now we want to filter source_ids against what we actually
3667
# have in target, but don't try to check for existence where we know
3668
# we do not have a revision as that would be pointless.
3669
target_ids = set(self.target.all_revision_ids())
3670
possibly_present_revisions = target_ids.intersection(source_ids_set)
3671
actually_present_revisions = set(
3672
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3673
required_revisions = source_ids_set.difference(actually_present_revisions)
3674
if revision_id is not None:
3675
# we used get_ancestry to determine source_ids then we are assured all
3676
# revisions referenced are present as they are installed in topological order.
3677
# and the tip revision was validated by get_ancestry.
3678
result_set = required_revisions
3680
# if we just grabbed the possibly available ids, then
3681
# we only have an estimate of whats available and need to validate
3682
# that against the revision records.
3684
self.source._eliminate_revisions_not_present(required_revisions))
3685
return self.source.revision_ids_to_search_result(result_set)
3688
class InterDifferingSerializer(InterRepository):
3691
def _get_repo_format_to_test(self):
3695
def is_compatible(source, target):
3696
"""Be compatible with Knit2 source and Knit3 target"""
3697
# This is redundant with format.check_conversion_target(), however that
3698
# raises an exception, and we just want to say "False" as in we won't
3699
# support converting between these formats.
3700
if 'IDS:never' in debug.debug_flags:
3702
if source.supports_rich_root() and not target.supports_rich_root():
3704
if (source._format.supports_tree_reference
3705
and not target._format.supports_tree_reference):
3707
if target._fallback_repositories and target._format.supports_chks:
3708
# IDS doesn't know how to copy CHKs for the parent inventories it
3709
# adds to stacked repos.
3711
if 'IDS:always' in debug.debug_flags:
3713
# Only use this code path for local source and target. IDS does far
3714
# too much IO (both bandwidth and roundtrips) over a network.
3715
if not source.bzrdir.transport.base.startswith('file:///'):
3717
if not target.bzrdir.transport.base.startswith('file:///'):
3721
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3722
"""Get the best delta and base for this revision.
3724
:return: (basis_id, delta)
3726
possible_trees = [(parent_id, cache[parent_id])
3727
for parent_id in parent_ids
3728
if parent_id in cache]
3729
if len(possible_trees) == 0:
3730
# There either aren't any parents, or the parents aren't in the
3731
# cache, so just use the last converted tree
3732
possible_trees.append((basis_id, cache[basis_id]))
3734
for basis_id, basis_tree in possible_trees:
3735
delta = tree.inventory._make_delta(basis_tree.inventory)
3736
deltas.append((len(delta), basis_id, delta))
3738
return deltas[0][1:]
3740
def _fetch_batch(self, revision_ids, basis_id, cache):
3741
"""Fetch across a few revisions.
3743
:param revision_ids: The revisions to copy
3744
:param basis_id: The revision_id of a tree that must be in cache, used
3745
as a basis for delta when no other base is available
3746
:param cache: A cache of RevisionTrees that we can use.
3747
:return: The revision_id of the last converted tree. The RevisionTree
3748
for it will be in cache
3750
# Walk though all revisions; get inventory deltas, copy referenced
3751
# texts that delta references, insert the delta, revision and
3753
root_keys_to_create = set()
3756
pending_revisions = []
3757
parent_map = self.source.get_parent_map(revision_ids)
3758
for tree in self.source.revision_trees(revision_ids):
3759
current_revision_id = tree.get_revision_id()
3760
parent_ids = parent_map.get(current_revision_id, ())
3761
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3763
if self._converting_to_rich_root:
3764
self._revision_id_to_root_id[current_revision_id] = \
3766
# Find text entries that need to be copied
3767
for old_path, new_path, file_id, entry in delta:
3768
if new_path is not None:
3771
if not self.target.supports_rich_root():
3772
# The target doesn't support rich root, so we don't
3775
if self._converting_to_rich_root:
3776
# This can't be copied normally, we have to insert
3778
root_keys_to_create.add((file_id, entry.revision))
3780
text_keys.add((file_id, entry.revision))
3781
revision = self.source.get_revision(current_revision_id)
3782
pending_deltas.append((basis_id, delta,
3783
current_revision_id, revision.parent_ids))
3784
pending_revisions.append(revision)
3785
cache[current_revision_id] = tree
3786
basis_id = current_revision_id
3788
from_texts = self.source.texts
3789
to_texts = self.target.texts
3790
if root_keys_to_create:
3791
from bzrlib.fetch import _new_root_data_stream
3792
root_stream = _new_root_data_stream(
3793
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3795
to_texts.insert_record_stream(root_stream)
3796
to_texts.insert_record_stream(from_texts.get_record_stream(
3797
text_keys, self.target._format._fetch_order,
3798
not self.target._format._fetch_uses_deltas))
3799
# insert inventory deltas
3800
for delta in pending_deltas:
3801
self.target.add_inventory_by_delta(*delta)
3802
if self.target._fallback_repositories:
3803
# Make sure this stacked repository has all the parent inventories
3804
# for the new revisions that we are about to insert. We do this
3805
# before adding the revisions so that no revision is added until
3806
# all the inventories it may depend on are added.
3808
revision_ids = set()
3809
for revision in pending_revisions:
3810
revision_ids.add(revision.revision_id)
3811
parent_ids.update(revision.parent_ids)
3812
parent_ids.difference_update(revision_ids)
3813
parent_ids.discard(_mod_revision.NULL_REVISION)
3814
parent_map = self.source.get_parent_map(parent_ids)
3815
for parent_tree in self.source.revision_trees(parent_ids):
3816
basis_id, delta = self._get_delta_for_revision(tree, parent_ids, basis_id, cache)
3817
current_revision_id = parent_tree.get_revision_id()
3818
parents_parents = parent_map[current_revision_id]
3819
self.target.add_inventory_by_delta(
3820
basis_id, delta, current_revision_id, parents_parents)
3821
# insert signatures and revisions
3822
for revision in pending_revisions:
3824
signature = self.source.get_signature_text(
3825
revision.revision_id)
3826
self.target.add_signature_text(revision.revision_id,
3828
except errors.NoSuchRevision:
3830
self.target.add_revision(revision.revision_id, revision)
3833
def _fetch_all_revisions(self, revision_ids, pb):
3834
"""Fetch everything for the list of revisions.
3836
:param revision_ids: The list of revisions to fetch. Must be in
3838
:param pb: A ProgressTask
3841
basis_id, basis_tree = self._get_basis(revision_ids[0])
3843
cache = lru_cache.LRUCache(100)
3844
cache[basis_id] = basis_tree
3845
del basis_tree # We don't want to hang on to it here
3847
for offset in range(0, len(revision_ids), batch_size):
3848
self.target.start_write_group()
3850
pb.update('Transferring revisions', offset,
3852
batch = revision_ids[offset:offset+batch_size]
3853
basis_id = self._fetch_batch(batch, basis_id, cache)
3855
self.target.abort_write_group()
3858
hint = self.target.commit_write_group()
3861
if hints and self.target._format.pack_compresses:
3862
self.target.pack(hint=hints)
3863
pb.update('Transferring revisions', len(revision_ids),
3867
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3869
"""See InterRepository.fetch()."""
3870
if fetch_spec is not None:
3871
raise AssertionError("Not implemented yet...")
3872
if (not self.source.supports_rich_root()
3873
and self.target.supports_rich_root()):
3874
self._converting_to_rich_root = True
3875
self._revision_id_to_root_id = {}
3877
self._converting_to_rich_root = False
3878
revision_ids = self.target.search_missing_revision_ids(self.source,
3879
revision_id, find_ghosts=find_ghosts).get_keys()
3880
if not revision_ids:
3882
revision_ids = tsort.topo_sort(
3883
self.source.get_graph().get_parent_map(revision_ids))
3884
if not revision_ids:
3886
# Walk though all revisions; get inventory deltas, copy referenced
3887
# texts that delta references, insert the delta, revision and
3890
my_pb = ui.ui_factory.nested_progress_bar()
3893
symbol_versioning.warn(
3894
symbol_versioning.deprecated_in((1, 14, 0))
3895
% "pb parameter to fetch()")
3898
self._fetch_all_revisions(revision_ids, pb)
3900
if my_pb is not None:
3902
return len(revision_ids), 0
3904
def _get_basis(self, first_revision_id):
3905
"""Get a revision and tree which exists in the target.
