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
if self._fallback_repositories:
1709
raise errors.BzrError("Cannot commit from a lightweight checkout "
1710
"to a stacked branch. See "
1711
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1712
result = self._commit_builder_class(self, parents, config,
1713
timestamp, timezone, committer, revprops, revision_id)
1714
self.start_write_group()
1718
if (self.control_files._lock_count == 1 and
1719
self.control_files._lock_mode == 'w'):
1720
if self._write_group is not None:
1721
self.abort_write_group()
1722
self.control_files.unlock()
1723
raise errors.BzrError(
1724
'Must end write groups before releasing write locks.')
1725
self.control_files.unlock()
1726
if self.control_files._lock_count == 0:
1727
self._inventory_entry_cache.clear()
1728
for repo in self._fallback_repositories:
1732
def clone(self, a_bzrdir, revision_id=None):
1733
"""Clone this repository into a_bzrdir using the current format.
1735
Currently no check is made that the format of this repository and
1736
the bzrdir format are compatible. FIXME RBC 20060201.
1738
:return: The newly created destination repository.
1740
# TODO: deprecate after 0.16; cloning this with all its settings is
1741
# probably not very useful -- mbp 20070423
1742
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1743
self.copy_content_into(dest_repo, revision_id)
1746
def start_write_group(self):
1747
"""Start a write group in the repository.
1749
Write groups are used by repositories which do not have a 1:1 mapping
1750
between file ids and backend store to manage the insertion of data from
1751
both fetch and commit operations.
1753
A write lock is required around the start_write_group/commit_write_group
1754
for the support of lock-requiring repository formats.
1756
One can only insert data into a repository inside a write group.
1760
if not self.is_write_locked():
1761
raise errors.NotWriteLocked(self)
1762
if self._write_group:
1763
raise errors.BzrError('already in a write group')
1764
self._start_write_group()
1765
# so we can detect unlock/relock - the write group is now entered.
1766
self._write_group = self.get_transaction()
1768
def _start_write_group(self):
1769
"""Template method for per-repository write group startup.
1771
This is called before the write group is considered to be
1776
def sprout(self, to_bzrdir, revision_id=None):
1777
"""Create a descendent repository for new development.
1779
Unlike clone, this does not copy the settings of the repository.
1781
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1782
dest_repo.fetch(self, revision_id=revision_id)
1785
def _create_sprouting_repo(self, a_bzrdir, shared):
1786
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1787
# use target default format.
1788
dest_repo = a_bzrdir.create_repository()
1790
# Most control formats need the repository to be specifically
1791
# created, but on some old all-in-one formats it's not needed
1793
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1794
except errors.UninitializableFormat:
1795
dest_repo = a_bzrdir.open_repository()
1798
def _get_sink(self):
1799
"""Return a sink for streaming into this repository."""
1800
return StreamSink(self)
1802
def _get_source(self, to_format):
1803
"""Return a source for streaming from this repository."""
1804
return StreamSource(self, to_format)
1807
def has_revision(self, revision_id):
1808
"""True if this repository has a copy of the revision."""
1809
return revision_id in self.has_revisions((revision_id,))
1812
def has_revisions(self, revision_ids):
1813
"""Probe to find out the presence of multiple revisions.
1815
:param revision_ids: An iterable of revision_ids.
1816
:return: A set of the revision_ids that were present.
1818
parent_map = self.revisions.get_parent_map(
1819
[(rev_id,) for rev_id in revision_ids])
1821
if _mod_revision.NULL_REVISION in revision_ids:
1822
result.add(_mod_revision.NULL_REVISION)
1823
result.update([key[0] for key in parent_map])
1827
def get_revision(self, revision_id):
1828
"""Return the Revision object for a named revision."""
1829
return self.get_revisions([revision_id])[0]
1832
def get_revision_reconcile(self, revision_id):
1833
"""'reconcile' helper routine that allows access to a revision always.
1835
This variant of get_revision does not cross check the weave graph
1836
against the revision one as get_revision does: but it should only
1837
be used by reconcile, or reconcile-alike commands that are correcting
1838
or testing the revision graph.
1840
return self._get_revisions([revision_id])[0]
1843
def get_revisions(self, revision_ids):
1844
"""Get many revisions at once.
1846
Repositories that need to check data on every revision read should
1847
subclass this method.
1849
return self._get_revisions(revision_ids)
1852
def _get_revisions(self, revision_ids):
1853
"""Core work logic to get many revisions without sanity checks."""
1855
for revid, rev in self._iter_revisions(revision_ids):
1857
raise errors.NoSuchRevision(self, revid)
1859
return [revs[revid] for revid in revision_ids]
1861
def _iter_revisions(self, revision_ids):
1862
"""Iterate over revision objects.
1864
:param revision_ids: An iterable of revisions to examine. None may be
1865
passed to request all revisions known to the repository. Note that
1866
not all repositories can find unreferenced revisions; for those
1867
repositories only referenced ones will be returned.
1868
:return: An iterator of (revid, revision) tuples. Absent revisions (
1869
those asked for but not available) are returned as (revid, None).
1871
if revision_ids is None:
1872
revision_ids = self.all_revision_ids()
1874
for rev_id in revision_ids:
1875
if not rev_id or not isinstance(rev_id, basestring):
1876
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1877
keys = [(key,) for key in revision_ids]
1878
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1879
for record in stream:
1880
revid = record.key[0]
1881
if record.storage_kind == 'absent':
1884
text = record.get_bytes_as('fulltext')
1885
rev = self._serializer.read_revision_from_string(text)
1889
def get_revision_xml(self, revision_id):
1890
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1891
# would have already do it.
1892
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1893
# TODO: this can't just be replaced by:
1894
# return self._serializer.write_revision_to_string(
1895
# self.get_revision(revision_id))
1896
# as cStringIO preservers the encoding unlike write_revision_to_string
1897
# or some other call down the path.
1898
rev = self.get_revision(revision_id)
1899
rev_tmp = cStringIO.StringIO()
1900
# the current serializer..
1901
self._serializer.write_revision(rev, rev_tmp)
1903
return rev_tmp.getvalue()
1905
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1906
"""Produce a generator of revision deltas.
1908
Note that the input is a sequence of REVISIONS, not revision_ids.
1909
Trees will be held in memory until the generator exits.
1910
Each delta is relative to the revision's lefthand predecessor.
1912
:param specific_fileids: if not None, the result is filtered
1913
so that only those file-ids, their parents and their
1914
children are included.
1916
# Get the revision-ids of interest
1917
required_trees = set()
1918
for revision in revisions:
1919
required_trees.add(revision.revision_id)
1920
required_trees.update(revision.parent_ids[:1])
1922
# Get the matching filtered trees. Note that it's more
1923
# efficient to pass filtered trees to changes_from() rather
1924
# than doing the filtering afterwards. changes_from() could
1925
# arguably do the filtering itself but it's path-based, not
1926
# file-id based, so filtering before or afterwards is
1928
if specific_fileids is None:
1929
trees = dict((t.get_revision_id(), t) for
1930
t in self.revision_trees(required_trees))
1932
trees = dict((t.get_revision_id(), t) for
1933
t in self._filtered_revision_trees(required_trees,
1936
# Calculate the deltas
1937
for revision in revisions:
1938
if not revision.parent_ids:
1939
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1941
old_tree = trees[revision.parent_ids[0]]
1942
yield trees[revision.revision_id].changes_from(old_tree)
1945
def get_revision_delta(self, revision_id, specific_fileids=None):
1946
"""Return the delta for one revision.
1948
The delta is relative to the left-hand predecessor of the
1951
:param specific_fileids: if not None, the result is filtered
1952
so that only those file-ids, their parents and their
1953
children are included.
1955
r = self.get_revision(revision_id)
1956
return list(self.get_deltas_for_revisions([r],
1957
specific_fileids=specific_fileids))[0]
1960
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1961
signature = gpg_strategy.sign(plaintext)
1962
self.add_signature_text(revision_id, signature)
1965
def add_signature_text(self, revision_id, signature):
1966
self.signatures.add_lines((revision_id,), (),
1967
osutils.split_lines(signature))
1969
def find_text_key_references(self):
1970
"""Find the text key references within the repository.
1972
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1973
to whether they were referred to by the inventory of the
1974
revision_id that they contain. The inventory texts from all present
1975
revision ids are assessed to generate this report.
1977
revision_keys = self.revisions.keys()
1978
w = self.inventories
1979
pb = ui.ui_factory.nested_progress_bar()
1981
return self._find_text_key_references_from_xml_inventory_lines(
1982
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1986
def _find_text_key_references_from_xml_inventory_lines(self,
1988
"""Core routine for extracting references to texts from inventories.
1990
This performs the translation of xml lines to revision ids.
1992
:param line_iterator: An iterator of lines, origin_version_id
1993
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1994
to whether they were referred to by the inventory of the
1995
revision_id that they contain. Note that if that revision_id was
1996
not part of the line_iterator's output then False will be given -
1997
even though it may actually refer to that key.
1999
if not self._serializer.support_altered_by_hack:
2000
raise AssertionError(
2001
"_find_text_key_references_from_xml_inventory_lines only "
2002
"supported for branches which store inventory as unnested xml"
2003
", not on %r" % self)
2006
# this code needs to read every new line in every inventory for the
2007
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2008
# not present in one of those inventories is unnecessary but not
2009
# harmful because we are filtering by the revision id marker in the
2010
# inventory lines : we only select file ids altered in one of those
2011
# revisions. We don't need to see all lines in the inventory because
2012
# only those added in an inventory in rev X can contain a revision=X
2014
unescape_revid_cache = {}
2015
unescape_fileid_cache = {}
2017
# jam 20061218 In a big fetch, this handles hundreds of thousands
2018
# of lines, so it has had a lot of inlining and optimizing done.
2019
# Sorry that it is a little bit messy.
2020
# Move several functions to be local variables, since this is a long
2022
search = self._file_ids_altered_regex.search
2023
unescape = _unescape_xml
2024
setdefault = result.setdefault
2025
for line, line_key in line_iterator:
2026
match = search(line)
2029
# One call to match.group() returning multiple items is quite a
2030
# bit faster than 2 calls to match.group() each returning 1
2031
file_id, revision_id = match.group('file_id', 'revision_id')
2033
# Inlining the cache lookups helps a lot when you make 170,000
2034
# lines and 350k ids, versus 8.4 unique ids.
2035
# Using a cache helps in 2 ways:
2036
# 1) Avoids unnecessary decoding calls
2037
# 2) Re-uses cached strings, which helps in future set and
2039
# (2) is enough that removing encoding entirely along with
2040
# the cache (so we are using plain strings) results in no
2041
# performance improvement.
2043
revision_id = unescape_revid_cache[revision_id]
2045
unescaped = unescape(revision_id)
2046
unescape_revid_cache[revision_id] = unescaped
2047
revision_id = unescaped
2049
# Note that unconditionally unescaping means that we deserialise
2050
# every fileid, which for general 'pull' is not great, but we don't
2051
# really want to have some many fulltexts that this matters anyway.
2054
file_id = unescape_fileid_cache[file_id]
2056
unescaped = unescape(file_id)
2057
unescape_fileid_cache[file_id] = unescaped
2060
key = (file_id, revision_id)
2061
setdefault(key, False)
2062
if revision_id == line_key[-1]:
2066
def _inventory_xml_lines_for_keys(self, keys):
2067
"""Get a line iterator of the sort needed for findind references.
2069
Not relevant for non-xml inventory repositories.
2071
Ghosts in revision_keys are ignored.
2073
:param revision_keys: The revision keys for the inventories to inspect.
2074
:return: An iterator over (inventory line, revid) for the fulltexts of
2075
all of the xml inventories specified by revision_keys.
2077
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2078
for record in stream:
2079
if record.storage_kind != 'absent':
2080
chunks = record.get_bytes_as('chunked')
2081
revid = record.key[-1]
2082
lines = osutils.chunks_to_lines(chunks)
2086
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2088
"""Helper routine for fileids_altered_by_revision_ids.
2090
This performs the translation of xml lines to revision ids.
2092
:param line_iterator: An iterator of lines, origin_version_id
2093
:param revision_keys: The revision ids to filter for. This should be a
2094
set or other type which supports efficient __contains__ lookups, as
2095
the revision key from each parsed line will be looked up in the
2096
revision_keys filter.
