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(), """
39
revision as _mod_revision,
45
from bzrlib.bundle import serializer
46
from bzrlib.revisiontree import RevisionTree
47
from bzrlib.store.versioned import VersionedFileStore
48
from bzrlib.testament import Testament
51
from bzrlib.decorators import needs_read_lock, needs_write_lock
52
from bzrlib.inter import InterObject
53
from bzrlib.inventory import (
59
from bzrlib import registry
60
from bzrlib.symbol_versioning import (
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
self.inv_sha1 = self.repository.add_inventory_by_delta(
211
basis_id, self._basis_delta, self._new_revision_id,
214
if self.new_inventory.root is None:
215
raise AssertionError('Root entry should be supplied to'
216
' record_entry_contents, as of bzr 0.10.')
217
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
218
self.new_inventory.revision_id = self._new_revision_id
219
self.inv_sha1 = self.repository.add_inventory(
220
self._new_revision_id,
224
return self._new_revision_id
226
def _gen_revision_id(self):
227
"""Return new revision-id."""
228
return generate_ids.gen_revision_id(self._config.username(),
231
def _generate_revision_if_needed(self):
232
"""Create a revision id if None was supplied.
234
If the repository can not support user-specified revision ids
235
they should override this function and raise CannotSetRevisionId
236
if _new_revision_id is not None.
238
:raises: CannotSetRevisionId
240
if self._new_revision_id is None:
241
self._new_revision_id = self._gen_revision_id()
242
self.random_revid = True
244
self.random_revid = False
246
def _heads(self, file_id, revision_ids):
247
"""Calculate the graph heads for revision_ids in the graph of file_id.
249
This can use either a per-file graph or a global revision graph as we
250
have an identity relationship between the two graphs.
252
return self.__heads(revision_ids)
254
def _check_root(self, ie, parent_invs, tree):
255
"""Helper for record_entry_contents.
257
:param ie: An entry being added.
258
:param parent_invs: The inventories of the parent revisions of the
260
:param tree: The tree that is being committed.
262
# In this revision format, root entries have no knit or weave When
263
# serializing out to disk and back in root.revision is always
265
ie.revision = self._new_revision_id
267
def _require_root_change(self, tree):
268
"""Enforce an appropriate root object change.
270
This is called once when record_iter_changes is called, if and only if
271
the root was not in the delta calculated by record_iter_changes.
273
:param tree: The tree which is being committed.
275
# NB: if there are no parents then this method is not called, so no
276
# need to guard on parents having length.
277
entry = entry_factory['directory'](tree.path2id(''), '',
279
entry.revision = self._new_revision_id
280
self._basis_delta.append(('', '', entry.file_id, entry))
282
def _get_delta(self, ie, basis_inv, path):
283
"""Get a delta against the basis inventory for ie."""
284
if ie.file_id not in basis_inv:
286
result = (None, path, ie.file_id, ie)
287
self._basis_delta.append(result)
289
elif ie != basis_inv[ie.file_id]:
291
# TODO: avoid tis id2path call.
292
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
293
self._basis_delta.append(result)
299
def get_basis_delta(self):
300
"""Return the complete inventory delta versus the basis inventory.
302
This has been built up with the calls to record_delete and
303
record_entry_contents. The client must have already called
304
will_record_deletes() to indicate that they will be generating a
307
:return: An inventory delta, suitable for use with apply_delta, or
308
Repository.add_inventory_by_delta, etc.
310
if not self._recording_deletes:
311
raise AssertionError("recording deletes not activated.")
312
return self._basis_delta
314
def record_delete(self, path, file_id):
315
"""Record that a delete occured against a basis tree.
317
This is an optional API - when used it adds items to the basis_delta
318
being accumulated by the commit builder. It cannot be called unless the
319
method will_record_deletes() has been called to inform the builder that
320
a delta is being supplied.
322
:param path: The path of the thing deleted.
323
:param file_id: The file id that was deleted.
325
if not self._recording_deletes:
326
raise AssertionError("recording deletes not activated.")
327
delta = (path, None, file_id, None)
328
self._basis_delta.append(delta)
329
self._any_changes = True
332
def will_record_deletes(self):
333
"""Tell the commit builder that deletes are being notified.
335
This enables the accumulation of an inventory delta; for the resulting
336
commit to be valid, deletes against the basis MUST be recorded via
337
builder.record_delete().
339
self._recording_deletes = True
341
basis_id = self.parents[0]
343
basis_id = _mod_revision.NULL_REVISION
344
self.basis_delta_revision = basis_id
346
def record_entry_contents(self, ie, parent_invs, path, tree,
348
"""Record the content of ie from tree into the commit if needed.
350
Side effect: sets ie.revision when unchanged
352
:param ie: An inventory entry present in the commit.
353
:param parent_invs: The inventories of the parent revisions of the
355
:param path: The path the entry is at in the tree.
356
:param tree: The tree which contains this entry and should be used to
358
:param content_summary: Summary data from the tree about the paths
359
content - stat, length, exec, sha/link target. This is only
360
accessed when the entry has a revision of None - that is when it is
361
a candidate to commit.
362
:return: A tuple (change_delta, version_recorded, fs_hash).
363
change_delta is an inventory_delta change for this entry against
364
the basis tree of the commit, or None if no change occured against
366
version_recorded is True if a new version of the entry has been
367
recorded. For instance, committing a merge where a file was only
368
changed on the other side will return (delta, False).
369
fs_hash is either None, or the hash details for the path (currently
370
a tuple of the contents sha1 and the statvalue returned by
371
tree.get_file_with_stat()).
373
if self.new_inventory.root is None:
374
if ie.parent_id is not None:
375
raise errors.RootMissing()
376
self._check_root(ie, parent_invs, tree)
377
if ie.revision is None:
378
kind = content_summary[0]
380
# ie is carried over from a prior commit
382
# XXX: repository specific check for nested tree support goes here - if
383
# the repo doesn't want nested trees we skip it ?
384
if (kind == 'tree-reference' and
385
not self.repository._format.supports_tree_reference):
386
# mismatch between commit builder logic and repository:
387
# this needs the entry creation pushed down into the builder.
388
raise NotImplementedError('Missing repository subtree support.')
389
self.new_inventory.add(ie)
391
# TODO: slow, take it out of the inner loop.
393
basis_inv = parent_invs[0]
395
basis_inv = Inventory(root_id=None)
397
# ie.revision is always None if the InventoryEntry is considered
398
# for committing. We may record the previous parents revision if the
399
# content is actually unchanged against a sole head.
400
if ie.revision is not None:
401
if not self._versioned_root and path == '':
402
# repositories that do not version the root set the root's
403
# revision to the new commit even when no change occurs (more
404
# specifically, they do not record a revision on the root; and
405
# the rev id is assigned to the root during deserialisation -
406
# this masks when a change may have occurred against the basis.
407
# To match this we always issue a delta, because the revision
408
# of the root will always be changing.
409
if ie.file_id in basis_inv:
410
delta = (basis_inv.id2path(ie.file_id), path,
414
delta = (None, path, ie.file_id, ie)
415
self._basis_delta.append(delta)
416
return delta, False, None
418
# we don't need to commit this, because the caller already
419
# determined that an existing revision of this file is
420
# appropriate. If its not being considered for committing then
421
# it and all its parents to the root must be unaltered so
422
# no-change against the basis.
423
if ie.revision == self._new_revision_id:
424
raise AssertionError("Impossible situation, a skipped "
425
"inventory entry (%r) claims to be modified in this "
426
"commit (%r).", (ie, self._new_revision_id))
427
return None, False, None
428
# XXX: Friction: parent_candidates should return a list not a dict
429
# so that we don't have to walk the inventories again.
430
parent_candiate_entries = ie.parent_candidates(parent_invs)
431
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
433
for inv in parent_invs:
434
if ie.file_id in inv:
435
old_rev = inv[ie.file_id].revision
436
if old_rev in head_set:
437
heads.append(inv[ie.file_id].revision)
438
head_set.remove(inv[ie.file_id].revision)
441
# now we check to see if we need to write a new record to the
443
# We write a new entry unless there is one head to the ancestors, and
444
# the kind-derived content is unchanged.
446
# Cheapest check first: no ancestors, or more the one head in the
447
# ancestors, we write a new node.
451
# There is a single head, look it up for comparison
452
parent_entry = parent_candiate_entries[heads[0]]
453
# if the non-content specific data has changed, we'll be writing a
455
if (parent_entry.parent_id != ie.parent_id or
456
parent_entry.name != ie.name):
458
# now we need to do content specific checks:
460
# if the kind changed the content obviously has
461
if kind != parent_entry.kind:
463
# Stat cache fingerprint feedback for the caller - None as we usually
464
# don't generate one.
467
if content_summary[2] is None:
468
raise ValueError("Files must not have executable = None")
470
if (# if the file length changed we have to store:
471
parent_entry.text_size != content_summary[1] or
472
# if the exec bit has changed we have to store:
473
parent_entry.executable != content_summary[2]):
475
elif parent_entry.text_sha1 == content_summary[3]:
476
# all meta and content is unchanged (using a hash cache
477
# hit to check the sha)
478
ie.revision = parent_entry.revision
479
ie.text_size = parent_entry.text_size
480
ie.text_sha1 = parent_entry.text_sha1
481
ie.executable = parent_entry.executable
482
return self._get_delta(ie, basis_inv, path), False, None
484
# Either there is only a hash change(no hash cache entry,
485
# or same size content change), or there is no change on
487
# Provide the parent's hash to the store layer, so that the
488
# content is unchanged we will not store a new node.
489
nostore_sha = parent_entry.text_sha1
491
# We want to record a new node regardless of the presence or
492
# absence of a content change in the file.
494
ie.executable = content_summary[2]
495
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
497
lines = file_obj.readlines()
501
ie.text_sha1, ie.text_size = self._add_text_to_weave(
502
ie.file_id, lines, heads, nostore_sha)
503
# Let the caller know we generated a stat fingerprint.
504
fingerprint = (ie.text_sha1, stat_value)
505
except errors.ExistingContent:
506
# Turns out that the file content was unchanged, and we were
507
# only going to store a new node if it was changed. Carry over
509
ie.revision = parent_entry.revision
510
ie.text_size = parent_entry.text_size
511
ie.text_sha1 = parent_entry.text_sha1
512
ie.executable = parent_entry.executable
513
return self._get_delta(ie, basis_inv, path), False, None
514
elif kind == 'directory':
516
# all data is meta here, nothing specific to directory, so
518
ie.revision = parent_entry.revision
519
return self._get_delta(ie, basis_inv, path), False, None
521
self._add_text_to_weave(ie.file_id, lines, heads, None)
522
elif kind == 'symlink':
523
current_link_target = content_summary[3]
525
# symlink target is not generic metadata, check if it has
527
if current_link_target != parent_entry.symlink_target:
530
# unchanged, carry over.
531
ie.revision = parent_entry.revision
532
ie.symlink_target = parent_entry.symlink_target
533
return self._get_delta(ie, basis_inv, path), False, None
534
ie.symlink_target = current_link_target
536
self._add_text_to_weave(ie.file_id, lines, heads, None)
537
elif kind == 'tree-reference':
539
if content_summary[3] != parent_entry.reference_revision:
542
# unchanged, carry over.
543
ie.reference_revision = parent_entry.reference_revision
544
ie.revision = parent_entry.revision
545
return self._get_delta(ie, basis_inv, path), False, None
546
ie.reference_revision = content_summary[3]
548
self._add_text_to_weave(ie.file_id, lines, heads, None)
550
raise NotImplementedError('unknown kind')
551
ie.revision = self._new_revision_id
552
self._any_changes = True
553
return self._get_delta(ie, basis_inv, path), True, fingerprint
555
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
556
_entry_factory=entry_factory):
557
"""Record a new tree via iter_changes.
559
:param tree: The tree to obtain text contents from for changed objects.
560
:param basis_revision_id: The revision id of the tree the iter_changes
561
has been generated against. Currently assumed to be the same
562
as self.parents[0] - if it is not, errors may occur.
563
:param iter_changes: An iter_changes iterator with the changes to apply
564
to basis_revision_id. The iterator must not include any items with
565
a current kind of None - missing items must be either filtered out
566
or errored-on beefore record_iter_changes sees the item.
567
:param _entry_factory: Private method to bind entry_factory locally for
569
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
572
# Create an inventory delta based on deltas between all the parents and
573
# deltas between all the parent inventories. We use inventory delta's
574
# between the inventory objects because iter_changes masks
575
# last-changed-field only changes.
577
# file_id -> change map, change is fileid, paths, changed, versioneds,
578
# parents, names, kinds, executables
580
# {file_id -> revision_id -> inventory entry, for entries in parent
581
# trees that are not parents[0]
585
revtrees = list(self.repository.revision_trees(self.parents))
586
except errors.NoSuchRevision:
587
# one or more ghosts, slow path.
589
for revision_id in self.parents:
591
revtrees.append(self.repository.revision_tree(revision_id))
592
except errors.NoSuchRevision:
594
basis_revision_id = _mod_revision.NULL_REVISION
596
revtrees.append(self.repository.revision_tree(
597
_mod_revision.NULL_REVISION))
598
# The basis inventory from a repository
600
basis_inv = revtrees[0].inventory
602
basis_inv = self.repository.revision_tree(
603
_mod_revision.NULL_REVISION).inventory
604
if len(self.parents) > 0:
605
if basis_revision_id != self.parents[0] and not ghost_basis:
607
"arbitrary basis parents not yet supported with merges")
608
for revtree in revtrees[1:]:
609
for change in revtree.inventory._make_delta(basis_inv):
610
if change[1] is None:
611
# Not present in this parent.
613
if change[2] not in merged_ids:
614
if change[0] is not None:
615
basis_entry = basis_inv[change[2]]
616
merged_ids[change[2]] = [
618
basis_entry.revision,
621
parent_entries[change[2]] = {
623
basis_entry.revision:basis_entry,
625
change[3].revision:change[3],
628
merged_ids[change[2]] = [change[3].revision]
629
parent_entries[change[2]] = {change[3].revision:change[3]}
631
merged_ids[change[2]].append(change[3].revision)
632
parent_entries[change[2]][change[3].revision] = change[3]
635
# Setup the changes from the tree:
636
# changes maps file_id -> (change, [parent revision_ids])
638
for change in iter_changes:
639
# This probably looks up in basis_inv way to much.
640
if change[1][0] is not None:
641
head_candidate = [basis_inv[change[0]].revision]
644
changes[change[0]] = change, merged_ids.get(change[0],
646
unchanged_merged = set(merged_ids) - set(changes)
647
# Extend the changes dict with synthetic changes to record merges of
649
for file_id in unchanged_merged:
650
# Record a merged version of these items that did not change vs the
651
# basis. This can be either identical parallel changes, or a revert
652
# of a specific file after a merge. The recorded content will be
653
# that of the current tree (which is the same as the basis), but
654
# the per-file graph will reflect a merge.
655
# NB:XXX: We are reconstructing path information we had, this
656
# should be preserved instead.
657
# inv delta change: (file_id, (path_in_source, path_in_target),
658
# changed_content, versioned, parent, name, kind,
661
basis_entry = basis_inv[file_id]
662
except errors.NoSuchId:
663
# a change from basis->some_parents but file_id isn't in basis
664
# so was new in the merge, which means it must have changed
665
# from basis -> current, and as it hasn't the add was reverted
666
# by the user. So we discard this change.
670
(basis_inv.id2path(file_id), tree.id2path(file_id)),
672
(basis_entry.parent_id, basis_entry.parent_id),
673
(basis_entry.name, basis_entry.name),
674
(basis_entry.kind, basis_entry.kind),
675
(basis_entry.executable, basis_entry.executable))
676
changes[file_id] = (change, merged_ids[file_id])
677
# changes contains tuples with the change and a set of inventory
678
# candidates for the file.
680
# old_path, new_path, file_id, new_inventory_entry
681
seen_root = False # Is the root in the basis delta?
682
inv_delta = self._basis_delta
683
modified_rev = self._new_revision_id
684
for change, head_candidates in changes.values():
685
if change[3][1]: # versioned in target.
