1
# Copyright (C) 2005-2010 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(), """
42
revision as _mod_revision,
50
from bzrlib.bundle import serializer
51
from bzrlib.revisiontree import RevisionTree
52
from bzrlib.store.versioned import VersionedFileStore
53
from bzrlib.testament import Testament
56
from bzrlib.decorators import needs_read_lock, needs_write_lock, only_raises
57
from bzrlib.inter import InterObject
58
from bzrlib.inventory import (
64
from bzrlib.lock import _RelockDebugMixin
65
from bzrlib import registry
66
from bzrlib.trace import (
67
log_exception_quietly, note, mutter, mutter_callsite, warning)
70
# Old formats display a warning, but only once
71
_deprecation_warning_done = False
74
class CommitBuilder(object):
75
"""Provides an interface to build up a commit.
77
This allows describing a tree to be committed without needing to
78
know the internals of the format of the repository.
81
# all clients should supply tree roots.
82
record_root_entry = True
83
# the default CommitBuilder does not manage trees whose root is versioned.
84
_versioned_root = False
86
def __init__(self, repository, parents, config, timestamp=None,
87
timezone=None, committer=None, revprops=None,
89
"""Initiate a CommitBuilder.
91
:param repository: Repository to commit to.
92
:param parents: Revision ids of the parents of the new revision.
93
:param config: Configuration to use.
94
:param timestamp: Optional timestamp recorded for commit.
95
:param timezone: Optional timezone for timestamp.
96
:param committer: Optional committer to set for commit.
97
:param revprops: Optional dictionary of revision properties.
98
:param revision_id: Optional revision id.
100
self._config = config
102
if committer is None:
103
self._committer = self._config.username()
105
self._committer = committer
107
self.new_inventory = Inventory(None)
108
self._new_revision_id = revision_id
109
self.parents = parents
110
self.repository = repository
113
if revprops is not None:
114
self._validate_revprops(revprops)
115
self._revprops.update(revprops)
117
if timestamp is None:
118
timestamp = time.time()
119
# Restrict resolution to 1ms
120
self._timestamp = round(timestamp, 3)
123
self._timezone = osutils.local_time_offset()
125
self._timezone = int(timezone)
127
self._generate_revision_if_needed()
128
self.__heads = graph.HeadsCache(repository.get_graph()).heads
129
self._basis_delta = []
130
# API compatibility, older code that used CommitBuilder did not call
131
# .record_delete(), which means the delta that is computed would not be
132
# valid. Callers that will call record_delete() should call
133
# .will_record_deletes() to indicate that.
134
self._recording_deletes = False
135
# memo'd check for no-op commits.
136
self._any_changes = False
138
def any_changes(self):
139
"""Return True if any entries were changed.
141
This includes merge-only changes. It is the core for the --unchanged
144
:return: True if any changes have occured.
146
return self._any_changes
148
def _validate_unicode_text(self, text, context):
149
"""Verify things like commit messages don't have bogus characters."""
151
raise ValueError('Invalid value for %s: %r' % (context, text))
153
def _validate_revprops(self, revprops):
154
for key, value in revprops.iteritems():
155
# We know that the XML serializers do not round trip '\r'
156
# correctly, so refuse to accept them
157
if not isinstance(value, basestring):
158
raise ValueError('revision property (%s) is not a valid'
159
' (unicode) string: %r' % (key, value))
160
self._validate_unicode_text(value,
161
'revision property (%s)' % (key,))
163
def commit(self, message):
164
"""Make the actual commit.
166
:return: The revision id of the recorded revision.
168
self._validate_unicode_text(message, 'commit message')
169
rev = _mod_revision.Revision(
170
timestamp=self._timestamp,
171
timezone=self._timezone,
172
committer=self._committer,
174
inventory_sha1=self.inv_sha1,
175
revision_id=self._new_revision_id,
176
properties=self._revprops)
177
rev.parent_ids = self.parents
178
self.repository.add_revision(self._new_revision_id, rev,
179
self.new_inventory, self._config)
180
self.repository.commit_write_group()
181
return self._new_revision_id
184
"""Abort the commit that is being built.
186
self.repository.abort_write_group()
188
def revision_tree(self):
189
"""Return the tree that was just committed.
191
After calling commit() this can be called to get a RevisionTree
192
representing the newly committed tree. This is preferred to
193
calling Repository.revision_tree() because that may require
194
deserializing the inventory, while we already have a copy in
197
if self.new_inventory is None:
198
self.new_inventory = self.repository.get_inventory(
199
self._new_revision_id)
200
return RevisionTree(self.repository, self.new_inventory,
201
self._new_revision_id)
203
def finish_inventory(self):
204
"""Tell the builder that the inventory is finished.
206
:return: The inventory id in the repository, which can be used with
207
repository.get_inventory.
209
if self.new_inventory is None:
210
# an inventory delta was accumulated without creating a new
212
basis_id = self.basis_delta_revision
213
# We ignore the 'inventory' returned by add_inventory_by_delta
214
# because self.new_inventory is used to hint to the rest of the
215
# system what code path was taken
216
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
217
basis_id, self._basis_delta, self._new_revision_id,
220
if self.new_inventory.root is None:
221
raise AssertionError('Root entry should be supplied to'
222
' record_entry_contents, as of bzr 0.10.')
223
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
224
self.new_inventory.revision_id = self._new_revision_id
225
self.inv_sha1 = self.repository.add_inventory(
226
self._new_revision_id,
230
return self._new_revision_id
232
def _gen_revision_id(self):
233
"""Return new revision-id."""
234
return generate_ids.gen_revision_id(self._config.username(),
237
def _generate_revision_if_needed(self):
238
"""Create a revision id if None was supplied.
240
If the repository can not support user-specified revision ids
241
they should override this function and raise CannotSetRevisionId
242
if _new_revision_id is not None.
244
:raises: CannotSetRevisionId
246
if self._new_revision_id is None:
247
self._new_revision_id = self._gen_revision_id()
248
self.random_revid = True
250
self.random_revid = False
252
def _heads(self, file_id, revision_ids):
253
"""Calculate the graph heads for revision_ids in the graph of file_id.
255
This can use either a per-file graph or a global revision graph as we
256
have an identity relationship between the two graphs.
258
return self.__heads(revision_ids)
260
def _check_root(self, ie, parent_invs, tree):
261
"""Helper for record_entry_contents.
263
:param ie: An entry being added.
264
:param parent_invs: The inventories of the parent revisions of the
266
:param tree: The tree that is being committed.
268
# In this revision format, root entries have no knit or weave When
269
# serializing out to disk and back in root.revision is always
271
ie.revision = self._new_revision_id
273
def _require_root_change(self, tree):
274
"""Enforce an appropriate root object change.
276
This is called once when record_iter_changes is called, if and only if
277
the root was not in the delta calculated by record_iter_changes.
279
:param tree: The tree which is being committed.
281
# NB: if there are no parents then this method is not called, so no
282
# need to guard on parents having length.
283
entry = entry_factory['directory'](tree.path2id(''), '',
285
entry.revision = self._new_revision_id
286
self._basis_delta.append(('', '', entry.file_id, entry))
288
def _get_delta(self, ie, basis_inv, path):
289
"""Get a delta against the basis inventory for ie."""
290
if ie.file_id not in basis_inv:
292
result = (None, path, ie.file_id, ie)
293
self._basis_delta.append(result)
295
elif ie != basis_inv[ie.file_id]:
297
# TODO: avoid tis id2path call.
298
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
299
self._basis_delta.append(result)
305
def get_basis_delta(self):
306
"""Return the complete inventory delta versus the basis inventory.
308
This has been built up with the calls to record_delete and
309
record_entry_contents. The client must have already called
310
will_record_deletes() to indicate that they will be generating a
313
:return: An inventory delta, suitable for use with apply_delta, or
314
Repository.add_inventory_by_delta, etc.
316
if not self._recording_deletes:
317
raise AssertionError("recording deletes not activated.")
318
return self._basis_delta
320
def record_delete(self, path, file_id):
321
"""Record that a delete occured against a basis tree.
323
This is an optional API - when used it adds items to the basis_delta
324
being accumulated by the commit builder. It cannot be called unless the
325
method will_record_deletes() has been called to inform the builder that
326
a delta is being supplied.
328
:param path: The path of the thing deleted.
329
:param file_id: The file id that was deleted.
331
if not self._recording_deletes:
332
raise AssertionError("recording deletes not activated.")
333
delta = (path, None, file_id, None)
334
self._basis_delta.append(delta)
335
self._any_changes = True
338
def will_record_deletes(self):
339
"""Tell the commit builder that deletes are being notified.
341
This enables the accumulation of an inventory delta; for the resulting
342
commit to be valid, deletes against the basis MUST be recorded via
343
builder.record_delete().
345
self._recording_deletes = True
347
basis_id = self.parents[0]
349
basis_id = _mod_revision.NULL_REVISION
350
self.basis_delta_revision = basis_id
352
def record_entry_contents(self, ie, parent_invs, path, tree,
354
"""Record the content of ie from tree into the commit if needed.
356
Side effect: sets ie.revision when unchanged
358
:param ie: An inventory entry present in the commit.
359
:param parent_invs: The inventories of the parent revisions of the
361
:param path: The path the entry is at in the tree.
362
:param tree: The tree which contains this entry and should be used to
364
:param content_summary: Summary data from the tree about the paths
365
content - stat, length, exec, sha/link target. This is only
366
accessed when the entry has a revision of None - that is when it is
367
a candidate to commit.
368
:return: A tuple (change_delta, version_recorded, fs_hash).
369
change_delta is an inventory_delta change for this entry against
370
the basis tree of the commit, or None if no change occured against
372
version_recorded is True if a new version of the entry has been
373
recorded. For instance, committing a merge where a file was only
374
changed on the other side will return (delta, False).
375
fs_hash is either None, or the hash details for the path (currently
376
a tuple of the contents sha1 and the statvalue returned by
377
tree.get_file_with_stat()).
379
if self.new_inventory.root is None:
380
if ie.parent_id is not None:
381
raise errors.RootMissing()
382
self._check_root(ie, parent_invs, tree)
383
if ie.revision is None:
384
kind = content_summary[0]
386
# ie is carried over from a prior commit
388
# XXX: repository specific check for nested tree support goes here - if
389
# the repo doesn't want nested trees we skip it ?
390
if (kind == 'tree-reference' and
391
not self.repository._format.supports_tree_reference):
392
# mismatch between commit builder logic and repository:
393
# this needs the entry creation pushed down into the builder.
394
raise NotImplementedError('Missing repository subtree support.')
395
self.new_inventory.add(ie)
397
# TODO: slow, take it out of the inner loop.
399
basis_inv = parent_invs[0]
401
basis_inv = Inventory(root_id=None)
403
# ie.revision is always None if the InventoryEntry is considered
404
# for committing. We may record the previous parents revision if the
405
# content is actually unchanged against a sole head.
406
if ie.revision is not None:
407
if not self._versioned_root and path == '':
408
# repositories that do not version the root set the root's
409
# revision to the new commit even when no change occurs (more
410
# specifically, they do not record a revision on the root; and
411
# the rev id is assigned to the root during deserialisation -
412
# this masks when a change may have occurred against the basis.
413
# To match this we always issue a delta, because the revision
414
# of the root will always be changing.
415
if ie.file_id in basis_inv:
416
delta = (basis_inv.id2path(ie.file_id), path,
420
delta = (None, path, ie.file_id, ie)
421
self._basis_delta.append(delta)
422
return delta, False, None
424
# we don't need to commit this, because the caller already
425
# determined that an existing revision of this file is
426
# appropriate. If its not being considered for committing then
427
# it and all its parents to the root must be unaltered so
428
# no-change against the basis.
429
if ie.revision == self._new_revision_id:
430
raise AssertionError("Impossible situation, a skipped "
431
"inventory entry (%r) claims to be modified in this "
432
"commit (%r).", (ie, self._new_revision_id))
433
return None, False, None
434
# XXX: Friction: parent_candidates should return a list not a dict
435
# so that we don't have to walk the inventories again.
436
parent_candiate_entries = ie.parent_candidates(parent_invs)
437
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
439
for inv in parent_invs:
440
if ie.file_id in inv:
441
old_rev = inv[ie.file_id].revision
442
if old_rev in head_set:
443
heads.append(inv[ie.file_id].revision)
444
head_set.remove(inv[ie.file_id].revision)
447
# now we check to see if we need to write a new record to the
449
# We write a new entry unless there is one head to the ancestors, and
450
# the kind-derived content is unchanged.
452
# Cheapest check first: no ancestors, or more the one head in the
453
# ancestors, we write a new node.
457
# There is a single head, look it up for comparison
458
parent_entry = parent_candiate_entries[heads[0]]
459
# if the non-content specific data has changed, we'll be writing a
461
if (parent_entry.parent_id != ie.parent_id or
462
parent_entry.name != ie.name):
464
# now we need to do content specific checks:
466
# if the kind changed the content obviously has
467
if kind != parent_entry.kind:
469
# Stat cache fingerprint feedback for the caller - None as we usually
470
# don't generate one.
473
if content_summary[2] is None:
474
raise ValueError("Files must not have executable = None")
476
# We can't trust a check of the file length because of content
478
if (# if the exec bit has changed we have to store:
479
parent_entry.executable != content_summary[2]):
481
elif parent_entry.text_sha1 == content_summary[3]:
482
# all meta and content is unchanged (using a hash cache
483
# hit to check the sha)
484
ie.revision = parent_entry.revision
485
ie.text_size = parent_entry.text_size
486
ie.text_sha1 = parent_entry.text_sha1
487
ie.executable = parent_entry.executable
488
return self._get_delta(ie, basis_inv, path), False, None
490
# Either there is only a hash change(no hash cache entry,
491
# or same size content change), or there is no change on
493
# Provide the parent's hash to the store layer, so that the
494
# content is unchanged we will not store a new node.
495
nostore_sha = parent_entry.text_sha1
497
# We want to record a new node regardless of the presence or
498
# absence of a content change in the file.
500
ie.executable = content_summary[2]
501
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
503
text = file_obj.read()
507
ie.text_sha1, ie.text_size = self._add_text_to_weave(
508
ie.file_id, text, heads, nostore_sha)
509
# Let the caller know we generated a stat fingerprint.
510
fingerprint = (ie.text_sha1, stat_value)
511
except errors.ExistingContent:
512
# Turns out that the file content was unchanged, and we were
513
# only going to store a new node if it was changed. Carry over
515
ie.revision = parent_entry.revision
516
ie.text_size = parent_entry.text_size
517
ie.text_sha1 = parent_entry.text_sha1
518
ie.executable = parent_entry.executable
519
return self._get_delta(ie, basis_inv, path), False, None
520
elif kind == 'directory':
522
# all data is meta here, nothing specific to directory, so
524
ie.revision = parent_entry.revision
525
return self._get_delta(ie, basis_inv, path), False, None
526
self._add_text_to_weave(ie.file_id, '', heads, None)
527
elif kind == 'symlink':
528
current_link_target = content_summary[3]
530
# symlink target is not generic metadata, check if it has
532
if current_link_target != parent_entry.symlink_target:
535
# unchanged, carry over.
536
ie.revision = parent_entry.revision
537
ie.symlink_target = parent_entry.symlink_target
538
return self._get_delta(ie, basis_inv, path), False, None
539
ie.symlink_target = current_link_target
540
self._add_text_to_weave(ie.file_id, '', heads, None)
541
elif kind == 'tree-reference':
543
if content_summary[3] != parent_entry.reference_revision:
546
# unchanged, carry over.
547
ie.reference_revision = parent_entry.reference_revision
548
ie.revision = parent_entry.revision
549
return self._get_delta(ie, basis_inv, path), False, None
550
ie.reference_revision = content_summary[3]
551
if ie.reference_revision is None:
552
raise AssertionError("invalid content_summary for nested tree: %r"
553
% (content_summary,))
554
self._add_text_to_weave(ie.file_id, '', heads, None)
556
raise NotImplementedError('unknown kind')
557
ie.revision = self._new_revision_id
558
self._any_changes = True
559
return self._get_delta(ie, basis_inv, path), True, fingerprint
561
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
562
_entry_factory=entry_factory):
563
"""Record a new tree via iter_changes.
565
:param tree: The tree to obtain text contents from for changed objects.
566
:param basis_revision_id: The revision id of the tree the iter_changes
567
has been generated against. Currently assumed to be the same
568
as self.parents[0] - if it is not, errors may occur.
569
:param iter_changes: An iter_changes iterator with the changes to apply
570
to basis_revision_id. The iterator must not include any items with
571
a current kind of None - missing items must be either filtered out
572
or errored-on beefore record_iter_changes sees the item.
573
:param _entry_factory: Private method to bind entry_factory locally for
575
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
578
# Create an inventory delta based on deltas between all the parents and
579
# deltas between all the parent inventories. We use inventory delta's
580
# between the inventory objects because iter_changes masks
581
# last-changed-field only changes.
583
# file_id -> change map, change is fileid, paths, changed, versioneds,
584
# parents, names, kinds, executables
586
# {file_id -> revision_id -> inventory entry, for entries in parent
587
# trees that are not parents[0]
591
revtrees = list(self.repository.revision_trees(self.parents))
592
except errors.NoSuchRevision:
593
# one or more ghosts, slow path.
595
for revision_id in self.parents:
597
revtrees.append(self.repository.revision_tree(revision_id))
598
except errors.NoSuchRevision:
600
basis_revision_id = _mod_revision.NULL_REVISION
602
revtrees.append(self.repository.revision_tree(
603
_mod_revision.NULL_REVISION))
604
# The basis inventory from a repository
606
basis_inv = revtrees[0].inventory
608
basis_inv = self.repository.revision_tree(
609
_mod_revision.NULL_REVISION).inventory
610
if len(self.parents) > 0:
611
if basis_revision_id != self.parents[0] and not ghost_basis:
613
"arbitrary basis parents not yet supported with merges")
614
for revtree in revtrees[1:]:
615
for change in revtree.inventory._make_delta(basis_inv):
616
if change[1] is None:
617
# Not present in this parent.
619
if change[2] not in merged_ids:
620
if change[0] is not None:
621
basis_entry = basis_inv[change[2]]
622
merged_ids[change[2]] = [
624
basis_entry.revision,
627
parent_entries[change[2]] = {
629
basis_entry.revision:basis_entry,
631
change[3].revision:change[3],
634
merged_ids[change[2]] = [change[3].revision]
635
parent_entries[change[2]] = {change[3].revision:change[3]}
637
merged_ids[change[2]].append(change[3].revision)
638
parent_entries[change[2]][change[3].revision] = change[3]
641
# Setup the changes from the tree:
642
# changes maps file_id -> (change, [parent revision_ids])
644
for change in iter_changes:
645
# This probably looks up in basis_inv way to much.
646
if change[1][0] is not None:
647
head_candidate = [basis_inv[change[0]].revision]
650
changes[change[0]] = change, merged_ids.get(change[0],
652
unchanged_merged = set(merged_ids) - set(changes)
653
# Extend the changes dict with synthetic changes to record merges of
655
for file_id in unchanged_merged:
656
# Record a merged version of these items that did not change vs the
657
# basis. This can be either identical parallel changes, or a revert
658
# of a specific file after a merge. The recorded content will be
659
# that of the current tree (which is the same as the basis), but
660
# the per-file graph will reflect a merge.
661
# NB:XXX: We are reconstructing path information we had, this
662
# should be preserved instead.
663
# inv delta change: (file_id, (path_in_source, path_in_target),
664
# changed_content, versioned, parent, name, kind,
667
basis_entry = basis_inv[file_id]
668
except errors.NoSuchId:
669
# a change from basis->some_parents but file_id isn't in basis
670
# so was new in the merge, which means it must have changed
671
# from basis -> current, and as it hasn't the add was reverted
672
# by the user. So we discard this change.
676
(basis_inv.id2path(file_id), tree.id2path(file_id)),
678
(basis_entry.parent_id, basis_entry.parent_id),
679
(basis_entry.name, basis_entry.name),
680
(basis_entry.kind, basis_entry.kind),
681
(basis_entry.executable, basis_entry.executable))
682
changes[file_id] = (change, merged_ids[file_id])
683
# changes contains tuples with the change and a set of inventory
684
# candidates for the file.
686
# old_path, new_path, file_id, new_inventory_entry
687
seen_root = False # Is the root in the basis delta?