2097
:return: a dictionary mapping altered file-ids to an iterable of
2098
revision_ids. Each altered file-ids has the exact revision_ids that
2099
altered it listed explicitly.
2101
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2102
line_iterator).iterkeys())
2103
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2104
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2105
self._inventory_xml_lines_for_keys(parent_keys)))
2106
new_keys = seen - parent_seen
2108
setdefault = result.setdefault
2109
for key in new_keys:
2110
setdefault(key[0], set()).add(key[-1])
2113
def _find_parent_ids_of_revisions(self, revision_ids):
2114
"""Find all parent ids that are mentioned in the revision graph.
2116
:return: set of revisions that are parents of revision_ids which are
2117
not part of revision_ids themselves
2119
parent_map = self.get_parent_map(revision_ids)
2121
map(parent_ids.update, parent_map.itervalues())
2122
parent_ids.difference_update(revision_ids)
2123
parent_ids.discard(_mod_revision.NULL_REVISION)
2126
def _find_parent_keys_of_revisions(self, revision_keys):
2127
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2129
:param revision_keys: An iterable of revision_keys.
2130
:return: The parents of all revision_keys that are not already in
2133
parent_map = self.revisions.get_parent_map(revision_keys)
2135
map(parent_keys.update, parent_map.itervalues())
2136
parent_keys.difference_update(revision_keys)
2137
parent_keys.discard(_mod_revision.NULL_REVISION)
2140
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2141
"""Find the file ids and versions affected by revisions.
2143
:param revisions: an iterable containing revision ids.
2144
:param _inv_weave: The inventory weave from this repository or None.
2145
If None, the inventory weave will be opened automatically.
2146
:return: a dictionary mapping altered file-ids to an iterable of
2147
revision_ids. Each altered file-ids has the exact revision_ids that
2148
altered it listed explicitly.
2150
selected_keys = set((revid,) for revid in revision_ids)
2151
w = _inv_weave or self.inventories
2152
pb = ui.ui_factory.nested_progress_bar()
2154
return self._find_file_ids_from_xml_inventory_lines(
2155
w.iter_lines_added_or_present_in_keys(
2156
selected_keys, pb=pb),
2161
def iter_files_bytes(self, desired_files):
2162
"""Iterate through file versions.
2164
Files will not necessarily be returned in the order they occur in
2165
desired_files. No specific order is guaranteed.
2167
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2168
value supplied by the caller as part of desired_files. It should
2169
uniquely identify the file version in the caller's context. (Examples:
2170
an index number or a TreeTransform trans_id.)
2172
bytes_iterator is an iterable of bytestrings for the file. The
2173
kind of iterable and length of the bytestrings are unspecified, but for
2174
this implementation, it is a list of bytes produced by
2175
VersionedFile.get_record_stream().
2177
:param desired_files: a list of (file_id, revision_id, identifier)
2181
for file_id, revision_id, callable_data in desired_files:
2182
text_keys[(file_id, revision_id)] = callable_data
2183
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2184
if record.storage_kind == 'absent':
2185
raise errors.RevisionNotPresent(record.key, self)
2186
yield text_keys[record.key], record.get_bytes_as('chunked')
2188
def _generate_text_key_index(self, text_key_references=None,
2190
"""Generate a new text key index for the repository.
2192
This is an expensive function that will take considerable time to run.
2194
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2195
list of parents, also text keys. When a given key has no parents,
2196
the parents list will be [NULL_REVISION].
2198
# All revisions, to find inventory parents.
2199
if ancestors is None:
2200
graph = self.get_graph()
2201
ancestors = graph.get_parent_map(self.all_revision_ids())
2202
if text_key_references is None:
2203
text_key_references = self.find_text_key_references()
2204
pb = ui.ui_factory.nested_progress_bar()
2206
return self._do_generate_text_key_index(ancestors,
2207
text_key_references, pb)
2211
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2212
"""Helper for _generate_text_key_index to avoid deep nesting."""
2213
revision_order = tsort.topo_sort(ancestors)
2214
invalid_keys = set()
2216
for revision_id in revision_order:
2217
revision_keys[revision_id] = set()
2218
text_count = len(text_key_references)
2219
# a cache of the text keys to allow reuse; costs a dict of all the
2220
# keys, but saves a 2-tuple for every child of a given key.
2222
for text_key, valid in text_key_references.iteritems():
2224
invalid_keys.add(text_key)
2226
revision_keys[text_key[1]].add(text_key)
2227
text_key_cache[text_key] = text_key
2228
del text_key_references
2230
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2231
NULL_REVISION = _mod_revision.NULL_REVISION
2232
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2233
# too small for large or very branchy trees. However, for 55K path
2234
# trees, it would be easy to use too much memory trivially. Ideally we
2235
# could gauge this by looking at available real memory etc, but this is
2236
# always a tricky proposition.
2237
inventory_cache = lru_cache.LRUCache(10)
2238
batch_size = 10 # should be ~150MB on a 55K path tree
2239
batch_count = len(revision_order) / batch_size + 1
2241
pb.update("Calculating text parents", processed_texts, text_count)
2242
for offset in xrange(batch_count):
2243
to_query = revision_order[offset * batch_size:(offset + 1) *
2247
for revision_id in to_query:
2248
parent_ids = ancestors[revision_id]
2249
for text_key in revision_keys[revision_id]:
2250
pb.update("Calculating text parents", processed_texts)
2251
processed_texts += 1
2252
candidate_parents = []
2253
for parent_id in parent_ids:
2254
parent_text_key = (text_key[0], parent_id)
2256
check_parent = parent_text_key not in \
2257
revision_keys[parent_id]
2259
# the parent parent_id is a ghost:
2260
check_parent = False
2261
# truncate the derived graph against this ghost.
2262
parent_text_key = None
2264
# look at the parent commit details inventories to
2265
# determine possible candidates in the per file graph.
2268
inv = inventory_cache[parent_id]
2270
inv = self.revision_tree(parent_id).inventory
2271
inventory_cache[parent_id] = inv
2273
parent_entry = inv[text_key[0]]
2274
except (KeyError, errors.NoSuchId):
2276
if parent_entry is not None:
2278
text_key[0], parent_entry.revision)
2280
parent_text_key = None
2281
if parent_text_key is not None:
2282
candidate_parents.append(
2283
text_key_cache[parent_text_key])
2284
parent_heads = text_graph.heads(candidate_parents)
2285
new_parents = list(parent_heads)
2286
new_parents.sort(key=lambda x:candidate_parents.index(x))
2287
if new_parents == []:
2288
new_parents = [NULL_REVISION]
2289
text_index[text_key] = new_parents
2291
for text_key in invalid_keys:
2292
text_index[text_key] = [NULL_REVISION]
2295
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2296
"""Get an iterable listing the keys of all the data introduced by a set
2299
The keys will be ordered so that the corresponding items can be safely
2300
fetched and inserted in that order.
2302
:returns: An iterable producing tuples of (knit-kind, file-id,
2303
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2304
'revisions'. file-id is None unless knit-kind is 'file'.
2306
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2309
for result in self._find_non_file_keys_to_fetch(revision_ids):
2312
def _find_file_keys_to_fetch(self, revision_ids, pb):
2313
# XXX: it's a bit weird to control the inventory weave caching in this
2314
# generator. Ideally the caching would be done in fetch.py I think. Or
2315
# maybe this generator should explicitly have the contract that it
2316
# should not be iterated until the previously yielded item has been
2318
inv_w = self.inventories
2320
# file ids that changed
2321
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2323
num_file_ids = len(file_ids)
2324
for file_id, altered_versions in file_ids.iteritems():
2326
pb.update("fetch texts", count, num_file_ids)
2328
yield ("file", file_id, altered_versions)
2330
def _find_non_file_keys_to_fetch(self, revision_ids):
2332
yield ("inventory", None, revision_ids)
2335
# XXX: Note ATM no callers actually pay attention to this return
2336
# instead they just use the list of revision ids and ignore
2337
# missing sigs. Consider removing this work entirely
2338
revisions_with_signatures = set(self.signatures.get_parent_map(
2339
[(r,) for r in revision_ids]))
2340
revisions_with_signatures = set(
2341
[r for (r,) in revisions_with_signatures])
2342
revisions_with_signatures.intersection_update(revision_ids)
2343
yield ("signatures", None, revisions_with_signatures)
2346
yield ("revisions", None, revision_ids)
2349
def get_inventory(self, revision_id):
2350
"""Get Inventory object by revision id."""
2351
return self.iter_inventories([revision_id]).next()
2353
def iter_inventories(self, revision_ids, ordering=None):
2354
"""Get many inventories by revision_ids.
2356
This will buffer some or all of the texts used in constructing the
2357
inventories in memory, but will only parse a single inventory at a
2360
:param revision_ids: The expected revision ids of the inventories.
2361
:param ordering: optional ordering, e.g. 'topological'. If not
2362
specified, the order of revision_ids will be preserved (by
2363
buffering if necessary).
2364
:return: An iterator of inventories.
2366
if ((None in revision_ids)
2367
or (_mod_revision.NULL_REVISION in revision_ids)):
2368
raise ValueError('cannot get null revision inventory')
2369
return self._iter_inventories(revision_ids, ordering)
2371
def _iter_inventories(self, revision_ids, ordering):
2372
"""single-document based inventory iteration."""
2373
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2374
for text, revision_id in inv_xmls:
2375
yield self.deserialise_inventory(revision_id, text)
2377
def _iter_inventory_xmls(self, revision_ids, ordering):
2378
if ordering is None:
2379
order_as_requested = True
2380
ordering = 'unordered'
2382
order_as_requested = False
2383
keys = [(revision_id,) for revision_id in revision_ids]
2386
if order_as_requested:
2387
key_iter = iter(keys)
2388
next_key = key_iter.next()
2389
stream = self.inventories.get_record_stream(keys, ordering, True)
2391
for record in stream:
2392
if record.storage_kind != 'absent':
2393
chunks = record.get_bytes_as('chunked')
2394
if order_as_requested:
2395
text_chunks[record.key] = chunks
2397
yield ''.join(chunks), record.key[-1]
2399
raise errors.NoSuchRevision(self, record.key)
2400
if order_as_requested:
2401
# Yield as many results as we can while preserving order.
2402
while next_key in text_chunks:
2403
chunks = text_chunks.pop(next_key)
2404
yield ''.join(chunks), next_key[-1]
2406
next_key = key_iter.next()
2407
except StopIteration:
2408
# We still want to fully consume the get_record_stream,
2409
# just in case it is not actually finished at this point
2413
def deserialise_inventory(self, revision_id, xml):
2414
"""Transform the xml into an inventory object.
2416
:param revision_id: The expected revision id of the inventory.
2417
:param xml: A serialised inventory.
2419
result = self._serializer.read_inventory_from_string(xml, revision_id,
2420
entry_cache=self._inventory_entry_cache)
2421
if result.revision_id != revision_id:
2422
raise AssertionError('revision id mismatch %s != %s' % (
2423
result.revision_id, revision_id))
2426
def serialise_inventory(self, inv):
2427
return self._serializer.write_inventory_to_string(inv)
2429
def _serialise_inventory_to_lines(self, inv):
2430
return self._serializer.write_inventory_to_lines(inv)
2432
def get_serializer_format(self):
2433
return self._serializer.format_num
2436
def get_inventory_xml(self, revision_id):
2437
"""Get inventory XML as a file object."""
2438
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2440
text, revision_id = texts.next()
2441
except StopIteration:
2442
raise errors.HistoryMissing(self, 'inventory', revision_id)
2446
def get_inventory_sha1(self, revision_id):
2447
"""Return the sha1 hash of the inventory entry
2449
return self.get_revision(revision_id).inventory_sha1
2451
def get_rev_id_for_revno(self, revno, known_pair):
2452
"""Return the revision id of a revno, given a later (revno, revid)
2453
pair in the same history.
2455
:return: if found (True, revid). If the available history ran out
2456
before reaching the revno, then this returns
2457
(False, (closest_revno, closest_revid)).
