1
# Copyright (C) 2005, 2006, 2007, 2008, 2009 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
from bzrlib.lazy_import import lazy_import
18
lazy_import(globals(), """
41
revision as _mod_revision,
47
from bzrlib.bundle import serializer
48
from bzrlib.revisiontree import RevisionTree
49
from bzrlib.store.versioned import VersionedFileStore
50
from bzrlib.testament import Testament
53
from bzrlib.decorators import needs_read_lock, needs_write_lock, only_raises
54
from bzrlib.inter import InterObject
55
from bzrlib.inventory import (
61
from bzrlib.lock import _RelockDebugMixin
62
from bzrlib import registry
63
from bzrlib.trace import (
64
log_exception_quietly, note, mutter, mutter_callsite, warning)
67
# Old formats display a warning, but only once
68
_deprecation_warning_done = False
71
class CommitBuilder(object):
72
"""Provides an interface to build up a commit.
74
This allows describing a tree to be committed without needing to
75
know the internals of the format of the repository.
78
# all clients should supply tree roots.
79
record_root_entry = True
80
# the default CommitBuilder does not manage trees whose root is versioned.
81
_versioned_root = False
83
def __init__(self, repository, parents, config, timestamp=None,
84
timezone=None, committer=None, revprops=None,
86
"""Initiate a CommitBuilder.
88
:param repository: Repository to commit to.
89
:param parents: Revision ids of the parents of the new revision.
90
:param config: Configuration to use.
91
:param timestamp: Optional timestamp recorded for commit.
92
:param timezone: Optional timezone for timestamp.
93
:param committer: Optional committer to set for commit.
94
:param revprops: Optional dictionary of revision properties.
95
:param revision_id: Optional revision id.
100
self._committer = self._config.username()
102
self._committer = committer
104
self.new_inventory = Inventory(None)
105
self._new_revision_id = revision_id
106
self.parents = parents
107
self.repository = repository
110
if revprops is not None:
111
self._validate_revprops(revprops)
112
self._revprops.update(revprops)
114
if timestamp is None:
115
timestamp = time.time()
116
# Restrict resolution to 1ms
117
self._timestamp = round(timestamp, 3)
120
self._timezone = osutils.local_time_offset()
122
self._timezone = int(timezone)
124
self._generate_revision_if_needed()
125
self.__heads = graph.HeadsCache(repository.get_graph()).heads
126
self._basis_delta = []
127
# API compatibility, older code that used CommitBuilder did not call
128
# .record_delete(), which means the delta that is computed would not be
129
# valid. Callers that will call record_delete() should call
130
# .will_record_deletes() to indicate that.
131
self._recording_deletes = False
132
# memo'd check for no-op commits.
133
self._any_changes = False
135
def any_changes(self):
136
"""Return True if any entries were changed.
138
This includes merge-only changes. It is the core for the --unchanged
141
:return: True if any changes have occured.
143
return self._any_changes
145
def _validate_unicode_text(self, text, context):
146
"""Verify things like commit messages don't have bogus characters."""
148
raise ValueError('Invalid value for %s: %r' % (context, text))
150
def _validate_revprops(self, revprops):
151
for key, value in revprops.iteritems():
152
# We know that the XML serializers do not round trip '\r'
153
# correctly, so refuse to accept them
154
if not isinstance(value, basestring):
155
raise ValueError('revision property (%s) is not a valid'
156
' (unicode) string: %r' % (key, value))
157
self._validate_unicode_text(value,
158
'revision property (%s)' % (key,))
160
def commit(self, message):
161
"""Make the actual commit.
163
:return: The revision id of the recorded revision.
165
self._validate_unicode_text(message, 'commit message')
166
rev = _mod_revision.Revision(
167
timestamp=self._timestamp,
168
timezone=self._timezone,
169
committer=self._committer,
171
inventory_sha1=self.inv_sha1,
172
revision_id=self._new_revision_id,
173
properties=self._revprops)
174
rev.parent_ids = self.parents
175
self.repository.add_revision(self._new_revision_id, rev,
176
self.new_inventory, self._config)
177
self.repository.commit_write_group()
178
return self._new_revision_id
181
"""Abort the commit that is being built.
183
self.repository.abort_write_group()
185
def revision_tree(self):
186
"""Return the tree that was just committed.
188
After calling commit() this can be called to get a RevisionTree
189
representing the newly committed tree. This is preferred to
190
calling Repository.revision_tree() because that may require
191
deserializing the inventory, while we already have a copy in
194
if self.new_inventory is None:
195
self.new_inventory = self.repository.get_inventory(
196
self._new_revision_id)
197
return RevisionTree(self.repository, self.new_inventory,
198
self._new_revision_id)
200
def finish_inventory(self):
201
"""Tell the builder that the inventory is finished.
203
:return: The inventory id in the repository, which can be used with
204
repository.get_inventory.
206
if self.new_inventory is None:
207
# an inventory delta was accumulated without creating a new
209
basis_id = self.basis_delta_revision
210
# We ignore the 'inventory' returned by add_inventory_by_delta
211
# because self.new_inventory is used to hint to the rest of the
212
# system what code path was taken
213
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
214
basis_id, self._basis_delta, self._new_revision_id,
217
if self.new_inventory.root is None:
218
raise AssertionError('Root entry should be supplied to'
219
' record_entry_contents, as of bzr 0.10.')
220
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
221
self.new_inventory.revision_id = self._new_revision_id
222
self.inv_sha1 = self.repository.add_inventory(
223
self._new_revision_id,
227
return self._new_revision_id
229
def _gen_revision_id(self):
230
"""Return new revision-id."""
231
return generate_ids.gen_revision_id(self._config.username(),
234
def _generate_revision_if_needed(self):
235
"""Create a revision id if None was supplied.
237
If the repository can not support user-specified revision ids
238
they should override this function and raise CannotSetRevisionId
239
if _new_revision_id is not None.
241
:raises: CannotSetRevisionId
243
if self._new_revision_id is None:
244
self._new_revision_id = self._gen_revision_id()
245
self.random_revid = True
247
self.random_revid = False
249
def _heads(self, file_id, revision_ids):
250
"""Calculate the graph heads for revision_ids in the graph of file_id.
252
This can use either a per-file graph or a global revision graph as we
253
have an identity relationship between the two graphs.
255
return self.__heads(revision_ids)
257
def _check_root(self, ie, parent_invs, tree):
258
"""Helper for record_entry_contents.
260
:param ie: An entry being added.
261
:param parent_invs: The inventories of the parent revisions of the
263
:param tree: The tree that is being committed.
265
# In this revision format, root entries have no knit or weave When
266
# serializing out to disk and back in root.revision is always
268
ie.revision = self._new_revision_id
270
def _require_root_change(self, tree):
271
"""Enforce an appropriate root object change.
273
This is called once when record_iter_changes is called, if and only if
274
the root was not in the delta calculated by record_iter_changes.
276
:param tree: The tree which is being committed.
278
# NB: if there are no parents then this method is not called, so no
279
# need to guard on parents having length.
280
entry = entry_factory['directory'](tree.path2id(''), '',
282
entry.revision = self._new_revision_id
283
self._basis_delta.append(('', '', entry.file_id, entry))
285
def _get_delta(self, ie, basis_inv, path):
286
"""Get a delta against the basis inventory for ie."""
287
if ie.file_id not in basis_inv:
289
result = (None, path, ie.file_id, ie)
290
self._basis_delta.append(result)
292
elif ie != basis_inv[ie.file_id]:
294
# TODO: avoid tis id2path call.
295
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
296
self._basis_delta.append(result)
302
def get_basis_delta(self):
303
"""Return the complete inventory delta versus the basis inventory.
305
This has been built up with the calls to record_delete and
306
record_entry_contents. The client must have already called
307
will_record_deletes() to indicate that they will be generating a
310
:return: An inventory delta, suitable for use with apply_delta, or
311
Repository.add_inventory_by_delta, etc.
313
if not self._recording_deletes:
314
raise AssertionError("recording deletes not activated.")
315
return self._basis_delta
317
def record_delete(self, path, file_id):
318
"""Record that a delete occured against a basis tree.
320
This is an optional API - when used it adds items to the basis_delta
321
being accumulated by the commit builder. It cannot be called unless the
322
method will_record_deletes() has been called to inform the builder that
323
a delta is being supplied.
325
:param path: The path of the thing deleted.
326
:param file_id: The file id that was deleted.
328
if not self._recording_deletes:
329
raise AssertionError("recording deletes not activated.")
330
delta = (path, None, file_id, None)
331
self._basis_delta.append(delta)
332
self._any_changes = True
335
def will_record_deletes(self):
336
"""Tell the commit builder that deletes are being notified.
338
This enables the accumulation of an inventory delta; for the resulting
339
commit to be valid, deletes against the basis MUST be recorded via
340
builder.record_delete().
342
self._recording_deletes = True
344
basis_id = self.parents[0]
346
basis_id = _mod_revision.NULL_REVISION
347
self.basis_delta_revision = basis_id
349
def record_entry_contents(self, ie, parent_invs, path, tree,
351
"""Record the content of ie from tree into the commit if needed.
353
Side effect: sets ie.revision when unchanged
355
:param ie: An inventory entry present in the commit.
356
:param parent_invs: The inventories of the parent revisions of the
358
:param path: The path the entry is at in the tree.
359
:param tree: The tree which contains this entry and should be used to
361
:param content_summary: Summary data from the tree about the paths
362
content - stat, length, exec, sha/link target. This is only
363
accessed when the entry has a revision of None - that is when it is
364
a candidate to commit.
365
:return: A tuple (change_delta, version_recorded, fs_hash).
366
change_delta is an inventory_delta change for this entry against
367
the basis tree of the commit, or None if no change occured against
369
version_recorded is True if a new version of the entry has been
370
recorded. For instance, committing a merge where a file was only
371
changed on the other side will return (delta, False).
372
fs_hash is either None, or the hash details for the path (currently
373
a tuple of the contents sha1 and the statvalue returned by
374
tree.get_file_with_stat()).
376
if self.new_inventory.root is None:
377
if ie.parent_id is not None:
378
raise errors.RootMissing()
379
self._check_root(ie, parent_invs, tree)
380
if ie.revision is None:
381
kind = content_summary[0]
383
# ie is carried over from a prior commit
385
# XXX: repository specific check for nested tree support goes here - if
386
# the repo doesn't want nested trees we skip it ?
387
if (kind == 'tree-reference' and
388
not self.repository._format.supports_tree_reference):
389
# mismatch between commit builder logic and repository:
390
# this needs the entry creation pushed down into the builder.
391
raise NotImplementedError('Missing repository subtree support.')
392
self.new_inventory.add(ie)
394
# TODO: slow, take it out of the inner loop.
396
basis_inv = parent_invs[0]
398
basis_inv = Inventory(root_id=None)
400
# ie.revision is always None if the InventoryEntry is considered
401
# for committing. We may record the previous parents revision if the
402
# content is actually unchanged against a sole head.
403
if ie.revision is not None:
404
if not self._versioned_root and path == '':
405
# repositories that do not version the root set the root's
406
# revision to the new commit even when no change occurs (more
407
# specifically, they do not record a revision on the root; and
408
# the rev id is assigned to the root during deserialisation -
409
# this masks when a change may have occurred against the basis.
410
# To match this we always issue a delta, because the revision
411
# of the root will always be changing.
412
if ie.file_id in basis_inv:
413
delta = (basis_inv.id2path(ie.file_id), path,
417
delta = (None, path, ie.file_id, ie)
418
self._basis_delta.append(delta)
419
return delta, False, None
421
# we don't need to commit this, because the caller already
422
# determined that an existing revision of this file is
423
# appropriate. If its not being considered for committing then
424
# it and all its parents to the root must be unaltered so
425
# no-change against the basis.
426
if ie.revision == self._new_revision_id:
427
raise AssertionError("Impossible situation, a skipped "
428
"inventory entry (%r) claims to be modified in this "
429
"commit (%r).", (ie, self._new_revision_id))
430
return None, False, None
431
# XXX: Friction: parent_candidates should return a list not a dict
432
# so that we don't have to walk the inventories again.
433
parent_candiate_entries = ie.parent_candidates(parent_invs)
434
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
436
for inv in parent_invs:
437
if ie.file_id in inv:
438
old_rev = inv[ie.file_id].revision
439
if old_rev in head_set:
440
heads.append(inv[ie.file_id].revision)
441
head_set.remove(inv[ie.file_id].revision)
444
# now we check to see if we need to write a new record to the
446
# We write a new entry unless there is one head to the ancestors, and
447
# the kind-derived content is unchanged.
449
# Cheapest check first: no ancestors, or more the one head in the
450
# ancestors, we write a new node.
454
# There is a single head, look it up for comparison
455
parent_entry = parent_candiate_entries[heads[0]]
456
# if the non-content specific data has changed, we'll be writing a
458
if (parent_entry.parent_id != ie.parent_id or
459
parent_entry.name != ie.name):
461
# now we need to do content specific checks:
463
# if the kind changed the content obviously has
464
if kind != parent_entry.kind:
466
# Stat cache fingerprint feedback for the caller - None as we usually
467
# don't generate one.
470
if content_summary[2] is None:
471
raise ValueError("Files must not have executable = None")
473
# We can't trust a check of the file length because of content
475
if (# if the exec bit has changed we have to store:
476
parent_entry.executable != content_summary[2]):
478
elif parent_entry.text_sha1 == content_summary[3]:
479
# all meta and content is unchanged (using a hash cache
480
# hit to check the sha)
481
ie.revision = parent_entry.revision
482
ie.text_size = parent_entry.text_size
483
ie.text_sha1 = parent_entry.text_sha1
484
ie.executable = parent_entry.executable
485
return self._get_delta(ie, basis_inv, path), False, None
487
# Either there is only a hash change(no hash cache entry,
488
# or same size content change), or there is no change on
490
# Provide the parent's hash to the store layer, so that the
491
# content is unchanged we will not store a new node.
492
nostore_sha = parent_entry.text_sha1
494
# We want to record a new node regardless of the presence or
495
# absence of a content change in the file.
497
ie.executable = content_summary[2]
498
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
500
text = file_obj.read()
504
ie.text_sha1, ie.text_size = self._add_text_to_weave(
505
ie.file_id, text, heads, nostore_sha)
506
# Let the caller know we generated a stat fingerprint.
507
fingerprint = (ie.text_sha1, stat_value)
508
except errors.ExistingContent:
509
# Turns out that the file content was unchanged, and we were
510
# only going to store a new node if it was changed. Carry over
512
ie.revision = parent_entry.revision
513
ie.text_size = parent_entry.text_size
514
ie.text_sha1 = parent_entry.text_sha1
515
ie.executable = parent_entry.executable
516
return self._get_delta(ie, basis_inv, path), False, None
517
elif kind == 'directory':
519
# all data is meta here, nothing specific to directory, so
521
ie.revision = parent_entry.revision
522
return self._get_delta(ie, basis_inv, path), False, None
523
self._add_text_to_weave(ie.file_id, '', heads, None)
524
elif kind == 'symlink':
525
current_link_target = content_summary[3]
527
# symlink target is not generic metadata, check if it has
529
if current_link_target != parent_entry.symlink_target:
532
# unchanged, carry over.
533
ie.revision = parent_entry.revision
534
ie.symlink_target = parent_entry.symlink_target
535
return self._get_delta(ie, basis_inv, path), False, None
536
ie.symlink_target = current_link_target
537
self._add_text_to_weave(ie.file_id, '', heads, None)
538
elif kind == 'tree-reference':
540
if content_summary[3] != parent_entry.reference_revision:
543
# unchanged, carry over.
544
ie.reference_revision = parent_entry.reference_revision
545
ie.revision = parent_entry.revision
546
return self._get_delta(ie, basis_inv, path), False, None
547
ie.reference_revision = content_summary[3]
548
if ie.reference_revision is None:
549
raise AssertionError("invalid content_summary for nested tree: %r"
550
% (content_summary,))
551
self._add_text_to_weave(ie.file_id, '', heads, None)
553
raise NotImplementedError('unknown kind')
554
ie.revision = self._new_revision_id
555
self._any_changes = True
556
return self._get_delta(ie, basis_inv, path), True, fingerprint
558
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
559
_entry_factory=entry_factory):
560
"""Record a new tree via iter_changes.
562
:param tree: The tree to obtain text contents from for changed objects.
563
:param basis_revision_id: The revision id of the tree the iter_changes
564
has been generated against. Currently assumed to be the same
565
as self.parents[0] - if it is not, errors may occur.
566
:param iter_changes: An iter_changes iterator with the changes to apply
567
to basis_revision_id. The iterator must not include any items with
568
a current kind of None - missing items must be either filtered out
569
or errored-on beefore record_iter_changes sees the item.
570
:param _entry_factory: Private method to bind entry_factory locally for
572
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
575
# Create an inventory delta based on deltas between all the parents and
576
# deltas between all the parent inventories. We use inventory delta's
577
# between the inventory objects because iter_changes masks
578
# last-changed-field only changes.
580
# file_id -> change map, change is fileid, paths, changed, versioneds,
581
# parents, names, kinds, executables
583
# {file_id -> revision_id -> inventory entry, for entries in parent
584
# trees that are not parents[0]
588
revtrees = list(self.repository.revision_trees(self.parents))
589
except errors.NoSuchRevision:
590
# one or more ghosts, slow path.
592
for revision_id in self.parents:
594
revtrees.append(self.repository.revision_tree(revision_id))
595
except errors.NoSuchRevision:
597
basis_revision_id = _mod_revision.NULL_REVISION
599
revtrees.append(self.repository.revision_tree(
600
_mod_revision.NULL_REVISION))
601
# The basis inventory from a repository
603
basis_inv = revtrees[0].inventory
605
basis_inv = self.repository.revision_tree(
606
_mod_revision.NULL_REVISION).inventory
607
if len(self.parents) > 0:
608
if basis_revision_id != self.parents[0] and not ghost_basis:
610
"arbitrary basis parents not yet supported with merges")
611
for revtree in revtrees[1:]:
612
for change in revtree.inventory._make_delta(basis_inv):
613
if change[1] is None:
614
# Not present in this parent.
616
if change[2] not in merged_ids:
617
if change[0] is not None:
618
basis_entry = basis_inv[change[2]]
619
merged_ids[change[2]] = [
621
basis_entry.revision,
624
parent_entries[change[2]] = {
626
basis_entry.revision:basis_entry,
628
change[3].revision:change[3],
631
merged_ids[change[2]] = [change[3].revision]
632
parent_entries[change[2]] = {change[3].revision:change[3]}
634
merged_ids[change[2]].append(change[3].revision)
635
parent_entries[change[2]][change[3].revision] = change[3]
638
# Setup the changes from the tree:
639
# changes maps file_id -> (change, [parent revision_ids])
641
for change in iter_changes:
642
# This probably looks up in basis_inv way to much.
643
if change[1][0] is not None:
644
head_candidate = [basis_inv[change[0]].revision]
647
changes[change[0]] = change, merged_ids.get(change[0],
649
unchanged_merged = set(merged_ids) - set(changes)
650
# Extend the changes dict with synthetic changes to record merges of
652
for file_id in unchanged_merged:
653
# Record a merged version of these items that did not change vs the
654
# basis. This can be either identical parallel changes, or a revert
655
# of a specific file after a merge. The recorded content will be
656
# that of the current tree (which is the same as the basis), but
657
# the per-file graph will reflect a merge.
658
# NB:XXX: We are reconstructing path information we had, this
659
# should be preserved instead.
660
# inv delta change: (file_id, (path_in_source, path_in_target),
661
# changed_content, versioned, parent, name, kind,
664
basis_entry = basis_inv[file_id]
665
except errors.NoSuchId:
666
# a change from basis->some_parents but file_id isn't in basis
667
# so was new in the merge, which means it must have changed
668
# from basis -> current, and as it hasn't the add was reverted
669
# by the user. So we discard this change.
673
(basis_inv.id2path(file_id), tree.id2path(file_id)),
675
(basis_entry.parent_id, basis_entry.parent_id),
676
(basis_entry.name, basis_entry.name),
677
(basis_entry.kind, basis_entry.kind),
678
(basis_entry.executable, basis_entry.executable))
679
changes[file_id] = (change, merged_ids[file_id])
680
# changes contains tuples with the change and a set of inventory
681
# candidates for the file.
683
# old_path, new_path, file_id, new_inventory_entry
684
seen_root = False # Is the root in the basis delta?
