1
# Copyright (C) 2005, 2006, 2007, 2008, 2009 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
from bzrlib.lazy_import import lazy_import
18
lazy_import(globals(), """
40
revision as _mod_revision,
46
from bzrlib.bundle import serializer
47
from bzrlib.revisiontree import RevisionTree
48
from bzrlib.store.versioned import VersionedFileStore
49
from bzrlib.testament import Testament
52
from bzrlib.decorators import needs_read_lock, needs_write_lock
53
from bzrlib.inter import InterObject
54
from bzrlib.inventory import (
60
from bzrlib import registry
61
from bzrlib.trace import (
62
log_exception_quietly, note, mutter, mutter_callsite, warning)
65
# Old formats display a warning, but only once
66
_deprecation_warning_done = False
69
class CommitBuilder(object):
70
"""Provides an interface to build up a commit.
72
This allows describing a tree to be committed without needing to
73
know the internals of the format of the repository.
76
# all clients should supply tree roots.
77
record_root_entry = True
78
# the default CommitBuilder does not manage trees whose root is versioned.
79
_versioned_root = False
81
def __init__(self, repository, parents, config, timestamp=None,
82
timezone=None, committer=None, revprops=None,
84
"""Initiate a CommitBuilder.
86
:param repository: Repository to commit to.
87
:param parents: Revision ids of the parents of the new revision.
88
:param config: Configuration to use.
89
:param timestamp: Optional timestamp recorded for commit.
90
:param timezone: Optional timezone for timestamp.
91
:param committer: Optional committer to set for commit.
92
:param revprops: Optional dictionary of revision properties.
93
:param revision_id: Optional revision id.
98
self._committer = self._config.username()
100
self._committer = committer
102
self.new_inventory = Inventory(None)
103
self._new_revision_id = revision_id
104
self.parents = parents
105
self.repository = repository
108
if revprops is not None:
109
self._validate_revprops(revprops)
110
self._revprops.update(revprops)
112
if timestamp is None:
113
timestamp = time.time()
114
# Restrict resolution to 1ms
115
self._timestamp = round(timestamp, 3)
118
self._timezone = osutils.local_time_offset()
120
self._timezone = int(timezone)
122
self._generate_revision_if_needed()
123
self.__heads = graph.HeadsCache(repository.get_graph()).heads
124
self._basis_delta = []
125
# API compatibility, older code that used CommitBuilder did not call
126
# .record_delete(), which means the delta that is computed would not be
127
# valid. Callers that will call record_delete() should call
128
# .will_record_deletes() to indicate that.
129
self._recording_deletes = False
130
# memo'd check for no-op commits.
131
self._any_changes = False
133
def any_changes(self):
134
"""Return True if any entries were changed.
136
This includes merge-only changes. It is the core for the --unchanged
139
:return: True if any changes have occured.
141
return self._any_changes
143
def _validate_unicode_text(self, text, context):
144
"""Verify things like commit messages don't have bogus characters."""
146
raise ValueError('Invalid value for %s: %r' % (context, text))
148
def _validate_revprops(self, revprops):
149
for key, value in revprops.iteritems():
150
# We know that the XML serializers do not round trip '\r'
151
# correctly, so refuse to accept them
152
if not isinstance(value, basestring):
153
raise ValueError('revision property (%s) is not a valid'
154
' (unicode) string: %r' % (key, value))
155
self._validate_unicode_text(value,
156
'revision property (%s)' % (key,))
158
def commit(self, message):
159
"""Make the actual commit.
161
:return: The revision id of the recorded revision.
163
self._validate_unicode_text(message, 'commit message')
164
rev = _mod_revision.Revision(
165
timestamp=self._timestamp,
166
timezone=self._timezone,
167
committer=self._committer,
169
inventory_sha1=self.inv_sha1,
170
revision_id=self._new_revision_id,
171
properties=self._revprops)
172
rev.parent_ids = self.parents
173
self.repository.add_revision(self._new_revision_id, rev,
174
self.new_inventory, self._config)
175
self.repository.commit_write_group()
176
return self._new_revision_id
179
"""Abort the commit that is being built.
181
self.repository.abort_write_group()
183
def revision_tree(self):
184
"""Return the tree that was just committed.
186
After calling commit() this can be called to get a RevisionTree
187
representing the newly committed tree. This is preferred to
188
calling Repository.revision_tree() because that may require
189
deserializing the inventory, while we already have a copy in
192
if self.new_inventory is None:
193
self.new_inventory = self.repository.get_inventory(
194
self._new_revision_id)
195
return RevisionTree(self.repository, self.new_inventory,
196
self._new_revision_id)
198
def finish_inventory(self):
199
"""Tell the builder that the inventory is finished.
201
:return: The inventory id in the repository, which can be used with
202
repository.get_inventory.
204
if self.new_inventory is None:
205
# an inventory delta was accumulated without creating a new
207
basis_id = self.basis_delta_revision
208
self.inv_sha1 = self.repository.add_inventory_by_delta(
209
basis_id, self._basis_delta, self._new_revision_id,
212
if self.new_inventory.root is None:
213
raise AssertionError('Root entry should be supplied to'
214
' record_entry_contents, as of bzr 0.10.')
215
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
216
self.new_inventory.revision_id = self._new_revision_id
217
self.inv_sha1 = self.repository.add_inventory(
218
self._new_revision_id,
222
return self._new_revision_id
224
def _gen_revision_id(self):
225
"""Return new revision-id."""
226
return generate_ids.gen_revision_id(self._config.username(),
229
def _generate_revision_if_needed(self):
230
"""Create a revision id if None was supplied.
232
If the repository can not support user-specified revision ids
233
they should override this function and raise CannotSetRevisionId
234
if _new_revision_id is not None.
236
:raises: CannotSetRevisionId
238
if self._new_revision_id is None:
239
self._new_revision_id = self._gen_revision_id()
240
self.random_revid = True
242
self.random_revid = False
244
def _heads(self, file_id, revision_ids):
245
"""Calculate the graph heads for revision_ids in the graph of file_id.
247
This can use either a per-file graph or a global revision graph as we
248
have an identity relationship between the two graphs.
250
return self.__heads(revision_ids)
252
def _check_root(self, ie, parent_invs, tree):
253
"""Helper for record_entry_contents.
255
:param ie: An entry being added.
256
:param parent_invs: The inventories of the parent revisions of the
258
:param tree: The tree that is being committed.
260
# In this revision format, root entries have no knit or weave When
261
# serializing out to disk and back in root.revision is always
263
ie.revision = self._new_revision_id
265
def _require_root_change(self, tree):
266
"""Enforce an appropriate root object change.
268
This is called once when record_iter_changes is called, if and only if
269
the root was not in the delta calculated by record_iter_changes.
271
:param tree: The tree which is being committed.
273
# NB: if there are no parents then this method is not called, so no
274
# need to guard on parents having length.
275
entry = entry_factory['directory'](tree.path2id(''), '',
277
entry.revision = self._new_revision_id
278
self._basis_delta.append(('', '', entry.file_id, entry))
280
def _get_delta(self, ie, basis_inv, path):
281
"""Get a delta against the basis inventory for ie."""
282
if ie.file_id not in basis_inv:
284
result = (None, path, ie.file_id, ie)
285
self._basis_delta.append(result)
287
elif ie != basis_inv[ie.file_id]:
289
# TODO: avoid tis id2path call.
290
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
291
self._basis_delta.append(result)
297
def get_basis_delta(self):
298
"""Return the complete inventory delta versus the basis inventory.
300
This has been built up with the calls to record_delete and
301
record_entry_contents. The client must have already called
302
will_record_deletes() to indicate that they will be generating a
305
:return: An inventory delta, suitable for use with apply_delta, or
306
Repository.add_inventory_by_delta, etc.
308
if not self._recording_deletes:
309
raise AssertionError("recording deletes not activated.")
310
return self._basis_delta
312
def record_delete(self, path, file_id):
313
"""Record that a delete occured against a basis tree.
315
This is an optional API - when used it adds items to the basis_delta
316
being accumulated by the commit builder. It cannot be called unless the
317
method will_record_deletes() has been called to inform the builder that
318
a delta is being supplied.
320
:param path: The path of the thing deleted.
321
:param file_id: The file id that was deleted.
323
if not self._recording_deletes:
324
raise AssertionError("recording deletes not activated.")
325
delta = (path, None, file_id, None)
326
self._basis_delta.append(delta)
327
self._any_changes = True
330
def will_record_deletes(self):
331
"""Tell the commit builder that deletes are being notified.
333
This enables the accumulation of an inventory delta; for the resulting
334
commit to be valid, deletes against the basis MUST be recorded via
335
builder.record_delete().
337
self._recording_deletes = True
339
basis_id = self.parents[0]
341
basis_id = _mod_revision.NULL_REVISION
342
self.basis_delta_revision = basis_id
344
def record_entry_contents(self, ie, parent_invs, path, tree,
346
"""Record the content of ie from tree into the commit if needed.
348
Side effect: sets ie.revision when unchanged
350
:param ie: An inventory entry present in the commit.
351
:param parent_invs: The inventories of the parent revisions of the
353
:param path: The path the entry is at in the tree.
354
:param tree: The tree which contains this entry and should be used to
356
:param content_summary: Summary data from the tree about the paths
357
content - stat, length, exec, sha/link target. This is only
358
accessed when the entry has a revision of None - that is when it is
359
a candidate to commit.
360
:return: A tuple (change_delta, version_recorded, fs_hash).
361
change_delta is an inventory_delta change for this entry against
362
the basis tree of the commit, or None if no change occured against
364
version_recorded is True if a new version of the entry has been
365
recorded. For instance, committing a merge where a file was only
366
changed on the other side will return (delta, False).
367
fs_hash is either None, or the hash details for the path (currently
368
a tuple of the contents sha1 and the statvalue returned by
369
tree.get_file_with_stat()).
371
if self.new_inventory.root is None:
372
if ie.parent_id is not None:
373
raise errors.RootMissing()
374
self._check_root(ie, parent_invs, tree)
375
if ie.revision is None:
376
kind = content_summary[0]
378
# ie is carried over from a prior commit
380
# XXX: repository specific check for nested tree support goes here - if
381
# the repo doesn't want nested trees we skip it ?
382
if (kind == 'tree-reference' and
383
not self.repository._format.supports_tree_reference):
384
# mismatch between commit builder logic and repository:
385
# this needs the entry creation pushed down into the builder.
386
raise NotImplementedError('Missing repository subtree support.')
387
self.new_inventory.add(ie)
389
# TODO: slow, take it out of the inner loop.
391
basis_inv = parent_invs[0]
393
basis_inv = Inventory(root_id=None)
395
# ie.revision is always None if the InventoryEntry is considered
396
# for committing. We may record the previous parents revision if the
397
# content is actually unchanged against a sole head.
398
if ie.revision is not None:
399
if not self._versioned_root and path == '':
400
# repositories that do not version the root set the root's
401
# revision to the new commit even when no change occurs (more
402
# specifically, they do not record a revision on the root; and
403
# the rev id is assigned to the root during deserialisation -
404
# this masks when a change may have occurred against the basis.
405
# To match this we always issue a delta, because the revision
406
# of the root will always be changing.
407
if ie.file_id in basis_inv:
408
delta = (basis_inv.id2path(ie.file_id), path,
412
delta = (None, path, ie.file_id, ie)
413
self._basis_delta.append(delta)
414
return delta, False, None
416
# we don't need to commit this, because the caller already
417
# determined that an existing revision of this file is
418
# appropriate. If its not being considered for committing then
419
# it and all its parents to the root must be unaltered so
420
# no-change against the basis.
421
if ie.revision == self._new_revision_id:
422
raise AssertionError("Impossible situation, a skipped "
423
"inventory entry (%r) claims to be modified in this "
424
"commit (%r).", (ie, self._new_revision_id))
425
return None, False, None
426
# XXX: Friction: parent_candidates should return a list not a dict
427
# so that we don't have to walk the inventories again.
428
parent_candiate_entries = ie.parent_candidates(parent_invs)
429
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
431
for inv in parent_invs:
432
if ie.file_id in inv:
433
old_rev = inv[ie.file_id].revision
434
if old_rev in head_set:
435
heads.append(inv[ie.file_id].revision)
436
head_set.remove(inv[ie.file_id].revision)
439
# now we check to see if we need to write a new record to the
441
# We write a new entry unless there is one head to the ancestors, and
442
# the kind-derived content is unchanged.
444
# Cheapest check first: no ancestors, or more the one head in the
445
# ancestors, we write a new node.
449
# There is a single head, look it up for comparison
450
parent_entry = parent_candiate_entries[heads[0]]
451
# if the non-content specific data has changed, we'll be writing a
453
if (parent_entry.parent_id != ie.parent_id or
454
parent_entry.name != ie.name):
456
# now we need to do content specific checks:
458
# if the kind changed the content obviously has
459
if kind != parent_entry.kind:
461
# Stat cache fingerprint feedback for the caller - None as we usually
462
# don't generate one.
465
if content_summary[2] is None:
466
raise ValueError("Files must not have executable = None")
468
if (# if the file length changed we have to store:
469
parent_entry.text_size != content_summary[1] or
470
# if the exec bit has changed we have to store:
471
parent_entry.executable != content_summary[2]):
473
elif parent_entry.text_sha1 == content_summary[3]:
474
# all meta and content is unchanged (using a hash cache
475
# hit to check the sha)
476
ie.revision = parent_entry.revision
477
ie.text_size = parent_entry.text_size
478
ie.text_sha1 = parent_entry.text_sha1
479
ie.executable = parent_entry.executable
480
return self._get_delta(ie, basis_inv, path), False, None
482
# Either there is only a hash change(no hash cache entry,
483
# or same size content change), or there is no change on
485
# Provide the parent's hash to the store layer, so that the
486
# content is unchanged we will not store a new node.
487
nostore_sha = parent_entry.text_sha1
489
# We want to record a new node regardless of the presence or
490
# absence of a content change in the file.
492
ie.executable = content_summary[2]
493
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
495
text = file_obj.read()
499
ie.text_sha1, ie.text_size = self._add_text_to_weave(
500
ie.file_id, text, heads, nostore_sha)
501
# Let the caller know we generated a stat fingerprint.
502
fingerprint = (ie.text_sha1, stat_value)
503
except errors.ExistingContent:
504
# Turns out that the file content was unchanged, and we were
505
# only going to store a new node if it was changed. Carry over
507
ie.revision = parent_entry.revision
508
ie.text_size = parent_entry.text_size
509
ie.text_sha1 = parent_entry.text_sha1
510
ie.executable = parent_entry.executable
511
return self._get_delta(ie, basis_inv, path), False, None
512
elif kind == 'directory':
514
# all data is meta here, nothing specific to directory, so
516
ie.revision = parent_entry.revision
517
return self._get_delta(ie, basis_inv, path), False, None
518
self._add_text_to_weave(ie.file_id, '', heads, None)
519
elif kind == 'symlink':
520
current_link_target = content_summary[3]
522
# symlink target is not generic metadata, check if it has
524
if current_link_target != parent_entry.symlink_target:
527
# unchanged, carry over.
528
ie.revision = parent_entry.revision
529
ie.symlink_target = parent_entry.symlink_target
530
return self._get_delta(ie, basis_inv, path), False, None
531
ie.symlink_target = current_link_target
532
self._add_text_to_weave(ie.file_id, '', heads, None)
533
elif kind == 'tree-reference':
535
if content_summary[3] != parent_entry.reference_revision:
538
# unchanged, carry over.
539
ie.reference_revision = parent_entry.reference_revision
540
ie.revision = parent_entry.revision
541
return self._get_delta(ie, basis_inv, path), False, None
542
ie.reference_revision = content_summary[3]
543
self._add_text_to_weave(ie.file_id, '', heads, None)
545
raise NotImplementedError('unknown kind')
546
ie.revision = self._new_revision_id
547
self._any_changes = True
548
return self._get_delta(ie, basis_inv, path), True, fingerprint
550
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
551
_entry_factory=entry_factory):
552
"""Record a new tree via iter_changes.
554
:param tree: The tree to obtain text contents from for changed objects.
555
:param basis_revision_id: The revision id of the tree the iter_changes
556
has been generated against. Currently assumed to be the same
557
as self.parents[0] - if it is not, errors may occur.
558
:param iter_changes: An iter_changes iterator with the changes to apply
559
to basis_revision_id. The iterator must not include any items with
560
a current kind of None - missing items must be either filtered out
561
or errored-on beefore record_iter_changes sees the item.
562
:param _entry_factory: Private method to bind entry_factory locally for
564
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
567
# Create an inventory delta based on deltas between all the parents and
568
# deltas between all the parent inventories. We use inventory delta's
569
# between the inventory objects because iter_changes masks
570
# last-changed-field only changes.
572
# file_id -> change map, change is fileid, paths, changed, versioneds,
573
# parents, names, kinds, executables
575
# {file_id -> revision_id -> inventory entry, for entries in parent
576
# trees that are not parents[0]
580
revtrees = list(self.repository.revision_trees(self.parents))
581
except errors.NoSuchRevision:
582
# one or more ghosts, slow path.
584
for revision_id in self.parents:
586
revtrees.append(self.repository.revision_tree(revision_id))
587
except errors.NoSuchRevision:
589
basis_revision_id = _mod_revision.NULL_REVISION
591
revtrees.append(self.repository.revision_tree(
592
_mod_revision.NULL_REVISION))
593
# The basis inventory from a repository
595
basis_inv = revtrees[0].inventory
597
basis_inv = self.repository.revision_tree(
598
_mod_revision.NULL_REVISION).inventory
599
if len(self.parents) > 0:
600
if basis_revision_id != self.parents[0] and not ghost_basis:
602
"arbitrary basis parents not yet supported with merges")
603
for revtree in revtrees[1:]:
604
for change in revtree.inventory._make_delta(basis_inv):
605
if change[1] is None:
606
# Not present in this parent.
608
if change[2] not in merged_ids:
609
if change[0] is not None:
610
basis_entry = basis_inv[change[2]]
611
merged_ids[change[2]] = [
613
basis_entry.revision,
616
parent_entries[change[2]] = {
618
basis_entry.revision:basis_entry,
620
change[3].revision:change[3],
623
merged_ids[change[2]] = [change[3].revision]
624
parent_entries[change[2]] = {change[3].revision:change[3]}
626
merged_ids[change[2]].append(change[3].revision)
627
parent_entries[change[2]][change[3].revision] = change[3]
630
# Setup the changes from the tree:
631
# changes maps file_id -> (change, [parent revision_ids])
633
for change in iter_changes:
634
# This probably looks up in basis_inv way to much.
635
if change[1][0] is not None:
636
head_candidate = [basis_inv[change[0]].revision]
639
changes[change[0]] = change, merged_ids.get(change[0],
641
unchanged_merged = set(merged_ids) - set(changes)
642
# Extend the changes dict with synthetic changes to record merges of
644
for file_id in unchanged_merged:
645
# Record a merged version of these items that did not change vs the
646
# basis. This can be either identical parallel changes, or a revert
647
# of a specific file after a merge. The recorded content will be
648
# that of the current tree (which is the same as the basis), but
649
# the per-file graph will reflect a merge.
650
# NB:XXX: We are reconstructing path information we had, this
651
# should be preserved instead.
652
# inv delta change: (file_id, (path_in_source, path_in_target),
653
# changed_content, versioned, parent, name, kind,
656
basis_entry = basis_inv[file_id]
657
except errors.NoSuchId:
658
# a change from basis->some_parents but file_id isn't in basis
659
# so was new in the merge, which means it must have changed
660
# from basis -> current, and as it hasn't the add was reverted
661
# by the user. So we discard this change.
665
(basis_inv.id2path(file_id), tree.id2path(file_id)),
667
(basis_entry.parent_id, basis_entry.parent_id),
668
(basis_entry.name, basis_entry.name),
669
(basis_entry.kind, basis_entry.kind),
670
(basis_entry.executable, basis_entry.executable))
671
changes[file_id] = (change, merged_ids[file_id])
672
# changes contains tuples with the change and a set of inventory
673
# candidates for the file.
675
# old_path, new_path, file_id, new_inventory_entry
676
seen_root = False # Is the root in the basis delta?
