1
# Copyright (C) 2005, 2006, 2007, 2008, 2009 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
from bzrlib.lazy_import import lazy_import
18
lazy_import(globals(), """
39
revision as _mod_revision,
45
from bzrlib.bundle import serializer
46
from bzrlib.revisiontree import RevisionTree
47
from bzrlib.store.versioned import VersionedFileStore
48
from bzrlib.testament import Testament
51
from bzrlib.decorators import needs_read_lock, needs_write_lock
52
from bzrlib.inter import InterObject
53
from bzrlib.inventory import (
59
from bzrlib import registry
60
from bzrlib.symbol_versioning import (
63
from bzrlib.trace import (
64
log_exception_quietly, note, mutter, mutter_callsite, warning)
67
# Old formats display a warning, but only once
68
_deprecation_warning_done = False
71
class CommitBuilder(object):
72
"""Provides an interface to build up a commit.
74
This allows describing a tree to be committed without needing to
75
know the internals of the format of the repository.
78
# all clients should supply tree roots.
79
record_root_entry = True
80
# the default CommitBuilder does not manage trees whose root is versioned.
81
_versioned_root = False
83
def __init__(self, repository, parents, config, timestamp=None,
84
timezone=None, committer=None, revprops=None,
86
"""Initiate a CommitBuilder.
88
:param repository: Repository to commit to.
89
:param parents: Revision ids of the parents of the new revision.
90
:param config: Configuration to use.
91
:param timestamp: Optional timestamp recorded for commit.
92
:param timezone: Optional timezone for timestamp.
93
:param committer: Optional committer to set for commit.
94
:param revprops: Optional dictionary of revision properties.
95
:param revision_id: Optional revision id.
100
self._committer = self._config.username()
102
self._committer = committer
104
self.new_inventory = Inventory(None)
105
self._new_revision_id = revision_id
106
self.parents = parents
107
self.repository = repository
110
if revprops is not None:
111
self._validate_revprops(revprops)
112
self._revprops.update(revprops)
114
if timestamp is None:
115
timestamp = time.time()
116
# Restrict resolution to 1ms
117
self._timestamp = round(timestamp, 3)
120
self._timezone = osutils.local_time_offset()
122
self._timezone = int(timezone)
124
self._generate_revision_if_needed()
125
self.__heads = graph.HeadsCache(repository.get_graph()).heads
126
self._basis_delta = []
127
# API compatibility, older code that used CommitBuilder did not call
128
# .record_delete(), which means the delta that is computed would not be
129
# valid. Callers that will call record_delete() should call
130
# .will_record_deletes() to indicate that.
131
self._recording_deletes = False
132
# memo'd check for no-op commits.
133
self._any_changes = False
135
def any_changes(self):
136
"""Return True if any entries were changed.
138
This includes merge-only changes. It is the core for the --unchanged
141
:return: True if any changes have occured.
143
return self._any_changes
145
def _validate_unicode_text(self, text, context):
146
"""Verify things like commit messages don't have bogus characters."""
148
raise ValueError('Invalid value for %s: %r' % (context, text))
150
def _validate_revprops(self, revprops):
151
for key, value in revprops.iteritems():
152
# We know that the XML serializers do not round trip '\r'
153
# correctly, so refuse to accept them
154
if not isinstance(value, basestring):
155
raise ValueError('revision property (%s) is not a valid'
156
' (unicode) string: %r' % (key, value))
157
self._validate_unicode_text(value,
158
'revision property (%s)' % (key,))
160
def commit(self, message):
161
"""Make the actual commit.
163
:return: The revision id of the recorded revision.
165
self._validate_unicode_text(message, 'commit message')
166
rev = _mod_revision.Revision(
167
timestamp=self._timestamp,
168
timezone=self._timezone,
169
committer=self._committer,
171
inventory_sha1=self.inv_sha1,
172
revision_id=self._new_revision_id,
173
properties=self._revprops)
174
rev.parent_ids = self.parents
175
self.repository.add_revision(self._new_revision_id, rev,
176
self.new_inventory, self._config)
177
self.repository.commit_write_group()
178
return self._new_revision_id
181
"""Abort the commit that is being built.