686
# Several things may be happening here:
687
# We may have a fork in the per-file graph
688
# - record a change with the content from tree
689
# We may have a change against < all trees
690
# - carry over the tree that hasn't changed
691
# We may have a change against all trees
692
# - record the change with the content from tree
695
entry = _entry_factory[kind](file_id, change[5][1],
697
head_set = self._heads(change[0], set(head_candidates))
700
for head_candidate in head_candidates:
701
if head_candidate in head_set:
702
heads.append(head_candidate)
703
head_set.remove(head_candidate)
706
# Could be a carry-over situation:
707
parent_entry_revs = parent_entries.get(file_id, None)
708
if parent_entry_revs:
709
parent_entry = parent_entry_revs.get(heads[0], None)
712
if parent_entry is None:
713
# The parent iter_changes was called against is the one
714
# that is the per-file head, so any change is relevant
715
# iter_changes is valid.
716
carry_over_possible = False
718
# could be a carry over situation
719
# A change against the basis may just indicate a merge,
720
# we need to check the content against the source of the
721
# merge to determine if it was changed after the merge
723
if (parent_entry.kind != entry.kind or
724
parent_entry.parent_id != entry.parent_id or
725
parent_entry.name != entry.name):
726
# Metadata common to all entries has changed
727
# against per-file parent
728
carry_over_possible = False
730
carry_over_possible = True
731
# per-type checks for changes against the parent_entry
734
# Cannot be a carry-over situation
735
carry_over_possible = False
736
# Populate the entry in the delta
738
# XXX: There is still a small race here: If someone reverts the content of a file
739
# after iter_changes examines and decides it has changed,
740
# we will unconditionally record a new version even if some
741
# other process reverts it while commit is running (with
742
# the revert happening after iter_changes did it's
745
entry.executable = True
747
entry.executable = False
748
if (carry_over_possible and
749
parent_entry.executable == entry.executable):
750
# Check the file length, content hash after reading
752
nostore_sha = parent_entry.text_sha1
755
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
757
lines = file_obj.readlines()
761
entry.text_sha1, entry.text_size = self._add_text_to_weave(
762
file_id, lines, heads, nostore_sha)
763
yield file_id, change[1][1], (entry.text_sha1, stat_value)
764
except errors.ExistingContent:
765
# No content change against a carry_over parent
766
# Perhaps this should also yield a fs hash update?
768
entry.text_size = parent_entry.text_size
769
entry.text_sha1 = parent_entry.text_sha1
770
elif kind == 'symlink':
772
entry.symlink_target = tree.get_symlink_target(file_id)
773
if (carry_over_possible and
774
parent_entry.symlink_target == entry.symlink_target):
777
self._add_text_to_weave(change[0], [], heads, None)
778
elif kind == 'directory':
779
if carry_over_possible:
782
# Nothing to set on the entry.
783
# XXX: split into the Root and nonRoot versions.
784
if change[1][1] != '' or self.repository.supports_rich_root():
785
self._add_text_to_weave(change[0], [], heads, None)
786
elif kind == 'tree-reference':
787
if not self.repository._format.supports_tree_reference:
788
# This isn't quite sane as an error, but we shouldn't
789
# ever see this code path in practice: tree's don't
790
# permit references when the repo doesn't support tree
792
raise errors.UnsupportedOperation(tree.add_reference,
794
entry.reference_revision = \
795
tree.get_reference_revision(change[0])
796
if (carry_over_possible and
797
parent_entry.reference_revision == reference_revision):
800
self._add_text_to_weave(change[0], [], heads, None)
802
raise AssertionError('unknown kind %r' % kind)
804
entry.revision = modified_rev
806
entry.revision = parent_entry.revision
809
new_path = change[1][1]
810
inv_delta.append((change[1][0], new_path, change[0], entry))
813
self.new_inventory = None
815
self._any_changes = True
817
# housekeeping root entry changes do not affect no-change commits.
818
self._require_root_change(tree)
819
self.basis_delta_revision = basis_revision_id
821
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
822
# Note: as we read the content directly from the tree, we know its not
823
# been turned into unicode or badly split - but a broken tree
824
# implementation could give us bad output from readlines() so this is
825
# not a guarantee of safety. What would be better is always checking
826
# the content during test suite execution. RBC 20070912
827
parent_keys = tuple((file_id, parent) for parent in parents)
828
return self.repository.texts.add_lines(
829
(file_id, self._new_revision_id), parent_keys, new_lines,
830
nostore_sha=nostore_sha, random_id=self.random_revid,
831
check_content=False)[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
######################################################################
863
class Repository(object):
864
"""Repository holding history for one or more branches.
866
The repository holds and retrieves historical information including
867
revisions and file history. It's normally accessed only by the Branch,
868
which views a particular line of development through that history.
870
The Repository builds on top of some byte storage facilies (the revisions,
871
signatures, inventories, texts and chk_bytes attributes) and a Transport,
872
which respectively provide byte storage and a means to access the (possibly
875
The byte storage facilities are addressed via tuples, which we refer to
876
as 'keys' throughout the code base. Revision_keys, inventory_keys and
877
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
878
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
879
byte string made up of a hash identifier and a hash value.
880
We use this interface because it allows low friction with the underlying
881
code that implements disk indices, network encoding and other parts of
884
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
885
the serialised revisions for the repository. This can be used to obtain
886
revision graph information or to access raw serialised revisions.
887
The result of trying to insert data into the repository via this store
888
is undefined: it should be considered read-only except for implementors
890
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
891
the serialised signatures for the repository. This can be used to
892
obtain access to raw serialised signatures. The result of trying to
893
insert data into the repository via this store is undefined: it should
894
be considered read-only except for implementors of repositories.
895
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
896
the serialised inventories for the repository. This can be used to
897
obtain unserialised inventories. The result of trying to insert data
898
into the repository via this store is undefined: it should be
899
considered read-only except for implementors of repositories.
900
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
901
texts of files and directories for the repository. This can be used to
902
obtain file texts or file graphs. Note that Repository.iter_file_bytes
903
is usually a better interface for accessing file texts.
904
The result of trying to insert data into the repository via this store
905
is undefined: it should be considered read-only except for implementors
907
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
908
any data the repository chooses to store or have indexed by its hash.
909
The result of trying to insert data into the repository via this store
910
is undefined: it should be considered read-only except for implementors
912
:ivar _transport: Transport for file access to repository, typically
913
pointing to .bzr/repository.
916
# What class to use for a CommitBuilder. Often its simpler to change this
917
# in a Repository class subclass rather than to override
918
# get_commit_builder.
919
_commit_builder_class = CommitBuilder
920
# The search regex used by xml based repositories to determine what things
921
# where changed in a single commit.
922
_file_ids_altered_regex = lazy_regex.lazy_compile(
923
r'file_id="(?P<file_id>[^"]+)"'
924
r'.* revision="(?P<revision_id>[^"]+)"'
927
def abort_write_group(self, suppress_errors=False):
928
"""Commit the contents accrued within the current write group.
930
:param suppress_errors: if true, abort_write_group will catch and log
931
unexpected errors that happen during the abort, rather than
932
allowing them to propagate. Defaults to False.
934
:seealso: start_write_group.
936
if self._write_group is not self.get_transaction():
937
# has an unlock or relock occured ?
938
raise errors.BzrError(
939
'mismatched lock context and write group. %r, %r' %
940
(self._write_group, self.get_transaction()))
942
self._abort_write_group()
943
except Exception, exc:
944
self._write_group = None
945
if not suppress_errors:
947
mutter('abort_write_group failed')
948
log_exception_quietly()
949
note('bzr: ERROR (ignored): %s', exc)
950
self._write_group = None
952
def _abort_write_group(self):
953
"""Template method for per-repository write group cleanup.
955
This is called during abort before the write group is considered to be
956
finished and should cleanup any internal state accrued during the write
957
group. There is no requirement that data handed to the repository be
958
*not* made available - this is not a rollback - but neither should any
959
attempt be made to ensure that data added is fully commited. Abort is
960
invoked when an error has occured so futher disk or network operations
961
may not be possible or may error and if possible should not be
965
def add_fallback_repository(self, repository):
966
"""Add a repository to use for looking up data not held locally.
968
:param repository: A repository.
970
if not self._format.supports_external_lookups:
971
raise errors.UnstackableRepositoryFormat(self._format, self.base)
972
self._check_fallback_repository(repository)
973
self._fallback_repositories.append(repository)
974
self.texts.add_fallback_versioned_files(repository.texts)
975
self.inventories.add_fallback_versioned_files(repository.inventories)
976
self.revisions.add_fallback_versioned_files(repository.revisions)
977
self.signatures.add_fallback_versioned_files(repository.signatures)
978
if self.chk_bytes is not None:
979
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
981
def _check_fallback_repository(self, repository):
982
"""Check that this repository can fallback to repository safely.
984
Raise an error if not.
986
:param repository: A repository to fallback to.
988
return InterRepository._assert_same_model(self, repository)
990
def add_inventory(self, revision_id, inv, parents):
991
"""Add the inventory inv to the repository as revision_id.
993
:param parents: The revision ids of the parents that revision_id
994
is known to have and are in the repository already.
996
:returns: The validator(which is a sha1 digest, though what is sha'd is
997
repository format specific) of the serialized inventory.
999
if not self.is_in_write_group():
1000
raise AssertionError("%r not in write group" % (self,))
1001
_mod_revision.check_not_reserved_id(revision_id)
1002
if not (inv.revision_id is None or inv.revision_id == revision_id):
1003
raise AssertionError(
1004
"Mismatch between inventory revision"
1005
" id and insertion revid (%r, %r)"
1006
% (inv.revision_id, revision_id))
1007
if inv.root is None:
1008
raise AssertionError()
1009
return self._add_inventory_checked(revision_id, inv, parents)
1011
def _add_inventory_checked(self, revision_id, inv, parents):
1012
"""Add inv to the repository after checking the inputs.
1014
This function can be overridden to allow different inventory styles.
1016
:seealso: add_inventory, for the contract.
1018
inv_lines = self._serialise_inventory_to_lines(inv)
1019
return self._inventory_add_lines(revision_id, parents,
1020
inv_lines, check_content=False)
1022
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1023
parents, basis_inv=None, propagate_caches=False):
1024
"""Add a new inventory expressed as a delta against another revision.
1026
:param basis_revision_id: The inventory id the delta was created
1027
against. (This does not have to be a direct parent.)
1028
:param delta: The inventory delta (see Inventory.apply_delta for
1030
:param new_revision_id: The revision id that the inventory is being
1032
:param parents: The revision ids of the parents that revision_id is
1033
known to have and are in the repository already. These are supplied
1034
for repositories that depend on the inventory graph for revision
1035
graph access, as well as for those that pun ancestry with delta
1037
:param basis_inv: The basis inventory if it is already known,
1039
:param propagate_caches: If True, the caches for this inventory are
1040
copied to and updated for the result if possible.
1042
:returns: (validator, new_inv)
1043
The validator(which is a sha1 digest, though what is sha'd is
1044
repository format specific) of the serialized inventory, and the
1045
resulting inventory.
1047
if not self.is_in_write_group():
1048
raise AssertionError("%r not in write group" % (self,))
1049
_mod_revision.check_not_reserved_id(new_revision_id)
1050
basis_tree = self.revision_tree(basis_revision_id)
1051
basis_tree.lock_read()
1053
# Note that this mutates the inventory of basis_tree, which not all
1054
# inventory implementations may support: A better idiom would be to
1055
# return a new inventory, but as there is no revision tree cache in
1056
# repository this is safe for now - RBC 20081013
1057
if basis_inv is None:
1058
basis_inv = basis_tree.inventory
1059
basis_inv.apply_delta(delta)
1060
basis_inv.revision_id = new_revision_id
1061
return (self.add_inventory(new_revision_id, basis_inv, parents),
1066
def _inventory_add_lines(self, revision_id, parents, lines,
1067
check_content=True):
1068
"""Store lines in inv_vf and return the sha1 of the inventory."""
1069
parents = [(parent,) for parent in parents]
1070
return self.inventories.add_lines((revision_id,), parents, lines,
1071
check_content=check_content)[0]
1073
def add_revision(self, revision_id, rev, inv=None, config=None):
1074
"""Add rev to the revision store as revision_id.
1076
:param revision_id: the revision id to use.
1077
:param rev: The revision object.
1078
:param inv: The inventory for the revision. if None, it will be looked
1079
up in the inventory storer
1080
:param config: If None no digital signature will be created.
1081
If supplied its signature_needed method will be used
1082
to determine if a signature should be made.
1084
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1086
_mod_revision.check_not_reserved_id(revision_id)
1087
if config is not None and config.signature_needed():
1089
inv = self.get_inventory(revision_id)
1090
plaintext = Testament(rev, inv).as_short_text()
1091
self.store_revision_signature(
1092
gpg.GPGStrategy(config), plaintext, revision_id)
1093
# check inventory present
1094
if not self.inventories.get_parent_map([(revision_id,)]):
1096
raise errors.WeaveRevisionNotPresent(revision_id,
1099
# yes, this is not suitable for adding with ghosts.
1100
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1103
key = (revision_id,)
1104
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1105
self._add_revision(rev)
1107
def _add_revision(self, revision):
1108
text = self._serializer.write_revision_to_string(revision)
1109
key = (revision.revision_id,)
1110
parents = tuple((parent,) for parent in revision.parent_ids)
1111
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1113
def all_revision_ids(self):
1114
"""Returns a list of all the revision ids in the repository.
1116
This is conceptually deprecated because code should generally work on
1117
the graph reachable from a particular revision, and ignore any other
1118
revisions that might be present. There is no direct replacement
1121
if 'evil' in debug.debug_flags:
1122
mutter_callsite(2, "all_revision_ids is linear with history.")
1123
return self._all_revision_ids()
1125
def _all_revision_ids(self):
1126
"""Returns a list of all the revision ids in the repository.
1128
These are in as much topological order as the underlying store can
1131
raise NotImplementedError(self._all_revision_ids)
1133
def break_lock(self):
1134
"""Break a lock if one is present from another instance.
1136
Uses the ui factory to ask for confirmation if the lock may be from
1139
self.control_files.break_lock()
1142
def _eliminate_revisions_not_present(self, revision_ids):
1143
"""Check every revision id in revision_ids to see if we have it.
1145
Returns a set of the present revisions.
1148
graph = self.get_graph()
1149
parent_map = graph.get_parent_map(revision_ids)
1150
# The old API returned a list, should this actually be a set?
1151
return parent_map.keys()
1154
def create(a_bzrdir):
1155
"""Construct the current default format repository in a_bzrdir."""
1156
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1158
def __init__(self, _format, a_bzrdir, control_files):
1159
"""instantiate a Repository.
1161
:param _format: The format of the repository on disk.
1162
:param a_bzrdir: The BzrDir of the repository.
1164
In the future we will have a single api for all stores for
1165
getting file texts, inventories and revisions, then
1166
this construct will accept instances of those things.
1168
super(Repository, self).__init__()
1169
self._format = _format
1170
# the following are part of the public API for Repository:
1171
self.bzrdir = a_bzrdir
1172
self.control_files = control_files
1173
self._transport = control_files._transport
1174
self.base = self._transport.base
1176
self._reconcile_does_inventory_gc = True
1177
self._reconcile_fixes_text_parents = False
1178
self._reconcile_backsup_inventory = True
1179
# not right yet - should be more semantically clear ?
1181
# TODO: make sure to construct the right store classes, etc, depending
1182
# on whether escaping is required.
1183
self._warn_if_deprecated()
1184
self._write_group = None
1185
# Additional places to query for data.
1186
self._fallback_repositories = []
1187
# An InventoryEntry cache, used during deserialization
1188
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1191
return '%s(%r)' % (self.__class__.__name__,
1194
def has_same_location(self, other):
1195
"""Returns a boolean indicating if this repository is at the same
1196
location as another repository.
1198
This might return False even when two repository objects are accessing
1199
the same physical repository via different URLs.
1201
if self.__class__ is not other.__class__:
1203
return (self._transport.base == other._transport.base)
1205
def is_in_write_group(self):
1206
"""Return True if there is an open write group.
1208
:seealso: start_write_group.
1210
return self._write_group is not None
1212
def is_locked(self):
1213
return self.control_files.is_locked()
1215
def is_write_locked(self):
1216
"""Return True if this object is write locked."""
1217
return self.is_locked() and self.control_files._lock_mode == 'w'
1219
def lock_write(self, token=None):
1220
"""Lock this repository for writing.
1222
This causes caching within the repository obejct to start accumlating
1223
data during reads, and allows a 'write_group' to be obtained. Write
1224
groups must be used for actual data insertion.
1226
:param token: if this is already locked, then lock_write will fail
1227
unless the token matches the existing lock.
1228
:returns: a token if this instance supports tokens, otherwise None.
1229
:raises TokenLockingNotSupported: when a token is given but this
1230
instance doesn't support using token locks.
1231
:raises MismatchedToken: if the specified token doesn't match the token
1232
of the existing lock.
1233
:seealso: start_write_group.
1235
A token should be passed in if you know that you have locked the object
1236
some other way, and need to synchronise this object's state with that
1239
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1241
locked = self.is_locked()
1242
result = self.control_files.lock_write(token=token)
1243
for repo in self._fallback_repositories:
1244
# Writes don't affect fallback repos
1247
self._refresh_data()
1250
def lock_read(self):
1251
locked = self.is_locked()
1252
self.control_files.lock_read()
1253
for repo in self._fallback_repositories:
1256
self._refresh_data()
1258
def get_physical_lock_status(self):
1259
return self.control_files.get_physical_lock_status()
1261
def leave_lock_in_place(self):
1262
"""Tell this repository not to release the physical lock when this
1265
If lock_write doesn't return a token, then this method is not supported.
1267
self.control_files.leave_in_place()
1269
def dont_leave_lock_in_place(self):
1270
"""Tell this repository to release the physical lock when this
1271
object is unlocked, even if it didn't originally acquire it.
1273
If lock_write doesn't return a token, then this method is not supported.