2459
known_revno, known_revid = known_pair
2460
partial_history = [known_revid]
2461
distance_from_known = known_revno - revno
2462
if distance_from_known < 0:
2464
'requested revno (%d) is later than given known revno (%d)'
2465
% (revno, known_revno))
2468
self, partial_history, stop_index=distance_from_known)
2469
except errors.RevisionNotPresent, err:
2470
if err.revision_id == known_revid:
2471
# The start revision (known_revid) wasn't found.
2473
# This is a stacked repository with no fallbacks, or a there's a
2474
# left-hand ghost. Either way, even though the revision named in
2475
# the error isn't in this repo, we know it's the next step in this
2476
# left-hand history.
2477
partial_history.append(err.revision_id)
2478
if len(partial_history) <= distance_from_known:
2479
# Didn't find enough history to get a revid for the revno.
2480
earliest_revno = known_revno - len(partial_history) + 1
2481
return (False, (earliest_revno, partial_history[-1]))
2482
if len(partial_history) - 1 > distance_from_known:
2483
raise AssertionError('_iter_for_revno returned too much history')
2484
return (True, partial_history[-1])
2486
def iter_reverse_revision_history(self, revision_id):
2487
"""Iterate backwards through revision ids in the lefthand history
2489
:param revision_id: The revision id to start with. All its lefthand
2490
ancestors will be traversed.
2492
graph = self.get_graph()
2493
next_id = revision_id
2495
if next_id in (None, _mod_revision.NULL_REVISION):
2498
parents = graph.get_parent_map([next_id])[next_id]
2500
raise errors.RevisionNotPresent(next_id, self)
2502
if len(parents) == 0:
2505
next_id = parents[0]
2508
def get_revision_inventory(self, revision_id):
2509
"""Return inventory of a past revision."""
2510
# TODO: Unify this with get_inventory()
2511
# bzr 0.0.6 and later imposes the constraint that the inventory_id
2512
# must be the same as its revision, so this is trivial.
2513
if revision_id is None:
2514
# This does not make sense: if there is no revision,
2515
# then it is the current tree inventory surely ?!
2516
# and thus get_root_id() is something that looks at the last
2517
# commit on the branch, and the get_root_id is an inventory check.
2518
raise NotImplementedError
2519
# return Inventory(self.get_root_id())
2521
return self.get_inventory(revision_id)
2523
def is_shared(self):
2524
"""Return True if this repository is flagged as a shared repository."""
2525
raise NotImplementedError(self.is_shared)
2528
def reconcile(self, other=None, thorough=False):
2529
"""Reconcile this repository."""
2530
from bzrlib.reconcile import RepoReconciler
2531
reconciler = RepoReconciler(self, thorough=thorough)
2532
reconciler.reconcile()
2535
def _refresh_data(self):
2536
"""Helper called from lock_* to ensure coherency with disk.
2538
The default implementation does nothing; it is however possible
2539
for repositories to maintain loaded indices across multiple locks
2540
by checking inside their implementation of this method to see
2541
whether their indices are still valid. This depends of course on
2542
the disk format being validatable in this manner. This method is
2543
also called by the refresh_data() public interface to cause a refresh
2544
to occur while in a write lock so that data inserted by a smart server
2545
push operation is visible on the client's instance of the physical
2550
def revision_tree(self, revision_id):
2551
"""Return Tree for a revision on this branch.
2553
`revision_id` may be NULL_REVISION for the empty tree revision.
2555
revision_id = _mod_revision.ensure_null(revision_id)
2556
# TODO: refactor this to use an existing revision object
2557
# so we don't need to read it in twice.
2558
if revision_id == _mod_revision.NULL_REVISION:
2559
return RevisionTree(self, Inventory(root_id=None),
2560
_mod_revision.NULL_REVISION)
2562
inv = self.get_revision_inventory(revision_id)
2563
return RevisionTree(self, inv, revision_id)
2565
def revision_trees(self, revision_ids):
2566
"""Return Trees for revisions in this repository.
2568
:param revision_ids: a sequence of revision-ids;
2569
a revision-id may not be None or 'null:'
2571
inventories = self.iter_inventories(revision_ids)
2572
for inv in inventories:
2573
yield RevisionTree(self, inv, inv.revision_id)
2575
def _filtered_revision_trees(self, revision_ids, file_ids):
2576
"""Return Tree for a revision on this branch with only some files.
2578
:param revision_ids: a sequence of revision-ids;
2579
a revision-id may not be None or 'null:'
2580
:param file_ids: if not None, the result is filtered
2581
so that only those file-ids, their parents and their
2582
children are included.
2584
inventories = self.iter_inventories(revision_ids)
2585
for inv in inventories:
2586
# Should we introduce a FilteredRevisionTree class rather
2587
# than pre-filter the inventory here?
2588
filtered_inv = inv.filter(file_ids)
2589
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2592
def get_ancestry(self, revision_id, topo_sorted=True):
2593
"""Return a list of revision-ids integrated by a revision.
2595
The first element of the list is always None, indicating the origin
2596
revision. This might change when we have history horizons, or
2597
perhaps we should have a new API.
2599
This is topologically sorted.
2601
if _mod_revision.is_null(revision_id):
2603
if not self.has_revision(revision_id):
2604
raise errors.NoSuchRevision(self, revision_id)
2605
graph = self.get_graph()
2607
search = graph._make_breadth_first_searcher([revision_id])
2610
found, ghosts = search.next_with_ghosts()
2611
except StopIteration:
2614
if _mod_revision.NULL_REVISION in keys:
2615
keys.remove(_mod_revision.NULL_REVISION)
2617
parent_map = graph.get_parent_map(keys)
2618
keys = tsort.topo_sort(parent_map)
2619
return [None] + list(keys)
2621
def pack(self, hint=None):
2622
"""Compress the data within the repository.
2624
This operation only makes sense for some repository types. For other
2625
types it should be a no-op that just returns.
2627
This stub method does not require a lock, but subclasses should use
2628
@needs_write_lock as this is a long running call its reasonable to
2629
implicitly lock for the user.
2631
:param hint: If not supplied, the whole repository is packed.
2632
If supplied, the repository may use the hint parameter as a
2633
hint for the parts of the repository to pack. A hint can be
2634
obtained from the result of commit_write_group(). Out of
2635
date hints are simply ignored, because concurrent operations
2636
can obsolete them rapidly.
2639
def get_transaction(self):
2640
return self.control_files.get_transaction()
2642
def get_parent_map(self, revision_ids):
2643
"""See graph.StackedParentsProvider.get_parent_map"""
2644
# revisions index works in keys; this just works in revisions
2645
# therefore wrap and unwrap
2648
for revision_id in revision_ids:
2649
if revision_id == _mod_revision.NULL_REVISION:
2650
result[revision_id] = ()
2651
elif revision_id is None:
2652
raise ValueError('get_parent_map(None) is not valid')
2654
query_keys.append((revision_id ,))
2655
for ((revision_id,), parent_keys) in \
2656
self.revisions.get_parent_map(query_keys).iteritems():
2658
result[revision_id] = tuple(parent_revid
2659
for (parent_revid,) in parent_keys)
2661
result[revision_id] = (_mod_revision.NULL_REVISION,)
2664
def _make_parents_provider(self):
2667
def get_graph(self, other_repository=None):
2668
"""Return the graph walker for this repository format"""
2669
parents_provider = self._make_parents_provider()
2670
if (other_repository is not None and
2671
not self.has_same_location(other_repository)):
2672
parents_provider = graph.StackedParentsProvider(
2673
[parents_provider, other_repository._make_parents_provider()])
2674
return graph.Graph(parents_provider)
2676
def _get_versioned_file_checker(self, text_key_references=None,
2678
"""Return an object suitable for checking versioned files.
2680
:param text_key_references: if non-None, an already built
2681
dictionary mapping text keys ((fileid, revision_id) tuples)
2682
to whether they were referred to by the inventory of the
2683
revision_id that they contain. If None, this will be
2685
:param ancestors: Optional result from
2686
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2689
return _VersionedFileChecker(self,
2690
text_key_references=text_key_references, ancestors=ancestors)
2692
def revision_ids_to_search_result(self, result_set):
2693
"""Convert a set of revision ids to a graph SearchResult."""
2694
result_parents = set()
2695
for parents in self.get_graph().get_parent_map(
2696
result_set).itervalues():
2697
result_parents.update(parents)
2698
included_keys = result_set.intersection(result_parents)
2699
start_keys = result_set.difference(included_keys)
2700
exclude_keys = result_parents.difference(result_set)
2701
result = graph.SearchResult(start_keys, exclude_keys,
2702
len(result_set), result_set)
2706
def set_make_working_trees(self, new_value):
2707
"""Set the policy flag for making working trees when creating branches.
2709
This only applies to branches that use this repository.
2711
The default is 'True'.
2712
:param new_value: True to restore the default, False to disable making
2715
raise NotImplementedError(self.set_make_working_trees)
2717
def make_working_trees(self):
2718
"""Returns the policy for making working trees on new branches."""
2719
raise NotImplementedError(self.make_working_trees)
2722
def sign_revision(self, revision_id, gpg_strategy):
2723
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2724
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2727
def has_signature_for_revision_id(self, revision_id):
2728
"""Query for a revision signature for revision_id in the repository."""
2729
if not self.has_revision(revision_id):
2730
raise errors.NoSuchRevision(self, revision_id)
2731
sig_present = (1 == len(
2732
self.signatures.get_parent_map([(revision_id,)])))
2736
def get_signature_text(self, revision_id):
2737
"""Return the text for a signature."""
2738
stream = self.signatures.get_record_stream([(revision_id,)],
2740
record = stream.next()
2741
if record.storage_kind == 'absent':
2742
raise errors.NoSuchRevision(self, revision_id)
2743
return record.get_bytes_as('fulltext')
2746
def check(self, revision_ids=None, callback_refs=None, check_repo=True):
2747
"""Check consistency of all history of given revision_ids.
2749
Different repository implementations should override _check().
2751
:param revision_ids: A non-empty list of revision_ids whose ancestry
2752
will be checked. Typically the last revision_id of a branch.
2753
:param callback_refs: A dict of check-refs to resolve and callback
2754
the check/_check method on the items listed as wanting the ref.
2756
:param check_repo: If False do not check the repository contents, just
2757
calculate the data callback_refs requires and call them back.
2759
return self._check(revision_ids, callback_refs=callback_refs,
2760
check_repo=check_repo)
2762
def _check(self, revision_ids, callback_refs, check_repo):
2763
result = check.Check(self, check_repo=check_repo)
2764
result.check(callback_refs)
2767
def _warn_if_deprecated(self):
2768
global _deprecation_warning_done
2769
if _deprecation_warning_done:
2771
_deprecation_warning_done = True
2772
warning("Format %s for %s is deprecated - please use 'bzr upgrade' to get better performance"
2773
% (self._format, self.bzrdir.transport.base))
2775
def supports_rich_root(self):
2776
return self._format.rich_root_data
2778
def _check_ascii_revisionid(self, revision_id, method):
2779
"""Private helper for ascii-only repositories."""
2780
# weave repositories refuse to store revisionids that are non-ascii.
2781
if revision_id is not None:
2782
# weaves require ascii revision ids.
2783
if isinstance(revision_id, unicode):
2785
revision_id.encode('ascii')
2786
except UnicodeEncodeError:
2787
raise errors.NonAsciiRevisionId(method, self)
2790
revision_id.decode('ascii')
2791
except UnicodeDecodeError:
2792
raise errors.NonAsciiRevisionId(method, self)
2794
def revision_graph_can_have_wrong_parents(self):
2795
"""Is it possible for this repository to have a revision graph with
2798
If True, then this repository must also implement
2799
_find_inconsistent_revision_parents so that check and reconcile can
2800
check for inconsistencies before proceeding with other checks that may
2801
depend on the revision index being consistent.
2803
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2806
# remove these delegates a while after bzr 0.15
2807
def __make_delegated(name, from_module):
2808
def _deprecated_repository_forwarder():
2809
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2810
% (name, from_module),
2813
m = __import__(from_module, globals(), locals(), [name])
2815
return getattr(m, name)
2816
except AttributeError:
2817
raise AttributeError('module %s has no name %s'
2819
globals()[name] = _deprecated_repository_forwarder
2822
'AllInOneRepository',
2823
'WeaveMetaDirRepository',
2824
'PreSplitOutRepositoryFormat',
2825
'RepositoryFormat4',
2826
'RepositoryFormat5',
2827
'RepositoryFormat6',
2828
'RepositoryFormat7',
2830
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2834
'RepositoryFormatKnit',
2835
'RepositoryFormatKnit1',
2837
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2840
def install_revision(repository, rev, revision_tree):
2841
"""Install all revision data into a repository."""
