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(), """
41
revision as _mod_revision,
47
from bzrlib.bundle import serializer
48
from bzrlib.revisiontree import RevisionTree
49
from bzrlib.store.versioned import VersionedFileStore
50
from bzrlib.testament import Testament
53
from bzrlib.decorators import needs_read_lock, needs_write_lock, only_raises
54
from bzrlib.inter import InterObject
55
from bzrlib.inventory import (
61
from bzrlib.lock import _RelockDebugMixin
62
from bzrlib import registry
63
from bzrlib.trace import (
64
log_exception_quietly, note, mutter, mutter_callsite, warning)
67
# Old formats display a warning, but only once
68
_deprecation_warning_done = False
71
class CommitBuilder(object):
72
"""Provides an interface to build up a commit.
74
This allows describing a tree to be committed without needing to
75
know the internals of the format of the repository.
78
# all clients should supply tree roots.
79
record_root_entry = True
80
# the default CommitBuilder does not manage trees whose root is versioned.
81
_versioned_root = False
83
def __init__(self, repository, parents, config, timestamp=None,
84
timezone=None, committer=None, revprops=None,
86
"""Initiate a CommitBuilder.
88
:param repository: Repository to commit to.
89
:param parents: Revision ids of the parents of the new revision.
90
:param config: Configuration to use.
91
:param timestamp: Optional timestamp recorded for commit.
92
:param timezone: Optional timezone for timestamp.
93
:param committer: Optional committer to set for commit.
94
:param revprops: Optional dictionary of revision properties.
95
:param revision_id: Optional revision id.
100
self._committer = self._config.username()
102
self._committer = committer
104
self.new_inventory = Inventory(None)
105
self._new_revision_id = revision_id
106
self.parents = parents
107
self.repository = repository
110
if revprops is not None:
111
self._validate_revprops(revprops)
112
self._revprops.update(revprops)
114
if timestamp is None:
115
timestamp = time.time()
116
# Restrict resolution to 1ms
117
self._timestamp = round(timestamp, 3)
120
self._timezone = osutils.local_time_offset()
122
self._timezone = int(timezone)
124
self._generate_revision_if_needed()
125
self.__heads = graph.HeadsCache(repository.get_graph()).heads
126
self._basis_delta = []
127
# API compatibility, older code that used CommitBuilder did not call
128
# .record_delete(), which means the delta that is computed would not be
129
# valid. Callers that will call record_delete() should call
130
# .will_record_deletes() to indicate that.
131
self._recording_deletes = False
132
# memo'd check for no-op commits.
133
self._any_changes = False
135
def any_changes(self):
136
"""Return True if any entries were changed.
138
This includes merge-only changes. It is the core for the --unchanged
141
:return: True if any changes have occured.
143
return self._any_changes
145
def _validate_unicode_text(self, text, context):
146
"""Verify things like commit messages don't have bogus characters."""
148
raise ValueError('Invalid value for %s: %r' % (context, text))
150
def _validate_revprops(self, revprops):
151
for key, value in revprops.iteritems():
152
# We know that the XML serializers do not round trip '\r'
153
# correctly, so refuse to accept them
154
if not isinstance(value, basestring):
155
raise ValueError('revision property (%s) is not a valid'
156
' (unicode) string: %r' % (key, value))
157
self._validate_unicode_text(value,
158
'revision property (%s)' % (key,))
160
def commit(self, message):
161
"""Make the actual commit.
163
:return: The revision id of the recorded revision.
165
self._validate_unicode_text(message, 'commit message')
166
rev = _mod_revision.Revision(
167
timestamp=self._timestamp,
168
timezone=self._timezone,
169
committer=self._committer,
171
inventory_sha1=self.inv_sha1,
172
revision_id=self._new_revision_id,
173
properties=self._revprops)
174
rev.parent_ids = self.parents
175
self.repository.add_revision(self._new_revision_id, rev,
176
self.new_inventory, self._config)
177
self.repository.commit_write_group()
178
return self._new_revision_id
181
"""Abort the commit that is being built.
183
self.repository.abort_write_group()
185
def revision_tree(self):
186
"""Return the tree that was just committed.
188
After calling commit() this can be called to get a RevisionTree
189
representing the newly committed tree. This is preferred to
190
calling Repository.revision_tree() because that may require
191
deserializing the inventory, while we already have a copy in
194
if self.new_inventory is None:
195
self.new_inventory = self.repository.get_inventory(
196
self._new_revision_id)
197
return RevisionTree(self.repository, self.new_inventory,
198
self._new_revision_id)
200
def finish_inventory(self):
201
"""Tell the builder that the inventory is finished.
203
:return: The inventory id in the repository, which can be used with
204
repository.get_inventory.
206
if self.new_inventory is None:
207
# an inventory delta was accumulated without creating a new
209
basis_id = self.basis_delta_revision
210
# We ignore the 'inventory' returned by add_inventory_by_delta
211
# because self.new_inventory is used to hint to the rest of the
212
# system what code path was taken
213
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
214
basis_id, self._basis_delta, self._new_revision_id,
217
if self.new_inventory.root is None:
218
raise AssertionError('Root entry should be supplied to'
219
' record_entry_contents, as of bzr 0.10.')
220
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
221
self.new_inventory.revision_id = self._new_revision_id
222
self.inv_sha1 = self.repository.add_inventory(
223
self._new_revision_id,
227
return self._new_revision_id
229
def _gen_revision_id(self):
230
"""Return new revision-id."""
231
return generate_ids.gen_revision_id(self._config.username(),
234
def _generate_revision_if_needed(self):
235
"""Create a revision id if None was supplied.
237
If the repository can not support user-specified revision ids
238
they should override this function and raise CannotSetRevisionId
239
if _new_revision_id is not None.
241
:raises: CannotSetRevisionId
243
if self._new_revision_id is None:
244
self._new_revision_id = self._gen_revision_id()
245
self.random_revid = True
247
self.random_revid = False
249
def _heads(self, file_id, revision_ids):
250
"""Calculate the graph heads for revision_ids in the graph of file_id.
252
This can use either a per-file graph or a global revision graph as we
253
have an identity relationship between the two graphs.
255
return self.__heads(revision_ids)
257
def _check_root(self, ie, parent_invs, tree):
258
"""Helper for record_entry_contents.
260
:param ie: An entry being added.
261
:param parent_invs: The inventories of the parent revisions of the
263
:param tree: The tree that is being committed.
265
# In this revision format, root entries have no knit or weave When
266
# serializing out to disk and back in root.revision is always
268
ie.revision = self._new_revision_id
270
def _require_root_change(self, tree):
271
"""Enforce an appropriate root object change.
273
This is called once when record_iter_changes is called, if and only if
274
the root was not in the delta calculated by record_iter_changes.
276
:param tree: The tree which is being committed.
278
# NB: if there are no parents then this method is not called, so no
279
# need to guard on parents having length.
280
entry = entry_factory['directory'](tree.path2id(''), '',
282
entry.revision = self._new_revision_id
283
self._basis_delta.append(('', '', entry.file_id, entry))
285
def _get_delta(self, ie, basis_inv, path):
286
"""Get a delta against the basis inventory for ie."""
287
if ie.file_id not in basis_inv:
289
result = (None, path, ie.file_id, ie)
290
self._basis_delta.append(result)
292
elif ie != basis_inv[ie.file_id]:
294
# TODO: avoid tis id2path call.
295
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
296
self._basis_delta.append(result)
302
def get_basis_delta(self):
303
"""Return the complete inventory delta versus the basis inventory.
305
This has been built up with the calls to record_delete and
306
record_entry_contents. The client must have already called
307
will_record_deletes() to indicate that they will be generating a
310
:return: An inventory delta, suitable for use with apply_delta, or
311
Repository.add_inventory_by_delta, etc.
313
if not self._recording_deletes:
314
raise AssertionError("recording deletes not activated.")
315
return self._basis_delta
317
def record_delete(self, path, file_id):
318
"""Record that a delete occured against a basis tree.
320
This is an optional API - when used it adds items to the basis_delta
321
being accumulated by the commit builder. It cannot be called unless the
322
method will_record_deletes() has been called to inform the builder that
323
a delta is being supplied.
325
:param path: The path of the thing deleted.
326
:param file_id: The file id that was deleted.
328
if not self._recording_deletes:
329
raise AssertionError("recording deletes not activated.")
330
delta = (path, None, file_id, None)
331
self._basis_delta.append(delta)
332
self._any_changes = True
335
def will_record_deletes(self):
336
"""Tell the commit builder that deletes are being notified.
338
This enables the accumulation of an inventory delta; for the resulting
339
commit to be valid, deletes against the basis MUST be recorded via
340
builder.record_delete().
342
self._recording_deletes = True
344
basis_id = self.parents[0]
346
basis_id = _mod_revision.NULL_REVISION
347
self.basis_delta_revision = basis_id
349
def record_entry_contents(self, ie, parent_invs, path, tree,
351
"""Record the content of ie from tree into the commit if needed.
353
Side effect: sets ie.revision when unchanged
355
:param ie: An inventory entry present in the commit.
356
:param parent_invs: The inventories of the parent revisions of the
358
:param path: The path the entry is at in the tree.
359
:param tree: The tree which contains this entry and should be used to
361
:param content_summary: Summary data from the tree about the paths
362
content - stat, length, exec, sha/link target. This is only
363
accessed when the entry has a revision of None - that is when it is
364
a candidate to commit.
365
:return: A tuple (change_delta, version_recorded, fs_hash).
366
change_delta is an inventory_delta change for this entry against
367
the basis tree of the commit, or None if no change occured against
369
version_recorded is True if a new version of the entry has been
370
recorded. For instance, committing a merge where a file was only
371
changed on the other side will return (delta, False).
372
fs_hash is either None, or the hash details for the path (currently
373
a tuple of the contents sha1 and the statvalue returned by
374
tree.get_file_with_stat()).
376
if self.new_inventory.root is None:
377
if ie.parent_id is not None:
378
raise errors.RootMissing()
379
self._check_root(ie, parent_invs, tree)
380
if ie.revision is None:
381
kind = content_summary[0]
383
# ie is carried over from a prior commit
385
# XXX: repository specific check for nested tree support goes here - if
386
# the repo doesn't want nested trees we skip it ?
387
if (kind == 'tree-reference' and
388
not self.repository._format.supports_tree_reference):
389
# mismatch between commit builder logic and repository:
390
# this needs the entry creation pushed down into the builder.
391
raise NotImplementedError('Missing repository subtree support.')
392
self.new_inventory.add(ie)
394
# TODO: slow, take it out of the inner loop.
396
basis_inv = parent_invs[0]
398
basis_inv = Inventory(root_id=None)
400
# ie.revision is always None if the InventoryEntry is considered
401
# for committing. We may record the previous parents revision if the
402
# content is actually unchanged against a sole head.
403
if ie.revision is not None:
404
if not self._versioned_root and path == '':
405
# repositories that do not version the root set the root's
406
# revision to the new commit even when no change occurs (more
407
# specifically, they do not record a revision on the root; and
408
# the rev id is assigned to the root during deserialisation -
409
# this masks when a change may have occurred against the basis.
410
# To match this we always issue a delta, because the revision
411
# of the root will always be changing.
412
if ie.file_id in basis_inv:
413
delta = (basis_inv.id2path(ie.file_id), path,
417
delta = (None, path, ie.file_id, ie)
418
self._basis_delta.append(delta)
419
return delta, False, None
421
# we don't need to commit this, because the caller already
422
# determined that an existing revision of this file is
423
# appropriate. If its not being considered for committing then
424
# it and all its parents to the root must be unaltered so
425
# no-change against the basis.
426
if ie.revision == self._new_revision_id:
427
raise AssertionError("Impossible situation, a skipped "
428
"inventory entry (%r) claims to be modified in this "
429
"commit (%r).", (ie, self._new_revision_id))
430
return None, False, None
431
# XXX: Friction: parent_candidates should return a list not a dict
432
# so that we don't have to walk the inventories again.
433
parent_candiate_entries = ie.parent_candidates(parent_invs)
434
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
436
for inv in parent_invs:
437
if ie.file_id in inv:
438
old_rev = inv[ie.file_id].revision
439
if old_rev in head_set:
440
heads.append(inv[ie.file_id].revision)
441
head_set.remove(inv[ie.file_id].revision)
444
# now we check to see if we need to write a new record to the
446
# We write a new entry unless there is one head to the ancestors, and
447
# the kind-derived content is unchanged.
449
# Cheapest check first: no ancestors, or more the one head in the
450
# ancestors, we write a new node.
454
# There is a single head, look it up for comparison
455
parent_entry = parent_candiate_entries[heads[0]]
456
# if the non-content specific data has changed, we'll be writing a
458
if (parent_entry.parent_id != ie.parent_id or
459
parent_entry.name != ie.name):
461
# now we need to do content specific checks:
463
# if the kind changed the content obviously has
464
if kind != parent_entry.kind:
466
# Stat cache fingerprint feedback for the caller - None as we usually
467
# don't generate one.
470
if content_summary[2] is None:
471
raise ValueError("Files must not have executable = None")
473
# We can't trust a check of the file length because of content
475
if (# if the exec bit has changed we have to store:
476
parent_entry.executable != content_summary[2]):
478
elif parent_entry.text_sha1 == content_summary[3]:
479
# all meta and content is unchanged (using a hash cache
480
# hit to check the sha)
481
ie.revision = parent_entry.revision
482
ie.text_size = parent_entry.text_size
483
ie.text_sha1 = parent_entry.text_sha1
484
ie.executable = parent_entry.executable
485
return self._get_delta(ie, basis_inv, path), False, None
487
# Either there is only a hash change(no hash cache entry,
488
# or same size content change), or there is no change on
490
# Provide the parent's hash to the store layer, so that the
491
# content is unchanged we will not store a new node.
492
nostore_sha = parent_entry.text_sha1
494
# We want to record a new node regardless of the presence or
495
# absence of a content change in the file.
497
ie.executable = content_summary[2]
498
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
500
text = file_obj.read()
504
ie.text_sha1, ie.text_size = self._add_text_to_weave(
505
ie.file_id, text, heads, nostore_sha)
506
# Let the caller know we generated a stat fingerprint.
507
fingerprint = (ie.text_sha1, stat_value)
508
except errors.ExistingContent:
509
# Turns out that the file content was unchanged, and we were
510
# only going to store a new node if it was changed. Carry over
512
ie.revision = parent_entry.revision
513
ie.text_size = parent_entry.text_size
514
ie.text_sha1 = parent_entry.text_sha1
515
ie.executable = parent_entry.executable
516
return self._get_delta(ie, basis_inv, path), False, None
517
elif kind == 'directory':
519
# all data is meta here, nothing specific to directory, so
521
ie.revision = parent_entry.revision
522
return self._get_delta(ie, basis_inv, path), False, None
523
self._add_text_to_weave(ie.file_id, '', heads, None)
524
elif kind == 'symlink':
525
current_link_target = content_summary[3]
527
# symlink target is not generic metadata, check if it has
529
if current_link_target != parent_entry.symlink_target:
532
# unchanged, carry over.
533
ie.revision = parent_entry.revision
534
ie.symlink_target = parent_entry.symlink_target
535
return self._get_delta(ie, basis_inv, path), False, None
536
ie.symlink_target = current_link_target
537
self._add_text_to_weave(ie.file_id, '', heads, None)
538
elif kind == 'tree-reference':
540
if content_summary[3] != parent_entry.reference_revision:
543
# unchanged, carry over.
544
ie.reference_revision = parent_entry.reference_revision
545
ie.revision = parent_entry.revision
546
return self._get_delta(ie, basis_inv, path), False, None
547
ie.reference_revision = content_summary[3]
548
if ie.reference_revision is None:
549
raise AssertionError("invalid content_summary for nested tree: %r"
550
% (content_summary,))
551
self._add_text_to_weave(ie.file_id, '', heads, None)
553
raise NotImplementedError('unknown kind')
554
ie.revision = self._new_revision_id
555
self._any_changes = True
556
return self._get_delta(ie, basis_inv, path), True, fingerprint
558
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
559
_entry_factory=entry_factory):
560
"""Record a new tree via iter_changes.
562
:param tree: The tree to obtain text contents from for changed objects.
563
:param basis_revision_id: The revision id of the tree the iter_changes
564
has been generated against. Currently assumed to be the same
565
as self.parents[0] - if it is not, errors may occur.
566
:param iter_changes: An iter_changes iterator with the changes to apply
567
to basis_revision_id. The iterator must not include any items with
568
a current kind of None - missing items must be either filtered out
569
or errored-on beefore record_iter_changes sees the item.
570
:param _entry_factory: Private method to bind entry_factory locally for
572
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
575
# Create an inventory delta based on deltas between all the parents and
576
# deltas between all the parent inventories. We use inventory delta's
577
# between the inventory objects because iter_changes masks
578
# last-changed-field only changes.
580
# file_id -> change map, change is fileid, paths, changed, versioneds,
581
# parents, names, kinds, executables
583
# {file_id -> revision_id -> inventory entry, for entries in parent
584
# trees that are not parents[0]
588
revtrees = list(self.repository.revision_trees(self.parents))
589
except errors.NoSuchRevision:
590
# one or more ghosts, slow path.
592
for revision_id in self.parents:
594
revtrees.append(self.repository.revision_tree(revision_id))
595
except errors.NoSuchRevision:
597
basis_revision_id = _mod_revision.NULL_REVISION
599
revtrees.append(self.repository.revision_tree(
600
_mod_revision.NULL_REVISION))
601
# The basis inventory from a repository
603
basis_inv = revtrees[0].inventory
605
basis_inv = self.repository.revision_tree(
606
_mod_revision.NULL_REVISION).inventory
607
if len(self.parents) > 0:
608
if basis_revision_id != self.parents[0] and not ghost_basis:
610
"arbitrary basis parents not yet supported with merges")
611
for revtree in revtrees[1:]:
612
for change in revtree.inventory._make_delta(basis_inv):
613
if change[1] is None:
614
# Not present in this parent.
616
if change[2] not in merged_ids:
617
if change[0] is not None:
618
basis_entry = basis_inv[change[2]]
619
merged_ids[change[2]] = [
621
basis_entry.revision,
624
parent_entries[change[2]] = {
626
basis_entry.revision:basis_entry,
628
change[3].revision:change[3],
631
merged_ids[change[2]] = [change[3].revision]
632
parent_entries[change[2]] = {change[3].revision:change[3]}
634
merged_ids[change[2]].append(change[3].revision)
635
parent_entries[change[2]][change[3].revision] = change[3]
638
# Setup the changes from the tree:
639
# changes maps file_id -> (change, [parent revision_ids])
641
for change in iter_changes:
642
# This probably looks up in basis_inv way to much.
643
if change[1][0] is not None:
644
head_candidate = [basis_inv[change[0]].revision]
647
changes[change[0]] = change, merged_ids.get(change[0],
649
unchanged_merged = set(merged_ids) - set(changes)
650
# Extend the changes dict with synthetic changes to record merges of
652
for file_id in unchanged_merged:
653
# Record a merged version of these items that did not change vs the
654
# basis. This can be either identical parallel changes, or a revert
655
# of a specific file after a merge. The recorded content will be
656
# that of the current tree (which is the same as the basis), but
657
# the per-file graph will reflect a merge.
658
# NB:XXX: We are reconstructing path information we had, this
659
# should be preserved instead.
660
# inv delta change: (file_id, (path_in_source, path_in_target),
661
# changed_content, versioned, parent, name, kind,
664
basis_entry = basis_inv[file_id]
665
except errors.NoSuchId:
666
# a change from basis->some_parents but file_id isn't in basis
667
# so was new in the merge, which means it must have changed
668
# from basis -> current, and as it hasn't the add was reverted
669
# by the user. So we discard this change.
673
(basis_inv.id2path(file_id), tree.id2path(file_id)),
675
(basis_entry.parent_id, basis_entry.parent_id),
676
(basis_entry.name, basis_entry.name),
677
(basis_entry.kind, basis_entry.kind),
678
(basis_entry.executable, basis_entry.executable))
679
changes[file_id] = (change, merged_ids[file_id])
680
# changes contains tuples with the change and a set of inventory
681
# candidates for the file.
683
# old_path, new_path, file_id, new_inventory_entry
684
seen_root = False # Is the root in the basis delta?