677
inv_delta = self._basis_delta
678
modified_rev = self._new_revision_id
679
for change, head_candidates in changes.values():
680
if change[3][1]: # versioned in target.
681
# Several things may be happening here:
682
# We may have a fork in the per-file graph
683
# - record a change with the content from tree
684
# We may have a change against < all trees
685
# - carry over the tree that hasn't changed
686
# We may have a change against all trees
687
# - record the change with the content from tree
690
entry = _entry_factory[kind](file_id, change[5][1],
692
head_set = self._heads(change[0], set(head_candidates))
695
for head_candidate in head_candidates:
696
if head_candidate in head_set:
697
heads.append(head_candidate)
698
head_set.remove(head_candidate)
701
# Could be a carry-over situation:
702
parent_entry_revs = parent_entries.get(file_id, None)
703
if parent_entry_revs:
704
parent_entry = parent_entry_revs.get(heads[0], None)
707
if parent_entry is None:
708
# The parent iter_changes was called against is the one
709
# that is the per-file head, so any change is relevant
710
# iter_changes is valid.
711
carry_over_possible = False
713
# could be a carry over situation
714
# A change against the basis may just indicate a merge,
715
# we need to check the content against the source of the
716
# merge to determine if it was changed after the merge
718
if (parent_entry.kind != entry.kind or
719
parent_entry.parent_id != entry.parent_id or
720
parent_entry.name != entry.name):
721
# Metadata common to all entries has changed
722
# against per-file parent
723
carry_over_possible = False
725
carry_over_possible = True
726
# per-type checks for changes against the parent_entry
729
# Cannot be a carry-over situation
730
carry_over_possible = False
731
# Populate the entry in the delta
733
# XXX: There is still a small race here: If someone reverts the content of a file
734
# after iter_changes examines and decides it has changed,
735
# we will unconditionally record a new version even if some
736
# other process reverts it while commit is running (with
737
# the revert happening after iter_changes did it's
740
entry.executable = True
742
entry.executable = False
743
if (carry_over_possible and
744
parent_entry.executable == entry.executable):
745
# Check the file length, content hash after reading
747
nostore_sha = parent_entry.text_sha1
750
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
752
text = file_obj.read()
756
entry.text_sha1, entry.text_size = self._add_text_to_weave(
757
file_id, text, heads, nostore_sha)
758
yield file_id, change[1][1], (entry.text_sha1, stat_value)
759
except errors.ExistingContent:
760
# No content change against a carry_over parent
761
# Perhaps this should also yield a fs hash update?
763
entry.text_size = parent_entry.text_size
764
entry.text_sha1 = parent_entry.text_sha1
765
elif kind == 'symlink':
767
entry.symlink_target = tree.get_symlink_target(file_id)
768
if (carry_over_possible and
769
parent_entry.symlink_target == entry.symlink_target):
772
self._add_text_to_weave(change[0], '', heads, None)
773
elif kind == 'directory':
774
if carry_over_possible:
777
# Nothing to set on the entry.
778
# XXX: split into the Root and nonRoot versions.
779
if change[1][1] != '' or self.repository.supports_rich_root():
780
self._add_text_to_weave(change[0], '', heads, None)
781
elif kind == 'tree-reference':
782
if not self.repository._format.supports_tree_reference:
783
# This isn't quite sane as an error, but we shouldn't
784
# ever see this code path in practice: tree's don't
785
# permit references when the repo doesn't support tree
787
raise errors.UnsupportedOperation(tree.add_reference,
789
reference_revision = tree.get_reference_revision(change[0])
790
entry.reference_revision = reference_revision
791
if (carry_over_possible and
792
parent_entry.reference_revision == reference_revision):
795
self._add_text_to_weave(change[0], '', heads, None)
797
raise AssertionError('unknown kind %r' % kind)
799
entry.revision = modified_rev
801
entry.revision = parent_entry.revision
804
new_path = change[1][1]
805
inv_delta.append((change[1][0], new_path, change[0], entry))
808
self.new_inventory = None
810
# This should perhaps be guarded by a check that the basis we
811
# commit against is the basis for the commit and if not do a delta
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_text, parents, nostore_sha):
820
parent_keys = tuple([(file_id, parent) for parent in parents])
821
return self.repository.texts._add_text(
822
(file_id, self._new_revision_id), parent_keys, new_text,
823
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
826
class RootCommitBuilder(CommitBuilder):
827
"""This commitbuilder actually records the root id"""
829
# the root entry gets versioned properly by this builder.
830
_versioned_root = True
832
def _check_root(self, ie, parent_invs, tree):
833
"""Helper for record_entry_contents.
835
:param ie: An entry being added.
836
:param parent_invs: The inventories of the parent revisions of the
838
:param tree: The tree that is being committed.
841
def _require_root_change(self, tree):
842
"""Enforce an appropriate root object change.
844
This is called once when record_iter_changes is called, if and only if
845
the root was not in the delta calculated by record_iter_changes.
847
:param tree: The tree which is being committed.
849
# versioned roots do not change unless the tree found a change.
852
######################################################################
856
class Repository(object):
857
"""Repository holding history for one or more branches.
859
The repository holds and retrieves historical information including
860
revisions and file history. It's normally accessed only by the Branch,
861
which views a particular line of development through that history.
863
The Repository builds on top of some byte storage facilies (the revisions,
864
signatures, inventories, texts and chk_bytes attributes) and a Transport,
865
which respectively provide byte storage and a means to access the (possibly
868
The byte storage facilities are addressed via tuples, which we refer to
869
as 'keys' throughout the code base. Revision_keys, inventory_keys and
870
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
871
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
872
byte string made up of a hash identifier and a hash value.
873
We use this interface because it allows low friction with the underlying
874
code that implements disk indices, network encoding and other parts of
877
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
878
the serialised revisions for the repository. This can be used to obtain
879
revision graph information or to access raw serialised revisions.
880
The result of trying to insert data into the repository via this store
881
is undefined: it should be considered read-only except for implementors
883
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
884
the serialised signatures for the repository. This can be used to
885
obtain access to raw serialised signatures. The result of trying to
886
insert data into the repository via this store is undefined: it should
887
be considered read-only except for implementors of repositories.
888
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
889
the serialised inventories for the repository. This can be used to
890
obtain unserialised inventories. The result of trying to insert data
891
into the repository via this store is undefined: it should be
892
considered read-only except for implementors of repositories.
893
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
894
texts of files and directories for the repository. This can be used to
895
obtain file texts or file graphs. Note that Repository.iter_file_bytes
896
is usually a better interface for accessing file texts.
897
The result of trying to insert data into the repository via this store
898
is undefined: it should be considered read-only except for implementors
900
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
901
any data the repository chooses to store or have indexed by its hash.
902
The result of trying to insert data into the repository via this store
903
is undefined: it should be considered read-only except for implementors
905
:ivar _transport: Transport for file access to repository, typically
906
pointing to .bzr/repository.
909
# What class to use for a CommitBuilder. Often its simpler to change this
910
# in a Repository class subclass rather than to override
911
# get_commit_builder.
912
_commit_builder_class = CommitBuilder
913
# The search regex used by xml based repositories to determine what things
914
# where changed in a single commit.
915
_file_ids_altered_regex = lazy_regex.lazy_compile(
916
r'file_id="(?P<file_id>[^"]+)"'
917
r'.* revision="(?P<revision_id>[^"]+)"'
920
def abort_write_group(self, suppress_errors=False):
921
"""Commit the contents accrued within the current write group.
923
:param suppress_errors: if true, abort_write_group will catch and log
924
unexpected errors that happen during the abort, rather than
925
allowing them to propagate. Defaults to False.
927
:seealso: start_write_group.
929
if self._write_group is not self.get_transaction():
930
# has an unlock or relock occured ?
933
'(suppressed) mismatched lock context and write group. %r, %r',
934
self._write_group, self.get_transaction())
936
raise errors.BzrError(
937
'mismatched lock context and write group. %r, %r' %
938
(self._write_group, self.get_transaction()))
940
self._abort_write_group()
941
except Exception, exc:
942
self._write_group = None
943
if not suppress_errors:
945
mutter('abort_write_group failed')
946
log_exception_quietly()
947
note('bzr: ERROR (ignored): %s', exc)
948
self._write_group = None
950
def _abort_write_group(self):
951
"""Template method for per-repository write group cleanup.
953
This is called during abort before the write group is considered to be
954
finished and should cleanup any internal state accrued during the write
955
group. There is no requirement that data handed to the repository be
956
*not* made available - this is not a rollback - but neither should any
957
attempt be made to ensure that data added is fully commited. Abort is
958
invoked when an error has occured so futher disk or network operations
959
may not be possible or may error and if possible should not be
963
def add_fallback_repository(self, repository):
964
"""Add a repository to use for looking up data not held locally.
966
:param repository: A repository.
968
if not self._format.supports_external_lookups:
969
raise errors.UnstackableRepositoryFormat(self._format, self.base)
971
# This repository will call fallback.unlock() when we transition to
972
# the unlocked state, so we make sure to increment the lock count
973
repository.lock_read()
974
self._check_fallback_repository(repository)
975
self._fallback_repositories.append(repository)
976
self.texts.add_fallback_versioned_files(repository.texts)
977
self.inventories.add_fallback_versioned_files(repository.inventories)
978
self.revisions.add_fallback_versioned_files(repository.revisions)
979
self.signatures.add_fallback_versioned_files(repository.signatures)
980
if self.chk_bytes is not None:
981
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
983
def _check_fallback_repository(self, repository):
984
"""Check that this repository can fallback to repository safely.
986
Raise an error if not.
988
:param repository: A repository to fallback to.
990
return InterRepository._assert_same_model(self, repository)
992
def add_inventory(self, revision_id, inv, parents):
993
"""Add the inventory inv to the repository as revision_id.
995
:param parents: The revision ids of the parents that revision_id
996
is known to have and are in the repository already.
998
:returns: The validator(which is a sha1 digest, though what is sha'd is
999
repository format specific) of the serialized inventory.
1001
if not self.is_in_write_group():
1002
raise AssertionError("%r not in write group" % (self,))
1003
_mod_revision.check_not_reserved_id(revision_id)
1004
if not (inv.revision_id is None or inv.revision_id == revision_id):
1005
raise AssertionError(
1006
"Mismatch between inventory revision"
1007
" id and insertion revid (%r, %r)"
1008
% (inv.revision_id, revision_id))
1009
if inv.root is None:
1010
raise AssertionError()
1011
return self._add_inventory_checked(revision_id, inv, parents)
1013
def _add_inventory_checked(self, revision_id, inv, parents):
1014
"""Add inv to the repository after checking the inputs.
1016
This function can be overridden to allow different inventory styles.
1018
:seealso: add_inventory, for the contract.
1020
inv_lines = self._serialise_inventory_to_lines(inv)
1021
return self._inventory_add_lines(revision_id, parents,
1022
inv_lines, check_content=False)
1024
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1025
parents, basis_inv=None, propagate_caches=False):
1026
"""Add a new inventory expressed as a delta against another revision.
1028
See the inventory developers documentation for the theory behind
1031
:param basis_revision_id: The inventory id the delta was created
1032
against. (This does not have to be a direct parent.)
1033
:param delta: The inventory delta (see Inventory.apply_delta for
1035
:param new_revision_id: The revision id that the inventory is being
1037
:param parents: The revision ids of the parents that revision_id is
1038
known to have and are in the repository already. These are supplied
1039
for repositories that depend on the inventory graph for revision
1040
graph access, as well as for those that pun ancestry with delta
1042
:param basis_inv: The basis inventory if it is already known,
1044
:param propagate_caches: If True, the caches for this inventory are
1045
copied to and updated for the result if possible.
1047
:returns: (validator, new_inv)
1048
The validator(which is a sha1 digest, though what is sha'd is
1049
repository format specific) of the serialized inventory, and the
1050
resulting inventory.
1052
if not self.is_in_write_group():
1053
raise AssertionError("%r not in write group" % (self,))
1054
_mod_revision.check_not_reserved_id(new_revision_id)
1055
basis_tree = self.revision_tree(basis_revision_id)
1056
basis_tree.lock_read()
1058
# Note that this mutates the inventory of basis_tree, which not all
1059
# inventory implementations may support: A better idiom would be to
1060
# return a new inventory, but as there is no revision tree cache in
1061
# repository this is safe for now - RBC 20081013
1062
if basis_inv is None:
1063
basis_inv = basis_tree.inventory
1064
basis_inv.apply_delta(delta)
1065
basis_inv.revision_id = new_revision_id
1066
return (self.add_inventory(new_revision_id, basis_inv, parents),
1071
def _inventory_add_lines(self, revision_id, parents, lines,
1072
check_content=True):
1073
"""Store lines in inv_vf and return the sha1 of the inventory."""
1074
parents = [(parent,) for parent in parents]
1075
result = self.inventories.add_lines((revision_id,), parents, lines,
1076
check_content=check_content)[0]
1077
self.inventories._access.flush()
1080
def add_revision(self, revision_id, rev, inv=None, config=None):
1081
"""Add rev to the revision store as revision_id.
1083
:param revision_id: the revision id to use.
1084
:param rev: The revision object.
1085
:param inv: The inventory for the revision. if None, it will be looked
1086
up in the inventory storer
1087
:param config: If None no digital signature will be created.
1088
If supplied its signature_needed method will be used
1089
to determine if a signature should be made.
1091
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1093
_mod_revision.check_not_reserved_id(revision_id)
1094
if config is not None and config.signature_needed():
1096
inv = self.get_inventory(revision_id)
1097
plaintext = Testament(rev, inv).as_short_text()
1098
self.store_revision_signature(
1099
gpg.GPGStrategy(config), plaintext, revision_id)
1100
# check inventory present
1101
if not self.inventories.get_parent_map([(revision_id,)]):
1103
raise errors.WeaveRevisionNotPresent(revision_id,
1106
# yes, this is not suitable for adding with ghosts.
1107
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1110
key = (revision_id,)
1111
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1112
self._add_revision(rev)
1114
def _add_revision(self, revision):
1115
text = self._serializer.write_revision_to_string(revision)
1116
key = (revision.revision_id,)
1117
parents = tuple((parent,) for parent in revision.parent_ids)
1118
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1120
def all_revision_ids(self):
1121
"""Returns a list of all the revision ids in the repository.
1123
This is conceptually deprecated because code should generally work on
1124
the graph reachable from a particular revision, and ignore any other
1125
revisions that might be present. There is no direct replacement
1128
if 'evil' in debug.debug_flags:
1129
mutter_callsite(2, "all_revision_ids is linear with history.")
1130
return self._all_revision_ids()
1132
def _all_revision_ids(self):
1133
"""Returns a list of all the revision ids in the repository.
1135
These are in as much topological order as the underlying store can
1138
raise NotImplementedError(self._all_revision_ids)
1140
def break_lock(self):
1141
"""Break a lock if one is present from another instance.
1143
Uses the ui factory to ask for confirmation if the lock may be from
1146
self.control_files.break_lock()
1149
def _eliminate_revisions_not_present(self, revision_ids):
1150
"""Check every revision id in revision_ids to see if we have it.
1152
Returns a set of the present revisions.
1155
graph = self.get_graph()
1156
parent_map = graph.get_parent_map(revision_ids)
1157
# The old API returned a list, should this actually be a set?
1158
return parent_map.keys()
1160
def _check_inventories(self, checker):
1161
"""Check the inventories found from the revision scan.
1163
This is responsible for verifying the sha1 of inventories and
1164
creating a pending_keys set that covers data referenced by inventories.
1166
bar = ui.ui_factory.nested_progress_bar()
1168
self._do_check_inventories(checker, bar)
1172
def _do_check_inventories(self, checker, bar):
1173
"""Helper for _check_inventories."""
1175
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1176
kinds = ['chk_bytes', 'texts']
1177
count = len(checker.pending_keys)
1178
bar.update("inventories", 0, 2)
1179
current_keys = checker.pending_keys
1180
checker.pending_keys = {}
1181
# Accumulate current checks.
1182
for key in current_keys:
1183
if key[0] != 'inventories' and key[0] not in kinds:
1184
checker._report_items.append('unknown key type %r' % (key,))
1185
keys[key[0]].add(key[1:])
1186
if keys['inventories']:
1187
# NB: output order *should* be roughly sorted - topo or
1188
# inverse topo depending on repository - either way decent
1189
# to just delta against. However, pre-CHK formats didn't
1190
# try to optimise inventory layout on disk. As such the
1191
# pre-CHK code path does not use inventory deltas.
1193
for record in self.inventories.check(keys=keys['inventories']):
1194
if record.storage_kind == 'absent':
1195
checker._report_items.append(
1196
'Missing inventory {%s}' % (record.key,))
1198
last_object = self._check_record('inventories', record,
1199
checker, last_object,
1200
current_keys[('inventories',) + record.key])
1201
del keys['inventories']
1204
bar.update("texts", 1)
1205
while (checker.pending_keys or keys['chk_bytes']
1207
# Something to check.
1208
current_keys = checker.pending_keys
1209
checker.pending_keys = {}
1210
# Accumulate current checks.
1211
for key in current_keys:
1212
if key[0] not in kinds:
1213
checker._report_items.append('unknown key type %r' % (key,))
1214
keys[key[0]].add(key[1:])
1215
# Check the outermost kind only - inventories || chk_bytes || texts
1219
for record in getattr(self, kind).check(keys=keys[kind]):
1220
if record.storage_kind == 'absent':
1221
checker._report_items.append(
1222
'Missing inventory {%s}' % (record.key,))
1224
last_object = self._check_record(kind, record,
1225
checker, last_object, current_keys[(kind,) + record.key])
1229
def _check_record(self, kind, record, checker, last_object, item_data):
1230
"""Check a single text from this repository."""
