1
# Copyright (C) 2005, 2006, 2007, 2008, 2009 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
from bzrlib.lazy_import import lazy_import
18
lazy_import(globals(), """
37
revision as _mod_revision,
43
from bzrlib.bundle import serializer
44
from bzrlib.revisiontree import RevisionTree
45
from bzrlib.store.versioned import VersionedFileStore
46
from bzrlib.testament import Testament
49
from bzrlib.decorators import needs_read_lock, needs_write_lock
50
from bzrlib.inter import InterObject
51
from bzrlib.inventory import (
57
from bzrlib import registry
58
from bzrlib.symbol_versioning import (
61
from bzrlib.trace import (
62
log_exception_quietly, note, mutter, mutter_callsite, warning)
65
# Old formats display a warning, but only once
66
_deprecation_warning_done = False
69
class CommitBuilder(object):
70
"""Provides an interface to build up a commit.
72
This allows describing a tree to be committed without needing to
73
know the internals of the format of the repository.
76
# all clients should supply tree roots.
77
record_root_entry = True
78
# the default CommitBuilder does not manage trees whose root is versioned.
79
_versioned_root = False
81
def __init__(self, repository, parents, config, timestamp=None,
82
timezone=None, committer=None, revprops=None,
84
"""Initiate a CommitBuilder.
86
:param repository: Repository to commit to.
87
:param parents: Revision ids of the parents of the new revision.
88
:param config: Configuration to use.
89
:param timestamp: Optional timestamp recorded for commit.
90
:param timezone: Optional timezone for timestamp.
91
:param committer: Optional committer to set for commit.
92
:param revprops: Optional dictionary of revision properties.
93
:param revision_id: Optional revision id.
98
self._committer = self._config.username()
100
self._committer = committer
102
self.new_inventory = Inventory(None)
103
self._new_revision_id = revision_id
104
self.parents = parents
105
self.repository = repository
108
if revprops is not None:
109
self._validate_revprops(revprops)
110
self._revprops.update(revprops)
112
if timestamp is None:
113
timestamp = time.time()
114
# Restrict resolution to 1ms
115
self._timestamp = round(timestamp, 3)
118
self._timezone = osutils.local_time_offset()
120
self._timezone = int(timezone)
122
self._generate_revision_if_needed()
123
self.__heads = graph.HeadsCache(repository.get_graph()).heads
124
self._basis_delta = []
125
# API compatibility, older code that used CommitBuilder did not call
126
# .record_delete(), which means the delta that is computed would not be
127
# valid. Callers that will call record_delete() should call
128
# .will_record_deletes() to indicate that.
129
self._recording_deletes = False
130
# memo'd check for no-op commits.
131
self._any_changes = False
133
def any_changes(self):
134
"""Return True if any entries were changed.
136
This includes merge-only changes. It is the core for the --unchanged
139
:return: True if any changes have occured.
141
return self._any_changes
143
def _validate_unicode_text(self, text, context):
144
"""Verify things like commit messages don't have bogus characters."""
146
raise ValueError('Invalid value for %s: %r' % (context, text))
148
def _validate_revprops(self, revprops):
149
for key, value in revprops.iteritems():
150
# We know that the XML serializers do not round trip '\r'
151
# correctly, so refuse to accept them
152
if not isinstance(value, basestring):
153
raise ValueError('revision property (%s) is not a valid'
154
' (unicode) string: %r' % (key, value))
155
self._validate_unicode_text(value,
156
'revision property (%s)' % (key,))
158
def commit(self, message):
159
"""Make the actual commit.
161
:return: The revision id of the recorded revision.
163
self._validate_unicode_text(message, 'commit message')
164
rev = _mod_revision.Revision(
165
timestamp=self._timestamp,
166
timezone=self._timezone,
167
committer=self._committer,
169
inventory_sha1=self.inv_sha1,
170
revision_id=self._new_revision_id,
171
properties=self._revprops)
172
rev.parent_ids = self.parents
173
self.repository.add_revision(self._new_revision_id, rev,
174
self.new_inventory, self._config)
175
self.repository.commit_write_group()
176
return self._new_revision_id
179
"""Abort the commit that is being built.
181
self.repository.abort_write_group()
183
def revision_tree(self):
184
"""Return the tree that was just committed.
186
After calling commit() this can be called to get a RevisionTree
187
representing the newly committed tree. This is preferred to
188
calling Repository.revision_tree() because that may require
189
deserializing the inventory, while we already have a copy in
192
if self.new_inventory is None:
193
self.new_inventory = self.repository.get_inventory(
194
self._new_revision_id)
195
return RevisionTree(self.repository, self.new_inventory,
196
self._new_revision_id)
198
def finish_inventory(self):
199
"""Tell the builder that the inventory is finished.
201
:return: The inventory id in the repository, which can be used with
202
repository.get_inventory.
204
if self.new_inventory is None:
205
# an inventory delta was accumulated without creating a new
207
basis_id = self.basis_delta_revision
208
self.inv_sha1 = self.repository.add_inventory_by_delta(
209
basis_id, self._basis_delta, self._new_revision_id,
212
if self.new_inventory.root is None:
213
raise AssertionError('Root entry should be supplied to'
214
' record_entry_contents, as of bzr 0.10.')
215
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
216
self.new_inventory.revision_id = self._new_revision_id
217
self.inv_sha1 = self.repository.add_inventory(
218
self._new_revision_id,
222
return self._new_revision_id
224
def _gen_revision_id(self):
225
"""Return new revision-id."""
226
return generate_ids.gen_revision_id(self._config.username(),
229
def _generate_revision_if_needed(self):
230
"""Create a revision id if None was supplied.
232
If the repository can not support user-specified revision ids
233
they should override this function and raise CannotSetRevisionId
234
if _new_revision_id is not None.
236
:raises: CannotSetRevisionId
238
if self._new_revision_id is None:
239
self._new_revision_id = self._gen_revision_id()
240
self.random_revid = True
242
self.random_revid = False
244
def _heads(self, file_id, revision_ids):
245
"""Calculate the graph heads for revision_ids in the graph of file_id.
247
This can use either a per-file graph or a global revision graph as we
248
have an identity relationship between the two graphs.
250
return self.__heads(revision_ids)
252
def _check_root(self, ie, parent_invs, tree):
253
"""Helper for record_entry_contents.
255
:param ie: An entry being added.
256
:param parent_invs: The inventories of the parent revisions of the
258
:param tree: The tree that is being committed.
260
# In this revision format, root entries have no knit or weave When
261
# serializing out to disk and back in root.revision is always
263
ie.revision = self._new_revision_id
265
def _require_root_change(self, tree):
266
"""Enforce an appropriate root object change.
268
This is called once when record_iter_changes is called, if and only if
269
the root was not in the delta calculated by record_iter_changes.
271
:param tree: The tree which is being committed.
273
# NB: if there are no parents then this method is not called, so no
274
# need to guard on parents having length.
275
entry = entry_factory['directory'](tree.path2id(''), '',
277
entry.revision = self._new_revision_id
278
self._basis_delta.append(('', '', entry.file_id, entry))
280
def _get_delta(self, ie, basis_inv, path):
281
"""Get a delta against the basis inventory for ie."""
282
if ie.file_id not in basis_inv:
284
result = (None, path, ie.file_id, ie)
285
self._basis_delta.append(result)
287
elif ie != basis_inv[ie.file_id]:
289
# TODO: avoid tis id2path call.
290
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
291
self._basis_delta.append(result)
297
def get_basis_delta(self):
298
"""Return the complete inventory delta versus the basis inventory.
300
This has been built up with the calls to record_delete and
301
record_entry_contents. The client must have already called
302
will_record_deletes() to indicate that they will be generating a
305
:return: An inventory delta, suitable for use with apply_delta, or
306
Repository.add_inventory_by_delta, etc.
308
if not self._recording_deletes:
309
raise AssertionError("recording deletes not activated.")
310
return self._basis_delta
312
def record_delete(self, path, file_id):
313
"""Record that a delete occured against a basis tree.
315
This is an optional API - when used it adds items to the basis_delta
316
being accumulated by the commit builder. It cannot be called unless the
317
method will_record_deletes() has been called to inform the builder that
318
a delta is being supplied.
320
:param path: The path of the thing deleted.
321
:param file_id: The file id that was deleted.
323
if not self._recording_deletes:
324
raise AssertionError("recording deletes not activated.")
325
delta = (path, None, file_id, None)
326
self._basis_delta.append(delta)
327
self._any_changes = True
330
def will_record_deletes(self):
331
"""Tell the commit builder that deletes are being notified.
333
This enables the accumulation of an inventory delta; for the resulting
334
commit to be valid, deletes against the basis MUST be recorded via
335
builder.record_delete().
337
self._recording_deletes = True
339
basis_id = self.parents[0]
341
basis_id = _mod_revision.NULL_REVISION
342
self.basis_delta_revision = basis_id
344
def record_entry_contents(self, ie, parent_invs, path, tree,
346
"""Record the content of ie from tree into the commit if needed.
348
Side effect: sets ie.revision when unchanged
350
:param ie: An inventory entry present in the commit.
351
:param parent_invs: The inventories of the parent revisions of the
353
:param path: The path the entry is at in the tree.
354
:param tree: The tree which contains this entry and should be used to
356
:param content_summary: Summary data from the tree about the paths
357
content - stat, length, exec, sha/link target. This is only
358
accessed when the entry has a revision of None - that is when it is
359
a candidate to commit.
360
:return: A tuple (change_delta, version_recorded, fs_hash).
361
change_delta is an inventory_delta change for this entry against
362
the basis tree of the commit, or None if no change occured against
364
version_recorded is True if a new version of the entry has been
365
recorded. For instance, committing a merge where a file was only
366
changed on the other side will return (delta, False).
367
fs_hash is either None, or the hash details for the path (currently
368
a tuple of the contents sha1 and the statvalue returned by
369
tree.get_file_with_stat()).
371
if self.new_inventory.root is None:
372
if ie.parent_id is not None:
373
raise errors.RootMissing()
374
self._check_root(ie, parent_invs, tree)
375
if ie.revision is None:
376
kind = content_summary[0]
378
# ie is carried over from a prior commit
380
# XXX: repository specific check for nested tree support goes here - if
381
# the repo doesn't want nested trees we skip it ?
382
if (kind == 'tree-reference' and
383
not self.repository._format.supports_tree_reference):
384
# mismatch between commit builder logic and repository:
385
# this needs the entry creation pushed down into the builder.
386
raise NotImplementedError('Missing repository subtree support.')
387
self.new_inventory.add(ie)
389
# TODO: slow, take it out of the inner loop.
391
basis_inv = parent_invs[0]
393
basis_inv = Inventory(root_id=None)
395
# ie.revision is always None if the InventoryEntry is considered
396
# for committing. We may record the previous parents revision if the
397
# content is actually unchanged against a sole head.
398
if ie.revision is not None:
399
if not self._versioned_root and path == '':
400
# repositories that do not version the root set the root's
401
# revision to the new commit even when no change occurs (more
402
# specifically, they do not record a revision on the root; and
403
# the rev id is assigned to the root during deserialisation -
404
# this masks when a change may have occurred against the basis.
405
# To match this we always issue a delta, because the revision
406
# of the root will always be changing.
407
if ie.file_id in basis_inv:
408
delta = (basis_inv.id2path(ie.file_id), path,
412
delta = (None, path, ie.file_id, ie)
413
self._basis_delta.append(delta)
414
return delta, False, None
416
# we don't need to commit this, because the caller already
417
# determined that an existing revision of this file is
418
# appropriate. If its not being considered for committing then
419
# it and all its parents to the root must be unaltered so
420
# no-change against the basis.
421
if ie.revision == self._new_revision_id:
422
raise AssertionError("Impossible situation, a skipped "
423
"inventory entry (%r) claims to be modified in this "
424
"commit (%r).", (ie, self._new_revision_id))
425
return None, False, None
426
# XXX: Friction: parent_candidates should return a list not a dict
427
# so that we don't have to walk the inventories again.
428
parent_candiate_entries = ie.parent_candidates(parent_invs)
429
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
431
for inv in parent_invs:
432
if ie.file_id in inv:
433
old_rev = inv[ie.file_id].revision
434
if old_rev in head_set:
435
heads.append(inv[ie.file_id].revision)
436
head_set.remove(inv[ie.file_id].revision)
439
# now we check to see if we need to write a new record to the
441
# We write a new entry unless there is one head to the ancestors, and
442
# the kind-derived content is unchanged.
444
# Cheapest check first: no ancestors, or more the one head in the
445
# ancestors, we write a new node.
449
# There is a single head, look it up for comparison
450
parent_entry = parent_candiate_entries[heads[0]]
451
# if the non-content specific data has changed, we'll be writing a
453
if (parent_entry.parent_id != ie.parent_id or
454
parent_entry.name != ie.name):
456
# now we need to do content specific checks:
458
# if the kind changed the content obviously has
459
if kind != parent_entry.kind:
461
# Stat cache fingerprint feedback for the caller - None as we usually
462
# don't generate one.
465
if content_summary[2] is None:
466
raise ValueError("Files must not have executable = None")
468
if (# if the file length changed we have to store:
469
parent_entry.text_size != content_summary[1] or
470
# if the exec bit has changed we have to store:
471
parent_entry.executable != content_summary[2]):
473
elif parent_entry.text_sha1 == content_summary[3]:
474
# all meta and content is unchanged (using a hash cache
475
# hit to check the sha)
476
ie.revision = parent_entry.revision
477
ie.text_size = parent_entry.text_size
478
ie.text_sha1 = parent_entry.text_sha1
479
ie.executable = parent_entry.executable
480
return self._get_delta(ie, basis_inv, path), False, None
482
# Either there is only a hash change(no hash cache entry,
483
# or same size content change), or there is no change on
485
# Provide the parent's hash to the store layer, so that the
486
# content is unchanged we will not store a new node.
487
nostore_sha = parent_entry.text_sha1
489
# We want to record a new node regardless of the presence or
490
# absence of a content change in the file.
492
ie.executable = content_summary[2]
493
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
495
lines = file_obj.readlines()
499
ie.text_sha1, ie.text_size = self._add_text_to_weave(
500
ie.file_id, lines, heads, nostore_sha)
501
# Let the caller know we generated a stat fingerprint.
502
fingerprint = (ie.text_sha1, stat_value)
503
except errors.ExistingContent:
504
# Turns out that the file content was unchanged, and we were
505
# only going to store a new node if it was changed. Carry over
507
ie.revision = parent_entry.revision
508
ie.text_size = parent_entry.text_size
509
ie.text_sha1 = parent_entry.text_sha1
510
ie.executable = parent_entry.executable
511
return self._get_delta(ie, basis_inv, path), False, None
512
elif kind == 'directory':
514
# all data is meta here, nothing specific to directory, so
516
ie.revision = parent_entry.revision
517
return self._get_delta(ie, basis_inv, path), False, None
519
self._add_text_to_weave(ie.file_id, lines, heads, None)
520
elif kind == 'symlink':
521
current_link_target = content_summary[3]
523
# symlink target is not generic metadata, check if it has
525
if current_link_target != parent_entry.symlink_target:
528
# unchanged, carry over.
529
ie.revision = parent_entry.revision
530
ie.symlink_target = parent_entry.symlink_target
531
return self._get_delta(ie, basis_inv, path), False, None
532
ie.symlink_target = current_link_target
534
self._add_text_to_weave(ie.file_id, lines, heads, None)
535
elif kind == 'tree-reference':
537
if content_summary[3] != parent_entry.reference_revision:
540
# unchanged, carry over.
541
ie.reference_revision = parent_entry.reference_revision
542
ie.revision = parent_entry.revision
543
return self._get_delta(ie, basis_inv, path), False, None
544
ie.reference_revision = content_summary[3]
546
self._add_text_to_weave(ie.file_id, lines, heads, None)
548
raise NotImplementedError('unknown kind')
549
ie.revision = self._new_revision_id
550
self._any_changes = True
551
return self._get_delta(ie, basis_inv, path), True, fingerprint
553
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
554
_entry_factory=entry_factory):
555
"""Record a new tree via iter_changes.
557
:param tree: The tree to obtain text contents from for changed objects.