688
inv_delta = self._basis_delta
689
modified_rev = self._new_revision_id
690
for change, head_candidates in changes.values():
691
if change[3][1]: # versioned in target.
692
# Several things may be happening here:
693
# We may have a fork in the per-file graph
694
# - record a change with the content from tree
695
# We may have a change against < all trees
696
# - carry over the tree that hasn't changed
697
# We may have a change against all trees
698
# - record the change with the content from tree
701
entry = _entry_factory[kind](file_id, change[5][1],
703
head_set = self._heads(change[0], set(head_candidates))
706
for head_candidate in head_candidates:
707
if head_candidate in head_set:
708
heads.append(head_candidate)
709
head_set.remove(head_candidate)
712
# Could be a carry-over situation:
713
parent_entry_revs = parent_entries.get(file_id, None)
714
if parent_entry_revs:
715
parent_entry = parent_entry_revs.get(heads[0], None)
718
if parent_entry is None:
719
# The parent iter_changes was called against is the one
720
# that is the per-file head, so any change is relevant
721
# iter_changes is valid.
722
carry_over_possible = False
724
# could be a carry over situation
725
# A change against the basis may just indicate a merge,
726
# we need to check the content against the source of the
727
# merge to determine if it was changed after the merge
729
if (parent_entry.kind != entry.kind or
730
parent_entry.parent_id != entry.parent_id or
731
parent_entry.name != entry.name):
732
# Metadata common to all entries has changed
733
# against per-file parent
734
carry_over_possible = False
736
carry_over_possible = True
737
# per-type checks for changes against the parent_entry
740
# Cannot be a carry-over situation
741
carry_over_possible = False
742
# Populate the entry in the delta
744
# XXX: There is still a small race here: If someone reverts the content of a file
745
# after iter_changes examines and decides it has changed,
746
# we will unconditionally record a new version even if some
747
# other process reverts it while commit is running (with
748
# the revert happening after iter_changes did it's
751
entry.executable = True
753
entry.executable = False
754
if (carry_over_possible and
755
parent_entry.executable == entry.executable):
756
# Check the file length, content hash after reading
758
nostore_sha = parent_entry.text_sha1
761
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
763
text = file_obj.read()
767
entry.text_sha1, entry.text_size = self._add_text_to_weave(
768
file_id, text, heads, nostore_sha)
769
yield file_id, change[1][1], (entry.text_sha1, stat_value)
770
except errors.ExistingContent:
771
# No content change against a carry_over parent
772
# Perhaps this should also yield a fs hash update?
774
entry.text_size = parent_entry.text_size
775
entry.text_sha1 = parent_entry.text_sha1
776
elif kind == 'symlink':
778
entry.symlink_target = tree.get_symlink_target(file_id)
779
if (carry_over_possible and
780
parent_entry.symlink_target == entry.symlink_target):
783
self._add_text_to_weave(change[0], '', heads, None)
784
elif kind == 'directory':
785
if carry_over_possible:
788
# Nothing to set on the entry.
789
# XXX: split into the Root and nonRoot versions.
790
if change[1][1] != '' or self.repository.supports_rich_root():
791
self._add_text_to_weave(change[0], '', heads, None)
792
elif kind == 'tree-reference':
793
if not self.repository._format.supports_tree_reference:
794
# This isn't quite sane as an error, but we shouldn't
795
# ever see this code path in practice: tree's don't
796
# permit references when the repo doesn't support tree
798
raise errors.UnsupportedOperation(tree.add_reference,
800
reference_revision = tree.get_reference_revision(change[0])
801
entry.reference_revision = reference_revision
802
if (carry_over_possible and
803
parent_entry.reference_revision == reference_revision):
806
self._add_text_to_weave(change[0], '', heads, None)
808
raise AssertionError('unknown kind %r' % kind)
810
entry.revision = modified_rev
812
entry.revision = parent_entry.revision
815
new_path = change[1][1]
816
inv_delta.append((change[1][0], new_path, change[0], entry))
819
self.new_inventory = None
821
# This should perhaps be guarded by a check that the basis we
822
# commit against is the basis for the commit and if not do a delta
824
self._any_changes = True
826
# housekeeping root entry changes do not affect no-change commits.
827
self._require_root_change(tree)
828
self.basis_delta_revision = basis_revision_id
830
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
831
parent_keys = tuple([(file_id, parent) for parent in parents])
832
return self.repository.texts._add_text(
833
(file_id, self._new_revision_id), parent_keys, new_text,
834
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
837
class RootCommitBuilder(CommitBuilder):
838
"""This commitbuilder actually records the root id"""
840
# the root entry gets versioned properly by this builder.
841
_versioned_root = True
843
def _check_root(self, ie, parent_invs, tree):
844
"""Helper for record_entry_contents.
846
:param ie: An entry being added.
847
:param parent_invs: The inventories of the parent revisions of the
849
:param tree: The tree that is being committed.
852
def _require_root_change(self, tree):
853
"""Enforce an appropriate root object change.
855
This is called once when record_iter_changes is called, if and only if
856
the root was not in the delta calculated by record_iter_changes.
858
:param tree: The tree which is being committed.
860
# versioned roots do not change unless the tree found a change.
863
######################################################################
867
class Repository(_RelockDebugMixin):
868
"""Repository holding history for one or more branches.
870
The repository holds and retrieves historical information including
871
revisions and file history. It's normally accessed only by the Branch,
872
which views a particular line of development through that history.
874
The Repository builds on top of some byte storage facilies (the revisions,
875
signatures, inventories, texts and chk_bytes attributes) and a Transport,
876
which respectively provide byte storage and a means to access the (possibly
879
The byte storage facilities are addressed via tuples, which we refer to
880
as 'keys' throughout the code base. Revision_keys, inventory_keys and
881
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
882
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
883
byte string made up of a hash identifier and a hash value.
884
We use this interface because it allows low friction with the underlying
885
code that implements disk indices, network encoding and other parts of
888
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
889
the serialised revisions for the repository. This can be used to obtain
890
revision graph information or to access raw serialised revisions.
891
The result of trying to insert data into the repository via this store
892
is undefined: it should be considered read-only except for implementors
894
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
895
the serialised signatures for the repository. This can be used to
896
obtain access to raw serialised signatures. The result of trying to
897
insert data into the repository via this store is undefined: it should
898
be considered read-only except for implementors of repositories.
899
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
900
the serialised inventories for the repository. This can be used to
901
obtain unserialised inventories. The result of trying to insert data
902
into the repository via this store is undefined: it should be
903
considered read-only except for implementors of repositories.
904
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
905
texts of files and directories for the repository. This can be used to
906
obtain file texts or file graphs. Note that Repository.iter_file_bytes
907
is usually a better interface for accessing file texts.
908
The result of trying to insert data into the repository via this store
909
is undefined: it should be considered read-only except for implementors
911
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
912
any data the repository chooses to store or have indexed by its hash.
913
The result of trying to insert data into the repository via this store
914
is undefined: it should be considered read-only except for implementors
916
:ivar _transport: Transport for file access to repository, typically
917
pointing to .bzr/repository.
920
# What class to use for a CommitBuilder. Often its simpler to change this
921
# in a Repository class subclass rather than to override
922
# get_commit_builder.
923
_commit_builder_class = CommitBuilder
924
# The search regex used by xml based repositories to determine what things
925
# where changed in a single commit.
926
_file_ids_altered_regex = lazy_regex.lazy_compile(
927
r'file_id="(?P<file_id>[^"]+)"'
928
r'.* revision="(?P<revision_id>[^"]+)"'
931
def abort_write_group(self, suppress_errors=False):
932
"""Commit the contents accrued within the current write group.
934
:param suppress_errors: if true, abort_write_group will catch and log
935
unexpected errors that happen during the abort, rather than
936
allowing them to propagate. Defaults to False.
938
:seealso: start_write_group.
940
if self._write_group is not self.get_transaction():
941
# has an unlock or relock occured ?
944
'(suppressed) mismatched lock context and write group. %r, %r',
945
self._write_group, self.get_transaction())
947
raise errors.BzrError(
948
'mismatched lock context and write group. %r, %r' %
949
(self._write_group, self.get_transaction()))
951
self._abort_write_group()
952
except Exception, exc:
953
self._write_group = None
954
if not suppress_errors:
956
mutter('abort_write_group failed')
957
log_exception_quietly()
958
note('bzr: ERROR (ignored): %s', exc)
959
self._write_group = None
961
def _abort_write_group(self):
962
"""Template method for per-repository write group cleanup.
964
This is called during abort before the write group is considered to be
965
finished and should cleanup any internal state accrued during the write
966
group. There is no requirement that data handed to the repository be
967
*not* made available - this is not a rollback - but neither should any
968
attempt be made to ensure that data added is fully commited. Abort is
969
invoked when an error has occured so futher disk or network operations
970
may not be possible or may error and if possible should not be
974
def add_fallback_repository(self, repository):
975
"""Add a repository to use for looking up data not held locally.
977
:param repository: A repository.
979
if not self._format.supports_external_lookups:
980
raise errors.UnstackableRepositoryFormat(self._format, self.base)
982
# This repository will call fallback.unlock() when we transition to
983
# the unlocked state, so we make sure to increment the lock count
984
repository.lock_read()
985
self._check_fallback_repository(repository)
986
self._fallback_repositories.append(repository)
987
self.texts.add_fallback_versioned_files(repository.texts)
988
self.inventories.add_fallback_versioned_files(repository.inventories)
989
self.revisions.add_fallback_versioned_files(repository.revisions)
990
self.signatures.add_fallback_versioned_files(repository.signatures)
991
if self.chk_bytes is not None:
992
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
994
def _check_fallback_repository(self, repository):
995
"""Check that this repository can fallback to repository safely.
997
Raise an error if not.
999
:param repository: A repository to fallback to.
1001
return InterRepository._assert_same_model(self, repository)
1003
def add_inventory(self, revision_id, inv, parents):
1004
"""Add the inventory inv to the repository as revision_id.
1006
:param parents: The revision ids of the parents that revision_id
1007
is known to have and are in the repository already.
1009
:returns: The validator(which is a sha1 digest, though what is sha'd is
1010
repository format specific) of the serialized inventory.
1012
if not self.is_in_write_group():
1013
raise AssertionError("%r not in write group" % (self,))
1014
_mod_revision.check_not_reserved_id(revision_id)
1015
if not (inv.revision_id is None or inv.revision_id == revision_id):
1016
raise AssertionError(
1017
"Mismatch between inventory revision"
1018
" id and insertion revid (%r, %r)"
1019
% (inv.revision_id, revision_id))
1020
if inv.root is None:
1021
raise AssertionError()
1022
return self._add_inventory_checked(revision_id, inv, parents)
1024
def _add_inventory_checked(self, revision_id, inv, parents):
1025
"""Add inv to the repository after checking the inputs.
1027
This function can be overridden to allow different inventory styles.
1029
:seealso: add_inventory, for the contract.
1031
inv_lines = self._serializer.write_inventory_to_lines(inv)
1032
return self._inventory_add_lines(revision_id, parents,
1033
inv_lines, check_content=False)
1035
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1036
parents, basis_inv=None, propagate_caches=False):
1037
"""Add a new inventory expressed as a delta against another revision.
1039
See the inventory developers documentation for the theory behind
1042
:param basis_revision_id: The inventory id the delta was created
1043
against. (This does not have to be a direct parent.)
1044
:param delta: The inventory delta (see Inventory.apply_delta for
1046
:param new_revision_id: The revision id that the inventory is being
1048
:param parents: The revision ids of the parents that revision_id is
1049
known to have and are in the repository already. These are supplied
1050
for repositories that depend on the inventory graph for revision
1051
graph access, as well as for those that pun ancestry with delta
1053
:param basis_inv: The basis inventory if it is already known,
1055
:param propagate_caches: If True, the caches for this inventory are
1056
copied to and updated for the result if possible.
1058
:returns: (validator, new_inv)
1059
The validator(which is a sha1 digest, though what is sha'd is
1060
repository format specific) of the serialized inventory, and the
1061
resulting inventory.
1063
if not self.is_in_write_group():
1064
raise AssertionError("%r not in write group" % (self,))
1065
_mod_revision.check_not_reserved_id(new_revision_id)
1066
basis_tree = self.revision_tree(basis_revision_id)
1067
basis_tree.lock_read()
1069
# Note that this mutates the inventory of basis_tree, which not all
1070
# inventory implementations may support: A better idiom would be to
1071
# return a new inventory, but as there is no revision tree cache in
1072
# repository this is safe for now - RBC 20081013
1073
if basis_inv is None:
1074
basis_inv = basis_tree.inventory
1075
basis_inv.apply_delta(delta)
1076
basis_inv.revision_id = new_revision_id
1077
return (self.add_inventory(new_revision_id, basis_inv, parents),
1082
def _inventory_add_lines(self, revision_id, parents, lines,
1083
check_content=True):
1084
"""Store lines in inv_vf and return the sha1 of the inventory."""
1085
parents = [(parent,) for parent in parents]
1086
result = self.inventories.add_lines((revision_id,), parents, lines,
1087
check_content=check_content)[0]
1088
self.inventories._access.flush()
1091
def add_revision(self, revision_id, rev, inv=None, config=None):
1092
"""Add rev to the revision store as revision_id.
1094
:param revision_id: the revision id to use.
1095
:param rev: The revision object.
1096
:param inv: The inventory for the revision. if None, it will be looked
1097
up in the inventory storer
1098
:param config: If None no digital signature will be created.
1099
If supplied its signature_needed method will be used
1100
to determine if a signature should be made.
1102
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1104
_mod_revision.check_not_reserved_id(revision_id)
1105
if config is not None and config.signature_needed():
1107
inv = self.get_inventory(revision_id)
1108
plaintext = Testament(rev, inv).as_short_text()
1109
self.store_revision_signature(
1110
gpg.GPGStrategy(config), plaintext, revision_id)
1111
# check inventory present
1112
if not self.inventories.get_parent_map([(revision_id,)]):
1114
raise errors.WeaveRevisionNotPresent(revision_id,
1117
# yes, this is not suitable for adding with ghosts.
1118
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1121
key = (revision_id,)
1122
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1123
self._add_revision(rev)
1125
def _add_revision(self, revision):
1126
text = self._serializer.write_revision_to_string(revision)
1127
key = (revision.revision_id,)
1128
parents = tuple((parent,) for parent in revision.parent_ids)
1129
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1131
def all_revision_ids(self):
1132
"""Returns a list of all the revision ids in the repository.
1134
This is conceptually deprecated because code should generally work on
1135
the graph reachable from a particular revision, and ignore any other
1136
revisions that might be present. There is no direct replacement
1139
if 'evil' in debug.debug_flags:
1140
mutter_callsite(2, "all_revision_ids is linear with history.")
1141
return self._all_revision_ids()
1143
def _all_revision_ids(self):
1144
"""Returns a list of all the revision ids in the repository.
1146
These are in as much topological order as the underlying store can
1149
raise NotImplementedError(self._all_revision_ids)
1151
def break_lock(self):
1152
"""Break a lock if one is present from another instance.
1154
Uses the ui factory to ask for confirmation if the lock may be from
1157
self.control_files.break_lock()
1160
def _eliminate_revisions_not_present(self, revision_ids):
1161
"""Check every revision id in revision_ids to see if we have it.
1163
Returns a set of the present revisions.
1166
graph = self.get_graph()
1167
parent_map = graph.get_parent_map(revision_ids)
1168
# The old API returned a list, should this actually be a set?
1169
return parent_map.keys()
1171
def _check_inventories(self, checker):
1172
"""Check the inventories found from the revision scan.
1174
This is responsible for verifying the sha1 of inventories and
1175
creating a pending_keys set that covers data referenced by inventories.
1177
bar = ui.ui_factory.nested_progress_bar()
1179
self._do_check_inventories(checker, bar)
1183
def _do_check_inventories(self, checker, bar):
1184
"""Helper for _check_inventories."""
1186
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1187
kinds = ['chk_bytes', 'texts']
1188
count = len(checker.pending_keys)
1189
bar.update("inventories", 0, 2)
1190
current_keys = checker.pending_keys
1191
checker.pending_keys = {}
1192
# Accumulate current checks.
1193
for key in current_keys:
1194
if key[0] != 'inventories' and key[0] not in kinds:
1195
checker._report_items.append('unknown key type %r' % (key,))
1196
keys[key[0]].add(key[1:])
1197
if keys['inventories']:
1198
# NB: output order *should* be roughly sorted - topo or
1199
# inverse topo depending on repository - either way decent
1200
# to just delta against. However, pre-CHK formats didn't
1201
# try to optimise inventory layout on disk. As such the
1202
# pre-CHK code path does not use inventory deltas.
1204
for record in self.inventories.check(keys=keys['inventories']):
1205
if record.storage_kind == 'absent':
1206
checker._report_items.append(
1207
'Missing inventory {%s}' % (record.key,))
1209
last_object = self._check_record('inventories', record,
1210
checker, last_object,
1211
current_keys[('inventories',) + record.key])
1212
del keys['inventories']
1215
bar.update("texts", 1)
1216
while (checker.pending_keys or keys['chk_bytes']
1218
# Something to check.
1219
current_keys = checker.pending_keys
1220
checker.pending_keys = {}
1221
# Accumulate current checks.
1222
for key in current_keys:
1223
if key[0] not in kinds:
1224
checker._report_items.append('unknown key type %r' % (key,))
1225
keys[key[0]].add(key[1:])
1226
# Check the outermost kind only - inventories || chk_bytes || texts
1230
for record in getattr(self, kind).check(keys=keys[kind]):
1231
if record.storage_kind == 'absent':
1232
checker._report_items.append(
1233
'Missing %s {%s}' % (kind, record.key,))
1235
last_object = self._check_record(kind, record,
1236
checker, last_object, current_keys[(kind,) + record.key])
1240
def _check_record(self, kind, record, checker, last_object, item_data):
1241
"""Check a single text from this repository."""