1231
if kind == 'inventories':
1232
rev_id = record.key[0]
1233
inv = self.deserialise_inventory(rev_id,
1234
record.get_bytes_as('fulltext'))
1235
if last_object is not None:
1236
delta = inv._make_delta(last_object)
1237
for old_path, path, file_id, ie in delta:
1240
ie.check(checker, rev_id, inv)
1242
for path, ie in inv.iter_entries():
1243
ie.check(checker, rev_id, inv)
1244
if self._format.fast_deltas:
1246
elif kind == 'chk_bytes':
1247
# No code written to check chk_bytes for this repo format.
1248
checker._report_items.append(
1249
'unsupported key type chk_bytes for %s' % (record.key,))
1250
elif kind == 'texts':
1251
self._check_text(record, checker, item_data)
1253
checker._report_items.append(
1254
'unknown key type %s for %s' % (kind, record.key))
1256
def _check_text(self, record, checker, item_data):
1257
"""Check a single text."""
1258
# Check it is extractable.
1259
# TODO: check length.
1260
if record.storage_kind == 'chunked':
1261
chunks = record.get_bytes_as(record.storage_kind)
1262
sha1 = osutils.sha_strings(chunks)
1263
length = sum(map(len, chunks))
1265
content = record.get_bytes_as('fulltext')
1266
sha1 = osutils.sha_string(content)
1267
length = len(content)
1268
if item_data and sha1 != item_data[1]:
1269
checker._report_items.append(
1270
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1271
(record.key, sha1, item_data[1], item_data[2]))
1274
def create(a_bzrdir):
1275
"""Construct the current default format repository in a_bzrdir."""
1276
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1278
def __init__(self, _format, a_bzrdir, control_files):
1279
"""instantiate a Repository.
1281
:param _format: The format of the repository on disk.
1282
:param a_bzrdir: The BzrDir of the repository.
1284
In the future we will have a single api for all stores for
1285
getting file texts, inventories and revisions, then
1286
this construct will accept instances of those things.
1288
super(Repository, self).__init__()
1289
self._format = _format
1290
# the following are part of the public API for Repository:
1291
self.bzrdir = a_bzrdir
1292
self.control_files = control_files
1293
self._transport = control_files._transport
1294
self.base = self._transport.base
1296
self._reconcile_does_inventory_gc = True
1297
self._reconcile_fixes_text_parents = False
1298
self._reconcile_backsup_inventory = True
1299
# not right yet - should be more semantically clear ?
1301
# TODO: make sure to construct the right store classes, etc, depending
1302
# on whether escaping is required.
1303
self._warn_if_deprecated()
1304
self._write_group = None
1305
# Additional places to query for data.
1306
self._fallback_repositories = []
1307
# An InventoryEntry cache, used during deserialization
1308
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1311
if self._fallback_repositories:
1312
return '%s(%r, fallback_repositories=%r)' % (
1313
self.__class__.__name__,
1315
self._fallback_repositories)
1317
return '%s(%r)' % (self.__class__.__name__,
1320
def _has_same_fallbacks(self, other_repo):
1321
"""Returns true if the repositories have the same fallbacks."""
1322
my_fb = self._fallback_repositories
1323
other_fb = other_repo._fallback_repositories
1324
if len(my_fb) != len(other_fb):
1326
for f, g in zip(my_fb, other_fb):
1327
if not f.has_same_location(g):
1331
def has_same_location(self, other):
1332
"""Returns a boolean indicating if this repository is at the same
1333
location as another repository.
1335
This might return False even when two repository objects are accessing
1336
the same physical repository via different URLs.
1338
if self.__class__ is not other.__class__:
1340
return (self._transport.base == other._transport.base)
1342
def is_in_write_group(self):
1343
"""Return True if there is an open write group.
1345
:seealso: start_write_group.
1347
return self._write_group is not None
1349
def is_locked(self):
1350
return self.control_files.is_locked()
1352
def is_write_locked(self):
1353
"""Return True if this object is write locked."""
1354
return self.is_locked() and self.control_files._lock_mode == 'w'
1356
def lock_write(self, token=None):
1357
"""Lock this repository for writing.
1359
This causes caching within the repository obejct to start accumlating
1360
data during reads, and allows a 'write_group' to be obtained. Write
1361
groups must be used for actual data insertion.
1363
:param token: if this is already locked, then lock_write will fail
1364
unless the token matches the existing lock.
1365
:returns: a token if this instance supports tokens, otherwise None.
1366
:raises TokenLockingNotSupported: when a token is given but this
1367
instance doesn't support using token locks.
1368
:raises MismatchedToken: if the specified token doesn't match the token
1369
of the existing lock.
1370
:seealso: start_write_group.
1372
A token should be passed in if you know that you have locked the object
1373
some other way, and need to synchronise this object's state with that
1376
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1378
locked = self.is_locked()
1379
result = self.control_files.lock_write(token=token)
1381
for repo in self._fallback_repositories:
1382
# Writes don't affect fallback repos
1384
self._refresh_data()
1387
def lock_read(self):
1388
locked = self.is_locked()
1389
self.control_files.lock_read()
1391
for repo in self._fallback_repositories:
1393
self._refresh_data()
1395
def get_physical_lock_status(self):
1396
return self.control_files.get_physical_lock_status()
1398
def leave_lock_in_place(self):
1399
"""Tell this repository not to release the physical lock when this
1402
If lock_write doesn't return a token, then this method is not supported.
1404
self.control_files.leave_in_place()
1406
def dont_leave_lock_in_place(self):
1407
"""Tell this repository to release the physical lock when this
1408
object is unlocked, even if it didn't originally acquire it.
1410
If lock_write doesn't return a token, then this method is not supported.
1412
self.control_files.dont_leave_in_place()
1415
def gather_stats(self, revid=None, committers=None):
1416
"""Gather statistics from a revision id.
1418
:param revid: The revision id to gather statistics from, if None, then
1419
no revision specific statistics are gathered.
1420
:param committers: Optional parameter controlling whether to grab
1421
a count of committers from the revision specific statistics.
1422
:return: A dictionary of statistics. Currently this contains:
1423
committers: The number of committers if requested.
1424
firstrev: A tuple with timestamp, timezone for the penultimate left
1425
most ancestor of revid, if revid is not the NULL_REVISION.
1426
latestrev: A tuple with timestamp, timezone for revid, if revid is
1427
not the NULL_REVISION.
1428
revisions: The total revision count in the repository.
1429
size: An estimate disk size of the repository in bytes.
1432
if revid and committers:
1433
result['committers'] = 0
1434
if revid and revid != _mod_revision.NULL_REVISION:
1436
all_committers = set()
1437
revisions = self.get_ancestry(revid)
1438
# pop the leading None
1440
first_revision = None
1442
# ignore the revisions in the middle - just grab first and last
1443
revisions = revisions[0], revisions[-1]
1444
for revision in self.get_revisions(revisions):
1445
if not first_revision:
1446
first_revision = revision
1448
all_committers.add(revision.committer)
1449
last_revision = revision
1451
result['committers'] = len(all_committers)
1452
result['firstrev'] = (first_revision.timestamp,
1453
first_revision.timezone)
1454
result['latestrev'] = (last_revision.timestamp,
1455
last_revision.timezone)
1457
# now gather global repository information
1458
# XXX: This is available for many repos regardless of listability.
1459
if self.bzrdir.root_transport.listable():
1460
# XXX: do we want to __define len__() ?
1461
# Maybe the versionedfiles object should provide a different
1462
# method to get the number of keys.
1463
result['revisions'] = len(self.revisions.keys())
1464
# result['size'] = t
1467
def find_branches(self, using=False):
1468
"""Find branches underneath this repository.
1470
This will include branches inside other branches.
1472
:param using: If True, list only branches using this repository.
1474
if using and not self.is_shared():
1476
return [self.bzrdir.open_branch()]
1477
except errors.NotBranchError:
1479
class Evaluator(object):
1482
self.first_call = True
1484
def __call__(self, bzrdir):
1485
# On the first call, the parameter is always the bzrdir
1486
# containing the current repo.
1487
if not self.first_call:
1489
repository = bzrdir.open_repository()
1490
except errors.NoRepositoryPresent:
1493
return False, (None, repository)
1494
self.first_call = False
1496
value = (bzrdir.open_branch(), None)
1497
except errors.NotBranchError:
1498
value = (None, None)
1502
for branch, repository in bzrdir.BzrDir.find_bzrdirs(
1503
self.bzrdir.root_transport, evaluate=Evaluator()):
1504
if branch is not None:
1505
branches.append(branch)
1506
if not using and repository is not None:
1507
branches.extend(repository.find_branches())
1511
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
1512
"""Return the revision ids that other has that this does not.
1514
These are returned in topological order.
1516
revision_id: only return revision ids included by revision_id.
1518
return InterRepository.get(other, self).search_missing_revision_ids(
1519
revision_id, find_ghosts)
1523
"""Open the repository rooted at base.
1525
For instance, if the repository is at URL/.bzr/repository,
1526
Repository.open(URL) -> a Repository instance.
1528
control = bzrdir.BzrDir.open(base)
1529
return control.open_repository()
1531
def copy_content_into(self, destination, revision_id=None):
1532
"""Make a complete copy of the content in self into destination.
1534
This is a destructive operation! Do not use it on existing
1537
return InterRepository.get(self, destination).copy_content(revision_id)
1539
def commit_write_group(self):
1540
"""Commit the contents accrued within the current write group.
1542
:seealso: start_write_group.
1544
:return: it may return an opaque hint that can be passed to 'pack'.
1546
if self._write_group is not self.get_transaction():
1547
# has an unlock or relock occured ?
1548
raise errors.BzrError('mismatched lock context %r and '
1550
(self.get_transaction(), self._write_group))
1551
result = self._commit_write_group()
1552
self._write_group = None
1555
def _commit_write_group(self):
1556
"""Template method for per-repository write group cleanup.
1558
This is called before the write group is considered to be
1559
finished and should ensure that all data handed to the repository
1560
for writing during the write group is safely committed (to the
1561
extent possible considering file system caching etc).
1564
def suspend_write_group(self):
1565
raise errors.UnsuspendableWriteGroup(self)
1567
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1568
"""Return the keys of missing inventory parents for revisions added in
1571
A revision is not complete if the inventory delta for that revision
1572
cannot be calculated. Therefore if the parent inventories of a
1573
revision are not present, the revision is incomplete, and e.g. cannot
1574
be streamed by a smart server. This method finds missing inventory
1575
parents for revisions added in this write group.
1577
if not self._format.supports_external_lookups:
1578
# This is only an issue for stacked repositories
1580
if not self.is_in_write_group():
1581
raise AssertionError('not in a write group')
1583
# XXX: We assume that every added revision already has its
1584
# corresponding inventory, so we only check for parent inventories that
1585
# might be missing, rather than all inventories.
1586
parents = set(self.revisions._index.get_missing_parents())
1587
parents.discard(_mod_revision.NULL_REVISION)
1588
unstacked_inventories = self.inventories._index
1589
present_inventories = unstacked_inventories.get_parent_map(
1590
key[-1:] for key in parents)
1591
parents.difference_update(present_inventories)
1592
if len(parents) == 0:
1593
# No missing parent inventories.
1595
if not check_for_missing_texts:
1596
return set(('inventories', rev_id) for (rev_id,) in parents)
1597
# Ok, now we have a list of missing inventories. But these only matter
1598
# if the inventories that reference them are missing some texts they
1599
# appear to introduce.
1600
# XXX: Texts referenced by all added inventories need to be present,
1601
# but at the moment we're only checking for texts referenced by
1602
# inventories at the graph's edge.
1603
key_deps = self.revisions._index._key_dependencies
1604
key_deps.add_keys(present_inventories)
1605
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1606
file_ids = self.fileids_altered_by_revision_ids(referrers)
1607
missing_texts = set()
1608
for file_id, version_ids in file_ids.iteritems():
1609
missing_texts.update(
1610
(file_id, version_id) for version_id in version_ids)
1611
present_texts = self.texts.get_parent_map(missing_texts)
1612
missing_texts.difference_update(present_texts)
1613
if not missing_texts:
1614
# No texts are missing, so all revisions and their deltas are
1617
# Alternatively the text versions could be returned as the missing
1618
# keys, but this is likely to be less data.
1619
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1622
def refresh_data(self):
1623
"""Re-read any data needed to to synchronise with disk.
1625
This method is intended to be called after another repository instance
1626
(such as one used by a smart server) has inserted data into the
1627
repository. It may not be called during a write group, but may be
1628
called at any other time.
1630
if self.is_in_write_group():
1631
raise errors.InternalBzrError(
1632
"May not refresh_data while in a write group.")
1633
self._refresh_data()
1635
def resume_write_group(self, tokens):
1636
if not self.is_write_locked():
1637
raise errors.NotWriteLocked(self)
1638
if self._write_group:
1639
raise errors.BzrError('already in a write group')
1640
self._resume_write_group(tokens)
1641
# so we can detect unlock/relock - the write group is now entered.
1642
self._write_group = self.get_transaction()
1644
def _resume_write_group(self, tokens):
1645
raise errors.UnsuspendableWriteGroup(self)
1647
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1649
"""Fetch the content required to construct revision_id from source.
1651
If revision_id is None and fetch_spec is None, then all content is
1654
fetch() may not be used when the repository is in a write group -
1655
either finish the current write group before using fetch, or use
1656
fetch before starting the write group.
1658
:param find_ghosts: Find and copy revisions in the source that are
1659
ghosts in the target (and not reachable directly by walking out to
1660
the first-present revision in target from revision_id).
1661
:param revision_id: If specified, all the content needed for this
1662
revision ID will be copied to the target. Fetch will determine for
1663
itself which content needs to be copied.
1664
:param fetch_spec: If specified, a SearchResult or
1665
PendingAncestryResult that describes which revisions to copy. This
1666
allows copying multiple heads at once. Mutually exclusive with
1669
if fetch_spec is not None and revision_id is not None:
1670
raise AssertionError(
1671
"fetch_spec and revision_id are mutually exclusive.")
1672
if self.is_in_write_group():
1673
raise errors.InternalBzrError(
1674
"May not fetch while in a write group.")
1675
# fast path same-url fetch operations
1676
# TODO: lift out to somewhere common with RemoteRepository
1677
# <https://bugs.edge.launchpad.net/bzr/+bug/401646>
1678
if (self.has_same_location(source)
1679
and fetch_spec is None
1680
and self._has_same_fallbacks(source)):
1681
# check that last_revision is in 'from' and then return a
1683
if (revision_id is not None and
1684
not _mod_revision.is_null(revision_id)):
1685
self.get_revision(revision_id)
1687
# if there is no specific appropriate InterRepository, this will get
1688
# the InterRepository base class, which raises an
1689
# IncompatibleRepositories when asked to fetch.
1690
inter = InterRepository.get(source, self)
1691
return inter.fetch(revision_id=revision_id, pb=pb,
1692
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1694
def create_bundle(self, target, base, fileobj, format=None):
1695
return serializer.write_bundle(self, target, base, fileobj, format)
1697
def get_commit_builder(self, branch, parents, config, timestamp=None,
1698
timezone=None, committer=None, revprops=None,
1700
"""Obtain a CommitBuilder for this repository.
1702
:param branch: Branch to commit to.
1703
:param parents: Revision ids of the parents of the new revision.
1704
:param config: Configuration to use.
1705
:param timestamp: Optional timestamp recorded for commit.
1706
:param timezone: Optional timezone for timestamp.
1707
:param committer: Optional committer to set for commit.
1708
:param revprops: Optional dictionary of revision properties.
1709
:param revision_id: Optional revision id.
1711
if self._fallback_repositories:
1712
raise errors.BzrError("Cannot commit from a lightweight checkout "
1713
"to a stacked branch. See "
1714
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1715
result = self._commit_builder_class(self, parents, config,
1716
timestamp, timezone, committer, revprops, revision_id)
1717
self.start_write_group()
1721
if (self.control_files._lock_count == 1 and
1722
self.control_files._lock_mode == 'w'):
1723
if self._write_group is not None:
1724
self.abort_write_group()
1725
self.control_files.unlock()
1726
raise errors.BzrError(
1727
'Must end write groups before releasing write locks.')
1728
self.control_files.unlock()
1729
if self.control_files._lock_count == 0:
1730
self._inventory_entry_cache.clear()
1731
for repo in self._fallback_repositories:
1735
def clone(self, a_bzrdir, revision_id=None):
1736
"""Clone this repository into a_bzrdir using the current format.
1738
Currently no check is made that the format of this repository and
1739
the bzrdir format are compatible. FIXME RBC 20060201.
1741
:return: The newly created destination repository.
1743
# TODO: deprecate after 0.16; cloning this with all its settings is
1744
# probably not very useful -- mbp 20070423
1745
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1746
self.copy_content_into(dest_repo, revision_id)
1749
def start_write_group(self):
1750
"""Start a write group in the repository.
1752
Write groups are used by repositories which do not have a 1:1 mapping
1753
between file ids and backend store to manage the insertion of data from
1754
both fetch and commit operations.
1756
A write lock is required around the start_write_group/commit_write_group
1757
for the support of lock-requiring repository formats.
1759
One can only insert data into a repository inside a write group.
1763
if not self.is_write_locked():
1764
raise errors.NotWriteLocked(self)
1765
if self._write_group:
1766
raise errors.BzrError('already in a write group')
1767
self._start_write_group()
1768
# so we can detect unlock/relock - the write group is now entered.
1769
self._write_group = self.get_transaction()
1771
def _start_write_group(self):
1772
"""Template method for per-repository write group startup.
1774
This is called before the write group is considered to be
1779
def sprout(self, to_bzrdir, revision_id=None):
1780
"""Create a descendent repository for new development.
1782
Unlike clone, this does not copy the settings of the repository.
1784
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1785
dest_repo.fetch(self, revision_id=revision_id)
1788
def _create_sprouting_repo(self, a_bzrdir, shared):
1789
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1790
# use target default format.
1791
dest_repo = a_bzrdir.create_repository()
1793
# Most control formats need the repository to be specifically
1794
# created, but on some old all-in-one formats it's not needed
1796
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1797
except errors.UninitializableFormat:
1798
dest_repo = a_bzrdir.open_repository()
1801
def _get_sink(self):
1802
"""Return a sink for streaming into this repository."""
1803
return StreamSink(self)
1805
def _get_source(self, to_format):
1806
"""Return a source for streaming from this repository."""
1807
return StreamSource(self, to_format)
1810
def has_revision(self, revision_id):
1811
"""True if this repository has a copy of the revision."""
1812
return revision_id in self.has_revisions((revision_id,))
1815
def has_revisions(self, revision_ids):
1816
"""Probe to find out the presence of multiple revisions.
1818
:param revision_ids: An iterable of revision_ids.
1819
:return: A set of the revision_ids that were present.