685
inv_delta = self._basis_delta
686
modified_rev = self._new_revision_id
687
for change, head_candidates in changes.values():
688
if change[3][1]: # versioned in target.
689
# Several things may be happening here:
690
# We may have a fork in the per-file graph
691
# - record a change with the content from tree
692
# We may have a change against < all trees
693
# - carry over the tree that hasn't changed
694
# We may have a change against all trees
695
# - record the change with the content from tree
698
entry = _entry_factory[kind](file_id, change[5][1],
700
head_set = self._heads(change[0], set(head_candidates))
703
for head_candidate in head_candidates:
704
if head_candidate in head_set:
705
heads.append(head_candidate)
706
head_set.remove(head_candidate)
709
# Could be a carry-over situation:
710
parent_entry_revs = parent_entries.get(file_id, None)
711
if parent_entry_revs:
712
parent_entry = parent_entry_revs.get(heads[0], None)
715
if parent_entry is None:
716
# The parent iter_changes was called against is the one
717
# that is the per-file head, so any change is relevant
718
# iter_changes is valid.
719
carry_over_possible = False
721
# could be a carry over situation
722
# A change against the basis may just indicate a merge,
723
# we need to check the content against the source of the
724
# merge to determine if it was changed after the merge
726
if (parent_entry.kind != entry.kind or
727
parent_entry.parent_id != entry.parent_id or
728
parent_entry.name != entry.name):
729
# Metadata common to all entries has changed
730
# against per-file parent
731
carry_over_possible = False
733
carry_over_possible = True
734
# per-type checks for changes against the parent_entry
737
# Cannot be a carry-over situation
738
carry_over_possible = False
739
# Populate the entry in the delta
741
# XXX: There is still a small race here: If someone reverts the content of a file
742
# after iter_changes examines and decides it has changed,
743
# we will unconditionally record a new version even if some
744
# other process reverts it while commit is running (with
745
# the revert happening after iter_changes did it's
748
entry.executable = True
750
entry.executable = False
751
if (carry_over_possible and
752
parent_entry.executable == entry.executable):
753
# Check the file length, content hash after reading
755
nostore_sha = parent_entry.text_sha1
758
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
760
text = file_obj.read()
764
entry.text_sha1, entry.text_size = self._add_text_to_weave(
765
file_id, text, heads, nostore_sha)
766
yield file_id, change[1][1], (entry.text_sha1, stat_value)
767
except errors.ExistingContent:
768
# No content change against a carry_over parent
769
# Perhaps this should also yield a fs hash update?
771
entry.text_size = parent_entry.text_size
772
entry.text_sha1 = parent_entry.text_sha1
773
elif kind == 'symlink':
775
entry.symlink_target = tree.get_symlink_target(file_id)
776
if (carry_over_possible and
777
parent_entry.symlink_target == entry.symlink_target):
780
self._add_text_to_weave(change[0], '', heads, None)
781
elif kind == 'directory':
782
if carry_over_possible:
785
# Nothing to set on the entry.
786
# XXX: split into the Root and nonRoot versions.
787
if change[1][1] != '' or self.repository.supports_rich_root():
788
self._add_text_to_weave(change[0], '', heads, None)
789
elif kind == 'tree-reference':
790
if not self.repository._format.supports_tree_reference:
791
# This isn't quite sane as an error, but we shouldn't
792
# ever see this code path in practice: tree's don't
793
# permit references when the repo doesn't support tree
795
raise errors.UnsupportedOperation(tree.add_reference,
797
reference_revision = tree.get_reference_revision(change[0])
798
entry.reference_revision = reference_revision
799
if (carry_over_possible and
800
parent_entry.reference_revision == reference_revision):
803
self._add_text_to_weave(change[0], '', heads, None)
805
raise AssertionError('unknown kind %r' % kind)
807
entry.revision = modified_rev
809
entry.revision = parent_entry.revision
812
new_path = change[1][1]
813
inv_delta.append((change[1][0], new_path, change[0], entry))
816
self.new_inventory = None
818
# This should perhaps be guarded by a check that the basis we
819
# commit against is the basis for the commit and if not do a delta
821
self._any_changes = True
823
# housekeeping root entry changes do not affect no-change commits.
824
self._require_root_change(tree)
825
self.basis_delta_revision = basis_revision_id
827
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
828
parent_keys = tuple([(file_id, parent) for parent in parents])
829
return self.repository.texts._add_text(
830
(file_id, self._new_revision_id), parent_keys, new_text,
831
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
834
class RootCommitBuilder(CommitBuilder):
835
"""This commitbuilder actually records the root id"""
837
# the root entry gets versioned properly by this builder.
838
_versioned_root = True
840
def _check_root(self, ie, parent_invs, tree):
841
"""Helper for record_entry_contents.
843
:param ie: An entry being added.
844
:param parent_invs: The inventories of the parent revisions of the
846
:param tree: The tree that is being committed.
849
def _require_root_change(self, tree):
850
"""Enforce an appropriate root object change.
852
This is called once when record_iter_changes is called, if and only if
853
the root was not in the delta calculated by record_iter_changes.
855
:param tree: The tree which is being committed.
857
# versioned roots do not change unless the tree found a change.
860
######################################################################
864
class Repository(_RelockDebugMixin):
865
"""Repository holding history for one or more branches.
867
The repository holds and retrieves historical information including
868
revisions and file history. It's normally accessed only by the Branch,
869
which views a particular line of development through that history.
871
The Repository builds on top of some byte storage facilies (the revisions,
872
signatures, inventories, texts and chk_bytes attributes) and a Transport,
873
which respectively provide byte storage and a means to access the (possibly
876
The byte storage facilities are addressed via tuples, which we refer to
877
as 'keys' throughout the code base. Revision_keys, inventory_keys and
878
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
879
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
880
byte string made up of a hash identifier and a hash value.
881
We use this interface because it allows low friction with the underlying
882
code that implements disk indices, network encoding and other parts of
885
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
886
the serialised revisions for the repository. This can be used to obtain
887
revision graph information or to access raw serialised revisions.
888
The result of trying to insert data into the repository via this store
889
is undefined: it should be considered read-only except for implementors
891
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
892
the serialised signatures for the repository. This can be used to
893
obtain access to raw serialised signatures. The result of trying to
894
insert data into the repository via this store is undefined: it should
895
be considered read-only except for implementors of repositories.
896
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
897
the serialised inventories for the repository. This can be used to
898
obtain unserialised inventories. The result of trying to insert data
899
into the repository via this store is undefined: it should be
900
considered read-only except for implementors of repositories.
901
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
902
texts of files and directories for the repository. This can be used to
903
obtain file texts or file graphs. Note that Repository.iter_file_bytes
904
is usually a better interface for accessing file texts.
905
The result of trying to insert data into the repository via this store
906
is undefined: it should be considered read-only except for implementors
908
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
909
any data the repository chooses to store or have indexed by its hash.
910
The result of trying to insert data into the repository via this store
911
is undefined: it should be considered read-only except for implementors
913
:ivar _transport: Transport for file access to repository, typically
914
pointing to .bzr/repository.
917
# What class to use for a CommitBuilder. Often its simpler to change this
918
# in a Repository class subclass rather than to override
919
# get_commit_builder.
920
_commit_builder_class = CommitBuilder
921
# The search regex used by xml based repositories to determine what things
922
# where changed in a single commit.
923
_file_ids_altered_regex = lazy_regex.lazy_compile(
924
r'file_id="(?P<file_id>[^"]+)"'
925
r'.* revision="(?P<revision_id>[^"]+)"'
928
def abort_write_group(self, suppress_errors=False):
929
"""Commit the contents accrued within the current write group.
931
:param suppress_errors: if true, abort_write_group will catch and log
932
unexpected errors that happen during the abort, rather than
933
allowing them to propagate. Defaults to False.
935
:seealso: start_write_group.
937
if self._write_group is not self.get_transaction():
938
# has an unlock or relock occured ?
941
'(suppressed) mismatched lock context and write group. %r, %r',
942
self._write_group, self.get_transaction())
944
raise errors.BzrError(
945
'mismatched lock context and write group. %r, %r' %
946
(self._write_group, self.get_transaction()))
948
self._abort_write_group()
949
except Exception, exc:
950
self._write_group = None
951
if not suppress_errors:
953
mutter('abort_write_group failed')
954
log_exception_quietly()
955
note('bzr: ERROR (ignored): %s', exc)
956
self._write_group = None
958
def _abort_write_group(self):
959
"""Template method for per-repository write group cleanup.
961
This is called during abort before the write group is considered to be
962
finished and should cleanup any internal state accrued during the write
963
group. There is no requirement that data handed to the repository be
964
*not* made available - this is not a rollback - but neither should any
965
attempt be made to ensure that data added is fully commited. Abort is
966
invoked when an error has occured so futher disk or network operations
967
may not be possible or may error and if possible should not be
971
def add_fallback_repository(self, repository):
972
"""Add a repository to use for looking up data not held locally.
974
:param repository: A repository.
976
if not self._format.supports_external_lookups:
977
raise errors.UnstackableRepositoryFormat(self._format, self.base)
979
# This repository will call fallback.unlock() when we transition to
980
# the unlocked state, so we make sure to increment the lock count
981
repository.lock_read()
982
self._check_fallback_repository(repository)
983
self._fallback_repositories.append(repository)
984
self.texts.add_fallback_versioned_files(repository.texts)
985
self.inventories.add_fallback_versioned_files(repository.inventories)
986
self.revisions.add_fallback_versioned_files(repository.revisions)
987
self.signatures.add_fallback_versioned_files(repository.signatures)
988
if self.chk_bytes is not None:
989
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
991
def _check_fallback_repository(self, repository):
992
"""Check that this repository can fallback to repository safely.
994
Raise an error if not.
996
:param repository: A repository to fallback to.
998
return InterRepository._assert_same_model(self, repository)
1000
def add_inventory(self, revision_id, inv, parents):
1001
"""Add the inventory inv to the repository as revision_id.
1003
:param parents: The revision ids of the parents that revision_id
1004
is known to have and are in the repository already.
1006
:returns: The validator(which is a sha1 digest, though what is sha'd is
1007
repository format specific) of the serialized inventory.
1009
if not self.is_in_write_group():
1010
raise AssertionError("%r not in write group" % (self,))
1011
_mod_revision.check_not_reserved_id(revision_id)
1012
if not (inv.revision_id is None or inv.revision_id == revision_id):
1013
raise AssertionError(
1014
"Mismatch between inventory revision"
1015
" id and insertion revid (%r, %r)"
1016
% (inv.revision_id, revision_id))
1017
if inv.root is None:
1018
raise AssertionError()
1019
return self._add_inventory_checked(revision_id, inv, parents)
1021
def _add_inventory_checked(self, revision_id, inv, parents):
1022
"""Add inv to the repository after checking the inputs.
1024
This function can be overridden to allow different inventory styles.
1026
:seealso: add_inventory, for the contract.
1028
inv_lines = self._serialise_inventory_to_lines(inv)
1029
return self._inventory_add_lines(revision_id, parents,
1030
inv_lines, check_content=False)
1032
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1033
parents, basis_inv=None, propagate_caches=False):
1034
"""Add a new inventory expressed as a delta against another revision.
1036
See the inventory developers documentation for the theory behind
1039
:param basis_revision_id: The inventory id the delta was created
1040
against. (This does not have to be a direct parent.)
1041
:param delta: The inventory delta (see Inventory.apply_delta for
1043
:param new_revision_id: The revision id that the inventory is being
1045
:param parents: The revision ids of the parents that revision_id is
1046
known to have and are in the repository already. These are supplied
1047
for repositories that depend on the inventory graph for revision
1048
graph access, as well as for those that pun ancestry with delta
1050
:param basis_inv: The basis inventory if it is already known,
1052
:param propagate_caches: If True, the caches for this inventory are
1053
copied to and updated for the result if possible.
1055
:returns: (validator, new_inv)
1056
The validator(which is a sha1 digest, though what is sha'd is
1057
repository format specific) of the serialized inventory, and the
1058
resulting inventory.
1060
if not self.is_in_write_group():
1061
raise AssertionError("%r not in write group" % (self,))
1062
_mod_revision.check_not_reserved_id(new_revision_id)
1063
basis_tree = self.revision_tree(basis_revision_id)
1064
basis_tree.lock_read()
1066
# Note that this mutates the inventory of basis_tree, which not all
1067
# inventory implementations may support: A better idiom would be to
1068
# return a new inventory, but as there is no revision tree cache in
1069
# repository this is safe for now - RBC 20081013
1070
if basis_inv is None:
1071
basis_inv = basis_tree.inventory
1072
basis_inv.apply_delta(delta)
1073
basis_inv.revision_id = new_revision_id
1074
return (self.add_inventory(new_revision_id, basis_inv, parents),
1079
def _inventory_add_lines(self, revision_id, parents, lines,
1080
check_content=True):
1081
"""Store lines in inv_vf and return the sha1 of the inventory."""
1082
parents = [(parent,) for parent in parents]
1083
result = self.inventories.add_lines((revision_id,), parents, lines,
1084
check_content=check_content)[0]
1085
self.inventories._access.flush()
1088
def add_revision(self, revision_id, rev, inv=None, config=None):
1089
"""Add rev to the revision store as revision_id.
1091
:param revision_id: the revision id to use.
1092
:param rev: The revision object.
1093
:param inv: The inventory for the revision. if None, it will be looked
1094
up in the inventory storer
1095
:param config: If None no digital signature will be created.
1096
If supplied its signature_needed method will be used
1097
to determine if a signature should be made.
1099
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1101
_mod_revision.check_not_reserved_id(revision_id)
1102
if config is not None and config.signature_needed():
1104
inv = self.get_inventory(revision_id)
1105
plaintext = Testament(rev, inv).as_short_text()
1106
self.store_revision_signature(
1107
gpg.GPGStrategy(config), plaintext, revision_id)
1108
# check inventory present
1109
if not self.inventories.get_parent_map([(revision_id,)]):
1111
raise errors.WeaveRevisionNotPresent(revision_id,
1114
# yes, this is not suitable for adding with ghosts.
1115
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1118
key = (revision_id,)
1119
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1120
self._add_revision(rev)
1122
def _add_revision(self, revision):
1123
text = self._serializer.write_revision_to_string(revision)
1124
key = (revision.revision_id,)
1125
parents = tuple((parent,) for parent in revision.parent_ids)
1126
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1128
def all_revision_ids(self):
1129
"""Returns a list of all the revision ids in the repository.
1131
This is conceptually deprecated because code should generally work on
1132
the graph reachable from a particular revision, and ignore any other
1133
revisions that might be present. There is no direct replacement
1136
if 'evil' in debug.debug_flags:
1137
mutter_callsite(2, "all_revision_ids is linear with history.")
1138
return self._all_revision_ids()
1140
def _all_revision_ids(self):
1141
"""Returns a list of all the revision ids in the repository.
1143
These are in as much topological order as the underlying store can
1146
raise NotImplementedError(self._all_revision_ids)
1148
def break_lock(self):
1149
"""Break a lock if one is present from another instance.
1151
Uses the ui factory to ask for confirmation if the lock may be from
1154
self.control_files.break_lock()
1157
def _eliminate_revisions_not_present(self, revision_ids):
1158
"""Check every revision id in revision_ids to see if we have it.
1160
Returns a set of the present revisions.
1163
graph = self.get_graph()
1164
parent_map = graph.get_parent_map(revision_ids)
1165
# The old API returned a list, should this actually be a set?
1166
return parent_map.keys()
1168
def _check_inventories(self, checker):
1169
"""Check the inventories found from the revision scan.
1171
This is responsible for verifying the sha1 of inventories and
1172
creating a pending_keys set that covers data referenced by inventories.
1174
bar = ui.ui_factory.nested_progress_bar()
1176
self._do_check_inventories(checker, bar)
1180
def _do_check_inventories(self, checker, bar):
1181
"""Helper for _check_inventories."""
1183
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1184
kinds = ['chk_bytes', 'texts']
1185
count = len(checker.pending_keys)
1186
bar.update("inventories", 0, 2)
1187
current_keys = checker.pending_keys
1188
checker.pending_keys = {}
1189
# Accumulate current checks.
1190
for key in current_keys:
1191
if key[0] != 'inventories' and key[0] not in kinds:
1192
checker._report_items.append('unknown key type %r' % (key,))
1193
keys[key[0]].add(key[1:])
1194
if keys['inventories']:
1195
# NB: output order *should* be roughly sorted - topo or
1196
# inverse topo depending on repository - either way decent
1197
# to just delta against. However, pre-CHK formats didn't
1198
# try to optimise inventory layout on disk. As such the
1199
# pre-CHK code path does not use inventory deltas.
1201
for record in self.inventories.check(keys=keys['inventories']):
1202
if record.storage_kind == 'absent':
1203
checker._report_items.append(
1204
'Missing inventory {%s}' % (record.key,))
1206
last_object = self._check_record('inventories', record,
1207
checker, last_object,
1208
current_keys[('inventories',) + record.key])
1209
del keys['inventories']
1212
bar.update("texts", 1)
1213
while (checker.pending_keys or keys['chk_bytes']
1215
# Something to check.
1216
current_keys = checker.pending_keys
1217
checker.pending_keys = {}
1218
# Accumulate current checks.
1219
for key in current_keys:
1220
if key[0] not in kinds:
1221
checker._report_items.append('unknown key type %r' % (key,))
1222
keys[key[0]].add(key[1:])
1223
# Check the outermost kind only - inventories || chk_bytes || texts
1227
for record in getattr(self, kind).check(keys=keys[kind]):
1228
if record.storage_kind == 'absent':
1229
checker._report_items.append(
1230
'Missing %s {%s}' % (kind, record.key,))
1232
last_object = self._check_record(kind, record,
1233
checker, last_object, current_keys[(kind,) + record.key])
1237
def _check_record(self, kind, record, checker, last_object, item_data):
1238
"""Check a single text from this repository."""