3907
This assumes that first_revision_id is selected for transmission
3908
because all other ancestors are already present. If we can't find an
3909
ancestor we fall back to NULL_REVISION since we know that is safe.
3911
:return: (basis_id, basis_tree)
3913
first_rev = self.source.get_revision(first_revision_id)
3915
basis_id = first_rev.parent_ids[0]
3916
# only valid as a basis if the target has it
3917
self.target.get_revision(basis_id)
3918
# Try to get a basis tree - if its a ghost it will hit the
3919
# NoSuchRevision case.
3920
basis_tree = self.source.revision_tree(basis_id)
3921
except (IndexError, errors.NoSuchRevision):
3922
basis_id = _mod_revision.NULL_REVISION
3923
basis_tree = self.source.revision_tree(basis_id)
3924
return basis_id, basis_tree
3927
InterRepository.register_optimiser(InterDifferingSerializer)
3928
InterRepository.register_optimiser(InterSameDataRepository)
3929
InterRepository.register_optimiser(InterWeaveRepo)
3930
InterRepository.register_optimiser(InterKnitRepo)
3933
class CopyConverter(object):
3934
"""A repository conversion tool which just performs a copy of the content.
3936
This is slow but quite reliable.
3939
def __init__(self, target_format):
3940
"""Create a CopyConverter.
3942
:param target_format: The format the resulting repository should be.
3944
self.target_format = target_format
3946
def convert(self, repo, pb):
3947
"""Perform the conversion of to_convert, giving feedback via pb.
3949
:param to_convert: The disk object to convert.
3950
:param pb: a progress bar to use for progress information.
3955
# this is only useful with metadir layouts - separated repo content.
3956
# trigger an assertion if not such
3957
repo._format.get_format_string()
3958
self.repo_dir = repo.bzrdir
3959
self.step('Moving repository to repository.backup')
3960
self.repo_dir.transport.move('repository', 'repository.backup')
3961
backup_transport = self.repo_dir.transport.clone('repository.backup')
3962
repo._format.check_conversion_target(self.target_format)
3963
self.source_repo = repo._format.open(self.repo_dir,
3965
_override_transport=backup_transport)
3966
self.step('Creating new repository')
3967
converted = self.target_format.initialize(self.repo_dir,
3968
self.source_repo.is_shared())
3969
converted.lock_write()
3971
self.step('Copying content into repository.')
3972
self.source_repo.copy_content_into(converted)
3975
self.step('Deleting old repository content.')
3976
self.repo_dir.transport.delete_tree('repository.backup')
3977
self.pb.note('repository converted')
3979
def step(self, message):
3980
"""Update the pb by a step."""
3982
self.pb.update(message, self.count, self.total)
3994
def _unescaper(match, _map=_unescape_map):
3995
code = match.group(1)
3999
if not code.startswith('#'):
4001
return unichr(int(code[1:])).encode('utf8')
4007
def _unescape_xml(data):
4008
"""Unescape predefined XML entities in a string of data."""
4010
if _unescape_re is None:
4011
_unescape_re = re.compile('\&([^;]*);')
4012
return _unescape_re.sub(_unescaper, data)
4015
class _VersionedFileChecker(object):
4017
def __init__(self, repository, text_key_references=None, ancestors=None):
4018
self.repository = repository
4019
self.text_index = self.repository._generate_text_key_index(
4020
text_key_references=text_key_references, ancestors=ancestors)
4022
def calculate_file_version_parents(self, text_key):
4023
"""Calculate the correct parents for a file version according to
4026
parent_keys = self.text_index[text_key]
4027
if parent_keys == [_mod_revision.NULL_REVISION]:
4029
return tuple(parent_keys)
4031
def check_file_version_parents(self, texts, progress_bar=None):
4032
"""Check the parents stored in a versioned file are correct.