1275
self.control_files.dont_leave_in_place()
1278
def gather_stats(self, revid=None, committers=None):
1279
"""Gather statistics from a revision id.
1281
:param revid: The revision id to gather statistics from, if None, then
1282
no revision specific statistics are gathered.
1283
:param committers: Optional parameter controlling whether to grab
1284
a count of committers from the revision specific statistics.
1285
:return: A dictionary of statistics. Currently this contains:
1286
committers: The number of committers if requested.
1287
firstrev: A tuple with timestamp, timezone for the penultimate left
1288
most ancestor of revid, if revid is not the NULL_REVISION.
1289
latestrev: A tuple with timestamp, timezone for revid, if revid is
1290
not the NULL_REVISION.
1291
revisions: The total revision count in the repository.
1292
size: An estimate disk size of the repository in bytes.
1295
if revid and committers:
1296
result['committers'] = 0
1297
if revid and revid != _mod_revision.NULL_REVISION:
1299
all_committers = set()
1300
revisions = self.get_ancestry(revid)
1301
# pop the leading None
1303
first_revision = None
1305
# ignore the revisions in the middle - just grab first and last
1306
revisions = revisions[0], revisions[-1]
1307
for revision in self.get_revisions(revisions):
1308
if not first_revision:
1309
first_revision = revision
1311
all_committers.add(revision.committer)
1312
last_revision = revision
1314
result['committers'] = len(all_committers)
1315
result['firstrev'] = (first_revision.timestamp,
1316
first_revision.timezone)
1317
result['latestrev'] = (last_revision.timestamp,
1318
last_revision.timezone)
1320
# now gather global repository information
1321
# XXX: This is available for many repos regardless of listability.
1322
if self.bzrdir.root_transport.listable():
1323
# XXX: do we want to __define len__() ?
1324
# Maybe the versionedfiles object should provide a different
1325
# method to get the number of keys.
1326
result['revisions'] = len(self.revisions.keys())
1327
# result['size'] = t
1330
def find_branches(self, using=False):
1331
"""Find branches underneath this repository.
1333
This will include branches inside other branches.
1335
:param using: If True, list only branches using this repository.
1337
if using and not self.is_shared():
1339
return [self.bzrdir.open_branch()]
1340
except errors.NotBranchError:
1342
class Evaluator(object):
1345
self.first_call = True
1347
def __call__(self, bzrdir):
1348
# On the first call, the parameter is always the bzrdir
1349
# containing the current repo.
1350
if not self.first_call:
1352
repository = bzrdir.open_repository()
1353
except errors.NoRepositoryPresent:
1356
return False, (None, repository)
1357
self.first_call = False
1359
value = (bzrdir.open_branch(), None)
1360
except errors.NotBranchError:
1361
value = (None, None)
1365
for branch, repository in bzrdir.BzrDir.find_bzrdirs(
1366
self.bzrdir.root_transport, evaluate=Evaluator()):
1367
if branch is not None:
1368
branches.append(branch)
1369
if not using and repository is not None:
1370
branches.extend(repository.find_branches())
1374
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
1375
"""Return the revision ids that other has that this does not.
1377
These are returned in topological order.
1379
revision_id: only return revision ids included by revision_id.
1381
return InterRepository.get(other, self).search_missing_revision_ids(
1382
revision_id, find_ghosts)
1386
"""Open the repository rooted at base.
1388
For instance, if the repository is at URL/.bzr/repository,
1389
Repository.open(URL) -> a Repository instance.
1391
control = bzrdir.BzrDir.open(base)
1392
return control.open_repository()
1394
def copy_content_into(self, destination, revision_id=None):
1395
"""Make a complete copy of the content in self into destination.
1397
This is a destructive operation! Do not use it on existing
1400
return InterRepository.get(self, destination).copy_content(revision_id)
1402
def commit_write_group(self):
1403
"""Commit the contents accrued within the current write group.
1405
:seealso: start_write_group.
1407
if self._write_group is not self.get_transaction():
1408
# has an unlock or relock occured ?
1409
raise errors.BzrError('mismatched lock context %r and '
1411
(self.get_transaction(), self._write_group))
1412
self._commit_write_group()
1413
self._write_group = None
1415
def _commit_write_group(self):
1416
"""Template method for per-repository write group cleanup.
1418
This is called before the write group is considered to be
1419
finished and should ensure that all data handed to the repository
1420
for writing during the write group is safely committed (to the
1421
extent possible considering file system caching etc).
1424
def suspend_write_group(self):
1425
raise errors.UnsuspendableWriteGroup(self)
1427
def get_missing_parent_inventories(self):
1428
"""Return the keys of missing inventory parents for revisions added in
1431
A revision is not complete if the inventory delta for that revision
1432
cannot be calculated. Therefore if the parent inventories of a
1433
revision are not present, the revision is incomplete, and e.g. cannot
1434
be streamed by a smart server. This method finds missing inventory
1435
parents for revisions added in this write group.
1437
if not self._format.supports_external_lookups:
1438
# This is only an issue for stacked repositories
1440
if not self.is_in_write_group():
1441
raise AssertionError('not in a write group')
1443
# XXX: We assume that every added revision already has its
1444
# corresponding inventory, so we only check for parent inventories that
1445
# might be missing, rather than all inventories.
1446
parents = set(self.revisions._index.get_missing_parents())
1447
parents.discard(_mod_revision.NULL_REVISION)
1448
unstacked_inventories = self.inventories._index
1449
present_inventories = unstacked_inventories.get_parent_map(
1450
key[-1:] for key in parents)
1451
parents.difference_update(present_inventories)
1452
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1455
def refresh_data(self):
1456
"""Re-read any data needed to to synchronise with disk.
1458
This method is intended to be called after another repository instance
1459
(such as one used by a smart server) has inserted data into the
1460
repository. It may not be called during a write group, but may be
1461
called at any other time.
1463
if self.is_in_write_group():
1464
raise errors.InternalBzrError(
1465
"May not refresh_data while in a write group.")
1466
self._refresh_data()
1468
def resume_write_group(self, tokens):
1469
if not self.is_write_locked():
1470
raise errors.NotWriteLocked(self)
1471
if self._write_group:
1472
raise errors.BzrError('already in a write group')
1473
self._resume_write_group(tokens)
1474
# so we can detect unlock/relock - the write group is now entered.
1475
self._write_group = self.get_transaction()
1477
def _resume_write_group(self, tokens):
1478
raise errors.UnsuspendableWriteGroup(self)
1480
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1482
"""Fetch the content required to construct revision_id from source.
1484
If revision_id is None and fetch_spec is None, then all content is
1487
fetch() may not be used when the repository is in a write group -
1488
either finish the current write group before using fetch, or use
1489
fetch before starting the write group.
1491
:param find_ghosts: Find and copy revisions in the source that are
1492
ghosts in the target (and not reachable directly by walking out to
1493
the first-present revision in target from revision_id).
1494
:param revision_id: If specified, all the content needed for this
1495
revision ID will be copied to the target. Fetch will determine for
1496
itself which content needs to be copied.
1497
:param fetch_spec: If specified, a SearchResult or
1498
PendingAncestryResult that describes which revisions to copy. This
1499
allows copying multiple heads at once. Mutually exclusive with
1502
if fetch_spec is not None and revision_id is not None:
1503
raise AssertionError(
1504
"fetch_spec and revision_id are mutually exclusive.")
1505
if self.is_in_write_group():
1506
raise errors.InternalBzrError(
1507
"May not fetch while in a write group.")
1508
# fast path same-url fetch operations
1509
if self.has_same_location(source) and fetch_spec is None:
1510
# check that last_revision is in 'from' and then return a
1512
if (revision_id is not None and
1513
not _mod_revision.is_null(revision_id)):
1514
self.get_revision(revision_id)
1516
# if there is no specific appropriate InterRepository, this will get
1517
# the InterRepository base class, which raises an
1518
# IncompatibleRepositories when asked to fetch.
1519
inter = InterRepository.get(source, self)
1520
return inter.fetch(revision_id=revision_id, pb=pb,
1521
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1523
def create_bundle(self, target, base, fileobj, format=None):
1524
return serializer.write_bundle(self, target, base, fileobj, format)
1526
def get_commit_builder(self, branch, parents, config, timestamp=None,
1527
timezone=None, committer=None, revprops=None,
1529
"""Obtain a CommitBuilder for this repository.
1531
:param branch: Branch to commit to.
1532
:param parents: Revision ids of the parents of the new revision.
1533
:param config: Configuration to use.
1534
:param timestamp: Optional timestamp recorded for commit.
1535
:param timezone: Optional timezone for timestamp.
1536
:param committer: Optional committer to set for commit.
1537
:param revprops: Optional dictionary of revision properties.
1538
:param revision_id: Optional revision id.
1540
result = self._commit_builder_class(self, parents, config,
1541
timestamp, timezone, committer, revprops, revision_id)
1542
self.start_write_group()
1546
if (self.control_files._lock_count == 1 and
1547
self.control_files._lock_mode == 'w'):
1548
if self._write_group is not None:
1549
self.abort_write_group()
1550
self.control_files.unlock()
1551
raise errors.BzrError(
1552
'Must end write groups before releasing write locks.')
1553
self.control_files.unlock()
1554
if self.control_files._lock_count == 0:
1555
self._inventory_entry_cache.clear()
1556
for repo in self._fallback_repositories:
1560
def clone(self, a_bzrdir, revision_id=None):
1561
"""Clone this repository into a_bzrdir using the current format.
1563
Currently no check is made that the format of this repository and
1564
the bzrdir format are compatible. FIXME RBC 20060201.
1566
:return: The newly created destination repository.
1568
# TODO: deprecate after 0.16; cloning this with all its settings is
1569
# probably not very useful -- mbp 20070423
1570
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1571
self.copy_content_into(dest_repo, revision_id)
1574
def start_write_group(self):
1575
"""Start a write group in the repository.
1577
Write groups are used by repositories which do not have a 1:1 mapping
1578
between file ids and backend store to manage the insertion of data from
1579
both fetch and commit operations.
1581
A write lock is required around the start_write_group/commit_write_group
1582
for the support of lock-requiring repository formats.
1584
One can only insert data into a repository inside a write group.
1588
if not self.is_write_locked():
1589
raise errors.NotWriteLocked(self)
1590
if self._write_group:
1591
raise errors.BzrError('already in a write group')
1592
self._start_write_group()
1593
# so we can detect unlock/relock - the write group is now entered.
1594
self._write_group = self.get_transaction()
1596
def _start_write_group(self):
1597
"""Template method for per-repository write group startup.
1599
This is called before the write group is considered to be
1604
def sprout(self, to_bzrdir, revision_id=None):
1605
"""Create a descendent repository for new development.
1607
Unlike clone, this does not copy the settings of the repository.
1609
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1610
dest_repo.fetch(self, revision_id=revision_id)
1613
def _create_sprouting_repo(self, a_bzrdir, shared):
1614
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1615
# use target default format.
1616
dest_repo = a_bzrdir.create_repository()
1618
# Most control formats need the repository to be specifically
1619
# created, but on some old all-in-one formats it's not needed
1621
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1622
except errors.UninitializableFormat:
1623
dest_repo = a_bzrdir.open_repository()
1626
def _get_sink(self):
1627
"""Return a sink for streaming into this repository."""
1628
return StreamSink(self)
1630
def _get_source(self, to_format):
1631
"""Return a source for streaming from this repository."""
1632
return StreamSource(self, to_format)
1635
def has_revision(self, revision_id):
1636
"""True if this repository has a copy of the revision."""
1637
return revision_id in self.has_revisions((revision_id,))
1640
def has_revisions(self, revision_ids):
1641
"""Probe to find out the presence of multiple revisions.
1643
:param revision_ids: An iterable of revision_ids.
1644
:return: A set of the revision_ids that were present.
1646
parent_map = self.revisions.get_parent_map(
1647
[(rev_id,) for rev_id in revision_ids])
1649
if _mod_revision.NULL_REVISION in revision_ids:
1650
result.add(_mod_revision.NULL_REVISION)
1651
result.update([key[0] for key in parent_map])
1655
def get_revision(self, revision_id):
1656
"""Return the Revision object for a named revision."""
1657
return self.get_revisions([revision_id])[0]
1660
def get_revision_reconcile(self, revision_id):
1661
"""'reconcile' helper routine that allows access to a revision always.
1663
This variant of get_revision does not cross check the weave graph
1664
against the revision one as get_revision does: but it should only
1665
be used by reconcile, or reconcile-alike commands that are correcting
1666
or testing the revision graph.
1668
return self._get_revisions([revision_id])[0]
1671
def get_revisions(self, revision_ids):
1672
"""Get many revisions at once."""
1673
return self._get_revisions(revision_ids)
1676
def _get_revisions(self, revision_ids):
1677
"""Core work logic to get many revisions without sanity checks."""
1678
for rev_id in revision_ids:
1679
if not rev_id or not isinstance(rev_id, basestring):
1680
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1681
keys = [(key,) for key in revision_ids]
1682
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1684
for record in stream:
1685
if record.storage_kind == 'absent':
1686
raise errors.NoSuchRevision(self, record.key[0])
1687
text = record.get_bytes_as('fulltext')
1688
rev = self._serializer.read_revision_from_string(text)
1689
revs[record.key[0]] = rev
1690
return [revs[revid] for revid in revision_ids]
1693
def get_revision_xml(self, revision_id):
1694
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1695
# would have already do it.
1696
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1697
# TODO: this can't just be replaced by:
1698
# return self._serializer.write_revision_to_string(
1699
# self.get_revision(revision_id))
1700
# as cStringIO preservers the encoding unlike write_revision_to_string
1701
# or some other call down the path.
1702
rev = self.get_revision(revision_id)
1703
rev_tmp = cStringIO.StringIO()
1704
# the current serializer..
1705
self._serializer.write_revision(rev, rev_tmp)
1707
return rev_tmp.getvalue()
1709
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1710
"""Produce a generator of revision deltas.
1712
Note that the input is a sequence of REVISIONS, not revision_ids.
1713
Trees will be held in memory until the generator exits.
1714
Each delta is relative to the revision's lefthand predecessor.
1716
:param specific_fileids: if not None, the result is filtered
1717
so that only those file-ids, their parents and their
1718
children are included.
1720
# Get the revision-ids of interest
1721
required_trees = set()
1722
for revision in revisions:
1723
required_trees.add(revision.revision_id)
1724
required_trees.update(revision.parent_ids[:1])
1726
# Get the matching filtered trees. Note that it's more
1727
# efficient to pass filtered trees to changes_from() rather
1728
# than doing the filtering afterwards. changes_from() could
1729
# arguably do the filtering itself but it's path-based, not
1730
# file-id based, so filtering before or afterwards is
1732
if specific_fileids is None:
1733
trees = dict((t.get_revision_id(), t) for
1734
t in self.revision_trees(required_trees))
1736
trees = dict((t.get_revision_id(), t) for
1737
t in self._filtered_revision_trees(required_trees,
1740
# Calculate the deltas
1741
for revision in revisions:
1742
if not revision.parent_ids:
1743
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1745
old_tree = trees[revision.parent_ids[0]]
1746
yield trees[revision.revision_id].changes_from(old_tree)
1749
def get_revision_delta(self, revision_id, specific_fileids=None):
1750
"""Return the delta for one revision.
1752
The delta is relative to the left-hand predecessor of the
1755
:param specific_fileids: if not None, the result is filtered
1756
so that only those file-ids, their parents and their
1757
children are included.
1759
r = self.get_revision(revision_id)
1760
return list(self.get_deltas_for_revisions([r],
1761
specific_fileids=specific_fileids))[0]
1764
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1765
signature = gpg_strategy.sign(plaintext)
1766
self.add_signature_text(revision_id, signature)
1769
def add_signature_text(self, revision_id, signature):
1770
self.signatures.add_lines((revision_id,), (),
1771
osutils.split_lines(signature))
1773
def find_text_key_references(self):
1774
"""Find the text key references within the repository.
1776
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1777
to whether they were referred to by the inventory of the
1778
revision_id that they contain. The inventory texts from all present
1779
revision ids are assessed to generate this report.
1781
revision_keys = self.revisions.keys()
1782
w = self.inventories
1783
pb = ui.ui_factory.nested_progress_bar()
1785
return self._find_text_key_references_from_xml_inventory_lines(
1786
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1790
def _find_text_key_references_from_xml_inventory_lines(self,
1792
"""Core routine for extracting references to texts from inventories.
1794
This performs the translation of xml lines to revision ids.
1796
:param line_iterator: An iterator of lines, origin_version_id
1797
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1798
to whether they were referred to by the inventory of the
1799
revision_id that they contain. Note that if that revision_id was
1800
not part of the line_iterator's output then False will be given -
1801
even though it may actually refer to that key.