1821
parent_map = self.revisions.get_parent_map(
1822
[(rev_id,) for rev_id in revision_ids])
1824
if _mod_revision.NULL_REVISION in revision_ids:
1825
result.add(_mod_revision.NULL_REVISION)
1826
result.update([key[0] for key in parent_map])
1830
def get_revision(self, revision_id):
1831
"""Return the Revision object for a named revision."""
1832
return self.get_revisions([revision_id])[0]
1835
def get_revision_reconcile(self, revision_id):
1836
"""'reconcile' helper routine that allows access to a revision always.
1838
This variant of get_revision does not cross check the weave graph
1839
against the revision one as get_revision does: but it should only
1840
be used by reconcile, or reconcile-alike commands that are correcting
1841
or testing the revision graph.
1843
return self._get_revisions([revision_id])[0]
1846
def get_revisions(self, revision_ids):
1847
"""Get many revisions at once.
1849
Repositories that need to check data on every revision read should
1850
subclass this method.
1852
return self._get_revisions(revision_ids)
1855
def _get_revisions(self, revision_ids):
1856
"""Core work logic to get many revisions without sanity checks."""
1858
for revid, rev in self._iter_revisions(revision_ids):
1860
raise errors.NoSuchRevision(self, revid)
1862
return [revs[revid] for revid in revision_ids]
1864
def _iter_revisions(self, revision_ids):
1865
"""Iterate over revision objects.
1867
:param revision_ids: An iterable of revisions to examine. None may be
1868
passed to request all revisions known to the repository. Note that
1869
not all repositories can find unreferenced revisions; for those
1870
repositories only referenced ones will be returned.
1871
:return: An iterator of (revid, revision) tuples. Absent revisions (
1872
those asked for but not available) are returned as (revid, None).
1874
if revision_ids is None:
1875
revision_ids = self.all_revision_ids()
1877
for rev_id in revision_ids:
1878
if not rev_id or not isinstance(rev_id, basestring):
1879
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1880
keys = [(key,) for key in revision_ids]
1881
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1882
for record in stream:
1883
revid = record.key[0]
1884
if record.storage_kind == 'absent':
1887
text = record.get_bytes_as('fulltext')
1888
rev = self._serializer.read_revision_from_string(text)
1892
def get_revision_xml(self, revision_id):
1893
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1894
# would have already do it.
1895
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1896
# TODO: this can't just be replaced by:
1897
# return self._serializer.write_revision_to_string(
1898
# self.get_revision(revision_id))
1899
# as cStringIO preservers the encoding unlike write_revision_to_string
1900
# or some other call down the path.
1901
rev = self.get_revision(revision_id)
1902
rev_tmp = cStringIO.StringIO()
1903
# the current serializer..
1904
self._serializer.write_revision(rev, rev_tmp)
1906
return rev_tmp.getvalue()
1908
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1909
"""Produce a generator of revision deltas.
1911
Note that the input is a sequence of REVISIONS, not revision_ids.
1912
Trees will be held in memory until the generator exits.
1913
Each delta is relative to the revision's lefthand predecessor.
1915
:param specific_fileids: if not None, the result is filtered
1916
so that only those file-ids, their parents and their
1917
children are included.
1919
# Get the revision-ids of interest
1920
required_trees = set()
1921
for revision in revisions:
1922
required_trees.add(revision.revision_id)
1923
required_trees.update(revision.parent_ids[:1])
1925
# Get the matching filtered trees. Note that it's more
1926
# efficient to pass filtered trees to changes_from() rather
1927
# than doing the filtering afterwards. changes_from() could
1928
# arguably do the filtering itself but it's path-based, not
1929
# file-id based, so filtering before or afterwards is
1931
if specific_fileids is None:
1932
trees = dict((t.get_revision_id(), t) for
1933
t in self.revision_trees(required_trees))
1935
trees = dict((t.get_revision_id(), t) for
1936
t in self._filtered_revision_trees(required_trees,
1939
# Calculate the deltas
1940
for revision in revisions:
1941
if not revision.parent_ids:
1942
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1944
old_tree = trees[revision.parent_ids[0]]
1945
yield trees[revision.revision_id].changes_from(old_tree)
1948
def get_revision_delta(self, revision_id, specific_fileids=None):
1949
"""Return the delta for one revision.
1951
The delta is relative to the left-hand predecessor of the
1954
:param specific_fileids: if not None, the result is filtered
1955
so that only those file-ids, their parents and their
1956
children are included.
1958
r = self.get_revision(revision_id)
1959
return list(self.get_deltas_for_revisions([r],
1960
specific_fileids=specific_fileids))[0]
1963
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1964
signature = gpg_strategy.sign(plaintext)
1965
self.add_signature_text(revision_id, signature)
1968
def add_signature_text(self, revision_id, signature):
1969
self.signatures.add_lines((revision_id,), (),
1970
osutils.split_lines(signature))
1972
def find_text_key_references(self):
1973
"""Find the text key references within the repository.
1975
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1976
to whether they were referred to by the inventory of the
1977
revision_id that they contain. The inventory texts from all present
1978
revision ids are assessed to generate this report.
1980
revision_keys = self.revisions.keys()
1981
w = self.inventories
1982
pb = ui.ui_factory.nested_progress_bar()
1984
return self._find_text_key_references_from_xml_inventory_lines(
1985
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1989
def _find_text_key_references_from_xml_inventory_lines(self,
1991
"""Core routine for extracting references to texts from inventories.
1993
This performs the translation of xml lines to revision ids.
1995
:param line_iterator: An iterator of lines, origin_version_id
1996
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1997
to whether they were referred to by the inventory of the
1998
revision_id that they contain. Note that if that revision_id was
1999
not part of the line_iterator's output then False will be given -
2000
even though it may actually refer to that key.
2002
if not self._serializer.support_altered_by_hack:
2003
raise AssertionError(
2004
"_find_text_key_references_from_xml_inventory_lines only "
2005
"supported for branches which store inventory as unnested xml"
2006
", not on %r" % self)
2009
# this code needs to read every new line in every inventory for the
2010
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2011
# not present in one of those inventories is unnecessary but not
2012
# harmful because we are filtering by the revision id marker in the
2013
# inventory lines : we only select file ids altered in one of those
2014
# revisions. We don't need to see all lines in the inventory because
2015
# only those added in an inventory in rev X can contain a revision=X
2017
unescape_revid_cache = {}
2018
unescape_fileid_cache = {}
2020
# jam 20061218 In a big fetch, this handles hundreds of thousands
2021
# of lines, so it has had a lot of inlining and optimizing done.
2022
# Sorry that it is a little bit messy.
2023
# Move several functions to be local variables, since this is a long
2025
search = self._file_ids_altered_regex.search
2026
unescape = _unescape_xml
2027
setdefault = result.setdefault
2028
for line, line_key in line_iterator:
2029
match = search(line)
2032
# One call to match.group() returning multiple items is quite a
2033
# bit faster than 2 calls to match.group() each returning 1
2034
file_id, revision_id = match.group('file_id', 'revision_id')
2036
# Inlining the cache lookups helps a lot when you make 170,000
2037
# lines and 350k ids, versus 8.4 unique ids.
2038
# Using a cache helps in 2 ways:
2039
# 1) Avoids unnecessary decoding calls
2040
# 2) Re-uses cached strings, which helps in future set and
2042
# (2) is enough that removing encoding entirely along with
2043
# the cache (so we are using plain strings) results in no
2044
# performance improvement.
2046
revision_id = unescape_revid_cache[revision_id]
2048
unescaped = unescape(revision_id)
2049
unescape_revid_cache[revision_id] = unescaped
2050
revision_id = unescaped
2052
# Note that unconditionally unescaping means that we deserialise
2053
# every fileid, which for general 'pull' is not great, but we don't
2054
# really want to have some many fulltexts that this matters anyway.
2057
file_id = unescape_fileid_cache[file_id]
2059
unescaped = unescape(file_id)
2060
unescape_fileid_cache[file_id] = unescaped
2063
key = (file_id, revision_id)
2064
setdefault(key, False)
2065
if revision_id == line_key[-1]:
2069
def _inventory_xml_lines_for_keys(self, keys):
2070
"""Get a line iterator of the sort needed for findind references.
2072
Not relevant for non-xml inventory repositories.
2074
Ghosts in revision_keys are ignored.
2076
:param revision_keys: The revision keys for the inventories to inspect.
2077
:return: An iterator over (inventory line, revid) for the fulltexts of
2078
all of the xml inventories specified by revision_keys.
2080
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2081
for record in stream:
2082
if record.storage_kind != 'absent':
2083
chunks = record.get_bytes_as('chunked')
2084
revid = record.key[-1]
2085
lines = osutils.chunks_to_lines(chunks)
2089
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2091
"""Helper routine for fileids_altered_by_revision_ids.
2093
This performs the translation of xml lines to revision ids.
2095
:param line_iterator: An iterator of lines, origin_version_id
2096
:param revision_keys: The revision ids to filter for. This should be a
2097
set or other type which supports efficient __contains__ lookups, as
2098
the revision key from each parsed line will be looked up in the
2099
revision_keys filter.
2100
:return: a dictionary mapping altered file-ids to an iterable of
2101
revision_ids. Each altered file-ids has the exact revision_ids that
2102
altered it listed explicitly.
2104
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2105
line_iterator).iterkeys())
2106
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2107
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2108
self._inventory_xml_lines_for_keys(parent_keys)))
2109
new_keys = seen - parent_seen
2111
setdefault = result.setdefault
2112
for key in new_keys:
2113
setdefault(key[0], set()).add(key[-1])
2116
def _find_parent_ids_of_revisions(self, revision_ids):
2117
"""Find all parent ids that are mentioned in the revision graph.
2119
:return: set of revisions that are parents of revision_ids which are
2120
not part of revision_ids themselves
2122
parent_map = self.get_parent_map(revision_ids)
2124
map(parent_ids.update, parent_map.itervalues())
2125
parent_ids.difference_update(revision_ids)
2126
parent_ids.discard(_mod_revision.NULL_REVISION)
2129
def _find_parent_keys_of_revisions(self, revision_keys):
2130
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2132
:param revision_keys: An iterable of revision_keys.
2133
:return: The parents of all revision_keys that are not already in
2136
parent_map = self.revisions.get_parent_map(revision_keys)
2138
map(parent_keys.update, parent_map.itervalues())
2139
parent_keys.difference_update(revision_keys)
2140
parent_keys.discard(_mod_revision.NULL_REVISION)
2143
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2144
"""Find the file ids and versions affected by revisions.
2146
:param revisions: an iterable containing revision ids.
2147
:param _inv_weave: The inventory weave from this repository or None.
2148
If None, the inventory weave will be opened automatically.
2149
:return: a dictionary mapping altered file-ids to an iterable of
2150
revision_ids. Each altered file-ids has the exact revision_ids that
2151
altered it listed explicitly.
2153
selected_keys = set((revid,) for revid in revision_ids)
2154
w = _inv_weave or self.inventories
2155
pb = ui.ui_factory.nested_progress_bar()
2157
return self._find_file_ids_from_xml_inventory_lines(
2158
w.iter_lines_added_or_present_in_keys(
2159
selected_keys, pb=pb),
2164
def iter_files_bytes(self, desired_files):
2165
"""Iterate through file versions.
2167
Files will not necessarily be returned in the order they occur in
2168
desired_files. No specific order is guaranteed.
2170
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2171
value supplied by the caller as part of desired_files. It should
2172
uniquely identify the file version in the caller's context. (Examples:
2173
an index number or a TreeTransform trans_id.)
2175
bytes_iterator is an iterable of bytestrings for the file. The
2176
kind of iterable and length of the bytestrings are unspecified, but for
2177
this implementation, it is a list of bytes produced by
2178
VersionedFile.get_record_stream().
2180
:param desired_files: a list of (file_id, revision_id, identifier)
2184
for file_id, revision_id, callable_data in desired_files:
2185
text_keys[(file_id, revision_id)] = callable_data
2186
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2187
if record.storage_kind == 'absent':
2188
raise errors.RevisionNotPresent(record.key, self)
2189
yield text_keys[record.key], record.get_bytes_as('chunked')
2191
def _generate_text_key_index(self, text_key_references=None,
2193
"""Generate a new text key index for the repository.
2195
This is an expensive function that will take considerable time to run.
2197
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2198
list of parents, also text keys. When a given key has no parents,
2199
the parents list will be [NULL_REVISION].
2201
# All revisions, to find inventory parents.
2202
if ancestors is None:
2203
graph = self.get_graph()
2204
ancestors = graph.get_parent_map(self.all_revision_ids())
2205
if text_key_references is None:
2206
text_key_references = self.find_text_key_references()
2207
pb = ui.ui_factory.nested_progress_bar()
2209
return self._do_generate_text_key_index(ancestors,
2210
text_key_references, pb)
2214
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2215
"""Helper for _generate_text_key_index to avoid deep nesting."""
2216
revision_order = tsort.topo_sort(ancestors)
2217
invalid_keys = set()
2219
for revision_id in revision_order:
2220
revision_keys[revision_id] = set()
2221
text_count = len(text_key_references)
2222
# a cache of the text keys to allow reuse; costs a dict of all the
2223
# keys, but saves a 2-tuple for every child of a given key.
2225
for text_key, valid in text_key_references.iteritems():
2227
invalid_keys.add(text_key)
2229
revision_keys[text_key[1]].add(text_key)
2230
text_key_cache[text_key] = text_key
2231
del text_key_references
2233
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2234
NULL_REVISION = _mod_revision.NULL_REVISION
2235
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2236
# too small for large or very branchy trees. However, for 55K path
2237
# trees, it would be easy to use too much memory trivially. Ideally we
2238
# could gauge this by looking at available real memory etc, but this is
2239
# always a tricky proposition.
2240
inventory_cache = lru_cache.LRUCache(10)
2241
batch_size = 10 # should be ~150MB on a 55K path tree
2242
batch_count = len(revision_order) / batch_size + 1
2244
pb.update("Calculating text parents", processed_texts, text_count)
2245
for offset in xrange(batch_count):
2246
to_query = revision_order[offset * batch_size:(offset + 1) *
2250
for revision_id in to_query:
2251
parent_ids = ancestors[revision_id]
2252
for text_key in revision_keys[revision_id]:
2253
pb.update("Calculating text parents", processed_texts)
2254
processed_texts += 1
2255
candidate_parents = []
2256
for parent_id in parent_ids:
2257
parent_text_key = (text_key[0], parent_id)
2259
check_parent = parent_text_key not in \
2260
revision_keys[parent_id]
2262
# the parent parent_id is a ghost:
2263
check_parent = False
2264
# truncate the derived graph against this ghost.
2265
parent_text_key = None
2267
# look at the parent commit details inventories to
2268
# determine possible candidates in the per file graph.
2271
inv = inventory_cache[parent_id]
2273
inv = self.revision_tree(parent_id).inventory
2274
inventory_cache[parent_id] = inv
2276
parent_entry = inv[text_key[0]]
2277
except (KeyError, errors.NoSuchId):
2279
if parent_entry is not None:
2281
text_key[0], parent_entry.revision)
2283
parent_text_key = None
2284
if parent_text_key is not None:
2285
candidate_parents.append(
2286
text_key_cache[parent_text_key])
2287
parent_heads = text_graph.heads(candidate_parents)
2288
new_parents = list(parent_heads)
2289
new_parents.sort(key=lambda x:candidate_parents.index(x))
2290
if new_parents == []:
2291
new_parents = [NULL_REVISION]
2292
text_index[text_key] = new_parents
2294
for text_key in invalid_keys:
2295
text_index[text_key] = [NULL_REVISION]
2298
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2299
"""Get an iterable listing the keys of all the data introduced by a set
2302
The keys will be ordered so that the corresponding items can be safely
2303
fetched and inserted in that order.
2305
:returns: An iterable producing tuples of (knit-kind, file-id,
2306
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2307
'revisions'. file-id is None unless knit-kind is 'file'.
2309
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2312
for result in self._find_non_file_keys_to_fetch(revision_ids):
2315
def _find_file_keys_to_fetch(self, revision_ids, pb):
2316
# XXX: it's a bit weird to control the inventory weave caching in this
2317
# generator. Ideally the caching would be done in fetch.py I think. Or
2318
# maybe this generator should explicitly have the contract that it
2319
# should not be iterated until the previously yielded item has been
2321
inv_w = self.inventories
2323
# file ids that changed
2324
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2326
num_file_ids = len(file_ids)
2327
for file_id, altered_versions in file_ids.iteritems():
2329
pb.update("fetch texts", count, num_file_ids)
2331
yield ("file", file_id, altered_versions)
2333
def _find_non_file_keys_to_fetch(self, revision_ids):
2335
yield ("inventory", None, revision_ids)
2338
# XXX: Note ATM no callers actually pay attention to this return
2339
# instead they just use the list of revision ids and ignore
2340
# missing sigs. Consider removing this work entirely
2341
revisions_with_signatures = set(self.signatures.get_parent_map(
2342
[(r,) for r in revision_ids]))
2343
revisions_with_signatures = set(
2344
[r for (r,) in revisions_with_signatures])
2345
revisions_with_signatures.intersection_update(revision_ids)
2346
yield ("signatures", None, revisions_with_signatures)
2349
yield ("revisions", None, revision_ids)
2352
def get_inventory(self, revision_id):
2353
"""Get Inventory object by revision id."""
2354
return self.iter_inventories([revision_id]).next()
2356
def iter_inventories(self, revision_ids, ordering=None):
2357
"""Get many inventories by revision_ids.
2359
This will buffer some or all of the texts used in constructing the
2360
inventories in memory, but will only parse a single inventory at a
2363
:param revision_ids: The expected revision ids of the inventories.
2364
:param ordering: optional ordering, e.g. 'topological'. If not
2365
specified, the order of revision_ids will be preserved (by
2366
buffering if necessary).
2367
:return: An iterator of inventories.
2369
if ((None in revision_ids)
2370
or (_mod_revision.NULL_REVISION in revision_ids)):
2371
raise ValueError('cannot get null revision inventory')
2372
return self._iter_inventories(revision_ids, ordering)
2374
def _iter_inventories(self, revision_ids, ordering):
2375
"""single-document based inventory iteration."""
2376
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2377
for text, revision_id in inv_xmls:
2378
yield self.deserialise_inventory(revision_id, text)
2380
def _iter_inventory_xmls(self, revision_ids, ordering):
2381
if ordering is None:
2382
order_as_requested = True
2383
ordering = 'unordered'
2385
order_as_requested = False
2386
keys = [(revision_id,) for revision_id in revision_ids]
2389
if order_as_requested:
2390
key_iter = iter(keys)
2391
next_key = key_iter.next()
2392
stream = self.inventories.get_record_stream(keys, ordering, True)
2394
for record in stream:
2395
if record.storage_kind != 'absent':
2396
chunks = record.get_bytes_as('chunked')
2397
if order_as_requested:
2398
text_chunks[record.key] = chunks
2400
yield ''.join(chunks), record.key[-1]
2402
raise errors.NoSuchRevision(self, record.key)
2403
if order_as_requested:
2404
# Yield as many results as we can while preserving order.
2405
while next_key in text_chunks:
2406
chunks = text_chunks.pop(next_key)
2407
yield ''.join(chunks), next_key[-1]
2409
next_key = key_iter.next()
2410
except StopIteration:
2411
# We still want to fully consume the get_record_stream,
2412
# just in case it is not actually finished at this point
2416
def deserialise_inventory(self, revision_id, xml):
2417
"""Transform the xml into an inventory object.