685
inv_delta = self._basis_delta
686
modified_rev = self._new_revision_id
687
for change, head_candidates in changes.values():
688
if change[3][1]: # versioned in target.
689
# Several things may be happening here:
690
# We may have a fork in the per-file graph
691
# - record a change with the content from tree
692
# We may have a change against < all trees
693
# - carry over the tree that hasn't changed
694
# We may have a change against all trees
695
# - record the change with the content from tree
698
entry = _entry_factory[kind](file_id, change[5][1],
700
head_set = self._heads(change[0], set(head_candidates))
703
for head_candidate in head_candidates:
704
if head_candidate in head_set:
705
heads.append(head_candidate)
706
head_set.remove(head_candidate)
709
# Could be a carry-over situation:
710
parent_entry_revs = parent_entries.get(file_id, None)
711
if parent_entry_revs:
712
parent_entry = parent_entry_revs.get(heads[0], None)
715
if parent_entry is None:
716
# The parent iter_changes was called against is the one
717
# that is the per-file head, so any change is relevant
718
# iter_changes is valid.
719
carry_over_possible = False
721
# could be a carry over situation
722
# A change against the basis may just indicate a merge,
723
# we need to check the content against the source of the
724
# merge to determine if it was changed after the merge
726
if (parent_entry.kind != entry.kind or
727
parent_entry.parent_id != entry.parent_id or
728
parent_entry.name != entry.name):
729
# Metadata common to all entries has changed
730
# against per-file parent
731
carry_over_possible = False
733
carry_over_possible = True
734
# per-type checks for changes against the parent_entry
737
# Cannot be a carry-over situation
738
carry_over_possible = False
739
# Populate the entry in the delta
741
# XXX: There is still a small race here: If someone reverts the content of a file
742
# after iter_changes examines and decides it has changed,
743
# we will unconditionally record a new version even if some
744
# other process reverts it while commit is running (with
745
# the revert happening after iter_changes did it's
748
entry.executable = True
750
entry.executable = False
751
if (carry_over_possible and
752
parent_entry.executable == entry.executable):
753
# Check the file length, content hash after reading
755
nostore_sha = parent_entry.text_sha1
758
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
760
text = file_obj.read()
764
entry.text_sha1, entry.text_size = self._add_text_to_weave(
765
file_id, text, heads, nostore_sha)
766
yield file_id, change[1][1], (entry.text_sha1, stat_value)
767
except errors.ExistingContent:
768
# No content change against a carry_over parent
769
# Perhaps this should also yield a fs hash update?
771
entry.text_size = parent_entry.text_size
772
entry.text_sha1 = parent_entry.text_sha1
773
elif kind == 'symlink':
775
entry.symlink_target = tree.get_symlink_target(file_id)
776
if (carry_over_possible and
777
parent_entry.symlink_target == entry.symlink_target):
780
self._add_text_to_weave(change[0], '', heads, None)
781
elif kind == 'directory':
782
if carry_over_possible:
785
# Nothing to set on the entry.
786
# XXX: split into the Root and nonRoot versions.
787
if change[1][1] != '' or self.repository.supports_rich_root():
788
self._add_text_to_weave(change[0], '', heads, None)
789
elif kind == 'tree-reference':
790
if not self.repository._format.supports_tree_reference:
791
# This isn't quite sane as an error, but we shouldn't
792
# ever see this code path in practice: tree's don't
793
# permit references when the repo doesn't support tree
795
raise errors.UnsupportedOperation(tree.add_reference,
797
reference_revision = tree.get_reference_revision(change[0])
798
entry.reference_revision = reference_revision
799
if (carry_over_possible and
800
parent_entry.reference_revision == reference_revision):
803
self._add_text_to_weave(change[0], '', heads, None)
805
raise AssertionError('unknown kind %r' % kind)
807
entry.revision = modified_rev
809
entry.revision = parent_entry.revision
812
new_path = change[1][1]
813
inv_delta.append((change[1][0], new_path, change[0], entry))
816
self.new_inventory = None
818
# This should perhaps be guarded by a check that the basis we
819
# commit against is the basis for the commit and if not do a delta
821
self._any_changes = True
823
# housekeeping root entry changes do not affect no-change commits.
824
self._require_root_change(tree)
825
self.basis_delta_revision = basis_revision_id
827
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
828
parent_keys = tuple([(file_id, parent) for parent in parents])
829
return self.repository.texts._add_text(
830
(file_id, self._new_revision_id), parent_keys, new_text,
831
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
834
class RootCommitBuilder(CommitBuilder):
835
"""This commitbuilder actually records the root id"""
837
# the root entry gets versioned properly by this builder.
838
_versioned_root = True
840
def _check_root(self, ie, parent_invs, tree):
841
"""Helper for record_entry_contents.
843
:param ie: An entry being added.
844
:param parent_invs: The inventories of the parent revisions of the
846
:param tree: The tree that is being committed.
849
def _require_root_change(self, tree):
850
"""Enforce an appropriate root object change.
852
This is called once when record_iter_changes is called, if and only if
853
the root was not in the delta calculated by record_iter_changes.
855
:param tree: The tree which is being committed.
857
# versioned roots do not change unless the tree found a change.
860
######################################################################
864
class Repository(_RelockDebugMixin):
865
"""Repository holding history for one or more branches.
867
The repository holds and retrieves historical information including
868
revisions and file history. It's normally accessed only by the Branch,
869
which views a particular line of development through that history.
871
The Repository builds on top of some byte storage facilies (the revisions,
872
signatures, inventories, texts and chk_bytes attributes) and a Transport,
873
which respectively provide byte storage and a means to access the (possibly
876
The byte storage facilities are addressed via tuples, which we refer to
877
as 'keys' throughout the code base. Revision_keys, inventory_keys and
878
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
879
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
880
byte string made up of a hash identifier and a hash value.
881
We use this interface because it allows low friction with the underlying
882
code that implements disk indices, network encoding and other parts of
885
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
886
the serialised revisions for the repository. This can be used to obtain
887
revision graph information or to access raw serialised revisions.
888
The result of trying to insert data into the repository via this store
889
is undefined: it should be considered read-only except for implementors
891
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
892
the serialised signatures for the repository. This can be used to
893
obtain access to raw serialised signatures. The result of trying to
894
insert data into the repository via this store is undefined: it should
895
be considered read-only except for implementors of repositories.
896
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
897
the serialised inventories for the repository. This can be used to
898
obtain unserialised inventories. The result of trying to insert data
899
into the repository via this store is undefined: it should be
900
considered read-only except for implementors of repositories.
901
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
902
texts of files and directories for the repository. This can be used to
903
obtain file texts or file graphs. Note that Repository.iter_file_bytes
904
is usually a better interface for accessing file texts.
905
The result of trying to insert data into the repository via this store
906
is undefined: it should be considered read-only except for implementors
908
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
909
any data the repository chooses to store or have indexed by its hash.
910
The result of trying to insert data into the repository via this store
911
is undefined: it should be considered read-only except for implementors
913
:ivar _transport: Transport for file access to repository, typically
914
pointing to .bzr/repository.
917
# What class to use for a CommitBuilder. Often its simpler to change this
918
# in a Repository class subclass rather than to override
919
# get_commit_builder.
920
_commit_builder_class = CommitBuilder
921
# The search regex used by xml based repositories to determine what things
922
# where changed in a single commit.
923
_file_ids_altered_regex = lazy_regex.lazy_compile(
924
r'file_id="(?P<file_id>[^"]+)"'
925
r'.* revision="(?P<revision_id>[^"]+)"'
928
def abort_write_group(self, suppress_errors=False):
929
"""Commit the contents accrued within the current write group.
931
:param suppress_errors: if true, abort_write_group will catch and log
932
unexpected errors that happen during the abort, rather than
933
allowing them to propagate. Defaults to False.
935
:seealso: start_write_group.
937
if self._write_group is not self.get_transaction():
938
# has an unlock or relock occured ?
941
'(suppressed) mismatched lock context and write group. %r, %r',
942
self._write_group, self.get_transaction())
944
raise errors.BzrError(
945
'mismatched lock context and write group. %r, %r' %
946
(self._write_group, self.get_transaction()))
948
self._abort_write_group()
949
except Exception, exc:
950
self._write_group = None
951
if not suppress_errors:
953
mutter('abort_write_group failed')
954
log_exception_quietly()
955
note('bzr: ERROR (ignored): %s', exc)
956
self._write_group = None
958
def _abort_write_group(self):
959
"""Template method for per-repository write group cleanup.
961
This is called during abort before the write group is considered to be
962
finished and should cleanup any internal state accrued during the write
963
group. There is no requirement that data handed to the repository be
964
*not* made available - this is not a rollback - but neither should any
965
attempt be made to ensure that data added is fully commited. Abort is
966
invoked when an error has occured so futher disk or network operations
967
may not be possible or may error and if possible should not be
971
def add_fallback_repository(self, repository):
972
"""Add a repository to use for looking up data not held locally.
974
:param repository: A repository.
976
if not self._format.supports_external_lookups:
977
raise errors.UnstackableRepositoryFormat(self._format, self.base)
979
# This repository will call fallback.unlock() when we transition to
980
# the unlocked state, so we make sure to increment the lock count
981
repository.lock_read()
982
self._check_fallback_repository(repository)
983
self._fallback_repositories.append(repository)
984
self.texts.add_fallback_versioned_files(repository.texts)
985
self.inventories.add_fallback_versioned_files(repository.inventories)
986
self.revisions.add_fallback_versioned_files(repository.revisions)
987
self.signatures.add_fallback_versioned_files(repository.signatures)
988
if self.chk_bytes is not None:
989
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
991
def _check_fallback_repository(self, repository):
992
"""Check that this repository can fallback to repository safely.
994
Raise an error if not.
996
:param repository: A repository to fallback to.
998
return InterRepository._assert_same_model(self, repository)
1000
def add_inventory(self, revision_id, inv, parents):
1001
"""Add the inventory inv to the repository as revision_id.
1003
:param parents: The revision ids of the parents that revision_id
1004
is known to have and are in the repository already.
1006
:returns: The validator(which is a sha1 digest, though what is sha'd is
1007
repository format specific) of the serialized inventory.
1009
if not self.is_in_write_group():
1010
raise AssertionError("%r not in write group" % (self,))
1011
_mod_revision.check_not_reserved_id(revision_id)
1012
if not (inv.revision_id is None or inv.revision_id == revision_id):
1013
raise AssertionError(
1014
"Mismatch between inventory revision"
1015
" id and insertion revid (%r, %r)"
1016
% (inv.revision_id, revision_id))
1017
if inv.root is None:
1018
raise AssertionError()
1019
return self._add_inventory_checked(revision_id, inv, parents)
1021
def _add_inventory_checked(self, revision_id, inv, parents):
1022
"""Add inv to the repository after checking the inputs.
1024
This function can be overridden to allow different inventory styles.
1026
:seealso: add_inventory, for the contract.
1028
inv_lines = self._serialise_inventory_to_lines(inv)
1029
return self._inventory_add_lines(revision_id, parents,
1030
inv_lines, check_content=False)
1032
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1033
parents, basis_inv=None, propagate_caches=False):
1034
"""Add a new inventory expressed as a delta against another revision.
1036
See the inventory developers documentation for the theory behind
1039
:param basis_revision_id: The inventory id the delta was created
1040
against. (This does not have to be a direct parent.)
1041
:param delta: The inventory delta (see Inventory.apply_delta for
1043
:param new_revision_id: The revision id that the inventory is being
1045
:param parents: The revision ids of the parents that revision_id is
1046
known to have and are in the repository already. These are supplied
1047
for repositories that depend on the inventory graph for revision
1048
graph access, as well as for those that pun ancestry with delta
1050
:param basis_inv: The basis inventory if it is already known,
1052
:param propagate_caches: If True, the caches for this inventory are
1053
copied to and updated for the result if possible.
1055
:returns: (validator, new_inv)
1056
The validator(which is a sha1 digest, though what is sha'd is
1057
repository format specific) of the serialized inventory, and the
1058
resulting inventory.
1060
if not self.is_in_write_group():
1061
raise AssertionError("%r not in write group" % (self,))
1062
_mod_revision.check_not_reserved_id(new_revision_id)
1063
basis_tree = self.revision_tree(basis_revision_id)
1064
basis_tree.lock_read()
1066
# Note that this mutates the inventory of basis_tree, which not all
1067
# inventory implementations may support: A better idiom would be to
1068
# return a new inventory, but as there is no revision tree cache in
1069
# repository this is safe for now - RBC 20081013
1070
if basis_inv is None:
1071
basis_inv = basis_tree.inventory
1072
basis_inv.apply_delta(delta)
1073
basis_inv.revision_id = new_revision_id
1074
return (self.add_inventory(new_revision_id, basis_inv, parents),
1079
def _inventory_add_lines(self, revision_id, parents, lines,
1080
check_content=True):
1081
"""Store lines in inv_vf and return the sha1 of the inventory."""
1082
parents = [(parent,) for parent in parents]
1083
result = self.inventories.add_lines((revision_id,), parents, lines,
1084
check_content=check_content)[0]
1085
self.inventories._access.flush()
1088
def add_revision(self, revision_id, rev, inv=None, config=None):
1089
"""Add rev to the revision store as revision_id.
1091
:param revision_id: the revision id to use.
1092
:param rev: The revision object.
1093
:param inv: The inventory for the revision. if None, it will be looked
1094
up in the inventory storer
1095
:param config: If None no digital signature will be created.
1096
If supplied its signature_needed method will be used
1097
to determine if a signature should be made.
1099
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1101
_mod_revision.check_not_reserved_id(revision_id)
1102
if config is not None and config.signature_needed():
1104
inv = self.get_inventory(revision_id)
1105
plaintext = Testament(rev, inv).as_short_text()
1106
self.store_revision_signature(
1107
gpg.GPGStrategy(config), plaintext, revision_id)
1108
# check inventory present
1109
if not self.inventories.get_parent_map([(revision_id,)]):
1111
raise errors.WeaveRevisionNotPresent(revision_id,
1114
# yes, this is not suitable for adding with ghosts.
1115
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1118
key = (revision_id,)
1119
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1120
self._add_revision(rev)
1122
def _add_revision(self, revision):
1123
text = self._serializer.write_revision_to_string(revision)
1124
key = (revision.revision_id,)
1125
parents = tuple((parent,) for parent in revision.parent_ids)
1126
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1128
def all_revision_ids(self):
1129
"""Returns a list of all the revision ids in the repository.
1131
This is conceptually deprecated because code should generally work on
1132
the graph reachable from a particular revision, and ignore any other
1133
revisions that might be present. There is no direct replacement
1136
if 'evil' in debug.debug_flags:
1137
mutter_callsite(2, "all_revision_ids is linear with history.")
1138
return self._all_revision_ids()
1140
def _all_revision_ids(self):
1141
"""Returns a list of all the revision ids in the repository.
1143
These are in as much topological order as the underlying store can
1146
raise NotImplementedError(self._all_revision_ids)
1148
def break_lock(self):
1149
"""Break a lock if one is present from another instance.
1151
Uses the ui factory to ask for confirmation if the lock may be from
1154
self.control_files.break_lock()
1157
def _eliminate_revisions_not_present(self, revision_ids):
1158
"""Check every revision id in revision_ids to see if we have it.
1160
Returns a set of the present revisions.
1163
graph = self.get_graph()
1164
parent_map = graph.get_parent_map(revision_ids)
1165
# The old API returned a list, should this actually be a set?
1166
return parent_map.keys()
1168
def _check_inventories(self, checker):
1169
"""Check the inventories found from the revision scan.
1171
This is responsible for verifying the sha1 of inventories and
1172
creating a pending_keys set that covers data referenced by inventories.
1174
bar = ui.ui_factory.nested_progress_bar()
1176
self._do_check_inventories(checker, bar)
1180
def _do_check_inventories(self, checker, bar):
1181
"""Helper for _check_inventories."""
1183
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1184
kinds = ['chk_bytes', 'texts']
1185
count = len(checker.pending_keys)
1186
bar.update("inventories", 0, 2)
1187
current_keys = checker.pending_keys
1188
checker.pending_keys = {}
1189
# Accumulate current checks.
1190
for key in current_keys:
1191
if key[0] != 'inventories' and key[0] not in kinds:
1192
checker._report_items.append('unknown key type %r' % (key,))
1193
keys[key[0]].add(key[1:])
1194
if keys['inventories']:
1195
# NB: output order *should* be roughly sorted - topo or
1196
# inverse topo depending on repository - either way decent
1197
# to just delta against. However, pre-CHK formats didn't
1198
# try to optimise inventory layout on disk. As such the
1199
# pre-CHK code path does not use inventory deltas.
1201
for record in self.inventories.check(keys=keys['inventories']):
1202
if record.storage_kind == 'absent':
1203
checker._report_items.append(
1204
'Missing inventory {%s}' % (record.key,))
1206
last_object = self._check_record('inventories', record,
1207
checker, last_object,
1208
current_keys[('inventories',) + record.key])
1209
del keys['inventories']
1212
bar.update("texts", 1)
1213
while (checker.pending_keys or keys['chk_bytes']
1215
# Something to check.
1216
current_keys = checker.pending_keys
1217
checker.pending_keys = {}
1218
# Accumulate current checks.
1219
for key in current_keys:
1220
if key[0] not in kinds:
1221
checker._report_items.append('unknown key type %r' % (key,))
1222
keys[key[0]].add(key[1:])
1223
# Check the outermost kind only - inventories || chk_bytes || texts
1227
for record in getattr(self, kind).check(keys=keys[kind]):
1228
if record.storage_kind == 'absent':
1229
checker._report_items.append(
1230
'Missing %s {%s}' % (kind, record.key,))
1232
last_object = self._check_record(kind, record,
1233
checker, last_object, current_keys[(kind,) + record.key])
1237
def _check_record(self, kind, record, checker, last_object, item_data):
1238
"""Check a single text from this repository."""