1803
if not self._serializer.support_altered_by_hack:
1804
raise AssertionError(
1805
"_find_text_key_references_from_xml_inventory_lines only "
1806
"supported for branches which store inventory as unnested xml"
1807
", not on %r" % self)
1810
# this code needs to read every new line in every inventory for the
1811
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
1812
# not present in one of those inventories is unnecessary but not
1813
# harmful because we are filtering by the revision id marker in the
1814
# inventory lines : we only select file ids altered in one of those
1815
# revisions. We don't need to see all lines in the inventory because
1816
# only those added in an inventory in rev X can contain a revision=X
1818
unescape_revid_cache = {}
1819
unescape_fileid_cache = {}
1821
# jam 20061218 In a big fetch, this handles hundreds of thousands
1822
# of lines, so it has had a lot of inlining and optimizing done.
1823
# Sorry that it is a little bit messy.
1824
# Move several functions to be local variables, since this is a long
1826
search = self._file_ids_altered_regex.search
1827
unescape = _unescape_xml
1828
setdefault = result.setdefault
1829
for line, line_key in line_iterator:
1830
match = search(line)
1833
# One call to match.group() returning multiple items is quite a
1834
# bit faster than 2 calls to match.group() each returning 1
1835
file_id, revision_id = match.group('file_id', 'revision_id')
1837
# Inlining the cache lookups helps a lot when you make 170,000
1838
# lines and 350k ids, versus 8.4 unique ids.
1839
# Using a cache helps in 2 ways:
1840
# 1) Avoids unnecessary decoding calls
1841
# 2) Re-uses cached strings, which helps in future set and
1843
# (2) is enough that removing encoding entirely along with
1844
# the cache (so we are using plain strings) results in no
1845
# performance improvement.
1847
revision_id = unescape_revid_cache[revision_id]
1849
unescaped = unescape(revision_id)
1850
unescape_revid_cache[revision_id] = unescaped
1851
revision_id = unescaped
1853
# Note that unconditionally unescaping means that we deserialise
1854
# every fileid, which for general 'pull' is not great, but we don't
1855
# really want to have some many fulltexts that this matters anyway.
1858
file_id = unescape_fileid_cache[file_id]
1860
unescaped = unescape(file_id)
1861
unescape_fileid_cache[file_id] = unescaped
1864
key = (file_id, revision_id)
1865
setdefault(key, False)
1866
if revision_id == line_key[-1]:
1870
def _inventory_xml_lines_for_keys(self, keys):
1871
"""Get a line iterator of the sort needed for findind references.
1873
Not relevant for non-xml inventory repositories.
1875
Ghosts in revision_keys are ignored.
1877
:param revision_keys: The revision keys for the inventories to inspect.
1878
:return: An iterator over (inventory line, revid) for the fulltexts of
1879
all of the xml inventories specified by revision_keys.
1881
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1882
for record in stream:
1883
if record.storage_kind != 'absent':
1884
chunks = record.get_bytes_as('chunked')
1885
revid = record.key[-1]
1886
lines = osutils.chunks_to_lines(chunks)
1890
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1892
"""Helper routine for fileids_altered_by_revision_ids.
1894
This performs the translation of xml lines to revision ids.
1896
:param line_iterator: An iterator of lines, origin_version_id
1897
:param revision_ids: The revision ids to filter for. This should be a
1898
set or other type which supports efficient __contains__ lookups, as
1899
the revision id from each parsed line will be looked up in the
1900
revision_ids filter.
1901
:return: a dictionary mapping altered file-ids to an iterable of
1902
revision_ids. Each altered file-ids has the exact revision_ids that
1903
altered it listed explicitly.
1905
seen = set(self._find_text_key_references_from_xml_inventory_lines(
1906
line_iterator).iterkeys())
1907
# Note that revision_ids are revision keys.
1908
parent_maps = self.revisions.get_parent_map(revision_ids)
1910
map(parents.update, parent_maps.itervalues())
1911
parents.difference_update(revision_ids)
1912
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
1913
self._inventory_xml_lines_for_keys(parents)))
1914
new_keys = seen - parent_seen
1916
setdefault = result.setdefault
1917
for key in new_keys:
1918
setdefault(key[0], set()).add(key[-1])
1921
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
1922
"""Find the file ids and versions affected by revisions.
1924
:param revisions: an iterable containing revision ids.
1925
:param _inv_weave: The inventory weave from this repository or None.
1926
If None, the inventory weave will be opened automatically.
1927
:return: a dictionary mapping altered file-ids to an iterable of
1928
revision_ids. Each altered file-ids has the exact revision_ids that
1929
altered it listed explicitly.
1931
selected_keys = set((revid,) for revid in revision_ids)
1932
w = _inv_weave or self.inventories
1933
pb = ui.ui_factory.nested_progress_bar()
1935
return self._find_file_ids_from_xml_inventory_lines(
1936
w.iter_lines_added_or_present_in_keys(
1937
selected_keys, pb=pb),
1942
def iter_files_bytes(self, desired_files):
1943
"""Iterate through file versions.
1945
Files will not necessarily be returned in the order they occur in
1946
desired_files. No specific order is guaranteed.
1948
Yields pairs of identifier, bytes_iterator. identifier is an opaque
1949
value supplied by the caller as part of desired_files. It should
1950
uniquely identify the file version in the caller's context. (Examples:
1951
an index number or a TreeTransform trans_id.)
1953
bytes_iterator is an iterable of bytestrings for the file. The
1954
kind of iterable and length of the bytestrings are unspecified, but for
1955
this implementation, it is a list of bytes produced by
1956
VersionedFile.get_record_stream().
1958
:param desired_files: a list of (file_id, revision_id, identifier)
1962
for file_id, revision_id, callable_data in desired_files:
1963
text_keys[(file_id, revision_id)] = callable_data
1964
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
1965
if record.storage_kind == 'absent':
1966
raise errors.RevisionNotPresent(record.key, self)
1967
yield text_keys[record.key], record.get_bytes_as('chunked')
1969
def _generate_text_key_index(self, text_key_references=None,
1971
"""Generate a new text key index for the repository.
1973
This is an expensive function that will take considerable time to run.
1975
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
1976
list of parents, also text keys. When a given key has no parents,
1977
the parents list will be [NULL_REVISION].
1979
# All revisions, to find inventory parents.
1980
if ancestors is None:
1981
graph = self.get_graph()
1982
ancestors = graph.get_parent_map(self.all_revision_ids())
1983
if text_key_references is None:
1984
text_key_references = self.find_text_key_references()
1985
pb = ui.ui_factory.nested_progress_bar()
1987
return self._do_generate_text_key_index(ancestors,
1988
text_key_references, pb)
1992
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
1993
"""Helper for _generate_text_key_index to avoid deep nesting."""
1994
revision_order = tsort.topo_sort(ancestors)
1995
invalid_keys = set()
1997
for revision_id in revision_order:
1998
revision_keys[revision_id] = set()
1999
text_count = len(text_key_references)
2000
# a cache of the text keys to allow reuse; costs a dict of all the
2001
# keys, but saves a 2-tuple for every child of a given key.
2003
for text_key, valid in text_key_references.iteritems():
2005
invalid_keys.add(text_key)
2007
revision_keys[text_key[1]].add(text_key)
2008
text_key_cache[text_key] = text_key
2009
del text_key_references
2011
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2012
NULL_REVISION = _mod_revision.NULL_REVISION
2013
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2014
# too small for large or very branchy trees. However, for 55K path
2015
# trees, it would be easy to use too much memory trivially. Ideally we
2016
# could gauge this by looking at available real memory etc, but this is
2017
# always a tricky proposition.
2018
inventory_cache = lru_cache.LRUCache(10)
2019
batch_size = 10 # should be ~150MB on a 55K path tree
2020
batch_count = len(revision_order) / batch_size + 1
2022
pb.update("Calculating text parents", processed_texts, text_count)
2023
for offset in xrange(batch_count):
2024
to_query = revision_order[offset * batch_size:(offset + 1) *
2028
for rev_tree in self.revision_trees(to_query):
2029
revision_id = rev_tree.get_revision_id()
2030
parent_ids = ancestors[revision_id]
2031
for text_key in revision_keys[revision_id]:
2032
pb.update("Calculating text parents", processed_texts)
2033
processed_texts += 1
2034
candidate_parents = []
2035
for parent_id in parent_ids:
2036
parent_text_key = (text_key[0], parent_id)
2038
check_parent = parent_text_key not in \
2039
revision_keys[parent_id]
2041
# the parent parent_id is a ghost:
2042
check_parent = False
2043
# truncate the derived graph against this ghost.
2044
parent_text_key = None
2046
# look at the parent commit details inventories to
2047
# determine possible candidates in the per file graph.
2050
inv = inventory_cache[parent_id]
2052
inv = self.revision_tree(parent_id).inventory
2053
inventory_cache[parent_id] = inv
2055
parent_entry = inv[text_key[0]]
2056
except (KeyError, errors.NoSuchId):
2058
if parent_entry is not None:
2060
text_key[0], parent_entry.revision)
2062
parent_text_key = None
2063
if parent_text_key is not None:
2064
candidate_parents.append(
2065
text_key_cache[parent_text_key])
2066
parent_heads = text_graph.heads(candidate_parents)
2067
new_parents = list(parent_heads)
2068
new_parents.sort(key=lambda x:candidate_parents.index(x))
2069
if new_parents == []:
2070
new_parents = [NULL_REVISION]
2071
text_index[text_key] = new_parents
2073
for text_key in invalid_keys:
2074
text_index[text_key] = [NULL_REVISION]
2077
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2078
"""Get an iterable listing the keys of all the data introduced by a set
2081
The keys will be ordered so that the corresponding items can be safely
2082
fetched and inserted in that order.
2084
:returns: An iterable producing tuples of (knit-kind, file-id,
2085
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2086
'revisions'. file-id is None unless knit-kind is 'file'.
2088
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2091
for result in self._find_non_file_keys_to_fetch(revision_ids):
2094
def _find_file_keys_to_fetch(self, revision_ids, pb):
2095
# XXX: it's a bit weird to control the inventory weave caching in this
2096
# generator. Ideally the caching would be done in fetch.py I think. Or
2097
# maybe this generator should explicitly have the contract that it
2098
# should not be iterated until the previously yielded item has been
2100
inv_w = self.inventories
2102
# file ids that changed
2103
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2105
num_file_ids = len(file_ids)
2106
for file_id, altered_versions in file_ids.iteritems():
2108
pb.update("fetch texts", count, num_file_ids)
2110
yield ("file", file_id, altered_versions)
2112
def _find_non_file_keys_to_fetch(self, revision_ids):
2114
yield ("inventory", None, revision_ids)
2117
# XXX: Note ATM no callers actually pay attention to this return
2118
# instead they just use the list of revision ids and ignore
2119
# missing sigs. Consider removing this work entirely
2120
revisions_with_signatures = set(self.signatures.get_parent_map(
2121
[(r,) for r in revision_ids]))
2122
revisions_with_signatures = set(
2123
[r for (r,) in revisions_with_signatures])
2124
revisions_with_signatures.intersection_update(revision_ids)
2125
yield ("signatures", None, revisions_with_signatures)
2128
yield ("revisions", None, revision_ids)
2131
def get_inventory(self, revision_id):
2132
"""Get Inventory object by revision id."""
2133
return self.iter_inventories([revision_id]).next()
2135
def iter_inventories(self, revision_ids):
2136
"""Get many inventories by revision_ids.
2138
This will buffer some or all of the texts used in constructing the
2139
inventories in memory, but will only parse a single inventory at a
2142
:param revision_ids: The expected revision ids of the inventories.
2143
:return: An iterator of inventories.
2145
if ((None in revision_ids)
2146
or (_mod_revision.NULL_REVISION in revision_ids)):
2147
raise ValueError('cannot get null revision inventory')
2148
return self._iter_inventories(revision_ids)
2150
def _iter_inventories(self, revision_ids):
2151
"""single-document based inventory iteration."""
2152
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2153
yield self.deserialise_inventory(revision_id, text)
2155
def _iter_inventory_xmls(self, revision_ids):
2156
keys = [(revision_id,) for revision_id in revision_ids]
2157
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2159
for record in stream:
2160
if record.storage_kind != 'absent':
2161
text_chunks[record.key] = record.get_bytes_as('chunked')
2163
raise errors.NoSuchRevision(self, record.key)
2165
chunks = text_chunks.pop(key)
2166
yield ''.join(chunks), key[-1]
2168
def deserialise_inventory(self, revision_id, xml):
2169
"""Transform the xml into an inventory object.
2171
:param revision_id: The expected revision id of the inventory.
2172
:param xml: A serialised inventory.
2174
result = self._serializer.read_inventory_from_string(xml, revision_id,
2175
entry_cache=self._inventory_entry_cache)
2176
if result.revision_id != revision_id:
2177
raise AssertionError('revision id mismatch %s != %s' % (
2178
result.revision_id, revision_id))
2181
def serialise_inventory(self, inv):
2182
return self._serializer.write_inventory_to_string(inv)
2184
def _serialise_inventory_to_lines(self, inv):
2185
return self._serializer.write_inventory_to_lines(inv)
2187
def get_serializer_format(self):
2188
return self._serializer.format_num
2191
def get_inventory_xml(self, revision_id):
2192
"""Get inventory XML as a file object."""
2193
texts = self._iter_inventory_xmls([revision_id])
2195
text, revision_id = texts.next()
2196
except StopIteration:
2197
raise errors.HistoryMissing(self, 'inventory', revision_id)
2201
def get_inventory_sha1(self, revision_id):
2202
"""Return the sha1 hash of the inventory entry
2204
return self.get_revision(revision_id).inventory_sha1
2206
def iter_reverse_revision_history(self, revision_id):
2207
"""Iterate backwards through revision ids in the lefthand history
2209
:param revision_id: The revision id to start with. All its lefthand
2210
ancestors will be traversed.
2212
graph = self.get_graph()
2213
next_id = revision_id
2215
if next_id in (None, _mod_revision.NULL_REVISION):
2218
# Note: The following line may raise KeyError in the event of
2219
# truncated history. We decided not to have a try:except:raise
2220
# RevisionNotPresent here until we see a use for it, because of the
2221
# cost in an inner loop that is by its very nature O(history).
2222
# Robert Collins 20080326
2223
parents = graph.get_parent_map([next_id])[next_id]
2224
if len(parents) == 0:
2227
next_id = parents[0]
2230
def get_revision_inventory(self, revision_id):
2231
"""Return inventory of a past revision."""
2232
# TODO: Unify this with get_inventory()
2233
# bzr 0.0.6 and later imposes the constraint that the inventory_id
2234
# must be the same as its revision, so this is trivial.
2235
if revision_id is None:
2236
# This does not make sense: if there is no revision,
2237
# then it is the current tree inventory surely ?!
2238
# and thus get_root_id() is something that looks at the last
2239
# commit on the branch, and the get_root_id is an inventory check.
2240
raise NotImplementedError
2241
# return Inventory(self.get_root_id())
2243
return self.get_inventory(revision_id)
2245
def is_shared(self):
2246
"""Return True if this repository is flagged as a shared repository."""
2247
raise NotImplementedError(self.is_shared)
2250
def reconcile(self, other=None, thorough=False):
2251
"""Reconcile this repository."""
2252
from bzrlib.reconcile import RepoReconciler
2253
reconciler = RepoReconciler(self, thorough=thorough)
2254
reconciler.reconcile()
2257
def _refresh_data(self):
2258
"""Helper called from lock_* to ensure coherency with disk.
2260
The default implementation does nothing; it is however possible
2261
for repositories to maintain loaded indices across multiple locks
2262
by checking inside their implementation of this method to see
2263
whether their indices are still valid. This depends of course on
2264
the disk format being validatable in this manner. This method is
2265
also called by the refresh_data() public interface to cause a refresh
2266
to occur while in a write lock so that data inserted by a smart server
2267
push operation is visible on the client's instance of the physical
2272
def revision_tree(self, revision_id):
2273
"""Return Tree for a revision on this branch.
2275
`revision_id` may be NULL_REVISION for the empty tree revision.
2277
revision_id = _mod_revision.ensure_null(revision_id)
2278
# TODO: refactor this to use an existing revision object
2279
# so we don't need to read it in twice.
2280
if revision_id == _mod_revision.NULL_REVISION:
2281
return RevisionTree(self, Inventory(root_id=None),
2282
_mod_revision.NULL_REVISION)
2284
inv = self.get_revision_inventory(revision_id)
2285
return RevisionTree(self, inv, revision_id)
2287
def revision_trees(self, revision_ids):
2288
"""Return Trees for revisions in this repository.
2290
:param revision_ids: a sequence of revision-ids;
2291
a revision-id may not be None or 'null:'
2293
inventories = self.iter_inventories(revision_ids)
2294
for inv in inventories:
2295
yield RevisionTree(self, inv, inv.revision_id)
2297
def _filtered_revision_trees(self, revision_ids, file_ids):
2298
"""Return Tree for a revision on this branch with only some files.
2300
:param revision_ids: a sequence of revision-ids;
2301
a revision-id may not be None or 'null:'
2302
:param file_ids: if not None, the result is filtered
2303
so that only those file-ids, their parents and their
2304
children are included.
2306
inventories = self.iter_inventories(revision_ids)
2307
for inv in inventories:
2308
# Should we introduce a FilteredRevisionTree class rather
2309
# than pre-filter the inventory here?
2310
filtered_inv = inv.filter(file_ids)
2311
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2314
def get_ancestry(self, revision_id, topo_sorted=True):
2315
"""Return a list of revision-ids integrated by a revision.