2419
:param revision_id: The expected revision id of the inventory.
2420
:param xml: A serialised inventory.
2422
result = self._serializer.read_inventory_from_string(xml, revision_id,
2423
entry_cache=self._inventory_entry_cache)
2424
if result.revision_id != revision_id:
2425
raise AssertionError('revision id mismatch %s != %s' % (
2426
result.revision_id, revision_id))
2429
def serialise_inventory(self, inv):
2430
return self._serializer.write_inventory_to_string(inv)
2432
def _serialise_inventory_to_lines(self, inv):
2433
return self._serializer.write_inventory_to_lines(inv)
2435
def get_serializer_format(self):
2436
return self._serializer.format_num
2439
def get_inventory_xml(self, revision_id):
2440
"""Get inventory XML as a file object."""
2441
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2443
text, revision_id = texts.next()
2444
except StopIteration:
2445
raise errors.HistoryMissing(self, 'inventory', revision_id)
2449
def get_inventory_sha1(self, revision_id):
2450
"""Return the sha1 hash of the inventory entry
2452
return self.get_revision(revision_id).inventory_sha1
2454
def get_rev_id_for_revno(self, revno, known_pair):
2455
"""Return the revision id of a revno, given a later (revno, revid)
2456
pair in the same history.
2458
:return: if found (True, revid). If the available history ran out
2459
before reaching the revno, then this returns
2460
(False, (closest_revno, closest_revid)).
2462
known_revno, known_revid = known_pair
2463
partial_history = [known_revid]
2464
distance_from_known = known_revno - revno
2465
if distance_from_known < 0:
2467
'requested revno (%d) is later than given known revno (%d)'
2468
% (revno, known_revno))
2471
self, partial_history, stop_index=distance_from_known)
2472
except errors.RevisionNotPresent, err:
2473
if err.revision_id == known_revid:
2474
# The start revision (known_revid) wasn't found.
2476
# This is a stacked repository with no fallbacks, or a there's a
2477
# left-hand ghost. Either way, even though the revision named in
2478
# the error isn't in this repo, we know it's the next step in this
2479
# left-hand history.
2480
partial_history.append(err.revision_id)
2481
if len(partial_history) <= distance_from_known:
2482
# Didn't find enough history to get a revid for the revno.
2483
earliest_revno = known_revno - len(partial_history) + 1
2484
return (False, (earliest_revno, partial_history[-1]))
2485
if len(partial_history) - 1 > distance_from_known:
2486
raise AssertionError('_iter_for_revno returned too much history')
2487
return (True, partial_history[-1])
2489
def iter_reverse_revision_history(self, revision_id):
2490
"""Iterate backwards through revision ids in the lefthand history
2492
:param revision_id: The revision id to start with. All its lefthand
2493
ancestors will be traversed.
2495
graph = self.get_graph()
2496
next_id = revision_id
2498
if next_id in (None, _mod_revision.NULL_REVISION):
2501
parents = graph.get_parent_map([next_id])[next_id]
2503
raise errors.RevisionNotPresent(next_id, self)
2505
if len(parents) == 0:
2508
next_id = parents[0]
2511
def get_revision_inventory(self, revision_id):
2512
"""Return inventory of a past revision."""
2513
# TODO: Unify this with get_inventory()
2514
# bzr 0.0.6 and later imposes the constraint that the inventory_id
2515
# must be the same as its revision, so this is trivial.
2516
if revision_id is None:
2517
# This does not make sense: if there is no revision,
2518
# then it is the current tree inventory surely ?!
2519
# and thus get_root_id() is something that looks at the last
2520
# commit on the branch, and the get_root_id is an inventory check.
2521
raise NotImplementedError
2522
# return Inventory(self.get_root_id())
2524
return self.get_inventory(revision_id)
2526
def is_shared(self):
2527
"""Return True if this repository is flagged as a shared repository."""
2528
raise NotImplementedError(self.is_shared)
2531
def reconcile(self, other=None, thorough=False):
2532
"""Reconcile this repository."""
2533
from bzrlib.reconcile import RepoReconciler
2534
reconciler = RepoReconciler(self, thorough=thorough)
2535
reconciler.reconcile()
2538
def _refresh_data(self):
2539
"""Helper called from lock_* to ensure coherency with disk.
2541
The default implementation does nothing; it is however possible
2542
for repositories to maintain loaded indices across multiple locks
2543
by checking inside their implementation of this method to see
2544
whether their indices are still valid. This depends of course on
2545
the disk format being validatable in this manner. This method is
2546
also called by the refresh_data() public interface to cause a refresh
2547
to occur while in a write lock so that data inserted by a smart server
2548
push operation is visible on the client's instance of the physical
2553
def revision_tree(self, revision_id):
2554
"""Return Tree for a revision on this branch.
2556
`revision_id` may be NULL_REVISION for the empty tree revision.
2558
revision_id = _mod_revision.ensure_null(revision_id)
2559
# TODO: refactor this to use an existing revision object
2560
# so we don't need to read it in twice.
2561
if revision_id == _mod_revision.NULL_REVISION:
2562
return RevisionTree(self, Inventory(root_id=None),
2563
_mod_revision.NULL_REVISION)
2565
inv = self.get_revision_inventory(revision_id)
2566
return RevisionTree(self, inv, revision_id)
2568
def revision_trees(self, revision_ids):
2569
"""Return Trees for revisions in this repository.
2571
:param revision_ids: a sequence of revision-ids;
2572
a revision-id may not be None or 'null:'
2574
inventories = self.iter_inventories(revision_ids)
2575
for inv in inventories:
2576
yield RevisionTree(self, inv, inv.revision_id)
2578
def _filtered_revision_trees(self, revision_ids, file_ids):
2579
"""Return Tree for a revision on this branch with only some files.
2581
:param revision_ids: a sequence of revision-ids;
2582
a revision-id may not be None or 'null:'
2583
:param file_ids: if not None, the result is filtered
2584
so that only those file-ids, their parents and their
2585
children are included.
2587
inventories = self.iter_inventories(revision_ids)
2588
for inv in inventories:
2589
# Should we introduce a FilteredRevisionTree class rather
2590
# than pre-filter the inventory here?
2591
filtered_inv = inv.filter(file_ids)
2592
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2595
def get_ancestry(self, revision_id, topo_sorted=True):
2596
"""Return a list of revision-ids integrated by a revision.
2598
The first element of the list is always None, indicating the origin
2599
revision. This might change when we have history horizons, or
2600
perhaps we should have a new API.
2602
This is topologically sorted.
2604
if _mod_revision.is_null(revision_id):
2606
if not self.has_revision(revision_id):
2607
raise errors.NoSuchRevision(self, revision_id)
2608
graph = self.get_graph()
2610
search = graph._make_breadth_first_searcher([revision_id])
2613
found, ghosts = search.next_with_ghosts()
2614
except StopIteration:
2617
if _mod_revision.NULL_REVISION in keys:
2618
keys.remove(_mod_revision.NULL_REVISION)
2620
parent_map = graph.get_parent_map(keys)
2621
keys = tsort.topo_sort(parent_map)
2622
return [None] + list(keys)
2624
def pack(self, hint=None):
2625
"""Compress the data within the repository.
2627
This operation only makes sense for some repository types. For other
2628
types it should be a no-op that just returns.
2630
This stub method does not require a lock, but subclasses should use
2631
@needs_write_lock as this is a long running call its reasonable to
2632
implicitly lock for the user.
2634
:param hint: If not supplied, the whole repository is packed.
2635
If supplied, the repository may use the hint parameter as a
2636
hint for the parts of the repository to pack. A hint can be
2637
obtained from the result of commit_write_group(). Out of
2638
date hints are simply ignored, because concurrent operations
2639
can obsolete them rapidly.
2642
def get_transaction(self):
2643
return self.control_files.get_transaction()
2645
def get_parent_map(self, revision_ids):
2646
"""See graph.StackedParentsProvider.get_parent_map"""
2647
# revisions index works in keys; this just works in revisions
2648
# therefore wrap and unwrap
2651
for revision_id in revision_ids:
2652
if revision_id == _mod_revision.NULL_REVISION:
2653
result[revision_id] = ()
2654
elif revision_id is None:
2655
raise ValueError('get_parent_map(None) is not valid')
2657
query_keys.append((revision_id ,))
2658
for ((revision_id,), parent_keys) in \
2659
self.revisions.get_parent_map(query_keys).iteritems():
2661
result[revision_id] = tuple(parent_revid
2662
for (parent_revid,) in parent_keys)
2664
result[revision_id] = (_mod_revision.NULL_REVISION,)
2667
def _make_parents_provider(self):
2670
def get_graph(self, other_repository=None):
2671
"""Return the graph walker for this repository format"""
2672
parents_provider = self._make_parents_provider()
2673
if (other_repository is not None and
2674
not self.has_same_location(other_repository)):
2675
parents_provider = graph.StackedParentsProvider(
2676
[parents_provider, other_repository._make_parents_provider()])
2677
return graph.Graph(parents_provider)
2679
def _get_versioned_file_checker(self, text_key_references=None,
2681
"""Return an object suitable for checking versioned files.
2683
:param text_key_references: if non-None, an already built
2684
dictionary mapping text keys ((fileid, revision_id) tuples)
2685
to whether they were referred to by the inventory of the
2686
revision_id that they contain. If None, this will be
2688
:param ancestors: Optional result from
2689
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2692
return _VersionedFileChecker(self,
2693
text_key_references=text_key_references, ancestors=ancestors)
2695
def revision_ids_to_search_result(self, result_set):
2696
"""Convert a set of revision ids to a graph SearchResult."""
2697
result_parents = set()
2698
for parents in self.get_graph().get_parent_map(
2699
result_set).itervalues():
2700
result_parents.update(parents)
2701
included_keys = result_set.intersection(result_parents)
2702
start_keys = result_set.difference(included_keys)
2703
exclude_keys = result_parents.difference(result_set)
2704
result = graph.SearchResult(start_keys, exclude_keys,
2705
len(result_set), result_set)
2709
def set_make_working_trees(self, new_value):
2710
"""Set the policy flag for making working trees when creating branches.
2712
This only applies to branches that use this repository.
2714
The default is 'True'.
2715
:param new_value: True to restore the default, False to disable making
2718
raise NotImplementedError(self.set_make_working_trees)
2720
def make_working_trees(self):
2721
"""Returns the policy for making working trees on new branches."""
2722
raise NotImplementedError(self.make_working_trees)
2725
def sign_revision(self, revision_id, gpg_strategy):
2726
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2727
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2730
def has_signature_for_revision_id(self, revision_id):
2731
"""Query for a revision signature for revision_id in the repository."""
2732
if not self.has_revision(revision_id):
2733
raise errors.NoSuchRevision(self, revision_id)
2734
sig_present = (1 == len(
2735
self.signatures.get_parent_map([(revision_id,)])))
2739
def get_signature_text(self, revision_id):
2740
"""Return the text for a signature."""
2741
stream = self.signatures.get_record_stream([(revision_id,)],
2743
record = stream.next()
2744
if record.storage_kind == 'absent':
2745
raise errors.NoSuchRevision(self, revision_id)
2746
return record.get_bytes_as('fulltext')
2749
def check(self, revision_ids=None, callback_refs=None, check_repo=True):
2750
"""Check consistency of all history of given revision_ids.
2752
Different repository implementations should override _check().
2754
:param revision_ids: A non-empty list of revision_ids whose ancestry
2755
will be checked. Typically the last revision_id of a branch.
2756
:param callback_refs: A dict of check-refs to resolve and callback
2757
the check/_check method on the items listed as wanting the ref.
2759
:param check_repo: If False do not check the repository contents, just
2760
calculate the data callback_refs requires and call them back.
2762
return self._check(revision_ids, callback_refs=callback_refs,
2763
check_repo=check_repo)
2765
def _check(self, revision_ids, callback_refs, check_repo):
2766
result = check.Check(self, check_repo=check_repo)
2767
result.check(callback_refs)
2770
def _warn_if_deprecated(self):
2771
global _deprecation_warning_done
2772
if _deprecation_warning_done:
2774
_deprecation_warning_done = True
2775
warning("Format %s for %s is deprecated - please use 'bzr upgrade' to get better performance"
2776
% (self._format, self.bzrdir.transport.base))
2778
def supports_rich_root(self):
2779
return self._format.rich_root_data
2781
def _check_ascii_revisionid(self, revision_id, method):
2782
"""Private helper for ascii-only repositories."""
2783
# weave repositories refuse to store revisionids that are non-ascii.
2784
if revision_id is not None:
2785
# weaves require ascii revision ids.
2786
if isinstance(revision_id, unicode):
2788
revision_id.encode('ascii')
2789
except UnicodeEncodeError:
2790
raise errors.NonAsciiRevisionId(method, self)
2793
revision_id.decode('ascii')
2794
except UnicodeDecodeError:
2795
raise errors.NonAsciiRevisionId(method, self)
2797
def revision_graph_can_have_wrong_parents(self):
2798
"""Is it possible for this repository to have a revision graph with
2801
If True, then this repository must also implement
2802
_find_inconsistent_revision_parents so that check and reconcile can
2803
check for inconsistencies before proceeding with other checks that may
2804
depend on the revision index being consistent.
2806
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2809
# remove these delegates a while after bzr 0.15
2810
def __make_delegated(name, from_module):
2811
def _deprecated_repository_forwarder():
2812
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2813
% (name, from_module),
2816
m = __import__(from_module, globals(), locals(), [name])
2818
return getattr(m, name)
2819
except AttributeError:
2820
raise AttributeError('module %s has no name %s'
2822
globals()[name] = _deprecated_repository_forwarder
2825
'AllInOneRepository',
2826
'WeaveMetaDirRepository',
2827
'PreSplitOutRepositoryFormat',
2828
'RepositoryFormat4',
2829
'RepositoryFormat5',
2830
'RepositoryFormat6',
2831
'RepositoryFormat7',
2833
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2837
'RepositoryFormatKnit',
2838
'RepositoryFormatKnit1',
2840
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2843
def install_revision(repository, rev, revision_tree):
2844
"""Install all revision data into a repository."""
2845
install_revisions(repository, [(rev, revision_tree, None)])
2848
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2849
"""Install all revision data into a repository.
2851
Accepts an iterable of revision, tree, signature tuples. The signature
2854
repository.start_write_group()
2856
inventory_cache = lru_cache.LRUCache(10)
2857
for n, (revision, revision_tree, signature) in enumerate(iterable):
2858
_install_revision(repository, revision, revision_tree, signature,
2861
pb.update('Transferring revisions', n + 1, num_revisions)
2863
repository.abort_write_group()
2866
repository.commit_write_group()
2869
def _install_revision(repository, rev, revision_tree, signature,
2871
"""Install all revision data into a repository."""
2872
present_parents = []
2874
for p_id in rev.parent_ids:
2875
if repository.has_revision(p_id):
2876
present_parents.append(p_id)
2877
parent_trees[p_id] = repository.revision_tree(p_id)
2879
parent_trees[p_id] = repository.revision_tree(
2880
_mod_revision.NULL_REVISION)
2882
inv = revision_tree.inventory
2883
entries = inv.iter_entries()
2884
# backwards compatibility hack: skip the root id.
2885
if not repository.supports_rich_root():
2886
path, root = entries.next()
2887
if root.revision != rev.revision_id:
2888
raise errors.IncompatibleRevision(repr(repository))
2890
for path, ie in entries:
2891
text_keys[(ie.file_id, ie.revision)] = ie
2892
text_parent_map = repository.texts.get_parent_map(text_keys)
2893
missing_texts = set(text_keys) - set(text_parent_map)
2894
# Add the texts that are not already present
2895
for text_key in missing_texts:
2896
ie = text_keys[text_key]
2898
# FIXME: TODO: The following loop overlaps/duplicates that done by
2899
# commit to determine parents. There is a latent/real bug here where
2900
# the parents inserted are not those commit would do - in particular
2901
# they are not filtered by heads(). RBC, AB
2902
for revision, tree in parent_trees.iteritems():
2903
if ie.file_id not in tree:
2905
parent_id = tree.inventory[ie.file_id].revision
2906
if parent_id in text_parents:
2908
text_parents.append((ie.file_id, parent_id))
2909
lines = revision_tree.get_file(ie.file_id).readlines()
2910
repository.texts.add_lines(text_key, text_parents, lines)
2912
# install the inventory
2913
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2914
# Cache this inventory
2915
inventory_cache[rev.revision_id] = inv
2917
basis_inv = inventory_cache[rev.parent_ids[0]]
2919
repository.add_inventory(rev.revision_id, inv, present_parents)
2921
delta = inv._make_delta(basis_inv)
2922
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2923
rev.revision_id, present_parents)
2925
repository.add_inventory(rev.revision_id, inv, present_parents)
2926
except errors.RevisionAlreadyPresent:
2928
if signature is not None:
2929
repository.add_signature_text(rev.revision_id, signature)
2930
repository.add_revision(rev.revision_id, rev, inv)
2933
class MetaDirRepository(Repository):
2934
"""Repositories in the new meta-dir layout.
2936
:ivar _transport: Transport for access to repository control files,
2937
typically pointing to .bzr/repository.
2940
def __init__(self, _format, a_bzrdir, control_files):
2941
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2942
self._transport = control_files._transport
2944
def is_shared(self):
2945
"""Return True if this repository is flagged as a shared repository."""
2946
return self._transport.has('shared-storage')
2949
def set_make_working_trees(self, new_value):
2950
"""Set the policy flag for making working trees when creating branches.
2952
This only applies to branches that use this repository.
2954
The default is 'True'.
2955
:param new_value: True to restore the default, False to disable making
2960
self._transport.delete('no-working-trees')
2961
except errors.NoSuchFile:
2964
self._transport.put_bytes('no-working-trees', '',
2965
mode=self.bzrdir._get_file_mode())
2967
def make_working_trees(self):
2968
"""Returns the policy for making working trees on new branches."""
2969
return not self._transport.has('no-working-trees')
2972
class MetaDirVersionedFileRepository(MetaDirRepository):
2973
"""Repositories in a meta-dir, that work via versioned file objects."""
2975
def __init__(self, _format, a_bzrdir, control_files):
2976
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2980
network_format_registry = registry.FormatRegistry()
2981
"""Registry of formats indexed by their network name.
2983
The network name for a repository format is an identifier that can be used when
2984
referring to formats with smart server operations. See
2985
RepositoryFormat.network_name() for more detail.