4034
It also detects file versions that are not referenced by their
4035
corresponding revision's inventory.
4037
:returns: A tuple of (wrong_parents, dangling_file_versions).
4038
wrong_parents is a dict mapping {revision_id: (stored_parents,
4039
correct_parents)} for each revision_id where the stored parents
4040
are not correct. dangling_file_versions is a set of (file_id,
4041
revision_id) tuples for versions that are present in this versioned
4042
file, but not used by the corresponding inventory.
4044
local_progress = None
4045
if progress_bar is None:
4046
local_progress = ui.ui_factory.nested_progress_bar()
4047
progress_bar = local_progress
4049
return self._check_file_version_parents(texts, progress_bar)
4052
local_progress.finished()
4054
def _check_file_version_parents(self, texts, progress_bar):
4055
"""See check_file_version_parents."""
4057
self.file_ids = set([file_id for file_id, _ in
4058
self.text_index.iterkeys()])
4059
# text keys is now grouped by file_id
4060
n_versions = len(self.text_index)
4061
progress_bar.update('loading text store', 0, n_versions)
4062
parent_map = self.repository.texts.get_parent_map(self.text_index)
4063
# On unlistable transports this could well be empty/error...
4064
text_keys = self.repository.texts.keys()
4065
unused_keys = frozenset(text_keys) - set(self.text_index)
4066
for num, key in enumerate(self.text_index.iterkeys()):
4067
progress_bar.update('checking text graph', num, n_versions)
4068
correct_parents = self.calculate_file_version_parents(key)
4070
knit_parents = parent_map[key]
4071
except errors.RevisionNotPresent:
4074
if correct_parents != knit_parents:
4075
wrong_parents[key] = (knit_parents, correct_parents)
4076
return wrong_parents, unused_keys
4079
def _old_get_graph(repository, revision_id):
4080
"""DO NOT USE. That is all. I'm serious."""
4081
graph = repository.get_graph()
4082
revision_graph = dict(((key, value) for key, value in
4083
graph.iter_ancestry([revision_id]) if value is not None))
4084
return _strip_NULL_ghosts(revision_graph)
4087
def _strip_NULL_ghosts(revision_graph):
4088
"""Also don't use this. more compatibility code for unmigrated clients."""
4089
# Filter ghosts, and null:
4090
if _mod_revision.NULL_REVISION in revision_graph:
4091
del revision_graph[_mod_revision.NULL_REVISION]
4092
for key, parents in revision_graph.items():
4093
revision_graph[key] = tuple(parent for parent in parents if parent
4095
return revision_graph
4098
class StreamSink(object):
4099
"""An object that can insert a stream into a repository.
4101
This interface handles the complexity of reserialising inventories and
4102
revisions from different formats, and allows unidirectional insertion into
4103
stacked repositories without looking for the missing basis parents
4107
def __init__(self, target_repo):
4108
self.target_repo = target_repo
4110
def insert_stream(self, stream, src_format, resume_tokens):
4111
"""Insert a stream's content into the target repository.
4113
:param src_format: a bzr repository format.
4115
:return: a list of resume tokens and an iterable of keys additional
4116
items required before the insertion can be completed.
4118
self.target_repo.lock_write()
4121
self.target_repo.resume_write_group(resume_tokens)
4124
self.target_repo.start_write_group()
4127
# locked_insert_stream performs a commit|suspend.
4128
return self._locked_insert_stream(stream, src_format, is_resume)
4130
self.target_repo.abort_write_group(suppress_errors=True)
4133
self.target_repo.unlock()
4135
def _locked_insert_stream(self, stream, src_format, is_resume):
4136
to_serializer = self.target_repo._format._serializer
4137
src_serializer = src_format._serializer
4139
if to_serializer == src_serializer:
4140
# If serializers match and the target is a pack repository, set the
4141
# write cache size on the new pack. This avoids poor performance
4142
# on transports where append is unbuffered (such as
4143
# RemoteTransport). This is safe to do because nothing should read
4144
# back from the target repository while a stream with matching
4145
# serialization is being inserted.
4146
# The exception is that a delta record from the source that should
4147
# be a fulltext may need to be expanded by the target (see
4148
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4149
# explicitly flush any buffered writes first in that rare case.
4151
new_pack = self.target_repo._pack_collection._new_pack
4152
except AttributeError:
4153
# Not a pack repository
4156
new_pack.set_write_cache_size(1024*1024)
4157
for substream_type, substream in stream:
4158
if 'stream' in debug.debug_flags:
4159
mutter('inserting substream: %s', substream_type)
4160
if substream_type == 'texts':
4161
self.target_repo.texts.insert_record_stream(substream)
4162
elif substream_type == 'inventories':
4163
if src_serializer == to_serializer:
4164
self.target_repo.inventories.insert_record_stream(
4167
self._extract_and_insert_inventories(
4168
substream, src_serializer)
4169
elif substream_type == 'inventory-deltas':
4170
self._extract_and_insert_inventory_deltas(
4171
substream, src_serializer)
4172
elif substream_type == 'chk_bytes':
4173
# XXX: This doesn't support conversions, as it assumes the
4174
# conversion was done in the fetch code.
4175
self.target_repo.chk_bytes.insert_record_stream(substream)
4176
elif substream_type == 'revisions':
4177
# This may fallback to extract-and-insert more often than
4178
# required if the serializers are different only in terms of
4180
if src_serializer == to_serializer:
4181
self.target_repo.revisions.insert_record_stream(
4184
self._extract_and_insert_revisions(substream,
4186
elif substream_type == 'signatures':
4187
self.target_repo.signatures.insert_record_stream(substream)
4189
raise AssertionError('kaboom! %s' % (substream_type,))
4190
# Done inserting data, and the missing_keys calculations will try to
4191
# read back from the inserted data, so flush the writes to the new pack
4192
# (if this is pack format).
4193
if new_pack is not None:
4194
new_pack._write_data('', flush=True)
4195
# Find all the new revisions (including ones from resume_tokens)
4196
missing_keys = self.target_repo.get_missing_parent_inventories(
4197
check_for_missing_texts=is_resume)
4199
for prefix, versioned_file in (
4200
('texts', self.target_repo.texts),
4201
('inventories', self.target_repo.inventories),
4202
('revisions', self.target_repo.revisions),
4203
('signatures', self.target_repo.signatures),
4204
('chk_bytes', self.target_repo.chk_bytes),
4206
if versioned_file is None:
4208
missing_keys.update((prefix,) + key for key in
4209
versioned_file.get_missing_compression_parent_keys())
4210
except NotImplementedError:
4211
# cannot even attempt suspending, and missing would have failed
4212
# during stream insertion.
4213
missing_keys = set()
4216
# suspend the write group and tell the caller what we is
4217
# missing. We know we can suspend or else we would not have
4218
# entered this code path. (All repositories that can handle
4219
# missing keys can handle suspending a write group).
4220
write_group_tokens = self.target_repo.suspend_write_group()
4221
return write_group_tokens, missing_keys
4222
hint = self.target_repo.commit_write_group()
4223
if (to_serializer != src_serializer and
4224
self.target_repo._format.pack_compresses):
4225
self.target_repo.pack(hint=hint)
4228
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4229
target_rich_root = self.target_repo._format.rich_root_data
4230
target_tree_refs = self.target_repo._format.supports_tree_reference
4231
for record in substream:
4232
# Insert the delta directly
4233
inventory_delta_bytes = record.get_bytes_as('fulltext')
4234
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4236
parse_result = deserialiser.parse_text_bytes(
4237
inventory_delta_bytes)
4238
except inventory_delta.IncompatibleInventoryDelta, err:
4239
trace.mutter("Incompatible delta: %s", err.msg)
4240
raise errors.IncompatibleRevision(self.target_repo._format)
4241
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4242
revision_id = new_id
4243
parents = [key[0] for key in record.parents]
4244
self.target_repo.add_inventory_by_delta(
4245
basis_id, inv_delta, revision_id, parents)
4247
def _extract_and_insert_inventories(self, substream, serializer,
4249
"""Generate a new inventory versionedfile in target, converting data.
4251
The inventory is retrieved from the source, (deserializing it), and
4252
stored in the target (reserializing it in a different format).