2317
The first element of the list is always None, indicating the origin
2318
revision. This might change when we have history horizons, or
2319
perhaps we should have a new API.
2321
This is topologically sorted.
2323
if _mod_revision.is_null(revision_id):
2325
if not self.has_revision(revision_id):
2326
raise errors.NoSuchRevision(self, revision_id)
2327
graph = self.get_graph()
2329
search = graph._make_breadth_first_searcher([revision_id])
2332
found, ghosts = search.next_with_ghosts()
2333
except StopIteration:
2336
if _mod_revision.NULL_REVISION in keys:
2337
keys.remove(_mod_revision.NULL_REVISION)
2339
parent_map = graph.get_parent_map(keys)
2340
keys = tsort.topo_sort(parent_map)
2341
return [None] + list(keys)
2344
"""Compress the data within the repository.
2346
This operation only makes sense for some repository types. For other
2347
types it should be a no-op that just returns.
2349
This stub method does not require a lock, but subclasses should use
2350
@needs_write_lock as this is a long running call its reasonable to
2351
implicitly lock for the user.
2354
def get_transaction(self):
2355
return self.control_files.get_transaction()
2357
def get_parent_map(self, revision_ids):
2358
"""See graph._StackedParentsProvider.get_parent_map"""
2359
# revisions index works in keys; this just works in revisions
2360
# therefore wrap and unwrap
2363
for revision_id in revision_ids:
2364
if revision_id == _mod_revision.NULL_REVISION:
2365
result[revision_id] = ()
2366
elif revision_id is None:
2367
raise ValueError('get_parent_map(None) is not valid')
2369
query_keys.append((revision_id ,))
2370
for ((revision_id,), parent_keys) in \
2371
self.revisions.get_parent_map(query_keys).iteritems():
2373
result[revision_id] = tuple(parent_revid
2374
for (parent_revid,) in parent_keys)
2376
result[revision_id] = (_mod_revision.NULL_REVISION,)
2379
def _make_parents_provider(self):
2382
def get_graph(self, other_repository=None):
2383
"""Return the graph walker for this repository format"""
2384
parents_provider = self._make_parents_provider()
2385
if (other_repository is not None and
2386
not self.has_same_location(other_repository)):
2387
parents_provider = graph._StackedParentsProvider(
2388
[parents_provider, other_repository._make_parents_provider()])
2389
return graph.Graph(parents_provider)
2391
def _get_versioned_file_checker(self, text_key_references=None):
2392
"""Return an object suitable for checking versioned files.
2394
:param text_key_references: if non-None, an already built
2395
dictionary mapping text keys ((fileid, revision_id) tuples)
2396
to whether they were referred to by the inventory of the
2397
revision_id that they contain. If None, this will be
2400
return _VersionedFileChecker(self,
2401
text_key_references=text_key_references)
2403
def revision_ids_to_search_result(self, result_set):
2404
"""Convert a set of revision ids to a graph SearchResult."""
2405
result_parents = set()
2406
for parents in self.get_graph().get_parent_map(
2407
result_set).itervalues():
2408
result_parents.update(parents)
2409
included_keys = result_set.intersection(result_parents)
2410
start_keys = result_set.difference(included_keys)
2411
exclude_keys = result_parents.difference(result_set)
2412
result = graph.SearchResult(start_keys, exclude_keys,
2413
len(result_set), result_set)
2417
def set_make_working_trees(self, new_value):
2418
"""Set the policy flag for making working trees when creating branches.
2420
This only applies to branches that use this repository.
2422
The default is 'True'.
2423
:param new_value: True to restore the default, False to disable making
2426
raise NotImplementedError(self.set_make_working_trees)
2428
def make_working_trees(self):
2429
"""Returns the policy for making working trees on new branches."""
2430
raise NotImplementedError(self.make_working_trees)
2433
def sign_revision(self, revision_id, gpg_strategy):
2434
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2435
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2438
def has_signature_for_revision_id(self, revision_id):
2439
"""Query for a revision signature for revision_id in the repository."""
2440
if not self.has_revision(revision_id):
2441
raise errors.NoSuchRevision(self, revision_id)
2442
sig_present = (1 == len(
2443
self.signatures.get_parent_map([(revision_id,)])))
2447
def get_signature_text(self, revision_id):
2448
"""Return the text for a signature."""
2449
stream = self.signatures.get_record_stream([(revision_id,)],
2451
record = stream.next()
2452
if record.storage_kind == 'absent':
2453
raise errors.NoSuchRevision(self, revision_id)
2454
return record.get_bytes_as('fulltext')
2457
def check(self, revision_ids=None):
2458
"""Check consistency of all history of given revision_ids.
2460
Different repository implementations should override _check().
2462
:param revision_ids: A non-empty list of revision_ids whose ancestry
2463
will be checked. Typically the last revision_id of a branch.
2465
return self._check(revision_ids)
2467
def _check(self, revision_ids):
2468
result = check.Check(self)
2472
def _warn_if_deprecated(self):
2473
global _deprecation_warning_done
2474
if _deprecation_warning_done:
2476
_deprecation_warning_done = True
2477
warning("Format %s for %s is deprecated - please use 'bzr upgrade' to get better performance"
2478
% (self._format, self.bzrdir.transport.base))
2480
def supports_rich_root(self):
2481
return self._format.rich_root_data
2483
def _check_ascii_revisionid(self, revision_id, method):
2484
"""Private helper for ascii-only repositories."""
2485
# weave repositories refuse to store revisionids that are non-ascii.
2486
if revision_id is not None:
2487
# weaves require ascii revision ids.
2488
if isinstance(revision_id, unicode):
2490
revision_id.encode('ascii')
2491
except UnicodeEncodeError:
2492
raise errors.NonAsciiRevisionId(method, self)
2495
revision_id.decode('ascii')
2496
except UnicodeDecodeError:
2497
raise errors.NonAsciiRevisionId(method, self)
2499
def revision_graph_can_have_wrong_parents(self):
2500
"""Is it possible for this repository to have a revision graph with
2503
If True, then this repository must also implement
2504
_find_inconsistent_revision_parents so that check and reconcile can
2505
check for inconsistencies before proceeding with other checks that may
2506
depend on the revision index being consistent.
2508
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2511
# remove these delegates a while after bzr 0.15
2512
def __make_delegated(name, from_module):
2513
def _deprecated_repository_forwarder():
2514
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2515
% (name, from_module),
2518
m = __import__(from_module, globals(), locals(), [name])
2520
return getattr(m, name)
2521
except AttributeError:
2522
raise AttributeError('module %s has no name %s'
2524
globals()[name] = _deprecated_repository_forwarder
2527
'AllInOneRepository',
2528
'WeaveMetaDirRepository',
2529
'PreSplitOutRepositoryFormat',
2530
'RepositoryFormat4',
2531
'RepositoryFormat5',
2532
'RepositoryFormat6',
2533
'RepositoryFormat7',
2535
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2539
'RepositoryFormatKnit',
2540
'RepositoryFormatKnit1',
2542
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2545
def install_revision(repository, rev, revision_tree):
2546
"""Install all revision data into a repository."""
2547
install_revisions(repository, [(rev, revision_tree, None)])
2550
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2551
"""Install all revision data into a repository.
2553
Accepts an iterable of revision, tree, signature tuples. The signature
2556
repository.start_write_group()
2558
inventory_cache = lru_cache.LRUCache(10)
2559
for n, (revision, revision_tree, signature) in enumerate(iterable):
2560
_install_revision(repository, revision, revision_tree, signature,
2563
pb.update('Transferring revisions', n + 1, num_revisions)
2565
repository.abort_write_group()
2568
repository.commit_write_group()
2571
def _install_revision(repository, rev, revision_tree, signature,
2573
"""Install all revision data into a repository."""
2574
present_parents = []
2576
for p_id in rev.parent_ids:
2577
if repository.has_revision(p_id):
2578
present_parents.append(p_id)
2579
parent_trees[p_id] = repository.revision_tree(p_id)
2581
parent_trees[p_id] = repository.revision_tree(
2582
_mod_revision.NULL_REVISION)
2584
inv = revision_tree.inventory
2585
entries = inv.iter_entries()
2586
# backwards compatibility hack: skip the root id.
2587
if not repository.supports_rich_root():
2588
path, root = entries.next()
2589
if root.revision != rev.revision_id:
2590
raise errors.IncompatibleRevision(repr(repository))
2592
for path, ie in entries:
2593
text_keys[(ie.file_id, ie.revision)] = ie
2594
text_parent_map = repository.texts.get_parent_map(text_keys)
2595
missing_texts = set(text_keys) - set(text_parent_map)
2596
# Add the texts that are not already present
2597
for text_key in missing_texts:
2598
ie = text_keys[text_key]
2600
# FIXME: TODO: The following loop overlaps/duplicates that done by
2601
# commit to determine parents. There is a latent/real bug here where
2602
# the parents inserted are not those commit would do - in particular
2603
# they are not filtered by heads(). RBC, AB
2604
for revision, tree in parent_trees.iteritems():
2605
if ie.file_id not in tree:
2607
parent_id = tree.inventory[ie.file_id].revision
2608
if parent_id in text_parents:
2610
text_parents.append((ie.file_id, parent_id))
2611
lines = revision_tree.get_file(ie.file_id).readlines()
2612
repository.texts.add_lines(text_key, text_parents, lines)
2614
# install the inventory
2615
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2616
# Cache this inventory
2617
inventory_cache[rev.revision_id] = inv
2619
basis_inv = inventory_cache[rev.parent_ids[0]]
2621
repository.add_inventory(rev.revision_id, inv, present_parents)
2623
delta = inv._make_delta(basis_inv)
2624
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2625
rev.revision_id, present_parents)
2627
repository.add_inventory(rev.revision_id, inv, present_parents)
2628
except errors.RevisionAlreadyPresent:
2630
if signature is not None:
2631
repository.add_signature_text(rev.revision_id, signature)
2632
repository.add_revision(rev.revision_id, rev, inv)
2635
class MetaDirRepository(Repository):
2636
"""Repositories in the new meta-dir layout.
2638
:ivar _transport: Transport for access to repository control files,
2639
typically pointing to .bzr/repository.
2642
def __init__(self, _format, a_bzrdir, control_files):
2643
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2644
self._transport = control_files._transport
2646
def is_shared(self):
2647
"""Return True if this repository is flagged as a shared repository."""
2648
return self._transport.has('shared-storage')
2651
def set_make_working_trees(self, new_value):
2652
"""Set the policy flag for making working trees when creating branches.
2654
This only applies to branches that use this repository.
2656
The default is 'True'.
2657
:param new_value: True to restore the default, False to disable making
2662
self._transport.delete('no-working-trees')
2663
except errors.NoSuchFile:
2666
self._transport.put_bytes('no-working-trees', '',
2667
mode=self.bzrdir._get_file_mode())
2669
def make_working_trees(self):
2670
"""Returns the policy for making working trees on new branches."""
2671
return not self._transport.has('no-working-trees')
2674
class MetaDirVersionedFileRepository(MetaDirRepository):
2675
"""Repositories in a meta-dir, that work via versioned file objects."""
2677
def __init__(self, _format, a_bzrdir, control_files):
2678
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2682
network_format_registry = registry.FormatRegistry()
2683
"""Registry of formats indexed by their network name.
2685
The network name for a repository format is an identifier that can be used when
2686
referring to formats with smart server operations. See
2687
RepositoryFormat.network_name() for more detail.
2691
format_registry = registry.FormatRegistry(network_format_registry)
2692
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
2694
This can contain either format instances themselves, or classes/factories that
2695
can be called to obtain one.
2699
#####################################################################
2700
# Repository Formats
2702
class RepositoryFormat(object):
2703
"""A repository format.
2705
Formats provide four things:
2706
* An initialization routine to construct repository data on disk.
2707
* a optional format string which is used when the BzrDir supports
2709
* an open routine which returns a Repository instance.
2710
* A network name for referring to the format in smart server RPC
2713
There is one and only one Format subclass for each on-disk format. But
2714
there can be one Repository subclass that is used for several different
2715
formats. The _format attribute on a Repository instance can be used to
2716
determine the disk format.
2718
Formats are placed in a registry by their format string for reference
2719
during opening. These should be subclasses of RepositoryFormat for
2722
Once a format is deprecated, just deprecate the initialize and open
2723
methods on the format class. Do not deprecate the object, as the
2724
object may be created even when a repository instance hasn't been
2727
Common instance attributes:
2728
_matchingbzrdir - the bzrdir format that the repository format was
2729
originally written to work with. This can be used if manually
2730
constructing a bzrdir and repository, or more commonly for test suite
2734
# Set to True or False in derived classes. True indicates that the format
2735
# supports ghosts gracefully.
2736
supports_ghosts = None
2737
# Can this repository be given external locations to lookup additional
2738
# data. Set to True or False in derived classes.
2739
supports_external_lookups = None
2740
# Does this format support CHK bytestring lookups. Set to True or False in
2742
supports_chks = None
2743
# Should commit add an inventory, or an inventory delta to the repository.
2744
_commit_inv_deltas = True
2745
# What order should fetch operations request streams in?
2746
# The default is unordered as that is the cheapest for an origin to
2748
_fetch_order = 'unordered'
2749
# Does this repository format use deltas that can be fetched as-deltas ?
2750
# (E.g. knits, where the knit deltas can be transplanted intact.
2751
# We default to False, which will ensure that enough data to get
2752
# a full text out of any fetch stream will be grabbed.
2753
_fetch_uses_deltas = False
2754
# Should fetch trigger a reconcile after the fetch? Only needed for
2755
# some repository formats that can suffer internal inconsistencies.
2756
_fetch_reconcile = False
2757
# Does this format have < O(tree_size) delta generation. Used to hint what
2758
# code path for commit, amongst other things.
2762
return "<%s>" % self.__class__.__name__
2764
def __eq__(self, other):
2765
# format objects are generally stateless
2766
return isinstance(other, self.__class__)
2768
def __ne__(self, other):
2769
return not self == other
2772
def find_format(klass, a_bzrdir):
2773
"""Return the format for the repository object in a_bzrdir.
2775
This is used by bzr native formats that have a "format" file in
2776
the repository. Other methods may be used by different types of
2780
transport = a_bzrdir.get_repository_transport(None)
2781
format_string = transport.get("format").read()
2782
return format_registry.get(format_string)
2783
except errors.NoSuchFile:
2784
raise errors.NoRepositoryPresent(a_bzrdir)
2786
raise errors.UnknownFormatError(format=format_string,
2790
def register_format(klass, format):
2791
format_registry.register(format.get_format_string(), format)
2794
def unregister_format(klass, format):
2795
format_registry.remove(format.get_format_string())
2798
def get_default_format(klass):
2799
"""Return the current default format."""
2800
from bzrlib import bzrdir
2801
return bzrdir.format_registry.make_bzrdir('default').repository_format
2803
def get_format_string(self):
2804
"""Return the ASCII format string that identifies this format.
2806
Note that in pre format ?? repositories the format string is
2807
not permitted nor written to disk.
2809
raise NotImplementedError(self.get_format_string)
2811
def get_format_description(self):
2812
"""Return the short description for this format."""
2813
raise NotImplementedError(self.get_format_description)
2815
# TODO: this shouldn't be in the base class, it's specific to things that
2816
# use weaves or knits -- mbp 20070207
2817
def _get_versioned_file_store(self,
2822
versionedfile_class=None,
2823
versionedfile_kwargs={},
2825
if versionedfile_class is None:
2826
versionedfile_class = self._versionedfile_class
2827
weave_transport = control_files._transport.clone(name)
2828
dir_mode = control_files._dir_mode
2829
file_mode = control_files._file_mode
2830
return VersionedFileStore(weave_transport, prefixed=prefixed,
2832
file_mode=file_mode,
2833
versionedfile_class=versionedfile_class,
2834
versionedfile_kwargs=versionedfile_kwargs,
2837
def initialize(self, a_bzrdir, shared=False):
2838
"""Initialize a repository of this format in a_bzrdir.
2840
:param a_bzrdir: The bzrdir to put the new repository in it.
2841
:param shared: The repository should be initialized as a sharable one.
2842
:returns: The new repository object.
2844
This may raise UninitializableFormat if shared repository are not
2845
compatible the a_bzrdir.
2847
raise NotImplementedError(self.initialize)
2849
def is_supported(self):
2850
"""Is this format supported?
2852
Supported formats must be initializable and openable.
2853
Unsupported formats may not support initialization or committing or
2854
some other features depending on the reason for not being supported.
2858
def network_name(self):
2859
"""A simple byte string uniquely identifying this format for RPC calls.
2861
MetaDir repository formats use their disk format string to identify the
2862
repository over the wire. All in one formats such as bzr < 0.8, and
2863
foreign formats like svn/git and hg should use some marker which is
2864
unique and immutable.
2866
raise NotImplementedError(self.network_name)
2868
def check_conversion_target(self, target_format):
2869
raise NotImplementedError(self.check_conversion_target)
2871
def open(self, a_bzrdir, _found=False):
2872
"""Return an instance of this format for the bzrdir a_bzrdir.