558
:param basis_revision_id: The revision id of the tree the iter_changes
559
has been generated against. Currently assumed to be the same
560
as self.parents[0] - if it is not, errors may occur.
561
:param iter_changes: An iter_changes iterator with the changes to apply
562
to basis_revision_id. The iterator must not include any items with
563
a current kind of None - missing items must be either filtered out
564
or errored-on beefore record_iter_changes sees the item.
565
:param _entry_factory: Private method to bind entry_factory locally for
567
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
570
# Create an inventory delta based on deltas between all the parents and
571
# deltas between all the parent inventories. We use inventory delta's
572
# between the inventory objects because iter_changes masks
573
# last-changed-field only changes.
575
# file_id -> change map, change is fileid, paths, changed, versioneds,
576
# parents, names, kinds, executables
578
# {file_id -> revision_id -> inventory entry, for entries in parent
579
# trees that are not parents[0]
583
revtrees = list(self.repository.revision_trees(self.parents))
584
except errors.NoSuchRevision:
585
# one or more ghosts, slow path.
587
for revision_id in self.parents:
589
revtrees.append(self.repository.revision_tree(revision_id))
590
except errors.NoSuchRevision:
592
basis_revision_id = _mod_revision.NULL_REVISION
594
revtrees.append(self.repository.revision_tree(
595
_mod_revision.NULL_REVISION))
596
# The basis inventory from a repository
598
basis_inv = revtrees[0].inventory
600
basis_inv = self.repository.revision_tree(
601
_mod_revision.NULL_REVISION).inventory
602
if len(self.parents) > 0:
603
if basis_revision_id != self.parents[0] and not ghost_basis:
605
"arbitrary basis parents not yet supported with merges")
606
for revtree in revtrees[1:]:
607
for change in revtree.inventory._make_delta(basis_inv):
608
if change[1] is None:
609
# Not present in this parent.
611
if change[2] not in merged_ids:
612
if change[0] is not None:
613
basis_entry = basis_inv[change[2]]
614
merged_ids[change[2]] = [
616
basis_entry.revision,
619
parent_entries[change[2]] = {
621
basis_entry.revision:basis_entry,
623
change[3].revision:change[3],
626
merged_ids[change[2]] = [change[3].revision]
627
parent_entries[change[2]] = {change[3].revision:change[3]}
629
merged_ids[change[2]].append(change[3].revision)
630
parent_entries[change[2]][change[3].revision] = change[3]
633
# Setup the changes from the tree:
634
# changes maps file_id -> (change, [parent revision_ids])
636
for change in iter_changes:
637
# This probably looks up in basis_inv way to much.
638
if change[1][0] is not None:
639
head_candidate = [basis_inv[change[0]].revision]
642
changes[change[0]] = change, merged_ids.get(change[0],
644
unchanged_merged = set(merged_ids) - set(changes)
645
# Extend the changes dict with synthetic changes to record merges of
647
for file_id in unchanged_merged:
648
# Record a merged version of these items that did not change vs the
649
# basis. This can be either identical parallel changes, or a revert
650
# of a specific file after a merge. The recorded content will be
651
# that of the current tree (which is the same as the basis), but
652
# the per-file graph will reflect a merge.
653
# NB:XXX: We are reconstructing path information we had, this
654
# should be preserved instead.
655
# inv delta change: (file_id, (path_in_source, path_in_target),
656
# changed_content, versioned, parent, name, kind,
659
basis_entry = basis_inv[file_id]
660
except errors.NoSuchId:
661
# a change from basis->some_parents but file_id isn't in basis
662
# so was new in the merge, which means it must have changed
663
# from basis -> current, and as it hasn't the add was reverted
664
# by the user. So we discard this change.
668
(basis_inv.id2path(file_id), tree.id2path(file_id)),
670
(basis_entry.parent_id, basis_entry.parent_id),
671
(basis_entry.name, basis_entry.name),
672
(basis_entry.kind, basis_entry.kind),
673
(basis_entry.executable, basis_entry.executable))
674
changes[file_id] = (change, merged_ids[file_id])
675
# changes contains tuples with the change and a set of inventory
676
# candidates for the file.
678
# old_path, new_path, file_id, new_inventory_entry
679
seen_root = False # Is the root in the basis delta?
680
inv_delta = self._basis_delta
681
modified_rev = self._new_revision_id
682
for change, head_candidates in changes.values():
683
if change[3][1]: # versioned in target.
684
# Several things may be happening here:
685
# We may have a fork in the per-file graph
686
# - record a change with the content from tree
687
# We may have a change against < all trees
688
# - carry over the tree that hasn't changed
689
# We may have a change against all trees
690
# - record the change with the content from tree
693
entry = _entry_factory[kind](file_id, change[5][1],
695
head_set = self._heads(change[0], set(head_candidates))
698
for head_candidate in head_candidates:
699
if head_candidate in head_set:
700
heads.append(head_candidate)
701
head_set.remove(head_candidate)
704
# Could be a carry-over situation:
705
parent_entry_revs = parent_entries.get(file_id, None)
706
if parent_entry_revs:
707
parent_entry = parent_entry_revs.get(heads[0], None)
710
if parent_entry is None:
711
# The parent iter_changes was called against is the one
712
# that is the per-file head, so any change is relevant
713
# iter_changes is valid.
714
carry_over_possible = False
716
# could be a carry over situation
717
# A change against the basis may just indicate a merge,
718
# we need to check the content against the source of the
719
# merge to determine if it was changed after the merge
721
if (parent_entry.kind != entry.kind or
722
parent_entry.parent_id != entry.parent_id or
723
parent_entry.name != entry.name):
724
# Metadata common to all entries has changed
725
# against per-file parent
726
carry_over_possible = False
728
carry_over_possible = True
729
# per-type checks for changes against the parent_entry
732
# Cannot be a carry-over situation
733
carry_over_possible = False
734
# Populate the entry in the delta
736
# XXX: There is still a small race here: If someone reverts the content of a file
737
# after iter_changes examines and decides it has changed,
738
# we will unconditionally record a new version even if some
739
# other process reverts it while commit is running (with
740
# the revert happening after iter_changes did it's
743
entry.executable = True
745
entry.executable = False
746
if (carry_over_possible and
747
parent_entry.executable == entry.executable):
748
# Check the file length, content hash after reading
750
nostore_sha = parent_entry.text_sha1
753
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
755
lines = file_obj.readlines()
759
entry.text_sha1, entry.text_size = self._add_text_to_weave(
760
file_id, lines, heads, nostore_sha)
761
yield file_id, change[1][1], (entry.text_sha1, stat_value)
762
except errors.ExistingContent:
763
# No content change against a carry_over parent
764
# Perhaps this should also yield a fs hash update?
766
entry.text_size = parent_entry.text_size
767
entry.text_sha1 = parent_entry.text_sha1
768
elif kind == 'symlink':
770
entry.symlink_target = tree.get_symlink_target(file_id)
771
if (carry_over_possible and
772
parent_entry.symlink_target == entry.symlink_target):
775
self._add_text_to_weave(change[0], [], heads, None)
776
elif kind == 'directory':
777
if carry_over_possible:
780
# Nothing to set on the entry.
781
# XXX: split into the Root and nonRoot versions.
782
if change[1][1] != '' or self.repository.supports_rich_root():
783
self._add_text_to_weave(change[0], [], heads, None)
784
elif kind == 'tree-reference':
785
if not self.repository._format.supports_tree_reference:
786
# This isn't quite sane as an error, but we shouldn't
787
# ever see this code path in practice: tree's don't
788
# permit references when the repo doesn't support tree
790
raise errors.UnsupportedOperation(tree.add_reference,
792
entry.reference_revision = \
793
tree.get_reference_revision(change[0])
794
if (carry_over_possible and
795
parent_entry.reference_revision == reference_revision):
798
self._add_text_to_weave(change[0], [], heads, None)
800
raise AssertionError('unknown kind %r' % kind)
802
entry.revision = modified_rev
804
entry.revision = parent_entry.revision
807
new_path = change[1][1]
808
inv_delta.append((change[1][0], new_path, change[0], entry))
811
self.new_inventory = None
813
self._any_changes = True
815
# housekeeping root entry changes do not affect no-change commits.
816
self._require_root_change(tree)
817
self.basis_delta_revision = basis_revision_id
819
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
820
# Note: as we read the content directly from the tree, we know its not
821
# been turned into unicode or badly split - but a broken tree
822
# implementation could give us bad output from readlines() so this is
823
# not a guarantee of safety. What would be better is always checking
824
# the content during test suite execution. RBC 20070912
825
parent_keys = tuple((file_id, parent) for parent in parents)
826
return self.repository.texts.add_lines(
827
(file_id, self._new_revision_id), parent_keys, new_lines,
828
nostore_sha=nostore_sha, random_id=self.random_revid,
829
check_content=False)[0:2]
832
class RootCommitBuilder(CommitBuilder):
833
"""This commitbuilder actually records the root id"""
835
# the root entry gets versioned properly by this builder.
836
_versioned_root = True
838
def _check_root(self, ie, parent_invs, tree):
839
"""Helper for record_entry_contents.
841
:param ie: An entry being added.
842
:param parent_invs: The inventories of the parent revisions of the
844
:param tree: The tree that is being committed.
847
def _require_root_change(self, tree):
848
"""Enforce an appropriate root object change.
850
This is called once when record_iter_changes is called, if and only if
851
the root was not in the delta calculated by record_iter_changes.
853
:param tree: The tree which is being committed.
855
# versioned roots do not change unless the tree found a change.
858
######################################################################
861
class Repository(object):
862
"""Repository holding history for one or more branches.
864
The repository holds and retrieves historical information including
865
revisions and file history. It's normally accessed only by the Branch,
866
which views a particular line of development through that history.
868
The Repository builds on top of some byte storage facilies (the revisions,
869
signatures, inventories and texts attributes) and a Transport, which
870
respectively provide byte storage and a means to access the (possibly
873
The byte storage facilities are addressed via tuples, which we refer to
874
as 'keys' throughout the code base. Revision_keys, inventory_keys and
875
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
876
(file_id, revision_id). We use this interface because it allows low
877
friction with the underlying code that implements disk indices, network
878
encoding and other parts of bzrlib.
880
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
881
the serialised revisions for the repository. This can be used to obtain
882
revision graph information or to access raw serialised revisions.
883
The result of trying to insert data into the repository via this store
884
is undefined: it should be considered read-only except for implementors
886
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
887
the serialised signatures for the repository. This can be used to
888
obtain access to raw serialised signatures. The result of trying to
889
insert data into the repository via this store is undefined: it should
890
be considered read-only except for implementors of repositories.
891
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
892
the serialised inventories for the repository. This can be used to
893
obtain unserialised inventories. The result of trying to insert data
894
into the repository via this store is undefined: it should be
895
considered read-only except for implementors of repositories.
896
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
897
texts of files and directories for the repository. This can be used to
898
obtain file texts or file graphs. Note that Repository.iter_file_bytes
899
is usually a better interface for accessing file texts.
900
The result of trying to insert data into the repository via this store
901
is undefined: it should be considered read-only except for implementors
903
:ivar _transport: Transport for file access to repository, typically
904
pointing to .bzr/repository.
907
# What class to use for a CommitBuilder. Often its simpler to change this
908
# in a Repository class subclass rather than to override
909
# get_commit_builder.
910
_commit_builder_class = CommitBuilder
911
# The search regex used by xml based repositories to determine what things
912
# where changed in a single commit.
913
_file_ids_altered_regex = lazy_regex.lazy_compile(
914
r'file_id="(?P<file_id>[^"]+)"'
915
r'.* revision="(?P<revision_id>[^"]+)"'
918
def abort_write_group(self, suppress_errors=False):
919
"""Commit the contents accrued within the current write group.
921
:param suppress_errors: if true, abort_write_group will catch and log
922
unexpected errors that happen during the abort, rather than
923
allowing them to propagate. Defaults to False.
925
:seealso: start_write_group.
927
if self._write_group is not self.get_transaction():
928
# has an unlock or relock occured ?
929
raise errors.BzrError('mismatched lock context and write group.')
931
self._abort_write_group()
932
except Exception, exc:
933
self._write_group = None
934
if not suppress_errors:
936
mutter('abort_write_group failed')
937
log_exception_quietly()
938
note('bzr: ERROR (ignored): %s', exc)
939
self._write_group = None
941
def _abort_write_group(self):
942
"""Template method for per-repository write group cleanup.
944
This is called during abort before the write group is considered to be
945
finished and should cleanup any internal state accrued during the write
946
group. There is no requirement that data handed to the repository be
947
*not* made available - this is not a rollback - but neither should any
948
attempt be made to ensure that data added is fully commited. Abort is
949
invoked when an error has occured so futher disk or network operations
950
may not be possible or may error and if possible should not be
954
def add_fallback_repository(self, repository):
955
"""Add a repository to use for looking up data not held locally.
957
:param repository: A repository.
959
if not self._format.supports_external_lookups:
960
raise errors.UnstackableRepositoryFormat(self._format, self.base)
961
self._check_fallback_repository(repository)
962
self._fallback_repositories.append(repository)
963
self.texts.add_fallback_versioned_files(repository.texts)
964
self.inventories.add_fallback_versioned_files(repository.inventories)
965
self.revisions.add_fallback_versioned_files(repository.revisions)
966
self.signatures.add_fallback_versioned_files(repository.signatures)
968
def _check_fallback_repository(self, repository):
969
"""Check that this repository can fallback to repository safely.
971
Raise an error if not.
973
:param repository: A repository to fallback to.
975
return InterRepository._assert_same_model(self, repository)
977
def add_inventory(self, revision_id, inv, parents):
978
"""Add the inventory inv to the repository as revision_id.
980
:param parents: The revision ids of the parents that revision_id
981
is known to have and are in the repository already.
983
:returns: The validator(which is a sha1 digest, though what is sha'd is
984
repository format specific) of the serialized inventory.
986
if not self.is_in_write_group():
987
raise AssertionError("%r not in write group" % (self,))
988
_mod_revision.check_not_reserved_id(revision_id)
989
if not (inv.revision_id is None or inv.revision_id == revision_id):
990
raise AssertionError(
991
"Mismatch between inventory revision"
992
" id and insertion revid (%r, %r)"
993
% (inv.revision_id, revision_id))
995
raise AssertionError()
996
inv_lines = self._serialise_inventory_to_lines(inv)
997
return self._inventory_add_lines(revision_id, parents,
998
inv_lines, check_content=False)
1000
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1002
"""Add a new inventory expressed as a delta against another revision.
1004
:param basis_revision_id: The inventory id the delta was created
1005
against. (This does not have to be a direct parent.)
1006
:param delta: The inventory delta (see Inventory.apply_delta for
1008
:param new_revision_id: The revision id that the inventory is being
1010
:param parents: The revision ids of the parents that revision_id is
1011
known to have and are in the repository already. These are supplied
1012
for repositories that depend on the inventory graph for revision
1013
graph access, as well as for those that pun ancestry with delta
1016
:returns: (validator, new_inv)
1017
The validator(which is a sha1 digest, though what is sha'd is
1018
repository format specific) of the serialized inventory, and the
1019
resulting inventory.
1021
if not self.is_in_write_group():
1022
raise AssertionError("%r not in write group" % (self,))
1023
_mod_revision.check_not_reserved_id(new_revision_id)
1024
basis_tree = self.revision_tree(basis_revision_id)
1025
basis_tree.lock_read()
1027
# Note that this mutates the inventory of basis_tree, which not all
1028
# inventory implementations may support: A better idiom would be to
1029
# return a new inventory, but as there is no revision tree cache in
1030
# repository this is safe for now - RBC 20081013
1031
basis_inv = basis_tree.inventory
1032
basis_inv.apply_delta(delta)
1033
basis_inv.revision_id = new_revision_id
1034
return (self.add_inventory(new_revision_id, basis_inv, parents),
1039
def _inventory_add_lines(self, revision_id, parents, lines,
1040
check_content=True):
1041
"""Store lines in inv_vf and return the sha1 of the inventory."""
1042
parents = [(parent,) for parent in parents]
1043
return self.inventories.add_lines((revision_id,), parents, lines,
1044
check_content=check_content)[0]
1046
def add_revision(self, revision_id, rev, inv=None, config=None):
1047
"""Add rev to the revision store as revision_id.