1239
if kind == 'inventories':
1240
rev_id = record.key[0]
1241
inv = self.deserialise_inventory(rev_id,
1242
record.get_bytes_as('fulltext'))
1243
if last_object is not None:
1244
delta = inv._make_delta(last_object)
1245
for old_path, path, file_id, ie in delta:
1248
ie.check(checker, rev_id, inv)
1250
for path, ie in inv.iter_entries():
1251
ie.check(checker, rev_id, inv)
1252
if self._format.fast_deltas:
1254
elif kind == 'chk_bytes':
1255
# No code written to check chk_bytes for this repo format.
1256
checker._report_items.append(
1257
'unsupported key type chk_bytes for %s' % (record.key,))
1258
elif kind == 'texts':
1259
self._check_text(record, checker, item_data)
1261
checker._report_items.append(
1262
'unknown key type %s for %s' % (kind, record.key))
1264
def _check_text(self, record, checker, item_data):
1265
"""Check a single text."""
1266
# Check it is extractable.
1267
# TODO: check length.
1268
if record.storage_kind == 'chunked':
1269
chunks = record.get_bytes_as(record.storage_kind)
1270
sha1 = osutils.sha_strings(chunks)
1271
length = sum(map(len, chunks))
1273
content = record.get_bytes_as('fulltext')
1274
sha1 = osutils.sha_string(content)
1275
length = len(content)
1276
if item_data and sha1 != item_data[1]:
1277
checker._report_items.append(
1278
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1279
(record.key, sha1, item_data[1], item_data[2]))
1282
def create(a_bzrdir):
1283
"""Construct the current default format repository in a_bzrdir."""
1284
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1286
def __init__(self, _format, a_bzrdir, control_files):
1287
"""instantiate a Repository.
1289
:param _format: The format of the repository on disk.
1290
:param a_bzrdir: The BzrDir of the repository.
1292
In the future we will have a single api for all stores for
1293
getting file texts, inventories and revisions, then
1294
this construct will accept instances of those things.
1296
super(Repository, self).__init__()
1297
self._format = _format
1298
# the following are part of the public API for Repository:
1299
self.bzrdir = a_bzrdir
1300
self.control_files = control_files
1301
self._transport = control_files._transport
1302
self.base = self._transport.base
1304
self._reconcile_does_inventory_gc = True
1305
self._reconcile_fixes_text_parents = False
1306
self._reconcile_backsup_inventory = True
1307
# not right yet - should be more semantically clear ?
1309
# TODO: make sure to construct the right store classes, etc, depending
1310
# on whether escaping is required.
1311
self._warn_if_deprecated()
1312
self._write_group = None
1313
# Additional places to query for data.
1314
self._fallback_repositories = []
1315
# An InventoryEntry cache, used during deserialization
1316
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1319
if self._fallback_repositories:
1320
return '%s(%r, fallback_repositories=%r)' % (
1321
self.__class__.__name__,
1323
self._fallback_repositories)
1325
return '%s(%r)' % (self.__class__.__name__,
1328
def _has_same_fallbacks(self, other_repo):
1329
"""Returns true if the repositories have the same fallbacks."""
1330
my_fb = self._fallback_repositories
1331
other_fb = other_repo._fallback_repositories
1332
if len(my_fb) != len(other_fb):
1334
for f, g in zip(my_fb, other_fb):
1335
if not f.has_same_location(g):
1339
def has_same_location(self, other):
1340
"""Returns a boolean indicating if this repository is at the same
1341
location as another repository.
1343
This might return False even when two repository objects are accessing
1344
the same physical repository via different URLs.
1346
if self.__class__ is not other.__class__:
1348
return (self._transport.base == other._transport.base)
1350
def is_in_write_group(self):
1351
"""Return True if there is an open write group.
1353
:seealso: start_write_group.
1355
return self._write_group is not None
1357
def is_locked(self):
1358
return self.control_files.is_locked()
1360
def is_write_locked(self):
1361
"""Return True if this object is write locked."""
1362
return self.is_locked() and self.control_files._lock_mode == 'w'
1364
def lock_write(self, token=None):
1365
"""Lock this repository for writing.
1367
This causes caching within the repository obejct to start accumlating
1368
data during reads, and allows a 'write_group' to be obtained. Write
1369
groups must be used for actual data insertion.
1371
:param token: if this is already locked, then lock_write will fail
1372
unless the token matches the existing lock.
1373
:returns: a token if this instance supports tokens, otherwise None.
1374
:raises TokenLockingNotSupported: when a token is given but this
1375
instance doesn't support using token locks.
1376
:raises MismatchedToken: if the specified token doesn't match the token
1377
of the existing lock.
1378
:seealso: start_write_group.
1380
A token should be passed in if you know that you have locked the object
1381
some other way, and need to synchronise this object's state with that
1384
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1386
locked = self.is_locked()
1387
result = self.control_files.lock_write(token=token)
1389
self._note_lock('w')
1390
for repo in self._fallback_repositories:
1391
# Writes don't affect fallback repos
1393
self._refresh_data()
1396
def lock_read(self):
1397
locked = self.is_locked()
1398
self.control_files.lock_read()
1400
self._note_lock('r')
1401
for repo in self._fallback_repositories:
1403
self._refresh_data()
1405
def get_physical_lock_status(self):
1406
return self.control_files.get_physical_lock_status()
1408
def leave_lock_in_place(self):
1409
"""Tell this repository not to release the physical lock when this
1412
If lock_write doesn't return a token, then this method is not supported.
1414
self.control_files.leave_in_place()
1416
def dont_leave_lock_in_place(self):
1417
"""Tell this repository to release the physical lock when this
1418
object is unlocked, even if it didn't originally acquire it.
1420
If lock_write doesn't return a token, then this method is not supported.
1422
self.control_files.dont_leave_in_place()
1425
def gather_stats(self, revid=None, committers=None):
1426
"""Gather statistics from a revision id.
1428
:param revid: The revision id to gather statistics from, if None, then
1429
no revision specific statistics are gathered.
1430
:param committers: Optional parameter controlling whether to grab
1431
a count of committers from the revision specific statistics.
1432
:return: A dictionary of statistics. Currently this contains:
1433
committers: The number of committers if requested.
1434
firstrev: A tuple with timestamp, timezone for the penultimate left
1435
most ancestor of revid, if revid is not the NULL_REVISION.
1436
latestrev: A tuple with timestamp, timezone for revid, if revid is
1437
not the NULL_REVISION.
1438
revisions: The total revision count in the repository.
1439
size: An estimate disk size of the repository in bytes.
1442
if revid and committers:
1443
result['committers'] = 0
1444
if revid and revid != _mod_revision.NULL_REVISION:
1446
all_committers = set()
1447
revisions = self.get_ancestry(revid)
1448
# pop the leading None
1450
first_revision = None
1452
# ignore the revisions in the middle - just grab first and last
1453
revisions = revisions[0], revisions[-1]
1454
for revision in self.get_revisions(revisions):
1455
if not first_revision:
1456
first_revision = revision
1458
all_committers.add(revision.committer)
1459
last_revision = revision
1461
result['committers'] = len(all_committers)
1462
result['firstrev'] = (first_revision.timestamp,
1463
first_revision.timezone)
1464
result['latestrev'] = (last_revision.timestamp,
1465
last_revision.timezone)
1467
# now gather global repository information
1468
# XXX: This is available for many repos regardless of listability.
1469
if self.bzrdir.root_transport.listable():
1470
# XXX: do we want to __define len__() ?
1471
# Maybe the versionedfiles object should provide a different
1472
# method to get the number of keys.
1473
result['revisions'] = len(self.revisions.keys())
1474
# result['size'] = t
1477
def find_branches(self, using=False):
1478
"""Find branches underneath this repository.
1480
This will include branches inside other branches.
1482
:param using: If True, list only branches using this repository.
1484
if using and not self.is_shared():
1486
return [self.bzrdir.open_branch()]
1487
except errors.NotBranchError:
1489
class Evaluator(object):
1492
self.first_call = True
1494
def __call__(self, bzrdir):
1495
# On the first call, the parameter is always the bzrdir
1496
# containing the current repo.
1497
if not self.first_call:
1499
repository = bzrdir.open_repository()
1500
except errors.NoRepositoryPresent:
1503
return False, (None, repository)
1504
self.first_call = False
1506
value = (bzrdir.open_branch(), None)
1507
except errors.NotBranchError:
1508
value = (None, None)
1512
for branch, repository in bzrdir.BzrDir.find_bzrdirs(
1513
self.bzrdir.root_transport, evaluate=Evaluator()):
1514
if branch is not None:
1515
branches.append(branch)
1516
if not using and repository is not None:
1517
branches.extend(repository.find_branches())
1521
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
1522
"""Return the revision ids that other has that this does not.
1524
These are returned in topological order.
1526
revision_id: only return revision ids included by revision_id.
1528
return InterRepository.get(other, self).search_missing_revision_ids(
1529
revision_id, find_ghosts)
1533
"""Open the repository rooted at base.
1535
For instance, if the repository is at URL/.bzr/repository,
1536
Repository.open(URL) -> a Repository instance.
1538
control = bzrdir.BzrDir.open(base)
1539
return control.open_repository()
1541
def copy_content_into(self, destination, revision_id=None):
1542
"""Make a complete copy of the content in self into destination.
1544
This is a destructive operation! Do not use it on existing
1547
return InterRepository.get(self, destination).copy_content(revision_id)
1549
def commit_write_group(self):
1550
"""Commit the contents accrued within the current write group.
1552
:seealso: start_write_group.
1554
:return: it may return an opaque hint that can be passed to 'pack'.
1556
if self._write_group is not self.get_transaction():
1557
# has an unlock or relock occured ?
1558
raise errors.BzrError('mismatched lock context %r and '
1560
(self.get_transaction(), self._write_group))
1561
result = self._commit_write_group()
1562
self._write_group = None
1565
def _commit_write_group(self):
1566
"""Template method for per-repository write group cleanup.
1568
This is called before the write group is considered to be
1569
finished and should ensure that all data handed to the repository
1570
for writing during the write group is safely committed (to the
1571
extent possible considering file system caching etc).
1574
def suspend_write_group(self):
1575
raise errors.UnsuspendableWriteGroup(self)
1577
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1578
"""Return the keys of missing inventory parents for revisions added in
1581
A revision is not complete if the inventory delta for that revision
1582
cannot be calculated. Therefore if the parent inventories of a
1583
revision are not present, the revision is incomplete, and e.g. cannot
1584
be streamed by a smart server. This method finds missing inventory
1585
parents for revisions added in this write group.
1587
if not self._format.supports_external_lookups:
1588
# This is only an issue for stacked repositories
1590
if not self.is_in_write_group():
1591
raise AssertionError('not in a write group')
1593
# XXX: We assume that every added revision already has its
1594
# corresponding inventory, so we only check for parent inventories that
1595
# might be missing, rather than all inventories.
1596
parents = set(self.revisions._index.get_missing_parents())
1597
parents.discard(_mod_revision.NULL_REVISION)
1598
unstacked_inventories = self.inventories._index
1599
present_inventories = unstacked_inventories.get_parent_map(
1600
key[-1:] for key in parents)
1601
parents.difference_update(present_inventories)
1602
if len(parents) == 0:
1603
# No missing parent inventories.
1605
if not check_for_missing_texts:
1606
return set(('inventories', rev_id) for (rev_id,) in parents)
1607
# Ok, now we have a list of missing inventories. But these only matter
1608
# if the inventories that reference them are missing some texts they
1609
# appear to introduce.
1610
# XXX: Texts referenced by all added inventories need to be present,
1611
# but at the moment we're only checking for texts referenced by
1612
# inventories at the graph's edge.
1613
key_deps = self.revisions._index._key_dependencies
1614
key_deps.satisfy_refs_for_keys(present_inventories)
1615
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1616
file_ids = self.fileids_altered_by_revision_ids(referrers)
1617
missing_texts = set()
1618
for file_id, version_ids in file_ids.iteritems():
1619
missing_texts.update(
1620
(file_id, version_id) for version_id in version_ids)
1621
present_texts = self.texts.get_parent_map(missing_texts)
1622
missing_texts.difference_update(present_texts)
1623
if not missing_texts:
1624
# No texts are missing, so all revisions and their deltas are
1627
# Alternatively the text versions could be returned as the missing
1628
# keys, but this is likely to be less data.
1629
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1632
def refresh_data(self):
1633
"""Re-read any data needed to to synchronise with disk.
1635
This method is intended to be called after another repository instance
1636
(such as one used by a smart server) has inserted data into the
1637
repository. It may not be called during a write group, but may be
1638
called at any other time.
1640
if self.is_in_write_group():
1641
raise errors.InternalBzrError(
1642
"May not refresh_data while in a write group.")
1643
self._refresh_data()
1645
def resume_write_group(self, tokens):
1646
if not self.is_write_locked():
1647
raise errors.NotWriteLocked(self)
1648
if self._write_group:
1649
raise errors.BzrError('already in a write group')
1650
self._resume_write_group(tokens)
1651
# so we can detect unlock/relock - the write group is now entered.
1652
self._write_group = self.get_transaction()
1654
def _resume_write_group(self, tokens):
1655
raise errors.UnsuspendableWriteGroup(self)
1657
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1659
"""Fetch the content required to construct revision_id from source.
1661
If revision_id is None and fetch_spec is None, then all content is
1664
fetch() may not be used when the repository is in a write group -
1665
either finish the current write group before using fetch, or use
1666
fetch before starting the write group.
1668
:param find_ghosts: Find and copy revisions in the source that are
1669
ghosts in the target (and not reachable directly by walking out to
1670
the first-present revision in target from revision_id).
1671
:param revision_id: If specified, all the content needed for this
1672
revision ID will be copied to the target. Fetch will determine for
1673
itself which content needs to be copied.
1674
:param fetch_spec: If specified, a SearchResult or
1675
PendingAncestryResult that describes which revisions to copy. This
1676
allows copying multiple heads at once. Mutually exclusive with
1679
if fetch_spec is not None and revision_id is not None:
1680
raise AssertionError(
1681
"fetch_spec and revision_id are mutually exclusive.")
1682
if self.is_in_write_group():
1683
raise errors.InternalBzrError(
1684
"May not fetch while in a write group.")
1685
# fast path same-url fetch operations
1686
# TODO: lift out to somewhere common with RemoteRepository
1687
# <https://bugs.edge.launchpad.net/bzr/+bug/401646>
1688
if (self.has_same_location(source)
1689
and fetch_spec is None
1690
and self._has_same_fallbacks(source)):
1691
# check that last_revision is in 'from' and then return a
1693
if (revision_id is not None and
1694
not _mod_revision.is_null(revision_id)):
1695
self.get_revision(revision_id)
1697
# if there is no specific appropriate InterRepository, this will get
1698
# the InterRepository base class, which raises an
1699
# IncompatibleRepositories when asked to fetch.
1700
inter = InterRepository.get(source, self)
1701
return inter.fetch(revision_id=revision_id, pb=pb,
1702
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1704
def create_bundle(self, target, base, fileobj, format=None):
1705
return serializer.write_bundle(self, target, base, fileobj, format)
1707
def get_commit_builder(self, branch, parents, config, timestamp=None,
1708
timezone=None, committer=None, revprops=None,
1710
"""Obtain a CommitBuilder for this repository.
1712
:param branch: Branch to commit to.
1713
:param parents: Revision ids of the parents of the new revision.
1714
:param config: Configuration to use.
1715
:param timestamp: Optional timestamp recorded for commit.
1716
:param timezone: Optional timezone for timestamp.
1717
:param committer: Optional committer to set for commit.
1718
:param revprops: Optional dictionary of revision properties.
1719
:param revision_id: Optional revision id.
1721
if self._fallback_repositories:
1722
raise errors.BzrError("Cannot commit from a lightweight checkout "
1723
"to a stacked branch. See "
1724
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1725
result = self._commit_builder_class(self, parents, config,
1726
timestamp, timezone, committer, revprops, revision_id)
1727
self.start_write_group()
1730
@only_raises(errors.LockNotHeld, errors.LockBroken)
1732
if (self.control_files._lock_count == 1 and
1733
self.control_files._lock_mode == 'w'):
1734
if self._write_group is not None:
1735
self.abort_write_group()
1736
self.control_files.unlock()
1737
raise errors.BzrError(
1738
'Must end write groups before releasing write locks.')
1739
self.control_files.unlock()
1740
if self.control_files._lock_count == 0:
1741
self._inventory_entry_cache.clear()
1742
for repo in self._fallback_repositories:
1746
def clone(self, a_bzrdir, revision_id=None):
1747
"""Clone this repository into a_bzrdir using the current format.
1749
Currently no check is made that the format of this repository and
1750
the bzrdir format are compatible. FIXME RBC 20060201.
1752
:return: The newly created destination repository.
1754
# TODO: deprecate after 0.16; cloning this with all its settings is
1755
# probably not very useful -- mbp 20070423
1756
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1757
self.copy_content_into(dest_repo, revision_id)
1760
def start_write_group(self):
1761
"""Start a write group in the repository.
1763
Write groups are used by repositories which do not have a 1:1 mapping
1764
between file ids and backend store to manage the insertion of data from
1765
both fetch and commit operations.
1767
A write lock is required around the start_write_group/commit_write_group
1768
for the support of lock-requiring repository formats.
1770
One can only insert data into a repository inside a write group.
1774
if not self.is_write_locked():
1775
raise errors.NotWriteLocked(self)
1776
if self._write_group:
1777
raise errors.BzrError('already in a write group')
1778
self._start_write_group()
1779
# so we can detect unlock/relock - the write group is now entered.
1780
self._write_group = self.get_transaction()
1782
def _start_write_group(self):
1783
"""Template method for per-repository write group startup.
1785
This is called before the write group is considered to be
1790
def sprout(self, to_bzrdir, revision_id=None):
1791
"""Create a descendent repository for new development.
1793
Unlike clone, this does not copy the settings of the repository.
1795
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1796
dest_repo.fetch(self, revision_id=revision_id)
1799
def _create_sprouting_repo(self, a_bzrdir, shared):
1800
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1801
# use target default format.
1802
dest_repo = a_bzrdir.create_repository()
1804
# Most control formats need the repository to be specifically
1805
# created, but on some old all-in-one formats it's not needed
1807
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1808
except errors.UninitializableFormat:
1809
dest_repo = a_bzrdir.open_repository()
1812
def _get_sink(self):
1813
"""Return a sink for streaming into this repository."""
1814
return StreamSink(self)
1816
def _get_source(self, to_format):
1817
"""Return a source for streaming from this repository."""
1818
return StreamSource(self, to_format)
1821
def has_revision(self, revision_id):
1822
"""True if this repository has a copy of the revision."""
1823
return revision_id in self.has_revisions((revision_id,))
1826
def has_revisions(self, revision_ids):
1827
"""Probe to find out the presence of multiple revisions.