2874
_found is a private parameter, do not use it.
2876
raise NotImplementedError(self.open)
2879
class MetaDirRepositoryFormat(RepositoryFormat):
2880
"""Common base class for the new repositories using the metadir layout."""
2882
rich_root_data = False
2883
supports_tree_reference = False
2884
supports_external_lookups = False
2887
def _matchingbzrdir(self):
2888
matching = bzrdir.BzrDirMetaFormat1()
2889
matching.repository_format = self
2893
super(MetaDirRepositoryFormat, self).__init__()
2895
def _create_control_files(self, a_bzrdir):
2896
"""Create the required files and the initial control_files object."""
2897
# FIXME: RBC 20060125 don't peek under the covers
2898
# NB: no need to escape relative paths that are url safe.
2899
repository_transport = a_bzrdir.get_repository_transport(self)
2900
control_files = lockable_files.LockableFiles(repository_transport,
2901
'lock', lockdir.LockDir)
2902
control_files.create_lock()
2903
return control_files
2905
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
2906
"""Upload the initial blank content."""
2907
control_files = self._create_control_files(a_bzrdir)
2908
control_files.lock_write()
2909
transport = control_files._transport
2911
utf8_files += [('shared-storage', '')]
2913
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
2914
for (filename, content_stream) in files:
2915
transport.put_file(filename, content_stream,
2916
mode=a_bzrdir._get_file_mode())
2917
for (filename, content_bytes) in utf8_files:
2918
transport.put_bytes_non_atomic(filename, content_bytes,
2919
mode=a_bzrdir._get_file_mode())
2921
control_files.unlock()
2923
def network_name(self):
2924
"""Metadir formats have matching disk and network format strings."""
2925
return self.get_format_string()
2928
# Pre-0.8 formats that don't have a disk format string (because they are
2929
# versioned by the matching control directory). We use the control directories
2930
# disk format string as a key for the network_name because they meet the
2931
# constraints (simple string, unique, immutable).
2932
network_format_registry.register_lazy(
2933
"Bazaar-NG branch, format 5\n",
2934
'bzrlib.repofmt.weaverepo',
2935
'RepositoryFormat5',
2937
network_format_registry.register_lazy(
2938
"Bazaar-NG branch, format 6\n",
2939
'bzrlib.repofmt.weaverepo',
2940
'RepositoryFormat6',
2943
# formats which have no format string are not discoverable or independently
2944
# creatable on disk, so are not registered in format_registry. They're
2945
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
2946
# needed, it's constructed directly by the BzrDir. Non-native formats where
2947
# the repository is not separately opened are similar.
2949
format_registry.register_lazy(
2950
'Bazaar-NG Repository format 7',
2951
'bzrlib.repofmt.weaverepo',
2955
format_registry.register_lazy(
2956
'Bazaar-NG Knit Repository Format 1',
2957
'bzrlib.repofmt.knitrepo',
2958
'RepositoryFormatKnit1',
2961
format_registry.register_lazy(
2962
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
2963
'bzrlib.repofmt.knitrepo',
2964
'RepositoryFormatKnit3',
2967
format_registry.register_lazy(
2968
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
2969
'bzrlib.repofmt.knitrepo',
2970
'RepositoryFormatKnit4',
2973
# Pack-based formats. There is one format for pre-subtrees, and one for
2974
# post-subtrees to allow ease of testing.
2975
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
2976
format_registry.register_lazy(
2977
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
2978
'bzrlib.repofmt.pack_repo',
2979
'RepositoryFormatKnitPack1',
2981
format_registry.register_lazy(
2982
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
2983
'bzrlib.repofmt.pack_repo',
2984
'RepositoryFormatKnitPack3',
2986
format_registry.register_lazy(
2987
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
2988
'bzrlib.repofmt.pack_repo',
2989
'RepositoryFormatKnitPack4',
2991
format_registry.register_lazy(
2992
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
2993
'bzrlib.repofmt.pack_repo',
2994
'RepositoryFormatKnitPack5',
2996
format_registry.register_lazy(
2997
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
2998
'bzrlib.repofmt.pack_repo',
2999
'RepositoryFormatKnitPack5RichRoot',
3001
format_registry.register_lazy(
3002
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3003
'bzrlib.repofmt.pack_repo',
3004
'RepositoryFormatKnitPack5RichRootBroken',
3006
format_registry.register_lazy(
3007
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3008
'bzrlib.repofmt.pack_repo',
3009
'RepositoryFormatKnitPack6',
3011
format_registry.register_lazy(
3012
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3013
'bzrlib.repofmt.pack_repo',
3014
'RepositoryFormatKnitPack6RichRoot',
3017
# Development formats.
3018
# Obsolete but kept pending a CHK based subtree format.
3019
format_registry.register_lazy(
3020
("Bazaar development format 2 with subtree support "
3021
"(needs bzr.dev from before 1.8)\n"),
3022
'bzrlib.repofmt.pack_repo',
3023
'RepositoryFormatPackDevelopment2Subtree',
3026
# 1.14->1.16 go below here
3027
format_registry.register_lazy(
3028
'Bazaar development format - group compression and chk inventory'
3029
' (needs bzr.dev from 1.14)\n',
3030
'bzrlib.repofmt.groupcompress_repo',
3031
'RepositoryFormatCHK1',
3035
class InterRepository(InterObject):
3036
"""This class represents operations taking place between two repositories.
3038
Its instances have methods like copy_content and fetch, and contain
3039
references to the source and target repositories these operations can be
3042
Often we will provide convenience methods on 'repository' which carry out
3043
operations with another repository - they will always forward to
3044
InterRepository.get(other).method_name(parameters).
3047
_walk_to_common_revisions_batch_size = 50
3049
"""The available optimised InterRepository types."""
3052
def copy_content(self, revision_id=None):
3053
"""Make a complete copy of the content in self into destination.
3055
This is a destructive operation! Do not use it on existing
3058
:param revision_id: Only copy the content needed to construct
3059
revision_id and its parents.
3062
self.target.set_make_working_trees(self.source.make_working_trees())
3063
except NotImplementedError:
3065
self.target.fetch(self.source, revision_id=revision_id)
3068
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3070
"""Fetch the content required to construct revision_id.
3072
The content is copied from self.source to self.target.
3074
:param revision_id: if None all content is copied, if NULL_REVISION no
3076
:param pb: optional progress bar to use for progress reports. If not
3077
provided a default one will be created.
3080
from bzrlib.fetch import RepoFetcher
3081
f = RepoFetcher(to_repository=self.target,
3082
from_repository=self.source,
3083
last_revision=revision_id,
3084
fetch_spec=fetch_spec,
3085
pb=pb, find_ghosts=find_ghosts)
3087
def _walk_to_common_revisions(self, revision_ids):
3088
"""Walk out from revision_ids in source to revisions target has.
3090
:param revision_ids: The start point for the search.
3091
:return: A set of revision ids.
3093
target_graph = self.target.get_graph()
3094
revision_ids = frozenset(revision_ids)
3095
missing_revs = set()
3096
source_graph = self.source.get_graph()
3097
# ensure we don't pay silly lookup costs.
3098
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3099
null_set = frozenset([_mod_revision.NULL_REVISION])
3100
searcher_exhausted = False
3104
# Iterate the searcher until we have enough next_revs
3105
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3107
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3108
next_revs.update(next_revs_part)
3109
ghosts.update(ghosts_part)
3110
except StopIteration:
3111
searcher_exhausted = True
3113
# If there are ghosts in the source graph, and the caller asked for
3114
# them, make sure that they are present in the target.
3115
# We don't care about other ghosts as we can't fetch them and
3116
# haven't been asked to.
3117
ghosts_to_check = set(revision_ids.intersection(ghosts))
3118
revs_to_get = set(next_revs).union(ghosts_to_check)
3120
have_revs = set(target_graph.get_parent_map(revs_to_get))
3121
# we always have NULL_REVISION present.
3122
have_revs = have_revs.union(null_set)
3123
# Check if the target is missing any ghosts we need.
3124
ghosts_to_check.difference_update(have_revs)
3126
# One of the caller's revision_ids is a ghost in both the
3127
# source and the target.
3128
raise errors.NoSuchRevision(
3129
self.source, ghosts_to_check.pop())
3130
missing_revs.update(next_revs - have_revs)
3131
# Because we may have walked past the original stop point, make
3132
# sure everything is stopped
3133
stop_revs = searcher.find_seen_ancestors(have_revs)
3134
searcher.stop_searching_any(stop_revs)
3135
if searcher_exhausted:
3137
return searcher.get_result()
3140
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3141
"""Return the revision ids that source has that target does not.
3143
:param revision_id: only return revision ids included by this
3145
:param find_ghosts: If True find missing revisions in deep history
3146
rather than just finding the surface difference.
3147
:return: A bzrlib.graph.SearchResult.
3149
# stop searching at found target revisions.
3150
if not find_ghosts and revision_id is not None:
3151
return self._walk_to_common_revisions([revision_id])
3152
# generic, possibly worst case, slow code path.
3153
target_ids = set(self.target.all_revision_ids())
3154
if revision_id is not None:
3155
source_ids = self.source.get_ancestry(revision_id)
3156
if source_ids[0] is not None:
3157
raise AssertionError()
3160
source_ids = self.source.all_revision_ids()
3161
result_set = set(source_ids).difference(target_ids)
3162
return self.source.revision_ids_to_search_result(result_set)
3165
def _same_model(source, target):
3166
"""True if source and target have the same data representation.
3168
Note: this is always called on the base class; overriding it in a
3169
subclass will have no effect.
3172
InterRepository._assert_same_model(source, target)
3174
except errors.IncompatibleRepositories, e:
3178
def _assert_same_model(source, target):
3179
"""Raise an exception if two repositories do not use the same model.
3181
if source.supports_rich_root() != target.supports_rich_root():
3182
raise errors.IncompatibleRepositories(source, target,
3183
"different rich-root support")
3184
if source._serializer != target._serializer:
3185
raise errors.IncompatibleRepositories(source, target,
3186
"different serializers")
3189
class InterSameDataRepository(InterRepository):
3190
"""Code for converting between repositories that represent the same data.
3192
Data format and model must match for this to work.
3196
def _get_repo_format_to_test(self):
3197
"""Repository format for testing with.
3199
InterSameData can pull from subtree to subtree and from non-subtree to
3200
non-subtree, so we test this with the richest repository format.
3202
from bzrlib.repofmt import knitrepo
3203
return knitrepo.RepositoryFormatKnit3()
3206
def is_compatible(source, target):
3207
return InterRepository._same_model(source, target)
3210
class InterWeaveRepo(InterSameDataRepository):
3211
"""Optimised code paths between Weave based repositories.
3213
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3214
implemented lazy inter-object optimisation.
3218
def _get_repo_format_to_test(self):
3219
from bzrlib.repofmt import weaverepo
3220
return weaverepo.RepositoryFormat7()
3223
def is_compatible(source, target):
3224
"""Be compatible with known Weave formats.
3226
We don't test for the stores being of specific types because that
3227
could lead to confusing results, and there is no need to be
3230
from bzrlib.repofmt.weaverepo import (
3236
return (isinstance(source._format, (RepositoryFormat5,
3238
RepositoryFormat7)) and
3239
isinstance(target._format, (RepositoryFormat5,
3241
RepositoryFormat7)))
3242
except AttributeError:
3246
def copy_content(self, revision_id=None):
3247
"""See InterRepository.copy_content()."""
3248
# weave specific optimised path:
3250
self.target.set_make_working_trees(self.source.make_working_trees())
3251
except (errors.RepositoryUpgradeRequired, NotImplemented):
3253
# FIXME do not peek!
3254
if self.source._transport.listable():
3255
pb = ui.ui_factory.nested_progress_bar()
3257
self.target.texts.insert_record_stream(
3258
self.source.texts.get_record_stream(
3259
self.source.texts.keys(), 'topological', False))
3260
pb.update('copying inventory', 0, 1)
3261
self.target.inventories.insert_record_stream(
3262
self.source.inventories.get_record_stream(
3263
self.source.inventories.keys(), 'topological', False))
3264
self.target.signatures.insert_record_stream(
3265
self.source.signatures.get_record_stream(
3266
self.source.signatures.keys(),
3268
self.target.revisions.insert_record_stream(
3269
self.source.revisions.get_record_stream(
3270
self.source.revisions.keys(),
3271
'topological', True))
3275
self.target.fetch(self.source, revision_id=revision_id)
3278
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3279
"""See InterRepository.missing_revision_ids()."""
3280
# we want all revisions to satisfy revision_id in source.
3281
# but we don't want to stat every file here and there.
3282
# we want then, all revisions other needs to satisfy revision_id
3283
# checked, but not those that we have locally.
3284
# so the first thing is to get a subset of the revisions to
3285
# satisfy revision_id in source, and then eliminate those that
3286
# we do already have.
3287
# this is slow on high latency connection to self, but as this
3288
# disk format scales terribly for push anyway due to rewriting
3289
# inventory.weave, this is considered acceptable.
3291
if revision_id is not None:
3292
source_ids = self.source.get_ancestry(revision_id)
3293
if source_ids[0] is not None:
3294
raise AssertionError()
3297
source_ids = self.source._all_possible_ids()
3298
source_ids_set = set(source_ids)
3299
# source_ids is the worst possible case we may need to pull.
3300
# now we want to filter source_ids against what we actually
3301
# have in target, but don't try to check for existence where we know
3302
# we do not have a revision as that would be pointless.
3303
target_ids = set(self.target._all_possible_ids())
3304
possibly_present_revisions = target_ids.intersection(source_ids_set)
3305
actually_present_revisions = set(
3306
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3307
required_revisions = source_ids_set.difference(actually_present_revisions)
3308
if revision_id is not None:
3309
# we used get_ancestry to determine source_ids then we are assured all
3310
# revisions referenced are present as they are installed in topological order.
3311
# and the tip revision was validated by get_ancestry.
3312
result_set = required_revisions
3314
# if we just grabbed the possibly available ids, then
3315
# we only have an estimate of whats available and need to validate
3316
# that against the revision records.
3318
self.source._eliminate_revisions_not_present(required_revisions))
3319
return self.source.revision_ids_to_search_result(result_set)
3322
class InterKnitRepo(InterSameDataRepository):
3323
"""Optimised code paths between Knit based repositories."""
3326
def _get_repo_format_to_test(self):
3327
from bzrlib.repofmt import knitrepo
3328
return knitrepo.RepositoryFormatKnit1()
3331
def is_compatible(source, target):
3332
"""Be compatible with known Knit formats.
3334
We don't test for the stores being of specific types because that
3335
could lead to confusing results, and there is no need to be
3338
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3340
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3341
isinstance(target._format, RepositoryFormatKnit))
3342
except AttributeError:
3344
return are_knits and InterRepository._same_model(source, target)
3347
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3348
"""See InterRepository.missing_revision_ids()."""
3349
if revision_id is not None:
3350
source_ids = self.source.get_ancestry(revision_id)
3351
if source_ids[0] is not None:
3352
raise AssertionError()
3355
source_ids = self.source.all_revision_ids()
3356
source_ids_set = set(source_ids)
3357
# source_ids is the worst possible case we may need to pull.
3358
# now we want to filter source_ids against what we actually
3359
# have in target, but don't try to check for existence where we know
3360
# we do not have a revision as that would be pointless.
3361
target_ids = set(self.target.all_revision_ids())
3362
possibly_present_revisions = target_ids.intersection(source_ids_set)
3363
actually_present_revisions = set(
3364
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3365
required_revisions = source_ids_set.difference(actually_present_revisions)
3366
if revision_id is not None:
3367
# we used get_ancestry to determine source_ids then we are assured all
3368
# revisions referenced are present as they are installed in topological order.
3369
# and the tip revision was validated by get_ancestry.
3370
result_set = required_revisions
3372
# if we just grabbed the possibly available ids, then
3373
# we only have an estimate of whats available and need to validate
3374
# that against the revision records.
3376
self.source._eliminate_revisions_not_present(required_revisions))
3377
return self.source.revision_ids_to_search_result(result_set)
3380
class InterPackRepo(InterSameDataRepository):
3381
"""Optimised code paths between Pack based repositories."""
3384
def _get_repo_format_to_test(self):
3385
from bzrlib.repofmt import pack_repo
3386
return pack_repo.RepositoryFormatKnitPack6RichRoot()
3389
def is_compatible(source, target):
3390
"""Be compatible with known Pack formats.
3392
We don't test for the stores being of specific types because that
3393
could lead to confusing results, and there is no need to be
3396
InterPackRepo does not support CHK based repositories.
3398
from bzrlib.repofmt.pack_repo import RepositoryFormatPack
3399
from bzrlib.repofmt.groupcompress_repo import RepositoryFormatCHK1
3401
are_packs = (isinstance(source._format, RepositoryFormatPack) and
3402
isinstance(target._format, RepositoryFormatPack))
3403
not_packs = (isinstance(source._format, RepositoryFormatCHK1) or
3404
isinstance(target._format, RepositoryFormatCHK1))
3405
except AttributeError:
3407
if not_packs or not are_packs:
3409
return InterRepository._same_model(source, target)
3412
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3414
"""See InterRepository.fetch()."""