1049
:param revision_id: the revision id to use.
1050
:param rev: The revision object.
1051
:param inv: The inventory for the revision. if None, it will be looked
1052
up in the inventory storer
1053
:param config: If None no digital signature will be created.
1054
If supplied its signature_needed method will be used
1055
to determine if a signature should be made.
1057
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1059
_mod_revision.check_not_reserved_id(revision_id)
1060
if config is not None and config.signature_needed():
1062
inv = self.get_inventory(revision_id)
1063
plaintext = Testament(rev, inv).as_short_text()
1064
self.store_revision_signature(
1065
gpg.GPGStrategy(config), plaintext, revision_id)
1066
# check inventory present
1067
if not self.inventories.get_parent_map([(revision_id,)]):
1069
raise errors.WeaveRevisionNotPresent(revision_id,
1072
# yes, this is not suitable for adding with ghosts.
1073
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1076
key = (revision_id,)
1077
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1078
self._add_revision(rev)
1080
def _add_revision(self, revision):
1081
text = self._serializer.write_revision_to_string(revision)
1082
key = (revision.revision_id,)
1083
parents = tuple((parent,) for parent in revision.parent_ids)
1084
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1086
def all_revision_ids(self):
1087
"""Returns a list of all the revision ids in the repository.
1089
This is conceptually deprecated because code should generally work on
1090
the graph reachable from a particular revision, and ignore any other
1091
revisions that might be present. There is no direct replacement
1094
if 'evil' in debug.debug_flags:
1095
mutter_callsite(2, "all_revision_ids is linear with history.")
1096
return self._all_revision_ids()
1098
def _all_revision_ids(self):
1099
"""Returns a list of all the revision ids in the repository.
1101
These are in as much topological order as the underlying store can
1104
raise NotImplementedError(self._all_revision_ids)
1106
def break_lock(self):
1107
"""Break a lock if one is present from another instance.
1109
Uses the ui factory to ask for confirmation if the lock may be from
1112
self.control_files.break_lock()
1115
def _eliminate_revisions_not_present(self, revision_ids):
1116
"""Check every revision id in revision_ids to see if we have it.
1118
Returns a set of the present revisions.
1121
graph = self.get_graph()
1122
parent_map = graph.get_parent_map(revision_ids)
1123
# The old API returned a list, should this actually be a set?
1124
return parent_map.keys()
1127
def create(a_bzrdir):
1128
"""Construct the current default format repository in a_bzrdir."""
1129
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1131
def __init__(self, _format, a_bzrdir, control_files):
1132
"""instantiate a Repository.
1134
:param _format: The format of the repository on disk.
1135
:param a_bzrdir: The BzrDir of the repository.
1137
In the future we will have a single api for all stores for
1138
getting file texts, inventories and revisions, then
1139
this construct will accept instances of those things.
1141
super(Repository, self).__init__()
1142
self._format = _format
1143
# the following are part of the public API for Repository:
1144
self.bzrdir = a_bzrdir
1145
self.control_files = control_files
1146
self._transport = control_files._transport
1147
self.base = self._transport.base
1149
self._reconcile_does_inventory_gc = True
1150
self._reconcile_fixes_text_parents = False
1151
self._reconcile_backsup_inventory = True
1152
# not right yet - should be more semantically clear ?
1154
# TODO: make sure to construct the right store classes, etc, depending
1155
# on whether escaping is required.
1156
self._warn_if_deprecated()
1157
self._write_group = None
1158
# Additional places to query for data.
1159
self._fallback_repositories = []
1160
# An InventoryEntry cache, used during deserialization
1161
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1164
return '%s(%r)' % (self.__class__.__name__,
1167
def has_same_location(self, other):
1168
"""Returns a boolean indicating if this repository is at the same
1169
location as another repository.
1171
This might return False even when two repository objects are accessing
1172
the same physical repository via different URLs.
1174
if self.__class__ is not other.__class__:
1176
return (self._transport.base == other._transport.base)
1178
def is_in_write_group(self):
1179
"""Return True if there is an open write group.
1181
:seealso: start_write_group.
1183
return self._write_group is not None
1185
def is_locked(self):
1186
return self.control_files.is_locked()
1188
def is_write_locked(self):
1189
"""Return True if this object is write locked."""
1190
return self.is_locked() and self.control_files._lock_mode == 'w'
1192
def lock_write(self, token=None):
1193
"""Lock this repository for writing.
1195
This causes caching within the repository obejct to start accumlating
1196
data during reads, and allows a 'write_group' to be obtained. Write
1197
groups must be used for actual data insertion.
1199
:param token: if this is already locked, then lock_write will fail
1200
unless the token matches the existing lock.
1201
:returns: a token if this instance supports tokens, otherwise None.
1202
:raises TokenLockingNotSupported: when a token is given but this
1203
instance doesn't support using token locks.
1204
:raises MismatchedToken: if the specified token doesn't match the token
1205
of the existing lock.
1206
:seealso: start_write_group.
1208
A token should be passed in if you know that you have locked the object
1209
some other way, and need to synchronise this object's state with that
1212
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1214
locked = self.is_locked()
1215
result = self.control_files.lock_write(token=token)
1216
for repo in self._fallback_repositories:
1217
# Writes don't affect fallback repos
1220
self._refresh_data()
1223
def lock_read(self):
1224
locked = self.is_locked()
1225
self.control_files.lock_read()
1226
for repo in self._fallback_repositories:
1229
self._refresh_data()
1231
def get_physical_lock_status(self):
1232
return self.control_files.get_physical_lock_status()
1234
def leave_lock_in_place(self):
1235
"""Tell this repository not to release the physical lock when this
1238
If lock_write doesn't return a token, then this method is not supported.
1240
self.control_files.leave_in_place()
1242
def dont_leave_lock_in_place(self):
1243
"""Tell this repository to release the physical lock when this
1244
object is unlocked, even if it didn't originally acquire it.
1246
If lock_write doesn't return a token, then this method is not supported.
1248
self.control_files.dont_leave_in_place()
1251
def gather_stats(self, revid=None, committers=None):
1252
"""Gather statistics from a revision id.
1254
:param revid: The revision id to gather statistics from, if None, then
1255
no revision specific statistics are gathered.
1256
:param committers: Optional parameter controlling whether to grab
1257
a count of committers from the revision specific statistics.
1258
:return: A dictionary of statistics. Currently this contains:
1259
committers: The number of committers if requested.
1260
firstrev: A tuple with timestamp, timezone for the penultimate left
1261
most ancestor of revid, if revid is not the NULL_REVISION.
1262
latestrev: A tuple with timestamp, timezone for revid, if revid is
1263
not the NULL_REVISION.
1264
revisions: The total revision count in the repository.
1265
size: An estimate disk size of the repository in bytes.
1268
if revid and committers:
1269
result['committers'] = 0
1270
if revid and revid != _mod_revision.NULL_REVISION:
1272
all_committers = set()
1273
revisions = self.get_ancestry(revid)
1274
# pop the leading None
1276
first_revision = None
1278
# ignore the revisions in the middle - just grab first and last
1279
revisions = revisions[0], revisions[-1]
1280
for revision in self.get_revisions(revisions):
1281
if not first_revision:
1282
first_revision = revision
1284
all_committers.add(revision.committer)
1285
last_revision = revision
1287
result['committers'] = len(all_committers)
1288
result['firstrev'] = (first_revision.timestamp,
1289
first_revision.timezone)
1290
result['latestrev'] = (last_revision.timestamp,
1291
last_revision.timezone)
1293
# now gather global repository information
1294
# XXX: This is available for many repos regardless of listability.
1295
if self.bzrdir.root_transport.listable():
1296
# XXX: do we want to __define len__() ?
1297
# Maybe the versionedfiles object should provide a different
1298
# method to get the number of keys.
1299
result['revisions'] = len(self.revisions.keys())
1300
# result['size'] = t
1303
def find_branches(self, using=False):
1304
"""Find branches underneath this repository.
1306
This will include branches inside other branches.
1308
:param using: If True, list only branches using this repository.
1310
if using and not self.is_shared():
1312
return [self.bzrdir.open_branch()]
1313
except errors.NotBranchError:
1315
class Evaluator(object):
1318
self.first_call = True
1320
def __call__(self, bzrdir):
1321
# On the first call, the parameter is always the bzrdir
1322
# containing the current repo.
1323
if not self.first_call:
1325
repository = bzrdir.open_repository()
1326
except errors.NoRepositoryPresent:
1329
return False, (None, repository)
1330
self.first_call = False
1332
value = (bzrdir.open_branch(), None)
1333
except errors.NotBranchError:
1334
value = (None, None)
1338
for branch, repository in bzrdir.BzrDir.find_bzrdirs(
1339
self.bzrdir.root_transport, evaluate=Evaluator()):
1340
if branch is not None:
1341
branches.append(branch)
1342
if not using and repository is not None:
1343
branches.extend(repository.find_branches())
1347
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
1348
"""Return the revision ids that other has that this does not.
1350
These are returned in topological order.
1352
revision_id: only return revision ids included by revision_id.
1354
return InterRepository.get(other, self).search_missing_revision_ids(
1355
revision_id, find_ghosts)
1359
"""Open the repository rooted at base.
1361
For instance, if the repository is at URL/.bzr/repository,
1362
Repository.open(URL) -> a Repository instance.
1364
control = bzrdir.BzrDir.open(base)
1365
return control.open_repository()
1367
def copy_content_into(self, destination, revision_id=None):
1368
"""Make a complete copy of the content in self into destination.
1370
This is a destructive operation! Do not use it on existing
1373
return InterRepository.get(self, destination).copy_content(revision_id)
1375
def commit_write_group(self):
1376
"""Commit the contents accrued within the current write group.
1378
:seealso: start_write_group.
1380
if self._write_group is not self.get_transaction():
1381
# has an unlock or relock occured ?
1382
raise errors.BzrError('mismatched lock context %r and '
1384
(self.get_transaction(), self._write_group))
1385
self._commit_write_group()
1386
self._write_group = None
1388
def _commit_write_group(self):
1389
"""Template method for per-repository write group cleanup.
1391
This is called before the write group is considered to be
1392
finished and should ensure that all data handed to the repository
1393
for writing during the write group is safely committed (to the
1394
extent possible considering file system caching etc).
1397
def suspend_write_group(self):
1398
raise errors.UnsuspendableWriteGroup(self)
1400
def refresh_data(self):
1401
"""Re-read any data needed to to synchronise with disk.
1403
This method is intended to be called after another repository instance
1404
(such as one used by a smart server) has inserted data into the
1405
repository. It may not be called during a write group, but may be
1406
called at any other time.
1408
if self.is_in_write_group():
1409
raise errors.InternalBzrError(
1410
"May not refresh_data while in a write group.")
1411
self._refresh_data()
1413
def resume_write_group(self, tokens):
1414
if not self.is_write_locked():
1415
raise errors.NotWriteLocked(self)
1416
if self._write_group:
1417
raise errors.BzrError('already in a write group')
1418
self._resume_write_group(tokens)
1419
# so we can detect unlock/relock - the write group is now entered.
1420
self._write_group = self.get_transaction()
1422
def _resume_write_group(self, tokens):
1423
raise errors.UnsuspendableWriteGroup(self)
1425
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1427
"""Fetch the content required to construct revision_id from source.
1429
If revision_id is None and fetch_spec is None, then all content is
1432
fetch() may not be used when the repository is in a write group -
1433
either finish the current write group before using fetch, or use
1434
fetch before starting the write group.
1436
:param find_ghosts: Find and copy revisions in the source that are
1437
ghosts in the target (and not reachable directly by walking out to
1438
the first-present revision in target from revision_id).
1439
:param revision_id: If specified, all the content needed for this
1440
revision ID will be copied to the target. Fetch will determine for
1441
itself which content needs to be copied.
1442
:param fetch_spec: If specified, a SearchResult or
1443
PendingAncestryResult that describes which revisions to copy. This
1444
allows copying multiple heads at once. Mutually exclusive with
1447
if fetch_spec is not None and revision_id is not None:
1448
raise AssertionError(
1449
"fetch_spec and revision_id are mutually exclusive.")
1450
if self.is_in_write_group():
1451
raise errors.InternalBzrError(
1452
"May not fetch while in a write group.")
1453
# fast path same-url fetch operations
1454
if self.has_same_location(source) and fetch_spec is None:
1455
# check that last_revision is in 'from' and then return a
1457
if (revision_id is not None and
1458
not _mod_revision.is_null(revision_id)):
1459
self.get_revision(revision_id)
1461
# if there is no specific appropriate InterRepository, this will get
1462
# the InterRepository base class, which raises an
1463
# IncompatibleRepositories when asked to fetch.
1464
inter = InterRepository.get(source, self)
1465
return inter.fetch(revision_id=revision_id, pb=pb,
1466
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1468
def create_bundle(self, target, base, fileobj, format=None):
1469
return serializer.write_bundle(self, target, base, fileobj, format)
1471
def get_commit_builder(self, branch, parents, config, timestamp=None,
1472
timezone=None, committer=None, revprops=None,
1474
"""Obtain a CommitBuilder for this repository.
1476
:param branch: Branch to commit to.
1477
:param parents: Revision ids of the parents of the new revision.
1478
:param config: Configuration to use.
1479
:param timestamp: Optional timestamp recorded for commit.
1480
:param timezone: Optional timezone for timestamp.
1481
:param committer: Optional committer to set for commit.
1482
:param revprops: Optional dictionary of revision properties.
1483
:param revision_id: Optional revision id.
1485
result = self._commit_builder_class(self, parents, config,
1486
timestamp, timezone, committer, revprops, revision_id)
1487
self.start_write_group()
1491
if (self.control_files._lock_count == 1 and
1492
self.control_files._lock_mode == 'w'):
1493
if self._write_group is not None:
1494
self.abort_write_group()
1495
self.control_files.unlock()
1496
raise errors.BzrError(
1497
'Must end write groups before releasing write locks.')
1498
self.control_files.unlock()
1499
if self.control_files._lock_count == 0:
1500
self._inventory_entry_cache.clear()
1501
for repo in self._fallback_repositories:
1505
def clone(self, a_bzrdir, revision_id=None):
1506
"""Clone this repository into a_bzrdir using the current format.
1508
Currently no check is made that the format of this repository and
1509
the bzrdir format are compatible. FIXME RBC 20060201.
1511
:return: The newly created destination repository.
1513
# TODO: deprecate after 0.16; cloning this with all its settings is
1514
# probably not very useful -- mbp 20070423
1515
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1516
self.copy_content_into(dest_repo, revision_id)
1519
def start_write_group(self):
1520
"""Start a write group in the repository.
1522
Write groups are used by repositories which do not have a 1:1 mapping
1523
between file ids and backend store to manage the insertion of data from
1524
both fetch and commit operations.
1526
A write lock is required around the start_write_group/commit_write_group
1527
for the support of lock-requiring repository formats.
1529
One can only insert data into a repository inside a write group.
1533
if not self.is_write_locked():
1534
raise errors.NotWriteLocked(self)
1535
if self._write_group:
1536
raise errors.BzrError('already in a write group')
1537
self._start_write_group()
1538
# so we can detect unlock/relock - the write group is now entered.
1539
self._write_group = self.get_transaction()
1541
def _start_write_group(self):
1542
"""Template method for per-repository write group startup.
1544
This is called before the write group is considered to be
1549
def sprout(self, to_bzrdir, revision_id=None):
1550
"""Create a descendent repository for new development.
1552
Unlike clone, this does not copy the settings of the repository.
1554
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1555
dest_repo.fetch(self, revision_id=revision_id)
1558
def _create_sprouting_repo(self, a_bzrdir, shared):
1559
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1560
# use target default format.
1561
dest_repo = a_bzrdir.create_repository()
1563
# Most control formats need the repository to be specifically
1564
# created, but on some old all-in-one formats it's not needed
1566
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1567
except errors.UninitializableFormat:
1568
dest_repo = a_bzrdir.open_repository()
1571
def _get_sink(self):
1572
"""Return a sink for streaming into this repository."""
1573
return StreamSink(self)
1575
def _get_source(self, to_format):
1576
"""Return a source for streaming from this repository."""