1239
if kind == 'inventories':
1240
rev_id = record.key[0]
1241
inv = self.deserialise_inventory(rev_id,
1242
record.get_bytes_as('fulltext'))
1243
if last_object is not None:
1244
delta = inv._make_delta(last_object)
1245
for old_path, path, file_id, ie in delta:
1248
ie.check(checker, rev_id, inv)
1250
for path, ie in inv.iter_entries():
1251
ie.check(checker, rev_id, inv)
1252
if self._format.fast_deltas:
1254
elif kind == 'chk_bytes':
1255
# No code written to check chk_bytes for this repo format.
1256
checker._report_items.append(
1257
'unsupported key type chk_bytes for %s' % (record.key,))
1258
elif kind == 'texts':
1259
self._check_text(record, checker, item_data)
1261
checker._report_items.append(
1262
'unknown key type %s for %s' % (kind, record.key))
1264
def _check_text(self, record, checker, item_data):
1265
"""Check a single text."""
1266
# Check it is extractable.
1267
# TODO: check length.
1268
if record.storage_kind == 'chunked':
1269
chunks = record.get_bytes_as(record.storage_kind)
1270
sha1 = osutils.sha_strings(chunks)
1271
length = sum(map(len, chunks))
1273
content = record.get_bytes_as('fulltext')
1274
sha1 = osutils.sha_string(content)
1275
length = len(content)
1276
if item_data and sha1 != item_data[1]:
1277
checker._report_items.append(
1278
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1279
(record.key, sha1, item_data[1], item_data[2]))
1282
def create(a_bzrdir):
1283
"""Construct the current default format repository in a_bzrdir."""
1284
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1286
def __init__(self, _format, a_bzrdir, control_files):
1287
"""instantiate a Repository.
1289
:param _format: The format of the repository on disk.
1290
:param a_bzrdir: The BzrDir of the repository.
1292
In the future we will have a single api for all stores for
1293
getting file texts, inventories and revisions, then
1294
this construct will accept instances of those things.
1296
super(Repository, self).__init__()
1297
self._format = _format
1298
# the following are part of the public API for Repository:
1299
self.bzrdir = a_bzrdir
1300
self.control_files = control_files
1301
self._transport = control_files._transport
1302
self.base = self._transport.base
1304
self._reconcile_does_inventory_gc = True
1305
self._reconcile_fixes_text_parents = False
1306
self._reconcile_backsup_inventory = True
1307
# not right yet - should be more semantically clear ?
1309
# TODO: make sure to construct the right store classes, etc, depending
1310
# on whether escaping is required.
1311
self._warn_if_deprecated()
1312
self._write_group = None
1313
# Additional places to query for data.
1314
self._fallback_repositories = []
1315
# An InventoryEntry cache, used during deserialization
1316
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1317
# Is it safe to return inventory entries directly from the entry cache,
1318
# rather copying them?
1319
self._safe_to_return_from_cache = False
1322
if self._fallback_repositories:
1323
return '%s(%r, fallback_repositories=%r)' % (
1324
self.__class__.__name__,
1326
self._fallback_repositories)
1328
return '%s(%r)' % (self.__class__.__name__,
1331
def _has_same_fallbacks(self, other_repo):
1332
"""Returns true if the repositories have the same fallbacks."""
1333
my_fb = self._fallback_repositories
1334
other_fb = other_repo._fallback_repositories
1335
if len(my_fb) != len(other_fb):
1337
for f, g in zip(my_fb, other_fb):
1338
if not f.has_same_location(g):
1342
def has_same_location(self, other):
1343
"""Returns a boolean indicating if this repository is at the same
1344
location as another repository.
1346
This might return False even when two repository objects are accessing
1347
the same physical repository via different URLs.
1349
if self.__class__ is not other.__class__:
1351
return (self._transport.base == other._transport.base)
1353
def is_in_write_group(self):
1354
"""Return True if there is an open write group.
1356
:seealso: start_write_group.
1358
return self._write_group is not None
1360
def is_locked(self):
1361
return self.control_files.is_locked()
1363
def is_write_locked(self):
1364
"""Return True if this object is write locked."""
1365
return self.is_locked() and self.control_files._lock_mode == 'w'
1367
def lock_write(self, token=None):
1368
"""Lock this repository for writing.
1370
This causes caching within the repository obejct to start accumlating
1371
data during reads, and allows a 'write_group' to be obtained. Write
1372
groups must be used for actual data insertion.
1374
:param token: if this is already locked, then lock_write will fail
1375
unless the token matches the existing lock.
1376
:returns: a token if this instance supports tokens, otherwise None.
1377
:raises TokenLockingNotSupported: when a token is given but this
1378
instance doesn't support using token locks.
1379
:raises MismatchedToken: if the specified token doesn't match the token
1380
of the existing lock.
1381
:seealso: start_write_group.
1383
A token should be passed in if you know that you have locked the object
1384
some other way, and need to synchronise this object's state with that
1387
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1389
locked = self.is_locked()
1390
result = self.control_files.lock_write(token=token)
1392
self._note_lock('w')
1393
for repo in self._fallback_repositories:
1394
# Writes don't affect fallback repos
1396
self._refresh_data()
1399
def lock_read(self):
1400
locked = self.is_locked()
1401
self.control_files.lock_read()
1403
self._note_lock('r')
1404
for repo in self._fallback_repositories:
1406
self._refresh_data()
1408
def get_physical_lock_status(self):
1409
return self.control_files.get_physical_lock_status()
1411
def leave_lock_in_place(self):
1412
"""Tell this repository not to release the physical lock when this
1415
If lock_write doesn't return a token, then this method is not supported.
1417
self.control_files.leave_in_place()
1419
def dont_leave_lock_in_place(self):
1420
"""Tell this repository to release the physical lock when this
1421
object is unlocked, even if it didn't originally acquire it.
1423
If lock_write doesn't return a token, then this method is not supported.
1425
self.control_files.dont_leave_in_place()
1428
def gather_stats(self, revid=None, committers=None):
1429
"""Gather statistics from a revision id.
1431
:param revid: The revision id to gather statistics from, if None, then
1432
no revision specific statistics are gathered.
1433
:param committers: Optional parameter controlling whether to grab
1434
a count of committers from the revision specific statistics.
1435
:return: A dictionary of statistics. Currently this contains:
1436
committers: The number of committers if requested.
1437
firstrev: A tuple with timestamp, timezone for the penultimate left
1438
most ancestor of revid, if revid is not the NULL_REVISION.
1439
latestrev: A tuple with timestamp, timezone for revid, if revid is
1440
not the NULL_REVISION.
1441
revisions: The total revision count in the repository.
1442
size: An estimate disk size of the repository in bytes.
1445
if revid and committers:
1446
result['committers'] = 0
1447
if revid and revid != _mod_revision.NULL_REVISION:
1449
all_committers = set()
1450
revisions = self.get_ancestry(revid)
1451
# pop the leading None
1453
first_revision = None
1455
# ignore the revisions in the middle - just grab first and last
1456
revisions = revisions[0], revisions[-1]
1457
for revision in self.get_revisions(revisions):
1458
if not first_revision:
1459
first_revision = revision
1461
all_committers.add(revision.committer)
1462
last_revision = revision
1464
result['committers'] = len(all_committers)
1465
result['firstrev'] = (first_revision.timestamp,
1466
first_revision.timezone)
1467
result['latestrev'] = (last_revision.timestamp,
1468
last_revision.timezone)
1470
# now gather global repository information
1471
# XXX: This is available for many repos regardless of listability.
1472
if self.bzrdir.root_transport.listable():
1473
# XXX: do we want to __define len__() ?
1474
# Maybe the versionedfiles object should provide a different
1475
# method to get the number of keys.
1476
result['revisions'] = len(self.revisions.keys())
1477
# result['size'] = t
1480
def find_branches(self, using=False):
1481
"""Find branches underneath this repository.
1483
This will include branches inside other branches.
1485
:param using: If True, list only branches using this repository.
1487
if using and not self.is_shared():
1489
return [self.bzrdir.open_branch()]
1490
except errors.NotBranchError:
1492
class Evaluator(object):
1495
self.first_call = True
1497
def __call__(self, bzrdir):
1498
# On the first call, the parameter is always the bzrdir
1499
# containing the current repo.
1500
if not self.first_call:
1502
repository = bzrdir.open_repository()
1503
except errors.NoRepositoryPresent:
1506
return False, (None, repository)
1507
self.first_call = False
1509
value = (bzrdir.open_branch(), None)
1510
except errors.NotBranchError:
1511
value = (None, None)
1515
for branch, repository in bzrdir.BzrDir.find_bzrdirs(
1516
self.bzrdir.root_transport, evaluate=Evaluator()):
1517
if branch is not None:
1518
branches.append(branch)
1519
if not using and repository is not None:
1520
branches.extend(repository.find_branches())
1524
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
1525
"""Return the revision ids that other has that this does not.
1527
These are returned in topological order.
1529
revision_id: only return revision ids included by revision_id.
1531
return InterRepository.get(other, self).search_missing_revision_ids(
1532
revision_id, find_ghosts)
1536
"""Open the repository rooted at base.
1538
For instance, if the repository is at URL/.bzr/repository,
1539
Repository.open(URL) -> a Repository instance.
1541
control = bzrdir.BzrDir.open(base)
1542
return control.open_repository()
1544
def copy_content_into(self, destination, revision_id=None):
1545
"""Make a complete copy of the content in self into destination.
1547
This is a destructive operation! Do not use it on existing
1550
return InterRepository.get(self, destination).copy_content(revision_id)
1552
def commit_write_group(self):
1553
"""Commit the contents accrued within the current write group.
1555
:seealso: start_write_group.
1557
:return: it may return an opaque hint that can be passed to 'pack'.
1559
if self._write_group is not self.get_transaction():
1560
# has an unlock or relock occured ?
1561
raise errors.BzrError('mismatched lock context %r and '
1563
(self.get_transaction(), self._write_group))
1564
result = self._commit_write_group()
1565
self._write_group = None
1568
def _commit_write_group(self):
1569
"""Template method for per-repository write group cleanup.
1571
This is called before the write group is considered to be
1572
finished and should ensure that all data handed to the repository
1573
for writing during the write group is safely committed (to the
1574
extent possible considering file system caching etc).
1577
def suspend_write_group(self):
1578
raise errors.UnsuspendableWriteGroup(self)
1580
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1581
"""Return the keys of missing inventory parents for revisions added in
1584
A revision is not complete if the inventory delta for that revision
1585
cannot be calculated. Therefore if the parent inventories of a
1586
revision are not present, the revision is incomplete, and e.g. cannot
1587
be streamed by a smart server. This method finds missing inventory
1588
parents for revisions added in this write group.
1590
if not self._format.supports_external_lookups:
1591
# This is only an issue for stacked repositories
1593
if not self.is_in_write_group():
1594
raise AssertionError('not in a write group')
1596
# XXX: We assume that every added revision already has its
1597
# corresponding inventory, so we only check for parent inventories that
1598
# might be missing, rather than all inventories.
1599
parents = set(self.revisions._index.get_missing_parents())
1600
parents.discard(_mod_revision.NULL_REVISION)
1601
unstacked_inventories = self.inventories._index
1602
present_inventories = unstacked_inventories.get_parent_map(
1603
key[-1:] for key in parents)
1604
parents.difference_update(present_inventories)
1605
if len(parents) == 0:
1606
# No missing parent inventories.
1608
if not check_for_missing_texts:
1609
return set(('inventories', rev_id) for (rev_id,) in parents)
1610
# Ok, now we have a list of missing inventories. But these only matter
1611
# if the inventories that reference them are missing some texts they
1612
# appear to introduce.
1613
# XXX: Texts referenced by all added inventories need to be present,
1614
# but at the moment we're only checking for texts referenced by
1615
# inventories at the graph's edge.
1616
key_deps = self.revisions._index._key_dependencies
1617
key_deps.satisfy_refs_for_keys(present_inventories)
1618
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1619
file_ids = self.fileids_altered_by_revision_ids(referrers)
1620
missing_texts = set()
1621
for file_id, version_ids in file_ids.iteritems():
1622
missing_texts.update(
1623
(file_id, version_id) for version_id in version_ids)
1624
present_texts = self.texts.get_parent_map(missing_texts)
1625
missing_texts.difference_update(present_texts)
1626
if not missing_texts:
1627
# No texts are missing, so all revisions and their deltas are
1630
# Alternatively the text versions could be returned as the missing
1631
# keys, but this is likely to be less data.
1632
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1635
def refresh_data(self):
1636
"""Re-read any data needed to to synchronise with disk.
1638
This method is intended to be called after another repository instance
1639
(such as one used by a smart server) has inserted data into the
1640
repository. It may not be called during a write group, but may be
1641
called at any other time.
1643
if self.is_in_write_group():
1644
raise errors.InternalBzrError(
1645
"May not refresh_data while in a write group.")
1646
self._refresh_data()
1648
def resume_write_group(self, tokens):
1649
if not self.is_write_locked():
1650
raise errors.NotWriteLocked(self)
1651
if self._write_group:
1652
raise errors.BzrError('already in a write group')
1653
self._resume_write_group(tokens)
1654
# so we can detect unlock/relock - the write group is now entered.
1655
self._write_group = self.get_transaction()
1657
def _resume_write_group(self, tokens):
1658
raise errors.UnsuspendableWriteGroup(self)
1660
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1662
"""Fetch the content required to construct revision_id from source.
1664
If revision_id is None and fetch_spec is None, then all content is
1667
fetch() may not be used when the repository is in a write group -
1668
either finish the current write group before using fetch, or use
1669
fetch before starting the write group.
1671
:param find_ghosts: Find and copy revisions in the source that are
1672
ghosts in the target (and not reachable directly by walking out to
1673
the first-present revision in target from revision_id).
1674
:param revision_id: If specified, all the content needed for this
1675
revision ID will be copied to the target. Fetch will determine for
1676
itself which content needs to be copied.
1677
:param fetch_spec: If specified, a SearchResult or
1678
PendingAncestryResult that describes which revisions to copy. This
1679
allows copying multiple heads at once. Mutually exclusive with
1682
if fetch_spec is not None and revision_id is not None:
1683
raise AssertionError(
1684
"fetch_spec and revision_id are mutually exclusive.")
1685
if self.is_in_write_group():
1686
raise errors.InternalBzrError(
1687
"May not fetch while in a write group.")
1688
# fast path same-url fetch operations
1689
# TODO: lift out to somewhere common with RemoteRepository
1690
# <https://bugs.edge.launchpad.net/bzr/+bug/401646>
1691
if (self.has_same_location(source)
1692
and fetch_spec is None
1693
and self._has_same_fallbacks(source)):
1694
# check that last_revision is in 'from' and then return a
1696
if (revision_id is not None and
1697
not _mod_revision.is_null(revision_id)):
1698
self.get_revision(revision_id)
1700
# if there is no specific appropriate InterRepository, this will get
1701
# the InterRepository base class, which raises an
1702
# IncompatibleRepositories when asked to fetch.
1703
inter = InterRepository.get(source, self)
1704
return inter.fetch(revision_id=revision_id, pb=pb,
1705
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1707
def create_bundle(self, target, base, fileobj, format=None):
1708
return serializer.write_bundle(self, target, base, fileobj, format)
1710
def get_commit_builder(self, branch, parents, config, timestamp=None,
1711
timezone=None, committer=None, revprops=None,
1713
"""Obtain a CommitBuilder for this repository.
1715
:param branch: Branch to commit to.
1716
:param parents: Revision ids of the parents of the new revision.
1717
:param config: Configuration to use.
1718
:param timestamp: Optional timestamp recorded for commit.
1719
:param timezone: Optional timezone for timestamp.
1720
:param committer: Optional committer to set for commit.
1721
:param revprops: Optional dictionary of revision properties.
1722
:param revision_id: Optional revision id.
1724
if self._fallback_repositories:
1725
raise errors.BzrError("Cannot commit from a lightweight checkout "
1726
"to a stacked branch. See "
1727
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1728
result = self._commit_builder_class(self, parents, config,
1729
timestamp, timezone, committer, revprops, revision_id)
1730
self.start_write_group()
1733
@only_raises(errors.LockNotHeld, errors.LockBroken)
1735
if (self.control_files._lock_count == 1 and
1736
self.control_files._lock_mode == 'w'):
1737
if self._write_group is not None:
1738
self.abort_write_group()
1739
self.control_files.unlock()
1740
raise errors.BzrError(
1741
'Must end write groups before releasing write locks.')
1742
self.control_files.unlock()
1743
if self.control_files._lock_count == 0:
1744
self._inventory_entry_cache.clear()
1745
for repo in self._fallback_repositories:
1749
def clone(self, a_bzrdir, revision_id=None):
1750
"""Clone this repository into a_bzrdir using the current format.
1752
Currently no check is made that the format of this repository and
1753
the bzrdir format are compatible. FIXME RBC 20060201.
1755
:return: The newly created destination repository.
1757
# TODO: deprecate after 0.16; cloning this with all its settings is
1758
# probably not very useful -- mbp 20070423
1759
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1760
self.copy_content_into(dest_repo, revision_id)
1763
def start_write_group(self):
1764
"""Start a write group in the repository.
1766
Write groups are used by repositories which do not have a 1:1 mapping
1767
between file ids and backend store to manage the insertion of data from
1768
both fetch and commit operations.
1770
A write lock is required around the start_write_group/commit_write_group
1771
for the support of lock-requiring repository formats.
1773
One can only insert data into a repository inside a write group.
1777
if not self.is_write_locked():
1778
raise errors.NotWriteLocked(self)
1779
if self._write_group:
1780
raise errors.BzrError('already in a write group')
1781
self._start_write_group()
1782
# so we can detect unlock/relock - the write group is now entered.
1783
self._write_group = self.get_transaction()
1785
def _start_write_group(self):
1786
"""Template method for per-repository write group startup.
1788
This is called before the write group is considered to be
1793
def sprout(self, to_bzrdir, revision_id=None):
1794
"""Create a descendent repository for new development.
1796
Unlike clone, this does not copy the settings of the repository.
1798
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1799
dest_repo.fetch(self, revision_id=revision_id)
1802
def _create_sprouting_repo(self, a_bzrdir, shared):
1803
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1804
# use target default format.
1805
dest_repo = a_bzrdir.create_repository()
1807
# Most control formats need the repository to be specifically
1808
# created, but on some old all-in-one formats it's not needed
1810
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1811
except errors.UninitializableFormat:
1812
dest_repo = a_bzrdir.open_repository()
1815
def _get_sink(self):
1816
"""Return a sink for streaming into this repository."""
1817
return StreamSink(self)
1819
def _get_source(self, to_format):
1820
"""Return a source for streaming from this repository."""
1821
return StreamSource(self, to_format)
1824
def has_revision(self, revision_id):
1825
"""True if this repository has a copy of the revision."""
1826
return revision_id in self.has_revisions((revision_id,))
1829
def has_revisions(self, revision_ids):
1830
"""Probe to find out the presence of multiple revisions.
1832
:param revision_ids: An iterable of revision_ids.
1833
:return: A set of the revision_ids that were present.