1242
if kind == 'inventories':
1243
rev_id = record.key[0]
1244
inv = self._deserialise_inventory(rev_id,
1245
record.get_bytes_as('fulltext'))
1246
if last_object is not None:
1247
delta = inv._make_delta(last_object)
1248
for old_path, path, file_id, ie in delta:
1251
ie.check(checker, rev_id, inv)
1253
for path, ie in inv.iter_entries():
1254
ie.check(checker, rev_id, inv)
1255
if self._format.fast_deltas:
1257
elif kind == 'chk_bytes':
1258
# No code written to check chk_bytes for this repo format.
1259
checker._report_items.append(
1260
'unsupported key type chk_bytes for %s' % (record.key,))
1261
elif kind == 'texts':
1262
self._check_text(record, checker, item_data)
1264
checker._report_items.append(
1265
'unknown key type %s for %s' % (kind, record.key))
1267
def _check_text(self, record, checker, item_data):
1268
"""Check a single text."""
1269
# Check it is extractable.
1270
# TODO: check length.
1271
if record.storage_kind == 'chunked':
1272
chunks = record.get_bytes_as(record.storage_kind)
1273
sha1 = osutils.sha_strings(chunks)
1274
length = sum(map(len, chunks))
1276
content = record.get_bytes_as('fulltext')
1277
sha1 = osutils.sha_string(content)
1278
length = len(content)
1279
if item_data and sha1 != item_data[1]:
1280
checker._report_items.append(
1281
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1282
(record.key, sha1, item_data[1], item_data[2]))
1285
def create(a_bzrdir):
1286
"""Construct the current default format repository in a_bzrdir."""
1287
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1289
def __init__(self, _format, a_bzrdir, control_files):
1290
"""instantiate a Repository.
1292
:param _format: The format of the repository on disk.
1293
:param a_bzrdir: The BzrDir of the repository.
1295
In the future we will have a single api for all stores for
1296
getting file texts, inventories and revisions, then
1297
this construct will accept instances of those things.
1299
super(Repository, self).__init__()
1300
self._format = _format
1301
# the following are part of the public API for Repository:
1302
self.bzrdir = a_bzrdir
1303
self.control_files = control_files
1304
self._transport = control_files._transport
1305
self.base = self._transport.base
1307
self._reconcile_does_inventory_gc = True
1308
self._reconcile_fixes_text_parents = False
1309
self._reconcile_backsup_inventory = True
1310
self._write_group = None
1311
# Additional places to query for data.
1312
self._fallback_repositories = []
1313
# An InventoryEntry cache, used during deserialization
1314
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1315
# Is it safe to return inventory entries directly from the entry cache,
1316
# rather copying them?
1317
self._safe_to_return_from_cache = False
1320
if self._fallback_repositories:
1321
return '%s(%r, fallback_repositories=%r)' % (
1322
self.__class__.__name__,
1324
self._fallback_repositories)
1326
return '%s(%r)' % (self.__class__.__name__,
1329
def _has_same_fallbacks(self, other_repo):
1330
"""Returns true if the repositories have the same fallbacks."""
1331
my_fb = self._fallback_repositories
1332
other_fb = other_repo._fallback_repositories
1333
if len(my_fb) != len(other_fb):
1335
for f, g in zip(my_fb, other_fb):
1336
if not f.has_same_location(g):
1340
def has_same_location(self, other):
1341
"""Returns a boolean indicating if this repository is at the same
1342
location as another repository.
1344
This might return False even when two repository objects are accessing
1345
the same physical repository via different URLs.
1347
if self.__class__ is not other.__class__:
1349
return (self._transport.base == other._transport.base)
1351
def is_in_write_group(self):
1352
"""Return True if there is an open write group.
1354
:seealso: start_write_group.
1356
return self._write_group is not None
1358
def is_locked(self):
1359
return self.control_files.is_locked()
1361
def is_write_locked(self):
1362
"""Return True if this object is write locked."""
1363
return self.is_locked() and self.control_files._lock_mode == 'w'
1365
def lock_write(self, token=None):
1366
"""Lock this repository for writing.
1368
This causes caching within the repository obejct to start accumlating
1369
data during reads, and allows a 'write_group' to be obtained. Write
1370
groups must be used for actual data insertion.
1372
:param token: if this is already locked, then lock_write will fail
1373
unless the token matches the existing lock.
1374
:returns: a token if this instance supports tokens, otherwise None.
1375
:raises TokenLockingNotSupported: when a token is given but this
1376
instance doesn't support using token locks.
1377
:raises MismatchedToken: if the specified token doesn't match the token
1378
of the existing lock.
1379
:seealso: start_write_group.
1381
A token should be passed in if you know that you have locked the object
1382
some other way, and need to synchronise this object's state with that
1385
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1387
locked = self.is_locked()
1388
result = self.control_files.lock_write(token=token)
1390
self._warn_if_deprecated()
1391
self._note_lock('w')
1392
for repo in self._fallback_repositories:
1393
# Writes don't affect fallback repos
1395
self._refresh_data()
1398
def lock_read(self):
1399
locked = self.is_locked()
1400
self.control_files.lock_read()
1402
self._warn_if_deprecated()
1403
self._note_lock('r')
1404
for repo in self._fallback_repositories:
1406
self._refresh_data()
1408
def get_physical_lock_status(self):
1409
return self.control_files.get_physical_lock_status()
1411
def leave_lock_in_place(self):
1412
"""Tell this repository not to release the physical lock when this
1415
If lock_write doesn't return a token, then this method is not supported.
1417
self.control_files.leave_in_place()
1419
def dont_leave_lock_in_place(self):
1420
"""Tell this repository to release the physical lock when this
1421
object is unlocked, even if it didn't originally acquire it.
1423
If lock_write doesn't return a token, then this method is not supported.
1425
self.control_files.dont_leave_in_place()
1428
def gather_stats(self, revid=None, committers=None):
1429
"""Gather statistics from a revision id.
1431
:param revid: The revision id to gather statistics from, if None, then
1432
no revision specific statistics are gathered.
1433
:param committers: Optional parameter controlling whether to grab
1434
a count of committers from the revision specific statistics.
1435
:return: A dictionary of statistics. Currently this contains:
1436
committers: The number of committers if requested.
1437
firstrev: A tuple with timestamp, timezone for the penultimate left
1438
most ancestor of revid, if revid is not the NULL_REVISION.
1439
latestrev: A tuple with timestamp, timezone for revid, if revid is
1440
not the NULL_REVISION.
1441
revisions: The total revision count in the repository.
1442
size: An estimate disk size of the repository in bytes.
1445
if revid and committers:
1446
result['committers'] = 0
1447
if revid and revid != _mod_revision.NULL_REVISION:
1449
all_committers = set()
1450
revisions = self.get_ancestry(revid)
1451
# pop the leading None
1453
first_revision = None
1455
# ignore the revisions in the middle - just grab first and last
1456
revisions = revisions[0], revisions[-1]
1457
for revision in self.get_revisions(revisions):
1458
if not first_revision:
1459
first_revision = revision
1461
all_committers.add(revision.committer)
1462
last_revision = revision
1464
result['committers'] = len(all_committers)
1465
result['firstrev'] = (first_revision.timestamp,
1466
first_revision.timezone)
1467
result['latestrev'] = (last_revision.timestamp,
1468
last_revision.timezone)
1470
# now gather global repository information
1471
# XXX: This is available for many repos regardless of listability.
1472
if self.bzrdir.root_transport.listable():
1473
# XXX: do we want to __define len__() ?
1474
# Maybe the versionedfiles object should provide a different
1475
# method to get the number of keys.
1476
result['revisions'] = len(self.revisions.keys())
1477
# result['size'] = t
1480
def find_branches(self, using=False):
1481
"""Find branches underneath this repository.
1483
This will include branches inside other branches.
1485
:param using: If True, list only branches using this repository.
1487
if using and not self.is_shared():
1488
return self.bzrdir.list_branches()
1489
class Evaluator(object):
1492
self.first_call = True
1494
def __call__(self, bzrdir):
1495
# On the first call, the parameter is always the bzrdir
1496
# containing the current repo.
1497
if not self.first_call:
1499
repository = bzrdir.open_repository()
1500
except errors.NoRepositoryPresent:
1503
return False, ([], repository)
1504
self.first_call = False
1505
value = (bzrdir.list_branches(), None)
1509
for branches, repository in bzrdir.BzrDir.find_bzrdirs(
1510
self.bzrdir.root_transport, evaluate=Evaluator()):
1511
if branches is not None:
1512
ret.extend(branches)
1513
if not using and repository is not None:
1514
ret.extend(repository.find_branches())
1518
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
1519
"""Return the revision ids that other has that this does not.
1521
These are returned in topological order.
1523
revision_id: only return revision ids included by revision_id.
1525
return InterRepository.get(other, self).search_missing_revision_ids(
1526
revision_id, find_ghosts)
1530
"""Open the repository rooted at base.
1532
For instance, if the repository is at URL/.bzr/repository,
1533
Repository.open(URL) -> a Repository instance.
1535
control = bzrdir.BzrDir.open(base)
1536
return control.open_repository()
1538
def copy_content_into(self, destination, revision_id=None):
1539
"""Make a complete copy of the content in self into destination.
1541
This is a destructive operation! Do not use it on existing
1544
return InterRepository.get(self, destination).copy_content(revision_id)
1546
def commit_write_group(self):
1547
"""Commit the contents accrued within the current write group.
1549
:seealso: start_write_group.
1551
:return: it may return an opaque hint that can be passed to 'pack'.
1553
if self._write_group is not self.get_transaction():
1554
# has an unlock or relock occured ?
1555
raise errors.BzrError('mismatched lock context %r and '
1557
(self.get_transaction(), self._write_group))
1558
result = self._commit_write_group()
1559
self._write_group = None
1562
def _commit_write_group(self):
1563
"""Template method for per-repository write group cleanup.
1565
This is called before the write group is considered to be
1566
finished and should ensure that all data handed to the repository
1567
for writing during the write group is safely committed (to the
1568
extent possible considering file system caching etc).
1571
def suspend_write_group(self):
1572
raise errors.UnsuspendableWriteGroup(self)
1574
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1575
"""Return the keys of missing inventory parents for revisions added in
1578
A revision is not complete if the inventory delta for that revision
1579
cannot be calculated. Therefore if the parent inventories of a
1580
revision are not present, the revision is incomplete, and e.g. cannot
1581
be streamed by a smart server. This method finds missing inventory
1582
parents for revisions added in this write group.
1584
if not self._format.supports_external_lookups:
1585
# This is only an issue for stacked repositories
1587
if not self.is_in_write_group():
1588
raise AssertionError('not in a write group')
1590
# XXX: We assume that every added revision already has its
1591
# corresponding inventory, so we only check for parent inventories that
1592
# might be missing, rather than all inventories.
1593
parents = set(self.revisions._index.get_missing_parents())
1594
parents.discard(_mod_revision.NULL_REVISION)
1595
unstacked_inventories = self.inventories._index
1596
present_inventories = unstacked_inventories.get_parent_map(
1597
key[-1:] for key in parents)
1598
parents.difference_update(present_inventories)
1599
if len(parents) == 0:
1600
# No missing parent inventories.
1602
if not check_for_missing_texts:
1603
return set(('inventories', rev_id) for (rev_id,) in parents)
1604
# Ok, now we have a list of missing inventories. But these only matter
1605
# if the inventories that reference them are missing some texts they
1606
# appear to introduce.
1607
# XXX: Texts referenced by all added inventories need to be present,
1608
# but at the moment we're only checking for texts referenced by
1609
# inventories at the graph's edge.
1610
key_deps = self.revisions._index._key_dependencies
1611
key_deps.satisfy_refs_for_keys(present_inventories)
1612
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1613
file_ids = self.fileids_altered_by_revision_ids(referrers)
1614
missing_texts = set()
1615
for file_id, version_ids in file_ids.iteritems():
1616
missing_texts.update(
1617
(file_id, version_id) for version_id in version_ids)
1618
present_texts = self.texts.get_parent_map(missing_texts)
1619
missing_texts.difference_update(present_texts)
1620
if not missing_texts:
1621
# No texts are missing, so all revisions and their deltas are
1624
# Alternatively the text versions could be returned as the missing
1625
# keys, but this is likely to be less data.
1626
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1629
def refresh_data(self):
1630
"""Re-read any data needed to to synchronise with disk.
1632
This method is intended to be called after another repository instance
1633
(such as one used by a smart server) has inserted data into the
1634
repository. It may not be called during a write group, but may be
1635
called at any other time.
1637
if self.is_in_write_group():
1638
raise errors.InternalBzrError(
1639
"May not refresh_data while in a write group.")
1640
self._refresh_data()
1642
def resume_write_group(self, tokens):
1643
if not self.is_write_locked():
1644
raise errors.NotWriteLocked(self)
1645
if self._write_group:
1646
raise errors.BzrError('already in a write group')
1647
self._resume_write_group(tokens)
1648
# so we can detect unlock/relock - the write group is now entered.
1649
self._write_group = self.get_transaction()
1651
def _resume_write_group(self, tokens):
1652
raise errors.UnsuspendableWriteGroup(self)
1654
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1656
"""Fetch the content required to construct revision_id from source.
1658
If revision_id is None and fetch_spec is None, then all content is
1661
fetch() may not be used when the repository is in a write group -
1662
either finish the current write group before using fetch, or use
1663
fetch before starting the write group.
1665
:param find_ghosts: Find and copy revisions in the source that are
1666
ghosts in the target (and not reachable directly by walking out to
1667
the first-present revision in target from revision_id).
1668
:param revision_id: If specified, all the content needed for this
1669
revision ID will be copied to the target. Fetch will determine for
1670
itself which content needs to be copied.
1671
:param fetch_spec: If specified, a SearchResult or
1672
PendingAncestryResult that describes which revisions to copy. This
1673
allows copying multiple heads at once. Mutually exclusive with
1676
if fetch_spec is not None and revision_id is not None:
1677
raise AssertionError(
1678
"fetch_spec and revision_id are mutually exclusive.")
1679
if self.is_in_write_group():
1680
raise errors.InternalBzrError(
1681
"May not fetch while in a write group.")
1682
# fast path same-url fetch operations
1683
# TODO: lift out to somewhere common with RemoteRepository
1684
# <https://bugs.edge.launchpad.net/bzr/+bug/401646>
1685
if (self.has_same_location(source)
1686
and fetch_spec is None
1687
and self._has_same_fallbacks(source)):
1688
# check that last_revision is in 'from' and then return a
1690
if (revision_id is not None and
1691
not _mod_revision.is_null(revision_id)):
1692
self.get_revision(revision_id)
1694
# if there is no specific appropriate InterRepository, this will get
1695
# the InterRepository base class, which raises an
1696
# IncompatibleRepositories when asked to fetch.
1697
inter = InterRepository.get(source, self)
1698
return inter.fetch(revision_id=revision_id, pb=pb,
1699
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1701
def create_bundle(self, target, base, fileobj, format=None):
1702
return serializer.write_bundle(self, target, base, fileobj, format)
1704
def get_commit_builder(self, branch, parents, config, timestamp=None,
1705
timezone=None, committer=None, revprops=None,
1707
"""Obtain a CommitBuilder for this repository.
1709
:param branch: Branch to commit to.
1710
:param parents: Revision ids of the parents of the new revision.
1711
:param config: Configuration to use.
1712
:param timestamp: Optional timestamp recorded for commit.
1713
:param timezone: Optional timezone for timestamp.
1714
:param committer: Optional committer to set for commit.
1715
:param revprops: Optional dictionary of revision properties.
1716
:param revision_id: Optional revision id.
1718
if self._fallback_repositories:
1719
raise errors.BzrError("Cannot commit from a lightweight checkout "
1720
"to a stacked branch. See "
1721
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1722
result = self._commit_builder_class(self, parents, config,
1723
timestamp, timezone, committer, revprops, revision_id)
1724
self.start_write_group()
1727
@only_raises(errors.LockNotHeld, errors.LockBroken)
1729
if (self.control_files._lock_count == 1 and
1730
self.control_files._lock_mode == 'w'):
1731
if self._write_group is not None:
1732
self.abort_write_group()
1733
self.control_files.unlock()
1734
raise errors.BzrError(
1735
'Must end write groups before releasing write locks.')
1736
self.control_files.unlock()
1737
if self.control_files._lock_count == 0:
1738
self._inventory_entry_cache.clear()
1739
for repo in self._fallback_repositories:
1743
def clone(self, a_bzrdir, revision_id=None):
1744
"""Clone this repository into a_bzrdir using the current format.
1746
Currently no check is made that the format of this repository and
1747
the bzrdir format are compatible. FIXME RBC 20060201.
1749
:return: The newly created destination repository.
1751
# TODO: deprecate after 0.16; cloning this with all its settings is
1752
# probably not very useful -- mbp 20070423
1753
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1754
self.copy_content_into(dest_repo, revision_id)
1757
def start_write_group(self):
1758
"""Start a write group in the repository.
1760
Write groups are used by repositories which do not have a 1:1 mapping
1761
between file ids and backend store to manage the insertion of data from
1762
both fetch and commit operations.
1764
A write lock is required around the start_write_group/commit_write_group
1765
for the support of lock-requiring repository formats.
1767
One can only insert data into a repository inside a write group.
1771
if not self.is_write_locked():
1772
raise errors.NotWriteLocked(self)
1773
if self._write_group:
1774
raise errors.BzrError('already in a write group')
1775
self._start_write_group()
1776
# so we can detect unlock/relock - the write group is now entered.
1777
self._write_group = self.get_transaction()
1779
def _start_write_group(self):
1780
"""Template method for per-repository write group startup.
1782
This is called before the write group is considered to be
1787
def sprout(self, to_bzrdir, revision_id=None):
1788
"""Create a descendent repository for new development.
1790
Unlike clone, this does not copy the settings of the repository.
1792
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1793
dest_repo.fetch(self, revision_id=revision_id)
1796
def _create_sprouting_repo(self, a_bzrdir, shared):
1797
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1798
# use target default format.
1799
dest_repo = a_bzrdir.create_repository()
1801
# Most control formats need the repository to be specifically
1802
# created, but on some old all-in-one formats it's not needed
1804
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1805
except errors.UninitializableFormat:
1806
dest_repo = a_bzrdir.open_repository()
1809
def _get_sink(self):
1810
"""Return a sink for streaming into this repository."""
1811
return StreamSink(self)
1813
def _get_source(self, to_format):
1814
"""Return a source for streaming from this repository."""
1815
return StreamSource(self, to_format)
1818
def has_revision(self, revision_id):
1819
"""True if this repository has a copy of the revision."""
1820
return revision_id in self.has_revisions((revision_id,))
1823
def has_revisions(self, revision_ids):
1824
"""Probe to find out the presence of multiple revisions.