1829
:param revision_ids: An iterable of revision_ids.
1830
:return: A set of the revision_ids that were present.
1832
parent_map = self.revisions.get_parent_map(
1833
[(rev_id,) for rev_id in revision_ids])
1835
if _mod_revision.NULL_REVISION in revision_ids:
1836
result.add(_mod_revision.NULL_REVISION)
1837
result.update([key[0] for key in parent_map])
1841
def get_revision(self, revision_id):
1842
"""Return the Revision object for a named revision."""
1843
return self.get_revisions([revision_id])[0]
1846
def get_revision_reconcile(self, revision_id):
1847
"""'reconcile' helper routine that allows access to a revision always.
1849
This variant of get_revision does not cross check the weave graph
1850
against the revision one as get_revision does: but it should only
1851
be used by reconcile, or reconcile-alike commands that are correcting
1852
or testing the revision graph.
1854
return self._get_revisions([revision_id])[0]
1857
def get_revisions(self, revision_ids):
1858
"""Get many revisions at once.
1860
Repositories that need to check data on every revision read should
1861
subclass this method.
1863
return self._get_revisions(revision_ids)
1866
def _get_revisions(self, revision_ids):
1867
"""Core work logic to get many revisions without sanity checks."""
1869
for revid, rev in self._iter_revisions(revision_ids):
1871
raise errors.NoSuchRevision(self, revid)
1873
return [revs[revid] for revid in revision_ids]
1875
def _iter_revisions(self, revision_ids):
1876
"""Iterate over revision objects.
1878
:param revision_ids: An iterable of revisions to examine. None may be
1879
passed to request all revisions known to the repository. Note that
1880
not all repositories can find unreferenced revisions; for those
1881
repositories only referenced ones will be returned.
1882
:return: An iterator of (revid, revision) tuples. Absent revisions (
1883
those asked for but not available) are returned as (revid, None).
1885
if revision_ids is None:
1886
revision_ids = self.all_revision_ids()
1888
for rev_id in revision_ids:
1889
if not rev_id or not isinstance(rev_id, basestring):
1890
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1891
keys = [(key,) for key in revision_ids]
1892
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1893
for record in stream:
1894
revid = record.key[0]
1895
if record.storage_kind == 'absent':
1898
text = record.get_bytes_as('fulltext')
1899
rev = self._serializer.read_revision_from_string(text)
1903
def get_revision_xml(self, revision_id):
1904
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1905
# would have already do it.
1906
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1907
# TODO: this can't just be replaced by:
1908
# return self._serializer.write_revision_to_string(
1909
# self.get_revision(revision_id))
1910
# as cStringIO preservers the encoding unlike write_revision_to_string
1911
# or some other call down the path.
1912
rev = self.get_revision(revision_id)
1913
rev_tmp = cStringIO.StringIO()
1914
# the current serializer..
1915
self._serializer.write_revision(rev, rev_tmp)
1917
return rev_tmp.getvalue()
1919
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1920
"""Produce a generator of revision deltas.
1922
Note that the input is a sequence of REVISIONS, not revision_ids.
1923
Trees will be held in memory until the generator exits.
1924
Each delta is relative to the revision's lefthand predecessor.
1926
:param specific_fileids: if not None, the result is filtered
1927
so that only those file-ids, their parents and their
1928
children are included.
1930
# Get the revision-ids of interest
1931
required_trees = set()
1932
for revision in revisions:
1933
required_trees.add(revision.revision_id)
1934
required_trees.update(revision.parent_ids[:1])
1936
# Get the matching filtered trees. Note that it's more
1937
# efficient to pass filtered trees to changes_from() rather
1938
# than doing the filtering afterwards. changes_from() could
1939
# arguably do the filtering itself but it's path-based, not
1940
# file-id based, so filtering before or afterwards is
1942
if specific_fileids is None:
1943
trees = dict((t.get_revision_id(), t) for
1944
t in self.revision_trees(required_trees))
1946
trees = dict((t.get_revision_id(), t) for
1947
t in self._filtered_revision_trees(required_trees,
1950
# Calculate the deltas
1951
for revision in revisions:
1952
if not revision.parent_ids:
1953
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1955
old_tree = trees[revision.parent_ids[0]]
1956
yield trees[revision.revision_id].changes_from(old_tree)
1959
def get_revision_delta(self, revision_id, specific_fileids=None):
1960
"""Return the delta for one revision.
1962
The delta is relative to the left-hand predecessor of the
1965
:param specific_fileids: if not None, the result is filtered
1966
so that only those file-ids, their parents and their
1967
children are included.
1969
r = self.get_revision(revision_id)
1970
return list(self.get_deltas_for_revisions([r],
1971
specific_fileids=specific_fileids))[0]
1974
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1975
signature = gpg_strategy.sign(plaintext)
1976
self.add_signature_text(revision_id, signature)
1979
def add_signature_text(self, revision_id, signature):
1980
self.signatures.add_lines((revision_id,), (),
1981
osutils.split_lines(signature))
1983
def find_text_key_references(self):
1984
"""Find the text key references within the repository.
1986
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1987
to whether they were referred to by the inventory of the
1988
revision_id that they contain. The inventory texts from all present
1989
revision ids are assessed to generate this report.
1991
revision_keys = self.revisions.keys()
1992
w = self.inventories
1993
pb = ui.ui_factory.nested_progress_bar()
1995
return self._find_text_key_references_from_xml_inventory_lines(
1996
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
2000
def _find_text_key_references_from_xml_inventory_lines(self,
2002
"""Core routine for extracting references to texts from inventories.
2004
This performs the translation of xml lines to revision ids.
2006
:param line_iterator: An iterator of lines, origin_version_id
2007
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2008
to whether they were referred to by the inventory of the
2009
revision_id that they contain. Note that if that revision_id was
2010
not part of the line_iterator's output then False will be given -
2011
even though it may actually refer to that key.
2013
if not self._serializer.support_altered_by_hack:
2014
raise AssertionError(
2015
"_find_text_key_references_from_xml_inventory_lines only "
2016
"supported for branches which store inventory as unnested xml"
2017
", not on %r" % self)
2020
# this code needs to read every new line in every inventory for the
2021
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2022
# not present in one of those inventories is unnecessary but not
2023
# harmful because we are filtering by the revision id marker in the
2024
# inventory lines : we only select file ids altered in one of those
2025
# revisions. We don't need to see all lines in the inventory because
2026
# only those added in an inventory in rev X can contain a revision=X
2028
unescape_revid_cache = {}
2029
unescape_fileid_cache = {}
2031
# jam 20061218 In a big fetch, this handles hundreds of thousands
2032
# of lines, so it has had a lot of inlining and optimizing done.
2033
# Sorry that it is a little bit messy.
2034
# Move several functions to be local variables, since this is a long
2036
search = self._file_ids_altered_regex.search
2037
unescape = _unescape_xml
2038
setdefault = result.setdefault
2039
for line, line_key in line_iterator:
2040
match = search(line)
2043
# One call to match.group() returning multiple items is quite a
2044
# bit faster than 2 calls to match.group() each returning 1
2045
file_id, revision_id = match.group('file_id', 'revision_id')
2047
# Inlining the cache lookups helps a lot when you make 170,000
2048
# lines and 350k ids, versus 8.4 unique ids.
2049
# Using a cache helps in 2 ways:
2050
# 1) Avoids unnecessary decoding calls
2051
# 2) Re-uses cached strings, which helps in future set and
2053
# (2) is enough that removing encoding entirely along with
2054
# the cache (so we are using plain strings) results in no
2055
# performance improvement.
2057
revision_id = unescape_revid_cache[revision_id]
2059
unescaped = unescape(revision_id)
2060
unescape_revid_cache[revision_id] = unescaped
2061
revision_id = unescaped
2063
# Note that unconditionally unescaping means that we deserialise
2064
# every fileid, which for general 'pull' is not great, but we don't
2065
# really want to have some many fulltexts that this matters anyway.
2068
file_id = unescape_fileid_cache[file_id]
2070
unescaped = unescape(file_id)
2071
unescape_fileid_cache[file_id] = unescaped
2074
key = (file_id, revision_id)
2075
setdefault(key, False)
2076
if revision_id == line_key[-1]:
2080
def _inventory_xml_lines_for_keys(self, keys):
2081
"""Get a line iterator of the sort needed for findind references.
2083
Not relevant for non-xml inventory repositories.
2085
Ghosts in revision_keys are ignored.
2087
:param revision_keys: The revision keys for the inventories to inspect.
2088
:return: An iterator over (inventory line, revid) for the fulltexts of
2089
all of the xml inventories specified by revision_keys.
2091
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2092
for record in stream:
2093
if record.storage_kind != 'absent':
2094
chunks = record.get_bytes_as('chunked')
2095
revid = record.key[-1]
2096
lines = osutils.chunks_to_lines(chunks)
2100
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2102
"""Helper routine for fileids_altered_by_revision_ids.
2104
This performs the translation of xml lines to revision ids.
2106
:param line_iterator: An iterator of lines, origin_version_id
2107
:param revision_keys: The revision ids to filter for. This should be a
2108
set or other type which supports efficient __contains__ lookups, as
2109
the revision key from each parsed line will be looked up in the
2110
revision_keys filter.
2111
:return: a dictionary mapping altered file-ids to an iterable of
2112
revision_ids. Each altered file-ids has the exact revision_ids that
2113
altered it listed explicitly.
2115
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2116
line_iterator).iterkeys())
2117
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2118
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2119
self._inventory_xml_lines_for_keys(parent_keys)))
2120
new_keys = seen - parent_seen
2122
setdefault = result.setdefault
2123
for key in new_keys:
2124
setdefault(key[0], set()).add(key[-1])
2127
def _find_parent_ids_of_revisions(self, revision_ids):
2128
"""Find all parent ids that are mentioned in the revision graph.
2130
:return: set of revisions that are parents of revision_ids which are
2131
not part of revision_ids themselves
2133
parent_map = self.get_parent_map(revision_ids)
2135
map(parent_ids.update, parent_map.itervalues())
2136
parent_ids.difference_update(revision_ids)
2137
parent_ids.discard(_mod_revision.NULL_REVISION)
2140
def _find_parent_keys_of_revisions(self, revision_keys):
2141
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2143
:param revision_keys: An iterable of revision_keys.
2144
:return: The parents of all revision_keys that are not already in
2147
parent_map = self.revisions.get_parent_map(revision_keys)
2149
map(parent_keys.update, parent_map.itervalues())
2150
parent_keys.difference_update(revision_keys)
2151
parent_keys.discard(_mod_revision.NULL_REVISION)
2154
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2155
"""Find the file ids and versions affected by revisions.
2157
:param revisions: an iterable containing revision ids.
2158
:param _inv_weave: The inventory weave from this repository or None.
2159
If None, the inventory weave will be opened automatically.
2160
:return: a dictionary mapping altered file-ids to an iterable of
2161
revision_ids. Each altered file-ids has the exact revision_ids that
2162
altered it listed explicitly.
2164
selected_keys = set((revid,) for revid in revision_ids)
2165
w = _inv_weave or self.inventories
2166
pb = ui.ui_factory.nested_progress_bar()
2168
return self._find_file_ids_from_xml_inventory_lines(
2169
w.iter_lines_added_or_present_in_keys(
2170
selected_keys, pb=pb),
2175
def iter_files_bytes(self, desired_files):
2176
"""Iterate through file versions.
2178
Files will not necessarily be returned in the order they occur in
2179
desired_files. No specific order is guaranteed.
2181
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2182
value supplied by the caller as part of desired_files. It should
2183
uniquely identify the file version in the caller's context. (Examples:
2184
an index number or a TreeTransform trans_id.)
2186
bytes_iterator is an iterable of bytestrings for the file. The
2187
kind of iterable and length of the bytestrings are unspecified, but for
2188
this implementation, it is a list of bytes produced by
2189
VersionedFile.get_record_stream().
2191
:param desired_files: a list of (file_id, revision_id, identifier)
2195
for file_id, revision_id, callable_data in desired_files:
2196
text_keys[(file_id, revision_id)] = callable_data
2197
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2198
if record.storage_kind == 'absent':
2199
raise errors.RevisionNotPresent(record.key, self)
2200
yield text_keys[record.key], record.get_bytes_as('chunked')
2202
def _generate_text_key_index(self, text_key_references=None,
2204
"""Generate a new text key index for the repository.
2206
This is an expensive function that will take considerable time to run.
2208
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2209
list of parents, also text keys. When a given key has no parents,
2210
the parents list will be [NULL_REVISION].
2212
# All revisions, to find inventory parents.
2213
if ancestors is None:
2214
graph = self.get_graph()
2215
ancestors = graph.get_parent_map(self.all_revision_ids())
2216
if text_key_references is None:
2217
text_key_references = self.find_text_key_references()
2218
pb = ui.ui_factory.nested_progress_bar()
2220
return self._do_generate_text_key_index(ancestors,
2221
text_key_references, pb)
2225
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2226
"""Helper for _generate_text_key_index to avoid deep nesting."""
2227
revision_order = tsort.topo_sort(ancestors)
2228
invalid_keys = set()
2230
for revision_id in revision_order:
2231
revision_keys[revision_id] = set()
2232
text_count = len(text_key_references)
2233
# a cache of the text keys to allow reuse; costs a dict of all the
2234
# keys, but saves a 2-tuple for every child of a given key.
2236
for text_key, valid in text_key_references.iteritems():
2238
invalid_keys.add(text_key)
2240
revision_keys[text_key[1]].add(text_key)
2241
text_key_cache[text_key] = text_key
2242
del text_key_references
2244
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2245
NULL_REVISION = _mod_revision.NULL_REVISION
2246
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2247
# too small for large or very branchy trees. However, for 55K path
2248
# trees, it would be easy to use too much memory trivially. Ideally we
2249
# could gauge this by looking at available real memory etc, but this is
2250
# always a tricky proposition.
2251
inventory_cache = lru_cache.LRUCache(10)
2252
batch_size = 10 # should be ~150MB on a 55K path tree
2253
batch_count = len(revision_order) / batch_size + 1
2255
pb.update("Calculating text parents", processed_texts, text_count)
2256
for offset in xrange(batch_count):
2257
to_query = revision_order[offset * batch_size:(offset + 1) *
2261
for revision_id in to_query:
2262
parent_ids = ancestors[revision_id]
2263
for text_key in revision_keys[revision_id]:
2264
pb.update("Calculating text parents", processed_texts)
2265
processed_texts += 1
2266
candidate_parents = []
2267
for parent_id in parent_ids:
2268
parent_text_key = (text_key[0], parent_id)
2270
check_parent = parent_text_key not in \
2271
revision_keys[parent_id]
2273
# the parent parent_id is a ghost:
2274
check_parent = False
2275
# truncate the derived graph against this ghost.
2276
parent_text_key = None
2278
# look at the parent commit details inventories to
2279
# determine possible candidates in the per file graph.
2282
inv = inventory_cache[parent_id]
2284
inv = self.revision_tree(parent_id).inventory
2285
inventory_cache[parent_id] = inv
2287
parent_entry = inv[text_key[0]]
2288
except (KeyError, errors.NoSuchId):
2290
if parent_entry is not None:
2292
text_key[0], parent_entry.revision)
2294
parent_text_key = None
2295
if parent_text_key is not None:
2296
candidate_parents.append(
2297
text_key_cache[parent_text_key])
2298
parent_heads = text_graph.heads(candidate_parents)
2299
new_parents = list(parent_heads)
2300
new_parents.sort(key=lambda x:candidate_parents.index(x))
2301
if new_parents == []:
2302
new_parents = [NULL_REVISION]
2303
text_index[text_key] = new_parents
2305
for text_key in invalid_keys:
2306
text_index[text_key] = [NULL_REVISION]
2309
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2310
"""Get an iterable listing the keys of all the data introduced by a set
2313
The keys will be ordered so that the corresponding items can be safely
2314
fetched and inserted in that order.
2316
:returns: An iterable producing tuples of (knit-kind, file-id,
2317
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2318
'revisions'. file-id is None unless knit-kind is 'file'.
2320
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2323
for result in self._find_non_file_keys_to_fetch(revision_ids):
2326
def _find_file_keys_to_fetch(self, revision_ids, pb):
2327
# XXX: it's a bit weird to control the inventory weave caching in this
2328
# generator. Ideally the caching would be done in fetch.py I think. Or
2329
# maybe this generator should explicitly have the contract that it
2330
# should not be iterated until the previously yielded item has been
2332
inv_w = self.inventories
2334
# file ids that changed
2335
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2337
num_file_ids = len(file_ids)
2338
for file_id, altered_versions in file_ids.iteritems():
2340
pb.update("Fetch texts", count, num_file_ids)
2342
yield ("file", file_id, altered_versions)
2344
def _find_non_file_keys_to_fetch(self, revision_ids):
2346
yield ("inventory", None, revision_ids)
2349
# XXX: Note ATM no callers actually pay attention to this return
2350
# instead they just use the list of revision ids and ignore
2351
# missing sigs. Consider removing this work entirely
2352
revisions_with_signatures = set(self.signatures.get_parent_map(
2353
[(r,) for r in revision_ids]))
2354
revisions_with_signatures = set(
2355
[r for (r,) in revisions_with_signatures])
2356
revisions_with_signatures.intersection_update(revision_ids)
2357
yield ("signatures", None, revisions_with_signatures)
2360
yield ("revisions", None, revision_ids)
2363
def get_inventory(self, revision_id):
2364
"""Get Inventory object by revision id."""
2365
return self.iter_inventories([revision_id]).next()
2367
def iter_inventories(self, revision_ids, ordering=None):
2368
"""Get many inventories by revision_ids.
2370
This will buffer some or all of the texts used in constructing the
2371
inventories in memory, but will only parse a single inventory at a
2374
:param revision_ids: The expected revision ids of the inventories.
2375
:param ordering: optional ordering, e.g. 'topological'. If not
2376
specified, the order of revision_ids will be preserved (by
2377
buffering if necessary).
2378
:return: An iterator of inventories.
2380
if ((None in revision_ids)
2381
or (_mod_revision.NULL_REVISION in revision_ids)):
2382
raise ValueError('cannot get null revision inventory')
2383
return self._iter_inventories(revision_ids, ordering)
2385
def _iter_inventories(self, revision_ids, ordering):
2386
"""single-document based inventory iteration."""
2387
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2388
for text, revision_id in inv_xmls:
2389
yield self.deserialise_inventory(revision_id, text)
2391
def _iter_inventory_xmls(self, revision_ids, ordering):
2392
if ordering is None:
2393
order_as_requested = True
2394
ordering = 'unordered'
2396
order_as_requested = False
2397
keys = [(revision_id,) for revision_id in revision_ids]
2400
if order_as_requested:
2401
key_iter = iter(keys)
2402
next_key = key_iter.next()
2403
stream = self.inventories.get_record_stream(keys, ordering, True)
2405
for record in stream:
2406
if record.storage_kind != 'absent':
2407
chunks = record.get_bytes_as('chunked')
2408
if order_as_requested:
2409
text_chunks[record.key] = chunks
2411
yield ''.join(chunks), record.key[-1]
2413
raise errors.NoSuchRevision(self, record.key)
2414
if order_as_requested:
2415
# Yield as many results as we can while preserving order.
2416
while next_key in text_chunks:
2417
chunks = text_chunks.pop(next_key)
2418
yield ''.join(chunks), next_key[-1]
2420
next_key = key_iter.next()
2421
except StopIteration:
2422
# We still want to fully consume the get_record_stream,
2423
# just in case it is not actually finished at this point
2427
def deserialise_inventory(self, revision_id, xml):
2428
"""Transform the xml into an inventory object.