183
self.repository.abort_write_group()
185
def revision_tree(self):
186
"""Return the tree that was just committed.
188
After calling commit() this can be called to get a RevisionTree
189
representing the newly committed tree. This is preferred to
190
calling Repository.revision_tree() because that may require
191
deserializing the inventory, while we already have a copy in
194
if self.new_inventory is None:
195
self.new_inventory = self.repository.get_inventory(
196
self._new_revision_id)
197
return RevisionTree(self.repository, self.new_inventory,
198
self._new_revision_id)
200
def finish_inventory(self):
201
"""Tell the builder that the inventory is finished.
203
:return: The inventory id in the repository, which can be used with
204
repository.get_inventory.
206
if self.new_inventory is None:
207
# an inventory delta was accumulated without creating a new
209
basis_id = self.basis_delta_revision
210
self.inv_sha1 = self.repository.add_inventory_by_delta(
211
basis_id, self._basis_delta, self._new_revision_id,
214
if self.new_inventory.root is None:
215
raise AssertionError('Root entry should be supplied to'
216
' record_entry_contents, as of bzr 0.10.')
217
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
218
self.new_inventory.revision_id = self._new_revision_id
219
self.inv_sha1 = self.repository.add_inventory(
220
self._new_revision_id,
224
return self._new_revision_id
226
def _gen_revision_id(self):
227
"""Return new revision-id."""
228
return generate_ids.gen_revision_id(self._config.username(),
231
def _generate_revision_if_needed(self):
232
"""Create a revision id if None was supplied.
234
If the repository can not support user-specified revision ids
235
they should override this function and raise CannotSetRevisionId
236
if _new_revision_id is not None.
238
:raises: CannotSetRevisionId
240
if self._new_revision_id is None:
241
self._new_revision_id = self._gen_revision_id()
242
self.random_revid = True
244
self.random_revid = False
246
def _heads(self, file_id, revision_ids):
247
"""Calculate the graph heads for revision_ids in the graph of file_id.
249
This can use either a per-file graph or a global revision graph as we
250
have an identity relationship between the two graphs.
252
return self.__heads(revision_ids)
254
def _check_root(self, ie, parent_invs, tree):
255
"""Helper for record_entry_contents.
257
:param ie: An entry being added.
258
:param parent_invs: The inventories of the parent revisions of the
260
:param tree: The tree that is being committed.
262
# In this revision format, root entries have no knit or weave When
263
# serializing out to disk and back in root.revision is always
265
ie.revision = self._new_revision_id
267
def _require_root_change(self, tree):
268
"""Enforce an appropriate root object change.
270
This is called once when record_iter_changes is called, if and only if
271
the root was not in the delta calculated by record_iter_changes.
273
:param tree: The tree which is being committed.
275
# NB: if there are no parents then this method is not called, so no
276
# need to guard on parents having length.
277
entry = entry_factory['directory'](tree.path2id(''), '',
279
entry.revision = self._new_revision_id
280
self._basis_delta.append(('', '', entry.file_id, entry))
282
def _get_delta(self, ie, basis_inv, path):
283
"""Get a delta against the basis inventory for ie."""
284
if ie.file_id not in basis_inv:
286
result = (None, path, ie.file_id, ie)
287
self._basis_delta.append(result)
289
elif ie != basis_inv[ie.file_id]:
291
# TODO: avoid tis id2path call.
292
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
293
self._basis_delta.append(result)
299
def get_basis_delta(self):
300
"""Return the complete inventory delta versus the basis inventory.
302
This has been built up with the calls to record_delete and
303
record_entry_contents. The client must have already called
304
will_record_deletes() to indicate that they will be generating a
307
:return: An inventory delta, suitable for use with apply_delta, or
308
Repository.add_inventory_by_delta, etc.
310
if not self._recording_deletes:
311
raise AssertionError("recording deletes not activated.")
312
return self._basis_delta
314
def record_delete(self, path, file_id):
315
"""Record that a delete occured against a basis tree.