1826
:param revision_ids: An iterable of revision_ids.
1827
:return: A set of the revision_ids that were present.
1829
parent_map = self.revisions.get_parent_map(
1830
[(rev_id,) for rev_id in revision_ids])
1832
if _mod_revision.NULL_REVISION in revision_ids:
1833
result.add(_mod_revision.NULL_REVISION)
1834
result.update([key[0] for key in parent_map])
1838
def get_revision(self, revision_id):
1839
"""Return the Revision object for a named revision."""
1840
return self.get_revisions([revision_id])[0]
1843
def get_revision_reconcile(self, revision_id):
1844
"""'reconcile' helper routine that allows access to a revision always.
1846
This variant of get_revision does not cross check the weave graph
1847
against the revision one as get_revision does: but it should only
1848
be used by reconcile, or reconcile-alike commands that are correcting
1849
or testing the revision graph.
1851
return self._get_revisions([revision_id])[0]
1854
def get_revisions(self, revision_ids):
1855
"""Get many revisions at once.
1857
Repositories that need to check data on every revision read should
1858
subclass this method.
1860
return self._get_revisions(revision_ids)
1863
def _get_revisions(self, revision_ids):
1864
"""Core work logic to get many revisions without sanity checks."""
1866
for revid, rev in self._iter_revisions(revision_ids):
1868
raise errors.NoSuchRevision(self, revid)
1870
return [revs[revid] for revid in revision_ids]
1872
def _iter_revisions(self, revision_ids):
1873
"""Iterate over revision objects.
1875
:param revision_ids: An iterable of revisions to examine. None may be
1876
passed to request all revisions known to the repository. Note that
1877
not all repositories can find unreferenced revisions; for those
1878
repositories only referenced ones will be returned.
1879
:return: An iterator of (revid, revision) tuples. Absent revisions (
1880
those asked for but not available) are returned as (revid, None).
1882
if revision_ids is None:
1883
revision_ids = self.all_revision_ids()
1885
for rev_id in revision_ids:
1886
if not rev_id or not isinstance(rev_id, basestring):
1887
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1888
keys = [(key,) for key in revision_ids]
1889
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1890
for record in stream:
1891
revid = record.key[0]
1892
if record.storage_kind == 'absent':
1895
text = record.get_bytes_as('fulltext')
1896
rev = self._serializer.read_revision_from_string(text)
1899
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1900
"""Produce a generator of revision deltas.
1902
Note that the input is a sequence of REVISIONS, not revision_ids.
1903
Trees will be held in memory until the generator exits.
1904
Each delta is relative to the revision's lefthand predecessor.
1906
:param specific_fileids: if not None, the result is filtered
1907
so that only those file-ids, their parents and their
1908
children are included.
1910
# Get the revision-ids of interest
1911
required_trees = set()
1912
for revision in revisions:
1913
required_trees.add(revision.revision_id)
1914
required_trees.update(revision.parent_ids[:1])
1916
# Get the matching filtered trees. Note that it's more
1917
# efficient to pass filtered trees to changes_from() rather
1918
# than doing the filtering afterwards. changes_from() could
1919
# arguably do the filtering itself but it's path-based, not
1920
# file-id based, so filtering before or afterwards is
1922
if specific_fileids is None:
1923
trees = dict((t.get_revision_id(), t) for
1924
t in self.revision_trees(required_trees))
1926
trees = dict((t.get_revision_id(), t) for
1927
t in self._filtered_revision_trees(required_trees,
1930
# Calculate the deltas
1931
for revision in revisions:
1932
if not revision.parent_ids:
1933
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1935
old_tree = trees[revision.parent_ids[0]]
1936
yield trees[revision.revision_id].changes_from(old_tree)
1939
def get_revision_delta(self, revision_id, specific_fileids=None):
1940
"""Return the delta for one revision.
1942
The delta is relative to the left-hand predecessor of the
1945
:param specific_fileids: if not None, the result is filtered
1946
so that only those file-ids, their parents and their
1947
children are included.
1949
r = self.get_revision(revision_id)
1950
return list(self.get_deltas_for_revisions([r],
1951
specific_fileids=specific_fileids))[0]
1954
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1955
signature = gpg_strategy.sign(plaintext)
1956
self.add_signature_text(revision_id, signature)
1959
def add_signature_text(self, revision_id, signature):
1960
self.signatures.add_lines((revision_id,), (),
1961
osutils.split_lines(signature))
1963
def find_text_key_references(self):
1964
"""Find the text key references within the repository.
1966
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1967
to whether they were referred to by the inventory of the
1968
revision_id that they contain. The inventory texts from all present
1969
revision ids are assessed to generate this report.
1971
revision_keys = self.revisions.keys()
1972
w = self.inventories
1973
pb = ui.ui_factory.nested_progress_bar()
1975
return self._find_text_key_references_from_xml_inventory_lines(
1976
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1980
def _find_text_key_references_from_xml_inventory_lines(self,
1982
"""Core routine for extracting references to texts from inventories.
1984
This performs the translation of xml lines to revision ids.
1986
:param line_iterator: An iterator of lines, origin_version_id
1987
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1988
to whether they were referred to by the inventory of the
1989
revision_id that they contain. Note that if that revision_id was
1990
not part of the line_iterator's output then False will be given -
1991
even though it may actually refer to that key.
1993
if not self._serializer.support_altered_by_hack:
1994
raise AssertionError(
1995
"_find_text_key_references_from_xml_inventory_lines only "
1996
"supported for branches which store inventory as unnested xml"
1997
", not on %r" % self)
2000
# this code needs to read every new line in every inventory for the
2001
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2002
# not present in one of those inventories is unnecessary but not
2003
# harmful because we are filtering by the revision id marker in the
2004
# inventory lines : we only select file ids altered in one of those
2005
# revisions. We don't need to see all lines in the inventory because
2006
# only those added in an inventory in rev X can contain a revision=X
2008
unescape_revid_cache = {}
2009
unescape_fileid_cache = {}
2011
# jam 20061218 In a big fetch, this handles hundreds of thousands
2012
# of lines, so it has had a lot of inlining and optimizing done.
2013
# Sorry that it is a little bit messy.
2014
# Move several functions to be local variables, since this is a long
2016
search = self._file_ids_altered_regex.search
2017
unescape = _unescape_xml
2018
setdefault = result.setdefault
2019
for line, line_key in line_iterator:
2020
match = search(line)
2023
# One call to match.group() returning multiple items is quite a
2024
# bit faster than 2 calls to match.group() each returning 1
2025
file_id, revision_id = match.group('file_id', 'revision_id')
2027
# Inlining the cache lookups helps a lot when you make 170,000
2028
# lines and 350k ids, versus 8.4 unique ids.
2029
# Using a cache helps in 2 ways:
2030
# 1) Avoids unnecessary decoding calls
2031
# 2) Re-uses cached strings, which helps in future set and
2033
# (2) is enough that removing encoding entirely along with
2034
# the cache (so we are using plain strings) results in no
2035
# performance improvement.
2037
revision_id = unescape_revid_cache[revision_id]
2039
unescaped = unescape(revision_id)
2040
unescape_revid_cache[revision_id] = unescaped
2041
revision_id = unescaped
2043
# Note that unconditionally unescaping means that we deserialise
2044
# every fileid, which for general 'pull' is not great, but we don't
2045
# really want to have some many fulltexts that this matters anyway.
2048
file_id = unescape_fileid_cache[file_id]
2050
unescaped = unescape(file_id)
2051
unescape_fileid_cache[file_id] = unescaped
2054
key = (file_id, revision_id)
2055
setdefault(key, False)
2056
if revision_id == line_key[-1]:
2060
def _inventory_xml_lines_for_keys(self, keys):
2061
"""Get a line iterator of the sort needed for findind references.
2063
Not relevant for non-xml inventory repositories.
2065
Ghosts in revision_keys are ignored.
2067
:param revision_keys: The revision keys for the inventories to inspect.
2068
:return: An iterator over (inventory line, revid) for the fulltexts of
2069
all of the xml inventories specified by revision_keys.
2071
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2072
for record in stream:
2073
if record.storage_kind != 'absent':
2074
chunks = record.get_bytes_as('chunked')
2075
revid = record.key[-1]
2076
lines = osutils.chunks_to_lines(chunks)
2080
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2082
"""Helper routine for fileids_altered_by_revision_ids.
2084
This performs the translation of xml lines to revision ids.
2086
:param line_iterator: An iterator of lines, origin_version_id
2087
:param revision_keys: The revision ids to filter for. This should be a
2088
set or other type which supports efficient __contains__ lookups, as
2089
the revision key from each parsed line will be looked up in the
2090
revision_keys filter.
2091
:return: a dictionary mapping altered file-ids to an iterable of
2092
revision_ids. Each altered file-ids has the exact revision_ids that
2093
altered it listed explicitly.
2095
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2096
line_iterator).iterkeys())
2097
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2098
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2099
self._inventory_xml_lines_for_keys(parent_keys)))
2100
new_keys = seen - parent_seen
2102
setdefault = result.setdefault
2103
for key in new_keys:
2104
setdefault(key[0], set()).add(key[-1])
2107
def _find_parent_ids_of_revisions(self, revision_ids):
2108
"""Find all parent ids that are mentioned in the revision graph.
2110
:return: set of revisions that are parents of revision_ids which are
2111
not part of revision_ids themselves
2113
parent_map = self.get_parent_map(revision_ids)
2115
map(parent_ids.update, parent_map.itervalues())
2116
parent_ids.difference_update(revision_ids)
2117
parent_ids.discard(_mod_revision.NULL_REVISION)
2120
def _find_parent_keys_of_revisions(self, revision_keys):
2121
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2123
:param revision_keys: An iterable of revision_keys.
2124
:return: The parents of all revision_keys that are not already in
2127
parent_map = self.revisions.get_parent_map(revision_keys)
2129
map(parent_keys.update, parent_map.itervalues())
2130
parent_keys.difference_update(revision_keys)
2131
parent_keys.discard(_mod_revision.NULL_REVISION)
2134
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2135
"""Find the file ids and versions affected by revisions.
2137
:param revisions: an iterable containing revision ids.
2138
:param _inv_weave: The inventory weave from this repository or None.
2139
If None, the inventory weave will be opened automatically.
2140
:return: a dictionary mapping altered file-ids to an iterable of
2141
revision_ids. Each altered file-ids has the exact revision_ids that
2142
altered it listed explicitly.
2144
selected_keys = set((revid,) for revid in revision_ids)
2145
w = _inv_weave or self.inventories
2146
return self._find_file_ids_from_xml_inventory_lines(
2147
w.iter_lines_added_or_present_in_keys(
2148
selected_keys, pb=None),
2151
def iter_files_bytes(self, desired_files):
2152
"""Iterate through file versions.
2154
Files will not necessarily be returned in the order they occur in
2155
desired_files. No specific order is guaranteed.
2157
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2158
value supplied by the caller as part of desired_files. It should
2159
uniquely identify the file version in the caller's context. (Examples:
2160
an index number or a TreeTransform trans_id.)
2162
bytes_iterator is an iterable of bytestrings for the file. The
2163
kind of iterable and length of the bytestrings are unspecified, but for
2164
this implementation, it is a list of bytes produced by
2165
VersionedFile.get_record_stream().
2167
:param desired_files: a list of (file_id, revision_id, identifier)
2171
for file_id, revision_id, callable_data in desired_files:
2172
text_keys[(file_id, revision_id)] = callable_data
2173
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2174
if record.storage_kind == 'absent':
2175
raise errors.RevisionNotPresent(record.key, self)
2176
yield text_keys[record.key], record.get_bytes_as('chunked')
2178
def _generate_text_key_index(self, text_key_references=None,
2180
"""Generate a new text key index for the repository.
2182
This is an expensive function that will take considerable time to run.
2184
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2185
list of parents, also text keys. When a given key has no parents,
2186
the parents list will be [NULL_REVISION].
2188
# All revisions, to find inventory parents.
2189
if ancestors is None:
2190
graph = self.get_graph()
2191
ancestors = graph.get_parent_map(self.all_revision_ids())
2192
if text_key_references is None:
2193
text_key_references = self.find_text_key_references()
2194
pb = ui.ui_factory.nested_progress_bar()
2196
return self._do_generate_text_key_index(ancestors,
2197
text_key_references, pb)
2201
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2202
"""Helper for _generate_text_key_index to avoid deep nesting."""
2203
revision_order = tsort.topo_sort(ancestors)
2204
invalid_keys = set()
2206
for revision_id in revision_order:
2207
revision_keys[revision_id] = set()
2208
text_count = len(text_key_references)
2209
# a cache of the text keys to allow reuse; costs a dict of all the
2210
# keys, but saves a 2-tuple for every child of a given key.
2212
for text_key, valid in text_key_references.iteritems():
2214
invalid_keys.add(text_key)
2216
revision_keys[text_key[1]].add(text_key)
2217
text_key_cache[text_key] = text_key
2218
del text_key_references
2220
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2221
NULL_REVISION = _mod_revision.NULL_REVISION
2222
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2223
# too small for large or very branchy trees. However, for 55K path
2224
# trees, it would be easy to use too much memory trivially. Ideally we
2225
# could gauge this by looking at available real memory etc, but this is
2226
# always a tricky proposition.
2227
inventory_cache = lru_cache.LRUCache(10)
2228
batch_size = 10 # should be ~150MB on a 55K path tree
2229
batch_count = len(revision_order) / batch_size + 1
2231
pb.update("Calculating text parents", processed_texts, text_count)
2232
for offset in xrange(batch_count):
2233
to_query = revision_order[offset * batch_size:(offset + 1) *
2237
for revision_id in to_query:
2238
parent_ids = ancestors[revision_id]
2239
for text_key in revision_keys[revision_id]:
2240
pb.update("Calculating text parents", processed_texts)
2241
processed_texts += 1
2242
candidate_parents = []
2243
for parent_id in parent_ids:
2244
parent_text_key = (text_key[0], parent_id)
2246
check_parent = parent_text_key not in \
2247
revision_keys[parent_id]
2249
# the parent parent_id is a ghost:
2250
check_parent = False
2251
# truncate the derived graph against this ghost.
2252
parent_text_key = None
2254
# look at the parent commit details inventories to
2255
# determine possible candidates in the per file graph.
2258
inv = inventory_cache[parent_id]
2260
inv = self.revision_tree(parent_id).inventory
2261
inventory_cache[parent_id] = inv
2263
parent_entry = inv[text_key[0]]
2264
except (KeyError, errors.NoSuchId):
2266
if parent_entry is not None:
2268
text_key[0], parent_entry.revision)
2270
parent_text_key = None
2271
if parent_text_key is not None:
2272
candidate_parents.append(
2273
text_key_cache[parent_text_key])
2274
parent_heads = text_graph.heads(candidate_parents)
2275
new_parents = list(parent_heads)
2276
new_parents.sort(key=lambda x:candidate_parents.index(x))
2277
if new_parents == []:
2278
new_parents = [NULL_REVISION]
2279
text_index[text_key] = new_parents
2281
for text_key in invalid_keys:
2282
text_index[text_key] = [NULL_REVISION]
2285
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2286
"""Get an iterable listing the keys of all the data introduced by a set
2289
The keys will be ordered so that the corresponding items can be safely
2290
fetched and inserted in that order.
2292
:returns: An iterable producing tuples of (knit-kind, file-id,
2293
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2294
'revisions'. file-id is None unless knit-kind is 'file'.
2296
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2299
for result in self._find_non_file_keys_to_fetch(revision_ids):
2302
def _find_file_keys_to_fetch(self, revision_ids, pb):
2303
# XXX: it's a bit weird to control the inventory weave caching in this
2304
# generator. Ideally the caching would be done in fetch.py I think. Or
2305
# maybe this generator should explicitly have the contract that it
2306
# should not be iterated until the previously yielded item has been
2308
inv_w = self.inventories
2310
# file ids that changed
2311
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2313
num_file_ids = len(file_ids)
2314
for file_id, altered_versions in file_ids.iteritems():
2316
pb.update("Fetch texts", count, num_file_ids)
2318
yield ("file", file_id, altered_versions)
2320
def _find_non_file_keys_to_fetch(self, revision_ids):
2322
yield ("inventory", None, revision_ids)
2325
# XXX: Note ATM no callers actually pay attention to this return
2326
# instead they just use the list of revision ids and ignore
2327
# missing sigs. Consider removing this work entirely
2328
revisions_with_signatures = set(self.signatures.get_parent_map(
2329
[(r,) for r in revision_ids]))
2330
revisions_with_signatures = set(
2331
[r for (r,) in revisions_with_signatures])
2332
revisions_with_signatures.intersection_update(revision_ids)
2333
yield ("signatures", None, revisions_with_signatures)
2336
yield ("revisions", None, revision_ids)
2339
def get_inventory(self, revision_id):
2340
"""Get Inventory object by revision id."""
2341
return self.iter_inventories([revision_id]).next()
2343
def iter_inventories(self, revision_ids, ordering=None):
2344
"""Get many inventories by revision_ids.
2346
This will buffer some or all of the texts used in constructing the
2347
inventories in memory, but will only parse a single inventory at a
2350
:param revision_ids: The expected revision ids of the inventories.
2351
:param ordering: optional ordering, e.g. 'topological'. If not
2352
specified, the order of revision_ids will be preserved (by
2353
buffering if necessary).
2354
:return: An iterator of inventories.
2356
if ((None in revision_ids)
2357
or (_mod_revision.NULL_REVISION in revision_ids)):
2358
raise ValueError('cannot get null revision inventory')
2359
return self._iter_inventories(revision_ids, ordering)
2361
def _iter_inventories(self, revision_ids, ordering):
2362
"""single-document based inventory iteration."""
2363
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2364
for text, revision_id in inv_xmls:
2365
yield self._deserialise_inventory(revision_id, text)
2367
def _iter_inventory_xmls(self, revision_ids, ordering):
2368
if ordering is None:
2369
order_as_requested = True
2370
ordering = 'unordered'
2372
order_as_requested = False
2373
keys = [(revision_id,) for revision_id in revision_ids]
2376
if order_as_requested:
2377
key_iter = iter(keys)
2378
next_key = key_iter.next()
2379
stream = self.inventories.get_record_stream(keys, ordering, True)
2381
for record in stream:
2382
if record.storage_kind != 'absent':
2383
chunks = record.get_bytes_as('chunked')
2384
if order_as_requested:
2385
text_chunks[record.key] = chunks
2387
yield ''.join(chunks), record.key[-1]
2389
raise errors.NoSuchRevision(self, record.key)
2390
if order_as_requested:
2391
# Yield as many results as we can while preserving order.
2392
while next_key in text_chunks:
2393
chunks = text_chunks.pop(next_key)
2394
yield ''.join(chunks), next_key[-1]
2396
next_key = key_iter.next()
2397
except StopIteration:
2398
# We still want to fully consume the get_record_stream,
2399
# just in case it is not actually finished at this point
2403
def _deserialise_inventory(self, revision_id, xml):
2404
"""Transform the xml into an inventory object.