2430
:param revision_id: The expected revision id of the inventory.
2431
:param xml: A serialised inventory.
2433
result = self._serializer.read_inventory_from_string(xml, revision_id,
2434
entry_cache=self._inventory_entry_cache)
2435
if result.revision_id != revision_id:
2436
raise AssertionError('revision id mismatch %s != %s' % (
2437
result.revision_id, revision_id))
2440
def serialise_inventory(self, inv):
2441
return self._serializer.write_inventory_to_string(inv)
2443
def _serialise_inventory_to_lines(self, inv):
2444
return self._serializer.write_inventory_to_lines(inv)
2446
def get_serializer_format(self):
2447
return self._serializer.format_num
2450
def get_inventory_xml(self, revision_id):
2451
"""Get inventory XML as a file object."""
2452
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2454
text, revision_id = texts.next()
2455
except StopIteration:
2456
raise errors.HistoryMissing(self, 'inventory', revision_id)
2460
def get_inventory_sha1(self, revision_id):
2461
"""Return the sha1 hash of the inventory entry
2463
return self.get_revision(revision_id).inventory_sha1
2465
def get_rev_id_for_revno(self, revno, known_pair):
2466
"""Return the revision id of a revno, given a later (revno, revid)
2467
pair in the same history.
2469
:return: if found (True, revid). If the available history ran out
2470
before reaching the revno, then this returns
2471
(False, (closest_revno, closest_revid)).
2473
known_revno, known_revid = known_pair
2474
partial_history = [known_revid]
2475
distance_from_known = known_revno - revno
2476
if distance_from_known < 0:
2478
'requested revno (%d) is later than given known revno (%d)'
2479
% (revno, known_revno))
2482
self, partial_history, stop_index=distance_from_known)
2483
except errors.RevisionNotPresent, err:
2484
if err.revision_id == known_revid:
2485
# The start revision (known_revid) wasn't found.
2487
# This is a stacked repository with no fallbacks, or a there's a
2488
# left-hand ghost. Either way, even though the revision named in
2489
# the error isn't in this repo, we know it's the next step in this
2490
# left-hand history.
2491
partial_history.append(err.revision_id)
2492
if len(partial_history) <= distance_from_known:
2493
# Didn't find enough history to get a revid for the revno.
2494
earliest_revno = known_revno - len(partial_history) + 1
2495
return (False, (earliest_revno, partial_history[-1]))
2496
if len(partial_history) - 1 > distance_from_known:
2497
raise AssertionError('_iter_for_revno returned too much history')
2498
return (True, partial_history[-1])
2500
def iter_reverse_revision_history(self, revision_id):
2501
"""Iterate backwards through revision ids in the lefthand history
2503
:param revision_id: The revision id to start with. All its lefthand
2504
ancestors will be traversed.
2506
graph = self.get_graph()
2507
next_id = revision_id
2509
if next_id in (None, _mod_revision.NULL_REVISION):
2512
parents = graph.get_parent_map([next_id])[next_id]
2514
raise errors.RevisionNotPresent(next_id, self)
2516
if len(parents) == 0:
2519
next_id = parents[0]
2522
def get_revision_inventory(self, revision_id):
2523
"""Return inventory of a past revision."""
2524
# TODO: Unify this with get_inventory()
2525
# bzr 0.0.6 and later imposes the constraint that the inventory_id
2526
# must be the same as its revision, so this is trivial.
2527
if revision_id is None:
2528
# This does not make sense: if there is no revision,
2529
# then it is the current tree inventory surely ?!
2530
# and thus get_root_id() is something that looks at the last
2531
# commit on the branch, and the get_root_id is an inventory check.
2532
raise NotImplementedError
2533
# return Inventory(self.get_root_id())
2535
return self.get_inventory(revision_id)
2537
def is_shared(self):
2538
"""Return True if this repository is flagged as a shared repository."""
2539
raise NotImplementedError(self.is_shared)
2542
def reconcile(self, other=None, thorough=False):
2543
"""Reconcile this repository."""
2544
from bzrlib.reconcile import RepoReconciler
2545
reconciler = RepoReconciler(self, thorough=thorough)
2546
reconciler.reconcile()
2549
def _refresh_data(self):
2550
"""Helper called from lock_* to ensure coherency with disk.
2552
The default implementation does nothing; it is however possible
2553
for repositories to maintain loaded indices across multiple locks
2554
by checking inside their implementation of this method to see
2555
whether their indices are still valid. This depends of course on
2556
the disk format being validatable in this manner. This method is
2557
also called by the refresh_data() public interface to cause a refresh
2558
to occur while in a write lock so that data inserted by a smart server
2559
push operation is visible on the client's instance of the physical
2564
def revision_tree(self, revision_id):
2565
"""Return Tree for a revision on this branch.
2567
`revision_id` may be NULL_REVISION for the empty tree revision.
2569
revision_id = _mod_revision.ensure_null(revision_id)
2570
# TODO: refactor this to use an existing revision object
2571
# so we don't need to read it in twice.
2572
if revision_id == _mod_revision.NULL_REVISION:
2573
return RevisionTree(self, Inventory(root_id=None),
2574
_mod_revision.NULL_REVISION)
2576
inv = self.get_revision_inventory(revision_id)
2577
return RevisionTree(self, inv, revision_id)
2579
def revision_trees(self, revision_ids):
2580
"""Return Trees for revisions in this repository.
2582
:param revision_ids: a sequence of revision-ids;
2583
a revision-id may not be None or 'null:'
2585
inventories = self.iter_inventories(revision_ids)
2586
for inv in inventories:
2587
yield RevisionTree(self, inv, inv.revision_id)
2589
def _filtered_revision_trees(self, revision_ids, file_ids):
2590
"""Return Tree for a revision on this branch with only some files.
2592
:param revision_ids: a sequence of revision-ids;
2593
a revision-id may not be None or 'null:'
2594
:param file_ids: if not None, the result is filtered
2595
so that only those file-ids, their parents and their
2596
children are included.
2598
inventories = self.iter_inventories(revision_ids)
2599
for inv in inventories:
2600
# Should we introduce a FilteredRevisionTree class rather
2601
# than pre-filter the inventory here?
2602
filtered_inv = inv.filter(file_ids)
2603
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2606
def get_ancestry(self, revision_id, topo_sorted=True):
2607
"""Return a list of revision-ids integrated by a revision.
2609
The first element of the list is always None, indicating the origin
2610
revision. This might change when we have history horizons, or
2611
perhaps we should have a new API.
2613
This is topologically sorted.
2615
if _mod_revision.is_null(revision_id):
2617
if not self.has_revision(revision_id):
2618
raise errors.NoSuchRevision(self, revision_id)
2619
graph = self.get_graph()
2621
search = graph._make_breadth_first_searcher([revision_id])
2624
found, ghosts = search.next_with_ghosts()
2625
except StopIteration:
2628
if _mod_revision.NULL_REVISION in keys:
2629
keys.remove(_mod_revision.NULL_REVISION)
2631
parent_map = graph.get_parent_map(keys)
2632
keys = tsort.topo_sort(parent_map)
2633
return [None] + list(keys)
2635
def pack(self, hint=None):
2636
"""Compress the data within the repository.
2638
This operation only makes sense for some repository types. For other
2639
types it should be a no-op that just returns.
2641
This stub method does not require a lock, but subclasses should use
2642
@needs_write_lock as this is a long running call its reasonable to
2643
implicitly lock for the user.
2645
:param hint: If not supplied, the whole repository is packed.
2646
If supplied, the repository may use the hint parameter as a
2647
hint for the parts of the repository to pack. A hint can be
2648
obtained from the result of commit_write_group(). Out of
2649
date hints are simply ignored, because concurrent operations
2650
can obsolete them rapidly.
2653
def get_transaction(self):
2654
return self.control_files.get_transaction()
2656
def get_parent_map(self, revision_ids):
2657
"""See graph.StackedParentsProvider.get_parent_map"""
2658
# revisions index works in keys; this just works in revisions
2659
# therefore wrap and unwrap
2662
for revision_id in revision_ids:
2663
if revision_id == _mod_revision.NULL_REVISION:
2664
result[revision_id] = ()
2665
elif revision_id is None:
2666
raise ValueError('get_parent_map(None) is not valid')
2668
query_keys.append((revision_id ,))
2669
for ((revision_id,), parent_keys) in \
2670
self.revisions.get_parent_map(query_keys).iteritems():
2672
result[revision_id] = tuple([parent_revid
2673
for (parent_revid,) in parent_keys])
2675
result[revision_id] = (_mod_revision.NULL_REVISION,)
2678
def _make_parents_provider(self):
2681
def get_graph(self, other_repository=None):
2682
"""Return the graph walker for this repository format"""
2683
parents_provider = self._make_parents_provider()
2684
if (other_repository is not None and
2685
not self.has_same_location(other_repository)):
2686
parents_provider = graph.StackedParentsProvider(
2687
[parents_provider, other_repository._make_parents_provider()])
2688
return graph.Graph(parents_provider)
2690
def _get_versioned_file_checker(self, text_key_references=None,
2692
"""Return an object suitable for checking versioned files.
2694
:param text_key_references: if non-None, an already built
2695
dictionary mapping text keys ((fileid, revision_id) tuples)
2696
to whether they were referred to by the inventory of the
2697
revision_id that they contain. If None, this will be
2699
:param ancestors: Optional result from
2700
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2703
return _VersionedFileChecker(self,
2704
text_key_references=text_key_references, ancestors=ancestors)
2706
def revision_ids_to_search_result(self, result_set):
2707
"""Convert a set of revision ids to a graph SearchResult."""
2708
result_parents = set()
2709
for parents in self.get_graph().get_parent_map(
2710
result_set).itervalues():
2711
result_parents.update(parents)
2712
included_keys = result_set.intersection(result_parents)
2713
start_keys = result_set.difference(included_keys)
2714
exclude_keys = result_parents.difference(result_set)
2715
result = graph.SearchResult(start_keys, exclude_keys,
2716
len(result_set), result_set)
2720
def set_make_working_trees(self, new_value):
2721
"""Set the policy flag for making working trees when creating branches.
2723
This only applies to branches that use this repository.
2725
The default is 'True'.
2726
:param new_value: True to restore the default, False to disable making
2729
raise NotImplementedError(self.set_make_working_trees)
2731
def make_working_trees(self):
2732
"""Returns the policy for making working trees on new branches."""
2733
raise NotImplementedError(self.make_working_trees)
2736
def sign_revision(self, revision_id, gpg_strategy):
2737
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2738
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2741
def has_signature_for_revision_id(self, revision_id):
2742
"""Query for a revision signature for revision_id in the repository."""
2743
if not self.has_revision(revision_id):
2744
raise errors.NoSuchRevision(self, revision_id)
2745
sig_present = (1 == len(
2746
self.signatures.get_parent_map([(revision_id,)])))
2750
def get_signature_text(self, revision_id):
2751
"""Return the text for a signature."""
2752
stream = self.signatures.get_record_stream([(revision_id,)],
2754
record = stream.next()
2755
if record.storage_kind == 'absent':
2756
raise errors.NoSuchRevision(self, revision_id)
2757
return record.get_bytes_as('fulltext')
2760
def check(self, revision_ids=None, callback_refs=None, check_repo=True):
2761
"""Check consistency of all history of given revision_ids.
2763
Different repository implementations should override _check().
2765
:param revision_ids: A non-empty list of revision_ids whose ancestry
2766
will be checked. Typically the last revision_id of a branch.
2767
:param callback_refs: A dict of check-refs to resolve and callback
2768
the check/_check method on the items listed as wanting the ref.
2770
:param check_repo: If False do not check the repository contents, just
2771
calculate the data callback_refs requires and call them back.
2773
return self._check(revision_ids, callback_refs=callback_refs,
2774
check_repo=check_repo)
2776
def _check(self, revision_ids, callback_refs, check_repo):
2777
result = check.Check(self, check_repo=check_repo)
2778
result.check(callback_refs)
2781
def _warn_if_deprecated(self):
2782
global _deprecation_warning_done
2783
if _deprecation_warning_done:
2785
_deprecation_warning_done = True
2786
warning("Format %s for %s is deprecated - please use 'bzr upgrade' to get better performance"
2787
% (self._format, self.bzrdir.transport.base))
2789
def supports_rich_root(self):
2790
return self._format.rich_root_data
2792
def _check_ascii_revisionid(self, revision_id, method):
2793
"""Private helper for ascii-only repositories."""
2794
# weave repositories refuse to store revisionids that are non-ascii.
2795
if revision_id is not None:
2796
# weaves require ascii revision ids.
2797
if isinstance(revision_id, unicode):
2799
revision_id.encode('ascii')
2800
except UnicodeEncodeError:
2801
raise errors.NonAsciiRevisionId(method, self)
2804
revision_id.decode('ascii')
2805
except UnicodeDecodeError:
2806
raise errors.NonAsciiRevisionId(method, self)
2808
def revision_graph_can_have_wrong_parents(self):
2809
"""Is it possible for this repository to have a revision graph with
2812
If True, then this repository must also implement
2813
_find_inconsistent_revision_parents so that check and reconcile can
2814
check for inconsistencies before proceeding with other checks that may
2815
depend on the revision index being consistent.
2817
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2820
# remove these delegates a while after bzr 0.15
2821
def __make_delegated(name, from_module):
2822
def _deprecated_repository_forwarder():
2823
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2824
% (name, from_module),
2827
m = __import__(from_module, globals(), locals(), [name])
2829
return getattr(m, name)
2830
except AttributeError:
2831
raise AttributeError('module %s has no name %s'
2833
globals()[name] = _deprecated_repository_forwarder
2836
'AllInOneRepository',
2837
'WeaveMetaDirRepository',
2838
'PreSplitOutRepositoryFormat',
2839
'RepositoryFormat4',
2840
'RepositoryFormat5',
2841
'RepositoryFormat6',
2842
'RepositoryFormat7',
2844
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2848
'RepositoryFormatKnit',
2849
'RepositoryFormatKnit1',
2851
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2854
def install_revision(repository, rev, revision_tree):
2855
"""Install all revision data into a repository."""
2856
install_revisions(repository, [(rev, revision_tree, None)])
2859
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2860
"""Install all revision data into a repository.
2862
Accepts an iterable of revision, tree, signature tuples. The signature
2865
repository.start_write_group()
2867
inventory_cache = lru_cache.LRUCache(10)
2868
for n, (revision, revision_tree, signature) in enumerate(iterable):
2869
_install_revision(repository, revision, revision_tree, signature,
2872
pb.update('Transferring revisions', n + 1, num_revisions)
2874
repository.abort_write_group()
2877
repository.commit_write_group()
2880
def _install_revision(repository, rev, revision_tree, signature,
2882
"""Install all revision data into a repository."""
2883
present_parents = []
2885
for p_id in rev.parent_ids:
2886
if repository.has_revision(p_id):
2887
present_parents.append(p_id)
2888
parent_trees[p_id] = repository.revision_tree(p_id)
2890
parent_trees[p_id] = repository.revision_tree(
2891
_mod_revision.NULL_REVISION)
2893
inv = revision_tree.inventory
2894
entries = inv.iter_entries()
2895
# backwards compatibility hack: skip the root id.
2896
if not repository.supports_rich_root():
2897
path, root = entries.next()
2898
if root.revision != rev.revision_id:
2899
raise errors.IncompatibleRevision(repr(repository))
2901
for path, ie in entries:
2902
text_keys[(ie.file_id, ie.revision)] = ie
2903
text_parent_map = repository.texts.get_parent_map(text_keys)
2904
missing_texts = set(text_keys) - set(text_parent_map)
2905
# Add the texts that are not already present
2906
for text_key in missing_texts:
2907
ie = text_keys[text_key]
2909
# FIXME: TODO: The following loop overlaps/duplicates that done by
2910
# commit to determine parents. There is a latent/real bug here where
2911
# the parents inserted are not those commit would do - in particular
2912
# they are not filtered by heads(). RBC, AB
2913
for revision, tree in parent_trees.iteritems():
2914
if ie.file_id not in tree:
2916
parent_id = tree.inventory[ie.file_id].revision
2917
if parent_id in text_parents:
2919
text_parents.append((ie.file_id, parent_id))
2920
lines = revision_tree.get_file(ie.file_id).readlines()
2921
repository.texts.add_lines(text_key, text_parents, lines)
2923
# install the inventory
2924
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2925
# Cache this inventory
2926
inventory_cache[rev.revision_id] = inv
2928
basis_inv = inventory_cache[rev.parent_ids[0]]
2930
repository.add_inventory(rev.revision_id, inv, present_parents)
2932
delta = inv._make_delta(basis_inv)
2933
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2934
rev.revision_id, present_parents)
2936
repository.add_inventory(rev.revision_id, inv, present_parents)
2937
except errors.RevisionAlreadyPresent:
2939
if signature is not None:
2940
repository.add_signature_text(rev.revision_id, signature)
2941
repository.add_revision(rev.revision_id, rev, inv)
2944
class MetaDirRepository(Repository):
2945
"""Repositories in the new meta-dir layout.
2947
:ivar _transport: Transport for access to repository control files,
2948
typically pointing to .bzr/repository.
2951
def __init__(self, _format, a_bzrdir, control_files):
2952
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2953
self._transport = control_files._transport
2955
def is_shared(self):
2956
"""Return True if this repository is flagged as a shared repository."""
2957
return self._transport.has('shared-storage')
2960
def set_make_working_trees(self, new_value):
2961
"""Set the policy flag for making working trees when creating branches.
2963
This only applies to branches that use this repository.
2965
The default is 'True'.
2966
:param new_value: True to restore the default, False to disable making
2971
self._transport.delete('no-working-trees')
2972
except errors.NoSuchFile:
2975
self._transport.put_bytes('no-working-trees', '',
2976
mode=self.bzrdir._get_file_mode())
2978
def make_working_trees(self):
2979
"""Returns the policy for making working trees on new branches."""
2980
return not self._transport.has('no-working-trees')
2983
class MetaDirVersionedFileRepository(MetaDirRepository):
2984
"""Repositories in a meta-dir, that work via versioned file objects."""
2986
def __init__(self, _format, a_bzrdir, control_files):
2987
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2991
network_format_registry = registry.FormatRegistry()
2992
"""Registry of formats indexed by their network name.
2994
The network name for a repository format is an identifier that can be used when
2995
referring to formats with smart server operations. See
2996
RepositoryFormat.network_name() for more detail.
3000
format_registry = registry.FormatRegistry(network_format_registry)
3001
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
3003
This can contain either format instances themselves, or classes/factories that
3004
can be called to obtain one.