1835
parent_map = self.revisions.get_parent_map(
1836
[(rev_id,) for rev_id in revision_ids])
1838
if _mod_revision.NULL_REVISION in revision_ids:
1839
result.add(_mod_revision.NULL_REVISION)
1840
result.update([key[0] for key in parent_map])
1844
def get_revision(self, revision_id):
1845
"""Return the Revision object for a named revision."""
1846
return self.get_revisions([revision_id])[0]
1849
def get_revision_reconcile(self, revision_id):
1850
"""'reconcile' helper routine that allows access to a revision always.
1852
This variant of get_revision does not cross check the weave graph
1853
against the revision one as get_revision does: but it should only
1854
be used by reconcile, or reconcile-alike commands that are correcting
1855
or testing the revision graph.
1857
return self._get_revisions([revision_id])[0]
1860
def get_revisions(self, revision_ids):
1861
"""Get many revisions at once.
1863
Repositories that need to check data on every revision read should
1864
subclass this method.
1866
return self._get_revisions(revision_ids)
1869
def _get_revisions(self, revision_ids):
1870
"""Core work logic to get many revisions without sanity checks."""
1872
for revid, rev in self._iter_revisions(revision_ids):
1874
raise errors.NoSuchRevision(self, revid)
1876
return [revs[revid] for revid in revision_ids]
1878
def _iter_revisions(self, revision_ids):
1879
"""Iterate over revision objects.
1881
:param revision_ids: An iterable of revisions to examine. None may be
1882
passed to request all revisions known to the repository. Note that
1883
not all repositories can find unreferenced revisions; for those
1884
repositories only referenced ones will be returned.
1885
:return: An iterator of (revid, revision) tuples. Absent revisions (
1886
those asked for but not available) are returned as (revid, None).
1888
if revision_ids is None:
1889
revision_ids = self.all_revision_ids()
1891
for rev_id in revision_ids:
1892
if not rev_id or not isinstance(rev_id, basestring):
1893
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1894
keys = [(key,) for key in revision_ids]
1895
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1896
for record in stream:
1897
revid = record.key[0]
1898
if record.storage_kind == 'absent':
1901
text = record.get_bytes_as('fulltext')
1902
rev = self._serializer.read_revision_from_string(text)
1906
def get_revision_xml(self, revision_id):
1907
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1908
# would have already do it.
1909
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1910
# TODO: this can't just be replaced by:
1911
# return self._serializer.write_revision_to_string(
1912
# self.get_revision(revision_id))
1913
# as cStringIO preservers the encoding unlike write_revision_to_string
1914
# or some other call down the path.
1915
rev = self.get_revision(revision_id)
1916
rev_tmp = cStringIO.StringIO()
1917
# the current serializer..
1918
self._serializer.write_revision(rev, rev_tmp)
1920
return rev_tmp.getvalue()
1922
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1923
"""Produce a generator of revision deltas.
1925
Note that the input is a sequence of REVISIONS, not revision_ids.
1926
Trees will be held in memory until the generator exits.
1927
Each delta is relative to the revision's lefthand predecessor.
1929
:param specific_fileids: if not None, the result is filtered
1930
so that only those file-ids, their parents and their
1931
children are included.
1933
# Get the revision-ids of interest
1934
required_trees = set()
1935
for revision in revisions:
1936
required_trees.add(revision.revision_id)
1937
required_trees.update(revision.parent_ids[:1])
1939
# Get the matching filtered trees. Note that it's more
1940
# efficient to pass filtered trees to changes_from() rather
1941
# than doing the filtering afterwards. changes_from() could
1942
# arguably do the filtering itself but it's path-based, not
1943
# file-id based, so filtering before or afterwards is
1945
if specific_fileids is None:
1946
trees = dict((t.get_revision_id(), t) for
1947
t in self.revision_trees(required_trees))
1949
trees = dict((t.get_revision_id(), t) for
1950
t in self._filtered_revision_trees(required_trees,
1953
# Calculate the deltas
1954
for revision in revisions:
1955
if not revision.parent_ids:
1956
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1958
old_tree = trees[revision.parent_ids[0]]
1959
yield trees[revision.revision_id].changes_from(old_tree)
1962
def get_revision_delta(self, revision_id, specific_fileids=None):
1963
"""Return the delta for one revision.
1965
The delta is relative to the left-hand predecessor of the
1968
:param specific_fileids: if not None, the result is filtered
1969
so that only those file-ids, their parents and their
1970
children are included.
1972
r = self.get_revision(revision_id)
1973
return list(self.get_deltas_for_revisions([r],
1974
specific_fileids=specific_fileids))[0]
1977
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1978
signature = gpg_strategy.sign(plaintext)
1979
self.add_signature_text(revision_id, signature)
1982
def add_signature_text(self, revision_id, signature):
1983
self.signatures.add_lines((revision_id,), (),
1984
osutils.split_lines(signature))
1986
def find_text_key_references(self):
1987
"""Find the text key references within the repository.
1989
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1990
to whether they were referred to by the inventory of the
1991
revision_id that they contain. The inventory texts from all present
1992
revision ids are assessed to generate this report.
1994
revision_keys = self.revisions.keys()
1995
w = self.inventories
1996
pb = ui.ui_factory.nested_progress_bar()
1998
return self._find_text_key_references_from_xml_inventory_lines(
1999
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
2003
def _find_text_key_references_from_xml_inventory_lines(self,
2005
"""Core routine for extracting references to texts from inventories.
2007
This performs the translation of xml lines to revision ids.
2009
:param line_iterator: An iterator of lines, origin_version_id
2010
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2011
to whether they were referred to by the inventory of the
2012
revision_id that they contain. Note that if that revision_id was
2013
not part of the line_iterator's output then False will be given -
2014
even though it may actually refer to that key.
2016
if not self._serializer.support_altered_by_hack:
2017
raise AssertionError(
2018
"_find_text_key_references_from_xml_inventory_lines only "
2019
"supported for branches which store inventory as unnested xml"
2020
", not on %r" % self)
2023
# this code needs to read every new line in every inventory for the
2024
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2025
# not present in one of those inventories is unnecessary but not
2026
# harmful because we are filtering by the revision id marker in the
2027
# inventory lines : we only select file ids altered in one of those
2028
# revisions. We don't need to see all lines in the inventory because
2029
# only those added in an inventory in rev X can contain a revision=X
2031
unescape_revid_cache = {}
2032
unescape_fileid_cache = {}
2034
# jam 20061218 In a big fetch, this handles hundreds of thousands
2035
# of lines, so it has had a lot of inlining and optimizing done.
2036
# Sorry that it is a little bit messy.
2037
# Move several functions to be local variables, since this is a long
2039
search = self._file_ids_altered_regex.search
2040
unescape = _unescape_xml
2041
setdefault = result.setdefault
2042
for line, line_key in line_iterator:
2043
match = search(line)
2046
# One call to match.group() returning multiple items is quite a
2047
# bit faster than 2 calls to match.group() each returning 1
2048
file_id, revision_id = match.group('file_id', 'revision_id')
2050
# Inlining the cache lookups helps a lot when you make 170,000
2051
# lines and 350k ids, versus 8.4 unique ids.
2052
# Using a cache helps in 2 ways:
2053
# 1) Avoids unnecessary decoding calls
2054
# 2) Re-uses cached strings, which helps in future set and
2056
# (2) is enough that removing encoding entirely along with
2057
# the cache (so we are using plain strings) results in no
2058
# performance improvement.
2060
revision_id = unescape_revid_cache[revision_id]
2062
unescaped = unescape(revision_id)
2063
unescape_revid_cache[revision_id] = unescaped
2064
revision_id = unescaped
2066
# Note that unconditionally unescaping means that we deserialise
2067
# every fileid, which for general 'pull' is not great, but we don't
2068
# really want to have some many fulltexts that this matters anyway.
2071
file_id = unescape_fileid_cache[file_id]
2073
unescaped = unescape(file_id)
2074
unescape_fileid_cache[file_id] = unescaped
2077
key = (file_id, revision_id)
2078
setdefault(key, False)
2079
if revision_id == line_key[-1]:
2083
def _inventory_xml_lines_for_keys(self, keys):
2084
"""Get a line iterator of the sort needed for findind references.
2086
Not relevant for non-xml inventory repositories.
2088
Ghosts in revision_keys are ignored.
2090
:param revision_keys: The revision keys for the inventories to inspect.
2091
:return: An iterator over (inventory line, revid) for the fulltexts of
2092
all of the xml inventories specified by revision_keys.
2094
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2095
for record in stream:
2096
if record.storage_kind != 'absent':
2097
chunks = record.get_bytes_as('chunked')
2098
revid = record.key[-1]
2099
lines = osutils.chunks_to_lines(chunks)
2103
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2105
"""Helper routine for fileids_altered_by_revision_ids.
2107
This performs the translation of xml lines to revision ids.
2109
:param line_iterator: An iterator of lines, origin_version_id
2110
:param revision_keys: The revision ids to filter for. This should be a
2111
set or other type which supports efficient __contains__ lookups, as
2112
the revision key from each parsed line will be looked up in the
2113
revision_keys filter.
2114
:return: a dictionary mapping altered file-ids to an iterable of
2115
revision_ids. Each altered file-ids has the exact revision_ids that
2116
altered it listed explicitly.
2118
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2119
line_iterator).iterkeys())
2120
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2121
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2122
self._inventory_xml_lines_for_keys(parent_keys)))
2123
new_keys = seen - parent_seen
2125
setdefault = result.setdefault
2126
for key in new_keys:
2127
setdefault(key[0], set()).add(key[-1])
2130
def _find_parent_ids_of_revisions(self, revision_ids):
2131
"""Find all parent ids that are mentioned in the revision graph.
2133
:return: set of revisions that are parents of revision_ids which are
2134
not part of revision_ids themselves
2136
parent_map = self.get_parent_map(revision_ids)
2138
map(parent_ids.update, parent_map.itervalues())
2139
parent_ids.difference_update(revision_ids)
2140
parent_ids.discard(_mod_revision.NULL_REVISION)
2143
def _find_parent_keys_of_revisions(self, revision_keys):
2144
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2146
:param revision_keys: An iterable of revision_keys.
2147
:return: The parents of all revision_keys that are not already in
2150
parent_map = self.revisions.get_parent_map(revision_keys)
2152
map(parent_keys.update, parent_map.itervalues())
2153
parent_keys.difference_update(revision_keys)
2154
parent_keys.discard(_mod_revision.NULL_REVISION)
2157
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2158
"""Find the file ids and versions affected by revisions.
2160
:param revisions: an iterable containing revision ids.
2161
:param _inv_weave: The inventory weave from this repository or None.
2162
If None, the inventory weave will be opened automatically.
2163
:return: a dictionary mapping altered file-ids to an iterable of
2164
revision_ids. Each altered file-ids has the exact revision_ids that
2165
altered it listed explicitly.
2167
selected_keys = set((revid,) for revid in revision_ids)
2168
w = _inv_weave or self.inventories
2169
pb = ui.ui_factory.nested_progress_bar()
2171
return self._find_file_ids_from_xml_inventory_lines(
2172
w.iter_lines_added_or_present_in_keys(
2173
selected_keys, pb=pb),
2178
def iter_files_bytes(self, desired_files):
2179
"""Iterate through file versions.
2181
Files will not necessarily be returned in the order they occur in
2182
desired_files. No specific order is guaranteed.
2184
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2185
value supplied by the caller as part of desired_files. It should
2186
uniquely identify the file version in the caller's context. (Examples:
2187
an index number or a TreeTransform trans_id.)
2189
bytes_iterator is an iterable of bytestrings for the file. The
2190
kind of iterable and length of the bytestrings are unspecified, but for
2191
this implementation, it is a list of bytes produced by
2192
VersionedFile.get_record_stream().
2194
:param desired_files: a list of (file_id, revision_id, identifier)
2198
for file_id, revision_id, callable_data in desired_files:
2199
text_keys[(file_id, revision_id)] = callable_data
2200
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2201
if record.storage_kind == 'absent':
2202
raise errors.RevisionNotPresent(record.key, self)
2203
yield text_keys[record.key], record.get_bytes_as('chunked')
2205
def _generate_text_key_index(self, text_key_references=None,
2207
"""Generate a new text key index for the repository.
2209
This is an expensive function that will take considerable time to run.
2211
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2212
list of parents, also text keys. When a given key has no parents,
2213
the parents list will be [NULL_REVISION].
2215
# All revisions, to find inventory parents.
2216
if ancestors is None:
2217
graph = self.get_graph()
2218
ancestors = graph.get_parent_map(self.all_revision_ids())
2219
if text_key_references is None:
2220
text_key_references = self.find_text_key_references()
2221
pb = ui.ui_factory.nested_progress_bar()
2223
return self._do_generate_text_key_index(ancestors,
2224
text_key_references, pb)
2228
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2229
"""Helper for _generate_text_key_index to avoid deep nesting."""
2230
revision_order = tsort.topo_sort(ancestors)
2231
invalid_keys = set()
2233
for revision_id in revision_order:
2234
revision_keys[revision_id] = set()
2235
text_count = len(text_key_references)
2236
# a cache of the text keys to allow reuse; costs a dict of all the
2237
# keys, but saves a 2-tuple for every child of a given key.
2239
for text_key, valid in text_key_references.iteritems():
2241
invalid_keys.add(text_key)
2243
revision_keys[text_key[1]].add(text_key)
2244
text_key_cache[text_key] = text_key
2245
del text_key_references
2247
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2248
NULL_REVISION = _mod_revision.NULL_REVISION
2249
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2250
# too small for large or very branchy trees. However, for 55K path
2251
# trees, it would be easy to use too much memory trivially. Ideally we
2252
# could gauge this by looking at available real memory etc, but this is
2253
# always a tricky proposition.
2254
inventory_cache = lru_cache.LRUCache(10)
2255
batch_size = 10 # should be ~150MB on a 55K path tree
2256
batch_count = len(revision_order) / batch_size + 1
2258
pb.update("Calculating text parents", processed_texts, text_count)
2259
for offset in xrange(batch_count):
2260
to_query = revision_order[offset * batch_size:(offset + 1) *
2264
for revision_id in to_query:
2265
parent_ids = ancestors[revision_id]
2266
for text_key in revision_keys[revision_id]:
2267
pb.update("Calculating text parents", processed_texts)
2268
processed_texts += 1
2269
candidate_parents = []
2270
for parent_id in parent_ids:
2271
parent_text_key = (text_key[0], parent_id)
2273
check_parent = parent_text_key not in \
2274
revision_keys[parent_id]
2276
# the parent parent_id is a ghost:
2277
check_parent = False
2278
# truncate the derived graph against this ghost.
2279
parent_text_key = None
2281
# look at the parent commit details inventories to
2282
# determine possible candidates in the per file graph.
2285
inv = inventory_cache[parent_id]
2287
inv = self.revision_tree(parent_id).inventory
2288
inventory_cache[parent_id] = inv
2290
parent_entry = inv[text_key[0]]
2291
except (KeyError, errors.NoSuchId):
2293
if parent_entry is not None:
2295
text_key[0], parent_entry.revision)
2297
parent_text_key = None
2298
if parent_text_key is not None:
2299
candidate_parents.append(
2300
text_key_cache[parent_text_key])
2301
parent_heads = text_graph.heads(candidate_parents)
2302
new_parents = list(parent_heads)
2303
new_parents.sort(key=lambda x:candidate_parents.index(x))
2304
if new_parents == []:
2305
new_parents = [NULL_REVISION]
2306
text_index[text_key] = new_parents
2308
for text_key in invalid_keys:
2309
text_index[text_key] = [NULL_REVISION]
2312
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2313
"""Get an iterable listing the keys of all the data introduced by a set
2316
The keys will be ordered so that the corresponding items can be safely
2317
fetched and inserted in that order.
2319
:returns: An iterable producing tuples of (knit-kind, file-id,
2320
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2321
'revisions'. file-id is None unless knit-kind is 'file'.
2323
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2326
for result in self._find_non_file_keys_to_fetch(revision_ids):
2329
def _find_file_keys_to_fetch(self, revision_ids, pb):
2330
# XXX: it's a bit weird to control the inventory weave caching in this
2331
# generator. Ideally the caching would be done in fetch.py I think. Or
2332
# maybe this generator should explicitly have the contract that it
2333
# should not be iterated until the previously yielded item has been
2335
inv_w = self.inventories
2337
# file ids that changed
2338
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2340
num_file_ids = len(file_ids)
2341
for file_id, altered_versions in file_ids.iteritems():
2343
pb.update("Fetch texts", count, num_file_ids)
2345
yield ("file", file_id, altered_versions)
2347
def _find_non_file_keys_to_fetch(self, revision_ids):
2349
yield ("inventory", None, revision_ids)
2352
# XXX: Note ATM no callers actually pay attention to this return
2353
# instead they just use the list of revision ids and ignore
2354
# missing sigs. Consider removing this work entirely
2355
revisions_with_signatures = set(self.signatures.get_parent_map(
2356
[(r,) for r in revision_ids]))
2357
revisions_with_signatures = set(
2358
[r for (r,) in revisions_with_signatures])
2359
revisions_with_signatures.intersection_update(revision_ids)
2360
yield ("signatures", None, revisions_with_signatures)
2363
yield ("revisions", None, revision_ids)
2366
def get_inventory(self, revision_id):
2367
"""Get Inventory object by revision id."""
2368
return self.iter_inventories([revision_id]).next()
2370
def iter_inventories(self, revision_ids, ordering=None):
2371
"""Get many inventories by revision_ids.
2373
This will buffer some or all of the texts used in constructing the
2374
inventories in memory, but will only parse a single inventory at a
2377
:param revision_ids: The expected revision ids of the inventories.
2378
:param ordering: optional ordering, e.g. 'topological'. If not
2379
specified, the order of revision_ids will be preserved (by
2380
buffering if necessary).
2381
:return: An iterator of inventories.
2383
if ((None in revision_ids)
2384
or (_mod_revision.NULL_REVISION in revision_ids)):
2385
raise ValueError('cannot get null revision inventory')
2386
return self._iter_inventories(revision_ids, ordering)
2388
def _iter_inventories(self, revision_ids, ordering):
2389
"""single-document based inventory iteration."""
2390
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2391
for text, revision_id in inv_xmls:
2392
yield self.deserialise_inventory(revision_id, text)
2394
def _iter_inventory_xmls(self, revision_ids, ordering):
2395
if ordering is None:
2396
order_as_requested = True
2397
ordering = 'unordered'
2399
order_as_requested = False
2400
keys = [(revision_id,) for revision_id in revision_ids]
2403
if order_as_requested:
2404
key_iter = iter(keys)
2405
next_key = key_iter.next()
2406
stream = self.inventories.get_record_stream(keys, ordering, True)
2408
for record in stream:
2409
if record.storage_kind != 'absent':
2410
chunks = record.get_bytes_as('chunked')
2411
if order_as_requested:
2412
text_chunks[record.key] = chunks
2414
yield ''.join(chunks), record.key[-1]
2416
raise errors.NoSuchRevision(self, record.key)
2417
if order_as_requested:
2418
# Yield as many results as we can while preserving order.
2419
while next_key in text_chunks:
2420
chunks = text_chunks.pop(next_key)
2421
yield ''.join(chunks), next_key[-1]
2423
next_key = key_iter.next()
2424
except StopIteration:
2425
# We still want to fully consume the get_record_stream,
2426
# just in case it is not actually finished at this point
2430
def deserialise_inventory(self, revision_id, xml):
2431
"""Transform the xml into an inventory object.