2406
:param revision_id: The expected revision id of the inventory.
2407
:param xml: A serialised inventory.
2409
result = self._serializer.read_inventory_from_string(xml, revision_id,
2410
entry_cache=self._inventory_entry_cache,
2411
return_from_cache=self._safe_to_return_from_cache)
2412
if result.revision_id != revision_id:
2413
raise AssertionError('revision id mismatch %s != %s' % (
2414
result.revision_id, revision_id))
2417
def get_serializer_format(self):
2418
return self._serializer.format_num
2421
def _get_inventory_xml(self, revision_id):
2422
"""Get serialized inventory as a string."""
2423
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2425
text, revision_id = texts.next()
2426
except StopIteration:
2427
raise errors.HistoryMissing(self, 'inventory', revision_id)
2430
def get_rev_id_for_revno(self, revno, known_pair):
2431
"""Return the revision id of a revno, given a later (revno, revid)
2432
pair in the same history.
2434
:return: if found (True, revid). If the available history ran out
2435
before reaching the revno, then this returns
2436
(False, (closest_revno, closest_revid)).
2438
known_revno, known_revid = known_pair
2439
partial_history = [known_revid]
2440
distance_from_known = known_revno - revno
2441
if distance_from_known < 0:
2443
'requested revno (%d) is later than given known revno (%d)'
2444
% (revno, known_revno))
2447
self, partial_history, stop_index=distance_from_known)
2448
except errors.RevisionNotPresent, err:
2449
if err.revision_id == known_revid:
2450
# The start revision (known_revid) wasn't found.
2452
# This is a stacked repository with no fallbacks, or a there's a
2453
# left-hand ghost. Either way, even though the revision named in
2454
# the error isn't in this repo, we know it's the next step in this
2455
# left-hand history.
2456
partial_history.append(err.revision_id)
2457
if len(partial_history) <= distance_from_known:
2458
# Didn't find enough history to get a revid for the revno.
2459
earliest_revno = known_revno - len(partial_history) + 1
2460
return (False, (earliest_revno, partial_history[-1]))
2461
if len(partial_history) - 1 > distance_from_known:
2462
raise AssertionError('_iter_for_revno returned too much history')
2463
return (True, partial_history[-1])
2465
def iter_reverse_revision_history(self, revision_id):
2466
"""Iterate backwards through revision ids in the lefthand history
2468
:param revision_id: The revision id to start with. All its lefthand
2469
ancestors will be traversed.
2471
graph = self.get_graph()
2472
next_id = revision_id
2474
if next_id in (None, _mod_revision.NULL_REVISION):
2477
parents = graph.get_parent_map([next_id])[next_id]
2479
raise errors.RevisionNotPresent(next_id, self)
2481
if len(parents) == 0:
2484
next_id = parents[0]
2486
def is_shared(self):
2487
"""Return True if this repository is flagged as a shared repository."""
2488
raise NotImplementedError(self.is_shared)
2491
def reconcile(self, other=None, thorough=False):
2492
"""Reconcile this repository."""
2493
from bzrlib.reconcile import RepoReconciler
2494
reconciler = RepoReconciler(self, thorough=thorough)
2495
reconciler.reconcile()
2498
def _refresh_data(self):
2499
"""Helper called from lock_* to ensure coherency with disk.
2501
The default implementation does nothing; it is however possible
2502
for repositories to maintain loaded indices across multiple locks
2503
by checking inside their implementation of this method to see
2504
whether their indices are still valid. This depends of course on
2505
the disk format being validatable in this manner. This method is
2506
also called by the refresh_data() public interface to cause a refresh
2507
to occur while in a write lock so that data inserted by a smart server
2508
push operation is visible on the client's instance of the physical
2513
def revision_tree(self, revision_id):
2514
"""Return Tree for a revision on this branch.
2516
`revision_id` may be NULL_REVISION for the empty tree revision.
2518
revision_id = _mod_revision.ensure_null(revision_id)
2519
# TODO: refactor this to use an existing revision object
2520
# so we don't need to read it in twice.
2521
if revision_id == _mod_revision.NULL_REVISION:
2522
return RevisionTree(self, Inventory(root_id=None),
2523
_mod_revision.NULL_REVISION)
2525
inv = self.get_inventory(revision_id)
2526
return RevisionTree(self, inv, revision_id)
2528
def revision_trees(self, revision_ids):
2529
"""Return Trees for revisions in this repository.
2531
:param revision_ids: a sequence of revision-ids;
2532
a revision-id may not be None or 'null:'
2534
inventories = self.iter_inventories(revision_ids)
2535
for inv in inventories:
2536
yield RevisionTree(self, inv, inv.revision_id)
2538
def _filtered_revision_trees(self, revision_ids, file_ids):
2539
"""Return Tree for a revision on this branch with only some files.
2541
:param revision_ids: a sequence of revision-ids;
2542
a revision-id may not be None or 'null:'
2543
:param file_ids: if not None, the result is filtered
2544
so that only those file-ids, their parents and their
2545
children are included.
2547
inventories = self.iter_inventories(revision_ids)
2548
for inv in inventories:
2549
# Should we introduce a FilteredRevisionTree class rather
2550
# than pre-filter the inventory here?
2551
filtered_inv = inv.filter(file_ids)
2552
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2555
def get_ancestry(self, revision_id, topo_sorted=True):
2556
"""Return a list of revision-ids integrated by a revision.
2558
The first element of the list is always None, indicating the origin
2559
revision. This might change when we have history horizons, or
2560
perhaps we should have a new API.
2562
This is topologically sorted.
2564
if _mod_revision.is_null(revision_id):
2566
if not self.has_revision(revision_id):
2567
raise errors.NoSuchRevision(self, revision_id)
2568
graph = self.get_graph()
2570
search = graph._make_breadth_first_searcher([revision_id])
2573
found, ghosts = search.next_with_ghosts()
2574
except StopIteration:
2577
if _mod_revision.NULL_REVISION in keys:
2578
keys.remove(_mod_revision.NULL_REVISION)
2580
parent_map = graph.get_parent_map(keys)
2581
keys = tsort.topo_sort(parent_map)
2582
return [None] + list(keys)
2584
def pack(self, hint=None):
2585
"""Compress the data within the repository.
2587
This operation only makes sense for some repository types. For other
2588
types it should be a no-op that just returns.
2590
This stub method does not require a lock, but subclasses should use
2591
@needs_write_lock as this is a long running call its reasonable to
2592
implicitly lock for the user.
2594
:param hint: If not supplied, the whole repository is packed.
2595
If supplied, the repository may use the hint parameter as a
2596
hint for the parts of the repository to pack. A hint can be
2597
obtained from the result of commit_write_group(). Out of
2598
date hints are simply ignored, because concurrent operations
2599
can obsolete them rapidly.
2602
def get_transaction(self):
2603
return self.control_files.get_transaction()
2605
def get_parent_map(self, revision_ids):
2606
"""See graph.StackedParentsProvider.get_parent_map"""
2607
# revisions index works in keys; this just works in revisions
2608
# therefore wrap and unwrap
2611
for revision_id in revision_ids:
2612
if revision_id == _mod_revision.NULL_REVISION:
2613
result[revision_id] = ()
2614
elif revision_id is None:
2615
raise ValueError('get_parent_map(None) is not valid')
2617
query_keys.append((revision_id ,))
2618
for ((revision_id,), parent_keys) in \
2619
self.revisions.get_parent_map(query_keys).iteritems():
2621
result[revision_id] = tuple([parent_revid
2622
for (parent_revid,) in parent_keys])
2624
result[revision_id] = (_mod_revision.NULL_REVISION,)
2627
def _make_parents_provider(self):
2631
def get_known_graph_ancestry(self, revision_ids):
2632
"""Return the known graph for a set of revision ids and their ancestors.
2634
st = static_tuple.StaticTuple
2635
revision_keys = [st(r_id).intern() for r_id in revision_ids]
2636
known_graph = self.revisions.get_known_graph_ancestry(revision_keys)
2637
return graph.GraphThunkIdsToKeys(known_graph)
2639
def get_graph(self, other_repository=None):
2640
"""Return the graph walker for this repository format"""
2641
parents_provider = self._make_parents_provider()
2642
if (other_repository is not None and
2643
not self.has_same_location(other_repository)):
2644
parents_provider = graph.StackedParentsProvider(
2645
[parents_provider, other_repository._make_parents_provider()])
2646
return graph.Graph(parents_provider)
2648
def _get_versioned_file_checker(self, text_key_references=None,
2650
"""Return an object suitable for checking versioned files.
2652
:param text_key_references: if non-None, an already built
2653
dictionary mapping text keys ((fileid, revision_id) tuples)
2654
to whether they were referred to by the inventory of the
2655
revision_id that they contain. If None, this will be
2657
:param ancestors: Optional result from
2658
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2661
return _VersionedFileChecker(self,
2662
text_key_references=text_key_references, ancestors=ancestors)
2664
def revision_ids_to_search_result(self, result_set):
2665
"""Convert a set of revision ids to a graph SearchResult."""
2666
result_parents = set()
2667
for parents in self.get_graph().get_parent_map(
2668
result_set).itervalues():
2669
result_parents.update(parents)
2670
included_keys = result_set.intersection(result_parents)
2671
start_keys = result_set.difference(included_keys)
2672
exclude_keys = result_parents.difference(result_set)
2673
result = graph.SearchResult(start_keys, exclude_keys,
2674
len(result_set), result_set)
2678
def set_make_working_trees(self, new_value):
2679
"""Set the policy flag for making working trees when creating branches.
2681
This only applies to branches that use this repository.
2683
The default is 'True'.
2684
:param new_value: True to restore the default, False to disable making
2687
raise NotImplementedError(self.set_make_working_trees)
2689
def make_working_trees(self):
2690
"""Returns the policy for making working trees on new branches."""
2691
raise NotImplementedError(self.make_working_trees)
2694
def sign_revision(self, revision_id, gpg_strategy):
2695
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2696
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2699
def has_signature_for_revision_id(self, revision_id):
2700
"""Query for a revision signature for revision_id in the repository."""
2701
if not self.has_revision(revision_id):
2702
raise errors.NoSuchRevision(self, revision_id)
2703
sig_present = (1 == len(
2704
self.signatures.get_parent_map([(revision_id,)])))
2708
def get_signature_text(self, revision_id):
2709
"""Return the text for a signature."""
2710
stream = self.signatures.get_record_stream([(revision_id,)],
2712
record = stream.next()
2713
if record.storage_kind == 'absent':
2714
raise errors.NoSuchRevision(self, revision_id)
2715
return record.get_bytes_as('fulltext')
2718
def check(self, revision_ids=None, callback_refs=None, check_repo=True):
2719
"""Check consistency of all history of given revision_ids.
2721
Different repository implementations should override _check().
2723
:param revision_ids: A non-empty list of revision_ids whose ancestry
2724
will be checked. Typically the last revision_id of a branch.
2725
:param callback_refs: A dict of check-refs to resolve and callback
2726
the check/_check method on the items listed as wanting the ref.
2728
:param check_repo: If False do not check the repository contents, just
2729
calculate the data callback_refs requires and call them back.
2731
return self._check(revision_ids, callback_refs=callback_refs,
2732
check_repo=check_repo)
2734
def _check(self, revision_ids, callback_refs, check_repo):
2735
result = check.Check(self, check_repo=check_repo)
2736
result.check(callback_refs)
2739
def _warn_if_deprecated(self, branch=None):
2740
global _deprecation_warning_done
2741
if _deprecation_warning_done:
2745
conf = config.GlobalConfig()
2747
conf = branch.get_config()
2748
if conf.suppress_warning('format_deprecation'):
2750
warning("Format %s for %s is deprecated -"
2751
" please use 'bzr upgrade' to get better performance"
2752
% (self._format, self.bzrdir.transport.base))
2754
_deprecation_warning_done = True
2756
def supports_rich_root(self):
2757
return self._format.rich_root_data
2759
def _check_ascii_revisionid(self, revision_id, method):
2760
"""Private helper for ascii-only repositories."""
2761
# weave repositories refuse to store revisionids that are non-ascii.
2762
if revision_id is not None:
2763
# weaves require ascii revision ids.
2764
if isinstance(revision_id, unicode):
2766
revision_id.encode('ascii')
2767
except UnicodeEncodeError:
2768
raise errors.NonAsciiRevisionId(method, self)
2771
revision_id.decode('ascii')
2772
except UnicodeDecodeError:
2773
raise errors.NonAsciiRevisionId(method, self)
2775
def revision_graph_can_have_wrong_parents(self):
2776
"""Is it possible for this repository to have a revision graph with
2779
If True, then this repository must also implement
2780
_find_inconsistent_revision_parents so that check and reconcile can
2781
check for inconsistencies before proceeding with other checks that may
2782
depend on the revision index being consistent.
2784
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2787
# remove these delegates a while after bzr 0.15
2788
def __make_delegated(name, from_module):
2789
def _deprecated_repository_forwarder():
2790
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2791
% (name, from_module),
2794
m = __import__(from_module, globals(), locals(), [name])
2796
return getattr(m, name)
2797
except AttributeError:
2798
raise AttributeError('module %s has no name %s'
2800
globals()[name] = _deprecated_repository_forwarder
2803
'AllInOneRepository',
2804
'WeaveMetaDirRepository',
2805
'PreSplitOutRepositoryFormat',
2806
'RepositoryFormat4',
2807
'RepositoryFormat5',
2808
'RepositoryFormat6',
2809
'RepositoryFormat7',
2811
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2815
'RepositoryFormatKnit',
2816
'RepositoryFormatKnit1',
2818
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2821
def install_revision(repository, rev, revision_tree):
2822
"""Install all revision data into a repository."""
2823
install_revisions(repository, [(rev, revision_tree, None)])
2826
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2827
"""Install all revision data into a repository.
2829
Accepts an iterable of revision, tree, signature tuples. The signature
2832
repository.start_write_group()
2834
inventory_cache = lru_cache.LRUCache(10)
2835
for n, (revision, revision_tree, signature) in enumerate(iterable):
2836
_install_revision(repository, revision, revision_tree, signature,
2839
pb.update('Transferring revisions', n + 1, num_revisions)
2841
repository.abort_write_group()
2844
repository.commit_write_group()
2847
def _install_revision(repository, rev, revision_tree, signature,
2849
"""Install all revision data into a repository."""
2850
present_parents = []
2852
for p_id in rev.parent_ids:
2853
if repository.has_revision(p_id):
2854
present_parents.append(p_id)
2855
parent_trees[p_id] = repository.revision_tree(p_id)
2857
parent_trees[p_id] = repository.revision_tree(
2858
_mod_revision.NULL_REVISION)
2860
inv = revision_tree.inventory
2861
entries = inv.iter_entries()
2862
# backwards compatibility hack: skip the root id.
2863
if not repository.supports_rich_root():
2864
path, root = entries.next()
2865
if root.revision != rev.revision_id:
2866
raise errors.IncompatibleRevision(repr(repository))
2868
for path, ie in entries:
2869
text_keys[(ie.file_id, ie.revision)] = ie
2870
text_parent_map = repository.texts.get_parent_map(text_keys)
2871
missing_texts = set(text_keys) - set(text_parent_map)
2872
# Add the texts that are not already present
2873
for text_key in missing_texts:
2874
ie = text_keys[text_key]
2876
# FIXME: TODO: The following loop overlaps/duplicates that done by
2877
# commit to determine parents. There is a latent/real bug here where
2878
# the parents inserted are not those commit would do - in particular
2879
# they are not filtered by heads(). RBC, AB
2880
for revision, tree in parent_trees.iteritems():
2881
if ie.file_id not in tree:
2883
parent_id = tree.inventory[ie.file_id].revision
2884
if parent_id in text_parents:
2886
text_parents.append((ie.file_id, parent_id))
2887
lines = revision_tree.get_file(ie.file_id).readlines()
2888
repository.texts.add_lines(text_key, text_parents, lines)
2890
# install the inventory
2891
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2892
# Cache this inventory
2893
inventory_cache[rev.revision_id] = inv
2895
basis_inv = inventory_cache[rev.parent_ids[0]]
2897
repository.add_inventory(rev.revision_id, inv, present_parents)
2899
delta = inv._make_delta(basis_inv)
2900
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2901
rev.revision_id, present_parents)
2903
repository.add_inventory(rev.revision_id, inv, present_parents)
2904
except errors.RevisionAlreadyPresent:
2906
if signature is not None:
2907
repository.add_signature_text(rev.revision_id, signature)
2908
repository.add_revision(rev.revision_id, rev, inv)
2911
class MetaDirRepository(Repository):
2912
"""Repositories in the new meta-dir layout.
2914
:ivar _transport: Transport for access to repository control files,
2915
typically pointing to .bzr/repository.
2918
def __init__(self, _format, a_bzrdir, control_files):
2919
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2920
self._transport = control_files._transport
2922
def is_shared(self):
2923
"""Return True if this repository is flagged as a shared repository."""
2924
return self._transport.has('shared-storage')
2927
def set_make_working_trees(self, new_value):
2928
"""Set the policy flag for making working trees when creating branches.
2930
This only applies to branches that use this repository.
2932
The default is 'True'.
2933
:param new_value: True to restore the default, False to disable making
2938
self._transport.delete('no-working-trees')
2939
except errors.NoSuchFile:
2942
self._transport.put_bytes('no-working-trees', '',
2943
mode=self.bzrdir._get_file_mode())
2945
def make_working_trees(self):
2946
"""Returns the policy for making working trees on new branches."""
2947
return not self._transport.has('no-working-trees')
2950
class MetaDirVersionedFileRepository(MetaDirRepository):
2951
"""Repositories in a meta-dir, that work via versioned file objects."""
2953
def __init__(self, _format, a_bzrdir, control_files):
2954
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2958
network_format_registry = registry.FormatRegistry()
2959
"""Registry of formats indexed by their network name.
2961
The network name for a repository format is an identifier that can be used when
2962
referring to formats with smart server operations. See
2963
RepositoryFormat.network_name() for more detail.
2967
format_registry = registry.FormatRegistry(network_format_registry)
2968
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
2970
This can contain either format instances themselves, or classes/factories that
2971
can be called to obtain one.
2975
#####################################################################
2976
# Repository Formats
2978
class RepositoryFormat(object):
2979
"""A repository format.
2981
Formats provide four things:
2982
* An initialization routine to construct repository data on disk.
2983
* a optional format string which is used when the BzrDir supports
2985
* an open routine which returns a Repository instance.
2986
* A network name for referring to the format in smart server RPC
2989
There is one and only one Format subclass for each on-disk format. But
2990
there can be one Repository subclass that is used for several different
2991
formats. The _format attribute on a Repository instance can be used to
2992
determine the disk format.