1577
return StreamSource(self, to_format)
1580
def has_revision(self, revision_id):
1581
"""True if this repository has a copy of the revision."""
1582
return revision_id in self.has_revisions((revision_id,))
1585
def has_revisions(self, revision_ids):
1586
"""Probe to find out the presence of multiple revisions.
1588
:param revision_ids: An iterable of revision_ids.
1589
:return: A set of the revision_ids that were present.
1591
parent_map = self.revisions.get_parent_map(
1592
[(rev_id,) for rev_id in revision_ids])
1594
if _mod_revision.NULL_REVISION in revision_ids:
1595
result.add(_mod_revision.NULL_REVISION)
1596
result.update([key[0] for key in parent_map])
1600
def get_revision(self, revision_id):
1601
"""Return the Revision object for a named revision."""
1602
return self.get_revisions([revision_id])[0]
1605
def get_revision_reconcile(self, revision_id):
1606
"""'reconcile' helper routine that allows access to a revision always.
1608
This variant of get_revision does not cross check the weave graph
1609
against the revision one as get_revision does: but it should only
1610
be used by reconcile, or reconcile-alike commands that are correcting
1611
or testing the revision graph.
1613
return self._get_revisions([revision_id])[0]
1616
def get_revisions(self, revision_ids):
1617
"""Get many revisions at once."""
1618
return self._get_revisions(revision_ids)
1621
def _get_revisions(self, revision_ids):
1622
"""Core work logic to get many revisions without sanity checks."""
1623
for rev_id in revision_ids:
1624
if not rev_id or not isinstance(rev_id, basestring):
1625
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1626
keys = [(key,) for key in revision_ids]
1627
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1629
for record in stream:
1630
if record.storage_kind == 'absent':
1631
raise errors.NoSuchRevision(self, record.key[0])
1632
text = record.get_bytes_as('fulltext')
1633
rev = self._serializer.read_revision_from_string(text)
1634
revs[record.key[0]] = rev
1635
return [revs[revid] for revid in revision_ids]
1638
def get_revision_xml(self, revision_id):
1639
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1640
# would have already do it.
1641
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1642
# TODO: this can't just be replaced by:
1643
# return self._serializer.write_revision_to_string(
1644
# self.get_revision(revision_id))
1645
# as cStringIO preservers the encoding unlike write_revision_to_string
1646
# or some other call down the path.
1647
rev = self.get_revision(revision_id)
1648
rev_tmp = cStringIO.StringIO()
1649
# the current serializer..
1650
self._serializer.write_revision(rev, rev_tmp)
1652
return rev_tmp.getvalue()
1654
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1655
"""Produce a generator of revision deltas.
1657
Note that the input is a sequence of REVISIONS, not revision_ids.
1658
Trees will be held in memory until the generator exits.
1659
Each delta is relative to the revision's lefthand predecessor.
1661
:param specific_fileids: if not None, the result is filtered
1662
so that only those file-ids, their parents and their
1663
children are included.
1665
# Get the revision-ids of interest
1666
required_trees = set()
1667
for revision in revisions:
1668
required_trees.add(revision.revision_id)
1669
required_trees.update(revision.parent_ids[:1])
1671
# Get the matching filtered trees. Note that it's more
1672
# efficient to pass filtered trees to changes_from() rather
1673
# than doing the filtering afterwards. changes_from() could
1674
# arguably do the filtering itself but it's path-based, not
1675
# file-id based, so filtering before or afterwards is
1677
if specific_fileids is None:
1678
trees = dict((t.get_revision_id(), t) for
1679
t in self.revision_trees(required_trees))
1681
trees = dict((t.get_revision_id(), t) for
1682
t in self._filtered_revision_trees(required_trees,
1685
# Calculate the deltas
1686
for revision in revisions:
1687
if not revision.parent_ids:
1688
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1690
old_tree = trees[revision.parent_ids[0]]
1691
yield trees[revision.revision_id].changes_from(old_tree)
1694
def get_revision_delta(self, revision_id, specific_fileids=None):
1695
"""Return the delta for one revision.
1697
The delta is relative to the left-hand predecessor of the
1700
:param specific_fileids: if not None, the result is filtered
1701
so that only those file-ids, their parents and their
1702
children are included.
1704
r = self.get_revision(revision_id)
1705
return list(self.get_deltas_for_revisions([r],
1706
specific_fileids=specific_fileids))[0]
1709
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1710
signature = gpg_strategy.sign(plaintext)
1711
self.add_signature_text(revision_id, signature)
1714
def add_signature_text(self, revision_id, signature):
1715
self.signatures.add_lines((revision_id,), (),
1716
osutils.split_lines(signature))
1718
def find_text_key_references(self):
1719
"""Find the text key references within the repository.
1721
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1722
to whether they were referred to by the inventory of the
1723
revision_id that they contain. The inventory texts from all present
1724
revision ids are assessed to generate this report.
1726
revision_keys = self.revisions.keys()
1727
w = self.inventories
1728
pb = ui.ui_factory.nested_progress_bar()
1730
return self._find_text_key_references_from_xml_inventory_lines(
1731
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1735
def _find_text_key_references_from_xml_inventory_lines(self,
1737
"""Core routine for extracting references to texts from inventories.
1739
This performs the translation of xml lines to revision ids.
1741
:param line_iterator: An iterator of lines, origin_version_id
1742
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1743
to whether they were referred to by the inventory of the
1744
revision_id that they contain. Note that if that revision_id was
1745
not part of the line_iterator's output then False will be given -
1746
even though it may actually refer to that key.
1748
if not self._serializer.support_altered_by_hack:
1749
raise AssertionError(
1750
"_find_text_key_references_from_xml_inventory_lines only "
1751
"supported for branches which store inventory as unnested xml"
1752
", not on %r" % self)
1755
# this code needs to read every new line in every inventory for the
1756
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
1757
# not present in one of those inventories is unnecessary but not
1758
# harmful because we are filtering by the revision id marker in the
1759
# inventory lines : we only select file ids altered in one of those
1760
# revisions. We don't need to see all lines in the inventory because
1761
# only those added in an inventory in rev X can contain a revision=X
1763
unescape_revid_cache = {}
1764
unescape_fileid_cache = {}
1766
# jam 20061218 In a big fetch, this handles hundreds of thousands
1767
# of lines, so it has had a lot of inlining and optimizing done.
1768
# Sorry that it is a little bit messy.
1769
# Move several functions to be local variables, since this is a long
1771
search = self._file_ids_altered_regex.search
1772
unescape = _unescape_xml
1773
setdefault = result.setdefault
1774
for line, line_key in line_iterator:
1775
match = search(line)
1778
# One call to match.group() returning multiple items is quite a
1779
# bit faster than 2 calls to match.group() each returning 1
1780
file_id, revision_id = match.group('file_id', 'revision_id')
1782
# Inlining the cache lookups helps a lot when you make 170,000
1783
# lines and 350k ids, versus 8.4 unique ids.
1784
# Using a cache helps in 2 ways:
1785
# 1) Avoids unnecessary decoding calls
1786
# 2) Re-uses cached strings, which helps in future set and
1788
# (2) is enough that removing encoding entirely along with
1789
# the cache (so we are using plain strings) results in no
1790
# performance improvement.
1792
revision_id = unescape_revid_cache[revision_id]
1794
unescaped = unescape(revision_id)
1795
unescape_revid_cache[revision_id] = unescaped
1796
revision_id = unescaped
1798
# Note that unconditionally unescaping means that we deserialise
1799
# every fileid, which for general 'pull' is not great, but we don't
1800
# really want to have some many fulltexts that this matters anyway.
1803
file_id = unescape_fileid_cache[file_id]
1805
unescaped = unescape(file_id)
1806
unescape_fileid_cache[file_id] = unescaped
1809
key = (file_id, revision_id)
1810
setdefault(key, False)
1811
if revision_id == line_key[-1]:
1815
def _inventory_xml_lines_for_keys(self, keys):
1816
"""Get a line iterator of the sort needed for findind references.
1818
Not relevant for non-xml inventory repositories.
1820
Ghosts in revision_keys are ignored.
1822
:param revision_keys: The revision keys for the inventories to inspect.
1823
:return: An iterator over (inventory line, revid) for the fulltexts of
1824
all of the xml inventories specified by revision_keys.
1826
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1827
for record in stream:
1828
if record.storage_kind != 'absent':
1829
chunks = record.get_bytes_as('chunked')
1830
revid = record.key[-1]
1831
lines = osutils.chunks_to_lines(chunks)
1835
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1837
"""Helper routine for fileids_altered_by_revision_ids.
1839
This performs the translation of xml lines to revision ids.
1841
:param line_iterator: An iterator of lines, origin_version_id
1842
:param revision_ids: The revision ids to filter for. This should be a
1843
set or other type which supports efficient __contains__ lookups, as
1844
the revision id from each parsed line will be looked up in the
1845
revision_ids filter.
1846
:return: a dictionary mapping altered file-ids to an iterable of
1847
revision_ids. Each altered file-ids has the exact revision_ids that
1848
altered it listed explicitly.
1850
seen = set(self._find_text_key_references_from_xml_inventory_lines(
1851
line_iterator).iterkeys())
1852
# Note that revision_ids are revision keys.
1853
parent_maps = self.revisions.get_parent_map(revision_ids)
1855
map(parents.update, parent_maps.itervalues())
1856
parents.difference_update(revision_ids)
1857
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
1858
self._inventory_xml_lines_for_keys(parents)))
1859
new_keys = seen - parent_seen
1861
setdefault = result.setdefault
1862
for key in new_keys:
1863
setdefault(key[0], set()).add(key[-1])
1866
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
1867
"""Find the file ids and versions affected by revisions.
1869
:param revisions: an iterable containing revision ids.
1870
:param _inv_weave: The inventory weave from this repository or None.
1871
If None, the inventory weave will be opened automatically.
1872
:return: a dictionary mapping altered file-ids to an iterable of
1873
revision_ids. Each altered file-ids has the exact revision_ids that
1874
altered it listed explicitly.
1876
selected_keys = set((revid,) for revid in revision_ids)
1877
w = _inv_weave or self.inventories
1878
pb = ui.ui_factory.nested_progress_bar()
1880
return self._find_file_ids_from_xml_inventory_lines(
1881
w.iter_lines_added_or_present_in_keys(
1882
selected_keys, pb=pb),
1887
def iter_files_bytes(self, desired_files):
1888
"""Iterate through file versions.
1890
Files will not necessarily be returned in the order they occur in
1891
desired_files. No specific order is guaranteed.
1893
Yields pairs of identifier, bytes_iterator. identifier is an opaque
1894
value supplied by the caller as part of desired_files. It should
1895
uniquely identify the file version in the caller's context. (Examples:
1896
an index number or a TreeTransform trans_id.)
1898
bytes_iterator is an iterable of bytestrings for the file. The
1899
kind of iterable and length of the bytestrings are unspecified, but for
1900
this implementation, it is a list of bytes produced by
1901
VersionedFile.get_record_stream().
1903
:param desired_files: a list of (file_id, revision_id, identifier)
1907
for file_id, revision_id, callable_data in desired_files:
1908
text_keys[(file_id, revision_id)] = callable_data
1909
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
1910
if record.storage_kind == 'absent':
1911
raise errors.RevisionNotPresent(record.key, self)
1912
yield text_keys[record.key], record.get_bytes_as('chunked')
1914
def _generate_text_key_index(self, text_key_references=None,
1916
"""Generate a new text key index for the repository.
1918
This is an expensive function that will take considerable time to run.
1920
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
1921
list of parents, also text keys. When a given key has no parents,
1922
the parents list will be [NULL_REVISION].
1924
# All revisions, to find inventory parents.
1925
if ancestors is None:
1926
graph = self.get_graph()
1927
ancestors = graph.get_parent_map(self.all_revision_ids())
1928
if text_key_references is None:
1929
text_key_references = self.find_text_key_references()
1930
pb = ui.ui_factory.nested_progress_bar()
1932
return self._do_generate_text_key_index(ancestors,
1933
text_key_references, pb)
1937
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
1938
"""Helper for _generate_text_key_index to avoid deep nesting."""
1939
revision_order = tsort.topo_sort(ancestors)
1940
invalid_keys = set()
1942
for revision_id in revision_order:
1943
revision_keys[revision_id] = set()
1944
text_count = len(text_key_references)
1945
# a cache of the text keys to allow reuse; costs a dict of all the
1946
# keys, but saves a 2-tuple for every child of a given key.
1948
for text_key, valid in text_key_references.iteritems():
1950
invalid_keys.add(text_key)
1952
revision_keys[text_key[1]].add(text_key)
1953
text_key_cache[text_key] = text_key
1954
del text_key_references
1956
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
1957
NULL_REVISION = _mod_revision.NULL_REVISION
1958
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
1959
# too small for large or very branchy trees. However, for 55K path
1960
# trees, it would be easy to use too much memory trivially. Ideally we
1961
# could gauge this by looking at available real memory etc, but this is
1962
# always a tricky proposition.
1963
inventory_cache = lru_cache.LRUCache(10)
1964
batch_size = 10 # should be ~150MB on a 55K path tree
1965
batch_count = len(revision_order) / batch_size + 1
1967
pb.update("Calculating text parents", processed_texts, text_count)
1968
for offset in xrange(batch_count):
1969
to_query = revision_order[offset * batch_size:(offset + 1) *
1973
for rev_tree in self.revision_trees(to_query):
1974
revision_id = rev_tree.get_revision_id()
1975
parent_ids = ancestors[revision_id]
1976
for text_key in revision_keys[revision_id]:
1977
pb.update("Calculating text parents", processed_texts)
1978
processed_texts += 1
1979
candidate_parents = []
1980
for parent_id in parent_ids:
1981
parent_text_key = (text_key[0], parent_id)
1983
check_parent = parent_text_key not in \
1984
revision_keys[parent_id]
1986
# the parent parent_id is a ghost:
1987
check_parent = False
1988
# truncate the derived graph against this ghost.
1989
parent_text_key = None
1991
# look at the parent commit details inventories to
1992
# determine possible candidates in the per file graph.
1995
inv = inventory_cache[parent_id]
1997
inv = self.revision_tree(parent_id).inventory
1998
inventory_cache[parent_id] = inv
1999
parent_entry = inv._byid.get(text_key[0], None)
2000
if parent_entry is not None:
2002
text_key[0], parent_entry.revision)
2004
parent_text_key = None
2005
if parent_text_key is not None:
2006
candidate_parents.append(
2007
text_key_cache[parent_text_key])
2008
parent_heads = text_graph.heads(candidate_parents)
2009
new_parents = list(parent_heads)
2010
new_parents.sort(key=lambda x:candidate_parents.index(x))
2011
if new_parents == []:
2012
new_parents = [NULL_REVISION]
2013
text_index[text_key] = new_parents
2015
for text_key in invalid_keys:
2016
text_index[text_key] = [NULL_REVISION]
2019
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2020
"""Get an iterable listing the keys of all the data introduced by a set
2023
The keys will be ordered so that the corresponding items can be safely
2024
fetched and inserted in that order.
2026
:returns: An iterable producing tuples of (knit-kind, file-id,
2027
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2028
'revisions'. file-id is None unless knit-kind is 'file'.
2030
# XXX: it's a bit weird to control the inventory weave caching in this
2031
# generator. Ideally the caching would be done in fetch.py I think. Or
2032
# maybe this generator should explicitly have the contract that it
2033
# should not be iterated until the previously yielded item has been
2035
inv_w = self.inventories
2037
# file ids that changed
2038
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2040
num_file_ids = len(file_ids)
2041
for file_id, altered_versions in file_ids.iteritems():
2042
if _files_pb is not None:
2043
_files_pb.update("fetch texts", count, num_file_ids)
2045
yield ("file", file_id, altered_versions)
2046
# We're done with the files_pb. Note that it finished by the caller,
2047
# just as it was created by the caller.
2051
yield ("inventory", None, revision_ids)
2054
# XXX: Note ATM no callers actually pay attention to this return
2055
# instead they just use the list of revision ids and ignore
2056
# missing sigs. Consider removing this work entirely
2057
revisions_with_signatures = set(self.signatures.get_parent_map(
2058
[(r,) for r in revision_ids]))
2059
revisions_with_signatures = set(
2060
[r for (r,) in revisions_with_signatures])
2061
revisions_with_signatures.intersection_update(revision_ids)
2062
yield ("signatures", None, revisions_with_signatures)
2065
yield ("revisions", None, revision_ids)
2068
def get_inventory(self, revision_id):
2069
"""Get Inventory object by revision id."""