2842
install_revisions(repository, [(rev, revision_tree, None)])
2845
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2846
"""Install all revision data into a repository.
2848
Accepts an iterable of revision, tree, signature tuples. The signature
2851
repository.start_write_group()
2853
inventory_cache = lru_cache.LRUCache(10)
2854
for n, (revision, revision_tree, signature) in enumerate(iterable):
2855
_install_revision(repository, revision, revision_tree, signature,
2858
pb.update('Transferring revisions', n + 1, num_revisions)
2860
repository.abort_write_group()
2863
repository.commit_write_group()
2866
def _install_revision(repository, rev, revision_tree, signature,
2868
"""Install all revision data into a repository."""
2869
present_parents = []
2871
for p_id in rev.parent_ids:
2872
if repository.has_revision(p_id):
2873
present_parents.append(p_id)
2874
parent_trees[p_id] = repository.revision_tree(p_id)
2876
parent_trees[p_id] = repository.revision_tree(
2877
_mod_revision.NULL_REVISION)
2879
inv = revision_tree.inventory
2880
entries = inv.iter_entries()
2881
# backwards compatibility hack: skip the root id.
2882
if not repository.supports_rich_root():
2883
path, root = entries.next()
2884
if root.revision != rev.revision_id:
2885
raise errors.IncompatibleRevision(repr(repository))
2887
for path, ie in entries:
2888
text_keys[(ie.file_id, ie.revision)] = ie
2889
text_parent_map = repository.texts.get_parent_map(text_keys)
2890
missing_texts = set(text_keys) - set(text_parent_map)
2891
# Add the texts that are not already present
2892
for text_key in missing_texts:
2893
ie = text_keys[text_key]
2895
# FIXME: TODO: The following loop overlaps/duplicates that done by
2896
# commit to determine parents. There is a latent/real bug here where
2897
# the parents inserted are not those commit would do - in particular
2898
# they are not filtered by heads(). RBC, AB
2899
for revision, tree in parent_trees.iteritems():
2900
if ie.file_id not in tree:
2902
parent_id = tree.inventory[ie.file_id].revision
2903
if parent_id in text_parents:
2905
text_parents.append((ie.file_id, parent_id))
2906
lines = revision_tree.get_file(ie.file_id).readlines()
2907
repository.texts.add_lines(text_key, text_parents, lines)
2909
# install the inventory
2910
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2911
# Cache this inventory
2912
inventory_cache[rev.revision_id] = inv
2914
basis_inv = inventory_cache[rev.parent_ids[0]]
2916
repository.add_inventory(rev.revision_id, inv, present_parents)
2918
delta = inv._make_delta(basis_inv)
2919
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2920
rev.revision_id, present_parents)
2922
repository.add_inventory(rev.revision_id, inv, present_parents)
2923
except errors.RevisionAlreadyPresent:
2925
if signature is not None:
2926
repository.add_signature_text(rev.revision_id, signature)
2927
repository.add_revision(rev.revision_id, rev, inv)
2930
class MetaDirRepository(Repository):
2931
"""Repositories in the new meta-dir layout.
2933
:ivar _transport: Transport for access to repository control files,
2934
typically pointing to .bzr/repository.
2937
def __init__(self, _format, a_bzrdir, control_files):
2938
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2939
self._transport = control_files._transport
2941
def is_shared(self):
2942
"""Return True if this repository is flagged as a shared repository."""
2943
return self._transport.has('shared-storage')
2946
def set_make_working_trees(self, new_value):
2947
"""Set the policy flag for making working trees when creating branches.
2949
This only applies to branches that use this repository.
2951
The default is 'True'.
2952
:param new_value: True to restore the default, False to disable making
2957
self._transport.delete('no-working-trees')
2958
except errors.NoSuchFile:
2961
self._transport.put_bytes('no-working-trees', '',
2962
mode=self.bzrdir._get_file_mode())
2964
def make_working_trees(self):
2965
"""Returns the policy for making working trees on new branches."""
2966
return not self._transport.has('no-working-trees')
2969
class MetaDirVersionedFileRepository(MetaDirRepository):
2970
"""Repositories in a meta-dir, that work via versioned file objects."""
2972
def __init__(self, _format, a_bzrdir, control_files):
2973
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2977
network_format_registry = registry.FormatRegistry()
2978
"""Registry of formats indexed by their network name.
2980
The network name for a repository format is an identifier that can be used when
2981
referring to formats with smart server operations. See
2982
RepositoryFormat.network_name() for more detail.
2986
format_registry = registry.FormatRegistry(network_format_registry)
2987
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
2989
This can contain either format instances themselves, or classes/factories that
2990
can be called to obtain one.
2994
#####################################################################
2995
# Repository Formats
2997
class RepositoryFormat(object):
2998
"""A repository format.
3000
Formats provide four things:
3001
* An initialization routine to construct repository data on disk.
3002
* a optional format string which is used when the BzrDir supports
3004
* an open routine which returns a Repository instance.
3005
* A network name for referring to the format in smart server RPC
3008
There is one and only one Format subclass for each on-disk format. But
3009
there can be one Repository subclass that is used for several different
3010
formats. The _format attribute on a Repository instance can be used to
3011
determine the disk format.
3013
Formats are placed in a registry by their format string for reference
3014
during opening. These should be subclasses of RepositoryFormat for
3017
Once a format is deprecated, just deprecate the initialize and open
3018
methods on the format class. Do not deprecate the object, as the
3019
object may be created even when a repository instance hasn't been
3022
Common instance attributes:
3023
_matchingbzrdir - the bzrdir format that the repository format was
3024
originally written to work with. This can be used if manually
3025
constructing a bzrdir and repository, or more commonly for test suite
3029
# Set to True or False in derived classes. True indicates that the format
3030
# supports ghosts gracefully.
3031
supports_ghosts = None
3032
# Can this repository be given external locations to lookup additional
3033
# data. Set to True or False in derived classes.
3034
supports_external_lookups = None
3035
# Does this format support CHK bytestring lookups. Set to True or False in
3037
supports_chks = None
3038
# Should commit add an inventory, or an inventory delta to the repository.
3039
_commit_inv_deltas = True
3040
# What order should fetch operations request streams in?
3041
# The default is unordered as that is the cheapest for an origin to
3043
_fetch_order = 'unordered'
3044
# Does this repository format use deltas that can be fetched as-deltas ?
3045
# (E.g. knits, where the knit deltas can be transplanted intact.
3046
# We default to False, which will ensure that enough data to get
3047
# a full text out of any fetch stream will be grabbed.
3048
_fetch_uses_deltas = False
3049
# Should fetch trigger a reconcile after the fetch? Only needed for
3050
# some repository formats that can suffer internal inconsistencies.
3051
_fetch_reconcile = False
3052
# Does this format have < O(tree_size) delta generation. Used to hint what
3053
# code path for commit, amongst other things.
3055
# Does doing a pack operation compress data? Useful for the pack UI command
3056
# (so if there is one pack, the operation can still proceed because it may
3057
# help), and for fetching when data won't have come from the same
3059
pack_compresses = False
3060
# Does the repository inventory storage understand references to trees?
3061
supports_tree_reference = None
3064
return "<%s>" % self.__class__.__name__
3066
def __eq__(self, other):
3067
# format objects are generally stateless
3068
return isinstance(other, self.__class__)
3070
def __ne__(self, other):
3071
return not self == other
3074
def find_format(klass, a_bzrdir):
3075
"""Return the format for the repository object in a_bzrdir.
3077
This is used by bzr native formats that have a "format" file in
3078
the repository. Other methods may be used by different types of
3082
transport = a_bzrdir.get_repository_transport(None)
3083
format_string = transport.get("format").read()
3084
return format_registry.get(format_string)
3085
except errors.NoSuchFile:
3086
raise errors.NoRepositoryPresent(a_bzrdir)
3088
raise errors.UnknownFormatError(format=format_string,
3092
def register_format(klass, format):
3093
format_registry.register(format.get_format_string(), format)
3096
def unregister_format(klass, format):
3097
format_registry.remove(format.get_format_string())
3100
def get_default_format(klass):
3101
"""Return the current default format."""
3102
from bzrlib import bzrdir
3103
return bzrdir.format_registry.make_bzrdir('default').repository_format
3105
def get_format_string(self):
3106
"""Return the ASCII format string that identifies this format.
3108
Note that in pre format ?? repositories the format string is
3109
not permitted nor written to disk.
3111
raise NotImplementedError(self.get_format_string)
3113
def get_format_description(self):
3114
"""Return the short description for this format."""
3115
raise NotImplementedError(self.get_format_description)
3117
# TODO: this shouldn't be in the base class, it's specific to things that
3118
# use weaves or knits -- mbp 20070207
3119
def _get_versioned_file_store(self,
3124
versionedfile_class=None,
3125
versionedfile_kwargs={},
3127
if versionedfile_class is None:
3128
versionedfile_class = self._versionedfile_class
3129
weave_transport = control_files._transport.clone(name)
3130
dir_mode = control_files._dir_mode
3131
file_mode = control_files._file_mode
3132
return VersionedFileStore(weave_transport, prefixed=prefixed,
3134
file_mode=file_mode,
3135
versionedfile_class=versionedfile_class,
3136
versionedfile_kwargs=versionedfile_kwargs,
3139
def initialize(self, a_bzrdir, shared=False):
3140
"""Initialize a repository of this format in a_bzrdir.
3142
:param a_bzrdir: The bzrdir to put the new repository in it.
3143
:param shared: The repository should be initialized as a sharable one.
3144
:returns: The new repository object.
3146
This may raise UninitializableFormat if shared repository are not
3147
compatible the a_bzrdir.
3149
raise NotImplementedError(self.initialize)
3151
def is_supported(self):
3152
"""Is this format supported?
3154
Supported formats must be initializable and openable.
3155
Unsupported formats may not support initialization or committing or
3156
some other features depending on the reason for not being supported.
3160
def network_name(self):
3161
"""A simple byte string uniquely identifying this format for RPC calls.
3163
MetaDir repository formats use their disk format string to identify the
3164
repository over the wire. All in one formats such as bzr < 0.8, and
3165
foreign formats like svn/git and hg should use some marker which is
3166
unique and immutable.
3168
raise NotImplementedError(self.network_name)
3170
def check_conversion_target(self, target_format):
3171
if self.rich_root_data and not target_format.rich_root_data:
3172
raise errors.BadConversionTarget(
3173
'Does not support rich root data.', target_format,
3175
if (self.supports_tree_reference and
3176
not getattr(target_format, 'supports_tree_reference', False)):
3177
raise errors.BadConversionTarget(
3178
'Does not support nested trees', target_format,
3181
def open(self, a_bzrdir, _found=False):
3182
"""Return an instance of this format for the bzrdir a_bzrdir.
3184
_found is a private parameter, do not use it.
3186
raise NotImplementedError(self.open)
3189
class MetaDirRepositoryFormat(RepositoryFormat):
3190
"""Common base class for the new repositories using the metadir layout."""
3192
rich_root_data = False
3193
supports_tree_reference = False
3194
supports_external_lookups = False
3197
def _matchingbzrdir(self):
3198
matching = bzrdir.BzrDirMetaFormat1()
3199
matching.repository_format = self
3203
super(MetaDirRepositoryFormat, self).__init__()
3205
def _create_control_files(self, a_bzrdir):
3206
"""Create the required files and the initial control_files object."""
3207
# FIXME: RBC 20060125 don't peek under the covers
3208
# NB: no need to escape relative paths that are url safe.
3209
repository_transport = a_bzrdir.get_repository_transport(self)
3210
control_files = lockable_files.LockableFiles(repository_transport,
3211
'lock', lockdir.LockDir)
3212
control_files.create_lock()
3213
return control_files
3215
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
3216
"""Upload the initial blank content."""
3217
control_files = self._create_control_files(a_bzrdir)
3218
control_files.lock_write()
3219
transport = control_files._transport
3221
utf8_files += [('shared-storage', '')]
3223
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
3224
for (filename, content_stream) in files:
3225
transport.put_file(filename, content_stream,
3226
mode=a_bzrdir._get_file_mode())
3227
for (filename, content_bytes) in utf8_files:
3228
transport.put_bytes_non_atomic(filename, content_bytes,
3229
mode=a_bzrdir._get_file_mode())
3231
control_files.unlock()
3233
def network_name(self):
3234
"""Metadir formats have matching disk and network format strings."""