3008
#####################################################################
3009
# Repository Formats
3011
class RepositoryFormat(object):
3012
"""A repository format.
3014
Formats provide four things:
3015
* An initialization routine to construct repository data on disk.
3016
* a optional format string which is used when the BzrDir supports
3018
* an open routine which returns a Repository instance.
3019
* A network name for referring to the format in smart server RPC
3022
There is one and only one Format subclass for each on-disk format. But
3023
there can be one Repository subclass that is used for several different
3024
formats. The _format attribute on a Repository instance can be used to
3025
determine the disk format.
3027
Formats are placed in a registry by their format string for reference
3028
during opening. These should be subclasses of RepositoryFormat for
3031
Once a format is deprecated, just deprecate the initialize and open
3032
methods on the format class. Do not deprecate the object, as the
3033
object may be created even when a repository instance hasn't been
3036
Common instance attributes:
3037
_matchingbzrdir - the bzrdir format that the repository format was
3038
originally written to work with. This can be used if manually
3039
constructing a bzrdir and repository, or more commonly for test suite
3043
# Set to True or False in derived classes. True indicates that the format
3044
# supports ghosts gracefully.
3045
supports_ghosts = None
3046
# Can this repository be given external locations to lookup additional
3047
# data. Set to True or False in derived classes.
3048
supports_external_lookups = None
3049
# Does this format support CHK bytestring lookups. Set to True or False in
3051
supports_chks = None
3052
# Should commit add an inventory, or an inventory delta to the repository.
3053
_commit_inv_deltas = True
3054
# What order should fetch operations request streams in?
3055
# The default is unordered as that is the cheapest for an origin to
3057
_fetch_order = 'unordered'
3058
# Does this repository format use deltas that can be fetched as-deltas ?
3059
# (E.g. knits, where the knit deltas can be transplanted intact.
3060
# We default to False, which will ensure that enough data to get
3061
# a full text out of any fetch stream will be grabbed.
3062
_fetch_uses_deltas = False
3063
# Should fetch trigger a reconcile after the fetch? Only needed for
3064
# some repository formats that can suffer internal inconsistencies.
3065
_fetch_reconcile = False
3066
# Does this format have < O(tree_size) delta generation. Used to hint what
3067
# code path for commit, amongst other things.
3069
# Does doing a pack operation compress data? Useful for the pack UI command
3070
# (so if there is one pack, the operation can still proceed because it may
3071
# help), and for fetching when data won't have come from the same
3073
pack_compresses = False
3074
# Does the repository inventory storage understand references to trees?
3075
supports_tree_reference = None
3078
return "<%s>" % self.__class__.__name__
3080
def __eq__(self, other):
3081
# format objects are generally stateless
3082
return isinstance(other, self.__class__)
3084
def __ne__(self, other):
3085
return not self == other
3088
def find_format(klass, a_bzrdir):
3089
"""Return the format for the repository object in a_bzrdir.
3091
This is used by bzr native formats that have a "format" file in
3092
the repository. Other methods may be used by different types of
3096
transport = a_bzrdir.get_repository_transport(None)
3097
format_string = transport.get_bytes("format")
3098
return format_registry.get(format_string)
3099
except errors.NoSuchFile:
3100
raise errors.NoRepositoryPresent(a_bzrdir)
3102
raise errors.UnknownFormatError(format=format_string,
3106
def register_format(klass, format):
3107
format_registry.register(format.get_format_string(), format)
3110
def unregister_format(klass, format):
3111
format_registry.remove(format.get_format_string())
3114
def get_default_format(klass):
3115
"""Return the current default format."""
3116
from bzrlib import bzrdir
3117
return bzrdir.format_registry.make_bzrdir('default').repository_format
3119
def get_format_string(self):
3120
"""Return the ASCII format string that identifies this format.
3122
Note that in pre format ?? repositories the format string is
3123
not permitted nor written to disk.
3125
raise NotImplementedError(self.get_format_string)
3127
def get_format_description(self):
3128
"""Return the short description for this format."""
3129
raise NotImplementedError(self.get_format_description)
3131
# TODO: this shouldn't be in the base class, it's specific to things that
3132
# use weaves or knits -- mbp 20070207
3133
def _get_versioned_file_store(self,
3138
versionedfile_class=None,
3139
versionedfile_kwargs={},
3141
if versionedfile_class is None:
3142
versionedfile_class = self._versionedfile_class
3143
weave_transport = control_files._transport.clone(name)
3144
dir_mode = control_files._dir_mode
3145
file_mode = control_files._file_mode
3146
return VersionedFileStore(weave_transport, prefixed=prefixed,
3148
file_mode=file_mode,
3149
versionedfile_class=versionedfile_class,
3150
versionedfile_kwargs=versionedfile_kwargs,
3153
def initialize(self, a_bzrdir, shared=False):
3154
"""Initialize a repository of this format in a_bzrdir.
3156
:param a_bzrdir: The bzrdir to put the new repository in it.
3157
:param shared: The repository should be initialized as a sharable one.
3158
:returns: The new repository object.
3160
This may raise UninitializableFormat if shared repository are not
3161
compatible the a_bzrdir.
3163
raise NotImplementedError(self.initialize)
3165
def is_supported(self):
3166
"""Is this format supported?
3168
Supported formats must be initializable and openable.
3169
Unsupported formats may not support initialization or committing or
3170
some other features depending on the reason for not being supported.
3174
def network_name(self):
3175
"""A simple byte string uniquely identifying this format for RPC calls.
3177
MetaDir repository formats use their disk format string to identify the
3178
repository over the wire. All in one formats such as bzr < 0.8, and
3179
foreign formats like svn/git and hg should use some marker which is
3180
unique and immutable.
3182
raise NotImplementedError(self.network_name)
3184
def check_conversion_target(self, target_format):
3185
if self.rich_root_data and not target_format.rich_root_data:
3186
raise errors.BadConversionTarget(
3187
'Does not support rich root data.', target_format,
3189
if (self.supports_tree_reference and
3190
not getattr(target_format, 'supports_tree_reference', False)):
3191
raise errors.BadConversionTarget(
3192
'Does not support nested trees', target_format,
3195
def open(self, a_bzrdir, _found=False):
3196
"""Return an instance of this format for the bzrdir a_bzrdir.
3198
_found is a private parameter, do not use it.
3200
raise NotImplementedError(self.open)
3203
class MetaDirRepositoryFormat(RepositoryFormat):
3204
"""Common base class for the new repositories using the metadir layout."""
3206
rich_root_data = False
3207
supports_tree_reference = False
3208
supports_external_lookups = False
3211
def _matchingbzrdir(self):
3212
matching = bzrdir.BzrDirMetaFormat1()
3213
matching.repository_format = self
3217
super(MetaDirRepositoryFormat, self).__init__()
3219
def _create_control_files(self, a_bzrdir):
3220
"""Create the required files and the initial control_files object."""
3221
# FIXME: RBC 20060125 don't peek under the covers
3222
# NB: no need to escape relative paths that are url safe.
3223
repository_transport = a_bzrdir.get_repository_transport(self)
3224
control_files = lockable_files.LockableFiles(repository_transport,
3225
'lock', lockdir.LockDir)
3226
control_files.create_lock()
3227
return control_files
3229
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
3230
"""Upload the initial blank content."""
3231
control_files = self._create_control_files(a_bzrdir)
3232
control_files.lock_write()
3233
transport = control_files._transport
3235
utf8_files += [('shared-storage', '')]
3237
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
3238
for (filename, content_stream) in files:
3239
transport.put_file(filename, content_stream,
3240
mode=a_bzrdir._get_file_mode())
3241
for (filename, content_bytes) in utf8_files:
3242
transport.put_bytes_non_atomic(filename, content_bytes,
3243
mode=a_bzrdir._get_file_mode())
3245
control_files.unlock()
3247
def network_name(self):
3248
"""Metadir formats have matching disk and network format strings."""
3249
return self.get_format_string()
3252
# Pre-0.8 formats that don't have a disk format string (because they are
3253
# versioned by the matching control directory). We use the control directories
3254
# disk format string as a key for the network_name because they meet the
3255
# constraints (simple string, unique, immutable).
3256
network_format_registry.register_lazy(
3257
"Bazaar-NG branch, format 5\n",
3258
'bzrlib.repofmt.weaverepo',
3259
'RepositoryFormat5',
3261
network_format_registry.register_lazy(
3262
"Bazaar-NG branch, format 6\n",
3263
'bzrlib.repofmt.weaverepo',
3264
'RepositoryFormat6',
3267
# formats which have no format string are not discoverable or independently
3268
# creatable on disk, so are not registered in format_registry. They're
3269
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
3270
# needed, it's constructed directly by the BzrDir. Non-native formats where
3271
# the repository is not separately opened are similar.
3273
format_registry.register_lazy(
3274
'Bazaar-NG Repository format 7',
3275
'bzrlib.repofmt.weaverepo',
3279
format_registry.register_lazy(
3280
'Bazaar-NG Knit Repository Format 1',
3281
'bzrlib.repofmt.knitrepo',
3282
'RepositoryFormatKnit1',
3285
format_registry.register_lazy(
3286
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
3287
'bzrlib.repofmt.knitrepo',
3288
'RepositoryFormatKnit3',
3291
format_registry.register_lazy(
3292
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
3293
'bzrlib.repofmt.knitrepo',
3294
'RepositoryFormatKnit4',
3297
# Pack-based formats. There is one format for pre-subtrees, and one for
3298
# post-subtrees to allow ease of testing.
3299
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
3300
format_registry.register_lazy(
3301
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
3302
'bzrlib.repofmt.pack_repo',
3303
'RepositoryFormatKnitPack1',
3305
format_registry.register_lazy(
3306
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
3307
'bzrlib.repofmt.pack_repo',
3308
'RepositoryFormatKnitPack3',
3310
format_registry.register_lazy(
3311
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
3312
'bzrlib.repofmt.pack_repo',
3313
'RepositoryFormatKnitPack4',
3315
format_registry.register_lazy(
3316
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
3317
'bzrlib.repofmt.pack_repo',
3318
'RepositoryFormatKnitPack5',
3320
format_registry.register_lazy(
3321
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
3322
'bzrlib.repofmt.pack_repo',
3323
'RepositoryFormatKnitPack5RichRoot',
3325
format_registry.register_lazy(
3326
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3327
'bzrlib.repofmt.pack_repo',
3328
'RepositoryFormatKnitPack5RichRootBroken',
3330
format_registry.register_lazy(
3331
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3332
'bzrlib.repofmt.pack_repo',
3333
'RepositoryFormatKnitPack6',
3335
format_registry.register_lazy(
3336
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3337
'bzrlib.repofmt.pack_repo',
3338
'RepositoryFormatKnitPack6RichRoot',
3341
# Development formats.
3342
# Obsolete but kept pending a CHK based subtree format.
3343
format_registry.register_lazy(
3344
("Bazaar development format 2 with subtree support "
3345
"(needs bzr.dev from before 1.8)\n"),
3346
'bzrlib.repofmt.pack_repo',
3347
'RepositoryFormatPackDevelopment2Subtree',
3350
# 1.14->1.16 go below here
3351
format_registry.register_lazy(
3352
'Bazaar development format - group compression and chk inventory'
3353
' (needs bzr.dev from 1.14)\n',
3354
'bzrlib.repofmt.groupcompress_repo',
3355
'RepositoryFormatCHK1',
3358
format_registry.register_lazy(
3359
'Bazaar development format - chk repository with bencode revision '
3360
'serialization (needs bzr.dev from 1.16)\n',
3361
'bzrlib.repofmt.groupcompress_repo',
3362
'RepositoryFormatCHK2',
3364
format_registry.register_lazy(
3365
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3366
'bzrlib.repofmt.groupcompress_repo',
3367
'RepositoryFormat2a',
3371
class InterRepository(InterObject):
3372
"""This class represents operations taking place between two repositories.
3374
Its instances have methods like copy_content and fetch, and contain
3375
references to the source and target repositories these operations can be
3378
Often we will provide convenience methods on 'repository' which carry out
3379
operations with another repository - they will always forward to
3380
InterRepository.get(other).method_name(parameters).
3383
_walk_to_common_revisions_batch_size = 50
3385
"""The available optimised InterRepository types."""
3388
def copy_content(self, revision_id=None):
3389
"""Make a complete copy of the content in self into destination.
3391
This is a destructive operation! Do not use it on existing
3394
:param revision_id: Only copy the content needed to construct
3395
revision_id and its parents.
3398
self.target.set_make_working_trees(self.source.make_working_trees())
3399
except NotImplementedError:
3401
self.target.fetch(self.source, revision_id=revision_id)
3404
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3406
"""Fetch the content required to construct revision_id.
3408
The content is copied from self.source to self.target.
3410
:param revision_id: if None all content is copied, if NULL_REVISION no
3412
:param pb: optional progress bar to use for progress reports. If not
3413
provided a default one will be created.
3416
f = _mod_fetch.RepoFetcher(to_repository=self.target,
3417
from_repository=self.source,
3418
last_revision=revision_id,
3419
fetch_spec=fetch_spec,
3420
pb=pb, find_ghosts=find_ghosts)
3422
def _walk_to_common_revisions(self, revision_ids):
3423
"""Walk out from revision_ids in source to revisions target has.
3425
:param revision_ids: The start point for the search.
3426
:return: A set of revision ids.
3428
target_graph = self.target.get_graph()
3429
revision_ids = frozenset(revision_ids)
3430
missing_revs = set()
3431
source_graph = self.source.get_graph()
3432
# ensure we don't pay silly lookup costs.
3433
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3434
null_set = frozenset([_mod_revision.NULL_REVISION])
3435
searcher_exhausted = False
3439
# Iterate the searcher until we have enough next_revs
3440
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3442
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3443
next_revs.update(next_revs_part)
3444
ghosts.update(ghosts_part)
3445
except StopIteration:
3446
searcher_exhausted = True
3448
# If there are ghosts in the source graph, and the caller asked for
3449
# them, make sure that they are present in the target.
3450
# We don't care about other ghosts as we can't fetch them and
3451
# haven't been asked to.
3452
ghosts_to_check = set(revision_ids.intersection(ghosts))
3453
revs_to_get = set(next_revs).union(ghosts_to_check)
3455
have_revs = set(target_graph.get_parent_map(revs_to_get))
3456
# we always have NULL_REVISION present.
3457
have_revs = have_revs.union(null_set)
3458
# Check if the target is missing any ghosts we need.
3459
ghosts_to_check.difference_update(have_revs)
3461
# One of the caller's revision_ids is a ghost in both the
3462
# source and the target.
3463
raise errors.NoSuchRevision(
3464
self.source, ghosts_to_check.pop())
3465
missing_revs.update(next_revs - have_revs)
3466
# Because we may have walked past the original stop point, make
3467
# sure everything is stopped
3468
stop_revs = searcher.find_seen_ancestors(have_revs)
3469
searcher.stop_searching_any(stop_revs)
3470
if searcher_exhausted:
3472
return searcher.get_result()
3475
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3476
"""Return the revision ids that source has that target does not.
3478
:param revision_id: only return revision ids included by this
3480
:param find_ghosts: If True find missing revisions in deep history
3481
rather than just finding the surface difference.
3482
:return: A bzrlib.graph.SearchResult.
3484
# stop searching at found target revisions.
3485
if not find_ghosts and revision_id is not None:
3486
return self._walk_to_common_revisions([revision_id])
3487
# generic, possibly worst case, slow code path.
3488
target_ids = set(self.target.all_revision_ids())
3489
if revision_id is not None:
3490
source_ids = self.source.get_ancestry(revision_id)
3491
if source_ids[0] is not None:
3492
raise AssertionError()
3495
source_ids = self.source.all_revision_ids()
3496
result_set = set(source_ids).difference(target_ids)
3497
return self.source.revision_ids_to_search_result(result_set)
3500
def _same_model(source, target):
3501
"""True if source and target have the same data representation.
3503
Note: this is always called on the base class; overriding it in a
3504
subclass will have no effect.
3507
InterRepository._assert_same_model(source, target)
3509
except errors.IncompatibleRepositories, e:
3513
def _assert_same_model(source, target):
3514
"""Raise an exception if two repositories do not use the same model.
3516
if source.supports_rich_root() != target.supports_rich_root():
3517
raise errors.IncompatibleRepositories(source, target,
3518
"different rich-root support")
3519
if source._serializer != target._serializer:
3520
raise errors.IncompatibleRepositories(source, target,
3521
"different serializers")
3524
class InterSameDataRepository(InterRepository):
3525
"""Code for converting between repositories that represent the same data.
3527
Data format and model must match for this to work.
3531
def _get_repo_format_to_test(self):
3532
"""Repository format for testing with.
3534
InterSameData can pull from subtree to subtree and from non-subtree to
3535
non-subtree, so we test this with the richest repository format.
3537
from bzrlib.repofmt import knitrepo
3538
return knitrepo.RepositoryFormatKnit3()
3541
def is_compatible(source, target):
3542
return InterRepository._same_model(source, target)
3545
class InterWeaveRepo(InterSameDataRepository):
3546
"""Optimised code paths between Weave based repositories.
3548
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3549
implemented lazy inter-object optimisation.
3553
def _get_repo_format_to_test(self):
3554
from bzrlib.repofmt import weaverepo
3555
return weaverepo.RepositoryFormat7()
3558
def is_compatible(source, target):
3559
"""Be compatible with known Weave formats.
3561
We don't test for the stores being of specific types because that
3562
could lead to confusing results, and there is no need to be
3565
from bzrlib.repofmt.weaverepo import (
3571
return (isinstance(source._format, (RepositoryFormat5,
3573
RepositoryFormat7)) and
3574
isinstance(target._format, (RepositoryFormat5,
3576
RepositoryFormat7)))
3577
except AttributeError:
3581
def copy_content(self, revision_id=None):
3582
"""See InterRepository.copy_content()."""
3583
# weave specific optimised path:
3585
self.target.set_make_working_trees(self.source.make_working_trees())
3586
except (errors.RepositoryUpgradeRequired, NotImplemented):
3588
# FIXME do not peek!
3589
if self.source._transport.listable():
3590
pb = ui.ui_factory.nested_progress_bar()
3592
self.target.texts.insert_record_stream(
3593
self.source.texts.get_record_stream(
3594
self.source.texts.keys(), 'topological', False))
3595
pb.update('Copying inventory', 0, 1)
3596
self.target.inventories.insert_record_stream(
3597
self.source.inventories.get_record_stream(
3598
self.source.inventories.keys(), 'topological', False))
3599
self.target.signatures.insert_record_stream(
3600
self.source.signatures.get_record_stream(
3601
self.source.signatures.keys(),
3603
self.target.revisions.insert_record_stream(
3604
self.source.revisions.get_record_stream(
3605
self.source.revisions.keys(),
3606
'topological', True))
3610
self.target.fetch(self.source, revision_id=revision_id)
3613
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3614
"""See InterRepository.missing_revision_ids()."""