2070
return self.iter_inventories([revision_id]).next()
2072
def iter_inventories(self, revision_ids):
2073
"""Get many inventories by revision_ids.
2075
This will buffer some or all of the texts used in constructing the
2076
inventories in memory, but will only parse a single inventory at a
2079
:param revision_ids: The expected revision ids of the inventories.
2080
:return: An iterator of inventories.
2082
if ((None in revision_ids)
2083
or (_mod_revision.NULL_REVISION in revision_ids)):
2084
raise ValueError('cannot get null revision inventory')
2085
return self._iter_inventories(revision_ids)
2087
def _iter_inventories(self, revision_ids):
2088
"""single-document based inventory iteration."""
2089
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2090
yield self.deserialise_inventory(revision_id, text)
2092
def _iter_inventory_xmls(self, revision_ids):
2093
keys = [(revision_id,) for revision_id in revision_ids]
2094
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2096
for record in stream:
2097
if record.storage_kind != 'absent':
2098
text_chunks[record.key] = record.get_bytes_as('chunked')
2100
raise errors.NoSuchRevision(self, record.key)
2102
chunks = text_chunks.pop(key)
2103
yield ''.join(chunks), key[-1]
2105
def deserialise_inventory(self, revision_id, xml):
2106
"""Transform the xml into an inventory object.
2108
:param revision_id: The expected revision id of the inventory.
2109
:param xml: A serialised inventory.
2111
result = self._serializer.read_inventory_from_string(xml, revision_id,
2112
entry_cache=self._inventory_entry_cache)
2113
if result.revision_id != revision_id:
2114
raise AssertionError('revision id mismatch %s != %s' % (
2115
result.revision_id, revision_id))
2118
def serialise_inventory(self, inv):
2119
return self._serializer.write_inventory_to_string(inv)
2121
def _serialise_inventory_to_lines(self, inv):
2122
return self._serializer.write_inventory_to_lines(inv)
2124
def get_serializer_format(self):
2125
return self._serializer.format_num
2128
def get_inventory_xml(self, revision_id):
2129
"""Get inventory XML as a file object."""
2130
texts = self._iter_inventory_xmls([revision_id])
2132
text, revision_id = texts.next()
2133
except StopIteration:
2134
raise errors.HistoryMissing(self, 'inventory', revision_id)
2138
def get_inventory_sha1(self, revision_id):
2139
"""Return the sha1 hash of the inventory entry
2141
return self.get_revision(revision_id).inventory_sha1
2143
def iter_reverse_revision_history(self, revision_id):
2144
"""Iterate backwards through revision ids in the lefthand history
2146
:param revision_id: The revision id to start with. All its lefthand
2147
ancestors will be traversed.
2149
graph = self.get_graph()
2150
next_id = revision_id
2152
if next_id in (None, _mod_revision.NULL_REVISION):
2155
# Note: The following line may raise KeyError in the event of
2156
# truncated history. We decided not to have a try:except:raise
2157
# RevisionNotPresent here until we see a use for it, because of the
2158
# cost in an inner loop that is by its very nature O(history).
2159
# Robert Collins 20080326
2160
parents = graph.get_parent_map([next_id])[next_id]
2161
if len(parents) == 0:
2164
next_id = parents[0]
2167
def get_revision_inventory(self, revision_id):
2168
"""Return inventory of a past revision."""
2169
# TODO: Unify this with get_inventory()
2170
# bzr 0.0.6 and later imposes the constraint that the inventory_id
2171
# must be the same as its revision, so this is trivial.
2172
if revision_id is None:
2173
# This does not make sense: if there is no revision,
2174
# then it is the current tree inventory surely ?!
2175
# and thus get_root_id() is something that looks at the last
2176
# commit on the branch, and the get_root_id is an inventory check.
2177
raise NotImplementedError
2178
# return Inventory(self.get_root_id())
2180
return self.get_inventory(revision_id)
2182
def is_shared(self):
2183
"""Return True if this repository is flagged as a shared repository."""
2184
raise NotImplementedError(self.is_shared)
2187
def reconcile(self, other=None, thorough=False):
2188
"""Reconcile this repository."""
2189
from bzrlib.reconcile import RepoReconciler
2190
reconciler = RepoReconciler(self, thorough=thorough)
2191
reconciler.reconcile()
2194
def _refresh_data(self):
2195
"""Helper called from lock_* to ensure coherency with disk.
2197
The default implementation does nothing; it is however possible
2198
for repositories to maintain loaded indices across multiple locks
2199
by checking inside their implementation of this method to see
2200
whether their indices are still valid. This depends of course on
2201
the disk format being validatable in this manner. This method is
2202
also called by the refresh_data() public interface to cause a refresh
2203
to occur while in a write lock so that data inserted by a smart server
2204
push operation is visible on the client's instance of the physical
2209
def revision_tree(self, revision_id):
2210
"""Return Tree for a revision on this branch.
2212
`revision_id` may be NULL_REVISION for the empty tree revision.
2214
revision_id = _mod_revision.ensure_null(revision_id)
2215
# TODO: refactor this to use an existing revision object
2216
# so we don't need to read it in twice.
2217
if revision_id == _mod_revision.NULL_REVISION:
2218
return RevisionTree(self, Inventory(root_id=None),
2219
_mod_revision.NULL_REVISION)
2221
inv = self.get_revision_inventory(revision_id)
2222
return RevisionTree(self, inv, revision_id)
2224
def revision_trees(self, revision_ids):
2225
"""Return Trees for revisions in this repository.
2227
:param revision_ids: a sequence of revision-ids;
2228
a revision-id may not be None or 'null:'
2230
inventories = self.iter_inventories(revision_ids)
2231
for inv in inventories:
2232
yield RevisionTree(self, inv, inv.revision_id)
2234
def _filtered_revision_trees(self, revision_ids, file_ids):
2235
"""Return Tree for a revision on this branch with only some files.
2237
:param revision_ids: a sequence of revision-ids;
2238
a revision-id may not be None or 'null:'
2239
:param file_ids: if not None, the result is filtered
2240
so that only those file-ids, their parents and their
2241
children are included.
2243
inventories = self.iter_inventories(revision_ids)
2244
for inv in inventories:
2245
# Should we introduce a FilteredRevisionTree class rather
2246
# than pre-filter the inventory here?
2247
filtered_inv = inv.filter(file_ids)
2248
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2251
def get_ancestry(self, revision_id, topo_sorted=True):
2252
"""Return a list of revision-ids integrated by a revision.
2254
The first element of the list is always None, indicating the origin
2255
revision. This might change when we have history horizons, or
2256
perhaps we should have a new API.
2258
This is topologically sorted.
2260
if _mod_revision.is_null(revision_id):
2262
if not self.has_revision(revision_id):
2263
raise errors.NoSuchRevision(self, revision_id)
2264
graph = self.get_graph()
2266
search = graph._make_breadth_first_searcher([revision_id])
2269
found, ghosts = search.next_with_ghosts()
2270
except StopIteration:
2273
if _mod_revision.NULL_REVISION in keys:
2274
keys.remove(_mod_revision.NULL_REVISION)
2276
parent_map = graph.get_parent_map(keys)
2277
keys = tsort.topo_sort(parent_map)
2278
return [None] + list(keys)
2281
"""Compress the data within the repository.
2283
This operation only makes sense for some repository types. For other
2284
types it should be a no-op that just returns.
2286
This stub method does not require a lock, but subclasses should use
2287
@needs_write_lock as this is a long running call its reasonable to
2288
implicitly lock for the user.
2291
def get_transaction(self):
2292
return self.control_files.get_transaction()
2294
def get_parent_map(self, revision_ids):
2295
"""See graph._StackedParentsProvider.get_parent_map"""
2296
# revisions index works in keys; this just works in revisions
2297
# therefore wrap and unwrap
2300
for revision_id in revision_ids:
2301
if revision_id == _mod_revision.NULL_REVISION:
2302
result[revision_id] = ()
2303
elif revision_id is None:
2304
raise ValueError('get_parent_map(None) is not valid')
2306
query_keys.append((revision_id ,))
2307
for ((revision_id,), parent_keys) in \
2308
self.revisions.get_parent_map(query_keys).iteritems():
2310
result[revision_id] = tuple(parent_revid
2311
for (parent_revid,) in parent_keys)
2313
result[revision_id] = (_mod_revision.NULL_REVISION,)
2316
def _make_parents_provider(self):
2319
def get_graph(self, other_repository=None):
2320
"""Return the graph walker for this repository format"""
2321
parents_provider = self._make_parents_provider()
2322
if (other_repository is not None and
2323
not self.has_same_location(other_repository)):
2324
parents_provider = graph._StackedParentsProvider(
2325
[parents_provider, other_repository._make_parents_provider()])
2326
return graph.Graph(parents_provider)
2328
def _get_versioned_file_checker(self, text_key_references=None):
2329
"""Return an object suitable for checking versioned files.
2331
:param text_key_references: if non-None, an already built
2332
dictionary mapping text keys ((fileid, revision_id) tuples)
2333
to whether they were referred to by the inventory of the
2334
revision_id that they contain. If None, this will be
2337
return _VersionedFileChecker(self,
2338
text_key_references=text_key_references)
2340
def revision_ids_to_search_result(self, result_set):
2341
"""Convert a set of revision ids to a graph SearchResult."""
2342
result_parents = set()
2343
for parents in self.get_graph().get_parent_map(
2344
result_set).itervalues():
2345
result_parents.update(parents)
2346
included_keys = result_set.intersection(result_parents)
2347
start_keys = result_set.difference(included_keys)
2348
exclude_keys = result_parents.difference(result_set)
2349
result = graph.SearchResult(start_keys, exclude_keys,
2350
len(result_set), result_set)
2354
def set_make_working_trees(self, new_value):
2355
"""Set the policy flag for making working trees when creating branches.
2357
This only applies to branches that use this repository.
2359
The default is 'True'.
2360
:param new_value: True to restore the default, False to disable making
2363
raise NotImplementedError(self.set_make_working_trees)
2365
def make_working_trees(self):
2366
"""Returns the policy for making working trees on new branches."""
2367
raise NotImplementedError(self.make_working_trees)
2370
def sign_revision(self, revision_id, gpg_strategy):
2371
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2372
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2375
def has_signature_for_revision_id(self, revision_id):
2376
"""Query for a revision signature for revision_id in the repository."""
2377
if not self.has_revision(revision_id):
2378
raise errors.NoSuchRevision(self, revision_id)
2379
sig_present = (1 == len(
2380
self.signatures.get_parent_map([(revision_id,)])))
2384
def get_signature_text(self, revision_id):
2385
"""Return the text for a signature."""
2386
stream = self.signatures.get_record_stream([(revision_id,)],
2388
record = stream.next()
2389
if record.storage_kind == 'absent':
2390
raise errors.NoSuchRevision(self, revision_id)
2391
return record.get_bytes_as('fulltext')
2394
def check(self, revision_ids=None):
2395
"""Check consistency of all history of given revision_ids.
2397
Different repository implementations should override _check().
2399
:param revision_ids: A non-empty list of revision_ids whose ancestry
2400
will be checked. Typically the last revision_id of a branch.
2402
return self._check(revision_ids)
2404
def _check(self, revision_ids):
2405
result = check.Check(self)
2409
def _warn_if_deprecated(self):
2410
global _deprecation_warning_done
2411
if _deprecation_warning_done:
2413
_deprecation_warning_done = True
2414
warning("Format %s for %s is deprecated - please use 'bzr upgrade' to get better performance"
2415
% (self._format, self.bzrdir.transport.base))
2417
def supports_rich_root(self):
2418
return self._format.rich_root_data
2420
def _check_ascii_revisionid(self, revision_id, method):
2421
"""Private helper for ascii-only repositories."""
2422
# weave repositories refuse to store revisionids that are non-ascii.
2423
if revision_id is not None:
2424
# weaves require ascii revision ids.
2425
if isinstance(revision_id, unicode):
2427
revision_id.encode('ascii')
2428
except UnicodeEncodeError:
2429
raise errors.NonAsciiRevisionId(method, self)
2432
revision_id.decode('ascii')
2433
except UnicodeDecodeError:
2434
raise errors.NonAsciiRevisionId(method, self)
2436
def revision_graph_can_have_wrong_parents(self):
2437
"""Is it possible for this repository to have a revision graph with
2440
If True, then this repository must also implement
2441
_find_inconsistent_revision_parents so that check and reconcile can
2442
check for inconsistencies before proceeding with other checks that may
2443
depend on the revision index being consistent.
2445
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2448
# remove these delegates a while after bzr 0.15
2449
def __make_delegated(name, from_module):
2450
def _deprecated_repository_forwarder():
2451
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2452
% (name, from_module),
2455
m = __import__(from_module, globals(), locals(), [name])
2457
return getattr(m, name)
2458
except AttributeError:
2459
raise AttributeError('module %s has no name %s'
2461
globals()[name] = _deprecated_repository_forwarder
2464
'AllInOneRepository',
2465
'WeaveMetaDirRepository',
2466
'PreSplitOutRepositoryFormat',
2467
'RepositoryFormat4',
2468
'RepositoryFormat5',
2469
'RepositoryFormat6',
2470
'RepositoryFormat7',
2472
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2476
'RepositoryFormatKnit',
2477
'RepositoryFormatKnit1',
2479
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2482
def install_revision(repository, rev, revision_tree):
2483
"""Install all revision data into a repository."""
2484
install_revisions(repository, [(rev, revision_tree, None)])
2487
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2488
"""Install all revision data into a repository.
2490
Accepts an iterable of revision, tree, signature tuples. The signature
2493
repository.start_write_group()
2495
for n, (revision, revision_tree, signature) in enumerate(iterable):
2496
_install_revision(repository, revision, revision_tree, signature)
2498
pb.update('Transferring revisions', n + 1, num_revisions)
2500
repository.abort_write_group()
2503
repository.commit_write_group()
2506
def _install_revision(repository, rev, revision_tree, signature):
2507
"""Install all revision data into a repository."""
2508
present_parents = []
2510
for p_id in rev.parent_ids:
2511
if repository.has_revision(p_id):
2512
present_parents.append(p_id)
2513
parent_trees[p_id] = repository.revision_tree(p_id)
2515
parent_trees[p_id] = repository.revision_tree(
2516
_mod_revision.NULL_REVISION)
2518
inv = revision_tree.inventory
2519
entries = inv.iter_entries()
2520
# backwards compatibility hack: skip the root id.
2521
if not repository.supports_rich_root():
2522
path, root = entries.next()
2523
if root.revision != rev.revision_id:
2524
raise errors.IncompatibleRevision(repr(repository))
2526
for path, ie in entries:
2527
text_keys[(ie.file_id, ie.revision)] = ie
2528
text_parent_map = repository.texts.get_parent_map(text_keys)
2529
missing_texts = set(text_keys) - set(text_parent_map)
2530
# Add the texts that are not already present
2531
for text_key in missing_texts:
2532
ie = text_keys[text_key]
2534
# FIXME: TODO: The following loop overlaps/duplicates that done by
2535
# commit to determine parents. There is a latent/real bug here where
2536
# the parents inserted are not those commit would do - in particular
2537
# they are not filtered by heads(). RBC, AB
2538
for revision, tree in parent_trees.iteritems():
2539
if ie.file_id not in tree:
2541
parent_id = tree.inventory[ie.file_id].revision
2542
if parent_id in text_parents:
2544
text_parents.append((ie.file_id, parent_id))
2545
lines = revision_tree.get_file(ie.file_id).readlines()
2546
repository.texts.add_lines(text_key, text_parents, lines)
2548
# install the inventory
2549
repository.add_inventory(rev.revision_id, inv, present_parents)
2550
except errors.RevisionAlreadyPresent:
2552
if signature is not None:
2553
repository.add_signature_text(rev.revision_id, signature)
2554
repository.add_revision(rev.revision_id, rev, inv)
2557
class MetaDirRepository(Repository):
2558
"""Repositories in the new meta-dir layout.