3235
return self.get_format_string()
3238
# Pre-0.8 formats that don't have a disk format string (because they are
3239
# versioned by the matching control directory). We use the control directories
3240
# disk format string as a key for the network_name because they meet the
3241
# constraints (simple string, unique, immutable).
3242
network_format_registry.register_lazy(
3243
"Bazaar-NG branch, format 5\n",
3244
'bzrlib.repofmt.weaverepo',
3245
'RepositoryFormat5',
3247
network_format_registry.register_lazy(
3248
"Bazaar-NG branch, format 6\n",
3249
'bzrlib.repofmt.weaverepo',
3250
'RepositoryFormat6',
3253
# formats which have no format string are not discoverable or independently
3254
# creatable on disk, so are not registered in format_registry. They're
3255
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
3256
# needed, it's constructed directly by the BzrDir. Non-native formats where
3257
# the repository is not separately opened are similar.
3259
format_registry.register_lazy(
3260
'Bazaar-NG Repository format 7',
3261
'bzrlib.repofmt.weaverepo',
3265
format_registry.register_lazy(
3266
'Bazaar-NG Knit Repository Format 1',
3267
'bzrlib.repofmt.knitrepo',
3268
'RepositoryFormatKnit1',
3271
format_registry.register_lazy(
3272
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
3273
'bzrlib.repofmt.knitrepo',
3274
'RepositoryFormatKnit3',
3277
format_registry.register_lazy(
3278
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
3279
'bzrlib.repofmt.knitrepo',
3280
'RepositoryFormatKnit4',
3283
# Pack-based formats. There is one format for pre-subtrees, and one for
3284
# post-subtrees to allow ease of testing.
3285
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
3286
format_registry.register_lazy(
3287
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
3288
'bzrlib.repofmt.pack_repo',
3289
'RepositoryFormatKnitPack1',
3291
format_registry.register_lazy(
3292
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
3293
'bzrlib.repofmt.pack_repo',
3294
'RepositoryFormatKnitPack3',
3296
format_registry.register_lazy(
3297
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
3298
'bzrlib.repofmt.pack_repo',
3299
'RepositoryFormatKnitPack4',
3301
format_registry.register_lazy(
3302
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
3303
'bzrlib.repofmt.pack_repo',
3304
'RepositoryFormatKnitPack5',
3306
format_registry.register_lazy(
3307
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
3308
'bzrlib.repofmt.pack_repo',
3309
'RepositoryFormatKnitPack5RichRoot',
3311
format_registry.register_lazy(
3312
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3313
'bzrlib.repofmt.pack_repo',
3314
'RepositoryFormatKnitPack5RichRootBroken',
3316
format_registry.register_lazy(
3317
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3318
'bzrlib.repofmt.pack_repo',
3319
'RepositoryFormatKnitPack6',
3321
format_registry.register_lazy(
3322
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3323
'bzrlib.repofmt.pack_repo',
3324
'RepositoryFormatKnitPack6RichRoot',
3327
# Development formats.
3328
# Obsolete but kept pending a CHK based subtree format.
3329
format_registry.register_lazy(
3330
("Bazaar development format 2 with subtree support "
3331
"(needs bzr.dev from before 1.8)\n"),
3332
'bzrlib.repofmt.pack_repo',
3333
'RepositoryFormatPackDevelopment2Subtree',
3336
# 1.14->1.16 go below here
3337
format_registry.register_lazy(
3338
'Bazaar development format - group compression and chk inventory'
3339
' (needs bzr.dev from 1.14)\n',
3340
'bzrlib.repofmt.groupcompress_repo',
3341
'RepositoryFormatCHK1',
3344
format_registry.register_lazy(
3345
'Bazaar development format - chk repository with bencode revision '
3346
'serialization (needs bzr.dev from 1.16)\n',
3347
'bzrlib.repofmt.groupcompress_repo',
3348
'RepositoryFormatCHK2',
3350
format_registry.register_lazy(
3351
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3352
'bzrlib.repofmt.groupcompress_repo',
3353
'RepositoryFormat2a',
3357
class InterRepository(InterObject):
3358
"""This class represents operations taking place between two repositories.
3360
Its instances have methods like copy_content and fetch, and contain
3361
references to the source and target repositories these operations can be
3364
Often we will provide convenience methods on 'repository' which carry out
3365
operations with another repository - they will always forward to
3366
InterRepository.get(other).method_name(parameters).
3369
_walk_to_common_revisions_batch_size = 50
3371
"""The available optimised InterRepository types."""
3374
def copy_content(self, revision_id=None):
3375
"""Make a complete copy of the content in self into destination.
3377
This is a destructive operation! Do not use it on existing
3380
:param revision_id: Only copy the content needed to construct
3381
revision_id and its parents.
3384
self.target.set_make_working_trees(self.source.make_working_trees())
3385
except NotImplementedError:
3387
self.target.fetch(self.source, revision_id=revision_id)
3390
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3392
"""Fetch the content required to construct revision_id.
3394
The content is copied from self.source to self.target.
3396
:param revision_id: if None all content is copied, if NULL_REVISION no
3398
:param pb: optional progress bar to use for progress reports. If not
3399
provided a default one will be created.
3402
from bzrlib.fetch import RepoFetcher
3403
f = RepoFetcher(to_repository=self.target,
3404
from_repository=self.source,
3405
last_revision=revision_id,
3406
fetch_spec=fetch_spec,
3407
pb=pb, find_ghosts=find_ghosts)
3409
def _walk_to_common_revisions(self, revision_ids):
3410
"""Walk out from revision_ids in source to revisions target has.
3412
:param revision_ids: The start point for the search.
3413
:return: A set of revision ids.
3415
target_graph = self.target.get_graph()
3416
revision_ids = frozenset(revision_ids)
3417
missing_revs = set()
3418
source_graph = self.source.get_graph()
3419
# ensure we don't pay silly lookup costs.
3420
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3421
null_set = frozenset([_mod_revision.NULL_REVISION])
3422
searcher_exhausted = False
3426
# Iterate the searcher until we have enough next_revs
3427
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3429
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3430
next_revs.update(next_revs_part)
3431
ghosts.update(ghosts_part)
3432
except StopIteration:
3433
searcher_exhausted = True
3435
# If there are ghosts in the source graph, and the caller asked for
3436
# them, make sure that they are present in the target.
3437
# We don't care about other ghosts as we can't fetch them and
3438
# haven't been asked to.
3439
ghosts_to_check = set(revision_ids.intersection(ghosts))
3440
revs_to_get = set(next_revs).union(ghosts_to_check)
3442
have_revs = set(target_graph.get_parent_map(revs_to_get))
3443
# we always have NULL_REVISION present.
3444
have_revs = have_revs.union(null_set)
3445
# Check if the target is missing any ghosts we need.
3446
ghosts_to_check.difference_update(have_revs)
3448
# One of the caller's revision_ids is a ghost in both the
3449
# source and the target.
3450
raise errors.NoSuchRevision(
3451
self.source, ghosts_to_check.pop())
3452
missing_revs.update(next_revs - have_revs)
3453
# Because we may have walked past the original stop point, make
3454
# sure everything is stopped
3455
stop_revs = searcher.find_seen_ancestors(have_revs)
3456
searcher.stop_searching_any(stop_revs)
3457
if searcher_exhausted:
3459
return searcher.get_result()
3462
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3463
"""Return the revision ids that source has that target does not.
3465
:param revision_id: only return revision ids included by this
3467
:param find_ghosts: If True find missing revisions in deep history
3468
rather than just finding the surface difference.
3469
:return: A bzrlib.graph.SearchResult.
3471
# stop searching at found target revisions.
3472
if not find_ghosts and revision_id is not None:
3473
return self._walk_to_common_revisions([revision_id])
3474
# generic, possibly worst case, slow code path.
3475
target_ids = set(self.target.all_revision_ids())
3476
if revision_id is not None:
3477
source_ids = self.source.get_ancestry(revision_id)
3478
if source_ids[0] is not None:
3479
raise AssertionError()
3482
source_ids = self.source.all_revision_ids()
3483
result_set = set(source_ids).difference(target_ids)
3484
return self.source.revision_ids_to_search_result(result_set)
3487
def _same_model(source, target):
3488
"""True if source and target have the same data representation.
3490
Note: this is always called on the base class; overriding it in a
3491
subclass will have no effect.
3494
InterRepository._assert_same_model(source, target)
3496
except errors.IncompatibleRepositories, e:
3500
def _assert_same_model(source, target):
3501
"""Raise an exception if two repositories do not use the same model.
3503
if source.supports_rich_root() != target.supports_rich_root():
3504
raise errors.IncompatibleRepositories(source, target,
3505
"different rich-root support")
3506
if source._serializer != target._serializer:
3507
raise errors.IncompatibleRepositories(source, target,
3508
"different serializers")
3511
class InterSameDataRepository(InterRepository):
3512
"""Code for converting between repositories that represent the same data.
3514
Data format and model must match for this to work.
3518
def _get_repo_format_to_test(self):
3519
"""Repository format for testing with.
3521
InterSameData can pull from subtree to subtree and from non-subtree to
3522
non-subtree, so we test this with the richest repository format.
3524
from bzrlib.repofmt import knitrepo
3525
return knitrepo.RepositoryFormatKnit3()
3528
def is_compatible(source, target):
3529
return InterRepository._same_model(source, target)
3532
class InterWeaveRepo(InterSameDataRepository):
3533
"""Optimised code paths between Weave based repositories.
3535
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3536
implemented lazy inter-object optimisation.
3540
def _get_repo_format_to_test(self):
3541
from bzrlib.repofmt import weaverepo
3542
return weaverepo.RepositoryFormat7()
3545
def is_compatible(source, target):
3546
"""Be compatible with known Weave formats.
3548
We don't test for the stores being of specific types because that
3549
could lead to confusing results, and there is no need to be
3552
from bzrlib.repofmt.weaverepo import (
3558
return (isinstance(source._format, (RepositoryFormat5,
3560
RepositoryFormat7)) and
3561
isinstance(target._format, (RepositoryFormat5,
3563
RepositoryFormat7)))
3564
except AttributeError:
3568
def copy_content(self, revision_id=None):
3569
"""See InterRepository.copy_content()."""
3570
# weave specific optimised path:
3572
self.target.set_make_working_trees(self.source.make_working_trees())
3573
except (errors.RepositoryUpgradeRequired, NotImplemented):
3575
# FIXME do not peek!
3576
if self.source._transport.listable():
3577
pb = ui.ui_factory.nested_progress_bar()
3579
self.target.texts.insert_record_stream(
3580
self.source.texts.get_record_stream(
3581
self.source.texts.keys(), 'topological', False))
3582
pb.update('copying inventory', 0, 1)
3583
self.target.inventories.insert_record_stream(
3584
self.source.inventories.get_record_stream(
3585
self.source.inventories.keys(), 'topological', False))
3586
self.target.signatures.insert_record_stream(
3587
self.source.signatures.get_record_stream(
3588
self.source.signatures.keys(),
3590
self.target.revisions.insert_record_stream(
3591
self.source.revisions.get_record_stream(
3592
self.source.revisions.keys(),
3593
'topological', True))
3597
self.target.fetch(self.source, revision_id=revision_id)
3600
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3601
"""See InterRepository.missing_revision_ids()."""
3602
# we want all revisions to satisfy revision_id in source.
3603
# but we don't want to stat every file here and there.
3604
# we want then, all revisions other needs to satisfy revision_id
3605
# checked, but not those that we have locally.
3606
# so the first thing is to get a subset of the revisions to
3607
# satisfy revision_id in source, and then eliminate those that
3608
# we do already have.
3609
# this is slow on high latency connection to self, but as this
3610
# disk format scales terribly for push anyway due to rewriting
3611
# inventory.weave, this is considered acceptable.
3613
if revision_id is not None:
3614
source_ids = self.source.get_ancestry(revision_id)
3615
if source_ids[0] is not None:
3616
raise AssertionError()
3619
source_ids = self.source._all_possible_ids()
3620
source_ids_set = set(source_ids)
3621
# source_ids is the worst possible case we may need to pull.
3622
# now we want to filter source_ids against what we actually
3623
# have in target, but don't try to check for existence where we know
3624
# we do not have a revision as that would be pointless.
3625
target_ids = set(self.target._all_possible_ids())
3626
possibly_present_revisions = target_ids.intersection(source_ids_set)
3627
actually_present_revisions = set(
3628
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3629
required_revisions = source_ids_set.difference(actually_present_revisions)
3630
if revision_id is not None:
3631
# we used get_ancestry to determine source_ids then we are assured all
3632
# revisions referenced are present as they are installed in topological order.
3633
# and the tip revision was validated by get_ancestry.
3634
result_set = required_revisions
3636
# if we just grabbed the possibly available ids, then
3637
# we only have an estimate of whats available and need to validate
3638
# that against the revision records.
3640
self.source._eliminate_revisions_not_present(required_revisions))
3641
return self.source.revision_ids_to_search_result(result_set)
3644
class InterKnitRepo(InterSameDataRepository):
3645
"""Optimised code paths between Knit based repositories."""