2989
format_registry = registry.FormatRegistry(network_format_registry)
2990
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
2992
This can contain either format instances themselves, or classes/factories that
2993
can be called to obtain one.
2997
#####################################################################
2998
# Repository Formats
3000
class RepositoryFormat(object):
3001
"""A repository format.
3003
Formats provide four things:
3004
* An initialization routine to construct repository data on disk.
3005
* a optional format string which is used when the BzrDir supports
3007
* an open routine which returns a Repository instance.
3008
* A network name for referring to the format in smart server RPC
3011
There is one and only one Format subclass for each on-disk format. But
3012
there can be one Repository subclass that is used for several different
3013
formats. The _format attribute on a Repository instance can be used to
3014
determine the disk format.
3016
Formats are placed in a registry by their format string for reference
3017
during opening. These should be subclasses of RepositoryFormat for
3020
Once a format is deprecated, just deprecate the initialize and open
3021
methods on the format class. Do not deprecate the object, as the
3022
object may be created even when a repository instance hasn't been
3025
Common instance attributes:
3026
_matchingbzrdir - the bzrdir format that the repository format was
3027
originally written to work with. This can be used if manually
3028
constructing a bzrdir and repository, or more commonly for test suite
3032
# Set to True or False in derived classes. True indicates that the format
3033
# supports ghosts gracefully.
3034
supports_ghosts = None
3035
# Can this repository be given external locations to lookup additional
3036
# data. Set to True or False in derived classes.
3037
supports_external_lookups = None
3038
# Does this format support CHK bytestring lookups. Set to True or False in
3040
supports_chks = None
3041
# Should commit add an inventory, or an inventory delta to the repository.
3042
_commit_inv_deltas = True
3043
# What order should fetch operations request streams in?
3044
# The default is unordered as that is the cheapest for an origin to
3046
_fetch_order = 'unordered'
3047
# Does this repository format use deltas that can be fetched as-deltas ?
3048
# (E.g. knits, where the knit deltas can be transplanted intact.
3049
# We default to False, which will ensure that enough data to get
3050
# a full text out of any fetch stream will be grabbed.
3051
_fetch_uses_deltas = False
3052
# Should fetch trigger a reconcile after the fetch? Only needed for
3053
# some repository formats that can suffer internal inconsistencies.
3054
_fetch_reconcile = False
3055
# Does this format have < O(tree_size) delta generation. Used to hint what
3056
# code path for commit, amongst other things.
3058
# Does doing a pack operation compress data? Useful for the pack UI command
3059
# (so if there is one pack, the operation can still proceed because it may
3060
# help), and for fetching when data won't have come from the same
3062
pack_compresses = False
3063
# Does the repository inventory storage understand references to trees?
3064
supports_tree_reference = None
3067
return "<%s>" % self.__class__.__name__
3069
def __eq__(self, other):
3070
# format objects are generally stateless
3071
return isinstance(other, self.__class__)
3073
def __ne__(self, other):
3074
return not self == other
3077
def find_format(klass, a_bzrdir):
3078
"""Return the format for the repository object in a_bzrdir.
3080
This is used by bzr native formats that have a "format" file in
3081
the repository. Other methods may be used by different types of
3085
transport = a_bzrdir.get_repository_transport(None)
3086
format_string = transport.get("format").read()
3087
return format_registry.get(format_string)
3088
except errors.NoSuchFile:
3089
raise errors.NoRepositoryPresent(a_bzrdir)
3091
raise errors.UnknownFormatError(format=format_string,
3095
def register_format(klass, format):
3096
format_registry.register(format.get_format_string(), format)
3099
def unregister_format(klass, format):
3100
format_registry.remove(format.get_format_string())
3103
def get_default_format(klass):
3104
"""Return the current default format."""
3105
from bzrlib import bzrdir
3106
return bzrdir.format_registry.make_bzrdir('default').repository_format
3108
def get_format_string(self):
3109
"""Return the ASCII format string that identifies this format.
3111
Note that in pre format ?? repositories the format string is
3112
not permitted nor written to disk.
3114
raise NotImplementedError(self.get_format_string)
3116
def get_format_description(self):
3117
"""Return the short description for this format."""
3118
raise NotImplementedError(self.get_format_description)
3120
# TODO: this shouldn't be in the base class, it's specific to things that
3121
# use weaves or knits -- mbp 20070207
3122
def _get_versioned_file_store(self,
3127
versionedfile_class=None,
3128
versionedfile_kwargs={},
3130
if versionedfile_class is None:
3131
versionedfile_class = self._versionedfile_class
3132
weave_transport = control_files._transport.clone(name)
3133
dir_mode = control_files._dir_mode
3134
file_mode = control_files._file_mode
3135
return VersionedFileStore(weave_transport, prefixed=prefixed,
3137
file_mode=file_mode,
3138
versionedfile_class=versionedfile_class,
3139
versionedfile_kwargs=versionedfile_kwargs,
3142
def initialize(self, a_bzrdir, shared=False):
3143
"""Initialize a repository of this format in a_bzrdir.
3145
:param a_bzrdir: The bzrdir to put the new repository in it.
3146
:param shared: The repository should be initialized as a sharable one.
3147
:returns: The new repository object.
3149
This may raise UninitializableFormat if shared repository are not
3150
compatible the a_bzrdir.
3152
raise NotImplementedError(self.initialize)
3154
def is_supported(self):
3155
"""Is this format supported?
3157
Supported formats must be initializable and openable.
3158
Unsupported formats may not support initialization or committing or
3159
some other features depending on the reason for not being supported.
3163
def network_name(self):
3164
"""A simple byte string uniquely identifying this format for RPC calls.
3166
MetaDir repository formats use their disk format string to identify the
3167
repository over the wire. All in one formats such as bzr < 0.8, and
3168
foreign formats like svn/git and hg should use some marker which is
3169
unique and immutable.
3171
raise NotImplementedError(self.network_name)
3173
def check_conversion_target(self, target_format):
3174
if self.rich_root_data and not target_format.rich_root_data:
3175
raise errors.BadConversionTarget(
3176
'Does not support rich root data.', target_format,
3178
if (self.supports_tree_reference and
3179
not getattr(target_format, 'supports_tree_reference', False)):
3180
raise errors.BadConversionTarget(
3181
'Does not support nested trees', target_format,
3184
def open(self, a_bzrdir, _found=False):
3185
"""Return an instance of this format for the bzrdir a_bzrdir.
3187
_found is a private parameter, do not use it.
3189
raise NotImplementedError(self.open)
3192
class MetaDirRepositoryFormat(RepositoryFormat):
3193
"""Common base class for the new repositories using the metadir layout."""
3195
rich_root_data = False
3196
supports_tree_reference = False
3197
supports_external_lookups = False
3200
def _matchingbzrdir(self):
3201
matching = bzrdir.BzrDirMetaFormat1()
3202
matching.repository_format = self
3206
super(MetaDirRepositoryFormat, self).__init__()
3208
def _create_control_files(self, a_bzrdir):
3209
"""Create the required files and the initial control_files object."""
3210
# FIXME: RBC 20060125 don't peek under the covers
3211
# NB: no need to escape relative paths that are url safe.
3212
repository_transport = a_bzrdir.get_repository_transport(self)
3213
control_files = lockable_files.LockableFiles(repository_transport,
3214
'lock', lockdir.LockDir)
3215
control_files.create_lock()
3216
return control_files
3218
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
3219
"""Upload the initial blank content."""
3220
control_files = self._create_control_files(a_bzrdir)
3221
control_files.lock_write()
3222
transport = control_files._transport
3224
utf8_files += [('shared-storage', '')]
3226
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
3227
for (filename, content_stream) in files:
3228
transport.put_file(filename, content_stream,
3229
mode=a_bzrdir._get_file_mode())
3230
for (filename, content_bytes) in utf8_files:
3231
transport.put_bytes_non_atomic(filename, content_bytes,
3232
mode=a_bzrdir._get_file_mode())
3234
control_files.unlock()
3236
def network_name(self):
3237
"""Metadir formats have matching disk and network format strings."""
3238
return self.get_format_string()
3241
# Pre-0.8 formats that don't have a disk format string (because they are
3242
# versioned by the matching control directory). We use the control directories
3243
# disk format string as a key for the network_name because they meet the
3244
# constraints (simple string, unique, immutable).
3245
network_format_registry.register_lazy(
3246
"Bazaar-NG branch, format 5\n",
3247
'bzrlib.repofmt.weaverepo',
3248
'RepositoryFormat5',
3250
network_format_registry.register_lazy(
3251
"Bazaar-NG branch, format 6\n",
3252
'bzrlib.repofmt.weaverepo',
3253
'RepositoryFormat6',
3256
# formats which have no format string are not discoverable or independently
3257
# creatable on disk, so are not registered in format_registry. They're
3258
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
3259
# needed, it's constructed directly by the BzrDir. Non-native formats where
3260
# the repository is not separately opened are similar.
3262
format_registry.register_lazy(
3263
'Bazaar-NG Repository format 7',
3264
'bzrlib.repofmt.weaverepo',
3268
format_registry.register_lazy(
3269
'Bazaar-NG Knit Repository Format 1',
3270
'bzrlib.repofmt.knitrepo',
3271
'RepositoryFormatKnit1',
3274
format_registry.register_lazy(
3275
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
3276
'bzrlib.repofmt.knitrepo',
3277
'RepositoryFormatKnit3',
3280
format_registry.register_lazy(
3281
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
3282
'bzrlib.repofmt.knitrepo',
3283
'RepositoryFormatKnit4',
3286
# Pack-based formats. There is one format for pre-subtrees, and one for
3287
# post-subtrees to allow ease of testing.
3288
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
3289
format_registry.register_lazy(
3290
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
3291
'bzrlib.repofmt.pack_repo',
3292
'RepositoryFormatKnitPack1',
3294
format_registry.register_lazy(
3295
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
3296
'bzrlib.repofmt.pack_repo',
3297
'RepositoryFormatKnitPack3',
3299
format_registry.register_lazy(
3300
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
3301
'bzrlib.repofmt.pack_repo',
3302
'RepositoryFormatKnitPack4',
3304
format_registry.register_lazy(
3305
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
3306
'bzrlib.repofmt.pack_repo',
3307
'RepositoryFormatKnitPack5',
3309
format_registry.register_lazy(
3310
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
3311
'bzrlib.repofmt.pack_repo',
3312
'RepositoryFormatKnitPack5RichRoot',
3314
format_registry.register_lazy(
3315
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3316
'bzrlib.repofmt.pack_repo',
3317
'RepositoryFormatKnitPack5RichRootBroken',
3319
format_registry.register_lazy(
3320
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3321
'bzrlib.repofmt.pack_repo',
3322
'RepositoryFormatKnitPack6',
3324
format_registry.register_lazy(
3325
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3326
'bzrlib.repofmt.pack_repo',
3327
'RepositoryFormatKnitPack6RichRoot',
3330
# Development formats.
3331
# Obsolete but kept pending a CHK based subtree format.
3332
format_registry.register_lazy(
3333
("Bazaar development format 2 with subtree support "
3334
"(needs bzr.dev from before 1.8)\n"),
3335
'bzrlib.repofmt.pack_repo',
3336
'RepositoryFormatPackDevelopment2Subtree',
3339
# 1.14->1.16 go below here
3340
format_registry.register_lazy(
3341
'Bazaar development format - group compression and chk inventory'
3342
' (needs bzr.dev from 1.14)\n',
3343
'bzrlib.repofmt.groupcompress_repo',
3344
'RepositoryFormatCHK1',
3347
format_registry.register_lazy(
3348
'Bazaar development format - chk repository with bencode revision '
3349
'serialization (needs bzr.dev from 1.16)\n',
3350
'bzrlib.repofmt.groupcompress_repo',
3351
'RepositoryFormatCHK2',
3353
format_registry.register_lazy(
3354
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3355
'bzrlib.repofmt.groupcompress_repo',
3356
'RepositoryFormat2a',
3360
class InterRepository(InterObject):
3361
"""This class represents operations taking place between two repositories.
3363
Its instances have methods like copy_content and fetch, and contain
3364
references to the source and target repositories these operations can be
3367
Often we will provide convenience methods on 'repository' which carry out
3368
operations with another repository - they will always forward to
3369
InterRepository.get(other).method_name(parameters).
3372
_walk_to_common_revisions_batch_size = 50
3374
"""The available optimised InterRepository types."""
3377
def copy_content(self, revision_id=None):
3378
"""Make a complete copy of the content in self into destination.
3380
This is a destructive operation! Do not use it on existing
3383
:param revision_id: Only copy the content needed to construct
3384
revision_id and its parents.
3387
self.target.set_make_working_trees(self.source.make_working_trees())
3388
except NotImplementedError:
3390
self.target.fetch(self.source, revision_id=revision_id)
3393
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3395
"""Fetch the content required to construct revision_id.
3397
The content is copied from self.source to self.target.
3399
:param revision_id: if None all content is copied, if NULL_REVISION no
3401
:param pb: optional progress bar to use for progress reports. If not
3402
provided a default one will be created.
3405
from bzrlib.fetch import RepoFetcher
3406
f = RepoFetcher(to_repository=self.target,
3407
from_repository=self.source,
3408
last_revision=revision_id,
3409
fetch_spec=fetch_spec,
3410
pb=pb, find_ghosts=find_ghosts)
3412
def _walk_to_common_revisions(self, revision_ids):
3413
"""Walk out from revision_ids in source to revisions target has.
3415
:param revision_ids: The start point for the search.
3416
:return: A set of revision ids.
3418
target_graph = self.target.get_graph()
3419
revision_ids = frozenset(revision_ids)
3420
missing_revs = set()
3421
source_graph = self.source.get_graph()
3422
# ensure we don't pay silly lookup costs.
3423
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3424
null_set = frozenset([_mod_revision.NULL_REVISION])
3425
searcher_exhausted = False
3429
# Iterate the searcher until we have enough next_revs
3430
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3432
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3433
next_revs.update(next_revs_part)
3434
ghosts.update(ghosts_part)
3435
except StopIteration:
3436
searcher_exhausted = True
3438
# If there are ghosts in the source graph, and the caller asked for
3439
# them, make sure that they are present in the target.
3440
# We don't care about other ghosts as we can't fetch them and
3441
# haven't been asked to.
3442
ghosts_to_check = set(revision_ids.intersection(ghosts))
3443
revs_to_get = set(next_revs).union(ghosts_to_check)
3445
have_revs = set(target_graph.get_parent_map(revs_to_get))
3446
# we always have NULL_REVISION present.
3447
have_revs = have_revs.union(null_set)
3448
# Check if the target is missing any ghosts we need.
3449
ghosts_to_check.difference_update(have_revs)
3451
# One of the caller's revision_ids is a ghost in both the
3452
# source and the target.
3453
raise errors.NoSuchRevision(
3454
self.source, ghosts_to_check.pop())
3455
missing_revs.update(next_revs - have_revs)
3456
# Because we may have walked past the original stop point, make
3457
# sure everything is stopped
3458
stop_revs = searcher.find_seen_ancestors(have_revs)
3459
searcher.stop_searching_any(stop_revs)
3460
if searcher_exhausted:
3462
return searcher.get_result()
3465
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3466
"""Return the revision ids that source has that target does not.
3468
:param revision_id: only return revision ids included by this
3470
:param find_ghosts: If True find missing revisions in deep history
3471
rather than just finding the surface difference.
3472
:return: A bzrlib.graph.SearchResult.
3474
# stop searching at found target revisions.
3475
if not find_ghosts and revision_id is not None:
3476
return self._walk_to_common_revisions([revision_id])
3477
# generic, possibly worst case, slow code path.
3478
target_ids = set(self.target.all_revision_ids())
3479
if revision_id is not None:
3480
source_ids = self.source.get_ancestry(revision_id)
3481
if source_ids[0] is not None:
3482
raise AssertionError()
3485
source_ids = self.source.all_revision_ids()
3486
result_set = set(source_ids).difference(target_ids)
3487
return self.source.revision_ids_to_search_result(result_set)
3490
def _same_model(source, target):
3491
"""True if source and target have the same data representation.
3493
Note: this is always called on the base class; overriding it in a
3494
subclass will have no effect.
3497
InterRepository._assert_same_model(source, target)
3499
except errors.IncompatibleRepositories, e:
3503
def _assert_same_model(source, target):
3504
"""Raise an exception if two repositories do not use the same model.
3506
if source.supports_rich_root() != target.supports_rich_root():
3507
raise errors.IncompatibleRepositories(source, target,
3508
"different rich-root support")
3509
if source._serializer != target._serializer:
3510
raise errors.IncompatibleRepositories(source, target,
3511
"different serializers")
3514
class InterSameDataRepository(InterRepository):
3515
"""Code for converting between repositories that represent the same data.
3517
Data format and model must match for this to work.
3521
def _get_repo_format_to_test(self):
3522
"""Repository format for testing with.
3524
InterSameData can pull from subtree to subtree and from non-subtree to
3525
non-subtree, so we test this with the richest repository format.
3527
from bzrlib.repofmt import knitrepo
3528
return knitrepo.RepositoryFormatKnit3()
3531
def is_compatible(source, target):
3532
return InterRepository._same_model(source, target)
3535
class InterWeaveRepo(InterSameDataRepository):
3536
"""Optimised code paths between Weave based repositories.
3538
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3539
implemented lazy inter-object optimisation.
3543
def _get_repo_format_to_test(self):
3544
from bzrlib.repofmt import weaverepo
3545
return weaverepo.RepositoryFormat7()
3548
def is_compatible(source, target):
3549
"""Be compatible with known Weave formats.
3551
We don't test for the stores being of specific types because that
3552
could lead to confusing results, and there is no need to be
3555
from bzrlib.repofmt.weaverepo import (
3561
return (isinstance(source._format, (RepositoryFormat5,
3563
RepositoryFormat7)) and
3564
isinstance(target._format, (RepositoryFormat5,
3566
RepositoryFormat7)))
3567
except AttributeError:
3571
def copy_content(self, revision_id=None):
3572
"""See InterRepository.copy_content()."""
3573
# weave specific optimised path:
3575
self.target.set_make_working_trees(self.source.make_working_trees())
3576
except (errors.RepositoryUpgradeRequired, NotImplemented):
3578
# FIXME do not peek!
3579
if self.source._transport.listable():
3580
pb = ui.ui_factory.nested_progress_bar()
3582
self.target.texts.insert_record_stream(
3583
self.source.texts.get_record_stream(
3584
self.source.texts.keys(), 'topological', False))
3585
pb.update('copying inventory', 0, 1)
3586
self.target.inventories.insert_record_stream(
3587
self.source.inventories.get_record_stream(
3588
self.source.inventories.keys(), 'topological', False))
3589
self.target.signatures.insert_record_stream(
3590
self.source.signatures.get_record_stream(
3591
self.source.signatures.keys(),
3593
self.target.revisions.insert_record_stream(
3594
self.source.revisions.get_record_stream(
3595
self.source.revisions.keys(),
3596
'topological', True))
3600
self.target.fetch(self.source, revision_id=revision_id)
3603
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3604
"""See InterRepository.missing_revision_ids()."""