2433
:param revision_id: The expected revision id of the inventory.
2434
:param xml: A serialised inventory.
2436
result = self._serializer.read_inventory_from_string(xml, revision_id,
2437
entry_cache=self._inventory_entry_cache,
2438
return_from_cache=self._safe_to_return_from_cache)
2439
if result.revision_id != revision_id:
2440
raise AssertionError('revision id mismatch %s != %s' % (
2441
result.revision_id, revision_id))
2444
def serialise_inventory(self, inv):
2445
return self._serializer.write_inventory_to_string(inv)
2447
def _serialise_inventory_to_lines(self, inv):
2448
return self._serializer.write_inventory_to_lines(inv)
2450
def get_serializer_format(self):
2451
return self._serializer.format_num
2454
def get_inventory_xml(self, revision_id):
2455
"""Get inventory XML as a file object."""
2456
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2458
text, revision_id = texts.next()
2459
except StopIteration:
2460
raise errors.HistoryMissing(self, 'inventory', revision_id)
2464
def get_inventory_sha1(self, revision_id):
2465
"""Return the sha1 hash of the inventory entry
2467
return self.get_revision(revision_id).inventory_sha1
2469
def get_rev_id_for_revno(self, revno, known_pair):
2470
"""Return the revision id of a revno, given a later (revno, revid)
2471
pair in the same history.
2473
:return: if found (True, revid). If the available history ran out
2474
before reaching the revno, then this returns
2475
(False, (closest_revno, closest_revid)).
2477
known_revno, known_revid = known_pair
2478
partial_history = [known_revid]
2479
distance_from_known = known_revno - revno
2480
if distance_from_known < 0:
2482
'requested revno (%d) is later than given known revno (%d)'
2483
% (revno, known_revno))
2486
self, partial_history, stop_index=distance_from_known)
2487
except errors.RevisionNotPresent, err:
2488
if err.revision_id == known_revid:
2489
# The start revision (known_revid) wasn't found.
2491
# This is a stacked repository with no fallbacks, or a there's a
2492
# left-hand ghost. Either way, even though the revision named in
2493
# the error isn't in this repo, we know it's the next step in this
2494
# left-hand history.
2495
partial_history.append(err.revision_id)
2496
if len(partial_history) <= distance_from_known:
2497
# Didn't find enough history to get a revid for the revno.
2498
earliest_revno = known_revno - len(partial_history) + 1
2499
return (False, (earliest_revno, partial_history[-1]))
2500
if len(partial_history) - 1 > distance_from_known:
2501
raise AssertionError('_iter_for_revno returned too much history')
2502
return (True, partial_history[-1])
2504
def iter_reverse_revision_history(self, revision_id):
2505
"""Iterate backwards through revision ids in the lefthand history
2507
:param revision_id: The revision id to start with. All its lefthand
2508
ancestors will be traversed.
2510
graph = self.get_graph()
2511
next_id = revision_id
2513
if next_id in (None, _mod_revision.NULL_REVISION):
2516
parents = graph.get_parent_map([next_id])[next_id]
2518
raise errors.RevisionNotPresent(next_id, self)
2520
if len(parents) == 0:
2523
next_id = parents[0]
2526
def get_revision_inventory(self, revision_id):
2527
"""Return inventory of a past revision."""
2528
# TODO: Unify this with get_inventory()
2529
# bzr 0.0.6 and later imposes the constraint that the inventory_id
2530
# must be the same as its revision, so this is trivial.
2531
if revision_id is None:
2532
# This does not make sense: if there is no revision,
2533
# then it is the current tree inventory surely ?!
2534
# and thus get_root_id() is something that looks at the last
2535
# commit on the branch, and the get_root_id is an inventory check.
2536
raise NotImplementedError
2537
# return Inventory(self.get_root_id())
2539
return self.get_inventory(revision_id)
2541
def is_shared(self):
2542
"""Return True if this repository is flagged as a shared repository."""
2543
raise NotImplementedError(self.is_shared)
2546
def reconcile(self, other=None, thorough=False):
2547
"""Reconcile this repository."""
2548
from bzrlib.reconcile import RepoReconciler
2549
reconciler = RepoReconciler(self, thorough=thorough)
2550
reconciler.reconcile()
2553
def _refresh_data(self):
2554
"""Helper called from lock_* to ensure coherency with disk.
2556
The default implementation does nothing; it is however possible
2557
for repositories to maintain loaded indices across multiple locks
2558
by checking inside their implementation of this method to see
2559
whether their indices are still valid. This depends of course on
2560
the disk format being validatable in this manner. This method is
2561
also called by the refresh_data() public interface to cause a refresh
2562
to occur while in a write lock so that data inserted by a smart server
2563
push operation is visible on the client's instance of the physical
2568
def revision_tree(self, revision_id):
2569
"""Return Tree for a revision on this branch.
2571
`revision_id` may be NULL_REVISION for the empty tree revision.
2573
revision_id = _mod_revision.ensure_null(revision_id)
2574
# TODO: refactor this to use an existing revision object
2575
# so we don't need to read it in twice.
2576
if revision_id == _mod_revision.NULL_REVISION:
2577
return RevisionTree(self, Inventory(root_id=None),
2578
_mod_revision.NULL_REVISION)
2580
inv = self.get_revision_inventory(revision_id)
2581
return RevisionTree(self, inv, revision_id)
2583
def revision_trees(self, revision_ids):
2584
"""Return Trees for revisions in this repository.
2586
:param revision_ids: a sequence of revision-ids;
2587
a revision-id may not be None or 'null:'
2589
inventories = self.iter_inventories(revision_ids)
2590
for inv in inventories:
2591
yield RevisionTree(self, inv, inv.revision_id)
2593
def _filtered_revision_trees(self, revision_ids, file_ids):
2594
"""Return Tree for a revision on this branch with only some files.
2596
:param revision_ids: a sequence of revision-ids;
2597
a revision-id may not be None or 'null:'
2598
:param file_ids: if not None, the result is filtered
2599
so that only those file-ids, their parents and their
2600
children are included.
2602
inventories = self.iter_inventories(revision_ids)
2603
for inv in inventories:
2604
# Should we introduce a FilteredRevisionTree class rather
2605
# than pre-filter the inventory here?
2606
filtered_inv = inv.filter(file_ids)
2607
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2610
def get_ancestry(self, revision_id, topo_sorted=True):
2611
"""Return a list of revision-ids integrated by a revision.
2613
The first element of the list is always None, indicating the origin
2614
revision. This might change when we have history horizons, or
2615
perhaps we should have a new API.
2617
This is topologically sorted.
2619
if _mod_revision.is_null(revision_id):
2621
if not self.has_revision(revision_id):
2622
raise errors.NoSuchRevision(self, revision_id)
2623
graph = self.get_graph()
2625
search = graph._make_breadth_first_searcher([revision_id])
2628
found, ghosts = search.next_with_ghosts()
2629
except StopIteration:
2632
if _mod_revision.NULL_REVISION in keys:
2633
keys.remove(_mod_revision.NULL_REVISION)
2635
parent_map = graph.get_parent_map(keys)
2636
keys = tsort.topo_sort(parent_map)
2637
return [None] + list(keys)
2639
def pack(self, hint=None):
2640
"""Compress the data within the repository.
2642
This operation only makes sense for some repository types. For other
2643
types it should be a no-op that just returns.
2645
This stub method does not require a lock, but subclasses should use
2646
@needs_write_lock as this is a long running call its reasonable to
2647
implicitly lock for the user.
2649
:param hint: If not supplied, the whole repository is packed.
2650
If supplied, the repository may use the hint parameter as a
2651
hint for the parts of the repository to pack. A hint can be
2652
obtained from the result of commit_write_group(). Out of
2653
date hints are simply ignored, because concurrent operations
2654
can obsolete them rapidly.
2657
def get_transaction(self):
2658
return self.control_files.get_transaction()
2660
def get_parent_map(self, revision_ids):
2661
"""See graph.StackedParentsProvider.get_parent_map"""
2662
# revisions index works in keys; this just works in revisions
2663
# therefore wrap and unwrap
2666
for revision_id in revision_ids:
2667
if revision_id == _mod_revision.NULL_REVISION:
2668
result[revision_id] = ()
2669
elif revision_id is None:
2670
raise ValueError('get_parent_map(None) is not valid')
2672
query_keys.append((revision_id ,))
2673
for ((revision_id,), parent_keys) in \
2674
self.revisions.get_parent_map(query_keys).iteritems():
2676
result[revision_id] = tuple([parent_revid
2677
for (parent_revid,) in parent_keys])
2679
result[revision_id] = (_mod_revision.NULL_REVISION,)
2682
def _make_parents_provider(self):
2685
def get_graph(self, other_repository=None):
2686
"""Return the graph walker for this repository format"""
2687
parents_provider = self._make_parents_provider()
2688
if (other_repository is not None and
2689
not self.has_same_location(other_repository)):
2690
parents_provider = graph.StackedParentsProvider(
2691
[parents_provider, other_repository._make_parents_provider()])
2692
return graph.Graph(parents_provider)
2694
def _get_versioned_file_checker(self, text_key_references=None,
2696
"""Return an object suitable for checking versioned files.
2698
:param text_key_references: if non-None, an already built
2699
dictionary mapping text keys ((fileid, revision_id) tuples)
2700
to whether they were referred to by the inventory of the
2701
revision_id that they contain. If None, this will be
2703
:param ancestors: Optional result from
2704
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2707
return _VersionedFileChecker(self,
2708
text_key_references=text_key_references, ancestors=ancestors)
2710
def revision_ids_to_search_result(self, result_set):
2711
"""Convert a set of revision ids to a graph SearchResult."""
2712
result_parents = set()
2713
for parents in self.get_graph().get_parent_map(
2714
result_set).itervalues():
2715
result_parents.update(parents)
2716
included_keys = result_set.intersection(result_parents)
2717
start_keys = result_set.difference(included_keys)
2718
exclude_keys = result_parents.difference(result_set)
2719
result = graph.SearchResult(start_keys, exclude_keys,
2720
len(result_set), result_set)
2724
def set_make_working_trees(self, new_value):
2725
"""Set the policy flag for making working trees when creating branches.
2727
This only applies to branches that use this repository.
2729
The default is 'True'.
2730
:param new_value: True to restore the default, False to disable making
2733
raise NotImplementedError(self.set_make_working_trees)
2735
def make_working_trees(self):
2736
"""Returns the policy for making working trees on new branches."""
2737
raise NotImplementedError(self.make_working_trees)
2740
def sign_revision(self, revision_id, gpg_strategy):
2741
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2742
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2745
def has_signature_for_revision_id(self, revision_id):
2746
"""Query for a revision signature for revision_id in the repository."""
2747
if not self.has_revision(revision_id):
2748
raise errors.NoSuchRevision(self, revision_id)
2749
sig_present = (1 == len(
2750
self.signatures.get_parent_map([(revision_id,)])))
2754
def get_signature_text(self, revision_id):
2755
"""Return the text for a signature."""
2756
stream = self.signatures.get_record_stream([(revision_id,)],
2758
record = stream.next()
2759
if record.storage_kind == 'absent':
2760
raise errors.NoSuchRevision(self, revision_id)
2761
return record.get_bytes_as('fulltext')
2764
def check(self, revision_ids=None, callback_refs=None, check_repo=True):
2765
"""Check consistency of all history of given revision_ids.
2767
Different repository implementations should override _check().
2769
:param revision_ids: A non-empty list of revision_ids whose ancestry
2770
will be checked. Typically the last revision_id of a branch.
2771
:param callback_refs: A dict of check-refs to resolve and callback
2772
the check/_check method on the items listed as wanting the ref.
2774
:param check_repo: If False do not check the repository contents, just
2775
calculate the data callback_refs requires and call them back.
2777
return self._check(revision_ids, callback_refs=callback_refs,
2778
check_repo=check_repo)
2780
def _check(self, revision_ids, callback_refs, check_repo):
2781
result = check.Check(self, check_repo=check_repo)
2782
result.check(callback_refs)
2785
def _warn_if_deprecated(self):
2786
global _deprecation_warning_done
2787
if _deprecation_warning_done:
2789
_deprecation_warning_done = True
2790
warning("Format %s for %s is deprecated - please use 'bzr upgrade' to get better performance"
2791
% (self._format, self.bzrdir.transport.base))
2793
def supports_rich_root(self):
2794
return self._format.rich_root_data
2796
def _check_ascii_revisionid(self, revision_id, method):
2797
"""Private helper for ascii-only repositories."""
2798
# weave repositories refuse to store revisionids that are non-ascii.
2799
if revision_id is not None:
2800
# weaves require ascii revision ids.
2801
if isinstance(revision_id, unicode):
2803
revision_id.encode('ascii')
2804
except UnicodeEncodeError:
2805
raise errors.NonAsciiRevisionId(method, self)
2808
revision_id.decode('ascii')
2809
except UnicodeDecodeError:
2810
raise errors.NonAsciiRevisionId(method, self)
2812
def revision_graph_can_have_wrong_parents(self):
2813
"""Is it possible for this repository to have a revision graph with
2816
If True, then this repository must also implement
2817
_find_inconsistent_revision_parents so that check and reconcile can
2818
check for inconsistencies before proceeding with other checks that may
2819
depend on the revision index being consistent.
2821
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2824
# remove these delegates a while after bzr 0.15
2825
def __make_delegated(name, from_module):
2826
def _deprecated_repository_forwarder():
2827
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2828
% (name, from_module),
2831
m = __import__(from_module, globals(), locals(), [name])
2833
return getattr(m, name)
2834
except AttributeError:
2835
raise AttributeError('module %s has no name %s'
2837
globals()[name] = _deprecated_repository_forwarder
2840
'AllInOneRepository',
2841
'WeaveMetaDirRepository',
2842
'PreSplitOutRepositoryFormat',
2843
'RepositoryFormat4',
2844
'RepositoryFormat5',
2845
'RepositoryFormat6',
2846
'RepositoryFormat7',
2848
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2852
'RepositoryFormatKnit',
2853
'RepositoryFormatKnit1',
2855
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2858
def install_revision(repository, rev, revision_tree):
2859
"""Install all revision data into a repository."""
2860
install_revisions(repository, [(rev, revision_tree, None)])
2863
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2864
"""Install all revision data into a repository.
2866
Accepts an iterable of revision, tree, signature tuples. The signature
2869
repository.start_write_group()
2871
inventory_cache = lru_cache.LRUCache(10)
2872
for n, (revision, revision_tree, signature) in enumerate(iterable):
2873
_install_revision(repository, revision, revision_tree, signature,
2876
pb.update('Transferring revisions', n + 1, num_revisions)
2878
repository.abort_write_group()
2881
repository.commit_write_group()
2884
def _install_revision(repository, rev, revision_tree, signature,
2886
"""Install all revision data into a repository."""
2887
present_parents = []
2889
for p_id in rev.parent_ids:
2890
if repository.has_revision(p_id):
2891
present_parents.append(p_id)
2892
parent_trees[p_id] = repository.revision_tree(p_id)
2894
parent_trees[p_id] = repository.revision_tree(
2895
_mod_revision.NULL_REVISION)
2897
inv = revision_tree.inventory
2898
entries = inv.iter_entries()
2899
# backwards compatibility hack: skip the root id.
2900
if not repository.supports_rich_root():
2901
path, root = entries.next()
2902
if root.revision != rev.revision_id:
2903
raise errors.IncompatibleRevision(repr(repository))
2905
for path, ie in entries:
2906
text_keys[(ie.file_id, ie.revision)] = ie
2907
text_parent_map = repository.texts.get_parent_map(text_keys)
2908
missing_texts = set(text_keys) - set(text_parent_map)
2909
# Add the texts that are not already present
2910
for text_key in missing_texts:
2911
ie = text_keys[text_key]
2913
# FIXME: TODO: The following loop overlaps/duplicates that done by
2914
# commit to determine parents. There is a latent/real bug here where
2915
# the parents inserted are not those commit would do - in particular
2916
# they are not filtered by heads(). RBC, AB
2917
for revision, tree in parent_trees.iteritems():
2918
if ie.file_id not in tree:
2920
parent_id = tree.inventory[ie.file_id].revision
2921
if parent_id in text_parents:
2923
text_parents.append((ie.file_id, parent_id))
2924
lines = revision_tree.get_file(ie.file_id).readlines()
2925
repository.texts.add_lines(text_key, text_parents, lines)
2927
# install the inventory
2928
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2929
# Cache this inventory
2930
inventory_cache[rev.revision_id] = inv
2932
basis_inv = inventory_cache[rev.parent_ids[0]]
2934
repository.add_inventory(rev.revision_id, inv, present_parents)
2936
delta = inv._make_delta(basis_inv)
2937
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2938
rev.revision_id, present_parents)
2940
repository.add_inventory(rev.revision_id, inv, present_parents)
2941
except errors.RevisionAlreadyPresent:
2943
if signature is not None:
2944
repository.add_signature_text(rev.revision_id, signature)
2945
repository.add_revision(rev.revision_id, rev, inv)
2948
class MetaDirRepository(Repository):
2949
"""Repositories in the new meta-dir layout.
2951
:ivar _transport: Transport for access to repository control files,
2952
typically pointing to .bzr/repository.
2955
def __init__(self, _format, a_bzrdir, control_files):
2956
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2957
self._transport = control_files._transport
2959
def is_shared(self):
2960
"""Return True if this repository is flagged as a shared repository."""
2961
return self._transport.has('shared-storage')
2964
def set_make_working_trees(self, new_value):
2965
"""Set the policy flag for making working trees when creating branches.
2967
This only applies to branches that use this repository.
2969
The default is 'True'.
2970
:param new_value: True to restore the default, False to disable making
2975
self._transport.delete('no-working-trees')
2976
except errors.NoSuchFile:
2979
self._transport.put_bytes('no-working-trees', '',
2980
mode=self.bzrdir._get_file_mode())
2982
def make_working_trees(self):
2983
"""Returns the policy for making working trees on new branches."""
2984
return not self._transport.has('no-working-trees')
2987
class MetaDirVersionedFileRepository(MetaDirRepository):
2988
"""Repositories in a meta-dir, that work via versioned file objects."""
2990
def __init__(self, _format, a_bzrdir, control_files):
2991
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2995
network_format_registry = registry.FormatRegistry()
2996
"""Registry of formats indexed by their network name.
2998
The network name for a repository format is an identifier that can be used when
2999
referring to formats with smart server operations. See
3000
RepositoryFormat.network_name() for more detail.
3004
format_registry = registry.FormatRegistry(network_format_registry)
3005
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
3007
This can contain either format instances themselves, or classes/factories that
3008
can be called to obtain one.
3012
#####################################################################
3013
# Repository Formats
3015
class RepositoryFormat(object):
3016
"""A repository format.