2994
Formats are placed in a registry by their format string for reference
2995
during opening. These should be subclasses of RepositoryFormat for
2998
Once a format is deprecated, just deprecate the initialize and open
2999
methods on the format class. Do not deprecate the object, as the
3000
object may be created even when a repository instance hasn't been
3003
Common instance attributes:
3004
_matchingbzrdir - the bzrdir format that the repository format was
3005
originally written to work with. This can be used if manually
3006
constructing a bzrdir and repository, or more commonly for test suite
3010
# Set to True or False in derived classes. True indicates that the format
3011
# supports ghosts gracefully.
3012
supports_ghosts = None
3013
# Can this repository be given external locations to lookup additional
3014
# data. Set to True or False in derived classes.
3015
supports_external_lookups = None
3016
# Does this format support CHK bytestring lookups. Set to True or False in
3018
supports_chks = None
3019
# Should commit add an inventory, or an inventory delta to the repository.
3020
_commit_inv_deltas = True
3021
# What order should fetch operations request streams in?
3022
# The default is unordered as that is the cheapest for an origin to
3024
_fetch_order = 'unordered'
3025
# Does this repository format use deltas that can be fetched as-deltas ?
3026
# (E.g. knits, where the knit deltas can be transplanted intact.
3027
# We default to False, which will ensure that enough data to get
3028
# a full text out of any fetch stream will be grabbed.
3029
_fetch_uses_deltas = False
3030
# Should fetch trigger a reconcile after the fetch? Only needed for
3031
# some repository formats that can suffer internal inconsistencies.
3032
_fetch_reconcile = False
3033
# Does this format have < O(tree_size) delta generation. Used to hint what
3034
# code path for commit, amongst other things.
3036
# Does doing a pack operation compress data? Useful for the pack UI command
3037
# (so if there is one pack, the operation can still proceed because it may
3038
# help), and for fetching when data won't have come from the same
3040
pack_compresses = False
3041
# Does the repository inventory storage understand references to trees?
3042
supports_tree_reference = None
3043
# Is the format experimental ?
3044
experimental = False
3047
return "<%s>" % self.__class__.__name__
3049
def __eq__(self, other):
3050
# format objects are generally stateless
3051
return isinstance(other, self.__class__)
3053
def __ne__(self, other):
3054
return not self == other
3057
def find_format(klass, a_bzrdir):
3058
"""Return the format for the repository object in a_bzrdir.
3060
This is used by bzr native formats that have a "format" file in
3061
the repository. Other methods may be used by different types of
3065
transport = a_bzrdir.get_repository_transport(None)
3066
format_string = transport.get_bytes("format")
3067
return format_registry.get(format_string)
3068
except errors.NoSuchFile:
3069
raise errors.NoRepositoryPresent(a_bzrdir)
3071
raise errors.UnknownFormatError(format=format_string,
3075
def register_format(klass, format):
3076
format_registry.register(format.get_format_string(), format)
3079
def unregister_format(klass, format):
3080
format_registry.remove(format.get_format_string())
3083
def get_default_format(klass):
3084
"""Return the current default format."""
3085
from bzrlib import bzrdir
3086
return bzrdir.format_registry.make_bzrdir('default').repository_format
3088
def get_format_string(self):
3089
"""Return the ASCII format string that identifies this format.
3091
Note that in pre format ?? repositories the format string is
3092
not permitted nor written to disk.
3094
raise NotImplementedError(self.get_format_string)
3096
def get_format_description(self):
3097
"""Return the short description for this format."""
3098
raise NotImplementedError(self.get_format_description)
3100
# TODO: this shouldn't be in the base class, it's specific to things that
3101
# use weaves or knits -- mbp 20070207
3102
def _get_versioned_file_store(self,
3107
versionedfile_class=None,
3108
versionedfile_kwargs={},
3110
if versionedfile_class is None:
3111
versionedfile_class = self._versionedfile_class
3112
weave_transport = control_files._transport.clone(name)
3113
dir_mode = control_files._dir_mode
3114
file_mode = control_files._file_mode
3115
return VersionedFileStore(weave_transport, prefixed=prefixed,
3117
file_mode=file_mode,
3118
versionedfile_class=versionedfile_class,
3119
versionedfile_kwargs=versionedfile_kwargs,
3122
def initialize(self, a_bzrdir, shared=False):
3123
"""Initialize a repository of this format in a_bzrdir.
3125
:param a_bzrdir: The bzrdir to put the new repository in it.
3126
:param shared: The repository should be initialized as a sharable one.
3127
:returns: The new repository object.
3129
This may raise UninitializableFormat if shared repository are not
3130
compatible the a_bzrdir.
3132
raise NotImplementedError(self.initialize)
3134
def is_supported(self):
3135
"""Is this format supported?
3137
Supported formats must be initializable and openable.
3138
Unsupported formats may not support initialization or committing or
3139
some other features depending on the reason for not being supported.
3143
def network_name(self):
3144
"""A simple byte string uniquely identifying this format for RPC calls.
3146
MetaDir repository formats use their disk format string to identify the
3147
repository over the wire. All in one formats such as bzr < 0.8, and
3148
foreign formats like svn/git and hg should use some marker which is
3149
unique and immutable.
3151
raise NotImplementedError(self.network_name)
3153
def check_conversion_target(self, target_format):
3154
if self.rich_root_data and not target_format.rich_root_data:
3155
raise errors.BadConversionTarget(
3156
'Does not support rich root data.', target_format,
3158
if (self.supports_tree_reference and
3159
not getattr(target_format, 'supports_tree_reference', False)):
3160
raise errors.BadConversionTarget(
3161
'Does not support nested trees', target_format,
3164
def open(self, a_bzrdir, _found=False):
3165
"""Return an instance of this format for the bzrdir a_bzrdir.
3167
_found is a private parameter, do not use it.
3169
raise NotImplementedError(self.open)
3172
class MetaDirRepositoryFormat(RepositoryFormat):
3173
"""Common base class for the new repositories using the metadir layout."""
3175
rich_root_data = False
3176
supports_tree_reference = False
3177
supports_external_lookups = False
3180
def _matchingbzrdir(self):
3181
matching = bzrdir.BzrDirMetaFormat1()
3182
matching.repository_format = self
3186
super(MetaDirRepositoryFormat, self).__init__()
3188
def _create_control_files(self, a_bzrdir):
3189
"""Create the required files and the initial control_files object."""
3190
# FIXME: RBC 20060125 don't peek under the covers
3191
# NB: no need to escape relative paths that are url safe.
3192
repository_transport = a_bzrdir.get_repository_transport(self)
3193
control_files = lockable_files.LockableFiles(repository_transport,
3194
'lock', lockdir.LockDir)
3195
control_files.create_lock()
3196
return control_files
3198
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
3199
"""Upload the initial blank content."""
3200
control_files = self._create_control_files(a_bzrdir)
3201
control_files.lock_write()
3202
transport = control_files._transport
3204
utf8_files += [('shared-storage', '')]
3206
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
3207
for (filename, content_stream) in files:
3208
transport.put_file(filename, content_stream,
3209
mode=a_bzrdir._get_file_mode())
3210
for (filename, content_bytes) in utf8_files:
3211
transport.put_bytes_non_atomic(filename, content_bytes,
3212
mode=a_bzrdir._get_file_mode())
3214
control_files.unlock()
3216
def network_name(self):
3217
"""Metadir formats have matching disk and network format strings."""
3218
return self.get_format_string()
3221
# Pre-0.8 formats that don't have a disk format string (because they are
3222
# versioned by the matching control directory). We use the control directories
3223
# disk format string as a key for the network_name because they meet the
3224
# constraints (simple string, unique, immutable).
3225
network_format_registry.register_lazy(
3226
"Bazaar-NG branch, format 5\n",
3227
'bzrlib.repofmt.weaverepo',
3228
'RepositoryFormat5',
3230
network_format_registry.register_lazy(
3231
"Bazaar-NG branch, format 6\n",
3232
'bzrlib.repofmt.weaverepo',
3233
'RepositoryFormat6',
3236
# formats which have no format string are not discoverable or independently
3237
# creatable on disk, so are not registered in format_registry. They're
3238
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
3239
# needed, it's constructed directly by the BzrDir. Non-native formats where
3240
# the repository is not separately opened are similar.
3242
format_registry.register_lazy(
3243
'Bazaar-NG Repository format 7',
3244
'bzrlib.repofmt.weaverepo',
3248
format_registry.register_lazy(
3249
'Bazaar-NG Knit Repository Format 1',
3250
'bzrlib.repofmt.knitrepo',
3251
'RepositoryFormatKnit1',
3254
format_registry.register_lazy(
3255
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
3256
'bzrlib.repofmt.knitrepo',
3257
'RepositoryFormatKnit3',
3260
format_registry.register_lazy(
3261
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
3262
'bzrlib.repofmt.knitrepo',
3263
'RepositoryFormatKnit4',
3266
# Pack-based formats. There is one format for pre-subtrees, and one for
3267
# post-subtrees to allow ease of testing.
3268
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
3269
format_registry.register_lazy(
3270
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
3271
'bzrlib.repofmt.pack_repo',
3272
'RepositoryFormatKnitPack1',
3274
format_registry.register_lazy(
3275
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
3276
'bzrlib.repofmt.pack_repo',
3277
'RepositoryFormatKnitPack3',
3279
format_registry.register_lazy(
3280
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
3281
'bzrlib.repofmt.pack_repo',
3282
'RepositoryFormatKnitPack4',
3284
format_registry.register_lazy(
3285
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
3286
'bzrlib.repofmt.pack_repo',
3287
'RepositoryFormatKnitPack5',
3289
format_registry.register_lazy(
3290
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
3291
'bzrlib.repofmt.pack_repo',
3292
'RepositoryFormatKnitPack5RichRoot',
3294
format_registry.register_lazy(
3295
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3296
'bzrlib.repofmt.pack_repo',
3297
'RepositoryFormatKnitPack5RichRootBroken',
3299
format_registry.register_lazy(
3300
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3301
'bzrlib.repofmt.pack_repo',
3302
'RepositoryFormatKnitPack6',
3304
format_registry.register_lazy(
3305
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3306
'bzrlib.repofmt.pack_repo',
3307
'RepositoryFormatKnitPack6RichRoot',
3310
# Development formats.
3311
# Obsolete but kept pending a CHK based subtree format.
3312
format_registry.register_lazy(
3313
("Bazaar development format 2 with subtree support "
3314
"(needs bzr.dev from before 1.8)\n"),
3315
'bzrlib.repofmt.pack_repo',
3316
'RepositoryFormatPackDevelopment2Subtree',
3319
# 1.14->1.16 go below here
3320
format_registry.register_lazy(
3321
'Bazaar development format - group compression and chk inventory'
3322
' (needs bzr.dev from 1.14)\n',
3323
'bzrlib.repofmt.groupcompress_repo',
3324
'RepositoryFormatCHK1',
3327
format_registry.register_lazy(
3328
'Bazaar development format - chk repository with bencode revision '
3329
'serialization (needs bzr.dev from 1.16)\n',
3330
'bzrlib.repofmt.groupcompress_repo',
3331
'RepositoryFormatCHK2',
3333
format_registry.register_lazy(
3334
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3335
'bzrlib.repofmt.groupcompress_repo',
3336
'RepositoryFormat2a',
3340
class InterRepository(InterObject):
3341
"""This class represents operations taking place between two repositories.
3343
Its instances have methods like copy_content and fetch, and contain
3344
references to the source and target repositories these operations can be
3347
Often we will provide convenience methods on 'repository' which carry out
3348
operations with another repository - they will always forward to
3349
InterRepository.get(other).method_name(parameters).
3352
_walk_to_common_revisions_batch_size = 50
3354
"""The available optimised InterRepository types."""
3357
def copy_content(self, revision_id=None):
3358
"""Make a complete copy of the content in self into destination.
3360
This is a destructive operation! Do not use it on existing
3363
:param revision_id: Only copy the content needed to construct
3364
revision_id and its parents.
3367
self.target.set_make_working_trees(self.source.make_working_trees())
3368
except NotImplementedError:
3370
self.target.fetch(self.source, revision_id=revision_id)
3373
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3375
"""Fetch the content required to construct revision_id.
3377
The content is copied from self.source to self.target.
3379
:param revision_id: if None all content is copied, if NULL_REVISION no
3384
ui.ui_factory.warn_experimental_format_fetch(self)
3385
f = _mod_fetch.RepoFetcher(to_repository=self.target,
3386
from_repository=self.source,
3387
last_revision=revision_id,
3388
fetch_spec=fetch_spec,
3389
find_ghosts=find_ghosts)
3391
def _walk_to_common_revisions(self, revision_ids):
3392
"""Walk out from revision_ids in source to revisions target has.
3394
:param revision_ids: The start point for the search.
3395
:return: A set of revision ids.
3397
target_graph = self.target.get_graph()
3398
revision_ids = frozenset(revision_ids)
3399
missing_revs = set()
3400
source_graph = self.source.get_graph()
3401
# ensure we don't pay silly lookup costs.
3402
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3403
null_set = frozenset([_mod_revision.NULL_REVISION])
3404
searcher_exhausted = False
3408
# Iterate the searcher until we have enough next_revs
3409
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3411
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3412
next_revs.update(next_revs_part)
3413
ghosts.update(ghosts_part)
3414
except StopIteration:
3415
searcher_exhausted = True
3417
# If there are ghosts in the source graph, and the caller asked for
3418
# them, make sure that they are present in the target.
3419
# We don't care about other ghosts as we can't fetch them and
3420
# haven't been asked to.
3421
ghosts_to_check = set(revision_ids.intersection(ghosts))
3422
revs_to_get = set(next_revs).union(ghosts_to_check)
3424
have_revs = set(target_graph.get_parent_map(revs_to_get))
3425
# we always have NULL_REVISION present.
3426
have_revs = have_revs.union(null_set)
3427
# Check if the target is missing any ghosts we need.
3428
ghosts_to_check.difference_update(have_revs)
3430
# One of the caller's revision_ids is a ghost in both the
3431
# source and the target.
3432
raise errors.NoSuchRevision(
3433
self.source, ghosts_to_check.pop())
3434
missing_revs.update(next_revs - have_revs)
3435
# Because we may have walked past the original stop point, make
3436
# sure everything is stopped
3437
stop_revs = searcher.find_seen_ancestors(have_revs)
3438
searcher.stop_searching_any(stop_revs)
3439
if searcher_exhausted:
3441
return searcher.get_result()
3444
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3445
"""Return the revision ids that source has that target does not.
3447
:param revision_id: only return revision ids included by this
3449
:param find_ghosts: If True find missing revisions in deep history
3450
rather than just finding the surface difference.
3451
:return: A bzrlib.graph.SearchResult.
3453
# stop searching at found target revisions.
3454
if not find_ghosts and revision_id is not None:
3455
return self._walk_to_common_revisions([revision_id])
3456
# generic, possibly worst case, slow code path.
3457
target_ids = set(self.target.all_revision_ids())
3458
if revision_id is not None:
3459
source_ids = self.source.get_ancestry(revision_id)
3460
if source_ids[0] is not None:
3461
raise AssertionError()
3464
source_ids = self.source.all_revision_ids()
3465
result_set = set(source_ids).difference(target_ids)
3466
return self.source.revision_ids_to_search_result(result_set)
3469
def _same_model(source, target):
3470
"""True if source and target have the same data representation.
3472
Note: this is always called on the base class; overriding it in a
3473
subclass will have no effect.
3476
InterRepository._assert_same_model(source, target)
3478
except errors.IncompatibleRepositories, e:
3482
def _assert_same_model(source, target):
3483
"""Raise an exception if two repositories do not use the same model.
3485
if source.supports_rich_root() != target.supports_rich_root():
3486
raise errors.IncompatibleRepositories(source, target,
3487
"different rich-root support")
3488
if source._serializer != target._serializer:
3489
raise errors.IncompatibleRepositories(source, target,
3490
"different serializers")
3493
class InterSameDataRepository(InterRepository):
3494
"""Code for converting between repositories that represent the same data.
3496
Data format and model must match for this to work.
3500
def _get_repo_format_to_test(self):
3501
"""Repository format for testing with.
3503
InterSameData can pull from subtree to subtree and from non-subtree to
3504
non-subtree, so we test this with the richest repository format.
3506
from bzrlib.repofmt import knitrepo
3507
return knitrepo.RepositoryFormatKnit3()
3510
def is_compatible(source, target):
3511
return InterRepository._same_model(source, target)
3514
class InterWeaveRepo(InterSameDataRepository):
3515
"""Optimised code paths between Weave based repositories.
3517
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3518
implemented lazy inter-object optimisation.
3522
def _get_repo_format_to_test(self):
3523
from bzrlib.repofmt import weaverepo
3524
return weaverepo.RepositoryFormat7()
3527
def is_compatible(source, target):
3528
"""Be compatible with known Weave formats.
3530
We don't test for the stores being of specific types because that
3531
could lead to confusing results, and there is no need to be
3534
from bzrlib.repofmt.weaverepo import (
3540
return (isinstance(source._format, (RepositoryFormat5,
3542
RepositoryFormat7)) and
3543
isinstance(target._format, (RepositoryFormat5,
3545
RepositoryFormat7)))
3546
except AttributeError:
3550
def copy_content(self, revision_id=None):
3551
"""See InterRepository.copy_content()."""
3552
# weave specific optimised path:
3554
self.target.set_make_working_trees(self.source.make_working_trees())
3555
except (errors.RepositoryUpgradeRequired, NotImplemented):
3557
# FIXME do not peek!
3558
if self.source._transport.listable():
3559
pb = ui.ui_factory.nested_progress_bar()
3561
self.target.texts.insert_record_stream(
3562
self.source.texts.get_record_stream(
3563
self.source.texts.keys(), 'topological', False))
3564
pb.update('Copying inventory', 0, 1)
3565
self.target.inventories.insert_record_stream(
3566
self.source.inventories.get_record_stream(
3567
self.source.inventories.keys(), 'topological', False))
3568
self.target.signatures.insert_record_stream(
3569
self.source.signatures.get_record_stream(
3570
self.source.signatures.keys(),
3572
self.target.revisions.insert_record_stream(
3573
self.source.revisions.get_record_stream(
3574
self.source.revisions.keys(),
3575
'topological', True))
3579
self.target.fetch(self.source, revision_id=revision_id)
3582
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3583
"""See InterRepository.missing_revision_ids()."""