3605
# we want all revisions to satisfy revision_id in source.
3606
# but we don't want to stat every file here and there.
3607
# we want then, all revisions other needs to satisfy revision_id
3608
# checked, but not those that we have locally.
3609
# so the first thing is to get a subset of the revisions to
3610
# satisfy revision_id in source, and then eliminate those that
3611
# we do already have.
3612
# this is slow on high latency connection to self, but as this
3613
# disk format scales terribly for push anyway due to rewriting
3614
# inventory.weave, this is considered acceptable.
3616
if revision_id is not None:
3617
source_ids = self.source.get_ancestry(revision_id)
3618
if source_ids[0] is not None:
3619
raise AssertionError()
3622
source_ids = self.source._all_possible_ids()
3623
source_ids_set = set(source_ids)
3624
# source_ids is the worst possible case we may need to pull.
3625
# now we want to filter source_ids against what we actually
3626
# have in target, but don't try to check for existence where we know
3627
# we do not have a revision as that would be pointless.
3628
target_ids = set(self.target._all_possible_ids())
3629
possibly_present_revisions = target_ids.intersection(source_ids_set)
3630
actually_present_revisions = set(
3631
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3632
required_revisions = source_ids_set.difference(actually_present_revisions)
3633
if revision_id is not None:
3634
# we used get_ancestry to determine source_ids then we are assured all
3635
# revisions referenced are present as they are installed in topological order.
3636
# and the tip revision was validated by get_ancestry.
3637
result_set = required_revisions
3639
# if we just grabbed the possibly available ids, then
3640
# we only have an estimate of whats available and need to validate
3641
# that against the revision records.
3643
self.source._eliminate_revisions_not_present(required_revisions))
3644
return self.source.revision_ids_to_search_result(result_set)
3647
class InterKnitRepo(InterSameDataRepository):
3648
"""Optimised code paths between Knit based repositories."""
3651
def _get_repo_format_to_test(self):
3652
from bzrlib.repofmt import knitrepo
3653
return knitrepo.RepositoryFormatKnit1()
3656
def is_compatible(source, target):
3657
"""Be compatible with known Knit formats.
3659
We don't test for the stores being of specific types because that
3660
could lead to confusing results, and there is no need to be
3663
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3665
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3666
isinstance(target._format, RepositoryFormatKnit))
3667
except AttributeError:
3669
return are_knits and InterRepository._same_model(source, target)
3672
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3673
"""See InterRepository.missing_revision_ids()."""
3674
if revision_id is not None:
3675
source_ids = self.source.get_ancestry(revision_id)
3676
if source_ids[0] is not None:
3677
raise AssertionError()
3680
source_ids = self.source.all_revision_ids()
3681
source_ids_set = set(source_ids)
3682
# source_ids is the worst possible case we may need to pull.
3683
# now we want to filter source_ids against what we actually
3684
# have in target, but don't try to check for existence where we know
3685
# we do not have a revision as that would be pointless.
3686
target_ids = set(self.target.all_revision_ids())
3687
possibly_present_revisions = target_ids.intersection(source_ids_set)
3688
actually_present_revisions = set(
3689
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3690
required_revisions = source_ids_set.difference(actually_present_revisions)
3691
if revision_id is not None:
3692
# we used get_ancestry to determine source_ids then we are assured all
3693
# revisions referenced are present as they are installed in topological order.
3694
# and the tip revision was validated by get_ancestry.
3695
result_set = required_revisions
3697
# if we just grabbed the possibly available ids, then
3698
# we only have an estimate of whats available and need to validate
3699
# that against the revision records.
3701
self.source._eliminate_revisions_not_present(required_revisions))
3702
return self.source.revision_ids_to_search_result(result_set)
3705
class InterDifferingSerializer(InterRepository):
3708
def _get_repo_format_to_test(self):
3712
def is_compatible(source, target):
3713
"""Be compatible with Knit2 source and Knit3 target"""
3714
# This is redundant with format.check_conversion_target(), however that
3715
# raises an exception, and we just want to say "False" as in we won't
3716
# support converting between these formats.
3717
if 'IDS_never' in debug.debug_flags:
3719
if source.supports_rich_root() and not target.supports_rich_root():
3721
if (source._format.supports_tree_reference
3722
and not target._format.supports_tree_reference):
3724
if target._fallback_repositories and target._format.supports_chks:
3725
# IDS doesn't know how to copy CHKs for the parent inventories it
3726
# adds to stacked repos.
3728
if 'IDS_always' in debug.debug_flags:
3730
# Only use this code path for local source and target. IDS does far
3731
# too much IO (both bandwidth and roundtrips) over a network.
3732
if not source.bzrdir.transport.base.startswith('file:///'):
3734
if not target.bzrdir.transport.base.startswith('file:///'):
3738
def _get_trees(self, revision_ids, cache):
3740
for rev_id in revision_ids:
3742
possible_trees.append((rev_id, cache[rev_id]))
3744
# Not cached, but inventory might be present anyway.
3746
tree = self.source.revision_tree(rev_id)
3747
except errors.NoSuchRevision:
3748
# Nope, parent is ghost.
3751
cache[rev_id] = tree
3752
possible_trees.append((rev_id, tree))
3753
return possible_trees
3755
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3756
"""Get the best delta and base for this revision.
3758
:return: (basis_id, delta)
3761
# Generate deltas against each tree, to find the shortest.
3762
texts_possibly_new_in_tree = set()
3763
for basis_id, basis_tree in possible_trees:
3764
delta = tree.inventory._make_delta(basis_tree.inventory)
3765
for old_path, new_path, file_id, new_entry in delta:
3766
if new_path is None:
3767
# This file_id isn't present in the new rev, so we don't
3771
# Rich roots are handled elsewhere...
3773
kind = new_entry.kind
3774
if kind != 'directory' and kind != 'file':
3775
# No text record associated with this inventory entry.
3777
# This is a directory or file that has changed somehow.
3778
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3779
deltas.append((len(delta), basis_id, delta))
3781
return deltas[0][1:]
3783
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3784
"""Find all parent revisions that are absent, but for which the
3785
inventory is present, and copy those inventories.
3787
This is necessary to preserve correctness when the source is stacked
3788
without fallbacks configured. (Note that in cases like upgrade the
3789
source may be not have _fallback_repositories even though it is
3793
for parents in parent_map.values():
3794
parent_revs.update(parents)
3795
present_parents = self.source.get_parent_map(parent_revs)
3796
absent_parents = set(parent_revs).difference(present_parents)
3797
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3798
(rev_id,) for rev_id in absent_parents)
3799
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3800
for parent_tree in self.source.revision_trees(parent_inv_ids):
3801
current_revision_id = parent_tree.get_revision_id()
3802
parents_parents_keys = parent_invs_keys_for_stacking[
3803
(current_revision_id,)]
3804
parents_parents = [key[-1] for key in parents_parents_keys]
3805
basis_id = _mod_revision.NULL_REVISION
3806
basis_tree = self.source.revision_tree(basis_id)
3807
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3808
self.target.add_inventory_by_delta(
3809
basis_id, delta, current_revision_id, parents_parents)
3810
cache[current_revision_id] = parent_tree
3812
def _fetch_batch(self, revision_ids, basis_id, cache):
3813
"""Fetch across a few revisions.
3815
:param revision_ids: The revisions to copy
3816
:param basis_id: The revision_id of a tree that must be in cache, used
3817
as a basis for delta when no other base is available
3818
:param cache: A cache of RevisionTrees that we can use.
3819
:return: The revision_id of the last converted tree. The RevisionTree
3820
for it will be in cache
3822
# Walk though all revisions; get inventory deltas, copy referenced
3823
# texts that delta references, insert the delta, revision and
3825
root_keys_to_create = set()
3828
pending_revisions = []
3829
parent_map = self.source.get_parent_map(revision_ids)
3830
self._fetch_parent_invs_for_stacking(parent_map, cache)
3831
for tree in self.source.revision_trees(revision_ids):
3832
# Find a inventory delta for this revision.
3833
# Find text entries that need to be copied, too.
3834
current_revision_id = tree.get_revision_id()
3835
parent_ids = parent_map.get(current_revision_id, ())
3836
parent_trees = self._get_trees(parent_ids, cache)
3837
possible_trees = list(parent_trees)
3838
if len(possible_trees) == 0:
3839
# There either aren't any parents, or the parents are ghosts,
3840
# so just use the last converted tree.
3841
possible_trees.append((basis_id, cache[basis_id]))
3842
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3844
if self._converting_to_rich_root:
3845
self._revision_id_to_root_id[current_revision_id] = \
3847
# Determine which texts are in present in this revision but not in
3848
# any of the available parents.
3849
texts_possibly_new_in_tree = set()
3850
for old_path, new_path, file_id, entry in delta:
3851
if new_path is None:
3852
# This file_id isn't present in the new rev
3856
if not self.target.supports_rich_root():
3857
# The target doesn't support rich root, so we don't
3860
if self._converting_to_rich_root:
3861
# This can't be copied normally, we have to insert
3863
root_keys_to_create.add((file_id, entry.revision))
3866
texts_possibly_new_in_tree.add((file_id, entry.revision))
3867
for basis_id, basis_tree in possible_trees:
3868
basis_inv = basis_tree.inventory
3869
for file_key in list(texts_possibly_new_in_tree):
3870
file_id, file_revision = file_key
3872
entry = basis_inv[file_id]
3873
except errors.NoSuchId:
3875
if entry.revision == file_revision:
3876
texts_possibly_new_in_tree.remove(file_key)
3877
text_keys.update(texts_possibly_new_in_tree)
3878
revision = self.source.get_revision(current_revision_id)
3879
pending_deltas.append((basis_id, delta,
3880
current_revision_id, revision.parent_ids))
3881
pending_revisions.append(revision)
3882
cache[current_revision_id] = tree
3883
basis_id = current_revision_id
3885
from_texts = self.source.texts
3886
to_texts = self.target.texts
3887
if root_keys_to_create:
3888
from bzrlib.fetch import _new_root_data_stream
3889
root_stream = _new_root_data_stream(
3890
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3892
to_texts.insert_record_stream(root_stream)
3893
to_texts.insert_record_stream(from_texts.get_record_stream(
3894
text_keys, self.target._format._fetch_order,
3895
not self.target._format._fetch_uses_deltas))
3896
# insert inventory deltas
3897
for delta in pending_deltas:
3898
self.target.add_inventory_by_delta(*delta)
3899
if self.target._fallback_repositories:
3900
# Make sure this stacked repository has all the parent inventories
3901
# for the new revisions that we are about to insert. We do this
3902
# before adding the revisions so that no revision is added until
3903
# all the inventories it may depend on are added.
3904
# Note that this is overzealous, as we may have fetched these in an
3907
revision_ids = set()
3908
for revision in pending_revisions:
3909
revision_ids.add(revision.revision_id)
3910
parent_ids.update(revision.parent_ids)
3911
parent_ids.difference_update(revision_ids)
3912
parent_ids.discard(_mod_revision.NULL_REVISION)
3913
parent_map = self.source.get_parent_map(parent_ids)
3914
# we iterate over parent_map and not parent_ids because we don't
3915
# want to try copying any revision which is a ghost
3916
for parent_tree in self.source.revision_trees(parent_map):
3917
current_revision_id = parent_tree.get_revision_id()
3918
parents_parents = parent_map[current_revision_id]
3919
possible_trees = self._get_trees(parents_parents, cache)
3920
if len(possible_trees) == 0:
3921
# There either aren't any parents, or the parents are
3922
# ghosts, so just use the last converted tree.
3923
possible_trees.append((basis_id, cache[basis_id]))
3924
basis_id, delta = self._get_delta_for_revision(parent_tree,
3925
parents_parents, possible_trees)
3926
self.target.add_inventory_by_delta(
3927
basis_id, delta, current_revision_id, parents_parents)
3928
# insert signatures and revisions
3929
for revision in pending_revisions:
3931
signature = self.source.get_signature_text(
3932
revision.revision_id)
3933
self.target.add_signature_text(revision.revision_id,
3935
except errors.NoSuchRevision:
3937
self.target.add_revision(revision.revision_id, revision)
3940
def _fetch_all_revisions(self, revision_ids, pb):
3941
"""Fetch everything for the list of revisions.
3943
:param revision_ids: The list of revisions to fetch. Must be in
3945
:param pb: A ProgressTask
3948
basis_id, basis_tree = self._get_basis(revision_ids[0])
3950
cache = lru_cache.LRUCache(100)
3951
cache[basis_id] = basis_tree
3952
del basis_tree # We don't want to hang on to it here
3954
for offset in range(0, len(revision_ids), batch_size):
3955
self.target.start_write_group()
3957
pb.update('Transferring revisions', offset,
3959
batch = revision_ids[offset:offset+batch_size]
3960
basis_id = self._fetch_batch(batch, basis_id, cache)
3962
self.target.abort_write_group()
3965
hint = self.target.commit_write_group()
3968
if hints and self.target._format.pack_compresses:
3969
self.target.pack(hint=hints)
3970
pb.update('Transferring revisions', len(revision_ids),
3974
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3976
"""See InterRepository.fetch()."""
3977
if fetch_spec is not None:
3978
raise AssertionError("Not implemented yet...")
3979
if (not self.source.supports_rich_root()
3980
and self.target.supports_rich_root()):
3981
self._converting_to_rich_root = True
3982
self._revision_id_to_root_id = {}
3984
self._converting_to_rich_root = False
3985
revision_ids = self.target.search_missing_revision_ids(self.source,
3986
revision_id, find_ghosts=find_ghosts).get_keys()
3987
if not revision_ids:
3989
revision_ids = tsort.topo_sort(
3990
self.source.get_graph().get_parent_map(revision_ids))
3991
if not revision_ids:
3993
# Walk though all revisions; get inventory deltas, copy referenced
3994
# texts that delta references, insert the delta, revision and
3997
my_pb = ui.ui_factory.nested_progress_bar()
4000
symbol_versioning.warn(
4001
symbol_versioning.deprecated_in((1, 14, 0))
4002
% "pb parameter to fetch()")
4005
self._fetch_all_revisions(revision_ids, pb)
4007
if my_pb is not None:
4009
return len(revision_ids), 0
4011
def _get_basis(self, first_revision_id):
4012
"""Get a revision and tree which exists in the target.
4014
This assumes that first_revision_id is selected for transmission
4015
because all other ancestors are already present. If we can't find an
4016
ancestor we fall back to NULL_REVISION since we know that is safe.
4018
:return: (basis_id, basis_tree)
4020
first_rev = self.source.get_revision(first_revision_id)
4022
basis_id = first_rev.parent_ids[0]
4023
# only valid as a basis if the target has it
4024
self.target.get_revision(basis_id)
4025
# Try to get a basis tree - if its a ghost it will hit the
4026
# NoSuchRevision case.
4027
basis_tree = self.source.revision_tree(basis_id)
4028
except (IndexError, errors.NoSuchRevision):
4029
basis_id = _mod_revision.NULL_REVISION
4030
basis_tree = self.source.revision_tree(basis_id)
4031
return basis_id, basis_tree
4034
InterRepository.register_optimiser(InterDifferingSerializer)
4035
InterRepository.register_optimiser(InterSameDataRepository)
4036
InterRepository.register_optimiser(InterWeaveRepo)
4037
InterRepository.register_optimiser(InterKnitRepo)
4040
class CopyConverter(object):
4041
"""A repository conversion tool which just performs a copy of the content.
4043
This is slow but quite reliable.
4046
def __init__(self, target_format):
4047
"""Create a CopyConverter.
4049
:param target_format: The format the resulting repository should be.
4051
self.target_format = target_format
4053
def convert(self, repo, pb):
4054
"""Perform the conversion of to_convert, giving feedback via pb.
4056
:param to_convert: The disk object to convert.
4057
:param pb: a progress bar to use for progress information.
4062
# this is only useful with metadir layouts - separated repo content.
4063
# trigger an assertion if not such
4064
repo._format.get_format_string()
4065
self.repo_dir = repo.bzrdir
4066
self.step('Moving repository to repository.backup')
4067
self.repo_dir.transport.move('repository', 'repository.backup')
4068
backup_transport = self.repo_dir.transport.clone('repository.backup')
4069
repo._format.check_conversion_target(self.target_format)
4070
self.source_repo = repo._format.open(self.repo_dir,
4072
_override_transport=backup_transport)
4073
self.step('Creating new repository')
4074
converted = self.target_format.initialize(self.repo_dir,
4075
self.source_repo.is_shared())
4076
converted.lock_write()
4078
self.step('Copying content into repository.')
4079
self.source_repo.copy_content_into(converted)
4082
self.step('Deleting old repository content.')
4083
self.repo_dir.transport.delete_tree('repository.backup')
4084
self.pb.note('repository converted')
4086
def step(self, message):
4087
"""Update the pb by a step."""
4089
self.pb.update(message, self.count, self.total)
4101
def _unescaper(match, _map=_unescape_map):
4102
code = match.group(1)
4106
if not code.startswith('#'):
4108
return unichr(int(code[1:])).encode('utf8')
4114
def _unescape_xml(data):
4115
"""Unescape predefined XML entities in a string of data."""
4117
if _unescape_re is None:
4118
_unescape_re = re.compile('\&([^;]*);')
4119
return _unescape_re.sub(_unescaper, data)
4122
class _VersionedFileChecker(object):
4124
def __init__(self, repository, text_key_references=None, ancestors=None):
4125
self.repository = repository
4126
self.text_index = self.repository._generate_text_key_index(
4127
text_key_references=text_key_references, ancestors=ancestors)
4129
def calculate_file_version_parents(self, text_key):
4130
"""Calculate the correct parents for a file version according to
4133
parent_keys = self.text_index[text_key]
4134
if parent_keys == [_mod_revision.NULL_REVISION]:
4136
return tuple(parent_keys)
4138
def check_file_version_parents(self, texts, progress_bar=None):
4139
"""Check the parents stored in a versioned file are correct.
4141
It also detects file versions that are not referenced by their
4142
corresponding revision's inventory.
4144
:returns: A tuple of (wrong_parents, dangling_file_versions).
4145
wrong_parents is a dict mapping {revision_id: (stored_parents,
4146
correct_parents)} for each revision_id where the stored parents
4147
are not correct. dangling_file_versions is a set of (file_id,
4148
revision_id) tuples for versions that are present in this versioned
4149
file, but not used by the corresponding inventory.
4151
local_progress = None
4152
if progress_bar is None:
4153
local_progress = ui.ui_factory.nested_progress_bar()
4154
progress_bar = local_progress
4156
return self._check_file_version_parents(texts, progress_bar)
4159
local_progress.finished()
4161
def _check_file_version_parents(self, texts, progress_bar):
4162
"""See check_file_version_parents."""