3018
Formats provide four things:
3019
* An initialization routine to construct repository data on disk.
3020
* a optional format string which is used when the BzrDir supports
3022
* an open routine which returns a Repository instance.
3023
* A network name for referring to the format in smart server RPC
3026
There is one and only one Format subclass for each on-disk format. But
3027
there can be one Repository subclass that is used for several different
3028
formats. The _format attribute on a Repository instance can be used to
3029
determine the disk format.
3031
Formats are placed in a registry by their format string for reference
3032
during opening. These should be subclasses of RepositoryFormat for
3035
Once a format is deprecated, just deprecate the initialize and open
3036
methods on the format class. Do not deprecate the object, as the
3037
object may be created even when a repository instance hasn't been
3040
Common instance attributes:
3041
_matchingbzrdir - the bzrdir format that the repository format was
3042
originally written to work with. This can be used if manually
3043
constructing a bzrdir and repository, or more commonly for test suite
3047
# Set to True or False in derived classes. True indicates that the format
3048
# supports ghosts gracefully.
3049
supports_ghosts = None
3050
# Can this repository be given external locations to lookup additional
3051
# data. Set to True or False in derived classes.
3052
supports_external_lookups = None
3053
# Does this format support CHK bytestring lookups. Set to True or False in
3055
supports_chks = None
3056
# Should commit add an inventory, or an inventory delta to the repository.
3057
_commit_inv_deltas = True
3058
# What order should fetch operations request streams in?
3059
# The default is unordered as that is the cheapest for an origin to
3061
_fetch_order = 'unordered'
3062
# Does this repository format use deltas that can be fetched as-deltas ?
3063
# (E.g. knits, where the knit deltas can be transplanted intact.
3064
# We default to False, which will ensure that enough data to get
3065
# a full text out of any fetch stream will be grabbed.
3066
_fetch_uses_deltas = False
3067
# Should fetch trigger a reconcile after the fetch? Only needed for
3068
# some repository formats that can suffer internal inconsistencies.
3069
_fetch_reconcile = False
3070
# Does this format have < O(tree_size) delta generation. Used to hint what
3071
# code path for commit, amongst other things.
3073
# Does doing a pack operation compress data? Useful for the pack UI command
3074
# (so if there is one pack, the operation can still proceed because it may
3075
# help), and for fetching when data won't have come from the same
3077
pack_compresses = False
3078
# Does the repository inventory storage understand references to trees?
3079
supports_tree_reference = None
3082
return "<%s>" % self.__class__.__name__
3084
def __eq__(self, other):
3085
# format objects are generally stateless
3086
return isinstance(other, self.__class__)
3088
def __ne__(self, other):
3089
return not self == other
3092
def find_format(klass, a_bzrdir):
3093
"""Return the format for the repository object in a_bzrdir.
3095
This is used by bzr native formats that have a "format" file in
3096
the repository. Other methods may be used by different types of
3100
transport = a_bzrdir.get_repository_transport(None)
3101
format_string = transport.get_bytes("format")
3102
return format_registry.get(format_string)
3103
except errors.NoSuchFile:
3104
raise errors.NoRepositoryPresent(a_bzrdir)
3106
raise errors.UnknownFormatError(format=format_string,
3110
def register_format(klass, format):
3111
format_registry.register(format.get_format_string(), format)
3114
def unregister_format(klass, format):
3115
format_registry.remove(format.get_format_string())
3118
def get_default_format(klass):
3119
"""Return the current default format."""
3120
from bzrlib import bzrdir
3121
return bzrdir.format_registry.make_bzrdir('default').repository_format
3123
def get_format_string(self):
3124
"""Return the ASCII format string that identifies this format.
3126
Note that in pre format ?? repositories the format string is
3127
not permitted nor written to disk.
3129
raise NotImplementedError(self.get_format_string)
3131
def get_format_description(self):
3132
"""Return the short description for this format."""
3133
raise NotImplementedError(self.get_format_description)
3135
# TODO: this shouldn't be in the base class, it's specific to things that
3136
# use weaves or knits -- mbp 20070207
3137
def _get_versioned_file_store(self,
3142
versionedfile_class=None,
3143
versionedfile_kwargs={},
3145
if versionedfile_class is None:
3146
versionedfile_class = self._versionedfile_class
3147
weave_transport = control_files._transport.clone(name)
3148
dir_mode = control_files._dir_mode
3149
file_mode = control_files._file_mode
3150
return VersionedFileStore(weave_transport, prefixed=prefixed,
3152
file_mode=file_mode,
3153
versionedfile_class=versionedfile_class,
3154
versionedfile_kwargs=versionedfile_kwargs,
3157
def initialize(self, a_bzrdir, shared=False):
3158
"""Initialize a repository of this format in a_bzrdir.
3160
:param a_bzrdir: The bzrdir to put the new repository in it.
3161
:param shared: The repository should be initialized as a sharable one.
3162
:returns: The new repository object.
3164
This may raise UninitializableFormat if shared repository are not
3165
compatible the a_bzrdir.
3167
raise NotImplementedError(self.initialize)
3169
def is_supported(self):
3170
"""Is this format supported?
3172
Supported formats must be initializable and openable.
3173
Unsupported formats may not support initialization or committing or
3174
some other features depending on the reason for not being supported.
3178
def network_name(self):
3179
"""A simple byte string uniquely identifying this format for RPC calls.
3181
MetaDir repository formats use their disk format string to identify the
3182
repository over the wire. All in one formats such as bzr < 0.8, and
3183
foreign formats like svn/git and hg should use some marker which is
3184
unique and immutable.
3186
raise NotImplementedError(self.network_name)
3188
def check_conversion_target(self, target_format):
3189
if self.rich_root_data and not target_format.rich_root_data:
3190
raise errors.BadConversionTarget(
3191
'Does not support rich root data.', target_format,
3193
if (self.supports_tree_reference and
3194
not getattr(target_format, 'supports_tree_reference', False)):
3195
raise errors.BadConversionTarget(
3196
'Does not support nested trees', target_format,
3199
def open(self, a_bzrdir, _found=False):
3200
"""Return an instance of this format for the bzrdir a_bzrdir.
3202
_found is a private parameter, do not use it.
3204
raise NotImplementedError(self.open)
3207
class MetaDirRepositoryFormat(RepositoryFormat):
3208
"""Common base class for the new repositories using the metadir layout."""
3210
rich_root_data = False
3211
supports_tree_reference = False
3212
supports_external_lookups = False
3215
def _matchingbzrdir(self):
3216
matching = bzrdir.BzrDirMetaFormat1()
3217
matching.repository_format = self
3221
super(MetaDirRepositoryFormat, self).__init__()
3223
def _create_control_files(self, a_bzrdir):
3224
"""Create the required files and the initial control_files object."""
3225
# FIXME: RBC 20060125 don't peek under the covers
3226
# NB: no need to escape relative paths that are url safe.
3227
repository_transport = a_bzrdir.get_repository_transport(self)
3228
control_files = lockable_files.LockableFiles(repository_transport,
3229
'lock', lockdir.LockDir)
3230
control_files.create_lock()
3231
return control_files
3233
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
3234
"""Upload the initial blank content."""
3235
control_files = self._create_control_files(a_bzrdir)
3236
control_files.lock_write()
3237
transport = control_files._transport
3239
utf8_files += [('shared-storage', '')]
3241
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
3242
for (filename, content_stream) in files:
3243
transport.put_file(filename, content_stream,
3244
mode=a_bzrdir._get_file_mode())
3245
for (filename, content_bytes) in utf8_files:
3246
transport.put_bytes_non_atomic(filename, content_bytes,
3247
mode=a_bzrdir._get_file_mode())
3249
control_files.unlock()
3251
def network_name(self):
3252
"""Metadir formats have matching disk and network format strings."""
3253
return self.get_format_string()
3256
# Pre-0.8 formats that don't have a disk format string (because they are
3257
# versioned by the matching control directory). We use the control directories
3258
# disk format string as a key for the network_name because they meet the
3259
# constraints (simple string, unique, immutable).
3260
network_format_registry.register_lazy(
3261
"Bazaar-NG branch, format 5\n",
3262
'bzrlib.repofmt.weaverepo',
3263
'RepositoryFormat5',
3265
network_format_registry.register_lazy(
3266
"Bazaar-NG branch, format 6\n",
3267
'bzrlib.repofmt.weaverepo',
3268
'RepositoryFormat6',
3271
# formats which have no format string are not discoverable or independently
3272
# creatable on disk, so are not registered in format_registry. They're
3273
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
3274
# needed, it's constructed directly by the BzrDir. Non-native formats where
3275
# the repository is not separately opened are similar.
3277
format_registry.register_lazy(
3278
'Bazaar-NG Repository format 7',
3279
'bzrlib.repofmt.weaverepo',
3283
format_registry.register_lazy(
3284
'Bazaar-NG Knit Repository Format 1',
3285
'bzrlib.repofmt.knitrepo',
3286
'RepositoryFormatKnit1',
3289
format_registry.register_lazy(
3290
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
3291
'bzrlib.repofmt.knitrepo',
3292
'RepositoryFormatKnit3',
3295
format_registry.register_lazy(
3296
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
3297
'bzrlib.repofmt.knitrepo',
3298
'RepositoryFormatKnit4',
3301
# Pack-based formats. There is one format for pre-subtrees, and one for
3302
# post-subtrees to allow ease of testing.
3303
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
3304
format_registry.register_lazy(
3305
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
3306
'bzrlib.repofmt.pack_repo',
3307
'RepositoryFormatKnitPack1',
3309
format_registry.register_lazy(
3310
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
3311
'bzrlib.repofmt.pack_repo',
3312
'RepositoryFormatKnitPack3',
3314
format_registry.register_lazy(
3315
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
3316
'bzrlib.repofmt.pack_repo',
3317
'RepositoryFormatKnitPack4',
3319
format_registry.register_lazy(
3320
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
3321
'bzrlib.repofmt.pack_repo',
3322
'RepositoryFormatKnitPack5',
3324
format_registry.register_lazy(
3325
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
3326
'bzrlib.repofmt.pack_repo',
3327
'RepositoryFormatKnitPack5RichRoot',
3329
format_registry.register_lazy(
3330
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3331
'bzrlib.repofmt.pack_repo',
3332
'RepositoryFormatKnitPack5RichRootBroken',
3334
format_registry.register_lazy(
3335
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3336
'bzrlib.repofmt.pack_repo',
3337
'RepositoryFormatKnitPack6',
3339
format_registry.register_lazy(
3340
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3341
'bzrlib.repofmt.pack_repo',
3342
'RepositoryFormatKnitPack6RichRoot',
3345
# Development formats.
3346
# Obsolete but kept pending a CHK based subtree format.
3347
format_registry.register_lazy(
3348
("Bazaar development format 2 with subtree support "
3349
"(needs bzr.dev from before 1.8)\n"),
3350
'bzrlib.repofmt.pack_repo',
3351
'RepositoryFormatPackDevelopment2Subtree',
3354
# 1.14->1.16 go below here
3355
format_registry.register_lazy(
3356
'Bazaar development format - group compression and chk inventory'
3357
' (needs bzr.dev from 1.14)\n',
3358
'bzrlib.repofmt.groupcompress_repo',
3359
'RepositoryFormatCHK1',
3362
format_registry.register_lazy(
3363
'Bazaar development format - chk repository with bencode revision '
3364
'serialization (needs bzr.dev from 1.16)\n',
3365
'bzrlib.repofmt.groupcompress_repo',
3366
'RepositoryFormatCHK2',
3368
format_registry.register_lazy(
3369
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3370
'bzrlib.repofmt.groupcompress_repo',
3371
'RepositoryFormat2a',
3375
class InterRepository(InterObject):
3376
"""This class represents operations taking place between two repositories.
3378
Its instances have methods like copy_content and fetch, and contain
3379
references to the source and target repositories these operations can be
3382
Often we will provide convenience methods on 'repository' which carry out
3383
operations with another repository - they will always forward to
3384
InterRepository.get(other).method_name(parameters).
3387
_walk_to_common_revisions_batch_size = 50
3389
"""The available optimised InterRepository types."""
3392
def copy_content(self, revision_id=None):
3393
"""Make a complete copy of the content in self into destination.
3395
This is a destructive operation! Do not use it on existing
3398
:param revision_id: Only copy the content needed to construct
3399
revision_id and its parents.
3402
self.target.set_make_working_trees(self.source.make_working_trees())
3403
except NotImplementedError:
3405
self.target.fetch(self.source, revision_id=revision_id)
3408
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3410
"""Fetch the content required to construct revision_id.
3412
The content is copied from self.source to self.target.
3414
:param revision_id: if None all content is copied, if NULL_REVISION no
3416
:param pb: optional progress bar to use for progress reports. If not
3417
provided a default one will be created.
3420
f = _mod_fetch.RepoFetcher(to_repository=self.target,
3421
from_repository=self.source,
3422
last_revision=revision_id,
3423
fetch_spec=fetch_spec,
3424
pb=pb, find_ghosts=find_ghosts)
3426
def _walk_to_common_revisions(self, revision_ids):
3427
"""Walk out from revision_ids in source to revisions target has.
3429
:param revision_ids: The start point for the search.
3430
:return: A set of revision ids.
3432
target_graph = self.target.get_graph()
3433
revision_ids = frozenset(revision_ids)
3434
missing_revs = set()
3435
source_graph = self.source.get_graph()
3436
# ensure we don't pay silly lookup costs.
3437
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3438
null_set = frozenset([_mod_revision.NULL_REVISION])
3439
searcher_exhausted = False
3443
# Iterate the searcher until we have enough next_revs
3444
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3446
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3447
next_revs.update(next_revs_part)
3448
ghosts.update(ghosts_part)
3449
except StopIteration:
3450
searcher_exhausted = True
3452
# If there are ghosts in the source graph, and the caller asked for
3453
# them, make sure that they are present in the target.
3454
# We don't care about other ghosts as we can't fetch them and
3455
# haven't been asked to.
3456
ghosts_to_check = set(revision_ids.intersection(ghosts))
3457
revs_to_get = set(next_revs).union(ghosts_to_check)
3459
have_revs = set(target_graph.get_parent_map(revs_to_get))
3460
# we always have NULL_REVISION present.
3461
have_revs = have_revs.union(null_set)
3462
# Check if the target is missing any ghosts we need.
3463
ghosts_to_check.difference_update(have_revs)
3465
# One of the caller's revision_ids is a ghost in both the
3466
# source and the target.
3467
raise errors.NoSuchRevision(
3468
self.source, ghosts_to_check.pop())
3469
missing_revs.update(next_revs - have_revs)
3470
# Because we may have walked past the original stop point, make
3471
# sure everything is stopped
3472
stop_revs = searcher.find_seen_ancestors(have_revs)
3473
searcher.stop_searching_any(stop_revs)
3474
if searcher_exhausted:
3476
return searcher.get_result()
3479
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3480
"""Return the revision ids that source has that target does not.
3482
:param revision_id: only return revision ids included by this
3484
:param find_ghosts: If True find missing revisions in deep history
3485
rather than just finding the surface difference.
3486
:return: A bzrlib.graph.SearchResult.
3488
# stop searching at found target revisions.
3489
if not find_ghosts and revision_id is not None:
3490
return self._walk_to_common_revisions([revision_id])
3491
# generic, possibly worst case, slow code path.
3492
target_ids = set(self.target.all_revision_ids())
3493
if revision_id is not None:
3494
source_ids = self.source.get_ancestry(revision_id)
3495
if source_ids[0] is not None:
3496
raise AssertionError()
3499
source_ids = self.source.all_revision_ids()
3500
result_set = set(source_ids).difference(target_ids)
3501
return self.source.revision_ids_to_search_result(result_set)
3504
def _same_model(source, target):
3505
"""True if source and target have the same data representation.
3507
Note: this is always called on the base class; overriding it in a
3508
subclass will have no effect.
3511
InterRepository._assert_same_model(source, target)
3513
except errors.IncompatibleRepositories, e:
3517
def _assert_same_model(source, target):
3518
"""Raise an exception if two repositories do not use the same model.
3520
if source.supports_rich_root() != target.supports_rich_root():
3521
raise errors.IncompatibleRepositories(source, target,
3522
"different rich-root support")
3523
if source._serializer != target._serializer:
3524
raise errors.IncompatibleRepositories(source, target,
3525
"different serializers")
3528
class InterSameDataRepository(InterRepository):
3529
"""Code for converting between repositories that represent the same data.
3531
Data format and model must match for this to work.
3535
def _get_repo_format_to_test(self):
3536
"""Repository format for testing with.
3538
InterSameData can pull from subtree to subtree and from non-subtree to
3539
non-subtree, so we test this with the richest repository format.
3541
from bzrlib.repofmt import knitrepo
3542
return knitrepo.RepositoryFormatKnit3()
3545
def is_compatible(source, target):
3546
return InterRepository._same_model(source, target)
3549
class InterWeaveRepo(InterSameDataRepository):
3550
"""Optimised code paths between Weave based repositories.
3552
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3553
implemented lazy inter-object optimisation.
3557
def _get_repo_format_to_test(self):
3558
from bzrlib.repofmt import weaverepo
3559
return weaverepo.RepositoryFormat7()
3562
def is_compatible(source, target):
3563
"""Be compatible with known Weave formats.
3565
We don't test for the stores being of specific types because that
3566
could lead to confusing results, and there is no need to be
3569
from bzrlib.repofmt.weaverepo import (
3575
return (isinstance(source._format, (RepositoryFormat5,
3577
RepositoryFormat7)) and
3578
isinstance(target._format, (RepositoryFormat5,
3580
RepositoryFormat7)))
3581
except AttributeError:
3585
def copy_content(self, revision_id=None):
3586
"""See InterRepository.copy_content()."""
3587
# weave specific optimised path:
3589
self.target.set_make_working_trees(self.source.make_working_trees())
3590
except (errors.RepositoryUpgradeRequired, NotImplemented):
3592
# FIXME do not peek!
3593
if self.source._transport.listable():
3594
pb = ui.ui_factory.nested_progress_bar()
3596
self.target.texts.insert_record_stream(
3597
self.source.texts.get_record_stream(
3598
self.source.texts.keys(), 'topological', False))
3599
pb.update('Copying inventory', 0, 1)
3600
self.target.inventories.insert_record_stream(
3601
self.source.inventories.get_record_stream(
3602
self.source.inventories.keys(), 'topological', False))
3603
self.target.signatures.insert_record_stream(
3604
self.source.signatures.get_record_stream(
3605
self.source.signatures.keys(),
3607
self.target.revisions.insert_record_stream(
3608
self.source.revisions.get_record_stream(
3609
self.source.revisions.keys(),
3610
'topological', True))
3614
self.target.fetch(self.source, revision_id=revision_id)
3617
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3618
"""See InterRepository.missing_revision_ids()."""