2560
:ivar _transport: Transport for access to repository control files,
2561
typically pointing to .bzr/repository.
2564
def __init__(self, _format, a_bzrdir, control_files):
2565
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2566
self._transport = control_files._transport
2568
def is_shared(self):
2569
"""Return True if this repository is flagged as a shared repository."""
2570
return self._transport.has('shared-storage')
2573
def set_make_working_trees(self, new_value):
2574
"""Set the policy flag for making working trees when creating branches.
2576
This only applies to branches that use this repository.
2578
The default is 'True'.
2579
:param new_value: True to restore the default, False to disable making
2584
self._transport.delete('no-working-trees')
2585
except errors.NoSuchFile:
2588
self._transport.put_bytes('no-working-trees', '',
2589
mode=self.bzrdir._get_file_mode())
2591
def make_working_trees(self):
2592
"""Returns the policy for making working trees on new branches."""
2593
return not self._transport.has('no-working-trees')
2596
class MetaDirVersionedFileRepository(MetaDirRepository):
2597
"""Repositories in a meta-dir, that work via versioned file objects."""
2599
def __init__(self, _format, a_bzrdir, control_files):
2600
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2604
network_format_registry = registry.FormatRegistry()
2605
"""Registry of formats indexed by their network name.
2607
The network name for a repository format is an identifier that can be used when
2608
referring to formats with smart server operations. See
2609
RepositoryFormat.network_name() for more detail.
2613
format_registry = registry.FormatRegistry(network_format_registry)
2614
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
2616
This can contain either format instances themselves, or classes/factories that
2617
can be called to obtain one.
2621
#####################################################################
2622
# Repository Formats
2624
class RepositoryFormat(object):
2625
"""A repository format.
2627
Formats provide four things:
2628
* An initialization routine to construct repository data on disk.
2629
* a optional format string which is used when the BzrDir supports
2631
* an open routine which returns a Repository instance.
2632
* A network name for referring to the format in smart server RPC
2635
There is one and only one Format subclass for each on-disk format. But
2636
there can be one Repository subclass that is used for several different
2637
formats. The _format attribute on a Repository instance can be used to
2638
determine the disk format.
2640
Formats are placed in a registry by their format string for reference
2641
during opening. These should be subclasses of RepositoryFormat for
2644
Once a format is deprecated, just deprecate the initialize and open
2645
methods on the format class. Do not deprecate the object, as the
2646
object may be created even when a repository instnace hasn't been
2649
Common instance attributes:
2650
_matchingbzrdir - the bzrdir format that the repository format was
2651
originally written to work with. This can be used if manually
2652
constructing a bzrdir and repository, or more commonly for test suite
2656
# Set to True or False in derived classes. True indicates that the format
2657
# supports ghosts gracefully.
2658
supports_ghosts = None
2659
# Can this repository be given external locations to lookup additional
2660
# data. Set to True or False in derived classes.
2661
supports_external_lookups = None
2662
# What order should fetch operations request streams in?
2663
# The default is unordered as that is the cheapest for an origin to
2665
_fetch_order = 'unordered'
2666
# Does this repository format use deltas that can be fetched as-deltas ?
2667
# (E.g. knits, where the knit deltas can be transplanted intact.
2668
# We default to False, which will ensure that enough data to get
2669
# a full text out of any fetch stream will be grabbed.
2670
_fetch_uses_deltas = False
2671
# Should fetch trigger a reconcile after the fetch? Only needed for
2672
# some repository formats that can suffer internal inconsistencies.
2673
_fetch_reconcile = False
2674
# Does this format have < O(tree_size) delta generation. Used to hint what
2675
# code path for commit, amongst other things.
2679
return "<%s>" % self.__class__.__name__
2681
def __eq__(self, other):
2682
# format objects are generally stateless
2683
return isinstance(other, self.__class__)
2685
def __ne__(self, other):
2686
return not self == other
2689
def find_format(klass, a_bzrdir):
2690
"""Return the format for the repository object in a_bzrdir.
2692
This is used by bzr native formats that have a "format" file in
2693
the repository. Other methods may be used by different types of
2697
transport = a_bzrdir.get_repository_transport(None)
2698
format_string = transport.get("format").read()
2699
return format_registry.get(format_string)
2700
except errors.NoSuchFile:
2701
raise errors.NoRepositoryPresent(a_bzrdir)
2703
raise errors.UnknownFormatError(format=format_string,
2707
def register_format(klass, format):
2708
format_registry.register(format.get_format_string(), format)
2711
def unregister_format(klass, format):
2712
format_registry.remove(format.get_format_string())
2715
def get_default_format(klass):
2716
"""Return the current default format."""
2717
from bzrlib import bzrdir
2718
return bzrdir.format_registry.make_bzrdir('default').repository_format
2720
def get_format_string(self):
2721
"""Return the ASCII format string that identifies this format.
2723
Note that in pre format ?? repositories the format string is
2724
not permitted nor written to disk.
2726
raise NotImplementedError(self.get_format_string)
2728
def get_format_description(self):
2729
"""Return the short description for this format."""
2730
raise NotImplementedError(self.get_format_description)
2732
# TODO: this shouldn't be in the base class, it's specific to things that
2733
# use weaves or knits -- mbp 20070207
2734
def _get_versioned_file_store(self,
2739
versionedfile_class=None,
2740
versionedfile_kwargs={},
2742
if versionedfile_class is None:
2743
versionedfile_class = self._versionedfile_class
2744
weave_transport = control_files._transport.clone(name)
2745
dir_mode = control_files._dir_mode
2746
file_mode = control_files._file_mode
2747
return VersionedFileStore(weave_transport, prefixed=prefixed,
2749
file_mode=file_mode,
2750
versionedfile_class=versionedfile_class,
2751
versionedfile_kwargs=versionedfile_kwargs,
2754
def initialize(self, a_bzrdir, shared=False):
2755
"""Initialize a repository of this format in a_bzrdir.
2757
:param a_bzrdir: The bzrdir to put the new repository in it.
2758
:param shared: The repository should be initialized as a sharable one.
2759
:returns: The new repository object.
2761
This may raise UninitializableFormat if shared repository are not
2762
compatible the a_bzrdir.
2764
raise NotImplementedError(self.initialize)
2766
def is_supported(self):
2767
"""Is this format supported?
2769
Supported formats must be initializable and openable.
2770
Unsupported formats may not support initialization or committing or
2771
some other features depending on the reason for not being supported.
2775
def network_name(self):
2776
"""A simple byte string uniquely identifying this format for RPC calls.
2778
MetaDir repository formats use their disk format string to identify the
2779
repository over the wire. All in one formats such as bzr < 0.8, and
2780
foreign formats like svn/git and hg should use some marker which is
2781
unique and immutable.
2783
raise NotImplementedError(self.network_name)
2785
def check_conversion_target(self, target_format):
2786
raise NotImplementedError(self.check_conversion_target)
2788
def open(self, a_bzrdir, _found=False):
2789
"""Return an instance of this format for the bzrdir a_bzrdir.
2791
_found is a private parameter, do not use it.
2793
raise NotImplementedError(self.open)
2796
class MetaDirRepositoryFormat(RepositoryFormat):
2797
"""Common base class for the new repositories using the metadir layout."""
2799
rich_root_data = False
2800
supports_tree_reference = False
2801
supports_external_lookups = False
2804
def _matchingbzrdir(self):
2805
matching = bzrdir.BzrDirMetaFormat1()
2806
matching.repository_format = self
2810
super(MetaDirRepositoryFormat, self).__init__()
2812
def _create_control_files(self, a_bzrdir):
2813
"""Create the required files and the initial control_files object."""
2814
# FIXME: RBC 20060125 don't peek under the covers
2815
# NB: no need to escape relative paths that are url safe.
2816
repository_transport = a_bzrdir.get_repository_transport(self)
2817
control_files = lockable_files.LockableFiles(repository_transport,
2818
'lock', lockdir.LockDir)
2819
control_files.create_lock()
2820
return control_files
2822
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
2823
"""Upload the initial blank content."""
2824
control_files = self._create_control_files(a_bzrdir)
2825
control_files.lock_write()
2826
transport = control_files._transport
2828
utf8_files += [('shared-storage', '')]
2830
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
2831
for (filename, content_stream) in files:
2832
transport.put_file(filename, content_stream,
2833
mode=a_bzrdir._get_file_mode())
2834
for (filename, content_bytes) in utf8_files:
2835
transport.put_bytes_non_atomic(filename, content_bytes,
2836
mode=a_bzrdir._get_file_mode())
2838
control_files.unlock()
2840
def network_name(self):
2841
"""Metadir formats have matching disk and network format strings."""
2842
return self.get_format_string()
2845
# Pre-0.8 formats that don't have a disk format string (because they are
2846
# versioned by the matching control directory). We use the control directories
2847
# disk format string as a key for the network_name because they meet the
2848
# constraints (simple string, unique, immmutable).
2849
network_format_registry.register_lazy(
2850
"Bazaar-NG branch, format 5\n",
2851
'bzrlib.repofmt.weaverepo',
2852
'RepositoryFormat5',
2854
network_format_registry.register_lazy(
2855
"Bazaar-NG branch, format 6\n",
2856
'bzrlib.repofmt.weaverepo',
2857
'RepositoryFormat6',
2860
# formats which have no format string are not discoverable or independently
2861
# creatable on disk, so are not registered in format_registry. They're
2862
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
2863
# needed, it's constructed directly by the BzrDir. Non-native formats where
2864
# the repository is not separately opened are similar.
2866
format_registry.register_lazy(
2867
'Bazaar-NG Repository format 7',
2868
'bzrlib.repofmt.weaverepo',
2872
format_registry.register_lazy(
2873
'Bazaar-NG Knit Repository Format 1',
2874
'bzrlib.repofmt.knitrepo',
2875
'RepositoryFormatKnit1',
2878
format_registry.register_lazy(
2879
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
2880
'bzrlib.repofmt.knitrepo',
2881
'RepositoryFormatKnit3',
2884
format_registry.register_lazy(
2885
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
2886
'bzrlib.repofmt.knitrepo',
2887
'RepositoryFormatKnit4',
2890
# Pack-based formats. There is one format for pre-subtrees, and one for
2891
# post-subtrees to allow ease of testing.
2892
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
2893
format_registry.register_lazy(
2894
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
2895
'bzrlib.repofmt.pack_repo',
2896
'RepositoryFormatKnitPack1',
2898
format_registry.register_lazy(
2899
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
2900
'bzrlib.repofmt.pack_repo',
2901
'RepositoryFormatKnitPack3',
2903
format_registry.register_lazy(
2904
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
2905
'bzrlib.repofmt.pack_repo',
2906
'RepositoryFormatKnitPack4',
2908
format_registry.register_lazy(
2909
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
2910
'bzrlib.repofmt.pack_repo',
2911
'RepositoryFormatKnitPack5',
2913
format_registry.register_lazy(
2914
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
2915
'bzrlib.repofmt.pack_repo',
2916
'RepositoryFormatKnitPack5RichRoot',
2918
format_registry.register_lazy(
2919
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
2920
'bzrlib.repofmt.pack_repo',
2921
'RepositoryFormatKnitPack5RichRootBroken',
2923
format_registry.register_lazy(
2924
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
2925
'bzrlib.repofmt.pack_repo',
2926
'RepositoryFormatKnitPack6',
2928
format_registry.register_lazy(
2929
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
2930
'bzrlib.repofmt.pack_repo',
2931
'RepositoryFormatKnitPack6RichRoot',
2934
# Development formats.
2935
# 1.7->1.8 go below here
2936
format_registry.register_lazy(
2937
"Bazaar development format 2 (needs bzr.dev from before 1.8)\n",
2938
'bzrlib.repofmt.pack_repo',
2939
'RepositoryFormatPackDevelopment2',
2941
format_registry.register_lazy(
2942
("Bazaar development format 2 with subtree support "
2943
"(needs bzr.dev from before 1.8)\n"),
2944
'bzrlib.repofmt.pack_repo',
2945
'RepositoryFormatPackDevelopment2Subtree',
2949
class InterRepository(InterObject):
2950
"""This class represents operations taking place between two repositories.
2952
Its instances have methods like copy_content and fetch, and contain
2953
references to the source and target repositories these operations can be
2956
Often we will provide convenience methods on 'repository' which carry out
2957
operations with another repository - they will always forward to
2958
InterRepository.get(other).method_name(parameters).
2961
_walk_to_common_revisions_batch_size = 50
2963
"""The available optimised InterRepository types."""
2966
def copy_content(self, revision_id=None):
2967
"""Make a complete copy of the content in self into destination.
2969
This is a destructive operation! Do not use it on existing
2972
:param revision_id: Only copy the content needed to construct
2973
revision_id and its parents.
2976
self.target.set_make_working_trees(self.source.make_working_trees())
2977
except NotImplementedError:
2979
self.target.fetch(self.source, revision_id=revision_id)
2982
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
2984
"""Fetch the content required to construct revision_id.
2986
The content is copied from self.source to self.target.
2988
:param revision_id: if None all content is copied, if NULL_REVISION no
2990
:param pb: optional progress bar to use for progress reports. If not
2991
provided a default one will be created.
2994
from bzrlib.fetch import RepoFetcher
2995
f = RepoFetcher(to_repository=self.target,
2996
from_repository=self.source,
2997
last_revision=revision_id,
2998
fetch_spec=fetch_spec,
2999
pb=pb, find_ghosts=find_ghosts)
3001
def _walk_to_common_revisions(self, revision_ids):
3002
"""Walk out from revision_ids in source to revisions target has.
3004
:param revision_ids: The start point for the search.
3005
:return: A set of revision ids.
3007
target_graph = self.target.get_graph()
3008
revision_ids = frozenset(revision_ids)
3009
# Fast path for the case where all the revisions are already in the
3011
# (Although this does incur an extra round trip for the
3012
# fairly common case where the target doesn't already have the revision
3014
if set(target_graph.get_parent_map(revision_ids)) == revision_ids:
3015
return graph.SearchResult(revision_ids, set(), 0, set())
3016
missing_revs = set()
3017
source_graph = self.source.get_graph()
3018
# ensure we don't pay silly lookup costs.
3019
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3020
null_set = frozenset([_mod_revision.NULL_REVISION])
3021
searcher_exhausted = False
3025
# Iterate the searcher until we have enough next_revs
3026
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3028
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3029
next_revs.update(next_revs_part)
3030
ghosts.update(ghosts_part)
3031
except StopIteration:
3032
searcher_exhausted = True
3034
# If there are ghosts in the source graph, and the caller asked for
3035
# them, make sure that they are present in the target.
3036
# We don't care about other ghosts as we can't fetch them and
3037
# haven't been asked to.
3038
ghosts_to_check = set(revision_ids.intersection(ghosts))
3039
revs_to_get = set(next_revs).union(ghosts_to_check)
3041
have_revs = set(target_graph.get_parent_map(revs_to_get))
3042
# we always have NULL_REVISION present.
3043
have_revs = have_revs.union(null_set)
3044
# Check if the target is missing any ghosts we need.
3045
ghosts_to_check.difference_update(have_revs)
3047
# One of the caller's revision_ids is a ghost in both the
3048
# source and the target.
3049
raise errors.NoSuchRevision(
3050
self.source, ghosts_to_check.pop())
3051
missing_revs.update(next_revs - have_revs)
3052
# Because we may have walked past the original stop point, make
3053
# sure everything is stopped
3054
stop_revs = searcher.find_seen_ancestors(have_revs)
3055
searcher.stop_searching_any(stop_revs)
3056
if searcher_exhausted:
3058
return searcher.get_result()
3061
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3062
"""Return the revision ids that source has that target does not.
3064
:param revision_id: only return revision ids included by this
3066
:param find_ghosts: If True find missing revisions in deep history
3067
rather than just finding the surface difference.
3068
:return: A bzrlib.graph.SearchResult.