3615
# we want all revisions to satisfy revision_id in source.
3616
# but we don't want to stat every file here and there.
3617
# we want then, all revisions other needs to satisfy revision_id
3618
# checked, but not those that we have locally.
3619
# so the first thing is to get a subset of the revisions to
3620
# satisfy revision_id in source, and then eliminate those that
3621
# we do already have.
3622
# this is slow on high latency connection to self, but as this
3623
# disk format scales terribly for push anyway due to rewriting
3624
# inventory.weave, this is considered acceptable.
3626
if revision_id is not None:
3627
source_ids = self.source.get_ancestry(revision_id)
3628
if source_ids[0] is not None:
3629
raise AssertionError()
3632
source_ids = self.source._all_possible_ids()
3633
source_ids_set = set(source_ids)
3634
# source_ids is the worst possible case we may need to pull.
3635
# now we want to filter source_ids against what we actually
3636
# have in target, but don't try to check for existence where we know
3637
# we do not have a revision as that would be pointless.
3638
target_ids = set(self.target._all_possible_ids())
3639
possibly_present_revisions = target_ids.intersection(source_ids_set)
3640
actually_present_revisions = set(
3641
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3642
required_revisions = source_ids_set.difference(actually_present_revisions)
3643
if revision_id is not None:
3644
# we used get_ancestry to determine source_ids then we are assured all
3645
# revisions referenced are present as they are installed in topological order.
3646
# and the tip revision was validated by get_ancestry.
3647
result_set = required_revisions
3649
# if we just grabbed the possibly available ids, then
3650
# we only have an estimate of whats available and need to validate
3651
# that against the revision records.
3653
self.source._eliminate_revisions_not_present(required_revisions))
3654
return self.source.revision_ids_to_search_result(result_set)
3657
class InterKnitRepo(InterSameDataRepository):
3658
"""Optimised code paths between Knit based repositories."""
3661
def _get_repo_format_to_test(self):
3662
from bzrlib.repofmt import knitrepo
3663
return knitrepo.RepositoryFormatKnit1()
3666
def is_compatible(source, target):
3667
"""Be compatible with known Knit formats.
3669
We don't test for the stores being of specific types because that
3670
could lead to confusing results, and there is no need to be
3673
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3675
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3676
isinstance(target._format, RepositoryFormatKnit))
3677
except AttributeError:
3679
return are_knits and InterRepository._same_model(source, target)
3682
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3683
"""See InterRepository.missing_revision_ids()."""
3684
if revision_id is not None:
3685
source_ids = self.source.get_ancestry(revision_id)
3686
if source_ids[0] is not None:
3687
raise AssertionError()
3690
source_ids = self.source.all_revision_ids()
3691
source_ids_set = set(source_ids)
3692
# source_ids is the worst possible case we may need to pull.
3693
# now we want to filter source_ids against what we actually
3694
# have in target, but don't try to check for existence where we know
3695
# we do not have a revision as that would be pointless.
3696
target_ids = set(self.target.all_revision_ids())
3697
possibly_present_revisions = target_ids.intersection(source_ids_set)
3698
actually_present_revisions = set(
3699
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3700
required_revisions = source_ids_set.difference(actually_present_revisions)
3701
if revision_id is not None:
3702
# we used get_ancestry to determine source_ids then we are assured all
3703
# revisions referenced are present as they are installed in topological order.
3704
# and the tip revision was validated by get_ancestry.
3705
result_set = required_revisions
3707
# if we just grabbed the possibly available ids, then
3708
# we only have an estimate of whats available and need to validate
3709
# that against the revision records.
3711
self.source._eliminate_revisions_not_present(required_revisions))
3712
return self.source.revision_ids_to_search_result(result_set)
3715
class InterDifferingSerializer(InterRepository):
3718
def _get_repo_format_to_test(self):
3722
def is_compatible(source, target):
3723
"""Be compatible with Knit2 source and Knit3 target"""
3724
# This is redundant with format.check_conversion_target(), however that
3725
# raises an exception, and we just want to say "False" as in we won't
3726
# support converting between these formats.
3727
if 'IDS_never' in debug.debug_flags:
3729
if source.supports_rich_root() and not target.supports_rich_root():
3731
if (source._format.supports_tree_reference
3732
and not target._format.supports_tree_reference):
3734
if target._fallback_repositories and target._format.supports_chks:
3735
# IDS doesn't know how to copy CHKs for the parent inventories it
3736
# adds to stacked repos.
3738
if 'IDS_always' in debug.debug_flags:
3740
# Only use this code path for local source and target. IDS does far
3741
# too much IO (both bandwidth and roundtrips) over a network.
3742
if not source.bzrdir.transport.base.startswith('file:///'):
3744
if not target.bzrdir.transport.base.startswith('file:///'):
3748
def _get_trees(self, revision_ids, cache):
3750
for rev_id in revision_ids:
3752
possible_trees.append((rev_id, cache[rev_id]))
3754
# Not cached, but inventory might be present anyway.
3756
tree = self.source.revision_tree(rev_id)
3757
except errors.NoSuchRevision:
3758
# Nope, parent is ghost.
3761
cache[rev_id] = tree
3762
possible_trees.append((rev_id, tree))
3763
return possible_trees
3765
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3766
"""Get the best delta and base for this revision.
3768
:return: (basis_id, delta)
3771
# Generate deltas against each tree, to find the shortest.
3772
texts_possibly_new_in_tree = set()
3773
for basis_id, basis_tree in possible_trees:
3774
delta = tree.inventory._make_delta(basis_tree.inventory)
3775
for old_path, new_path, file_id, new_entry in delta:
3776
if new_path is None:
3777
# This file_id isn't present in the new rev, so we don't
3781
# Rich roots are handled elsewhere...
3783
kind = new_entry.kind
3784
if kind != 'directory' and kind != 'file':
3785
# No text record associated with this inventory entry.
3787
# This is a directory or file that has changed somehow.
3788
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3789
deltas.append((len(delta), basis_id, delta))
3791
return deltas[0][1:]
3793
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3794
"""Find all parent revisions that are absent, but for which the
3795
inventory is present, and copy those inventories.
3797
This is necessary to preserve correctness when the source is stacked
3798
without fallbacks configured. (Note that in cases like upgrade the
3799
source may be not have _fallback_repositories even though it is
3803
for parents in parent_map.values():
3804
parent_revs.update(parents)
3805
present_parents = self.source.get_parent_map(parent_revs)
3806
absent_parents = set(parent_revs).difference(present_parents)
3807
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3808
(rev_id,) for rev_id in absent_parents)
3809
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3810
for parent_tree in self.source.revision_trees(parent_inv_ids):
3811
current_revision_id = parent_tree.get_revision_id()
3812
parents_parents_keys = parent_invs_keys_for_stacking[
3813
(current_revision_id,)]
3814
parents_parents = [key[-1] for key in parents_parents_keys]
3815
basis_id = _mod_revision.NULL_REVISION
3816
basis_tree = self.source.revision_tree(basis_id)
3817
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3818
self.target.add_inventory_by_delta(
3819
basis_id, delta, current_revision_id, parents_parents)
3820
cache[current_revision_id] = parent_tree
3822
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3823
"""Fetch across a few revisions.
3825
:param revision_ids: The revisions to copy
3826
:param basis_id: The revision_id of a tree that must be in cache, used
3827
as a basis for delta when no other base is available
3828
:param cache: A cache of RevisionTrees that we can use.
3829
:param a_graph: A Graph object to determine the heads() of the
3830
rich-root data stream.
3831
:return: The revision_id of the last converted tree. The RevisionTree
3832
for it will be in cache
3834
# Walk though all revisions; get inventory deltas, copy referenced
3835
# texts that delta references, insert the delta, revision and
3837
root_keys_to_create = set()
3840
pending_revisions = []
3841
parent_map = self.source.get_parent_map(revision_ids)
3842
self._fetch_parent_invs_for_stacking(parent_map, cache)
3843
for tree in self.source.revision_trees(revision_ids):
3844
# Find a inventory delta for this revision.
3845
# Find text entries that need to be copied, too.
3846
current_revision_id = tree.get_revision_id()
3847
parent_ids = parent_map.get(current_revision_id, ())
3848
parent_trees = self._get_trees(parent_ids, cache)
3849
possible_trees = list(parent_trees)
3850
if len(possible_trees) == 0:
3851
# There either aren't any parents, or the parents are ghosts,
3852
# so just use the last converted tree.
3853
possible_trees.append((basis_id, cache[basis_id]))
3854
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3856
revision = self.source.get_revision(current_revision_id)
3857
pending_deltas.append((basis_id, delta,
3858
current_revision_id, revision.parent_ids))
3859
if self._converting_to_rich_root:
3860
self._revision_id_to_root_id[current_revision_id] = \
3862
# Determine which texts are in present in this revision but not in
3863
# any of the available parents.
3864
texts_possibly_new_in_tree = set()
3865
for old_path, new_path, file_id, entry in delta:
3866
if new_path is None:
3867
# This file_id isn't present in the new rev
3871
if not self.target.supports_rich_root():
3872
# The target doesn't support rich root, so we don't
3875
if self._converting_to_rich_root:
3876
# This can't be copied normally, we have to insert
3878
root_keys_to_create.add((file_id, entry.revision))
3881
texts_possibly_new_in_tree.add((file_id, entry.revision))
3882
for basis_id, basis_tree in possible_trees:
3883
basis_inv = basis_tree.inventory
3884
for file_key in list(texts_possibly_new_in_tree):
3885
file_id, file_revision = file_key
3887
entry = basis_inv[file_id]
3888
except errors.NoSuchId:
3890
if entry.revision == file_revision:
3891
texts_possibly_new_in_tree.remove(file_key)
3892
text_keys.update(texts_possibly_new_in_tree)
3893
pending_revisions.append(revision)
3894
cache[current_revision_id] = tree
3895
basis_id = current_revision_id
3897
from_texts = self.source.texts
3898
to_texts = self.target.texts
3899
if root_keys_to_create:
3900
root_stream = _mod_fetch._new_root_data_stream(
3901
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3902
self.source, graph=a_graph)
3903
to_texts.insert_record_stream(root_stream)
3904
to_texts.insert_record_stream(from_texts.get_record_stream(
3905
text_keys, self.target._format._fetch_order,
3906
not self.target._format._fetch_uses_deltas))
3907
# insert inventory deltas
3908
for delta in pending_deltas:
3909
self.target.add_inventory_by_delta(*delta)
3910
if self.target._fallback_repositories:
3911
# Make sure this stacked repository has all the parent inventories
3912
# for the new revisions that we are about to insert. We do this
3913
# before adding the revisions so that no revision is added until
3914
# all the inventories it may depend on are added.
3915
# Note that this is overzealous, as we may have fetched these in an
3918
revision_ids = set()
3919
for revision in pending_revisions:
3920
revision_ids.add(revision.revision_id)
3921
parent_ids.update(revision.parent_ids)
3922
parent_ids.difference_update(revision_ids)
3923
parent_ids.discard(_mod_revision.NULL_REVISION)
3924
parent_map = self.source.get_parent_map(parent_ids)
3925
# we iterate over parent_map and not parent_ids because we don't
3926
# want to try copying any revision which is a ghost
3927
for parent_tree in self.source.revision_trees(parent_map):
3928
current_revision_id = parent_tree.get_revision_id()
3929
parents_parents = parent_map[current_revision_id]
3930
possible_trees = self._get_trees(parents_parents, cache)
3931
if len(possible_trees) == 0:
3932
# There either aren't any parents, or the parents are
3933
# ghosts, so just use the last converted tree.
3934
possible_trees.append((basis_id, cache[basis_id]))
3935
basis_id, delta = self._get_delta_for_revision(parent_tree,
3936
parents_parents, possible_trees)
3937
self.target.add_inventory_by_delta(
3938
basis_id, delta, current_revision_id, parents_parents)
3939
# insert signatures and revisions
3940
for revision in pending_revisions:
3942
signature = self.source.get_signature_text(
3943
revision.revision_id)
3944
self.target.add_signature_text(revision.revision_id,
3946
except errors.NoSuchRevision:
3948
self.target.add_revision(revision.revision_id, revision)
3951
def _fetch_all_revisions(self, revision_ids, pb):
3952
"""Fetch everything for the list of revisions.
3954
:param revision_ids: The list of revisions to fetch. Must be in
3956
:param pb: A ProgressTask
3959
basis_id, basis_tree = self._get_basis(revision_ids[0])
3961
cache = lru_cache.LRUCache(100)
3962
cache[basis_id] = basis_tree
3963
del basis_tree # We don't want to hang on to it here
3965
if self._converting_to_rich_root and len(revision_ids) > 100:
3966
a_graph = _mod_fetch._get_rich_root_heads_graph(self.source,
3971
for offset in range(0, len(revision_ids), batch_size):
3972
self.target.start_write_group()
3974
pb.update('Transferring revisions', offset,
3976
batch = revision_ids[offset:offset+batch_size]
3977
basis_id = self._fetch_batch(batch, basis_id, cache,
3980
self.target.abort_write_group()
3983
hint = self.target.commit_write_group()
3986
if hints and self.target._format.pack_compresses:
3987
self.target.pack(hint=hints)
3988
pb.update('Transferring revisions', len(revision_ids),
3992
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3994
"""See InterRepository.fetch()."""
3995
if fetch_spec is not None:
3996
raise AssertionError("Not implemented yet...")
3997
if (not self.source.supports_rich_root()
3998
and self.target.supports_rich_root()):
3999
self._converting_to_rich_root = True
4000
self._revision_id_to_root_id = {}
4002
self._converting_to_rich_root = False
4003
revision_ids = self.target.search_missing_revision_ids(self.source,
4004
revision_id, find_ghosts=find_ghosts).get_keys()
4005
if not revision_ids:
4007
revision_ids = tsort.topo_sort(
4008
self.source.get_graph().get_parent_map(revision_ids))
4009
if not revision_ids:
4011
# Walk though all revisions; get inventory deltas, copy referenced
4012
# texts that delta references, insert the delta, revision and
4015
my_pb = ui.ui_factory.nested_progress_bar()
4018
symbol_versioning.warn(
4019
symbol_versioning.deprecated_in((1, 14, 0))
4020
% "pb parameter to fetch()")
4023
self._fetch_all_revisions(revision_ids, pb)
4025
if my_pb is not None:
4027
return len(revision_ids), 0
4029
def _get_basis(self, first_revision_id):
4030
"""Get a revision and tree which exists in the target.
4032
This assumes that first_revision_id is selected for transmission
4033
because all other ancestors are already present. If we can't find an
4034
ancestor we fall back to NULL_REVISION since we know that is safe.
4036
:return: (basis_id, basis_tree)
4038
first_rev = self.source.get_revision(first_revision_id)
4040
basis_id = first_rev.parent_ids[0]
4041
# only valid as a basis if the target has it
4042
self.target.get_revision(basis_id)
4043
# Try to get a basis tree - if its a ghost it will hit the
4044
# NoSuchRevision case.
4045
basis_tree = self.source.revision_tree(basis_id)
4046
except (IndexError, errors.NoSuchRevision):
4047
basis_id = _mod_revision.NULL_REVISION
4048
basis_tree = self.source.revision_tree(basis_id)
4049
return basis_id, basis_tree
4052
InterRepository.register_optimiser(InterDifferingSerializer)
4053
InterRepository.register_optimiser(InterSameDataRepository)
4054
InterRepository.register_optimiser(InterWeaveRepo)
4055
InterRepository.register_optimiser(InterKnitRepo)
4058
class CopyConverter(object):
4059
"""A repository conversion tool which just performs a copy of the content.
4061
This is slow but quite reliable.
4064
def __init__(self, target_format):
4065
"""Create a CopyConverter.
4067
:param target_format: The format the resulting repository should be.
4069
self.target_format = target_format
4071
def convert(self, repo, pb):
4072
"""Perform the conversion of to_convert, giving feedback via pb.
4074
:param to_convert: The disk object to convert.
4075
:param pb: a progress bar to use for progress information.
4080
# this is only useful with metadir layouts - separated repo content.
4081
# trigger an assertion if not such
4082
repo._format.get_format_string()
4083
self.repo_dir = repo.bzrdir
4084
self.step('Moving repository to repository.backup')
4085
self.repo_dir.transport.move('repository', 'repository.backup')
4086
backup_transport = self.repo_dir.transport.clone('repository.backup')
4087
repo._format.check_conversion_target(self.target_format)
4088
self.source_repo = repo._format.open(self.repo_dir,
4090
_override_transport=backup_transport)
4091
self.step('Creating new repository')
4092
converted = self.target_format.initialize(self.repo_dir,
4093
self.source_repo.is_shared())
4094
converted.lock_write()
4096
self.step('Copying content')
4097
self.source_repo.copy_content_into(converted)
4100
self.step('Deleting old repository content')
4101
self.repo_dir.transport.delete_tree('repository.backup')
4102
ui.ui_factory.note('repository converted')
4104
def step(self, message):
4105
"""Update the pb by a step."""
4107
self.pb.update(message, self.count, self.total)
4119
def _unescaper(match, _map=_unescape_map):
4120
code = match.group(1)
4124
if not code.startswith('#'):
4126
return unichr(int(code[1:])).encode('utf8')
4132
def _unescape_xml(data):
4133
"""Unescape predefined XML entities in a string of data."""
4135
if _unescape_re is None:
4136
_unescape_re = re.compile('\&([^;]*);')
4137
return _unescape_re.sub(_unescaper, data)
4140
class _VersionedFileChecker(object):
4142
def __init__(self, repository, text_key_references=None, ancestors=None):
4143
self.repository = repository
4144
self.text_index = self.repository._generate_text_key_index(
4145
text_key_references=text_key_references, ancestors=ancestors)
4147
def calculate_file_version_parents(self, text_key):
4148
"""Calculate the correct parents for a file version according to
4151
parent_keys = self.text_index[text_key]
4152
if parent_keys == [_mod_revision.NULL_REVISION]:
4154
return tuple(parent_keys)
4156
def check_file_version_parents(self, texts, progress_bar=None):
4157
"""Check the parents stored in a versioned file are correct.
4159
It also detects file versions that are not referenced by their
4160
corresponding revision's inventory.
4162
:returns: A tuple of (wrong_parents, dangling_file_versions).
4163
wrong_parents is a dict mapping {revision_id: (stored_parents,
4164
correct_parents)} for each revision_id where the stored parents
4165
are not correct. dangling_file_versions is a set of (file_id,
4166
revision_id) tuples for versions that are present in this versioned
4167
file, but not used by the corresponding inventory.
4169
local_progress = None
4170
if progress_bar is None:
4171
local_progress = ui.ui_factory.nested_progress_bar()
4172
progress_bar = local_progress
4174
return self._check_file_version_parents(texts, progress_bar)
4177
local_progress.finished()
4179
def _check_file_version_parents(self, texts, progress_bar):
4180
"""See check_file_version_parents."""