3648
def _get_repo_format_to_test(self):
3649
from bzrlib.repofmt import knitrepo
3650
return knitrepo.RepositoryFormatKnit1()
3653
def is_compatible(source, target):
3654
"""Be compatible with known Knit formats.
3656
We don't test for the stores being of specific types because that
3657
could lead to confusing results, and there is no need to be
3660
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3662
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3663
isinstance(target._format, RepositoryFormatKnit))
3664
except AttributeError:
3666
return are_knits and InterRepository._same_model(source, target)
3669
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3670
"""See InterRepository.missing_revision_ids()."""
3671
if revision_id is not None:
3672
source_ids = self.source.get_ancestry(revision_id)
3673
if source_ids[0] is not None:
3674
raise AssertionError()
3677
source_ids = self.source.all_revision_ids()
3678
source_ids_set = set(source_ids)
3679
# source_ids is the worst possible case we may need to pull.
3680
# now we want to filter source_ids against what we actually
3681
# have in target, but don't try to check for existence where we know
3682
# we do not have a revision as that would be pointless.
3683
target_ids = set(self.target.all_revision_ids())
3684
possibly_present_revisions = target_ids.intersection(source_ids_set)
3685
actually_present_revisions = set(
3686
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3687
required_revisions = source_ids_set.difference(actually_present_revisions)
3688
if revision_id is not None:
3689
# we used get_ancestry to determine source_ids then we are assured all
3690
# revisions referenced are present as they are installed in topological order.
3691
# and the tip revision was validated by get_ancestry.
3692
result_set = required_revisions
3694
# if we just grabbed the possibly available ids, then
3695
# we only have an estimate of whats available and need to validate
3696
# that against the revision records.
3698
self.source._eliminate_revisions_not_present(required_revisions))
3699
return self.source.revision_ids_to_search_result(result_set)
3702
class InterDifferingSerializer(InterRepository):
3705
def _get_repo_format_to_test(self):
3709
def is_compatible(source, target):
3710
"""Be compatible with Knit2 source and Knit3 target"""
3711
# This is redundant with format.check_conversion_target(), however that
3712
# raises an exception, and we just want to say "False" as in we won't
3713
# support converting between these formats.
3714
if 'IDS_never' in debug.debug_flags:
3716
if source.supports_rich_root() and not target.supports_rich_root():
3718
if (source._format.supports_tree_reference
3719
and not target._format.supports_tree_reference):
3721
if target._fallback_repositories and target._format.supports_chks:
3722
# IDS doesn't know how to copy CHKs for the parent inventories it
3723
# adds to stacked repos.
3725
if 'IDS_always' in debug.debug_flags:
3727
# Only use this code path for local source and target. IDS does far
3728
# too much IO (both bandwidth and roundtrips) over a network.
3729
if not source.bzrdir.transport.base.startswith('file:///'):
3731
if not target.bzrdir.transport.base.startswith('file:///'):
3735
def _get_trees(self, revision_ids, cache):
3737
for rev_id in revision_ids:
3739
possible_trees.append((rev_id, cache[rev_id]))
3741
# Not cached, but inventory might be present anyway.
3743
tree = self.source.revision_tree(rev_id)
3744
except errors.NoSuchRevision:
3745
# Nope, parent is ghost.
3748
cache[rev_id] = tree
3749
possible_trees.append((rev_id, tree))
3750
return possible_trees
3752
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3753
"""Get the best delta and base for this revision.
3755
:return: (basis_id, delta)
3758
# Generate deltas against each tree, to find the shortest.
3759
texts_possibly_new_in_tree = set()
3760
for basis_id, basis_tree in possible_trees:
3761
delta = tree.inventory._make_delta(basis_tree.inventory)
3762
for old_path, new_path, file_id, new_entry in delta:
3763
if new_path is None:
3764
# This file_id isn't present in the new rev, so we don't
3768
# Rich roots are handled elsewhere...
3770
kind = new_entry.kind
3771
if kind != 'directory' and kind != 'file':
3772
# No text record associated with this inventory entry.
3774
# This is a directory or file that has changed somehow.
3775
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3776
deltas.append((len(delta), basis_id, delta))
3778
return deltas[0][1:]
3780
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3781
"""Find all parent revisions that are absent, but for which the
3782
inventory is present, and copy those inventories.
3784
This is necessary to preserve correctness when the source is stacked
3785
without fallbacks configured. (Note that in cases like upgrade the
3786
source may be not have _fallback_repositories even though it is
3790
for parents in parent_map.values():
3791
parent_revs.update(parents)
3792
present_parents = self.source.get_parent_map(parent_revs)
3793
absent_parents = set(parent_revs).difference(present_parents)
3794
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3795
(rev_id,) for rev_id in absent_parents)
3796
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3797
for parent_tree in self.source.revision_trees(parent_inv_ids):
3798
current_revision_id = parent_tree.get_revision_id()
3799
parents_parents_keys = parent_invs_keys_for_stacking[
3800
(current_revision_id,)]
3801
parents_parents = [key[-1] for key in parents_parents_keys]
3802
basis_id = _mod_revision.NULL_REVISION
3803
basis_tree = self.source.revision_tree(basis_id)
3804
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3805
self.target.add_inventory_by_delta(
3806
basis_id, delta, current_revision_id, parents_parents)
3807
cache[current_revision_id] = parent_tree
3809
def _fetch_batch(self, revision_ids, basis_id, cache):
3810
"""Fetch across a few revisions.
3812
:param revision_ids: The revisions to copy
3813
:param basis_id: The revision_id of a tree that must be in cache, used
3814
as a basis for delta when no other base is available
3815
:param cache: A cache of RevisionTrees that we can use.
3816
:return: The revision_id of the last converted tree. The RevisionTree
3817
for it will be in cache
3819
# Walk though all revisions; get inventory deltas, copy referenced
3820
# texts that delta references, insert the delta, revision and
3822
root_keys_to_create = set()
3825
pending_revisions = []
3826
parent_map = self.source.get_parent_map(revision_ids)
3827
self._fetch_parent_invs_for_stacking(parent_map, cache)
3828
for tree in self.source.revision_trees(revision_ids):
3829
# Find a inventory delta for this revision.
3830
# Find text entries that need to be copied, too.
3831
current_revision_id = tree.get_revision_id()
3832
parent_ids = parent_map.get(current_revision_id, ())
3833
parent_trees = self._get_trees(parent_ids, cache)
3834
possible_trees = list(parent_trees)
3835
if len(possible_trees) == 0:
3836
# There either aren't any parents, or the parents are ghosts,
3837
# so just use the last converted tree.
3838
possible_trees.append((basis_id, cache[basis_id]))
3839
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3841
if self._converting_to_rich_root:
3842
self._revision_id_to_root_id[current_revision_id] = \
3844
# Determine which texts are in present in this revision but not in
3845
# any of the available parents.
3846
texts_possibly_new_in_tree = set()
3847
for old_path, new_path, file_id, entry in delta:
3848
if new_path is None:
3849
# This file_id isn't present in the new rev
3853
if not self.target.supports_rich_root():
3854
# The target doesn't support rich root, so we don't
3857
if self._converting_to_rich_root:
3858
# This can't be copied normally, we have to insert
3860
root_keys_to_create.add((file_id, entry.revision))
3863
texts_possibly_new_in_tree.add((file_id, entry.revision))
3864
for basis_id, basis_tree in possible_trees:
3865
basis_inv = basis_tree.inventory
3866
for file_key in list(texts_possibly_new_in_tree):
3867
file_id, file_revision = file_key
3869
entry = basis_inv[file_id]
3870
except errors.NoSuchId:
3872
if entry.revision == file_revision:
3873
texts_possibly_new_in_tree.remove(file_key)
3874
text_keys.update(texts_possibly_new_in_tree)
3875
revision = self.source.get_revision(current_revision_id)
3876
pending_deltas.append((basis_id, delta,
3877
current_revision_id, revision.parent_ids))
3878
pending_revisions.append(revision)
3879
cache[current_revision_id] = tree
3880
basis_id = current_revision_id
3882
from_texts = self.source.texts
3883
to_texts = self.target.texts
3884
if root_keys_to_create:
3885
from bzrlib.fetch import _new_root_data_stream
3886
root_stream = _new_root_data_stream(
3887
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3889
to_texts.insert_record_stream(root_stream)
3890
to_texts.insert_record_stream(from_texts.get_record_stream(
3891
text_keys, self.target._format._fetch_order,
3892
not self.target._format._fetch_uses_deltas))
3893
# insert inventory deltas
3894
for delta in pending_deltas:
3895
self.target.add_inventory_by_delta(*delta)
3896
if self.target._fallback_repositories:
3897
# Make sure this stacked repository has all the parent inventories
3898
# for the new revisions that we are about to insert. We do this
3899
# before adding the revisions so that no revision is added until
3900
# all the inventories it may depend on are added.
3901
# Note that this is overzealous, as we may have fetched these in an
3904
revision_ids = set()
3905
for revision in pending_revisions:
3906
revision_ids.add(revision.revision_id)
3907
parent_ids.update(revision.parent_ids)
3908
parent_ids.difference_update(revision_ids)
3909
parent_ids.discard(_mod_revision.NULL_REVISION)
3910
parent_map = self.source.get_parent_map(parent_ids)
3911
# we iterate over parent_map and not parent_ids because we don't
3912
# want to try copying any revision which is a ghost
3913
for parent_tree in self.source.revision_trees(parent_map):
3914
current_revision_id = parent_tree.get_revision_id()
3915
parents_parents = parent_map[current_revision_id]
3916
possible_trees = self._get_trees(parents_parents, cache)
3917
if len(possible_trees) == 0:
3918
# There either aren't any parents, or the parents are
3919
# ghosts, so just use the last converted tree.
3920
possible_trees.append((basis_id, cache[basis_id]))
3921
basis_id, delta = self._get_delta_for_revision(parent_tree,
3922
parents_parents, possible_trees)
3923
self.target.add_inventory_by_delta(
3924
basis_id, delta, current_revision_id, parents_parents)
3925
# insert signatures and revisions
3926
for revision in pending_revisions:
3928
signature = self.source.get_signature_text(
3929
revision.revision_id)
3930
self.target.add_signature_text(revision.revision_id,
3932
except errors.NoSuchRevision:
3934
self.target.add_revision(revision.revision_id, revision)
3937
def _fetch_all_revisions(self, revision_ids, pb):
3938
"""Fetch everything for the list of revisions.
3940
:param revision_ids: The list of revisions to fetch. Must be in
3942
:param pb: A ProgressTask
3945
basis_id, basis_tree = self._get_basis(revision_ids[0])
3947
cache = lru_cache.LRUCache(100)
3948
cache[basis_id] = basis_tree
3949
del basis_tree # We don't want to hang on to it here
3951
for offset in range(0, len(revision_ids), batch_size):
3952
self.target.start_write_group()
3954
pb.update('Transferring revisions', offset,
3956
batch = revision_ids[offset:offset+batch_size]
3957
basis_id = self._fetch_batch(batch, basis_id, cache)
3959
self.target.abort_write_group()
3962
hint = self.target.commit_write_group()
3965
if hints and self.target._format.pack_compresses:
3966
self.target.pack(hint=hints)
3967
pb.update('Transferring revisions', len(revision_ids),
3971
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3973
"""See InterRepository.fetch()."""
3974
if fetch_spec is not None:
3975
raise AssertionError("Not implemented yet...")