3070
# stop searching at found target revisions.
3071
if not find_ghosts and revision_id is not None:
3072
return self._walk_to_common_revisions([revision_id])
3073
# generic, possibly worst case, slow code path.
3074
target_ids = set(self.target.all_revision_ids())
3075
if revision_id is not None:
3076
source_ids = self.source.get_ancestry(revision_id)
3077
if source_ids[0] is not None:
3078
raise AssertionError()
3081
source_ids = self.source.all_revision_ids()
3082
result_set = set(source_ids).difference(target_ids)
3083
return self.source.revision_ids_to_search_result(result_set)
3086
def _same_model(source, target):
3087
"""True if source and target have the same data representation.
3089
Note: this is always called on the base class; overriding it in a
3090
subclass will have no effect.
3093
InterRepository._assert_same_model(source, target)
3095
except errors.IncompatibleRepositories, e:
3099
def _assert_same_model(source, target):
3100
"""Raise an exception if two repositories do not use the same model.
3102
if source.supports_rich_root() != target.supports_rich_root():
3103
raise errors.IncompatibleRepositories(source, target,
3104
"different rich-root support")
3105
if source._serializer != target._serializer:
3106
raise errors.IncompatibleRepositories(source, target,
3107
"different serializers")
3110
class InterSameDataRepository(InterRepository):
3111
"""Code for converting between repositories that represent the same data.
3113
Data format and model must match for this to work.
3117
def _get_repo_format_to_test(self):
3118
"""Repository format for testing with.
3120
InterSameData can pull from subtree to subtree and from non-subtree to
3121
non-subtree, so we test this with the richest repository format.
3123
from bzrlib.repofmt import knitrepo
3124
return knitrepo.RepositoryFormatKnit3()
3127
def is_compatible(source, target):
3128
return InterRepository._same_model(source, target)
3131
class InterWeaveRepo(InterSameDataRepository):
3132
"""Optimised code paths between Weave based repositories.
3134
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3135
implemented lazy inter-object optimisation.
3139
def _get_repo_format_to_test(self):
3140
from bzrlib.repofmt import weaverepo
3141
return weaverepo.RepositoryFormat7()
3144
def is_compatible(source, target):
3145
"""Be compatible with known Weave formats.
3147
We don't test for the stores being of specific types because that
3148
could lead to confusing results, and there is no need to be
3151
from bzrlib.repofmt.weaverepo import (
3157
return (isinstance(source._format, (RepositoryFormat5,
3159
RepositoryFormat7)) and
3160
isinstance(target._format, (RepositoryFormat5,
3162
RepositoryFormat7)))
3163
except AttributeError:
3167
def copy_content(self, revision_id=None):
3168
"""See InterRepository.copy_content()."""
3169
# weave specific optimised path:
3171
self.target.set_make_working_trees(self.source.make_working_trees())
3172
except (errors.RepositoryUpgradeRequired, NotImplemented):
3174
# FIXME do not peek!
3175
if self.source._transport.listable():
3176
pb = ui.ui_factory.nested_progress_bar()
3178
self.target.texts.insert_record_stream(
3179
self.source.texts.get_record_stream(
3180
self.source.texts.keys(), 'topological', False))
3181
pb.update('copying inventory', 0, 1)
3182
self.target.inventories.insert_record_stream(
3183
self.source.inventories.get_record_stream(
3184
self.source.inventories.keys(), 'topological', False))
3185
self.target.signatures.insert_record_stream(
3186
self.source.signatures.get_record_stream(
3187
self.source.signatures.keys(),
3189
self.target.revisions.insert_record_stream(
3190
self.source.revisions.get_record_stream(
3191
self.source.revisions.keys(),
3192
'topological', True))
3196
self.target.fetch(self.source, revision_id=revision_id)
3199
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3200
"""See InterRepository.missing_revision_ids()."""
3201
# we want all revisions to satisfy revision_id in source.
3202
# but we don't want to stat every file here and there.
3203
# we want then, all revisions other needs to satisfy revision_id
3204
# checked, but not those that we have locally.
3205
# so the first thing is to get a subset of the revisions to
3206
# satisfy revision_id in source, and then eliminate those that
3207
# we do already have.
3208
# this is slow on high latency connection to self, but as as this
3209
# disk format scales terribly for push anyway due to rewriting
3210
# inventory.weave, this is considered acceptable.
3212
if revision_id is not None:
3213
source_ids = self.source.get_ancestry(revision_id)
3214
if source_ids[0] is not None:
3215
raise AssertionError()
3218
source_ids = self.source._all_possible_ids()
3219
source_ids_set = set(source_ids)
3220
# source_ids is the worst possible case we may need to pull.
3221
# now we want to filter source_ids against what we actually
3222
# have in target, but don't try to check for existence where we know
3223
# we do not have a revision as that would be pointless.
3224
target_ids = set(self.target._all_possible_ids())
3225
possibly_present_revisions = target_ids.intersection(source_ids_set)
3226
actually_present_revisions = set(
3227
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3228
required_revisions = source_ids_set.difference(actually_present_revisions)
3229
if revision_id is not None:
3230
# we used get_ancestry to determine source_ids then we are assured all
3231
# revisions referenced are present as they are installed in topological order.
3232
# and the tip revision was validated by get_ancestry.
3233
result_set = required_revisions
3235
# if we just grabbed the possibly available ids, then
3236
# we only have an estimate of whats available and need to validate
3237
# that against the revision records.
3239
self.source._eliminate_revisions_not_present(required_revisions))
3240
return self.source.revision_ids_to_search_result(result_set)
3243
class InterKnitRepo(InterSameDataRepository):
3244
"""Optimised code paths between Knit based repositories."""
3247
def _get_repo_format_to_test(self):
3248
from bzrlib.repofmt import knitrepo
3249
return knitrepo.RepositoryFormatKnit1()
3252
def is_compatible(source, target):
3253
"""Be compatible with known Knit formats.
3255
We don't test for the stores being of specific types because that
3256
could lead to confusing results, and there is no need to be
3259
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3261
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3262
isinstance(target._format, RepositoryFormatKnit))
3263
except AttributeError:
3265
return are_knits and InterRepository._same_model(source, target)
3268
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3269
"""See InterRepository.missing_revision_ids()."""
3270
if revision_id is not None:
3271
source_ids = self.source.get_ancestry(revision_id)
3272
if source_ids[0] is not None:
3273
raise AssertionError()
3276
source_ids = self.source.all_revision_ids()
3277
source_ids_set = set(source_ids)
3278
# source_ids is the worst possible case we may need to pull.
3279
# now we want to filter source_ids against what we actually
3280
# have in target, but don't try to check for existence where we know
3281
# we do not have a revision as that would be pointless.
3282
target_ids = set(self.target.all_revision_ids())
3283
possibly_present_revisions = target_ids.intersection(source_ids_set)
3284
actually_present_revisions = set(
3285
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3286
required_revisions = source_ids_set.difference(actually_present_revisions)
3287
if revision_id is not None:
3288
# we used get_ancestry to determine source_ids then we are assured all
3289
# revisions referenced are present as they are installed in topological order.
3290
# and the tip revision was validated by get_ancestry.
3291
result_set = required_revisions
3293
# if we just grabbed the possibly available ids, then
3294
# we only have an estimate of whats available and need to validate
3295
# that against the revision records.
3297
self.source._eliminate_revisions_not_present(required_revisions))
3298
return self.source.revision_ids_to_search_result(result_set)
3301
class InterPackRepo(InterSameDataRepository):
3302
"""Optimised code paths between Pack based repositories."""
3305
def _get_repo_format_to_test(self):
3306
from bzrlib.repofmt import pack_repo
3307
return pack_repo.RepositoryFormatKnitPack1()
3310
def is_compatible(source, target):
3311
"""Be compatible with known Pack formats.
3313
We don't test for the stores being of specific types because that
3314
could lead to confusing results, and there is no need to be
3317
from bzrlib.repofmt.pack_repo import RepositoryFormatPack
3319
are_packs = (isinstance(source._format, RepositoryFormatPack) and
3320
isinstance(target._format, RepositoryFormatPack))
3321
except AttributeError:
3323
return are_packs and InterRepository._same_model(source, target)
3326
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3328
"""See InterRepository.fetch()."""
3329
if (len(self.source._fallback_repositories) > 0 or
3330
len(self.target._fallback_repositories) > 0):
3331
# The pack layer is not aware of fallback repositories, so when
3332
# fetching from a stacked repository or into a stacked repository
3333
# we use the generic fetch logic which uses the VersionedFiles
3334
# attributes on repository.
3335
from bzrlib.fetch import RepoFetcher
3336
fetcher = RepoFetcher(self.target, self.source, revision_id,
3337
pb, find_ghosts, fetch_spec=fetch_spec)
3338
if fetch_spec is not None:
3339
if len(list(fetch_spec.heads)) != 1:
3340
raise AssertionError(
3341
"InterPackRepo.fetch doesn't support "
3342
"fetching multiple heads yet.")
3343
revision_id = list(fetch_spec.heads)[0]
3345
if revision_id is None:
3347
# everything to do - use pack logic
3348
# to fetch from all packs to one without
3349
# inventory parsing etc, IFF nothing to be copied is in the target.
3351
source_revision_ids = frozenset(self.source.all_revision_ids())
3352
revision_ids = source_revision_ids - \
3353
frozenset(self.target.get_parent_map(source_revision_ids))
3354
revision_keys = [(revid,) for revid in revision_ids]
3355
index = self.target._pack_collection.revision_index.combined_index
3356
present_revision_ids = set(item[1][0] for item in
3357
index.iter_entries(revision_keys))
3358
revision_ids = set(revision_ids) - present_revision_ids
3359
# implementing the TODO will involve:
3360
# - detecting when all of a pack is selected
3361
# - avoiding as much as possible pre-selection, so the
3362
# more-core routines such as create_pack_from_packs can filter in
3363
# a just-in-time fashion. (though having a HEADS list on a
3364
# repository might make this a lot easier, because we could
3365
# sensibly detect 'new revisions' without doing a full index scan.
3366
elif _mod_revision.is_null(revision_id):
3371
revision_ids = self.search_missing_revision_ids(revision_id,
3372
find_ghosts=find_ghosts).get_keys()
3373
except errors.NoSuchRevision:
3374
raise errors.InstallFailed([revision_id])
3375
if len(revision_ids) == 0:
3377
return self._pack(self.source, self.target, revision_ids)
3379
def _pack(self, source, target, revision_ids):
3380
from bzrlib.repofmt.pack_repo import Packer
3381
packs = source._pack_collection.all_packs()
3382
pack = Packer(self.target._pack_collection, packs, '.fetch',
3383
revision_ids).pack()
3384
if pack is not None:
3385
self.target._pack_collection._save_pack_names()
3386
copied_revs = pack.get_revision_count()
3387
# Trigger an autopack. This may duplicate effort as we've just done
3388
# a pack creation, but for now it is simpler to think about as
3389
# 'upload data, then repack if needed'.
3390
self.target._pack_collection.autopack()
3391
return (copied_revs, [])
3396
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3397
"""See InterRepository.missing_revision_ids().
3399
:param find_ghosts: Find ghosts throughout the ancestry of
3402
if not find_ghosts and revision_id is not None:
3403
return self._walk_to_common_revisions([revision_id])
3404
elif revision_id is not None:
3405
# Find ghosts: search for revisions pointing from one repository to
3406
# the other, and vice versa, anywhere in the history of revision_id.
3407
graph = self.target.get_graph(other_repository=self.source)
3408
searcher = graph._make_breadth_first_searcher([revision_id])
3412
next_revs, ghosts = searcher.next_with_ghosts()
3413
except StopIteration:
3415
if revision_id in ghosts:
3416
raise errors.NoSuchRevision(self.source, revision_id)
3417
found_ids.update(next_revs)
3418
found_ids.update(ghosts)
3419
found_ids = frozenset(found_ids)
3420
# Double query here: should be able to avoid this by changing the
3421
# graph api further.
3422
result_set = found_ids - frozenset(
3423
self.target.get_parent_map(found_ids))
3425
source_ids = self.source.all_revision_ids()
3426
# source_ids is the worst possible case we may need to pull.
3427
# now we want to filter source_ids against what we actually
3428
# have in target, but don't try to check for existence where we know
3429
# we do not have a revision as that would be pointless.
3430
target_ids = set(self.target.all_revision_ids())
3431
result_set = set(source_ids).difference(target_ids)
3432
return self.source.revision_ids_to_search_result(result_set)
3435
class InterDifferingSerializer(InterKnitRepo):
3438
def _get_repo_format_to_test(self):
3442
def is_compatible(source, target):
3443
"""Be compatible with Knit2 source and Knit3 target"""
3444
if source.supports_rich_root() != target.supports_rich_root():
3446
# Ideally, we'd support fetching if the source had no tree references
3447
# even if it supported them...
3448
if (getattr(source, '_format.supports_tree_reference', False) and
3449
not getattr(target, '_format.supports_tree_reference', False)):
3453
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3454
"""Get the best delta and base for this revision.
3456
:return: (basis_id, delta)
3458
possible_trees = [(parent_id, cache[parent_id])
3459
for parent_id in parent_ids
3460
if parent_id in cache]
3461
if len(possible_trees) == 0:
3462
# There either aren't any parents, or the parents aren't in the
3463
# cache, so just use the last converted tree
3464
possible_trees.append((basis_id, cache[basis_id]))
3466
for basis_id, basis_tree in possible_trees:
3467
delta = tree.inventory._make_delta(basis_tree.inventory)
3468
deltas.append((len(delta), basis_id, delta))
3470
return deltas[0][1:]
3472
def _fetch_batch(self, revision_ids, basis_id, cache):
3473
"""Fetch across a few revisions.
3475
:param revision_ids: The revisions to copy
3476
:param basis_id: The revision_id of a tree that must be in cache, used
3477
as a basis for delta when no other base is available
3478
:param cache: A cache of RevisionTrees that we can use.
3479
:return: The revision_id of the last converted tree. The RevisionTree
3480
for it will be in cache
3482
# Walk though all revisions; get inventory deltas, copy referenced
3483
# texts that delta references, insert the delta, revision and
3487
pending_revisions = []
3488
parent_map = self.source.get_parent_map(revision_ids)
3489
for tree in self.source.revision_trees(revision_ids):
3490
current_revision_id = tree.get_revision_id()
3491
parent_ids = parent_map.get(current_revision_id, ())
3492
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3494
# Find text entries that need to be copied
3495
for old_path, new_path, file_id, entry in delta:
3496
if new_path is not None:
3497
if not (new_path or self.target.supports_rich_root()):
3498
# We don't copy the text for the root node unless the
3499
# target supports_rich_root.
3501
text_keys.add((file_id, entry.revision))
3502
revision = self.source.get_revision(current_revision_id)
3503
pending_deltas.append((basis_id, delta,
3504
current_revision_id, revision.parent_ids))
3505
pending_revisions.append(revision)
3506
cache[current_revision_id] = tree
3507
basis_id = current_revision_id
3509
from_texts = self.source.texts
3510
to_texts = self.target.texts
3511
to_texts.insert_record_stream(from_texts.get_record_stream(
3512
text_keys, self.target._format._fetch_order,
3513
not self.target._format._fetch_uses_deltas))
3515
for delta in pending_deltas:
3516
self.target.add_inventory_by_delta(*delta)
3517
# insert signatures and revisions
3518
for revision in pending_revisions:
3520
signature = self.source.get_signature_text(
3521
revision.revision_id)
3522
self.target.add_signature_text(revision.revision_id,
3524
except errors.NoSuchRevision:
3526
self.target.add_revision(revision.revision_id, revision)
3529
def _fetch_all_revisions(self, revision_ids, pb):
3530
"""Fetch everything for the list of revisions.
3532
:param revision_ids: The list of revisions to fetch. Must be in
3534
:param pb: A ProgressBar
3537
basis_id, basis_tree = self._get_basis(revision_ids[0])
3539
cache = lru_cache.LRUCache(100)
3540
cache[basis_id] = basis_tree
3541
del basis_tree # We don't want to hang on to it here
3542
for offset in range(0, len(revision_ids), batch_size):
3543
self.target.start_write_group()
3545
pb.update('Transferring revisions', offset,
3547
batch = revision_ids[offset:offset+batch_size]
3548
basis_id = self._fetch_batch(batch, basis_id, cache)
3550
self.target.abort_write_group()
3553
self.target.commit_write_group()
3554
pb.update('Transferring revisions', len(revision_ids),
3558
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3560
"""See InterRepository.fetch()."""