4254
target_rich_root = self.target_repo._format.rich_root_data
4255
target_tree_refs = self.target_repo._format.supports_tree_reference
4256
for record in substream:
4257
# It's not a delta, so it must be a fulltext in the source
4258
# serializer's format.
4259
bytes = record.get_bytes_as('fulltext')
4260
revision_id = record.key[0]
4261
inv = serializer.read_inventory_from_string(bytes, revision_id)
4262
parents = [key[0] for key in record.parents]
4263
self.target_repo.add_inventory(revision_id, inv, parents)
4264
# No need to keep holding this full inv in memory when the rest of
4265
# the substream is likely to be all deltas.
4268
def _extract_and_insert_revisions(self, substream, serializer):
4269
for record in substream:
4270
bytes = record.get_bytes_as('fulltext')
4271
revision_id = record.key[0]
4272
rev = serializer.read_revision_from_string(bytes)
4273
if rev.revision_id != revision_id:
4274
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4275
self.target_repo.add_revision(revision_id, rev)
4278
if self.target_repo._format._fetch_reconcile:
4279
self.target_repo.reconcile()
4282
class StreamSource(object):
4283
"""A source of a stream for fetching between repositories."""
4285
def __init__(self, from_repository, to_format):
4286
"""Create a StreamSource streaming from from_repository."""
4287
self.from_repository = from_repository
4288
self.to_format = to_format
4290
def delta_on_metadata(self):
4291
"""Return True if delta's are permitted on metadata streams.
4293
That is on revisions and signatures.
4295
src_serializer = self.from_repository._format._serializer
4296
target_serializer = self.to_format._serializer
4297
return (self.to_format._fetch_uses_deltas and
4298
src_serializer == target_serializer)
4300
def _fetch_revision_texts(self, revs):
4301
# fetch signatures first and then the revision texts
4302
# may need to be a InterRevisionStore call here.
4303
from_sf = self.from_repository.signatures
4304
# A missing signature is just skipped.
4305
keys = [(rev_id,) for rev_id in revs]
4306
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4308
self.to_format._fetch_order,
4309
not self.to_format._fetch_uses_deltas))
4310
# If a revision has a delta, this is actually expanded inside the
4311
# insert_record_stream code now, which is an alternate fix for
4313
from_rf = self.from_repository.revisions
4314
revisions = from_rf.get_record_stream(
4316
self.to_format._fetch_order,
4317
not self.delta_on_metadata())
4318
return [('signatures', signatures), ('revisions', revisions)]
4320
def _generate_root_texts(self, revs):
4321
"""This will be called by get_stream between fetching weave texts and
4322
fetching the inventory weave.
4324
if self._rich_root_upgrade():
4326
return bzrlib.fetch.Inter1and2Helper(
4327
self.from_repository).generate_root_texts(revs)
4331
def get_stream(self, search):
4333
revs = search.get_keys()
4334
graph = self.from_repository.get_graph()
4335
revs = list(graph.iter_topo_order(revs))
4336
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4338
for knit_kind, file_id, revisions in data_to_fetch:
4339
if knit_kind != phase:
4341
# Make a new progress bar for this phase
4342
if knit_kind == "file":
4343
# Accumulate file texts
4344
text_keys.extend([(file_id, revision) for revision in
4346
elif knit_kind == "inventory":
4347
# Now copy the file texts.
4348
from_texts = self.from_repository.texts
4349
yield ('texts', from_texts.get_record_stream(
4350
text_keys, self.to_format._fetch_order,
4351
not self.to_format._fetch_uses_deltas))
4352
# Cause an error if a text occurs after we have done the
4355
# Before we process the inventory we generate the root
4356
# texts (if necessary) so that the inventories references
4358
for _ in self._generate_root_texts(revs):
4360
# we fetch only the referenced inventories because we do not
4361
# know for unselected inventories whether all their required
4362
# texts are present in the other repository - it could be
4364
for info in self._get_inventory_stream(revs):
4366
elif knit_kind == "signatures":