3619
# we want all revisions to satisfy revision_id in source.
3620
# but we don't want to stat every file here and there.
3621
# we want then, all revisions other needs to satisfy revision_id
3622
# checked, but not those that we have locally.
3623
# so the first thing is to get a subset of the revisions to
3624
# satisfy revision_id in source, and then eliminate those that
3625
# we do already have.
3626
# this is slow on high latency connection to self, but as this
3627
# disk format scales terribly for push anyway due to rewriting
3628
# inventory.weave, this is considered acceptable.
3630
if revision_id is not None:
3631
source_ids = self.source.get_ancestry(revision_id)
3632
if source_ids[0] is not None:
3633
raise AssertionError()
3636
source_ids = self.source._all_possible_ids()
3637
source_ids_set = set(source_ids)
3638
# source_ids is the worst possible case we may need to pull.
3639
# now we want to filter source_ids against what we actually
3640
# have in target, but don't try to check for existence where we know
3641
# we do not have a revision as that would be pointless.
3642
target_ids = set(self.target._all_possible_ids())
3643
possibly_present_revisions = target_ids.intersection(source_ids_set)
3644
actually_present_revisions = set(
3645
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3646
required_revisions = source_ids_set.difference(actually_present_revisions)
3647
if revision_id is not None:
3648
# we used get_ancestry to determine source_ids then we are assured all
3649
# revisions referenced are present as they are installed in topological order.
3650
# and the tip revision was validated by get_ancestry.
3651
result_set = required_revisions
3653
# if we just grabbed the possibly available ids, then
3654
# we only have an estimate of whats available and need to validate
3655
# that against the revision records.
3657
self.source._eliminate_revisions_not_present(required_revisions))
3658
return self.source.revision_ids_to_search_result(result_set)
3661
class InterKnitRepo(InterSameDataRepository):
3662
"""Optimised code paths between Knit based repositories."""
3665
def _get_repo_format_to_test(self):
3666
from bzrlib.repofmt import knitrepo
3667
return knitrepo.RepositoryFormatKnit1()
3670
def is_compatible(source, target):
3671
"""Be compatible with known Knit formats.
3673
We don't test for the stores being of specific types because that
3674
could lead to confusing results, and there is no need to be
3677
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3679
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3680
isinstance(target._format, RepositoryFormatKnit))
3681
except AttributeError:
3683
return are_knits and InterRepository._same_model(source, target)
3686
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3687
"""See InterRepository.missing_revision_ids()."""
3688
if revision_id is not None:
3689
source_ids = self.source.get_ancestry(revision_id)
3690
if source_ids[0] is not None:
3691
raise AssertionError()
3694
source_ids = self.source.all_revision_ids()
3695
source_ids_set = set(source_ids)
3696
# source_ids is the worst possible case we may need to pull.
3697
# now we want to filter source_ids against what we actually
3698
# have in target, but don't try to check for existence where we know
3699
# we do not have a revision as that would be pointless.
3700
target_ids = set(self.target.all_revision_ids())
3701
possibly_present_revisions = target_ids.intersection(source_ids_set)
3702
actually_present_revisions = set(
3703
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3704
required_revisions = source_ids_set.difference(actually_present_revisions)
3705
if revision_id is not None:
3706
# we used get_ancestry to determine source_ids then we are assured all
3707
# revisions referenced are present as they are installed in topological order.
3708
# and the tip revision was validated by get_ancestry.
3709
result_set = required_revisions
3711
# if we just grabbed the possibly available ids, then
3712
# we only have an estimate of whats available and need to validate
3713
# that against the revision records.
3715
self.source._eliminate_revisions_not_present(required_revisions))
3716
return self.source.revision_ids_to_search_result(result_set)
3719
class InterDifferingSerializer(InterRepository):
3722
def _get_repo_format_to_test(self):
3726
def is_compatible(source, target):
3727
"""Be compatible with Knit2 source and Knit3 target"""
3728
# This is redundant with format.check_conversion_target(), however that
3729
# raises an exception, and we just want to say "False" as in we won't
3730
# support converting between these formats.
3731
if 'IDS_never' in debug.debug_flags:
3733
if source.supports_rich_root() and not target.supports_rich_root():
3735
if (source._format.supports_tree_reference
3736
and not target._format.supports_tree_reference):
3738
if target._fallback_repositories and target._format.supports_chks:
3739
# IDS doesn't know how to copy CHKs for the parent inventories it
3740
# adds to stacked repos.
3742
if 'IDS_always' in debug.debug_flags:
3744
# Only use this code path for local source and target. IDS does far
3745
# too much IO (both bandwidth and roundtrips) over a network.
3746
if not source.bzrdir.transport.base.startswith('file:///'):
3748
if not target.bzrdir.transport.base.startswith('file:///'):
3752
def _get_trees(self, revision_ids, cache):
3754
for rev_id in revision_ids:
3756
possible_trees.append((rev_id, cache[rev_id]))
3758
# Not cached, but inventory might be present anyway.
3760
tree = self.source.revision_tree(rev_id)
3761
except errors.NoSuchRevision:
3762
# Nope, parent is ghost.
3765
cache[rev_id] = tree
3766
possible_trees.append((rev_id, tree))
3767
return possible_trees
3769
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3770
"""Get the best delta and base for this revision.
3772
:return: (basis_id, delta)
3775
# Generate deltas against each tree, to find the shortest.
3776
texts_possibly_new_in_tree = set()
3777
for basis_id, basis_tree in possible_trees:
3778
delta = tree.inventory._make_delta(basis_tree.inventory)
3779
for old_path, new_path, file_id, new_entry in delta:
3780
if new_path is None:
3781
# This file_id isn't present in the new rev, so we don't
3785
# Rich roots are handled elsewhere...
3787
kind = new_entry.kind
3788
if kind != 'directory' and kind != 'file':
3789
# No text record associated with this inventory entry.
3791
# This is a directory or file that has changed somehow.
3792
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3793
deltas.append((len(delta), basis_id, delta))
3795
return deltas[0][1:]
3797
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3798
"""Find all parent revisions that are absent, but for which the
3799
inventory is present, and copy those inventories.
3801
This is necessary to preserve correctness when the source is stacked
3802
without fallbacks configured. (Note that in cases like upgrade the
3803
source may be not have _fallback_repositories even though it is
3807
for parents in parent_map.values():
3808
parent_revs.update(parents)
3809
present_parents = self.source.get_parent_map(parent_revs)
3810
absent_parents = set(parent_revs).difference(present_parents)
3811
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3812
(rev_id,) for rev_id in absent_parents)
3813
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3814
for parent_tree in self.source.revision_trees(parent_inv_ids):
3815
current_revision_id = parent_tree.get_revision_id()
3816
parents_parents_keys = parent_invs_keys_for_stacking[
3817
(current_revision_id,)]
3818
parents_parents = [key[-1] for key in parents_parents_keys]
3819
basis_id = _mod_revision.NULL_REVISION
3820
basis_tree = self.source.revision_tree(basis_id)
3821
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3822
self.target.add_inventory_by_delta(
3823
basis_id, delta, current_revision_id, parents_parents)
3824
cache[current_revision_id] = parent_tree
3826
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3827
"""Fetch across a few revisions.
3829
:param revision_ids: The revisions to copy
3830
:param basis_id: The revision_id of a tree that must be in cache, used
3831
as a basis for delta when no other base is available
3832
:param cache: A cache of RevisionTrees that we can use.
3833
:param a_graph: A Graph object to determine the heads() of the
3834
rich-root data stream.
3835
:return: The revision_id of the last converted tree. The RevisionTree
3836
for it will be in cache
3838
# Walk though all revisions; get inventory deltas, copy referenced
3839
# texts that delta references, insert the delta, revision and
3841
root_keys_to_create = set()
3844
pending_revisions = []
3845
parent_map = self.source.get_parent_map(revision_ids)
3846
self._fetch_parent_invs_for_stacking(parent_map, cache)
3847
self.source._safe_to_return_from_cache = True
3848
for tree in self.source.revision_trees(revision_ids):
3849
# Find a inventory delta for this revision.
3850
# Find text entries that need to be copied, too.
3851
current_revision_id = tree.get_revision_id()
3852
parent_ids = parent_map.get(current_revision_id, ())
3853
parent_trees = self._get_trees(parent_ids, cache)
3854
possible_trees = list(parent_trees)
3855
if len(possible_trees) == 0:
3856
# There either aren't any parents, or the parents are ghosts,
3857
# so just use the last converted tree.
3858
possible_trees.append((basis_id, cache[basis_id]))
3859
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3861
revision = self.source.get_revision(current_revision_id)
3862
pending_deltas.append((basis_id, delta,
3863
current_revision_id, revision.parent_ids))
3864
if self._converting_to_rich_root:
3865
self._revision_id_to_root_id[current_revision_id] = \
3867
# Determine which texts are in present in this revision but not in
3868
# any of the available parents.
3869
texts_possibly_new_in_tree = set()
3870
for old_path, new_path, file_id, entry in delta:
3871
if new_path is None:
3872
# This file_id isn't present in the new rev
3876
if not self.target.supports_rich_root():
3877
# The target doesn't support rich root, so we don't
3880
if self._converting_to_rich_root:
3881
# This can't be copied normally, we have to insert
3883
root_keys_to_create.add((file_id, entry.revision))
3886
texts_possibly_new_in_tree.add((file_id, entry.revision))
3887
for basis_id, basis_tree in possible_trees:
3888
basis_inv = basis_tree.inventory
3889
for file_key in list(texts_possibly_new_in_tree):
3890
file_id, file_revision = file_key
3892
entry = basis_inv[file_id]
3893
except errors.NoSuchId:
3895
if entry.revision == file_revision:
3896
texts_possibly_new_in_tree.remove(file_key)
3897
text_keys.update(texts_possibly_new_in_tree)
3898
pending_revisions.append(revision)
3899
cache[current_revision_id] = tree
3900
basis_id = current_revision_id
3901
self.source._safe_to_return_from_cache = False
3903
from_texts = self.source.texts
3904
to_texts = self.target.texts
3905
if root_keys_to_create:
3906
root_stream = _mod_fetch._new_root_data_stream(
3907
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3908
self.source, graph=a_graph)
3909
to_texts.insert_record_stream(root_stream)
3910
to_texts.insert_record_stream(from_texts.get_record_stream(
3911
text_keys, self.target._format._fetch_order,
3912
not self.target._format._fetch_uses_deltas))
3913
# insert inventory deltas
3914
for delta in pending_deltas:
3915
self.target.add_inventory_by_delta(*delta)
3916
if self.target._fallback_repositories:
3917
# Make sure this stacked repository has all the parent inventories
3918
# for the new revisions that we are about to insert. We do this
3919
# before adding the revisions so that no revision is added until
3920
# all the inventories it may depend on are added.
3921
# Note that this is overzealous, as we may have fetched these in an
3924
revision_ids = set()
3925
for revision in pending_revisions:
3926
revision_ids.add(revision.revision_id)
3927
parent_ids.update(revision.parent_ids)
3928
parent_ids.difference_update(revision_ids)
3929
parent_ids.discard(_mod_revision.NULL_REVISION)
3930
parent_map = self.source.get_parent_map(parent_ids)
3931
# we iterate over parent_map and not parent_ids because we don't
3932
# want to try copying any revision which is a ghost
3933
for parent_tree in self.source.revision_trees(parent_map):
3934
current_revision_id = parent_tree.get_revision_id()
3935
parents_parents = parent_map[current_revision_id]
3936
possible_trees = self._get_trees(parents_parents, cache)
3937
if len(possible_trees) == 0:
3938
# There either aren't any parents, or the parents are
3939
# ghosts, so just use the last converted tree.
3940
possible_trees.append((basis_id, cache[basis_id]))
3941
basis_id, delta = self._get_delta_for_revision(parent_tree,
3942
parents_parents, possible_trees)
3943
self.target.add_inventory_by_delta(
3944
basis_id, delta, current_revision_id, parents_parents)
3945
# insert signatures and revisions
3946
for revision in pending_revisions:
3948
signature = self.source.get_signature_text(
3949
revision.revision_id)
3950
self.target.add_signature_text(revision.revision_id,
3952
except errors.NoSuchRevision:
3954
self.target.add_revision(revision.revision_id, revision)
3957
def _fetch_all_revisions(self, revision_ids, pb):
3958
"""Fetch everything for the list of revisions.
3960
:param revision_ids: The list of revisions to fetch. Must be in
3962
:param pb: A ProgressTask
3965
basis_id, basis_tree = self._get_basis(revision_ids[0])
3967
cache = lru_cache.LRUCache(100)
3968
cache[basis_id] = basis_tree
3969
del basis_tree # We don't want to hang on to it here
3971
if self._converting_to_rich_root and len(revision_ids) > 100:
3972
a_graph = _mod_fetch._get_rich_root_heads_graph(self.source,
3977
for offset in range(0, len(revision_ids), batch_size):
3978
self.target.start_write_group()
3980
pb.update('Transferring revisions', offset,
3982
batch = revision_ids[offset:offset+batch_size]
3983
basis_id = self._fetch_batch(batch, basis_id, cache,
3986
self.source._safe_to_return_from_cache = False
3987
self.target.abort_write_group()
3990
hint = self.target.commit_write_group()
3993
if hints and self.target._format.pack_compresses:
3994
self.target.pack(hint=hints)
3995
pb.update('Transferring revisions', len(revision_ids),
3999
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
4001
"""See InterRepository.fetch()."""
4002
if fetch_spec is not None:
4003
raise AssertionError("Not implemented yet...")
4004
if (not self.source.supports_rich_root()
4005
and self.target.supports_rich_root()):
4006
self._converting_to_rich_root = True
4007
self._revision_id_to_root_id = {}
4009
self._converting_to_rich_root = False
4010
revision_ids = self.target.search_missing_revision_ids(self.source,
4011
revision_id, find_ghosts=find_ghosts).get_keys()
4012
if not revision_ids:
4014
revision_ids = tsort.topo_sort(
4015
self.source.get_graph().get_parent_map(revision_ids))
4016
if not revision_ids:
4018
# Walk though all revisions; get inventory deltas, copy referenced
4019
# texts that delta references, insert the delta, revision and
4022
my_pb = ui.ui_factory.nested_progress_bar()
4025
symbol_versioning.warn(
4026
symbol_versioning.deprecated_in((1, 14, 0))
4027
% "pb parameter to fetch()")
4030
self._fetch_all_revisions(revision_ids, pb)
4032
if my_pb is not None:
4034
return len(revision_ids), 0
4036
def _get_basis(self, first_revision_id):
4037
"""Get a revision and tree which exists in the target.
4039
This assumes that first_revision_id is selected for transmission
4040
because all other ancestors are already present. If we can't find an
4041
ancestor we fall back to NULL_REVISION since we know that is safe.
4043
:return: (basis_id, basis_tree)
4045
first_rev = self.source.get_revision(first_revision_id)
4047
basis_id = first_rev.parent_ids[0]
4048
# only valid as a basis if the target has it
4049
self.target.get_revision(basis_id)
4050
# Try to get a basis tree - if its a ghost it will hit the
4051
# NoSuchRevision case.
4052
basis_tree = self.source.revision_tree(basis_id)
4053
except (IndexError, errors.NoSuchRevision):
4054
basis_id = _mod_revision.NULL_REVISION
4055
basis_tree = self.source.revision_tree(basis_id)
4056
return basis_id, basis_tree
4059
InterRepository.register_optimiser(InterDifferingSerializer)
4060
InterRepository.register_optimiser(InterSameDataRepository)
4061
InterRepository.register_optimiser(InterWeaveRepo)
4062
InterRepository.register_optimiser(InterKnitRepo)
4065
class CopyConverter(object):
4066
"""A repository conversion tool which just performs a copy of the content.
4068
This is slow but quite reliable.
4071
def __init__(self, target_format):
4072
"""Create a CopyConverter.
4074
:param target_format: The format the resulting repository should be.
4076
self.target_format = target_format
4078
def convert(self, repo, pb):
4079
"""Perform the conversion of to_convert, giving feedback via pb.
4081
:param to_convert: The disk object to convert.
4082
:param pb: a progress bar to use for progress information.
4087
# this is only useful with metadir layouts - separated repo content.
4088
# trigger an assertion if not such
4089
repo._format.get_format_string()
4090
self.repo_dir = repo.bzrdir
4091
self.step('Moving repository to repository.backup')
4092
self.repo_dir.transport.move('repository', 'repository.backup')
4093
backup_transport = self.repo_dir.transport.clone('repository.backup')
4094
repo._format.check_conversion_target(self.target_format)
4095
self.source_repo = repo._format.open(self.repo_dir,
4097
_override_transport=backup_transport)
4098
self.step('Creating new repository')
4099
converted = self.target_format.initialize(self.repo_dir,
4100
self.source_repo.is_shared())
4101
converted.lock_write()
4103
self.step('Copying content')
4104
self.source_repo.copy_content_into(converted)
4107
self.step('Deleting old repository content')
4108
self.repo_dir.transport.delete_tree('repository.backup')
4109
ui.ui_factory.note('repository converted')
4111
def step(self, message):
4112
"""Update the pb by a step."""
4114
self.pb.update(message, self.count, self.total)
4126
def _unescaper(match, _map=_unescape_map):
4127
code = match.group(1)
4131
if not code.startswith('#'):
4133
return unichr(int(code[1:])).encode('utf8')
4139
def _unescape_xml(data):
4140
"""Unescape predefined XML entities in a string of data."""
4142
if _unescape_re is None:
4143
_unescape_re = re.compile('\&([^;]*);')
4144
return _unescape_re.sub(_unescaper, data)
4147
class _VersionedFileChecker(object):
4149
def __init__(self, repository, text_key_references=None, ancestors=None):
4150
self.repository = repository
4151
self.text_index = self.repository._generate_text_key_index(
4152
text_key_references=text_key_references, ancestors=ancestors)
4154
def calculate_file_version_parents(self, text_key):
4155
"""Calculate the correct parents for a file version according to
4158
parent_keys = self.text_index[text_key]
4159
if parent_keys == [_mod_revision.NULL_REVISION]:
4161
return tuple(parent_keys)
4163
def check_file_version_parents(self, texts, progress_bar=None):
4164
"""Check the parents stored in a versioned file are correct.
4166
It also detects file versions that are not referenced by their
4167
corresponding revision's inventory.
4169
:returns: A tuple of (wrong_parents, dangling_file_versions).
4170
wrong_parents is a dict mapping {revision_id: (stored_parents,
4171
correct_parents)} for each revision_id where the stored parents
4172
are not correct. dangling_file_versions is a set of (file_id,
4173
revision_id) tuples for versions that are present in this versioned
4174
file, but not used by the corresponding inventory.
4176
local_progress = None
4177
if progress_bar is None:
4178
local_progress = ui.ui_factory.nested_progress_bar()
4179
progress_bar = local_progress
4181
return self._check_file_version_parents(texts, progress_bar)
4184
local_progress.finished()
4186
def _check_file_version_parents(self, texts, progress_bar):
4187
"""See check_file_version_parents."""