3976
if (not self.source.supports_rich_root()
3977
and self.target.supports_rich_root()):
3978
self._converting_to_rich_root = True
3979
self._revision_id_to_root_id = {}
3981
self._converting_to_rich_root = False
3982
revision_ids = self.target.search_missing_revision_ids(self.source,
3983
revision_id, find_ghosts=find_ghosts).get_keys()
3984
if not revision_ids:
3986
revision_ids = tsort.topo_sort(
3987
self.source.get_graph().get_parent_map(revision_ids))
3988
if not revision_ids:
3990
# Walk though all revisions; get inventory deltas, copy referenced
3991
# texts that delta references, insert the delta, revision and
3994
my_pb = ui.ui_factory.nested_progress_bar()
3997
symbol_versioning.warn(
3998
symbol_versioning.deprecated_in((1, 14, 0))
3999
% "pb parameter to fetch()")
4002
self._fetch_all_revisions(revision_ids, pb)
4004
if my_pb is not None:
4006
return len(revision_ids), 0
4008
def _get_basis(self, first_revision_id):
4009
"""Get a revision and tree which exists in the target.
4011
This assumes that first_revision_id is selected for transmission
4012
because all other ancestors are already present. If we can't find an
4013
ancestor we fall back to NULL_REVISION since we know that is safe.
4015
:return: (basis_id, basis_tree)
4017
first_rev = self.source.get_revision(first_revision_id)
4019
basis_id = first_rev.parent_ids[0]
4020
# only valid as a basis if the target has it
4021
self.target.get_revision(basis_id)
4022
# Try to get a basis tree - if its a ghost it will hit the
4023
# NoSuchRevision case.
4024
basis_tree = self.source.revision_tree(basis_id)
4025
except (IndexError, errors.NoSuchRevision):
4026
basis_id = _mod_revision.NULL_REVISION
4027
basis_tree = self.source.revision_tree(basis_id)
4028
return basis_id, basis_tree
4031
InterRepository.register_optimiser(InterDifferingSerializer)
4032
InterRepository.register_optimiser(InterSameDataRepository)
4033
InterRepository.register_optimiser(InterWeaveRepo)
4034
InterRepository.register_optimiser(InterKnitRepo)
4037
class CopyConverter(object):
4038
"""A repository conversion tool which just performs a copy of the content.
4040
This is slow but quite reliable.
4043
def __init__(self, target_format):
4044
"""Create a CopyConverter.
4046
:param target_format: The format the resulting repository should be.
4048
self.target_format = target_format
4050
def convert(self, repo, pb):
4051
"""Perform the conversion of to_convert, giving feedback via pb.
4053
:param to_convert: The disk object to convert.
4054
:param pb: a progress bar to use for progress information.
4059
# this is only useful with metadir layouts - separated repo content.
4060
# trigger an assertion if not such
4061
repo._format.get_format_string()
4062
self.repo_dir = repo.bzrdir
4063
self.step('Moving repository to repository.backup')
4064
self.repo_dir.transport.move('repository', 'repository.backup')
4065
backup_transport = self.repo_dir.transport.clone('repository.backup')
4066
repo._format.check_conversion_target(self.target_format)
4067
self.source_repo = repo._format.open(self.repo_dir,
4069
_override_transport=backup_transport)
4070
self.step('Creating new repository')
4071
converted = self.target_format.initialize(self.repo_dir,
4072
self.source_repo.is_shared())
4073
converted.lock_write()
4075
self.step('Copying content into repository.')
4076
self.source_repo.copy_content_into(converted)
4079
self.step('Deleting old repository content.')
4080
self.repo_dir.transport.delete_tree('repository.backup')
4081
self.pb.note('repository converted')
4083
def step(self, message):
4084
"""Update the pb by a step."""
4086
self.pb.update(message, self.count, self.total)
4098
def _unescaper(match, _map=_unescape_map):
4099
code = match.group(1)
4103
if not code.startswith('#'):
4105
return unichr(int(code[1:])).encode('utf8')
4111
def _unescape_xml(data):
4112
"""Unescape predefined XML entities in a string of data."""
4114
if _unescape_re is None:
4115
_unescape_re = re.compile('\&([^;]*);')
4116
return _unescape_re.sub(_unescaper, data)
4119
class _VersionedFileChecker(object):
4121
def __init__(self, repository, text_key_references=None, ancestors=None):
4122
self.repository = repository
4123
self.text_index = self.repository._generate_text_key_index(
4124
text_key_references=text_key_references, ancestors=ancestors)
4126
def calculate_file_version_parents(self, text_key):
4127
"""Calculate the correct parents for a file version according to
4130
parent_keys = self.text_index[text_key]
4131
if parent_keys == [_mod_revision.NULL_REVISION]:
4133
return tuple(parent_keys)
4135
def check_file_version_parents(self, texts, progress_bar=None):
4136
"""Check the parents stored in a versioned file are correct.
4138
It also detects file versions that are not referenced by their
4139
corresponding revision's inventory.
4141
:returns: A tuple of (wrong_parents, dangling_file_versions).
4142
wrong_parents is a dict mapping {revision_id: (stored_parents,
4143
correct_parents)} for each revision_id where the stored parents
4144
are not correct. dangling_file_versions is a set of (file_id,
4145
revision_id) tuples for versions that are present in this versioned
4146
file, but not used by the corresponding inventory.
4148
local_progress = None
4149
if progress_bar is None:
4150
local_progress = ui.ui_factory.nested_progress_bar()
4151
progress_bar = local_progress
4153
return self._check_file_version_parents(texts, progress_bar)
4156
local_progress.finished()
4158
def _check_file_version_parents(self, texts, progress_bar):
4159
"""See check_file_version_parents."""
4161
self.file_ids = set([file_id for file_id, _ in
4162
self.text_index.iterkeys()])
4163
# text keys is now grouped by file_id
4164
n_versions = len(self.text_index)
4165
progress_bar.update('loading text store', 0, n_versions)
4166
parent_map = self.repository.texts.get_parent_map(self.text_index)
4167
# On unlistable transports this could well be empty/error...
4168
text_keys = self.repository.texts.keys()
4169
unused_keys = frozenset(text_keys) - set(self.text_index)
4170
for num, key in enumerate(self.text_index.iterkeys()):
4171
progress_bar.update('checking text graph', num, n_versions)
4172
correct_parents = self.calculate_file_version_parents(key)
4174
knit_parents = parent_map[key]
4175
except errors.RevisionNotPresent:
4178
if correct_parents != knit_parents:
4179
wrong_parents[key] = (knit_parents, correct_parents)
4180
return wrong_parents, unused_keys
4183
def _old_get_graph(repository, revision_id):
4184
"""DO NOT USE. That is all. I'm serious."""
4185
graph = repository.get_graph()
4186
revision_graph = dict(((key, value) for key, value in
4187
graph.iter_ancestry([revision_id]) if value is not None))
4188
return _strip_NULL_ghosts(revision_graph)
4191
def _strip_NULL_ghosts(revision_graph):
4192
"""Also don't use this. more compatibility code for unmigrated clients."""
4193
# Filter ghosts, and null:
4194
if _mod_revision.NULL_REVISION in revision_graph:
4195
del revision_graph[_mod_revision.NULL_REVISION]
4196
for key, parents in revision_graph.items():
4197
revision_graph[key] = tuple(parent for parent in parents if parent
4199
return revision_graph
4202
class StreamSink(object):
4203
"""An object that can insert a stream into a repository.
4205
This interface handles the complexity of reserialising inventories and
4206
revisions from different formats, and allows unidirectional insertion into
4207
stacked repositories without looking for the missing basis parents
4211
def __init__(self, target_repo):
4212
self.target_repo = target_repo
4214
def insert_stream(self, stream, src_format, resume_tokens):
4215
"""Insert a stream's content into the target repository.
4217
:param src_format: a bzr repository format.
4219
:return: a list of resume tokens and an iterable of keys additional
4220
items required before the insertion can be completed.
4222
self.target_repo.lock_write()
4225
self.target_repo.resume_write_group(resume_tokens)
4228
self.target_repo.start_write_group()
4231
# locked_insert_stream performs a commit|suspend.
4232
return self._locked_insert_stream(stream, src_format, is_resume)
4234
self.target_repo.abort_write_group(suppress_errors=True)
4237
self.target_repo.unlock()
4239
def _locked_insert_stream(self, stream, src_format, is_resume):
4240
to_serializer = self.target_repo._format._serializer
4241
src_serializer = src_format._serializer
4243
if to_serializer == src_serializer:
4244
# If serializers match and the target is a pack repository, set the
4245
# write cache size on the new pack. This avoids poor performance
4246
# on transports where append is unbuffered (such as
4247
# RemoteTransport). This is safe to do because nothing should read
4248
# back from the target repository while a stream with matching
4249
# serialization is being inserted.
4250
# The exception is that a delta record from the source that should
4251
# be a fulltext may need to be expanded by the target (see
4252
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4253
# explicitly flush any buffered writes first in that rare case.
4255
new_pack = self.target_repo._pack_collection._new_pack
4256
except AttributeError:
4257
# Not a pack repository
4260
new_pack.set_write_cache_size(1024*1024)
4261
for substream_type, substream in stream:
4262
if 'stream' in debug.debug_flags:
4263
mutter('inserting substream: %s', substream_type)
4264
if substream_type == 'texts':
4265
self.target_repo.texts.insert_record_stream(substream)
4266
elif substream_type == 'inventories':
4267
if src_serializer == to_serializer:
4268
self.target_repo.inventories.insert_record_stream(
4271
self._extract_and_insert_inventories(
4272
substream, src_serializer)
4273
elif substream_type == 'inventory-deltas':
4274
self._extract_and_insert_inventory_deltas(
4275
substream, src_serializer)
4276
elif substream_type == 'chk_bytes':
4277
# XXX: This doesn't support conversions, as it assumes the
4278
# conversion was done in the fetch code.
4279
self.target_repo.chk_bytes.insert_record_stream(substream)
4280
elif substream_type == 'revisions':
4281
# This may fallback to extract-and-insert more often than
4282
# required if the serializers are different only in terms of
4284
if src_serializer == to_serializer:
4285
self.target_repo.revisions.insert_record_stream(
4288
self._extract_and_insert_revisions(substream,
4290
elif substream_type == 'signatures':
4291
self.target_repo.signatures.insert_record_stream(substream)
4293
raise AssertionError('kaboom! %s' % (substream_type,))
4294
# Done inserting data, and the missing_keys calculations will try to
4295
# read back from the inserted data, so flush the writes to the new pack
4296
# (if this is pack format).
4297
if new_pack is not None:
4298
new_pack._write_data('', flush=True)
4299
# Find all the new revisions (including ones from resume_tokens)
4300
missing_keys = self.target_repo.get_missing_parent_inventories(
4301
check_for_missing_texts=is_resume)
4303
for prefix, versioned_file in (
4304
('texts', self.target_repo.texts),
4305
('inventories', self.target_repo.inventories),
4306
('revisions', self.target_repo.revisions),
4307
('signatures', self.target_repo.signatures),
4308
('chk_bytes', self.target_repo.chk_bytes),
4310
if versioned_file is None:
4312
missing_keys.update((prefix,) + key for key in
4313
versioned_file.get_missing_compression_parent_keys())
4314
except NotImplementedError:
4315
# cannot even attempt suspending, and missing would have failed
4316
# during stream insertion.
4317
missing_keys = set()
4320
# suspend the write group and tell the caller what we is
4321
# missing. We know we can suspend or else we would not have
4322
# entered this code path. (All repositories that can handle
4323
# missing keys can handle suspending a write group).
4324
write_group_tokens = self.target_repo.suspend_write_group()
4325
return write_group_tokens, missing_keys
4326
hint = self.target_repo.commit_write_group()
4327
if (to_serializer != src_serializer and
4328
self.target_repo._format.pack_compresses):
4329
self.target_repo.pack(hint=hint)
4332
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4333
target_rich_root = self.target_repo._format.rich_root_data
4334
target_tree_refs = self.target_repo._format.supports_tree_reference
4335
for record in substream:
4336
# Insert the delta directly
4337
inventory_delta_bytes = record.get_bytes_as('fulltext')
4338
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4340
parse_result = deserialiser.parse_text_bytes(
4341
inventory_delta_bytes)
4342
except inventory_delta.IncompatibleInventoryDelta, err:
4343
trace.mutter("Incompatible delta: %s", err.msg)
4344
raise errors.IncompatibleRevision(self.target_repo._format)
4345
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4346
revision_id = new_id
4347
parents = [key[0] for key in record.parents]
4348
self.target_repo.add_inventory_by_delta(
4349
basis_id, inv_delta, revision_id, parents)
4351
def _extract_and_insert_inventories(self, substream, serializer,
4353
"""Generate a new inventory versionedfile in target, converting data.