317
This is an optional API - when used it adds items to the basis_delta
318
being accumulated by the commit builder. It cannot be called unless the
319
method will_record_deletes() has been called to inform the builder that
320
a delta is being supplied.
322
:param path: The path of the thing deleted.
323
:param file_id: The file id that was deleted.
325
if not self._recording_deletes:
326
raise AssertionError("recording deletes not activated.")
327
delta = (path, None, file_id, None)
328
self._basis_delta.append(delta)
329
self._any_changes = True
332
def will_record_deletes(self):
333
"""Tell the commit builder that deletes are being notified.
335
This enables the accumulation of an inventory delta; for the resulting
336
commit to be valid, deletes against the basis MUST be recorded via
337
builder.record_delete().
339
self._recording_deletes = True
341
basis_id = self.parents[0]
343
basis_id = _mod_revision.NULL_REVISION
344
self.basis_delta_revision = basis_id
346
def record_entry_contents(self, ie, parent_invs, path, tree,
348
"""Record the content of ie from tree into the commit if needed.
350
Side effect: sets ie.revision when unchanged
352
:param ie: An inventory entry present in the commit.
353
:param parent_invs: The inventories of the parent revisions of the
355
:param path: The path the entry is at in the tree.
356
:param tree: The tree which contains this entry and should be used to
358
:param content_summary: Summary data from the tree about the paths
359
content - stat, length, exec, sha/link target. This is only
360
accessed when the entry has a revision of None - that is when it is
361
a candidate to commit.
362
:return: A tuple (change_delta, version_recorded, fs_hash).
363
change_delta is an inventory_delta change for this entry against
364
the basis tree of the commit, or None if no change occured against
366
version_recorded is True if a new version of the entry has been
367
recorded. For instance, committing a merge where a file was only
368
changed on the other side will return (delta, False).
369
fs_hash is either None, or the hash details for the path (currently
370
a tuple of the contents sha1 and the statvalue returned by
371
tree.get_file_with_stat()).
373
if self.new_inventory.root is None:
374
if ie.parent_id is not None:
375
raise errors.RootMissing()
376
self._check_root(ie, parent_invs, tree)
377
if ie.revision is None:
378
kind = content_summary[0]
380
# ie is carried over from a prior commit
382
# XXX: repository specific check for nested tree support goes here - if
383
# the repo doesn't want nested trees we skip it ?
384
if (kind == 'tree-reference' and
385
not self.repository._format.supports_tree_reference):
386
# mismatch between commit builder logic and repository:
387
# this needs the entry creation pushed down into the builder.
388
raise NotImplementedError('Missing repository subtree support.')
389
self.new_inventory.add(ie)
391
# TODO: slow, take it out of the inner loop.
393
basis_inv = parent_invs[0]
395
basis_inv = Inventory(root_id=None)
397
# ie.revision is always None if the InventoryEntry is considered
398
# for committing. We may record the previous parents revision if the
399
# content is actually unchanged against a sole head.
400
if ie.revision is not None:
401
if not self._versioned_root and path == '':
402
# repositories that do not version the root set the root's
403
# revision to the new commit even when no change occurs (more
404
# specifically, they do not record a revision on the root; and
405
# the rev id is assigned to the root during deserialisation -
406
# this masks when a change may have occurred against the basis.
407
# To match this we always issue a delta, because the revision
408
# of the root will always be changing.
409
if ie.file_id in basis_inv:
410
delta = (basis_inv.id2path(ie.file_id), path,
414
delta = (None, path, ie.file_id, ie)
415
self._basis_delta.append(delta)
416
return delta, False, None
418
# we don't need to commit this, because the caller already
419
# determined that an existing revision of this file is
420
# appropriate. If its not being considered for committing then
421
# it and all its parents to the root must be unaltered so
422
# no-change against the basis.
423
if ie.revision == self._new_revision_id:
424
raise AssertionError("Impossible situation, a skipped "
425
"inventory entry (%r) claims to be modified in this "
426
"commit (%r).", (ie, self._new_revision_id))