4189
self.file_ids = set([file_id for file_id, _ in
4190
self.text_index.iterkeys()])
4191
# text keys is now grouped by file_id
4192
n_versions = len(self.text_index)
4193
progress_bar.update('loading text store', 0, n_versions)
4194
parent_map = self.repository.texts.get_parent_map(self.text_index)
4195
# On unlistable transports this could well be empty/error...
4196
text_keys = self.repository.texts.keys()
4197
unused_keys = frozenset(text_keys) - set(self.text_index)
4198
for num, key in enumerate(self.text_index.iterkeys()):
4199
progress_bar.update('checking text graph', num, n_versions)
4200
correct_parents = self.calculate_file_version_parents(key)
4202
knit_parents = parent_map[key]
4203
except errors.RevisionNotPresent:
4206
if correct_parents != knit_parents:
4207
wrong_parents[key] = (knit_parents, correct_parents)
4208
return wrong_parents, unused_keys
4211
def _old_get_graph(repository, revision_id):
4212
"""DO NOT USE. That is all. I'm serious."""
4213
graph = repository.get_graph()
4214
revision_graph = dict(((key, value) for key, value in
4215
graph.iter_ancestry([revision_id]) if value is not None))
4216
return _strip_NULL_ghosts(revision_graph)
4219
def _strip_NULL_ghosts(revision_graph):
4220
"""Also don't use this. more compatibility code for unmigrated clients."""
4221
# Filter ghosts, and null:
4222
if _mod_revision.NULL_REVISION in revision_graph:
4223
del revision_graph[_mod_revision.NULL_REVISION]
4224
for key, parents in revision_graph.items():
4225
revision_graph[key] = tuple(parent for parent in parents if parent
4227
return revision_graph
4230
class StreamSink(object):
4231
"""An object that can insert a stream into a repository.
4233
This interface handles the complexity of reserialising inventories and
4234
revisions from different formats, and allows unidirectional insertion into
4235
stacked repositories without looking for the missing basis parents
4239
def __init__(self, target_repo):
4240
self.target_repo = target_repo
4242
def insert_stream(self, stream, src_format, resume_tokens):
4243
"""Insert a stream's content into the target repository.
4245
:param src_format: a bzr repository format.
4247
:return: a list of resume tokens and an iterable of keys additional
4248
items required before the insertion can be completed.
4250
self.target_repo.lock_write()
4253
self.target_repo.resume_write_group(resume_tokens)
4256
self.target_repo.start_write_group()
4259
# locked_insert_stream performs a commit|suspend.
4260
return self._locked_insert_stream(stream, src_format, is_resume)
4262
self.target_repo.abort_write_group(suppress_errors=True)
4265
self.target_repo.unlock()
4267
def _locked_insert_stream(self, stream, src_format, is_resume):
4268
to_serializer = self.target_repo._format._serializer
4269
src_serializer = src_format._serializer
4271
if to_serializer == src_serializer:
4272
# If serializers match and the target is a pack repository, set the
4273
# write cache size on the new pack. This avoids poor performance
4274
# on transports where append is unbuffered (such as
4275
# RemoteTransport). This is safe to do because nothing should read
4276
# back from the target repository while a stream with matching
4277
# serialization is being inserted.
4278
# The exception is that a delta record from the source that should
4279
# be a fulltext may need to be expanded by the target (see
4280
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4281
# explicitly flush any buffered writes first in that rare case.
4283
new_pack = self.target_repo._pack_collection._new_pack
4284
except AttributeError:
4285
# Not a pack repository
4288
new_pack.set_write_cache_size(1024*1024)
4289
for substream_type, substream in stream:
4290
if 'stream' in debug.debug_flags:
4291
mutter('inserting substream: %s', substream_type)
4292
if substream_type == 'texts':
4293
self.target_repo.texts.insert_record_stream(substream)
4294
elif substream_type == 'inventories':
4295
if src_serializer == to_serializer:
4296
self.target_repo.inventories.insert_record_stream(
4299
self._extract_and_insert_inventories(
4300
substream, src_serializer)
4301
elif substream_type == 'inventory-deltas':
4302
self._extract_and_insert_inventory_deltas(
4303
substream, src_serializer)
4304
elif substream_type == 'chk_bytes':
4305
# XXX: This doesn't support conversions, as it assumes the
4306
# conversion was done in the fetch code.
4307
self.target_repo.chk_bytes.insert_record_stream(substream)
4308
elif substream_type == 'revisions':
4309
# This may fallback to extract-and-insert more often than
4310
# required if the serializers are different only in terms of
4312
if src_serializer == to_serializer:
4313
self.target_repo.revisions.insert_record_stream(
4316
self._extract_and_insert_revisions(substream,
4318
elif substream_type == 'signatures':
4319
self.target_repo.signatures.insert_record_stream(substream)
4321
raise AssertionError('kaboom! %s' % (substream_type,))
4322
# Done inserting data, and the missing_keys calculations will try to
4323
# read back from the inserted data, so flush the writes to the new pack
4324
# (if this is pack format).
4325
if new_pack is not None:
4326
new_pack._write_data('', flush=True)
4327
# Find all the new revisions (including ones from resume_tokens)
4328
missing_keys = self.target_repo.get_missing_parent_inventories(
4329
check_for_missing_texts=is_resume)
4331
for prefix, versioned_file in (
4332
('texts', self.target_repo.texts),
4333
('inventories', self.target_repo.inventories),
4334
('revisions', self.target_repo.revisions),
4335
('signatures', self.target_repo.signatures),
4336
('chk_bytes', self.target_repo.chk_bytes),
4338
if versioned_file is None:
4340
# TODO: key is often going to be a StaticTuple object
4341
# I don't believe we can define a method by which
4342
# (prefix,) + StaticTuple will work, though we could
4343
# define a StaticTuple.sq_concat that would allow you to
4344
# pass in either a tuple or a StaticTuple as the second
4345
# object, so instead we could have:
4346
# StaticTuple(prefix) + key here...
4347
missing_keys.update((prefix,) + key for key in
4348
versioned_file.get_missing_compression_parent_keys())
4349
except NotImplementedError:
4350
# cannot even attempt suspending, and missing would have failed
4351
# during stream insertion.
4352
missing_keys = set()
4355
# suspend the write group and tell the caller what we is
4356
# missing. We know we can suspend or else we would not have
4357
# entered this code path. (All repositories that can handle
4358
# missing keys can handle suspending a write group).
4359
write_group_tokens = self.target_repo.suspend_write_group()
4360
return write_group_tokens, missing_keys
4361
hint = self.target_repo.commit_write_group()
4362
if (to_serializer != src_serializer and
4363
self.target_repo._format.pack_compresses):
4364
self.target_repo.pack(hint=hint)
4367
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4368
target_rich_root = self.target_repo._format.rich_root_data
4369
target_tree_refs = self.target_repo._format.supports_tree_reference
4370
for record in substream:
4371
# Insert the delta directly
4372
inventory_delta_bytes = record.get_bytes_as('fulltext')
4373
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4375
parse_result = deserialiser.parse_text_bytes(
4376
inventory_delta_bytes)
4377
except inventory_delta.IncompatibleInventoryDelta, err:
4378
trace.mutter("Incompatible delta: %s", err.msg)
4379
raise errors.IncompatibleRevision(self.target_repo._format)
4380
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4381
revision_id = new_id
4382
parents = [key[0] for key in record.parents]
4383
self.target_repo.add_inventory_by_delta(
4384
basis_id, inv_delta, revision_id, parents)
4386
def _extract_and_insert_inventories(self, substream, serializer,
4388
"""Generate a new inventory versionedfile in target, converting data.
4390
The inventory is retrieved from the source, (deserializing it), and
4391
stored in the target (reserializing it in a different format).
4393
target_rich_root = self.target_repo._format.rich_root_data
4394
target_tree_refs = self.target_repo._format.supports_tree_reference
4395
for record in substream:
4396
# It's not a delta, so it must be a fulltext in the source
4397
# serializer's format.
4398
bytes = record.get_bytes_as('fulltext')
4399
revision_id = record.key[0]
4400
inv = serializer.read_inventory_from_string(bytes, revision_id)
4401
parents = [key[0] for key in record.parents]
4402
self.target_repo.add_inventory(revision_id, inv, parents)
4403
# No need to keep holding this full inv in memory when the rest of
4404
# the substream is likely to be all deltas.
4407
def _extract_and_insert_revisions(self, substream, serializer):
4408
for record in substream:
4409
bytes = record.get_bytes_as('fulltext')
4410
revision_id = record.key[0]
4411
rev = serializer.read_revision_from_string(bytes)
4412
if rev.revision_id != revision_id:
4413
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4414
self.target_repo.add_revision(revision_id, rev)
4417
if self.target_repo._format._fetch_reconcile:
4418
self.target_repo.reconcile()
4421
class StreamSource(object):
4422
"""A source of a stream for fetching between repositories."""
4424
def __init__(self, from_repository, to_format):
4425
"""Create a StreamSource streaming from from_repository."""
4426
self.from_repository = from_repository
4427
self.to_format = to_format
4429
def delta_on_metadata(self):
4430
"""Return True if delta's are permitted on metadata streams.
4432
That is on revisions and signatures.
4434
src_serializer = self.from_repository._format._serializer
4435
target_serializer = self.to_format._serializer
4436
return (self.to_format._fetch_uses_deltas and
4437
src_serializer == target_serializer)
4439
def _fetch_revision_texts(self, revs):
4440
# fetch signatures first and then the revision texts
4441
# may need to be a InterRevisionStore call here.
4442
from_sf = self.from_repository.signatures
4443
# A missing signature is just skipped.
4444
keys = [(rev_id,) for rev_id in revs]
4445
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4447
self.to_format._fetch_order,
4448
not self.to_format._fetch_uses_deltas))
4449
# If a revision has a delta, this is actually expanded inside the
4450
# insert_record_stream code now, which is an alternate fix for
4452
from_rf = self.from_repository.revisions
4453
revisions = from_rf.get_record_stream(
4455
self.to_format._fetch_order,
4456
not self.delta_on_metadata())
4457
return [('signatures', signatures), ('revisions', revisions)]
4459
def _generate_root_texts(self, revs):
4460
"""This will be called by get_stream between fetching weave texts and
4461
fetching the inventory weave.
4463
if self._rich_root_upgrade():
4464
return _mod_fetch.Inter1and2Helper(
4465
self.from_repository).generate_root_texts(revs)
4469
def get_stream(self, search):
4471
revs = search.get_keys()
4472
graph = self.from_repository.get_graph()
4473
revs = tsort.topo_sort(graph.get_parent_map(revs))
4474
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4476
for knit_kind, file_id, revisions in data_to_fetch:
4477
if knit_kind != phase:
4479
# Make a new progress bar for this phase
4480
if knit_kind == "file":
4481
# Accumulate file texts
4482
text_keys.extend([(file_id, revision) for revision in
4484
elif knit_kind == "inventory":
4485
# Now copy the file texts.
4486
from_texts = self.from_repository.texts
4487
yield ('texts', from_texts.get_record_stream(
4488
text_keys, self.to_format._fetch_order,
4489
not self.to_format._fetch_uses_deltas))
4490
# Cause an error if a text occurs after we have done the
4493
# Before we process the inventory we generate the root
4494
# texts (if necessary) so that the inventories references
4496
for _ in self._generate_root_texts(revs):
4498
# we fetch only the referenced inventories because we do not
4499
# know for unselected inventories whether all their required
4500
# texts are present in the other repository - it could be
4502
for info in self._get_inventory_stream(revs):
4504
elif knit_kind == "signatures":
4505
# Nothing to do here; this will be taken care of when
4506
# _fetch_revision_texts happens.
4508
elif knit_kind == "revisions":
4509
for record in self._fetch_revision_texts(revs):
4512
raise AssertionError("Unknown knit kind %r" % knit_kind)
4514
def get_stream_for_missing_keys(self, missing_keys):
4515
# missing keys can only occur when we are byte copying and not
4516
# translating (because translation means we don't send
4517
# unreconstructable deltas ever).
4519
keys['texts'] = set()
4520
keys['revisions'] = set()
4521
keys['inventories'] = set()
4522
keys['chk_bytes'] = set()
4523
keys['signatures'] = set()
4524
for key in missing_keys:
4525
keys[key[0]].add(key[1:])
4526
if len(keys['revisions']):
4527
# If we allowed copying revisions at this point, we could end up
4528
# copying a revision without copying its required texts: a
4529
# violation of the requirements for repository integrity.
4530
raise AssertionError(
4531
'cannot copy revisions to fill in missing deltas %s' % (
4532
keys['revisions'],))
4533
for substream_kind, keys in keys.iteritems():
4534
vf = getattr(self.from_repository, substream_kind)
4535
if vf is None and keys:
4536
raise AssertionError(
4537
"cannot fill in keys for a versioned file we don't"
4538
" have: %s needs %s" % (substream_kind, keys))
4540
# No need to stream something we don't have
4542
if substream_kind == 'inventories':
4543
# Some missing keys are genuinely ghosts, filter those out.
4544
present = self.from_repository.inventories.get_parent_map(keys)
4545
revs = [key[0] for key in present]
4546
# Get the inventory stream more-or-less as we do for the
4547
# original stream; there's no reason to assume that records
4548
# direct from the source will be suitable for the sink. (Think
4549
# e.g. 2a -> 1.9-rich-root).
4550
for info in self._get_inventory_stream(revs, missing=True):
4554
# Ask for full texts always so that we don't need more round trips
4555
# after this stream.
4556
# Some of the missing keys are genuinely ghosts, so filter absent
4557
# records. The Sink is responsible for doing another check to
4558
# ensure that ghosts don't introduce missing data for future
4560
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4561
self.to_format._fetch_order, True))
4562
yield substream_kind, stream
4564
def inventory_fetch_order(self):
4565
if self._rich_root_upgrade():
4566
return 'topological'
4568
return self.to_format._fetch_order
4570
def _rich_root_upgrade(self):
4571
return (not self.from_repository._format.rich_root_data and
4572
self.to_format.rich_root_data)
4574
def _get_inventory_stream(self, revision_ids, missing=False):
4575
from_format = self.from_repository._format
4576
if (from_format.supports_chks and self.to_format.supports_chks and
4577
from_format.network_name() == self.to_format.network_name()):
4578
raise AssertionError(
4579
"this case should be handled by GroupCHKStreamSource")
4580
elif 'forceinvdeltas' in debug.debug_flags:
4581
return self._get_convertable_inventory_stream(revision_ids,
4582
delta_versus_null=missing)
4583
elif from_format.network_name() == self.to_format.network_name():
4585
return self._get_simple_inventory_stream(revision_ids,
4587
elif (not from_format.supports_chks and not self.to_format.supports_chks
4588
and from_format._serializer == self.to_format._serializer):
4589
# Essentially the same format.
4590
return self._get_simple_inventory_stream(revision_ids,
4593
# Any time we switch serializations, we want to use an
4594
# inventory-delta based approach.
4595
return self._get_convertable_inventory_stream(revision_ids,
4596
delta_versus_null=missing)
4598
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4599
# NB: This currently reopens the inventory weave in source;
4600
# using a single stream interface instead would avoid this.
4601
from_weave = self.from_repository.inventories
4603
delta_closure = True
4605
delta_closure = not self.delta_on_metadata()
4606
yield ('inventories', from_weave.get_record_stream(
4607
[(rev_id,) for rev_id in revision_ids],
4608
self.inventory_fetch_order(), delta_closure))
4610
def _get_convertable_inventory_stream(self, revision_ids,
4611
delta_versus_null=False):
4612
# The source is using CHKs, but the target either doesn't or it has a
4613
# different serializer. The StreamSink code expects to be able to
4614
# convert on the target, so we need to put bytes-on-the-wire that can
4615
# be converted. That means inventory deltas (if the remote is <1.19,
4616
# RemoteStreamSink will fallback to VFS to insert the deltas).
4617
yield ('inventory-deltas',
4618
self._stream_invs_as_deltas(revision_ids,
4619
delta_versus_null=delta_versus_null))
4621
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4622
"""Return a stream of inventory-deltas for the given rev ids.
4624
:param revision_ids: The list of inventories to transmit
4625
:param delta_versus_null: Don't try to find a minimal delta for this
4626
entry, instead compute the delta versus the NULL_REVISION. This
4627
effectively streams a complete inventory. Used for stuff like
4628
filling in missing parents, etc.
4630
from_repo = self.from_repository
4631
revision_keys = [(rev_id,) for rev_id in revision_ids]
4632
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4633
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4635
inventories = self.from_repository.iter_inventories(
4636
revision_ids, 'topological')
4637
format = from_repo._format
4638
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4639
inventory_cache = lru_cache.LRUCache(50)
4640
null_inventory = from_repo.revision_tree(
4641
_mod_revision.NULL_REVISION).inventory
4642
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4643
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4644
# repo back into a non-rich-root repo ought to be allowed)
4645
serializer = inventory_delta.InventoryDeltaSerializer(
4646
versioned_root=format.rich_root_data,
4647
tree_references=format.supports_tree_reference)
4648
for inv in inventories:
4649
key = (inv.revision_id,)
4650
parent_keys = parent_map.get(key, ())
4652
if not delta_versus_null and parent_keys:
4653
# The caller did not ask for complete inventories and we have
4654
# some parents that we can delta against. Make a delta against
4655
# each parent so that we can find the smallest.
4656
parent_ids = [parent_key[0] for parent_key in parent_keys]
4657
for parent_id in parent_ids:
4658
if parent_id not in invs_sent_so_far:
4659
# We don't know that the remote side has this basis, so
4662
if parent_id == _mod_revision.NULL_REVISION:
4663
parent_inv = null_inventory
4665
parent_inv = inventory_cache.get(parent_id, None)
4666
if parent_inv is None:
4667
parent_inv = from_repo.get_inventory(parent_id)
4668
candidate_delta = inv._make_delta(parent_inv)
4669
if (delta is None or
4670
len(delta) > len(candidate_delta)):
4671
delta = candidate_delta
4672
basis_id = parent_id
4674
# Either none of the parents ended up being suitable, or we
4675
# were asked to delta against NULL
4676
basis_id = _mod_revision.NULL_REVISION
4677
delta = inv._make_delta(null_inventory)
4678
invs_sent_so_far.add(inv.revision_id)
4679
inventory_cache[inv.revision_id] = inv
4680
delta_serialized = ''.join(
4681
serializer.delta_to_lines(basis_id, key[-1], delta))
4682
yield versionedfile.FulltextContentFactory(
4683
key, parent_keys, None, delta_serialized)
4686
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4687
stop_revision=None):
4688
"""Extend the partial history to include a given index
4690
If a stop_index is supplied, stop when that index has been reached.
4691
If a stop_revision is supplied, stop when that revision is
4692
encountered. Otherwise, stop when the beginning of history is
4695
:param stop_index: The index which should be present. When it is
4696
present, history extension will stop.
4697
:param stop_revision: The revision id which should be present. When
4698
it is encountered, history extension will stop.
4700
start_revision = partial_history_cache[-1]
4701
iterator = repo.iter_reverse_revision_history(start_revision)
4703
#skip the last revision in the list
4706
if (stop_index is not None and
4707
len(partial_history_cache) > stop_index):
4709
if partial_history_cache[-1] == stop_revision:
4711
revision_id = iterator.next()
4712
partial_history_cache.append(revision_id)
4713
except StopIteration:
added
removed
bzrlib/repository.py