3415
if (len(self.source._fallback_repositories) > 0 or
3416
len(self.target._fallback_repositories) > 0):
3417
# The pack layer is not aware of fallback repositories, so when
3418
# fetching from a stacked repository or into a stacked repository
3419
# we use the generic fetch logic which uses the VersionedFiles
3420
# attributes on repository.
3421
from bzrlib.fetch import RepoFetcher
3422
fetcher = RepoFetcher(self.target, self.source, revision_id,
3423
pb, find_ghosts, fetch_spec=fetch_spec)
3424
if fetch_spec is not None:
3425
if len(list(fetch_spec.heads)) != 1:
3426
raise AssertionError(
3427
"InterPackRepo.fetch doesn't support "
3428
"fetching multiple heads yet.")
3429
revision_id = list(fetch_spec.heads)[0]
3431
if revision_id is None:
3433
# everything to do - use pack logic
3434
# to fetch from all packs to one without
3435
# inventory parsing etc, IFF nothing to be copied is in the target.
3437
source_revision_ids = frozenset(self.source.all_revision_ids())
3438
revision_ids = source_revision_ids - \
3439
frozenset(self.target.get_parent_map(source_revision_ids))
3440
revision_keys = [(revid,) for revid in revision_ids]
3441
index = self.target._pack_collection.revision_index.combined_index
3442
present_revision_ids = set(item[1][0] for item in
3443
index.iter_entries(revision_keys))
3444
revision_ids = set(revision_ids) - present_revision_ids
3445
# implementing the TODO will involve:
3446
# - detecting when all of a pack is selected
3447
# - avoiding as much as possible pre-selection, so the
3448
# more-core routines such as create_pack_from_packs can filter in
3449
# a just-in-time fashion. (though having a HEADS list on a
3450
# repository might make this a lot easier, because we could
3451
# sensibly detect 'new revisions' without doing a full index scan.
3452
elif _mod_revision.is_null(revision_id):
3456
revision_ids = self.search_missing_revision_ids(revision_id,
3457
find_ghosts=find_ghosts).get_keys()
3458
if len(revision_ids) == 0:
3460
return self._pack(self.source, self.target, revision_ids)
3462
def _pack(self, source, target, revision_ids):
3463
from bzrlib.repofmt.pack_repo import Packer
3464
packs = source._pack_collection.all_packs()
3465
pack = Packer(self.target._pack_collection, packs, '.fetch',
3466
revision_ids).pack()
3467
if pack is not None:
3468
self.target._pack_collection._save_pack_names()
3469
copied_revs = pack.get_revision_count()
3470
# Trigger an autopack. This may duplicate effort as we've just done
3471
# a pack creation, but for now it is simpler to think about as
3472
# 'upload data, then repack if needed'.
3473
self.target._pack_collection.autopack()
3474
return (copied_revs, [])
3479
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3480
"""See InterRepository.missing_revision_ids().
3482
:param find_ghosts: Find ghosts throughout the ancestry of
3485
if not find_ghosts and revision_id is not None:
3486
return self._walk_to_common_revisions([revision_id])
3487
elif revision_id is not None:
3488
# Find ghosts: search for revisions pointing from one repository to
3489
# the other, and vice versa, anywhere in the history of revision_id.
3490
graph = self.target.get_graph(other_repository=self.source)
3491
searcher = graph._make_breadth_first_searcher([revision_id])
3495
next_revs, ghosts = searcher.next_with_ghosts()
3496
except StopIteration:
3498
if revision_id in ghosts:
3499
raise errors.NoSuchRevision(self.source, revision_id)
3500
found_ids.update(next_revs)
3501
found_ids.update(ghosts)
3502
found_ids = frozenset(found_ids)
3503
# Double query here: should be able to avoid this by changing the
3504
# graph api further.
3505
result_set = found_ids - frozenset(
3506
self.target.get_parent_map(found_ids))
3508
source_ids = self.source.all_revision_ids()
3509
# source_ids is the worst possible case we may need to pull.
3510
# now we want to filter source_ids against what we actually
3511
# have in target, but don't try to check for existence where we know
3512
# we do not have a revision as that would be pointless.
3513
target_ids = set(self.target.all_revision_ids())
3514
result_set = set(source_ids).difference(target_ids)
3515
return self.source.revision_ids_to_search_result(result_set)
3518
class InterDifferingSerializer(InterRepository):
3521
def _get_repo_format_to_test(self):
3525
def is_compatible(source, target):
3526
"""Be compatible with Knit2 source and Knit3 target"""
3527
# This is redundant with format.check_conversion_target(), however that
3528
# raises an exception, and we just want to say "False" as in we won't
3529
# support converting between these formats.
3530
if source.supports_rich_root() and not target.supports_rich_root():
3532
if (source._format.supports_tree_reference
3533
and not target._format.supports_tree_reference):
3537
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3538
"""Get the best delta and base for this revision.
3540
:return: (basis_id, delta)
3542
possible_trees = [(parent_id, cache[parent_id])
3543
for parent_id in parent_ids
3544
if parent_id in cache]
3545
if len(possible_trees) == 0:
3546
# There either aren't any parents, or the parents aren't in the
3547
# cache, so just use the last converted tree
3548
possible_trees.append((basis_id, cache[basis_id]))
3550
for basis_id, basis_tree in possible_trees:
3551
delta = tree.inventory._make_delta(basis_tree.inventory)
3552
deltas.append((len(delta), basis_id, delta))
3554
return deltas[0][1:]
3556
def _get_parent_keys(self, root_key, parent_map):
3557
"""Get the parent keys for a given root id."""
3558
root_id, rev_id = root_key
3559
# Include direct parents of the revision, but only if they used
3562
for parent_id in parent_map[rev_id]:
3563
if parent_id == _mod_revision.NULL_REVISION:
3565
if parent_id not in self._revision_id_to_root_id:
3566
# We probably didn't read this revision, go spend the
3567
# extra effort to actually check
3569
tree = self.source.revision_tree(parent_id)
3570
except errors.NoSuchRevision:
3571
# Ghost, fill out _revision_id_to_root_id in case we
3572
# encounter this again.
3573
# But set parent_root_id to None since we don't really know
3574
parent_root_id = None
3576
parent_root_id = tree.get_root_id()
3577
self._revision_id_to_root_id[parent_id] = None
3579
parent_root_id = self._revision_id_to_root_id[parent_id]
3580
if root_id == parent_root_id or parent_root_id is None:
3581
parent_keys.append((root_id, parent_id))
3582
return tuple(parent_keys)
3584
def _new_root_data_stream(self, root_keys_to_create, parent_map):
3585
for root_key in root_keys_to_create:
3586
parent_keys = self._get_parent_keys(root_key, parent_map)
3587
yield versionedfile.FulltextContentFactory(root_key,
3588
parent_keys, None, '')
3590
def _fetch_batch(self, revision_ids, basis_id, cache):
3591
"""Fetch across a few revisions.
3593
:param revision_ids: The revisions to copy
3594
:param basis_id: The revision_id of a tree that must be in cache, used
3595
as a basis for delta when no other base is available
3596
:param cache: A cache of RevisionTrees that we can use.
3597
:return: The revision_id of the last converted tree. The RevisionTree
3598
for it will be in cache
3600
# Walk though all revisions; get inventory deltas, copy referenced
3601
# texts that delta references, insert the delta, revision and
3603
root_keys_to_create = set()
3606
pending_revisions = []
3607
parent_map = self.source.get_parent_map(revision_ids)
3608
for tree in self.source.revision_trees(revision_ids):
3609
current_revision_id = tree.get_revision_id()
3610
parent_ids = parent_map.get(current_revision_id, ())
3611
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3613
if self._converting_to_rich_root:
3614
self._revision_id_to_root_id[current_revision_id] = \
3616
# Find text entries that need to be copied
3617
for old_path, new_path, file_id, entry in delta:
3618
if new_path is not None:
3621
if not self.target.supports_rich_root():
3622
# The target doesn't support rich root, so we don't
3625
if self._converting_to_rich_root:
3626
# This can't be copied normally, we have to insert
3628
root_keys_to_create.add((file_id, entry.revision))
3630
text_keys.add((file_id, entry.revision))
3631
revision = self.source.get_revision(current_revision_id)
3632
pending_deltas.append((basis_id, delta,
3633
current_revision_id, revision.parent_ids))
3634
pending_revisions.append(revision)
3635
cache[current_revision_id] = tree
3636
basis_id = current_revision_id
3638
from_texts = self.source.texts
3639
to_texts = self.target.texts
3640
if root_keys_to_create:
3641
root_stream = self._new_root_data_stream(root_keys_to_create,
3643
to_texts.insert_record_stream(root_stream)
3644
to_texts.insert_record_stream(from_texts.get_record_stream(
3645
text_keys, self.target._format._fetch_order,
3646
not self.target._format._fetch_uses_deltas))
3647
# insert inventory deltas
3648
for delta in pending_deltas:
3649
self.target.add_inventory_by_delta(*delta)
3650
if self.target._fallback_repositories:
3651
# Make sure this stacked repository has all the parent inventories
3652
# for the new revisions that we are about to insert. We do this
3653
# before adding the revisions so that no revision is added until
3654
# all the inventories it may depend on are added.
3656
revision_ids = set()
3657
for revision in pending_revisions:
3658
revision_ids.add(revision.revision_id)
3659
parent_ids.update(revision.parent_ids)
3660
parent_ids.difference_update(revision_ids)
3661
parent_ids.discard(_mod_revision.NULL_REVISION)
3662
parent_map = self.source.get_parent_map(parent_ids)
3663
for parent_tree in self.source.revision_trees(parent_ids):
3664
basis_id, delta = self._get_delta_for_revision(tree, parent_ids, basis_id, cache)
3665
current_revision_id = parent_tree.get_revision_id()
3666
parents_parents = parent_map[current_revision_id]
3667
self.target.add_inventory_by_delta(
3668
basis_id, delta, current_revision_id, parents_parents)
3669
# insert signatures and revisions
3670
for revision in pending_revisions:
3672
signature = self.source.get_signature_text(
3673
revision.revision_id)
3674
self.target.add_signature_text(revision.revision_id,
3676
except errors.NoSuchRevision:
3678
self.target.add_revision(revision.revision_id, revision)
3681
def _fetch_all_revisions(self, revision_ids, pb):
3682
"""Fetch everything for the list of revisions.
3684
:param revision_ids: The list of revisions to fetch. Must be in
3686
:param pb: A ProgressBar
3689
basis_id, basis_tree = self._get_basis(revision_ids[0])
3691
cache = lru_cache.LRUCache(100)
3692
cache[basis_id] = basis_tree
3693
del basis_tree # We don't want to hang on to it here
3694
for offset in range(0, len(revision_ids), batch_size):
3695
self.target.start_write_group()
3697
pb.update('Transferring revisions', offset,
3699
batch = revision_ids[offset:offset+batch_size]
3700
basis_id = self._fetch_batch(batch, basis_id, cache)
3702
self.target.abort_write_group()
3705
self.target.commit_write_group()
3706
pb.update('Transferring revisions', len(revision_ids),
3710
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3712
"""See InterRepository.fetch()."""
3713
if fetch_spec is not None:
3714
raise AssertionError("Not implemented yet...")
3715
if (not self.source.supports_rich_root()
3716
and self.target.supports_rich_root()):
3717
self._converting_to_rich_root = True
3718
self._revision_id_to_root_id = {}
3720
self._converting_to_rich_root = False
3721
revision_ids = self.target.search_missing_revision_ids(self.source,
3722
revision_id, find_ghosts=find_ghosts).get_keys()
3723
if not revision_ids:
3725
revision_ids = tsort.topo_sort(
3726
self.source.get_graph().get_parent_map(revision_ids))
3727
if not revision_ids:
3729
# Walk though all revisions; get inventory deltas, copy referenced
3730
# texts that delta references, insert the delta, revision and
3732
first_rev = self.source.get_revision(revision_ids[0])
3734
my_pb = ui.ui_factory.nested_progress_bar()
3737
symbol_versioning.warn(
3738
symbol_versioning.deprecated_in((1, 14, 0))
3739
% "pb parameter to fetch()")
3742
self._fetch_all_revisions(revision_ids, pb)
3744
if my_pb is not None:
3746
return len(revision_ids), 0
3748
def _get_basis(self, first_revision_id):
3749
"""Get a revision and tree which exists in the target.
3751
This assumes that first_revision_id is selected for transmission
3752
because all other ancestors are already present. If we can't find an
3753
ancestor we fall back to NULL_REVISION since we know that is safe.
3755
:return: (basis_id, basis_tree)
3757
first_rev = self.source.get_revision(first_revision_id)
3759
basis_id = first_rev.parent_ids[0]
3760
# only valid as a basis if the target has it
3761
self.target.get_revision(basis_id)
3762
# Try to get a basis tree - if its a ghost it will hit the
3763
# NoSuchRevision case.
3764
basis_tree = self.source.revision_tree(basis_id)
3765
except (IndexError, errors.NoSuchRevision):
3766
basis_id = _mod_revision.NULL_REVISION
3767
basis_tree = self.source.revision_tree(basis_id)
3768
return basis_id, basis_tree
3771
InterRepository.register_optimiser(InterDifferingSerializer)
3772
InterRepository.register_optimiser(InterSameDataRepository)
3773
InterRepository.register_optimiser(InterWeaveRepo)
3774
InterRepository.register_optimiser(InterKnitRepo)
3775
InterRepository.register_optimiser(InterPackRepo)
3778
class CopyConverter(object):
3779
"""A repository conversion tool which just performs a copy of the content.
3781
This is slow but quite reliable.
3784
def __init__(self, target_format):
3785
"""Create a CopyConverter.
3787
:param target_format: The format the resulting repository should be.
3789
self.target_format = target_format
3791
def convert(self, repo, pb):
3792
"""Perform the conversion of to_convert, giving feedback via pb.
3794
:param to_convert: The disk object to convert.
3795
:param pb: a progress bar to use for progress information.
3800
# this is only useful with metadir layouts - separated repo content.
3801
# trigger an assertion if not such
3802
repo._format.get_format_string()
3803
self.repo_dir = repo.bzrdir
3804
self.step('Moving repository to repository.backup')
3805
self.repo_dir.transport.move('repository', 'repository.backup')
3806
backup_transport = self.repo_dir.transport.clone('repository.backup')
3807
repo._format.check_conversion_target(self.target_format)
3808
self.source_repo = repo._format.open(self.repo_dir,
3810
_override_transport=backup_transport)
3811
self.step('Creating new repository')
3812
converted = self.target_format.initialize(self.repo_dir,
3813
self.source_repo.is_shared())
3814
converted.lock_write()
3816
self.step('Copying content into repository.')
3817
self.source_repo.copy_content_into(converted)
3820
self.step('Deleting old repository content.')
3821
self.repo_dir.transport.delete_tree('repository.backup')
3822
self.pb.note('repository converted')
3824
def step(self, message):
3825
"""Update the pb by a step."""
3827
self.pb.update(message, self.count, self.total)
3839
def _unescaper(match, _map=_unescape_map):
3840
code = match.group(1)
3844
if not code.startswith('#'):
3846
return unichr(int(code[1:])).encode('utf8')
3852
def _unescape_xml(data):
3853
"""Unescape predefined XML entities in a string of data."""
3855
if _unescape_re is None:
3856
_unescape_re = re.compile('\&([^;]*);')
3857
return _unescape_re.sub(_unescaper, data)
3860
class _VersionedFileChecker(object):
3862
def __init__(self, repository, text_key_references=None):
3863
self.repository = repository
3864
self.text_index = self.repository._generate_text_key_index(
3865
text_key_references=text_key_references)
3867
def calculate_file_version_parents(self, text_key):
3868
"""Calculate the correct parents for a file version according to
3871
parent_keys = self.text_index[text_key]
3872
if parent_keys == [_mod_revision.NULL_REVISION]:
3874
return tuple(parent_keys)
3876
def check_file_version_parents(self, texts, progress_bar=None):
3877
"""Check the parents stored in a versioned file are correct.
3879
It also detects file versions that are not referenced by their
3880
corresponding revision's inventory.
3882
:returns: A tuple of (wrong_parents, dangling_file_versions).
3883
wrong_parents is a dict mapping {revision_id: (stored_parents,
3884
correct_parents)} for each revision_id where the stored parents
3885
are not correct. dangling_file_versions is a set of (file_id,
3886
revision_id) tuples for versions that are present in this versioned
3887
file, but not used by the corresponding inventory.