3561
if fetch_spec is not None:
3562
raise AssertionError("Not implemented yet...")
3563
revision_ids = self.target.search_missing_revision_ids(self.source,
3564
revision_id, find_ghosts=find_ghosts).get_keys()
3565
if not revision_ids:
3567
revision_ids = tsort.topo_sort(
3568
self.source.get_graph().get_parent_map(revision_ids))
3570
my_pb = ui.ui_factory.nested_progress_bar()
3573
symbol_versioning.warn(
3574
symbol_versioning.deprecated_in((1, 14, 0))
3575
% "pb parameter to fetch()")
3578
self._fetch_all_revisions(revision_ids, pb)
3580
if my_pb is not None:
3582
return len(revision_ids), 0
3584
def _get_basis(self, first_revision_id):
3585
"""Get a revision and tree which exists in the target.
3587
This assumes that first_revision_id is selected for transmission
3588
because all other ancestors are already present. If we can't find an
3589
ancestor we fall back to NULL_REVISION since we know that is safe.
3591
:return: (basis_id, basis_tree)
3593
first_rev = self.source.get_revision(first_revision_id)
3595
basis_id = first_rev.parent_ids[0]
3596
# only valid as a basis if the target has it
3597
self.target.get_revision(basis_id)
3598
# Try to get a basis tree - if its a ghost it will hit the
3599
# NoSuchRevision case.
3600
basis_tree = self.source.revision_tree(basis_id)
3601
except (IndexError, errors.NoSuchRevision):
3602
basis_id = _mod_revision.NULL_REVISION
3603
basis_tree = self.source.revision_tree(basis_id)
3604
return basis_id, basis_tree
3607
InterRepository.register_optimiser(InterDifferingSerializer)
3608
InterRepository.register_optimiser(InterSameDataRepository)
3609
InterRepository.register_optimiser(InterWeaveRepo)
3610
InterRepository.register_optimiser(InterKnitRepo)
3611
InterRepository.register_optimiser(InterPackRepo)
3614
class CopyConverter(object):
3615
"""A repository conversion tool which just performs a copy of the content.
3617
This is slow but quite reliable.
3620
def __init__(self, target_format):
3621
"""Create a CopyConverter.
3623
:param target_format: The format the resulting repository should be.
3625
self.target_format = target_format
3627
def convert(self, repo, pb):
3628
"""Perform the conversion of to_convert, giving feedback via pb.
3630
:param to_convert: The disk object to convert.
3631
:param pb: a progress bar to use for progress information.
3636
# this is only useful with metadir layouts - separated repo content.
3637
# trigger an assertion if not such
3638
repo._format.get_format_string()
3639
self.repo_dir = repo.bzrdir
3640
self.step('Moving repository to repository.backup')
3641
self.repo_dir.transport.move('repository', 'repository.backup')
3642
backup_transport = self.repo_dir.transport.clone('repository.backup')
3643
repo._format.check_conversion_target(self.target_format)
3644
self.source_repo = repo._format.open(self.repo_dir,
3646
_override_transport=backup_transport)
3647
self.step('Creating new repository')
3648
converted = self.target_format.initialize(self.repo_dir,
3649
self.source_repo.is_shared())
3650
converted.lock_write()
3652
self.step('Copying content into repository.')
3653
self.source_repo.copy_content_into(converted)
3656
self.step('Deleting old repository content.')
3657
self.repo_dir.transport.delete_tree('repository.backup')
3658
self.pb.note('repository converted')
3660
def step(self, message):
3661
"""Update the pb by a step."""
3663
self.pb.update(message, self.count, self.total)
3675
def _unescaper(match, _map=_unescape_map):
3676
code = match.group(1)
3680
if not code.startswith('#'):
3682
return unichr(int(code[1:])).encode('utf8')
3688
def _unescape_xml(data):
3689
"""Unescape predefined XML entities in a string of data."""
3691
if _unescape_re is None:
3692
_unescape_re = re.compile('\&([^;]*);')
3693
return _unescape_re.sub(_unescaper, data)
3696
class _VersionedFileChecker(object):
3698
def __init__(self, repository, text_key_references=None):
3699
self.repository = repository
3700
self.text_index = self.repository._generate_text_key_index(
3701
text_key_references=text_key_references)
3703
def calculate_file_version_parents(self, text_key):
3704
"""Calculate the correct parents for a file version according to
3707
parent_keys = self.text_index[text_key]
3708
if parent_keys == [_mod_revision.NULL_REVISION]:
3710
return tuple(parent_keys)
3712
def check_file_version_parents(self, texts, progress_bar=None):
3713
"""Check the parents stored in a versioned file are correct.
3715
It also detects file versions that are not referenced by their
3716
corresponding revision's inventory.
3718
:returns: A tuple of (wrong_parents, dangling_file_versions).
3719
wrong_parents is a dict mapping {revision_id: (stored_parents,
3720
correct_parents)} for each revision_id where the stored parents
3721
are not correct. dangling_file_versions is a set of (file_id,
3722
revision_id) tuples for versions that are present in this versioned
3723
file, but not used by the corresponding inventory.
3726
self.file_ids = set([file_id for file_id, _ in
3727
self.text_index.iterkeys()])
3728
# text keys is now grouped by file_id
3729
n_weaves = len(self.file_ids)
3730
files_in_revisions = {}
3731
revisions_of_files = {}
3732
n_versions = len(self.text_index)
3733
progress_bar.update('loading text store', 0, n_versions)
3734
parent_map = self.repository.texts.get_parent_map(self.text_index)
3735
# On unlistable transports this could well be empty/error...
3736
text_keys = self.repository.texts.keys()
3737
unused_keys = frozenset(text_keys) - set(self.text_index)
3738
for num, key in enumerate(self.text_index.iterkeys()):
3739
if progress_bar is not None:
3740
progress_bar.update('checking text graph', num, n_versions)
3741
correct_parents = self.calculate_file_version_parents(key)
3743
knit_parents = parent_map[key]
3744
except errors.RevisionNotPresent:
3747
if correct_parents != knit_parents:
3748
wrong_parents[key] = (knit_parents, correct_parents)
3749
return wrong_parents, unused_keys
3752
def _old_get_graph(repository, revision_id):
3753
"""DO NOT USE. That is all. I'm serious."""
3754
graph = repository.get_graph()
3755
revision_graph = dict(((key, value) for key, value in
3756
graph.iter_ancestry([revision_id]) if value is not None))
3757
return _strip_NULL_ghosts(revision_graph)
3760
def _strip_NULL_ghosts(revision_graph):
3761
"""Also don't use this. more compatibility code for unmigrated clients."""
3762
# Filter ghosts, and null:
3763
if _mod_revision.NULL_REVISION in revision_graph:
3764
del revision_graph[_mod_revision.NULL_REVISION]
3765
for key, parents in revision_graph.items():
3766
revision_graph[key] = tuple(parent for parent in parents if parent
3768
return revision_graph
3771
class StreamSink(object):
3772
"""An object that can insert a stream into a repository.
3774
This interface handles the complexity of reserialising inventories and
3775
revisions from different formats, and allows unidirectional insertion into
3776
stacked repositories without looking for the missing basis parents
3780
def __init__(self, target_repo):
3781
self.target_repo = target_repo
3783
def insert_stream(self, stream, src_format, resume_tokens):
3784
"""Insert a stream's content into the target repository.
3786
:param src_format: a bzr repository format.
3788
:return: a list of resume tokens and an iterable of keys additional
3789
items required before the insertion can be completed.
3791
self.target_repo.lock_write()
3794
self.target_repo.resume_write_group(resume_tokens)
3796
self.target_repo.start_write_group()
3798
# locked_insert_stream performs a commit|suspend.
3799
return self._locked_insert_stream(stream, src_format)
3801
self.target_repo.abort_write_group(suppress_errors=True)
3804
self.target_repo.unlock()
3806
def _locked_insert_stream(self, stream, src_format):
3807
to_serializer = self.target_repo._format._serializer
3808
src_serializer = src_format._serializer
3809
if to_serializer == src_serializer:
3810
# If serializers match and the target is a pack repository, set the
3811
# write cache size on the new pack. This avoids poor performance
3812
# on transports where append is unbuffered (such as
3813
# RemoteTransport). This is safe to do because nothing should read
3814
# back from the target repository while a stream with matching
3815
# serialization is being inserted.
3816
# The exception is that a delta record from the source that should
3817
# be a fulltext may need to be expanded by the target (see
3818
# test_fetch_revisions_with_deltas_into_pack); but we take care to
3819
# explicitly flush any buffered writes first in that rare case.
3821
new_pack = self.target_repo._pack_collection._new_pack
3822
except AttributeError:
3823
# Not a pack repository
3826
new_pack.set_write_cache_size(1024*1024)
3827
for substream_type, substream in stream:
3828
if substream_type == 'texts':
3829
self.target_repo.texts.insert_record_stream(substream)
3830
elif substream_type == 'inventories':
3831
if src_serializer == to_serializer:
3832
self.target_repo.inventories.insert_record_stream(
3835
self._extract_and_insert_inventories(
3836
substream, src_serializer)
3837
elif substream_type == 'revisions':
3838
# This may fallback to extract-and-insert more often than
3839
# required if the serializers are different only in terms of
3841
if src_serializer == to_serializer:
3842
self.target_repo.revisions.insert_record_stream(
3845
self._extract_and_insert_revisions(substream,
3847
elif substream_type == 'signatures':
3848
self.target_repo.signatures.insert_record_stream(substream)
3850
raise AssertionError('kaboom! %s' % (substream_type,))
3852
missing_keys = set()
3853
for prefix, versioned_file in (
3854
('texts', self.target_repo.texts),
3855
('inventories', self.target_repo.inventories),
3856
('revisions', self.target_repo.revisions),
3857
('signatures', self.target_repo.signatures),
3859
missing_keys.update((prefix,) + key for key in
3860
versioned_file.get_missing_compression_parent_keys())
3861
except NotImplementedError:
3862
# cannot even attempt suspending, and missing would have failed
3863
# during stream insertion.
3864
missing_keys = set()
3867
# suspend the write group and tell the caller what we is
3868
# missing. We know we can suspend or else we would not have
3869
# entered this code path. (All repositories that can handle
3870
# missing keys can handle suspending a write group).
3871
write_group_tokens = self.target_repo.suspend_write_group()
3872
return write_group_tokens, missing_keys
3873
self.target_repo.commit_write_group()
3876
def _extract_and_insert_inventories(self, substream, serializer):
3877
"""Generate a new inventory versionedfile in target, converting data.
3879
The inventory is retrieved from the source, (deserializing it), and
3880
stored in the target (reserializing it in a different format).
3882
for record in substream:
3883
bytes = record.get_bytes_as('fulltext')
3884
revision_id = record.key[0]
3885
inv = serializer.read_inventory_from_string(bytes, revision_id)
3886
parents = [key[0] for key in record.parents]
3887
self.target_repo.add_inventory(revision_id, inv, parents)
3889
def _extract_and_insert_revisions(self, substream, serializer):
3890
for record in substream:
3891
bytes = record.get_bytes_as('fulltext')
3892
revision_id = record.key[0]
3893
rev = serializer.read_revision_from_string(bytes)
3894
if rev.revision_id != revision_id:
3895
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
3896
self.target_repo.add_revision(revision_id, rev)
3899
if self.target_repo._format._fetch_reconcile:
3900
self.target_repo.reconcile()
3903
class StreamSource(object):
3904
"""A source of a stream for fetching between repositories."""
3906
def __init__(self, from_repository, to_format):
3907
"""Create a StreamSource streaming from from_repository."""
3908
self.from_repository = from_repository
3909
self.to_format = to_format
3911
def delta_on_metadata(self):
3912
"""Return True if delta's are permitted on metadata streams.
3914
That is on revisions and signatures.
3916
src_serializer = self.from_repository._format._serializer
3917
target_serializer = self.to_format._serializer
3918
return (self.to_format._fetch_uses_deltas and
3919
src_serializer == target_serializer)
3921
def _fetch_revision_texts(self, revs):
3922
# fetch signatures first and then the revision texts
3923
# may need to be a InterRevisionStore call here.
3924
from_sf = self.from_repository.signatures
3925
# A missing signature is just skipped.
3926
keys = [(rev_id,) for rev_id in revs]
3927
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
3929
self.to_format._fetch_order,
3930
not self.to_format._fetch_uses_deltas))
3931
# If a revision has a delta, this is actually expanded inside the
3932
# insert_record_stream code now, which is an alternate fix for
3934
from_rf = self.from_repository.revisions
3935
revisions = from_rf.get_record_stream(
3937
self.to_format._fetch_order,
3938
not self.delta_on_metadata())
3939
return [('signatures', signatures), ('revisions', revisions)]
3941
def _generate_root_texts(self, revs):
3942
"""This will be called by __fetch between fetching weave texts and
3943
fetching the inventory weave.
3945
Subclasses should override this if they need to generate root texts
3946
after fetching weave texts.
3948
if self._rich_root_upgrade():
3950
return bzrlib.fetch.Inter1and2Helper(
3951
self.from_repository).generate_root_texts(revs)
3955
def get_stream(self, search):
3957
revs = search.get_keys()
3958
graph = self.from_repository.get_graph()
3959
revs = list(graph.iter_topo_order(revs))
3960
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
3962
for knit_kind, file_id, revisions in data_to_fetch:
3963
if knit_kind != phase:
3965
# Make a new progress bar for this phase
3966
if knit_kind == "file":
3967
# Accumulate file texts
3968
text_keys.extend([(file_id, revision) for revision in
3970
elif knit_kind == "inventory":
3971
# Now copy the file texts.
3972
from_texts = self.from_repository.texts
3973
yield ('texts', from_texts.get_record_stream(
3974
text_keys, self.to_format._fetch_order,
3975
not self.to_format._fetch_uses_deltas))
3976
# Cause an error if a text occurs after we have done the
3979
# Before we process the inventory we generate the root
3980
# texts (if necessary) so that the inventories references
3982
for _ in self._generate_root_texts(revs):
3984
# NB: This currently reopens the inventory weave in source;
3985
# using a single stream interface instead would avoid this.
3986
from_weave = self.from_repository.inventories
3987
# we fetch only the referenced inventories because we do not
3988
# know for unselected inventories whether all their required
3989
# texts are present in the other repository - it could be
3991
yield ('inventories', from_weave.get_record_stream(
3992
[(rev_id,) for rev_id in revs],
3993
self.inventory_fetch_order(),
3994
not self.delta_on_metadata()))
3995
elif knit_kind == "signatures":
3996
# Nothing to do here; this will be taken care of when
3997
# _fetch_revision_texts happens.
3999
elif knit_kind == "revisions":
4000
for record in self._fetch_revision_texts(revs):
4003
raise AssertionError("Unknown knit kind %r" % knit_kind)
4005
def get_stream_for_missing_keys(self, missing_keys):
4006
# missing keys can only occur when we are byte copying and not
4007
# translating (because translation means we don't send
4008
# unreconstructable deltas ever).
4010
keys['texts'] = set()
4011
keys['revisions'] = set()
4012
keys['inventories'] = set()
4013
keys['signatures'] = set()
4014
for key in missing_keys:
4015
keys[key[0]].add(key[1:])
4016
if len(keys['revisions']):
4017
# If we allowed copying revisions at this point, we could end up
4018
# copying a revision without copying its required texts: a
4019
# violation of the requirements for repository integrity.
4020
raise AssertionError(
4021
'cannot copy revisions to fill in missing deltas %s' % (
4022
keys['revisions'],))
4023
for substream_kind, keys in keys.iteritems():
4024
vf = getattr(self.from_repository, substream_kind)
4025
# Ask for full texts always so that we don't need more round trips
4026
# after this stream.
4027
stream = vf.get_record_stream(keys,
4028
self.to_format._fetch_order, True)
4029
yield substream_kind, stream
4031
def inventory_fetch_order(self):
4032
if self._rich_root_upgrade():
4033
return 'topological'
4035
return self.to_format._fetch_order
4037
def _rich_root_upgrade(self):
4038
return (not self.from_repository._format.rich_root_data and
4039
self.to_format.rich_root_data)
added
removed
bzrlib/repository.py