4164
self.file_ids = set([file_id for file_id, _ in
4165
self.text_index.iterkeys()])
4166
# text keys is now grouped by file_id
4167
n_versions = len(self.text_index)
4168
progress_bar.update('loading text store', 0, n_versions)
4169
parent_map = self.repository.texts.get_parent_map(self.text_index)
4170
# On unlistable transports this could well be empty/error...
4171
text_keys = self.repository.texts.keys()
4172
unused_keys = frozenset(text_keys) - set(self.text_index)
4173
for num, key in enumerate(self.text_index.iterkeys()):
4174
progress_bar.update('checking text graph', num, n_versions)
4175
correct_parents = self.calculate_file_version_parents(key)
4177
knit_parents = parent_map[key]
4178
except errors.RevisionNotPresent:
4181
if correct_parents != knit_parents:
4182
wrong_parents[key] = (knit_parents, correct_parents)
4183
return wrong_parents, unused_keys
4186
def _old_get_graph(repository, revision_id):
4187
"""DO NOT USE. That is all. I'm serious."""
4188
graph = repository.get_graph()
4189
revision_graph = dict(((key, value) for key, value in
4190
graph.iter_ancestry([revision_id]) if value is not None))
4191
return _strip_NULL_ghosts(revision_graph)
4194
def _strip_NULL_ghosts(revision_graph):
4195
"""Also don't use this. more compatibility code for unmigrated clients."""
4196
# Filter ghosts, and null:
4197
if _mod_revision.NULL_REVISION in revision_graph:
4198
del revision_graph[_mod_revision.NULL_REVISION]
4199
for key, parents in revision_graph.items():
4200
revision_graph[key] = tuple(parent for parent in parents if parent
4202
return revision_graph
4205
class StreamSink(object):
4206
"""An object that can insert a stream into a repository.
4208
This interface handles the complexity of reserialising inventories and
4209
revisions from different formats, and allows unidirectional insertion into
4210
stacked repositories without looking for the missing basis parents
4214
def __init__(self, target_repo):
4215
self.target_repo = target_repo
4217
def insert_stream(self, stream, src_format, resume_tokens):
4218
"""Insert a stream's content into the target repository.
4220
:param src_format: a bzr repository format.
4222
:return: a list of resume tokens and an iterable of keys additional
4223
items required before the insertion can be completed.
4225
self.target_repo.lock_write()
4228
self.target_repo.resume_write_group(resume_tokens)
4231
self.target_repo.start_write_group()
4234
# locked_insert_stream performs a commit|suspend.
4235
return self._locked_insert_stream(stream, src_format, is_resume)
4237
self.target_repo.abort_write_group(suppress_errors=True)
4240
self.target_repo.unlock()
4242
def _locked_insert_stream(self, stream, src_format, is_resume):
4243
to_serializer = self.target_repo._format._serializer
4244
src_serializer = src_format._serializer
4246
if to_serializer == src_serializer:
4247
# If serializers match and the target is a pack repository, set the
4248
# write cache size on the new pack. This avoids poor performance
4249
# on transports where append is unbuffered (such as
4250
# RemoteTransport). This is safe to do because nothing should read
4251
# back from the target repository while a stream with matching
4252
# serialization is being inserted.
4253
# The exception is that a delta record from the source that should
4254
# be a fulltext may need to be expanded by the target (see
4255
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4256
# explicitly flush any buffered writes first in that rare case.
4258
new_pack = self.target_repo._pack_collection._new_pack
4259
except AttributeError:
4260
# Not a pack repository
4263
new_pack.set_write_cache_size(1024*1024)
4264
for substream_type, substream in stream:
4265
if 'stream' in debug.debug_flags:
4266
mutter('inserting substream: %s', substream_type)
4267
if substream_type == 'texts':
4268
self.target_repo.texts.insert_record_stream(substream)
4269
elif substream_type == 'inventories':
4270
if src_serializer == to_serializer:
4271
self.target_repo.inventories.insert_record_stream(
4274
self._extract_and_insert_inventories(
4275
substream, src_serializer)
4276
elif substream_type == 'inventory-deltas':
4277
self._extract_and_insert_inventory_deltas(
4278
substream, src_serializer)
4279
elif substream_type == 'chk_bytes':
4280
# XXX: This doesn't support conversions, as it assumes the
4281
# conversion was done in the fetch code.
4282
self.target_repo.chk_bytes.insert_record_stream(substream)
4283
elif substream_type == 'revisions':
4284
# This may fallback to extract-and-insert more often than
4285
# required if the serializers are different only in terms of
4287
if src_serializer == to_serializer:
4288
self.target_repo.revisions.insert_record_stream(
4291
self._extract_and_insert_revisions(substream,
4293
elif substream_type == 'signatures':
4294
self.target_repo.signatures.insert_record_stream(substream)
4296
raise AssertionError('kaboom! %s' % (substream_type,))
4297
# Done inserting data, and the missing_keys calculations will try to
4298
# read back from the inserted data, so flush the writes to the new pack
4299
# (if this is pack format).
4300
if new_pack is not None:
4301
new_pack._write_data('', flush=True)
4302
# Find all the new revisions (including ones from resume_tokens)
4303
missing_keys = self.target_repo.get_missing_parent_inventories(
4304
check_for_missing_texts=is_resume)
4306
for prefix, versioned_file in (
4307
('texts', self.target_repo.texts),
4308
('inventories', self.target_repo.inventories),
4309
('revisions', self.target_repo.revisions),
4310
('signatures', self.target_repo.signatures),
4311
('chk_bytes', self.target_repo.chk_bytes),
4313
if versioned_file is None:
4315
missing_keys.update((prefix,) + key for key in
4316
versioned_file.get_missing_compression_parent_keys())
4317
except NotImplementedError:
4318
# cannot even attempt suspending, and missing would have failed
4319
# during stream insertion.
4320
missing_keys = set()
4323
# suspend the write group and tell the caller what we is
4324
# missing. We know we can suspend or else we would not have
4325
# entered this code path. (All repositories that can handle
4326
# missing keys can handle suspending a write group).
4327
write_group_tokens = self.target_repo.suspend_write_group()
4328
return write_group_tokens, missing_keys
4329
hint = self.target_repo.commit_write_group()
4330
if (to_serializer != src_serializer and
4331
self.target_repo._format.pack_compresses):
4332
self.target_repo.pack(hint=hint)
4335
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4336
target_rich_root = self.target_repo._format.rich_root_data
4337
target_tree_refs = self.target_repo._format.supports_tree_reference
4338
for record in substream:
4339
# Insert the delta directly
4340
inventory_delta_bytes = record.get_bytes_as('fulltext')
4341
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4343
parse_result = deserialiser.parse_text_bytes(
4344
inventory_delta_bytes)
4345
except inventory_delta.IncompatibleInventoryDelta, err:
4346
trace.mutter("Incompatible delta: %s", err.msg)
4347
raise errors.IncompatibleRevision(self.target_repo._format)
4348
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4349
revision_id = new_id
4350
parents = [key[0] for key in record.parents]
4351
self.target_repo.add_inventory_by_delta(
4352
basis_id, inv_delta, revision_id, parents)
4354
def _extract_and_insert_inventories(self, substream, serializer,
4356
"""Generate a new inventory versionedfile in target, converting data.
4358
The inventory is retrieved from the source, (deserializing it), and
4359
stored in the target (reserializing it in a different format).
4361
target_rich_root = self.target_repo._format.rich_root_data
4362
target_tree_refs = self.target_repo._format.supports_tree_reference
4363
for record in substream:
4364
# It's not a delta, so it must be a fulltext in the source
4365
# serializer's format.
4366
bytes = record.get_bytes_as('fulltext')
4367
revision_id = record.key[0]
4368
inv = serializer.read_inventory_from_string(bytes, revision_id)
4369
parents = [key[0] for key in record.parents]
4370
self.target_repo.add_inventory(revision_id, inv, parents)
4371
# No need to keep holding this full inv in memory when the rest of
4372
# the substream is likely to be all deltas.
4375
def _extract_and_insert_revisions(self, substream, serializer):
4376
for record in substream:
4377
bytes = record.get_bytes_as('fulltext')
4378
revision_id = record.key[0]
4379
rev = serializer.read_revision_from_string(bytes)
4380
if rev.revision_id != revision_id:
4381
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4382
self.target_repo.add_revision(revision_id, rev)
4385
if self.target_repo._format._fetch_reconcile:
4386
self.target_repo.reconcile()
4389
class StreamSource(object):
4390
"""A source of a stream for fetching between repositories."""
4392
def __init__(self, from_repository, to_format):
4393
"""Create a StreamSource streaming from from_repository."""
4394
self.from_repository = from_repository
4395
self.to_format = to_format
4397
def delta_on_metadata(self):
4398
"""Return True if delta's are permitted on metadata streams.
4400
That is on revisions and signatures.
4402
src_serializer = self.from_repository._format._serializer
4403
target_serializer = self.to_format._serializer
4404
return (self.to_format._fetch_uses_deltas and
4405
src_serializer == target_serializer)
4407
def _fetch_revision_texts(self, revs):
4408
# fetch signatures first and then the revision texts
4409
# may need to be a InterRevisionStore call here.
4410
from_sf = self.from_repository.signatures
4411
# A missing signature is just skipped.
4412
keys = [(rev_id,) for rev_id in revs]
4413
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4415
self.to_format._fetch_order,
4416
not self.to_format._fetch_uses_deltas))
4417
# If a revision has a delta, this is actually expanded inside the
4418
# insert_record_stream code now, which is an alternate fix for
4420
from_rf = self.from_repository.revisions
4421
revisions = from_rf.get_record_stream(
4423
self.to_format._fetch_order,
4424
not self.delta_on_metadata())
4425
return [('signatures', signatures), ('revisions', revisions)]
4427
def _generate_root_texts(self, revs):
4428
"""This will be called by get_stream between fetching weave texts and
4429
fetching the inventory weave.
4431
if self._rich_root_upgrade():
4433
return bzrlib.fetch.Inter1and2Helper(
4434
self.from_repository).generate_root_texts(revs)
4438
def get_stream(self, search):
4440
revs = search.get_keys()
4441
graph = self.from_repository.get_graph()
4442
revs = tsort.topo_sort(graph.get_parent_map(revs))
4443
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4445
for knit_kind, file_id, revisions in data_to_fetch:
4446
if knit_kind != phase:
4448
# Make a new progress bar for this phase
4449
if knit_kind == "file":
4450
# Accumulate file texts
4451
text_keys.extend([(file_id, revision) for revision in
4453
elif knit_kind == "inventory":
4454
# Now copy the file texts.
4455
from_texts = self.from_repository.texts
4456
yield ('texts', from_texts.get_record_stream(
4457
text_keys, self.to_format._fetch_order,
4458
not self.to_format._fetch_uses_deltas))
4459
# Cause an error if a text occurs after we have done the
4462
# Before we process the inventory we generate the root
4463
# texts (if necessary) so that the inventories references
4465
for _ in self._generate_root_texts(revs):
4467
# we fetch only the referenced inventories because we do not
4468
# know for unselected inventories whether all their required
4469
# texts are present in the other repository - it could be
4471
for info in self._get_inventory_stream(revs):
4473
elif knit_kind == "signatures":
4474
# Nothing to do here; this will be taken care of when
4475
# _fetch_revision_texts happens.
4477
elif knit_kind == "revisions":
4478
for record in self._fetch_revision_texts(revs):
4481
raise AssertionError("Unknown knit kind %r" % knit_kind)
4483
def get_stream_for_missing_keys(self, missing_keys):
4484
# missing keys can only occur when we are byte copying and not
4485
# translating (because translation means we don't send
4486
# unreconstructable deltas ever).
4488
keys['texts'] = set()
4489
keys['revisions'] = set()
4490
keys['inventories'] = set()
4491
keys['chk_bytes'] = set()
4492
keys['signatures'] = set()
4493
for key in missing_keys:
4494
keys[key[0]].add(key[1:])
4495
if len(keys['revisions']):
4496
# If we allowed copying revisions at this point, we could end up
4497
# copying a revision without copying its required texts: a
4498
# violation of the requirements for repository integrity.
4499
raise AssertionError(
4500
'cannot copy revisions to fill in missing deltas %s' % (
4501
keys['revisions'],))
4502
for substream_kind, keys in keys.iteritems():
4503
vf = getattr(self.from_repository, substream_kind)
4504
if vf is None and keys:
4505
raise AssertionError(
4506
"cannot fill in keys for a versioned file we don't"
4507
" have: %s needs %s" % (substream_kind, keys))
4509
# No need to stream something we don't have
4511
if substream_kind == 'inventories':
4512
# Some missing keys are genuinely ghosts, filter those out.
4513
present = self.from_repository.inventories.get_parent_map(keys)
4514
revs = [key[0] for key in present]
4515
# Get the inventory stream more-or-less as we do for the
4516
# original stream; there's no reason to assume that records
4517
# direct from the source will be suitable for the sink. (Think
4518
# e.g. 2a -> 1.9-rich-root).
4519
for info in self._get_inventory_stream(revs, missing=True):
4523
# Ask for full texts always so that we don't need more round trips
4524
# after this stream.
4525
# Some of the missing keys are genuinely ghosts, so filter absent
4526
# records. The Sink is responsible for doing another check to
4527
# ensure that ghosts don't introduce missing data for future
4529
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4530
self.to_format._fetch_order, True))
4531
yield substream_kind, stream
4533
def inventory_fetch_order(self):
4534
if self._rich_root_upgrade():
4535
return 'topological'
4537
return self.to_format._fetch_order
4539
def _rich_root_upgrade(self):
4540
return (not self.from_repository._format.rich_root_data and
4541
self.to_format.rich_root_data)
4543
def _get_inventory_stream(self, revision_ids, missing=False):
4544
from_format = self.from_repository._format
4545
if (from_format.supports_chks and self.to_format.supports_chks and
4546
from_format.network_name() == self.to_format.network_name()):
4547
raise AssertionError(
4548
"this case should be handled by GroupCHKStreamSource")
4549
elif 'forceinvdeltas' in debug.debug_flags:
4550
return self._get_convertable_inventory_stream(revision_ids,
4551
delta_versus_null=missing)
4552
elif from_format.network_name() == self.to_format.network_name():
4554
return self._get_simple_inventory_stream(revision_ids,
4556
elif (not from_format.supports_chks and not self.to_format.supports_chks
4557
and from_format._serializer == self.to_format._serializer):
4558
# Essentially the same format.
4559
return self._get_simple_inventory_stream(revision_ids,
4562
# Any time we switch serializations, we want to use an
4563
# inventory-delta based approach.
4564
return self._get_convertable_inventory_stream(revision_ids,
4565
delta_versus_null=missing)
4567
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4568
# NB: This currently reopens the inventory weave in source;
4569
# using a single stream interface instead would avoid this.
4570
from_weave = self.from_repository.inventories
4572
delta_closure = True
4574
delta_closure = not self.delta_on_metadata()
4575
yield ('inventories', from_weave.get_record_stream(
4576
[(rev_id,) for rev_id in revision_ids],
4577
self.inventory_fetch_order(), delta_closure))
4579
def _get_convertable_inventory_stream(self, revision_ids,
4580
delta_versus_null=False):
4581
# The source is using CHKs, but the target either doesn't or it has a
4582
# different serializer. The StreamSink code expects to be able to
4583
# convert on the target, so we need to put bytes-on-the-wire that can
4584
# be converted. That means inventory deltas (if the remote is <1.19,
4585
# RemoteStreamSink will fallback to VFS to insert the deltas).
4586
yield ('inventory-deltas',
4587
self._stream_invs_as_deltas(revision_ids,
4588
delta_versus_null=delta_versus_null))
4590
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4591
"""Return a stream of inventory-deltas for the given rev ids.
4593
:param revision_ids: The list of inventories to transmit
4594
:param delta_versus_null: Don't try to find a minimal delta for this
4595
entry, instead compute the delta versus the NULL_REVISION. This
4596
effectively streams a complete inventory. Used for stuff like
4597
filling in missing parents, etc.
4599
from_repo = self.from_repository
4600
revision_keys = [(rev_id,) for rev_id in revision_ids]
4601
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4602
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4604
inventories = self.from_repository.iter_inventories(
4605
revision_ids, 'topological')
4606
format = from_repo._format
4607
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4608
inventory_cache = lru_cache.LRUCache(50)
4609
null_inventory = from_repo.revision_tree(
4610
_mod_revision.NULL_REVISION).inventory
4611
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4612
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4613
# repo back into a non-rich-root repo ought to be allowed)
4614
serializer = inventory_delta.InventoryDeltaSerializer(
4615
versioned_root=format.rich_root_data,
4616
tree_references=format.supports_tree_reference)
4617
for inv in inventories:
4618
key = (inv.revision_id,)
4619
parent_keys = parent_map.get(key, ())
4621
if not delta_versus_null and parent_keys:
4622
# The caller did not ask for complete inventories and we have
4623
# some parents that we can delta against. Make a delta against
4624
# each parent so that we can find the smallest.
4625
parent_ids = [parent_key[0] for parent_key in parent_keys]
4626
for parent_id in parent_ids:
4627
if parent_id not in invs_sent_so_far:
4628
# We don't know that the remote side has this basis, so
4631
if parent_id == _mod_revision.NULL_REVISION:
4632
parent_inv = null_inventory
4634
parent_inv = inventory_cache.get(parent_id, None)
4635
if parent_inv is None:
4636
parent_inv = from_repo.get_inventory(parent_id)
4637
candidate_delta = inv._make_delta(parent_inv)
4638
if (delta is None or
4639
len(delta) > len(candidate_delta)):
4640
delta = candidate_delta
4641
basis_id = parent_id
4643
# Either none of the parents ended up being suitable, or we
4644
# were asked to delta against NULL
4645
basis_id = _mod_revision.NULL_REVISION
4646
delta = inv._make_delta(null_inventory)
4647
invs_sent_so_far.add(inv.revision_id)
4648
inventory_cache[inv.revision_id] = inv
4649
delta_serialized = ''.join(
4650
serializer.delta_to_lines(basis_id, key[-1], delta))
4651
yield versionedfile.FulltextContentFactory(
4652
key, parent_keys, None, delta_serialized)
4655
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4656
stop_revision=None):
4657
"""Extend the partial history to include a given index
4659
If a stop_index is supplied, stop when that index has been reached.
4660
If a stop_revision is supplied, stop when that revision is
4661
encountered. Otherwise, stop when the beginning of history is
4664
:param stop_index: The index which should be present. When it is
4665
present, history extension will stop.
4666
:param stop_revision: The revision id which should be present. When
4667
it is encountered, history extension will stop.
4669
start_revision = partial_history_cache[-1]
4670
iterator = repo.iter_reverse_revision_history(start_revision)
4672
#skip the last revision in the list
4675
if (stop_index is not None and
4676
len(partial_history_cache) > stop_index):
4678
if partial_history_cache[-1] == stop_revision:
4680
revision_id = iterator.next()
4681
partial_history_cache.append(revision_id)
4682
except StopIteration:
added
removed
bzrlib/repository.py