3890
self.file_ids = set([file_id for file_id, _ in
3891
self.text_index.iterkeys()])
3892
# text keys is now grouped by file_id
3893
n_weaves = len(self.file_ids)
3894
files_in_revisions = {}
3895
revisions_of_files = {}
3896
n_versions = len(self.text_index)
3897
progress_bar.update('loading text store', 0, n_versions)
3898
parent_map = self.repository.texts.get_parent_map(self.text_index)
3899
# On unlistable transports this could well be empty/error...
3900
text_keys = self.repository.texts.keys()
3901
unused_keys = frozenset(text_keys) - set(self.text_index)
3902
for num, key in enumerate(self.text_index.iterkeys()):
3903
if progress_bar is not None:
3904
progress_bar.update('checking text graph', num, n_versions)
3905
correct_parents = self.calculate_file_version_parents(key)
3907
knit_parents = parent_map[key]
3908
except errors.RevisionNotPresent:
3911
if correct_parents != knit_parents:
3912
wrong_parents[key] = (knit_parents, correct_parents)
3913
return wrong_parents, unused_keys
3916
def _old_get_graph(repository, revision_id):
3917
"""DO NOT USE. That is all. I'm serious."""
3918
graph = repository.get_graph()
3919
revision_graph = dict(((key, value) for key, value in
3920
graph.iter_ancestry([revision_id]) if value is not None))
3921
return _strip_NULL_ghosts(revision_graph)
3924
def _strip_NULL_ghosts(revision_graph):
3925
"""Also don't use this. more compatibility code for unmigrated clients."""
3926
# Filter ghosts, and null:
3927
if _mod_revision.NULL_REVISION in revision_graph:
3928
del revision_graph[_mod_revision.NULL_REVISION]
3929
for key, parents in revision_graph.items():
3930
revision_graph[key] = tuple(parent for parent in parents if parent
3932
return revision_graph
3935
class StreamSink(object):
3936
"""An object that can insert a stream into a repository.
3938
This interface handles the complexity of reserialising inventories and
3939
revisions from different formats, and allows unidirectional insertion into
3940
stacked repositories without looking for the missing basis parents
3944
def __init__(self, target_repo):
3945
self.target_repo = target_repo
3947
def insert_stream(self, stream, src_format, resume_tokens):
3948
"""Insert a stream's content into the target repository.
3950
:param src_format: a bzr repository format.
3952
:return: a list of resume tokens and an iterable of keys additional
3953
items required before the insertion can be completed.
3955
self.target_repo.lock_write()
3958
self.target_repo.resume_write_group(resume_tokens)
3960
self.target_repo.start_write_group()
3962
# locked_insert_stream performs a commit|suspend.
3963
return self._locked_insert_stream(stream, src_format)
3965
self.target_repo.abort_write_group(suppress_errors=True)
3968
self.target_repo.unlock()
3970
def _locked_insert_stream(self, stream, src_format):
3971
to_serializer = self.target_repo._format._serializer
3972
src_serializer = src_format._serializer
3973
if to_serializer == src_serializer:
3974
# If serializers match and the target is a pack repository, set the
3975
# write cache size on the new pack. This avoids poor performance
3976
# on transports where append is unbuffered (such as
3977
# RemoteTransport). This is safe to do because nothing should read
3978
# back from the target repository while a stream with matching
3979
# serialization is being inserted.
3980
# The exception is that a delta record from the source that should
3981
# be a fulltext may need to be expanded by the target (see
3982
# test_fetch_revisions_with_deltas_into_pack); but we take care to
3983
# explicitly flush any buffered writes first in that rare case.
3985
new_pack = self.target_repo._pack_collection._new_pack
3986
except AttributeError:
3987
# Not a pack repository
3990
new_pack.set_write_cache_size(1024*1024)
3991
for substream_type, substream in stream:
3992
if substream_type == 'texts':
3993
self.target_repo.texts.insert_record_stream(substream)
3994
elif substream_type == 'inventories':
3995
if src_serializer == to_serializer:
3996
self.target_repo.inventories.insert_record_stream(
3999
self._extract_and_insert_inventories(
4000
substream, src_serializer)
4001
elif substream_type == 'chk_bytes':
4002
# XXX: This doesn't support conversions, as it assumes the
4003
# conversion was done in the fetch code.
4004
self.target_repo.chk_bytes.insert_record_stream(substream)
4005
elif substream_type == 'revisions':
4006
# This may fallback to extract-and-insert more often than
4007
# required if the serializers are different only in terms of
4009
if src_serializer == to_serializer:
4010
self.target_repo.revisions.insert_record_stream(
4013
self._extract_and_insert_revisions(substream,
4015
elif substream_type == 'signatures':
4016
self.target_repo.signatures.insert_record_stream(substream)
4018
raise AssertionError('kaboom! %s' % (substream_type,))
4019
# Find all the new revisions (including ones from resume_tokens)
4020
missing_keys = self.target_repo.get_missing_parent_inventories()
4022
for prefix, versioned_file in (
4023
('texts', self.target_repo.texts),
4024
('inventories', self.target_repo.inventories),
4025
('revisions', self.target_repo.revisions),
4026
('signatures', self.target_repo.signatures),
4028
missing_keys.update((prefix,) + key for key in
4029
versioned_file.get_missing_compression_parent_keys())
4030
except NotImplementedError:
4031
# cannot even attempt suspending, and missing would have failed
4032
# during stream insertion.
4033
missing_keys = set()
4036
# suspend the write group and tell the caller what we is
4037
# missing. We know we can suspend or else we would not have
4038
# entered this code path. (All repositories that can handle
4039
# missing keys can handle suspending a write group).
4040
write_group_tokens = self.target_repo.suspend_write_group()
4041
return write_group_tokens, missing_keys
4042
self.target_repo.commit_write_group()
4045
def _extract_and_insert_inventories(self, substream, serializer):
4046
"""Generate a new inventory versionedfile in target, converting data.
4048
The inventory is retrieved from the source, (deserializing it), and
4049
stored in the target (reserializing it in a different format).
4051
for record in substream:
4052
bytes = record.get_bytes_as('fulltext')
4053
revision_id = record.key[0]
4054
inv = serializer.read_inventory_from_string(bytes, revision_id)
4055
parents = [key[0] for key in record.parents]
4056
self.target_repo.add_inventory(revision_id, inv, parents)
4058
def _extract_and_insert_revisions(self, substream, serializer):
4059
for record in substream:
4060
bytes = record.get_bytes_as('fulltext')
4061
revision_id = record.key[0]
4062
rev = serializer.read_revision_from_string(bytes)
4063
if rev.revision_id != revision_id:
4064
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4065
self.target_repo.add_revision(revision_id, rev)
4068
if self.target_repo._format._fetch_reconcile:
4069
self.target_repo.reconcile()
4072
class StreamSource(object):
4073
"""A source of a stream for fetching between repositories."""
4075
def __init__(self, from_repository, to_format):
4076
"""Create a StreamSource streaming from from_repository."""
4077
self.from_repository = from_repository
4078
self.to_format = to_format
4080
def delta_on_metadata(self):
4081
"""Return True if delta's are permitted on metadata streams.
4083
That is on revisions and signatures.
4085
src_serializer = self.from_repository._format._serializer
4086
target_serializer = self.to_format._serializer
4087
return (self.to_format._fetch_uses_deltas and
4088
src_serializer == target_serializer)
4090
def _fetch_revision_texts(self, revs):
4091
# fetch signatures first and then the revision texts
4092
# may need to be a InterRevisionStore call here.
4093
from_sf = self.from_repository.signatures
4094
# A missing signature is just skipped.
4095
keys = [(rev_id,) for rev_id in revs]
4096
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4098
self.to_format._fetch_order,
4099
not self.to_format._fetch_uses_deltas))
4100
# If a revision has a delta, this is actually expanded inside the
4101
# insert_record_stream code now, which is an alternate fix for
4103
from_rf = self.from_repository.revisions
4104
revisions = from_rf.get_record_stream(
4106
self.to_format._fetch_order,
4107
not self.delta_on_metadata())
4108
return [('signatures', signatures), ('revisions', revisions)]
4110
def _generate_root_texts(self, revs):
4111
"""This will be called by __fetch between fetching weave texts and
4112
fetching the inventory weave.
4114
Subclasses should override this if they need to generate root texts
4115
after fetching weave texts.
4117
if self._rich_root_upgrade():
4119
return bzrlib.fetch.Inter1and2Helper(
4120
self.from_repository).generate_root_texts(revs)
4124
def get_stream(self, search):
4126
revs = search.get_keys()
4127
graph = self.from_repository.get_graph()
4128
revs = list(graph.iter_topo_order(revs))
4129
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4131
for knit_kind, file_id, revisions in data_to_fetch:
4132
if knit_kind != phase:
4134
# Make a new progress bar for this phase
4135
if knit_kind == "file":
4136
# Accumulate file texts
4137
text_keys.extend([(file_id, revision) for revision in
4139
elif knit_kind == "inventory":
4140
# Now copy the file texts.
4141
from_texts = self.from_repository.texts
4142
yield ('texts', from_texts.get_record_stream(
4143
text_keys, self.to_format._fetch_order,
4144
not self.to_format._fetch_uses_deltas))
4145
# Cause an error if a text occurs after we have done the
4148
# Before we process the inventory we generate the root
4149
# texts (if necessary) so that the inventories references
4151
for _ in self._generate_root_texts(revs):
4153
# NB: This currently reopens the inventory weave in source;
4154
# using a single stream interface instead would avoid this.
4155
from_weave = self.from_repository.inventories
4156
# we fetch only the referenced inventories because we do not
4157
# know for unselected inventories whether all their required
4158
# texts are present in the other repository - it could be
4160
for info in self._get_inventory_stream(revs):
4162
elif knit_kind == "signatures":
4163
# Nothing to do here; this will be taken care of when
4164
# _fetch_revision_texts happens.
4166
elif knit_kind == "revisions":
4167
for record in self._fetch_revision_texts(revs):
4170
raise AssertionError("Unknown knit kind %r" % knit_kind)
4172
def get_stream_for_missing_keys(self, missing_keys):
4173
# missing keys can only occur when we are byte copying and not
4174
# translating (because translation means we don't send
4175
# unreconstructable deltas ever).
4177
keys['texts'] = set()
4178
keys['revisions'] = set()
4179
keys['inventories'] = set()
4180
keys['signatures'] = set()
4181
for key in missing_keys:
4182
keys[key[0]].add(key[1:])
4183
if len(keys['revisions']):
4184
# If we allowed copying revisions at this point, we could end up
4185
# copying a revision without copying its required texts: a
4186
# violation of the requirements for repository integrity.
4187
raise AssertionError(
4188
'cannot copy revisions to fill in missing deltas %s' % (
4189
keys['revisions'],))
4190
for substream_kind, keys in keys.iteritems():
4191
vf = getattr(self.from_repository, substream_kind)
4192
# Ask for full texts always so that we don't need more round trips
4193
# after this stream.
4194
stream = vf.get_record_stream(keys,
4195
self.to_format._fetch_order, True)
4196
yield substream_kind, stream
4198
def inventory_fetch_order(self):
4199
if self._rich_root_upgrade():
4200
return 'topological'
4202
return self.to_format._fetch_order
4204
def _rich_root_upgrade(self):
4205
return (not self.from_repository._format.rich_root_data and
4206
self.to_format.rich_root_data)
4208
def _get_inventory_stream(self, revision_ids):
4209
from_format = self.from_repository._format
4210
if (from_format.supports_chks and self.to_format.supports_chks
4211
and (from_format._serializer == self.to_format._serializer)):
4212
# Both sides support chks, and they use the same serializer, so it
4213
# is safe to transmit the chk pages and inventory pages across
4215
return self._get_chk_inventory_stream(revision_ids)
4216
elif (not from_format.supports_chks):
4217
# Source repository doesn't support chks. So we can transmit the
4218
# inventories 'as-is' and either they are just accepted on the
4219
# target, or the Sink will properly convert it.
4220
return self._get_simple_inventory_stream(revision_ids)
4222
# XXX: Hack to make not-chk->chk fetch: copy the inventories as
4223
# inventories. Note that this should probably be done somehow
4224
# as part of bzrlib.repository.StreamSink. Except JAM couldn't
4225
# figure out how a non-chk repository could possibly handle
4226
# deserializing an inventory stream from a chk repo, as it
4227
# doesn't have a way to understand individual pages.
4228
return self._get_convertable_inventory_stream(revision_ids)
4230
def _get_simple_inventory_stream(self, revision_ids):
4231
from_weave = self.from_repository.inventories
4232
yield ('inventories', from_weave.get_record_stream(
4233
[(rev_id,) for rev_id in revision_ids],
4234
self.inventory_fetch_order(),
4235
not self.delta_on_metadata()))
4237
def _get_chk_inventory_stream(self, revision_ids):
4238
"""Fetch the inventory texts, along with the associated chk maps."""
4239
# We want an inventory outside of the search set, so that we can filter
4240
# out uninteresting chk pages. For now we use
4241
# _find_revision_outside_set, but if we had a Search with cut_revs, we
4242
# could use that instead.
4243
start_rev_id = self.from_repository._find_revision_outside_set(
4245
start_rev_key = (start_rev_id,)
4246
inv_keys_to_fetch = [(rev_id,) for rev_id in revision_ids]
4247
if start_rev_id != _mod_revision.NULL_REVISION:
4248
inv_keys_to_fetch.append((start_rev_id,))
4249
# Any repo that supports chk_bytes must also support out-of-order
4250
# insertion. At least, that is how we expect it to work
4251
# We use get_record_stream instead of iter_inventories because we want
4252
# to be able to insert the stream as well. We could instead fetch
4253
# allowing deltas, and then iter_inventories, but we don't know whether
4254
# source or target is more 'local' anway.
4255
inv_stream = self.from_repository.inventories.get_record_stream(
4256
inv_keys_to_fetch, 'unordered',
4257
True) # We need them as full-texts so we can find their references
4258
uninteresting_chk_roots = set()
4259
interesting_chk_roots = set()
4260
def filter_inv_stream(inv_stream):
4261
for idx, record in enumerate(inv_stream):
4262
### child_pb.update('fetch inv', idx, len(inv_keys_to_fetch))
4263
bytes = record.get_bytes_as('fulltext')
4264
chk_inv = inventory.CHKInventory.deserialise(
4265
self.from_repository.chk_bytes, bytes, record.key)
4266
if record.key == start_rev_key:
4267
uninteresting_chk_roots.add(chk_inv.id_to_entry.key())
4268
p_id_map = chk_inv.parent_id_basename_to_file_id
4269
if p_id_map is not None:
4270
uninteresting_chk_roots.add(p_id_map.key())
4273
interesting_chk_roots.add(chk_inv.id_to_entry.key())
4274
p_id_map = chk_inv.parent_id_basename_to_file_id
4275
if p_id_map is not None:
4276
interesting_chk_roots.add(p_id_map.key())
4277
### pb.update('fetch inventory', 0, 2)
4278
yield ('inventories', filter_inv_stream(inv_stream))
4279
# Now that we have worked out all of the interesting root nodes, grab
4280
# all of the interesting pages and insert them
4281
### pb.update('fetch inventory', 1, 2)
4282
interesting = chk_map.iter_interesting_nodes(
4283
self.from_repository.chk_bytes, interesting_chk_roots,
4284
uninteresting_chk_roots)
4285
def to_stream_adapter():
4286
"""Adapt the iter_interesting_nodes result to a single stream.
4288
iter_interesting_nodes returns records as it processes them, along
4289
with keys. However, we only want to return the records themselves.
4291
for record, items in interesting:
4292
if record is not None:
4294
# XXX: We could instead call get_record_stream(records.keys())
4295
# ATM, this will always insert the records as fulltexts, and
4296
# requires that you can hang on to records once you have gone
4297
# on to the next one. Further, it causes the target to
4298
# recompress the data. Testing shows it to be faster than
4299
# requesting the records again, though.
4300
yield ('chk_bytes', to_stream_adapter())
4301
### pb.update('fetch inventory', 2, 2)
4303
def _get_convertable_inventory_stream(self, revision_ids):
4304
# XXX: One of source or target is using chks, and they don't have
4305
# compatible serializations. The StreamSink code expects to be
4306
# able to convert on the target, so we need to put
4307
# bytes-on-the-wire that can be converted
4308
yield ('inventories', self._stream_invs_as_fulltexts(revision_ids))
4310
def _stream_invs_as_fulltexts(self, revision_ids):
4311
from_repo = self.from_repository
4312
from_serializer = from_repo._format._serializer
4313
revision_keys = [(rev_id,) for rev_id in revision_ids]
4314
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4315
for inv in self.from_repository.iter_inventories(revision_ids):
4316
# XXX: This is a bit hackish, but it works. Basically,
4317
# CHKSerializer 'accidentally' supports
4318
# read/write_inventory_to_string, even though that is never
4319
# the format that is stored on disk. It *does* give us a
4320
# single string representation for an inventory, so live with
4322
# This would be far better if we had a 'serialized inventory
4323
# delta' form. Then we could use 'inventory._make_delta', and
4324
# transmit that. This would both be faster to generate, and
4325
# result in fewer bytes-on-the-wire.
4326
as_bytes = from_serializer.write_inventory_to_string(inv)
4327
key = (inv.revision_id,)
4328
parent_keys = parent_map.get(key, ())
4329
yield versionedfile.FulltextContentFactory(
4330
key, parent_keys, None, as_bytes)
added
removed
bzrlib/repository.py