4367
# Nothing to do here; this will be taken care of when
4368
# _fetch_revision_texts happens.
4370
elif knit_kind == "revisions":
4371
for record in self._fetch_revision_texts(revs):
4374
raise AssertionError("Unknown knit kind %r" % knit_kind)
4376
def get_stream_for_missing_keys(self, missing_keys):
4377
# missing keys can only occur when we are byte copying and not
4378
# translating (because translation means we don't send
4379
# unreconstructable deltas ever).
4381
keys['texts'] = set()
4382
keys['revisions'] = set()
4383
keys['inventories'] = set()
4384
keys['chk_bytes'] = set()
4385
keys['signatures'] = set()
4386
for key in missing_keys:
4387
keys[key[0]].add(key[1:])
4388
if len(keys['revisions']):
4389
# If we allowed copying revisions at this point, we could end up
4390
# copying a revision without copying its required texts: a
4391
# violation of the requirements for repository integrity.
4392
raise AssertionError(
4393
'cannot copy revisions to fill in missing deltas %s' % (
4394
keys['revisions'],))
4395
for substream_kind, keys in keys.iteritems():
4396
vf = getattr(self.from_repository, substream_kind)
4397
if vf is None and keys:
4398
raise AssertionError(
4399
"cannot fill in keys for a versioned file we don't"
4400
" have: %s needs %s" % (substream_kind, keys))
4402
# No need to stream something we don't have
4404
if substream_kind == 'inventories':
4405
# Some missing keys are genuinely ghosts, filter those out.
4406
present = self.from_repository.inventories.get_parent_map(keys)
4407
revs = [key[0] for key in present]
4408
# As with the original stream, we may need to generate root
4409
# texts for the inventories we're about to stream.
4410
for _ in self._generate_root_texts(revs):
4412
# Get the inventory stream more-or-less as we do for the
4413
# original stream; there's no reason to assume that records
4414
# direct from the source will be suitable for the sink. (Think
4415
# e.g. 2a -> 1.9-rich-root).
4416
for info in self._get_inventory_stream(revs, missing=True):
4420
# Ask for full texts always so that we don't need more round trips
4421
# after this stream.
4422
# Some of the missing keys are genuinely ghosts, so filter absent
4423
# records. The Sink is responsible for doing another check to
4424
# ensure that ghosts don't introduce missing data for future
4426
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4427
self.to_format._fetch_order, True))
4428
yield substream_kind, stream
4430
def inventory_fetch_order(self):
4431
if self._rich_root_upgrade():
4432
return 'topological'
4434
return self.to_format._fetch_order
4436
def _rich_root_upgrade(self):
4437
return (not self.from_repository._format.rich_root_data and
4438
self.to_format.rich_root_data)
4440
def _get_inventory_stream(self, revision_ids, missing=False):
4441
from_format = self.from_repository._format
4442
if (from_format.supports_chks and self.to_format.supports_chks and
4443
from_format.network_name() == self.to_format.network_name()):
4444
raise AssertionError(
4445
"this case should be handled by GroupCHKStreamSource")
4446
elif 'forceinvdeltas' in debug.debug_flags:
4447
return self._get_convertable_inventory_stream(revision_ids,
4448
delta_versus_null=missing)
4449
elif from_format.network_name() == self.to_format.network_name():
4451
return self._get_simple_inventory_stream(revision_ids,
4453
elif (not from_format.supports_chks and not self.to_format.supports_chks
4454
and from_format._serializer == self.to_format._serializer):
4455
# Essentially the same format.
4456
return self._get_simple_inventory_stream(revision_ids,
4459
# Any time we switch serializations, we want to use an
4460
# inventory-delta based approach.
4461
return self._get_convertable_inventory_stream(revision_ids,
4462
delta_versus_null=missing)