3584
# we want all revisions to satisfy revision_id in source.
3585
# but we don't want to stat every file here and there.
3586
# we want then, all revisions other needs to satisfy revision_id
3587
# checked, but not those that we have locally.
3588
# so the first thing is to get a subset of the revisions to
3589
# satisfy revision_id in source, and then eliminate those that
3590
# we do already have.
3591
# this is slow on high latency connection to self, but as this
3592
# disk format scales terribly for push anyway due to rewriting
3593
# inventory.weave, this is considered acceptable.
3595
if revision_id is not None:
3596
source_ids = self.source.get_ancestry(revision_id)
3597
if source_ids[0] is not None:
3598
raise AssertionError()
3601
source_ids = self.source._all_possible_ids()
3602
source_ids_set = set(source_ids)
3603
# source_ids is the worst possible case we may need to pull.
3604
# now we want to filter source_ids against what we actually
3605
# have in target, but don't try to check for existence where we know
3606
# we do not have a revision as that would be pointless.
3607
target_ids = set(self.target._all_possible_ids())
3608
possibly_present_revisions = target_ids.intersection(source_ids_set)
3609
actually_present_revisions = set(
3610
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3611
required_revisions = source_ids_set.difference(actually_present_revisions)
3612
if revision_id is not None:
3613
# we used get_ancestry to determine source_ids then we are assured all
3614
# revisions referenced are present as they are installed in topological order.
3615
# and the tip revision was validated by get_ancestry.
3616
result_set = required_revisions
3618
# if we just grabbed the possibly available ids, then
3619
# we only have an estimate of whats available and need to validate
3620
# that against the revision records.
3622
self.source._eliminate_revisions_not_present(required_revisions))
3623
return self.source.revision_ids_to_search_result(result_set)
3626
class InterKnitRepo(InterSameDataRepository):
3627
"""Optimised code paths between Knit based repositories."""
3630
def _get_repo_format_to_test(self):
3631
from bzrlib.repofmt import knitrepo
3632
return knitrepo.RepositoryFormatKnit1()
3635
def is_compatible(source, target):
3636
"""Be compatible with known Knit formats.
3638
We don't test for the stores being of specific types because that
3639
could lead to confusing results, and there is no need to be
3642
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3644
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3645
isinstance(target._format, RepositoryFormatKnit))
3646
except AttributeError:
3648
return are_knits and InterRepository._same_model(source, target)
3651
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3652
"""See InterRepository.missing_revision_ids()."""
3653
if revision_id is not None:
3654
source_ids = self.source.get_ancestry(revision_id)
3655
if source_ids[0] is not None:
3656
raise AssertionError()
3659
source_ids = self.source.all_revision_ids()
3660
source_ids_set = set(source_ids)
3661
# source_ids is the worst possible case we may need to pull.
3662
# now we want to filter source_ids against what we actually
3663
# have in target, but don't try to check for existence where we know
3664
# we do not have a revision as that would be pointless.
3665
target_ids = set(self.target.all_revision_ids())
3666
possibly_present_revisions = target_ids.intersection(source_ids_set)
3667
actually_present_revisions = set(
3668
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3669
required_revisions = source_ids_set.difference(actually_present_revisions)
3670
if revision_id is not None:
3671
# we used get_ancestry to determine source_ids then we are assured all
3672
# revisions referenced are present as they are installed in topological order.
3673
# and the tip revision was validated by get_ancestry.
3674
result_set = required_revisions
3676
# if we just grabbed the possibly available ids, then
3677
# we only have an estimate of whats available and need to validate
3678
# that against the revision records.
3680
self.source._eliminate_revisions_not_present(required_revisions))
3681
return self.source.revision_ids_to_search_result(result_set)
3684
class InterDifferingSerializer(InterRepository):
3687
def _get_repo_format_to_test(self):
3691
def is_compatible(source, target):
3692
"""Be compatible with Knit2 source and Knit3 target"""
3693
# This is redundant with format.check_conversion_target(), however that
3694
# raises an exception, and we just want to say "False" as in we won't
3695
# support converting between these formats.
3696
if 'IDS_never' in debug.debug_flags:
3698
if source.supports_rich_root() and not target.supports_rich_root():
3700
if (source._format.supports_tree_reference
3701
and not target._format.supports_tree_reference):
3703
if target._fallback_repositories and target._format.supports_chks:
3704
# IDS doesn't know how to copy CHKs for the parent inventories it
3705
# adds to stacked repos.
3707
if 'IDS_always' in debug.debug_flags:
3709
# Only use this code path for local source and target. IDS does far
3710
# too much IO (both bandwidth and roundtrips) over a network.
3711
if not source.bzrdir.transport.base.startswith('file:///'):
3713
if not target.bzrdir.transport.base.startswith('file:///'):
3717
def _get_trees(self, revision_ids, cache):
3719
for rev_id in revision_ids:
3721
possible_trees.append((rev_id, cache[rev_id]))
3723
# Not cached, but inventory might be present anyway.
3725
tree = self.source.revision_tree(rev_id)
3726
except errors.NoSuchRevision:
3727
# Nope, parent is ghost.
3730
cache[rev_id] = tree
3731
possible_trees.append((rev_id, tree))
3732
return possible_trees
3734
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3735
"""Get the best delta and base for this revision.
3737
:return: (basis_id, delta)
3740
# Generate deltas against each tree, to find the shortest.
3741
texts_possibly_new_in_tree = set()
3742
for basis_id, basis_tree in possible_trees:
3743
delta = tree.inventory._make_delta(basis_tree.inventory)
3744
for old_path, new_path, file_id, new_entry in delta:
3745
if new_path is None:
3746
# This file_id isn't present in the new rev, so we don't
3750
# Rich roots are handled elsewhere...
3752
kind = new_entry.kind
3753
if kind != 'directory' and kind != 'file':
3754
# No text record associated with this inventory entry.
3756
# This is a directory or file that has changed somehow.
3757
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3758
deltas.append((len(delta), basis_id, delta))
3760
return deltas[0][1:]
3762
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3763
"""Find all parent revisions that are absent, but for which the
3764
inventory is present, and copy those inventories.
3766
This is necessary to preserve correctness when the source is stacked
3767
without fallbacks configured. (Note that in cases like upgrade the
3768
source may be not have _fallback_repositories even though it is
3772
for parents in parent_map.values():
3773
parent_revs.update(parents)
3774
present_parents = self.source.get_parent_map(parent_revs)
3775
absent_parents = set(parent_revs).difference(present_parents)
3776
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3777
(rev_id,) for rev_id in absent_parents)
3778
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3779
for parent_tree in self.source.revision_trees(parent_inv_ids):
3780
current_revision_id = parent_tree.get_revision_id()
3781
parents_parents_keys = parent_invs_keys_for_stacking[
3782
(current_revision_id,)]
3783
parents_parents = [key[-1] for key in parents_parents_keys]
3784
basis_id = _mod_revision.NULL_REVISION
3785
basis_tree = self.source.revision_tree(basis_id)
3786
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3787
self.target.add_inventory_by_delta(
3788
basis_id, delta, current_revision_id, parents_parents)
3789
cache[current_revision_id] = parent_tree
3791
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3792
"""Fetch across a few revisions.
3794
:param revision_ids: The revisions to copy
3795
:param basis_id: The revision_id of a tree that must be in cache, used
3796
as a basis for delta when no other base is available
3797
:param cache: A cache of RevisionTrees that we can use.
3798
:param a_graph: A Graph object to determine the heads() of the
3799
rich-root data stream.
3800
:return: The revision_id of the last converted tree. The RevisionTree
3801
for it will be in cache
3803
# Walk though all revisions; get inventory deltas, copy referenced
3804
# texts that delta references, insert the delta, revision and
3806
root_keys_to_create = set()
3809
pending_revisions = []
3810
parent_map = self.source.get_parent_map(revision_ids)
3811
self._fetch_parent_invs_for_stacking(parent_map, cache)
3812
self.source._safe_to_return_from_cache = True
3813
for tree in self.source.revision_trees(revision_ids):
3814
# Find a inventory delta for this revision.
3815
# Find text entries that need to be copied, too.
3816
current_revision_id = tree.get_revision_id()
3817
parent_ids = parent_map.get(current_revision_id, ())
3818
parent_trees = self._get_trees(parent_ids, cache)
3819
possible_trees = list(parent_trees)
3820
if len(possible_trees) == 0:
3821
# There either aren't any parents, or the parents are ghosts,
3822
# so just use the last converted tree.
3823
possible_trees.append((basis_id, cache[basis_id]))
3824
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3826
revision = self.source.get_revision(current_revision_id)
3827
pending_deltas.append((basis_id, delta,
3828
current_revision_id, revision.parent_ids))
3829
if self._converting_to_rich_root:
3830
self._revision_id_to_root_id[current_revision_id] = \
3832
# Determine which texts are in present in this revision but not in
3833
# any of the available parents.
3834
texts_possibly_new_in_tree = set()
3835
for old_path, new_path, file_id, entry in delta:
3836
if new_path is None:
3837
# This file_id isn't present in the new rev
3841
if not self.target.supports_rich_root():
3842
# The target doesn't support rich root, so we don't
3845
if self._converting_to_rich_root:
3846
# This can't be copied normally, we have to insert
3848
root_keys_to_create.add((file_id, entry.revision))
3851
texts_possibly_new_in_tree.add((file_id, entry.revision))
3852
for basis_id, basis_tree in possible_trees:
3853
basis_inv = basis_tree.inventory
3854
for file_key in list(texts_possibly_new_in_tree):
3855
file_id, file_revision = file_key
3857
entry = basis_inv[file_id]
3858
except errors.NoSuchId:
3860
if entry.revision == file_revision:
3861
texts_possibly_new_in_tree.remove(file_key)
3862
text_keys.update(texts_possibly_new_in_tree)
3863
pending_revisions.append(revision)
3864
cache[current_revision_id] = tree
3865
basis_id = current_revision_id
3866
self.source._safe_to_return_from_cache = False
3868
from_texts = self.source.texts
3869
to_texts = self.target.texts
3870
if root_keys_to_create:
3871
root_stream = _mod_fetch._new_root_data_stream(
3872
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3873
self.source, graph=a_graph)
3874
to_texts.insert_record_stream(root_stream)
3875
to_texts.insert_record_stream(from_texts.get_record_stream(
3876
text_keys, self.target._format._fetch_order,
3877
not self.target._format._fetch_uses_deltas))
3878
# insert inventory deltas
3879
for delta in pending_deltas:
3880
self.target.add_inventory_by_delta(*delta)
3881
if self.target._fallback_repositories:
3882
# Make sure this stacked repository has all the parent inventories
3883
# for the new revisions that we are about to insert. We do this
3884
# before adding the revisions so that no revision is added until
3885
# all the inventories it may depend on are added.
3886
# Note that this is overzealous, as we may have fetched these in an
3889
revision_ids = set()
3890
for revision in pending_revisions:
3891
revision_ids.add(revision.revision_id)
3892
parent_ids.update(revision.parent_ids)
3893
parent_ids.difference_update(revision_ids)
3894
parent_ids.discard(_mod_revision.NULL_REVISION)
3895
parent_map = self.source.get_parent_map(parent_ids)
3896
# we iterate over parent_map and not parent_ids because we don't
3897
# want to try copying any revision which is a ghost
3898
for parent_tree in self.source.revision_trees(parent_map):
3899
current_revision_id = parent_tree.get_revision_id()
3900
parents_parents = parent_map[current_revision_id]
3901
possible_trees = self._get_trees(parents_parents, cache)
3902
if len(possible_trees) == 0:
3903
# There either aren't any parents, or the parents are
3904
# ghosts, so just use the last converted tree.
3905
possible_trees.append((basis_id, cache[basis_id]))
3906
basis_id, delta = self._get_delta_for_revision(parent_tree,
3907
parents_parents, possible_trees)
3908
self.target.add_inventory_by_delta(
3909
basis_id, delta, current_revision_id, parents_parents)
3910
# insert signatures and revisions
3911
for revision in pending_revisions:
3913
signature = self.source.get_signature_text(
3914
revision.revision_id)
3915
self.target.add_signature_text(revision.revision_id,
3917
except errors.NoSuchRevision:
3919
self.target.add_revision(revision.revision_id, revision)
3922
def _fetch_all_revisions(self, revision_ids, pb):
3923
"""Fetch everything for the list of revisions.
3925
:param revision_ids: The list of revisions to fetch. Must be in
3927
:param pb: A ProgressTask
3930
basis_id, basis_tree = self._get_basis(revision_ids[0])
3932
cache = lru_cache.LRUCache(100)
3933
cache[basis_id] = basis_tree
3934
del basis_tree # We don't want to hang on to it here
3936
if self._converting_to_rich_root and len(revision_ids) > 100:
3937
a_graph = _mod_fetch._get_rich_root_heads_graph(self.source,
3942
for offset in range(0, len(revision_ids), batch_size):
3943
self.target.start_write_group()
3945
pb.update('Transferring revisions', offset,
3947
batch = revision_ids[offset:offset+batch_size]
3948
basis_id = self._fetch_batch(batch, basis_id, cache,
3951
self.source._safe_to_return_from_cache = False
3952
self.target.abort_write_group()
3955
hint = self.target.commit_write_group()
3958
if hints and self.target._format.pack_compresses:
3959
self.target.pack(hint=hints)
3960
pb.update('Transferring revisions', len(revision_ids),
3964
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3966
"""See InterRepository.fetch()."""
3967
if fetch_spec is not None:
3968
raise AssertionError("Not implemented yet...")
3969
# See <https://launchpad.net/bugs/456077> asking for a warning here
3971
# nb this is only active for local-local fetches; other things using
3973
ui.ui_factory.warn_cross_format_fetch(self.source._format,
3974
self.target._format)
3975
ui.ui_factory.warn_experimental_format_fetch(self)
3976
if (not self.source.supports_rich_root()
3977
and self.target.supports_rich_root()):
3978
self._converting_to_rich_root = True
3979
self._revision_id_to_root_id = {}
3981
self._converting_to_rich_root = False
3982
revision_ids = self.target.search_missing_revision_ids(self.source,
3983
revision_id, find_ghosts=find_ghosts).get_keys()
3984
if not revision_ids:
3986
revision_ids = tsort.topo_sort(
3987
self.source.get_graph().get_parent_map(revision_ids))
3988
if not revision_ids:
3990
# Walk though all revisions; get inventory deltas, copy referenced
3991
# texts that delta references, insert the delta, revision and
3994
my_pb = ui.ui_factory.nested_progress_bar()
3997
symbol_versioning.warn(
3998
symbol_versioning.deprecated_in((1, 14, 0))
3999
% "pb parameter to fetch()")
4002
self._fetch_all_revisions(revision_ids, pb)
4004
if my_pb is not None:
4006
return len(revision_ids), 0
4008
def _get_basis(self, first_revision_id):
4009
"""Get a revision and tree which exists in the target.
4011
This assumes that first_revision_id is selected for transmission
4012
because all other ancestors are already present. If we can't find an
4013
ancestor we fall back to NULL_REVISION since we know that is safe.
4015
:return: (basis_id, basis_tree)
4017
first_rev = self.source.get_revision(first_revision_id)
4019
basis_id = first_rev.parent_ids[0]
4020
# only valid as a basis if the target has it
4021
self.target.get_revision(basis_id)
4022
# Try to get a basis tree - if its a ghost it will hit the
4023
# NoSuchRevision case.
4024
basis_tree = self.source.revision_tree(basis_id)
4025
except (IndexError, errors.NoSuchRevision):
4026
basis_id = _mod_revision.NULL_REVISION
4027
basis_tree = self.source.revision_tree(basis_id)
4028
return basis_id, basis_tree
4031
InterRepository.register_optimiser(InterDifferingSerializer)
4032
InterRepository.register_optimiser(InterSameDataRepository)
4033
InterRepository.register_optimiser(InterWeaveRepo)
4034
InterRepository.register_optimiser(InterKnitRepo)
4037
class CopyConverter(object):
4038
"""A repository conversion tool which just performs a copy of the content.
4040
This is slow but quite reliable.
4043
def __init__(self, target_format):
4044
"""Create a CopyConverter.
4046
:param target_format: The format the resulting repository should be.
4048
self.target_format = target_format
4050
def convert(self, repo, pb):
4051
"""Perform the conversion of to_convert, giving feedback via pb.
4053
:param to_convert: The disk object to convert.
4054
:param pb: a progress bar to use for progress information.
4056
pb = ui.ui_factory.nested_progress_bar()
4059
# this is only useful with metadir layouts - separated repo content.
4060
# trigger an assertion if not such
4061
repo._format.get_format_string()
4062
self.repo_dir = repo.bzrdir
4063
pb.update('Moving repository to repository.backup')
4064
self.repo_dir.transport.move('repository', 'repository.backup')
4065
backup_transport = self.repo_dir.transport.clone('repository.backup')
4066
repo._format.check_conversion_target(self.target_format)
4067
self.source_repo = repo._format.open(self.repo_dir,
4069
_override_transport=backup_transport)
4070
pb.update('Creating new repository')
4071
converted = self.target_format.initialize(self.repo_dir,
4072
self.source_repo.is_shared())
4073
converted.lock_write()
4075
pb.update('Copying content')
4076
self.source_repo.copy_content_into(converted)
4079
pb.update('Deleting old repository content')
4080
self.repo_dir.transport.delete_tree('repository.backup')
4081
ui.ui_factory.note('repository converted')
4094
def _unescaper(match, _map=_unescape_map):
4095
code = match.group(1)
4099
if not code.startswith('#'):
4101
return unichr(int(code[1:])).encode('utf8')
4107
def _unescape_xml(data):
4108
"""Unescape predefined XML entities in a string of data."""
4110
if _unescape_re is None:
4111
_unescape_re = re.compile('\&([^;]*);')
4112
return _unescape_re.sub(_unescaper, data)
4115
class _VersionedFileChecker(object):
4117
def __init__(self, repository, text_key_references=None, ancestors=None):
4118
self.repository = repository
4119
self.text_index = self.repository._generate_text_key_index(
4120
text_key_references=text_key_references, ancestors=ancestors)
4122
def calculate_file_version_parents(self, text_key):
4123
"""Calculate the correct parents for a file version according to
4126
parent_keys = self.text_index[text_key]
4127
if parent_keys == [_mod_revision.NULL_REVISION]:
4129
return tuple(parent_keys)
4131
def check_file_version_parents(self, texts, progress_bar=None):
4132
"""Check the parents stored in a versioned file are correct.
4134
It also detects file versions that are not referenced by their
4135
corresponding revision's inventory.
4137
:returns: A tuple of (wrong_parents, dangling_file_versions).
4138
wrong_parents is a dict mapping {revision_id: (stored_parents,
4139
correct_parents)} for each revision_id where the stored parents
4140
are not correct. dangling_file_versions is a set of (file_id,
4141
revision_id) tuples for versions that are present in this versioned
4142
file, but not used by the corresponding inventory.
4144
local_progress = None
4145
if progress_bar is None:
4146
local_progress = ui.ui_factory.nested_progress_bar()
4147
progress_bar = local_progress
4149
return self._check_file_version_parents(texts, progress_bar)
4152
local_progress.finished()
4154
def _check_file_version_parents(self, texts, progress_bar):
4155
"""See check_file_version_parents."""
4157
self.file_ids = set([file_id for file_id, _ in
4158
self.text_index.iterkeys()])
4159
# text keys is now grouped by file_id
4160
n_versions = len(self.text_index)
4161
progress_bar.update('loading text store', 0, n_versions)
4162
parent_map = self.repository.texts.get_parent_map(self.text_index)
4163
# On unlistable transports this could well be empty/error...
4164
text_keys = self.repository.texts.keys()
4165
unused_keys = frozenset(text_keys) - set(self.text_index)
4166
for num, key in enumerate(self.text_index.iterkeys()):
4167
progress_bar.update('checking text graph', num, n_versions)
4168
correct_parents = self.calculate_file_version_parents(key)
4170
knit_parents = parent_map[key]
4171
except errors.RevisionNotPresent:
4174
if correct_parents != knit_parents:
4175
wrong_parents[key] = (knit_parents, correct_parents)
4176
return wrong_parents, unused_keys
4179
def _old_get_graph(repository, revision_id):
4180
"""DO NOT USE. That is all. I'm serious."""