427
return None, False, None
428
# XXX: Friction: parent_candidates should return a list not a dict
429
# so that we don't have to walk the inventories again.
430
parent_candiate_entries = ie.parent_candidates(parent_invs)
431
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
433
for inv in parent_invs:
434
if ie.file_id in inv:
435
old_rev = inv[ie.file_id].revision
436
if old_rev in head_set:
437
heads.append(inv[ie.file_id].revision)
438
head_set.remove(inv[ie.file_id].revision)
441
# now we check to see if we need to write a new record to the
443
# We write a new entry unless there is one head to the ancestors, and
444
# the kind-derived content is unchanged.
446
# Cheapest check first: no ancestors, or more the one head in the
447
# ancestors, we write a new node.
451
# There is a single head, look it up for comparison
452
parent_entry = parent_candiate_entries[heads[0]]
453
# if the non-content specific data has changed, we'll be writing a
455
if (parent_entry.parent_id != ie.parent_id or
456
parent_entry.name != ie.name):
458
# now we need to do content specific checks:
460
# if the kind changed the content obviously has
461
if kind != parent_entry.kind:
463
# Stat cache fingerprint feedback for the caller - None as we usually
464
# don't generate one.
467
if content_summary[2] is None:
468
raise ValueError("Files must not have executable = None")
470
if (# if the file length changed we have to store:
471
parent_entry.text_size != content_summary[1] or
472
# if the exec bit has changed we have to store:
473
parent_entry.executable != content_summary[2]):
475
elif parent_entry.text_sha1 == content_summary[3]:
476
# all meta and content is unchanged (using a hash cache
477
# hit to check the sha)
478
ie.revision = parent_entry.revision
479
ie.text_size = parent_entry.text_size
480
ie.text_sha1 = parent_entry.text_sha1
481
ie.executable = parent_entry.executable
482
return self._get_delta(ie, basis_inv, path), False, None
484
# Either there is only a hash change(no hash cache entry,
485
# or same size content change), or there is no change on
487
# Provide the parent's hash to the store layer, so that the
488
# content is unchanged we will not store a new node.
489
nostore_sha = parent_entry.text_sha1
491
# We want to record a new node regardless of the presence or
492
# absence of a content change in the file.
494
ie.executable = content_summary[2]
495
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
497
lines = file_obj.readlines()
501
ie.text_sha1, ie.text_size = self._add_text_to_weave(
502
ie.file_id, lines, heads, nostore_sha)
503
# Let the caller know we generated a stat fingerprint.
504
fingerprint = (ie.text_sha1, stat_value)
505
except errors.ExistingContent:
506
# Turns out that the file content was unchanged, and we were
507
# only going to store a new node if it was changed. Carry over
509
ie.revision = parent_entry.revision
510
ie.text_size = parent_entry.text_size
511
ie.text_sha1 = parent_entry.text_sha1
512
ie.executable = parent_entry.executable
513
return self._get_delta(ie, basis_inv, path), False, None
514
elif kind == 'directory':
516
# all data is meta here, nothing specific to directory, so
518
ie.revision = parent_entry.revision
519
return self._get_delta(ie, basis_inv, path), False, None
521
self._add_text_to_weave(ie.file_id, lines, heads, None)
522
elif kind == 'symlink':
523
current_link_target = content_summary[3]
525
# symlink target is not generic metadata, check if it has
527
if current_link_target != parent_entry.symlink_target:
530
# unchanged, carry over.
531
ie.revision = parent_entry.revision
532
ie.symlink_target = parent_entry.symlink_target
533
return self._get_delta(ie, basis_inv, path), False, None
534
ie.symlink_target = current_link_target
536
self._add_text_to_weave(ie.file_id, lines, heads, None)
537
elif kind == 'tree-reference':
539
if content_summary[3] != parent_entry.reference_revision:
542
# unchanged, carry over.
543
ie.reference_revision = parent_entry.reference_revision
544
ie.revision = parent_entry.revision
545
return self._get_delta(ie, basis_inv, path), False, None
546
ie.reference_revision = content_summary[3]
548
self._add_text_to_weave(ie.file_id, lines, heads, None)
550
raise NotImplementedError('unknown kind')
551
ie.revision = self._new_revision_id
552
self._any_changes = True
553
return self._get_delta(ie, basis_inv, path), True, fingerprint
555
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
556
_entry_factory=entry_factory):
557
"""Record a new tree via iter_changes.