4464
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4465
# NB: This currently reopens the inventory weave in source;
4466
# using a single stream interface instead would avoid this.
4467
from_weave = self.from_repository.inventories
4469
delta_closure = True
4471
delta_closure = not self.delta_on_metadata()
4472
yield ('inventories', from_weave.get_record_stream(
4473
[(rev_id,) for rev_id in revision_ids],
4474
self.inventory_fetch_order(), delta_closure))
4476
def _get_convertable_inventory_stream(self, revision_ids,
4477
delta_versus_null=False):
4478
# The source is using CHKs, but the target either doesn't or it has a
4479
# different serializer. The StreamSink code expects to be able to
4480
# convert on the target, so we need to put bytes-on-the-wire that can
4481
# be converted. That means inventory deltas (if the remote is <1.18,
4482
# RemoteStreamSink will fallback to VFS to insert the deltas).
4483
yield ('inventory-deltas',
4484
self._stream_invs_as_deltas(revision_ids,
4485
delta_versus_null=delta_versus_null))
4487
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4488
"""Return a stream of inventory-deltas for the given rev ids.
4490
:param revision_ids: The list of inventories to transmit
4491
:param delta_versus_null: Don't try to find a minimal delta for this
4492
entry, instead compute the delta versus the NULL_REVISION. This
4493
effectively streams a complete inventory. Used for stuff like
4494
filling in missing parents, etc.
4496
from_repo = self.from_repository
4497
revision_keys = [(rev_id,) for rev_id in revision_ids]
4498
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4499
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4501
inventories = self.from_repository.iter_inventories(
4502
revision_ids, 'topological')
4503
format = from_repo._format
4504
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4505
inventory_cache = lru_cache.LRUCache(50)
4506
null_inventory = from_repo.revision_tree(
4507
_mod_revision.NULL_REVISION).inventory
4508
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4509
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4510
# repo back into a non-rich-root repo ought to be allowed)
4511
serializer = inventory_delta.InventoryDeltaSerializer(
4512
versioned_root=format.rich_root_data,
4513
tree_references=format.supports_tree_reference)
4514
for inv in inventories:
4515
key = (inv.revision_id,)
4516
parent_keys = parent_map.get(key, ())
4518
if not delta_versus_null and parent_keys:
4519
# The caller did not ask for complete inventories and we have
4520
# some parents that we can delta against. Make a delta against
4521
# each parent so that we can find the smallest.
4522
parent_ids = [parent_key[0] for parent_key in parent_keys]
4523
for parent_id in parent_ids:
4524
if parent_id not in invs_sent_so_far:
4525
# We don't know that the remote side has this basis, so
4528
if parent_id == _mod_revision.NULL_REVISION:
4529
parent_inv = null_inventory
4531
parent_inv = inventory_cache.get(parent_id, None)
4532
if parent_inv is None:
4533
parent_inv = from_repo.get_inventory(parent_id)
4534
candidate_delta = inv._make_delta(parent_inv)
4535
if (delta is None or
4536
len(delta) > len(candidate_delta)):
4537
delta = candidate_delta
4538
basis_id = parent_id
4540
# Either none of the parents ended up being suitable, or we
4541
# were asked to delta against NULL
4542
basis_id = _mod_revision.NULL_REVISION
4543
delta = inv._make_delta(null_inventory)
4544
invs_sent_so_far.add(inv.revision_id)
4545
inventory_cache[inv.revision_id] = inv
4546
delta_serialized = ''.join(
4547
serializer.delta_to_lines(basis_id, key[-1], delta))
4548
yield versionedfile.FulltextContentFactory(
4549
key, parent_keys, None, delta_serialized)
4552
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4553
stop_revision=None):
4554
"""Extend the partial history to include a given index
4556
If a stop_index is supplied, stop when that index has been reached.
4557
If a stop_revision is supplied, stop when that revision is
4558
encountered. Otherwise, stop when the beginning of history is
4561
:param stop_index: The index which should be present. When it is
4562
present, history extension will stop.
4563
:param stop_revision: The revision id which should be present. When
4564
it is encountered, history extension will stop.
4566
start_revision = partial_history_cache[-1]
4567
iterator = repo.iter_reverse_revision_history(start_revision)
4569
#skip the last revision in the list
4572
if (stop_index is not None and
4573
len(partial_history_cache) > stop_index):
4575
if partial_history_cache[-1] == stop_revision:
4577
revision_id = iterator.next()
4578
partial_history_cache.append(revision_id)
4579
except StopIteration:
added
removed
bzrlib/repository.py