4355
The inventory is retrieved from the source, (deserializing it), and
4356
stored in the target (reserializing it in a different format).
4358
target_rich_root = self.target_repo._format.rich_root_data
4359
target_tree_refs = self.target_repo._format.supports_tree_reference
4360
for record in substream:
4361
# It's not a delta, so it must be a fulltext in the source
4362
# serializer's format.
4363
bytes = record.get_bytes_as('fulltext')
4364
revision_id = record.key[0]
4365
inv = serializer.read_inventory_from_string(bytes, revision_id)
4366
parents = [key[0] for key in record.parents]
4367
self.target_repo.add_inventory(revision_id, inv, parents)
4368
# No need to keep holding this full inv in memory when the rest of
4369
# the substream is likely to be all deltas.
4372
def _extract_and_insert_revisions(self, substream, serializer):
4373
for record in substream:
4374
bytes = record.get_bytes_as('fulltext')
4375
revision_id = record.key[0]
4376
rev = serializer.read_revision_from_string(bytes)
4377
if rev.revision_id != revision_id:
4378
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4379
self.target_repo.add_revision(revision_id, rev)
4382
if self.target_repo._format._fetch_reconcile:
4383
self.target_repo.reconcile()
4386
class StreamSource(object):
4387
"""A source of a stream for fetching between repositories."""
4389
def __init__(self, from_repository, to_format):
4390
"""Create a StreamSource streaming from from_repository."""
4391
self.from_repository = from_repository
4392
self.to_format = to_format
4394
def delta_on_metadata(self):
4395
"""Return True if delta's are permitted on metadata streams.
4397
That is on revisions and signatures.
4399
src_serializer = self.from_repository._format._serializer
4400
target_serializer = self.to_format._serializer
4401
return (self.to_format._fetch_uses_deltas and
4402
src_serializer == target_serializer)
4404
def _fetch_revision_texts(self, revs):
4405
# fetch signatures first and then the revision texts
4406
# may need to be a InterRevisionStore call here.
4407
from_sf = self.from_repository.signatures
4408
# A missing signature is just skipped.
4409
keys = [(rev_id,) for rev_id in revs]
4410
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4412
self.to_format._fetch_order,
4413
not self.to_format._fetch_uses_deltas))
4414
# If a revision has a delta, this is actually expanded inside the
4415
# insert_record_stream code now, which is an alternate fix for
4417
from_rf = self.from_repository.revisions
4418
revisions = from_rf.get_record_stream(
4420
self.to_format._fetch_order,
4421
not self.delta_on_metadata())
4422
return [('signatures', signatures), ('revisions', revisions)]
4424
def _generate_root_texts(self, revs):
4425
"""This will be called by get_stream between fetching weave texts and
4426
fetching the inventory weave.
4428
if self._rich_root_upgrade():
4430
return bzrlib.fetch.Inter1and2Helper(
4431
self.from_repository).generate_root_texts(revs)
4435
def get_stream(self, search):
4437
revs = search.get_keys()
4438
graph = self.from_repository.get_graph()
4439
revs = tsort.topo_sort(graph.get_parent_map(revs))
4440
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4442
for knit_kind, file_id, revisions in data_to_fetch:
4443
if knit_kind != phase:
4445
# Make a new progress bar for this phase
4446
if knit_kind == "file":
4447
# Accumulate file texts
4448
text_keys.extend([(file_id, revision) for revision in
4450
elif knit_kind == "inventory":
4451
# Now copy the file texts.
4452
from_texts = self.from_repository.texts
4453
yield ('texts', from_texts.get_record_stream(
4454
text_keys, self.to_format._fetch_order,
4455
not self.to_format._fetch_uses_deltas))
4456
# Cause an error if a text occurs after we have done the
4459
# Before we process the inventory we generate the root
4460
# texts (if necessary) so that the inventories references
4462
for _ in self._generate_root_texts(revs):
4464
# we fetch only the referenced inventories because we do not
4465
# know for unselected inventories whether all their required
4466
# texts are present in the other repository - it could be
4468
for info in self._get_inventory_stream(revs):
4470
elif knit_kind == "signatures":
4471
# Nothing to do here; this will be taken care of when
4472
# _fetch_revision_texts happens.
4474
elif knit_kind == "revisions":
4475
for record in self._fetch_revision_texts(revs):
4478
raise AssertionError("Unknown knit kind %r" % knit_kind)
4480
def get_stream_for_missing_keys(self, missing_keys):
4481
# missing keys can only occur when we are byte copying and not
4482
# translating (because translation means we don't send
4483
# unreconstructable deltas ever).
4485
keys['texts'] = set()
4486
keys['revisions'] = set()
4487
keys['inventories'] = set()
4488
keys['chk_bytes'] = set()
4489
keys['signatures'] = set()
4490
for key in missing_keys:
4491
keys[key[0]].add(key[1:])
4492
if len(keys['revisions']):
4493
# If we allowed copying revisions at this point, we could end up
4494
# copying a revision without copying its required texts: a
4495
# violation of the requirements for repository integrity.
4496
raise AssertionError(
4497
'cannot copy revisions to fill in missing deltas %s' % (
4498
keys['revisions'],))
4499
for substream_kind, keys in keys.iteritems():
4500
vf = getattr(self.from_repository, substream_kind)
4501
if vf is None and keys:
4502
raise AssertionError(
4503
"cannot fill in keys for a versioned file we don't"
4504
" have: %s needs %s" % (substream_kind, keys))
4506
# No need to stream something we don't have
4508
if substream_kind == 'inventories':
4509
# Some missing keys are genuinely ghosts, filter those out.
4510
present = self.from_repository.inventories.get_parent_map(keys)
4511
revs = [key[0] for key in present]
4512
# Get the inventory stream more-or-less as we do for the
4513
# original stream; there's no reason to assume that records
4514
# direct from the source will be suitable for the sink. (Think
4515
# e.g. 2a -> 1.9-rich-root).
4516
for info in self._get_inventory_stream(revs, missing=True):
4520
# Ask for full texts always so that we don't need more round trips
4521
# after this stream.
4522
# Some of the missing keys are genuinely ghosts, so filter absent
4523
# records. The Sink is responsible for doing another check to
4524
# ensure that ghosts don't introduce missing data for future
4526
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4527
self.to_format._fetch_order, True))
4528
yield substream_kind, stream
4530
def inventory_fetch_order(self):
4531
if self._rich_root_upgrade():
4532
return 'topological'
4534
return self.to_format._fetch_order
4536
def _rich_root_upgrade(self):
4537
return (not self.from_repository._format.rich_root_data and
4538
self.to_format.rich_root_data)
4540
def _get_inventory_stream(self, revision_ids, missing=False):
4541
from_format = self.from_repository._format
4542
if (from_format.supports_chks and self.to_format.supports_chks and
4543
from_format.network_name() == self.to_format.network_name()):
4544
raise AssertionError(
4545
"this case should be handled by GroupCHKStreamSource")
4546
elif 'forceinvdeltas' in debug.debug_flags:
4547
return self._get_convertable_inventory_stream(revision_ids,
4548
delta_versus_null=missing)
4549
elif from_format.network_name() == self.to_format.network_name():
4551
return self._get_simple_inventory_stream(revision_ids,
4553
elif (not from_format.supports_chks and not self.to_format.supports_chks
4554
and from_format._serializer == self.to_format._serializer):
4555
# Essentially the same format.
4556
return self._get_simple_inventory_stream(revision_ids,
4559
# Any time we switch serializations, we want to use an
4560
# inventory-delta based approach.
4561
return self._get_convertable_inventory_stream(revision_ids,
4562
delta_versus_null=missing)
4564
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4565
# NB: This currently reopens the inventory weave in source;
4566
# using a single stream interface instead would avoid this.
4567
from_weave = self.from_repository.inventories
4569
delta_closure = True
4571
delta_closure = not self.delta_on_metadata()
4572
yield ('inventories', from_weave.get_record_stream(
4573
[(rev_id,) for rev_id in revision_ids],
4574
self.inventory_fetch_order(), delta_closure))
4576
def _get_convertable_inventory_stream(self, revision_ids,
4577
delta_versus_null=False):
4578
# The source is using CHKs, but the target either doesn't or it has a
4579
# different serializer. The StreamSink code expects to be able to
4580
# convert on the target, so we need to put bytes-on-the-wire that can
4581
# be converted. That means inventory deltas (if the remote is <1.19,
4582
# RemoteStreamSink will fallback to VFS to insert the deltas).
4583
yield ('inventory-deltas',
4584
self._stream_invs_as_deltas(revision_ids,
4585
delta_versus_null=delta_versus_null))
4587
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4588
"""Return a stream of inventory-deltas for the given rev ids.
4590
:param revision_ids: The list of inventories to transmit
4591
:param delta_versus_null: Don't try to find a minimal delta for this
4592
entry, instead compute the delta versus the NULL_REVISION. This
4593
effectively streams a complete inventory. Used for stuff like
4594
filling in missing parents, etc.
4596
from_repo = self.from_repository
4597
revision_keys = [(rev_id,) for rev_id in revision_ids]
4598
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4599
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4601
inventories = self.from_repository.iter_inventories(
4602
revision_ids, 'topological')
4603
format = from_repo._format
4604
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4605
inventory_cache = lru_cache.LRUCache(50)
4606
null_inventory = from_repo.revision_tree(
4607
_mod_revision.NULL_REVISION).inventory
4608
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4609
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4610
# repo back into a non-rich-root repo ought to be allowed)
4611
serializer = inventory_delta.InventoryDeltaSerializer(
4612
versioned_root=format.rich_root_data,
4613
tree_references=format.supports_tree_reference)
4614
for inv in inventories:
4615
key = (inv.revision_id,)
4616
parent_keys = parent_map.get(key, ())
4618
if not delta_versus_null and parent_keys:
4619
# The caller did not ask for complete inventories and we have
4620
# some parents that we can delta against. Make a delta against
4621
# each parent so that we can find the smallest.
4622
parent_ids = [parent_key[0] for parent_key in parent_keys]
4623
for parent_id in parent_ids:
4624
if parent_id not in invs_sent_so_far:
4625
# We don't know that the remote side has this basis, so
4628
if parent_id == _mod_revision.NULL_REVISION:
4629
parent_inv = null_inventory
4631
parent_inv = inventory_cache.get(parent_id, None)
4632
if parent_inv is None:
4633
parent_inv = from_repo.get_inventory(parent_id)
4634
candidate_delta = inv._make_delta(parent_inv)
4635
if (delta is None or
4636
len(delta) > len(candidate_delta)):
4637
delta = candidate_delta
4638
basis_id = parent_id
4640
# Either none of the parents ended up being suitable, or we
4641
# were asked to delta against NULL
4642
basis_id = _mod_revision.NULL_REVISION
4643
delta = inv._make_delta(null_inventory)
4644
invs_sent_so_far.add(inv.revision_id)
4645
inventory_cache[inv.revision_id] = inv
4646
delta_serialized = ''.join(
4647
serializer.delta_to_lines(basis_id, key[-1], delta))
4648
yield versionedfile.FulltextContentFactory(
4649
key, parent_keys, None, delta_serialized)
4652
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4653
stop_revision=None):
4654
"""Extend the partial history to include a given index
4656
If a stop_index is supplied, stop when that index has been reached.
4657
If a stop_revision is supplied, stop when that revision is
4658
encountered. Otherwise, stop when the beginning of history is
4661
:param stop_index: The index which should be present. When it is
4662
present, history extension will stop.
4663
:param stop_revision: The revision id which should be present. When
4664
it is encountered, history extension will stop.
4666
start_revision = partial_history_cache[-1]
4667
iterator = repo.iter_reverse_revision_history(start_revision)
4669
#skip the last revision in the list
4672
if (stop_index is not None and
4673
len(partial_history_cache) > stop_index):
4675
if partial_history_cache[-1] == stop_revision:
4677
revision_id = iterator.next()
4678
partial_history_cache.append(revision_id)
4679
except StopIteration:
added
removed
bzrlib/repository.py