4181
graph = repository.get_graph()
4182
revision_graph = dict(((key, value) for key, value in
4183
graph.iter_ancestry([revision_id]) if value is not None))
4184
return _strip_NULL_ghosts(revision_graph)
4187
def _strip_NULL_ghosts(revision_graph):
4188
"""Also don't use this. more compatibility code for unmigrated clients."""
4189
# Filter ghosts, and null:
4190
if _mod_revision.NULL_REVISION in revision_graph:
4191
del revision_graph[_mod_revision.NULL_REVISION]
4192
for key, parents in revision_graph.items():
4193
revision_graph[key] = tuple(parent for parent in parents if parent
4195
return revision_graph
4198
class StreamSink(object):
4199
"""An object that can insert a stream into a repository.
4201
This interface handles the complexity of reserialising inventories and
4202
revisions from different formats, and allows unidirectional insertion into
4203
stacked repositories without looking for the missing basis parents
4207
def __init__(self, target_repo):
4208
self.target_repo = target_repo
4210
def insert_stream(self, stream, src_format, resume_tokens):
4211
"""Insert a stream's content into the target repository.
4213
:param src_format: a bzr repository format.
4215
:return: a list of resume tokens and an iterable of keys additional
4216
items required before the insertion can be completed.
4218
self.target_repo.lock_write()
4221
self.target_repo.resume_write_group(resume_tokens)
4224
self.target_repo.start_write_group()
4227
# locked_insert_stream performs a commit|suspend.
4228
return self._locked_insert_stream(stream, src_format, is_resume)
4230
self.target_repo.abort_write_group(suppress_errors=True)
4233
self.target_repo.unlock()
4235
def _locked_insert_stream(self, stream, src_format, is_resume):
4236
to_serializer = self.target_repo._format._serializer
4237
src_serializer = src_format._serializer
4239
if to_serializer == src_serializer:
4240
# If serializers match and the target is a pack repository, set the
4241
# write cache size on the new pack. This avoids poor performance
4242
# on transports where append is unbuffered (such as
4243
# RemoteTransport). This is safe to do because nothing should read
4244
# back from the target repository while a stream with matching
4245
# serialization is being inserted.
4246
# The exception is that a delta record from the source that should
4247
# be a fulltext may need to be expanded by the target (see
4248
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4249
# explicitly flush any buffered writes first in that rare case.
4251
new_pack = self.target_repo._pack_collection._new_pack
4252
except AttributeError:
4253
# Not a pack repository
4256
new_pack.set_write_cache_size(1024*1024)
4257
for substream_type, substream in stream:
4258
if 'stream' in debug.debug_flags:
4259
mutter('inserting substream: %s', substream_type)
4260
if substream_type == 'texts':
4261
self.target_repo.texts.insert_record_stream(substream)
4262
elif substream_type == 'inventories':
4263
if src_serializer == to_serializer:
4264
self.target_repo.inventories.insert_record_stream(
4267
self._extract_and_insert_inventories(
4268
substream, src_serializer)
4269
elif substream_type == 'inventory-deltas':
4270
ui.ui_factory.warn_cross_format_fetch(src_format,
4271
self.target_repo._format)
4272
self._extract_and_insert_inventory_deltas(
4273
substream, src_serializer)
4274
elif substream_type == 'chk_bytes':
4275
# XXX: This doesn't support conversions, as it assumes the
4276
# conversion was done in the fetch code.
4277
self.target_repo.chk_bytes.insert_record_stream(substream)
4278
elif substream_type == 'revisions':
4279
# This may fallback to extract-and-insert more often than
4280
# required if the serializers are different only in terms of
4282
if src_serializer == to_serializer:
4283
self.target_repo.revisions.insert_record_stream(
4286
self._extract_and_insert_revisions(substream,
4288
elif substream_type == 'signatures':
4289
self.target_repo.signatures.insert_record_stream(substream)
4291
raise AssertionError('kaboom! %s' % (substream_type,))
4292
# Done inserting data, and the missing_keys calculations will try to
4293
# read back from the inserted data, so flush the writes to the new pack
4294
# (if this is pack format).
4295
if new_pack is not None:
4296
new_pack._write_data('', flush=True)
4297
# Find all the new revisions (including ones from resume_tokens)
4298
missing_keys = self.target_repo.get_missing_parent_inventories(
4299
check_for_missing_texts=is_resume)
4301
for prefix, versioned_file in (
4302
('texts', self.target_repo.texts),
4303
('inventories', self.target_repo.inventories),
4304
('revisions', self.target_repo.revisions),
4305
('signatures', self.target_repo.signatures),
4306
('chk_bytes', self.target_repo.chk_bytes),
4308
if versioned_file is None:
4310
# TODO: key is often going to be a StaticTuple object
4311
# I don't believe we can define a method by which
4312
# (prefix,) + StaticTuple will work, though we could
4313
# define a StaticTuple.sq_concat that would allow you to
4314
# pass in either a tuple or a StaticTuple as the second
4315
# object, so instead we could have:
4316
# StaticTuple(prefix) + key here...
4317
missing_keys.update((prefix,) + key for key in
4318
versioned_file.get_missing_compression_parent_keys())
4319
except NotImplementedError:
4320
# cannot even attempt suspending, and missing would have failed
4321
# during stream insertion.
4322
missing_keys = set()
4325
# suspend the write group and tell the caller what we is
4326
# missing. We know we can suspend or else we would not have
4327
# entered this code path. (All repositories that can handle
4328
# missing keys can handle suspending a write group).
4329
write_group_tokens = self.target_repo.suspend_write_group()
4330
return write_group_tokens, missing_keys
4331
hint = self.target_repo.commit_write_group()
4332
if (to_serializer != src_serializer and
4333
self.target_repo._format.pack_compresses):
4334
self.target_repo.pack(hint=hint)
4337
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4338
target_rich_root = self.target_repo._format.rich_root_data
4339
target_tree_refs = self.target_repo._format.supports_tree_reference
4340
for record in substream:
4341
# Insert the delta directly
4342
inventory_delta_bytes = record.get_bytes_as('fulltext')
4343
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4345
parse_result = deserialiser.parse_text_bytes(
4346
inventory_delta_bytes)
4347
except inventory_delta.IncompatibleInventoryDelta, err:
4348
trace.mutter("Incompatible delta: %s", err.msg)
4349
raise errors.IncompatibleRevision(self.target_repo._format)
4350
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4351
revision_id = new_id
4352
parents = [key[0] for key in record.parents]
4353
self.target_repo.add_inventory_by_delta(
4354
basis_id, inv_delta, revision_id, parents)
4356
def _extract_and_insert_inventories(self, substream, serializer,
4358
"""Generate a new inventory versionedfile in target, converting data.
4360
The inventory is retrieved from the source, (deserializing it), and
4361
stored in the target (reserializing it in a different format).
4363
target_rich_root = self.target_repo._format.rich_root_data
4364
target_tree_refs = self.target_repo._format.supports_tree_reference
4365
for record in substream:
4366
# It's not a delta, so it must be a fulltext in the source
4367
# serializer's format.
4368
bytes = record.get_bytes_as('fulltext')
4369
revision_id = record.key[0]
4370
inv = serializer.read_inventory_from_string(bytes, revision_id)
4371
parents = [key[0] for key in record.parents]
4372
self.target_repo.add_inventory(revision_id, inv, parents)
4373
# No need to keep holding this full inv in memory when the rest of
4374
# the substream is likely to be all deltas.
4377
def _extract_and_insert_revisions(self, substream, serializer):
4378
for record in substream:
4379
bytes = record.get_bytes_as('fulltext')
4380
revision_id = record.key[0]
4381
rev = serializer.read_revision_from_string(bytes)
4382
if rev.revision_id != revision_id:
4383
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4384
self.target_repo.add_revision(revision_id, rev)
4387
if self.target_repo._format._fetch_reconcile:
4388
self.target_repo.reconcile()
4391
class StreamSource(object):
4392
"""A source of a stream for fetching between repositories."""
4394
def __init__(self, from_repository, to_format):
4395
"""Create a StreamSource streaming from from_repository."""
4396
self.from_repository = from_repository
4397
self.to_format = to_format
4399
def delta_on_metadata(self):
4400
"""Return True if delta's are permitted on metadata streams.
4402
That is on revisions and signatures.
4404
src_serializer = self.from_repository._format._serializer
4405
target_serializer = self.to_format._serializer
4406
return (self.to_format._fetch_uses_deltas and
4407
src_serializer == target_serializer)
4409
def _fetch_revision_texts(self, revs):
4410
# fetch signatures first and then the revision texts
4411
# may need to be a InterRevisionStore call here.
4412
from_sf = self.from_repository.signatures
4413
# A missing signature is just skipped.
4414
keys = [(rev_id,) for rev_id in revs]
4415
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4417
self.to_format._fetch_order,
4418
not self.to_format._fetch_uses_deltas))
4419
# If a revision has a delta, this is actually expanded inside the
4420
# insert_record_stream code now, which is an alternate fix for
4422
from_rf = self.from_repository.revisions
4423
revisions = from_rf.get_record_stream(
4425
self.to_format._fetch_order,
4426
not self.delta_on_metadata())
4427
return [('signatures', signatures), ('revisions', revisions)]
4429
def _generate_root_texts(self, revs):
4430
"""This will be called by get_stream between fetching weave texts and
4431
fetching the inventory weave.
4433
if self._rich_root_upgrade():
4434
return _mod_fetch.Inter1and2Helper(
4435
self.from_repository).generate_root_texts(revs)
4439
def get_stream(self, search):
4441
revs = search.get_keys()
4442
graph = self.from_repository.get_graph()
4443
revs = tsort.topo_sort(graph.get_parent_map(revs))
4444
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4446
for knit_kind, file_id, revisions in data_to_fetch:
4447
if knit_kind != phase:
4449
# Make a new progress bar for this phase
4450
if knit_kind == "file":
4451
# Accumulate file texts
4452
text_keys.extend([(file_id, revision) for revision in
4454
elif knit_kind == "inventory":
4455
# Now copy the file texts.
4456
from_texts = self.from_repository.texts
4457
yield ('texts', from_texts.get_record_stream(
4458
text_keys, self.to_format._fetch_order,
4459
not self.to_format._fetch_uses_deltas))
4460
# Cause an error if a text occurs after we have done the
4463
# Before we process the inventory we generate the root
4464
# texts (if necessary) so that the inventories references
4466
for _ in self._generate_root_texts(revs):
4468
# we fetch only the referenced inventories because we do not
4469
# know for unselected inventories whether all their required
4470
# texts are present in the other repository - it could be
4472
for info in self._get_inventory_stream(revs):
4474
elif knit_kind == "signatures":
4475
# Nothing to do here; this will be taken care of when
4476
# _fetch_revision_texts happens.
4478
elif knit_kind == "revisions":
4479
for record in self._fetch_revision_texts(revs):
4482
raise AssertionError("Unknown knit kind %r" % knit_kind)
4484
def get_stream_for_missing_keys(self, missing_keys):
4485
# missing keys can only occur when we are byte copying and not
4486
# translating (because translation means we don't send
4487
# unreconstructable deltas ever).
4489
keys['texts'] = set()
4490
keys['revisions'] = set()
4491
keys['inventories'] = set()
4492
keys['chk_bytes'] = set()
4493
keys['signatures'] = set()
4494
for key in missing_keys:
4495
keys[key[0]].add(key[1:])
4496
if len(keys['revisions']):
4497
# If we allowed copying revisions at this point, we could end up
4498
# copying a revision without copying its required texts: a
4499
# violation of the requirements for repository integrity.
4500
raise AssertionError(
4501
'cannot copy revisions to fill in missing deltas %s' % (
4502
keys['revisions'],))
4503
for substream_kind, keys in keys.iteritems():
4504
vf = getattr(self.from_repository, substream_kind)
4505
if vf is None and keys:
4506
raise AssertionError(
4507
"cannot fill in keys for a versioned file we don't"
4508
" have: %s needs %s" % (substream_kind, keys))
4510
# No need to stream something we don't have
4512
if substream_kind == 'inventories':
4513
# Some missing keys are genuinely ghosts, filter those out.
4514
present = self.from_repository.inventories.get_parent_map(keys)
4515
revs = [key[0] for key in present]
4516
# Get the inventory stream more-or-less as we do for the
4517
# original stream; there's no reason to assume that records
4518
# direct from the source will be suitable for the sink. (Think
4519
# e.g. 2a -> 1.9-rich-root).
4520
for info in self._get_inventory_stream(revs, missing=True):
4524
# Ask for full texts always so that we don't need more round trips
4525
# after this stream.
4526
# Some of the missing keys are genuinely ghosts, so filter absent
4527
# records. The Sink is responsible for doing another check to
4528
# ensure that ghosts don't introduce missing data for future
4530
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4531
self.to_format._fetch_order, True))
4532
yield substream_kind, stream
4534
def inventory_fetch_order(self):
4535
if self._rich_root_upgrade():
4536
return 'topological'
4538
return self.to_format._fetch_order
4540
def _rich_root_upgrade(self):
4541
return (not self.from_repository._format.rich_root_data and
4542
self.to_format.rich_root_data)
4544
def _get_inventory_stream(self, revision_ids, missing=False):
4545
from_format = self.from_repository._format
4546
if (from_format.supports_chks and self.to_format.supports_chks and
4547
from_format.network_name() == self.to_format.network_name()):
4548
raise AssertionError(
4549
"this case should be handled by GroupCHKStreamSource")
4550
elif 'forceinvdeltas' in debug.debug_flags:
4551
return self._get_convertable_inventory_stream(revision_ids,
4552
delta_versus_null=missing)
4553
elif from_format.network_name() == self.to_format.network_name():
4555
return self._get_simple_inventory_stream(revision_ids,
4557
elif (not from_format.supports_chks and not self.to_format.supports_chks
4558
and from_format._serializer == self.to_format._serializer):
4559
# Essentially the same format.
4560
return self._get_simple_inventory_stream(revision_ids,
4563
# Any time we switch serializations, we want to use an
4564
# inventory-delta based approach.
4565
return self._get_convertable_inventory_stream(revision_ids,
4566
delta_versus_null=missing)
4568
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4569
# NB: This currently reopens the inventory weave in source;
4570
# using a single stream interface instead would avoid this.
4571
from_weave = self.from_repository.inventories
4573
delta_closure = True
4575
delta_closure = not self.delta_on_metadata()
4576
yield ('inventories', from_weave.get_record_stream(
4577
[(rev_id,) for rev_id in revision_ids],
4578
self.inventory_fetch_order(), delta_closure))
4580
def _get_convertable_inventory_stream(self, revision_ids,
4581
delta_versus_null=False):
4582
# The two formats are sufficiently different that there is no fast
4583
# path, so we need to send just inventorydeltas, which any
4584
# sufficiently modern client can insert into any repository.
4585
# The StreamSink code expects to be able to
4586
# convert on the target, so we need to put bytes-on-the-wire that can
4587
# be converted. That means inventory deltas (if the remote is <1.19,
4588
# RemoteStreamSink will fallback to VFS to insert the deltas).
4589
yield ('inventory-deltas',
4590
self._stream_invs_as_deltas(revision_ids,
4591
delta_versus_null=delta_versus_null))
4593
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4594
"""Return a stream of inventory-deltas for the given rev ids.
4596
:param revision_ids: The list of inventories to transmit
4597
:param delta_versus_null: Don't try to find a minimal delta for this
4598
entry, instead compute the delta versus the NULL_REVISION. This
4599
effectively streams a complete inventory. Used for stuff like
4600
filling in missing parents, etc.
4602
from_repo = self.from_repository
4603
revision_keys = [(rev_id,) for rev_id in revision_ids]
4604
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4605
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4607
inventories = self.from_repository.iter_inventories(
4608
revision_ids, 'topological')
4609
format = from_repo._format
4610
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4611
inventory_cache = lru_cache.LRUCache(50)
4612
null_inventory = from_repo.revision_tree(
4613
_mod_revision.NULL_REVISION).inventory
4614
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4615
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4616
# repo back into a non-rich-root repo ought to be allowed)
4617
serializer = inventory_delta.InventoryDeltaSerializer(
4618
versioned_root=format.rich_root_data,
4619
tree_references=format.supports_tree_reference)
4620
for inv in inventories:
4621
key = (inv.revision_id,)
4622
parent_keys = parent_map.get(key, ())
4624
if not delta_versus_null and parent_keys:
4625
# The caller did not ask for complete inventories and we have
4626
# some parents that we can delta against. Make a delta against
4627
# each parent so that we can find the smallest.
4628
parent_ids = [parent_key[0] for parent_key in parent_keys]
4629
for parent_id in parent_ids:
4630
if parent_id not in invs_sent_so_far:
4631
# We don't know that the remote side has this basis, so
4634
if parent_id == _mod_revision.NULL_REVISION:
4635
parent_inv = null_inventory
4637
parent_inv = inventory_cache.get(parent_id, None)
4638
if parent_inv is None:
4639
parent_inv = from_repo.get_inventory(parent_id)
4640
candidate_delta = inv._make_delta(parent_inv)
4641
if (delta is None or
4642
len(delta) > len(candidate_delta)):
4643
delta = candidate_delta
4644
basis_id = parent_id
4646
# Either none of the parents ended up being suitable, or we
4647
# were asked to delta against NULL
4648
basis_id = _mod_revision.NULL_REVISION
4649
delta = inv._make_delta(null_inventory)
4650
invs_sent_so_far.add(inv.revision_id)
4651
inventory_cache[inv.revision_id] = inv
4652
delta_serialized = ''.join(
4653
serializer.delta_to_lines(basis_id, key[-1], delta))
4654
yield versionedfile.FulltextContentFactory(
4655
key, parent_keys, None, delta_serialized)
4658
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4659
stop_revision=None):
4660
"""Extend the partial history to include a given index
4662
If a stop_index is supplied, stop when that index has been reached.
4663
If a stop_revision is supplied, stop when that revision is
4664
encountered. Otherwise, stop when the beginning of history is
4667
:param stop_index: The index which should be present. When it is
4668
present, history extension will stop.
4669
:param stop_revision: The revision id which should be present. When
4670
it is encountered, history extension will stop.
4672
start_revision = partial_history_cache[-1]
4673
iterator = repo.iter_reverse_revision_history(start_revision)
4675
#skip the last revision in the list
4678
if (stop_index is not None and
4679
len(partial_history_cache) > stop_index):
4681
if partial_history_cache[-1] == stop_revision:
4683
revision_id = iterator.next()
4684
partial_history_cache.append(revision_id)
4685
except StopIteration:
added
removed
bzrlib/repository.py