4182
self.file_ids = set([file_id for file_id, _ in
4183
self.text_index.iterkeys()])
4184
# text keys is now grouped by file_id
4185
n_versions = len(self.text_index)
4186
progress_bar.update('loading text store', 0, n_versions)
4187
parent_map = self.repository.texts.get_parent_map(self.text_index)
4188
# On unlistable transports this could well be empty/error...
4189
text_keys = self.repository.texts.keys()
4190
unused_keys = frozenset(text_keys) - set(self.text_index)
4191
for num, key in enumerate(self.text_index.iterkeys()):
4192
progress_bar.update('checking text graph', num, n_versions)
4193
correct_parents = self.calculate_file_version_parents(key)
4195
knit_parents = parent_map[key]
4196
except errors.RevisionNotPresent:
4199
if correct_parents != knit_parents:
4200
wrong_parents[key] = (knit_parents, correct_parents)
4201
return wrong_parents, unused_keys
4204
def _old_get_graph(repository, revision_id):
4205
"""DO NOT USE. That is all. I'm serious."""
4206
graph = repository.get_graph()
4207
revision_graph = dict(((key, value) for key, value in
4208
graph.iter_ancestry([revision_id]) if value is not None))
4209
return _strip_NULL_ghosts(revision_graph)
4212
def _strip_NULL_ghosts(revision_graph):
4213
"""Also don't use this. more compatibility code for unmigrated clients."""
4214
# Filter ghosts, and null:
4215
if _mod_revision.NULL_REVISION in revision_graph:
4216
del revision_graph[_mod_revision.NULL_REVISION]
4217
for key, parents in revision_graph.items():
4218
revision_graph[key] = tuple(parent for parent in parents if parent
4220
return revision_graph
4223
class StreamSink(object):
4224
"""An object that can insert a stream into a repository.
4226
This interface handles the complexity of reserialising inventories and
4227
revisions from different formats, and allows unidirectional insertion into
4228
stacked repositories without looking for the missing basis parents
4232
def __init__(self, target_repo):
4233
self.target_repo = target_repo
4235
def insert_stream(self, stream, src_format, resume_tokens):
4236
"""Insert a stream's content into the target repository.
4238
:param src_format: a bzr repository format.
4240
:return: a list of resume tokens and an iterable of keys additional
4241
items required before the insertion can be completed.
4243
self.target_repo.lock_write()
4246
self.target_repo.resume_write_group(resume_tokens)
4249
self.target_repo.start_write_group()
4252
# locked_insert_stream performs a commit|suspend.
4253
return self._locked_insert_stream(stream, src_format, is_resume)
4255
self.target_repo.abort_write_group(suppress_errors=True)
4258
self.target_repo.unlock()
4260
def _locked_insert_stream(self, stream, src_format, is_resume):
4261
to_serializer = self.target_repo._format._serializer
4262
src_serializer = src_format._serializer
4264
if to_serializer == src_serializer:
4265
# If serializers match and the target is a pack repository, set the
4266
# write cache size on the new pack. This avoids poor performance
4267
# on transports where append is unbuffered (such as
4268
# RemoteTransport). This is safe to do because nothing should read
4269
# back from the target repository while a stream with matching
4270
# serialization is being inserted.
4271
# The exception is that a delta record from the source that should
4272
# be a fulltext may need to be expanded by the target (see
4273
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4274
# explicitly flush any buffered writes first in that rare case.
4276
new_pack = self.target_repo._pack_collection._new_pack
4277
except AttributeError:
4278
# Not a pack repository
4281
new_pack.set_write_cache_size(1024*1024)
4282
for substream_type, substream in stream:
4283
if 'stream' in debug.debug_flags:
4284
mutter('inserting substream: %s', substream_type)
4285
if substream_type == 'texts':
4286
self.target_repo.texts.insert_record_stream(substream)
4287
elif substream_type == 'inventories':
4288
if src_serializer == to_serializer:
4289
self.target_repo.inventories.insert_record_stream(
4292
self._extract_and_insert_inventories(
4293
substream, src_serializer)
4294
elif substream_type == 'inventory-deltas':
4295
self._extract_and_insert_inventory_deltas(
4296
substream, src_serializer)
4297
elif substream_type == 'chk_bytes':
4298
# XXX: This doesn't support conversions, as it assumes the
4299
# conversion was done in the fetch code.
4300
self.target_repo.chk_bytes.insert_record_stream(substream)
4301
elif substream_type == 'revisions':
4302
# This may fallback to extract-and-insert more often than
4303
# required if the serializers are different only in terms of
4305
if src_serializer == to_serializer:
4306
self.target_repo.revisions.insert_record_stream(
4309
self._extract_and_insert_revisions(substream,
4311
elif substream_type == 'signatures':
4312
self.target_repo.signatures.insert_record_stream(substream)
4314
raise AssertionError('kaboom! %s' % (substream_type,))
4315
# Done inserting data, and the missing_keys calculations will try to
4316
# read back from the inserted data, so flush the writes to the new pack
4317
# (if this is pack format).
4318
if new_pack is not None:
4319
new_pack._write_data('', flush=True)
4320
# Find all the new revisions (including ones from resume_tokens)
4321
missing_keys = self.target_repo.get_missing_parent_inventories(
4322
check_for_missing_texts=is_resume)
4324
for prefix, versioned_file in (
4325
('texts', self.target_repo.texts),
4326
('inventories', self.target_repo.inventories),
4327
('revisions', self.target_repo.revisions),
4328
('signatures', self.target_repo.signatures),
4329
('chk_bytes', self.target_repo.chk_bytes),
4331
if versioned_file is None:
4333
# TODO: key is often going to be a StaticTuple object
4334
# I don't believe we can define a method by which
4335
# (prefix,) + StaticTuple will work, though we could
4336
# define a StaticTuple.sq_concat that would allow you to
4337
# pass in either a tuple or a StaticTuple as the second
4338
# object, so instead we could have:
4339
# StaticTuple(prefix) + key here...
4340
missing_keys.update((prefix,) + key for key in
4341
versioned_file.get_missing_compression_parent_keys())
4342
except NotImplementedError:
4343
# cannot even attempt suspending, and missing would have failed
4344
# during stream insertion.
4345
missing_keys = set()
4348
# suspend the write group and tell the caller what we is
4349
# missing. We know we can suspend or else we would not have
4350
# entered this code path. (All repositories that can handle
4351
# missing keys can handle suspending a write group).
4352
write_group_tokens = self.target_repo.suspend_write_group()
4353
return write_group_tokens, missing_keys
4354
hint = self.target_repo.commit_write_group()
4355
if (to_serializer != src_serializer and
4356
self.target_repo._format.pack_compresses):
4357
self.target_repo.pack(hint=hint)
4360
def _extract_and_insert_inventory_deltas(self, substream, serializer):
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
# Insert the delta directly
4365
inventory_delta_bytes = record.get_bytes_as('fulltext')
4366
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4368
parse_result = deserialiser.parse_text_bytes(
4369
inventory_delta_bytes)
4370
except inventory_delta.IncompatibleInventoryDelta, err:
4371
trace.mutter("Incompatible delta: %s", err.msg)
4372
raise errors.IncompatibleRevision(self.target_repo._format)
4373
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4374
revision_id = new_id
4375
parents = [key[0] for key in record.parents]
4376
self.target_repo.add_inventory_by_delta(
4377
basis_id, inv_delta, revision_id, parents)
4379
def _extract_and_insert_inventories(self, substream, serializer,
4381
"""Generate a new inventory versionedfile in target, converting data.
4383
The inventory is retrieved from the source, (deserializing it), and
4384
stored in the target (reserializing it in a different format).
4386
target_rich_root = self.target_repo._format.rich_root_data
4387
target_tree_refs = self.target_repo._format.supports_tree_reference
4388
for record in substream:
4389
# It's not a delta, so it must be a fulltext in the source
4390
# serializer's format.
4391
bytes = record.get_bytes_as('fulltext')
4392
revision_id = record.key[0]
4393
inv = serializer.read_inventory_from_string(bytes, revision_id)
4394
parents = [key[0] for key in record.parents]
4395
self.target_repo.add_inventory(revision_id, inv, parents)
4396
# No need to keep holding this full inv in memory when the rest of
4397
# the substream is likely to be all deltas.
4400
def _extract_and_insert_revisions(self, substream, serializer):
4401
for record in substream:
4402
bytes = record.get_bytes_as('fulltext')
4403
revision_id = record.key[0]
4404
rev = serializer.read_revision_from_string(bytes)
4405
if rev.revision_id != revision_id:
4406
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4407
self.target_repo.add_revision(revision_id, rev)
4410
if self.target_repo._format._fetch_reconcile:
4411
self.target_repo.reconcile()
4414
class StreamSource(object):
4415
"""A source of a stream for fetching between repositories."""
4417
def __init__(self, from_repository, to_format):
4418
"""Create a StreamSource streaming from from_repository."""
4419
self.from_repository = from_repository
4420
self.to_format = to_format
4422
def delta_on_metadata(self):
4423
"""Return True if delta's are permitted on metadata streams.
4425
That is on revisions and signatures.
4427
src_serializer = self.from_repository._format._serializer
4428
target_serializer = self.to_format._serializer
4429
return (self.to_format._fetch_uses_deltas and
4430
src_serializer == target_serializer)
4432
def _fetch_revision_texts(self, revs):
4433
# fetch signatures first and then the revision texts
4434
# may need to be a InterRevisionStore call here.
4435
from_sf = self.from_repository.signatures
4436
# A missing signature is just skipped.
4437
keys = [(rev_id,) for rev_id in revs]
4438
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4440
self.to_format._fetch_order,
4441
not self.to_format._fetch_uses_deltas))
4442
# If a revision has a delta, this is actually expanded inside the
4443
# insert_record_stream code now, which is an alternate fix for
4445
from_rf = self.from_repository.revisions
4446
revisions = from_rf.get_record_stream(
4448
self.to_format._fetch_order,
4449
not self.delta_on_metadata())
4450
return [('signatures', signatures), ('revisions', revisions)]
4452
def _generate_root_texts(self, revs):
4453
"""This will be called by get_stream between fetching weave texts and
4454
fetching the inventory weave.
4456
if self._rich_root_upgrade():
4457
return _mod_fetch.Inter1and2Helper(
4458
self.from_repository).generate_root_texts(revs)
4462
def get_stream(self, search):
4464
revs = search.get_keys()
4465
graph = self.from_repository.get_graph()
4466
revs = tsort.topo_sort(graph.get_parent_map(revs))
4467
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4469
for knit_kind, file_id, revisions in data_to_fetch:
4470
if knit_kind != phase:
4472
# Make a new progress bar for this phase
4473
if knit_kind == "file":
4474
# Accumulate file texts
4475
text_keys.extend([(file_id, revision) for revision in
4477
elif knit_kind == "inventory":
4478
# Now copy the file texts.
4479
from_texts = self.from_repository.texts
4480
yield ('texts', from_texts.get_record_stream(
4481
text_keys, self.to_format._fetch_order,
4482
not self.to_format._fetch_uses_deltas))
4483
# Cause an error if a text occurs after we have done the
4486
# Before we process the inventory we generate the root
4487
# texts (if necessary) so that the inventories references
4489
for _ in self._generate_root_texts(revs):
4491
# we fetch only the referenced inventories because we do not
4492
# know for unselected inventories whether all their required
4493
# texts are present in the other repository - it could be
4495
for info in self._get_inventory_stream(revs):
4497
elif knit_kind == "signatures":
4498
# Nothing to do here; this will be taken care of when
4499
# _fetch_revision_texts happens.
4501
elif knit_kind == "revisions":
4502
for record in self._fetch_revision_texts(revs):
4505
raise AssertionError("Unknown knit kind %r" % knit_kind)
4507
def get_stream_for_missing_keys(self, missing_keys):
4508
# missing keys can only occur when we are byte copying and not
4509
# translating (because translation means we don't send
4510
# unreconstructable deltas ever).
4512
keys['texts'] = set()
4513
keys['revisions'] = set()
4514
keys['inventories'] = set()
4515
keys['chk_bytes'] = set()
4516
keys['signatures'] = set()
4517
for key in missing_keys:
4518
keys[key[0]].add(key[1:])
4519
if len(keys['revisions']):
4520
# If we allowed copying revisions at this point, we could end up
4521
# copying a revision without copying its required texts: a
4522
# violation of the requirements for repository integrity.
4523
raise AssertionError(
4524
'cannot copy revisions to fill in missing deltas %s' % (
4525
keys['revisions'],))
4526
for substream_kind, keys in keys.iteritems():
4527
vf = getattr(self.from_repository, substream_kind)
4528
if vf is None and keys:
4529
raise AssertionError(
4530
"cannot fill in keys for a versioned file we don't"
4531
" have: %s needs %s" % (substream_kind, keys))
4533
# No need to stream something we don't have
4535
if substream_kind == 'inventories':
4536
# Some missing keys are genuinely ghosts, filter those out.
4537
present = self.from_repository.inventories.get_parent_map(keys)
4538
revs = [key[0] for key in present]
4539
# Get the inventory stream more-or-less as we do for the
4540
# original stream; there's no reason to assume that records
4541
# direct from the source will be suitable for the sink. (Think
4542
# e.g. 2a -> 1.9-rich-root).
4543
for info in self._get_inventory_stream(revs, missing=True):
4547
# Ask for full texts always so that we don't need more round trips
4548
# after this stream.
4549
# Some of the missing keys are genuinely ghosts, so filter absent
4550
# records. The Sink is responsible for doing another check to
4551
# ensure that ghosts don't introduce missing data for future
4553
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4554
self.to_format._fetch_order, True))
4555
yield substream_kind, stream
4557
def inventory_fetch_order(self):
4558
if self._rich_root_upgrade():
4559
return 'topological'
4561
return self.to_format._fetch_order
4563
def _rich_root_upgrade(self):
4564
return (not self.from_repository._format.rich_root_data and
4565
self.to_format.rich_root_data)
4567
def _get_inventory_stream(self, revision_ids, missing=False):
4568
from_format = self.from_repository._format
4569
if (from_format.supports_chks and self.to_format.supports_chks and
4570
from_format.network_name() == self.to_format.network_name()):
4571
raise AssertionError(
4572
"this case should be handled by GroupCHKStreamSource")
4573
elif 'forceinvdeltas' in debug.debug_flags:
4574
return self._get_convertable_inventory_stream(revision_ids,
4575
delta_versus_null=missing)
4576
elif from_format.network_name() == self.to_format.network_name():
4578
return self._get_simple_inventory_stream(revision_ids,
4580
elif (not from_format.supports_chks and not self.to_format.supports_chks
4581
and from_format._serializer == self.to_format._serializer):
4582
# Essentially the same format.
4583
return self._get_simple_inventory_stream(revision_ids,
4586
# Any time we switch serializations, we want to use an
4587
# inventory-delta based approach.
4588
return self._get_convertable_inventory_stream(revision_ids,
4589
delta_versus_null=missing)
4591
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4592
# NB: This currently reopens the inventory weave in source;
4593
# using a single stream interface instead would avoid this.
4594
from_weave = self.from_repository.inventories
4596
delta_closure = True
4598
delta_closure = not self.delta_on_metadata()
4599
yield ('inventories', from_weave.get_record_stream(
4600
[(rev_id,) for rev_id in revision_ids],
4601
self.inventory_fetch_order(), delta_closure))
4603
def _get_convertable_inventory_stream(self, revision_ids,
4604
delta_versus_null=False):
4605
# The source is using CHKs, but the target either doesn't or it has a
4606
# different serializer. The StreamSink code expects to be able to
4607
# convert on the target, so we need to put bytes-on-the-wire that can
4608
# be converted. That means inventory deltas (if the remote is <1.19,
4609
# RemoteStreamSink will fallback to VFS to insert the deltas).
4610
yield ('inventory-deltas',
4611
self._stream_invs_as_deltas(revision_ids,
4612
delta_versus_null=delta_versus_null))
4614
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4615
"""Return a stream of inventory-deltas for the given rev ids.
4617
:param revision_ids: The list of inventories to transmit
4618
:param delta_versus_null: Don't try to find a minimal delta for this
4619
entry, instead compute the delta versus the NULL_REVISION. This
4620
effectively streams a complete inventory. Used for stuff like
4621
filling in missing parents, etc.
4623
from_repo = self.from_repository
4624
revision_keys = [(rev_id,) for rev_id in revision_ids]
4625
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4626
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4628
inventories = self.from_repository.iter_inventories(
4629
revision_ids, 'topological')
4630
format = from_repo._format
4631
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4632
inventory_cache = lru_cache.LRUCache(50)
4633
null_inventory = from_repo.revision_tree(
4634
_mod_revision.NULL_REVISION).inventory
4635
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4636
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4637
# repo back into a non-rich-root repo ought to be allowed)
4638
serializer = inventory_delta.InventoryDeltaSerializer(
4639
versioned_root=format.rich_root_data,
4640
tree_references=format.supports_tree_reference)
4641
for inv in inventories:
4642
key = (inv.revision_id,)
4643
parent_keys = parent_map.get(key, ())
4645
if not delta_versus_null and parent_keys:
4646
# The caller did not ask for complete inventories and we have
4647
# some parents that we can delta against. Make a delta against
4648
# each parent so that we can find the smallest.
4649
parent_ids = [parent_key[0] for parent_key in parent_keys]
4650
for parent_id in parent_ids:
4651
if parent_id not in invs_sent_so_far:
4652
# We don't know that the remote side has this basis, so
4655
if parent_id == _mod_revision.NULL_REVISION:
4656
parent_inv = null_inventory
4658
parent_inv = inventory_cache.get(parent_id, None)
4659
if parent_inv is None:
4660
parent_inv = from_repo.get_inventory(parent_id)
4661
candidate_delta = inv._make_delta(parent_inv)
4662
if (delta is None or
4663
len(delta) > len(candidate_delta)):
4664
delta = candidate_delta
4665
basis_id = parent_id
4667
# Either none of the parents ended up being suitable, or we
4668
# were asked to delta against NULL
4669
basis_id = _mod_revision.NULL_REVISION
4670
delta = inv._make_delta(null_inventory)
4671
invs_sent_so_far.add(inv.revision_id)
4672
inventory_cache[inv.revision_id] = inv
4673
delta_serialized = ''.join(
4674
serializer.delta_to_lines(basis_id, key[-1], delta))
4675
yield versionedfile.FulltextContentFactory(
4676
key, parent_keys, None, delta_serialized)
4679
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4680
stop_revision=None):
4681
"""Extend the partial history to include a given index
4683
If a stop_index is supplied, stop when that index has been reached.
4684
If a stop_revision is supplied, stop when that revision is
4685
encountered. Otherwise, stop when the beginning of history is
4688
:param stop_index: The index which should be present. When it is
4689
present, history extension will stop.
4690
:param stop_revision: The revision id which should be present. When
4691
it is encountered, history extension will stop.
4693
start_revision = partial_history_cache[-1]
4694
iterator = repo.iter_reverse_revision_history(start_revision)
4696
#skip the last revision in the list
4699
if (stop_index is not None and
4700
len(partial_history_cache) > stop_index):
4702
if partial_history_cache[-1] == stop_revision:
4704
revision_id = iterator.next()
4705
partial_history_cache.append(revision_id)
4706
except StopIteration:
added
removed
bzrlib/repository.py