559
:param tree: The tree to obtain text contents from for changed objects.
560
:param basis_revision_id: The revision id of the tree the iter_changes
561
has been generated against. Currently assumed to be the same
562
as self.parents[0] - if it is not, errors may occur.
563
:param iter_changes: An iter_changes iterator with the changes to apply
564
to basis_revision_id. The iterator must not include any items with
565
a current kind of None - missing items must be either filtered out
566
or errored-on beefore record_iter_changes sees the item.
567
:param _entry_factory: Private method to bind entry_factory locally for
569
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
572
# Create an inventory delta based on deltas between all the parents and
573
# deltas between all the parent inventories. We use inventory delta's
574
# between the inventory objects because iter_changes masks
575
# last-changed-field only changes.
577
# file_id -> change map, change is fileid, paths, changed, versioneds,
578
# parents, names, kinds, executables
580
# {file_id -> revision_id -> inventory entry, for entries in parent
581
# trees that are not parents[0]
585
revtrees = list(self.repository.revision_trees(self.parents))
586
except errors.NoSuchRevision:
587
# one or more ghosts, slow path.
589
for revision_id in self.parents:
591
revtrees.append(self.repository.revision_tree(revision_id))
592
except errors.NoSuchRevision:
594
basis_revision_id = _mod_revision.NULL_REVISION
596
revtrees.append(self.repository.revision_tree(
597
_mod_revision.NULL_REVISION))
598
# The basis inventory from a repository
600
basis_inv = revtrees[0].inventory
602
basis_inv = self.repository.revision_tree(
603
_mod_revision.NULL_REVISION).inventory
604
if len(self.parents) > 0:
605
if basis_revision_id != self.parents[0] and not ghost_basis:
607
"arbitrary basis parents not yet supported with merges")
608
for revtree in revtrees[1:]:
609
for change in revtree.inventory._make_delta(basis_inv):
610
if change[1] is None:
611
# Not present in this parent.
613
if change[2] not in merged_ids:
614
if change[0] is not None:
615
basis_entry = basis_inv[change[2]]
616
merged_ids[change[2]] = [
618
basis_entry.revision,
621
parent_entries[change[2]] = {
623
basis_entry.revision:basis_entry,
625
change[3].revision:change[3],
628
merged_ids[change[2]] = [change[3].revision]
629
parent_entries[change[2]] = {change[3].revision:change[3]}
631
merged_ids[change[2]].append(change[3].revision)
632
parent_entries[change[2]][change[3].revision] = change[3]
635
# Setup the changes from the tree:
636
# changes maps file_id -> (change, [parent revision_ids])
638
for change in iter_changes:
639
# This probably looks up in basis_inv way to much.
640
if change[1][0] is not None:
641
head_candidate = [basis_inv[change[0]].revision]
644
changes[change[0]] = change, merged_ids.get(change[0],
646
unchanged_merged = set(merged_ids) - set(changes)
647
# Extend the changes dict with synthetic changes to record merges of
649
for file_id in unchanged_merged:
650
# Record a merged version of these items that did not change vs the
651
# basis. This can be either identical parallel changes, or a revert
652
# of a specific file after a merge. The recorded content will be
653
# that of the current tree (which is the same as the basis), but
654
# the per-file graph will reflect a merge.
655
# NB:XXX: We are reconstructing path information we had, this
656
# should be preserved instead.
657
# inv delta change: (file_id, (path_in_source, path_in_target),
658
# changed_content, versioned, parent, name, kind,
661
basis_entry = basis_inv[file_id]
662
except errors.NoSuchId:
663
# a change from basis->some_parents but file_id isn't in basis
664
# so was new in the merge, which means it must have changed
665
# from basis -> current, and as it hasn't the add was reverted
666
# by the user. So we discard this change.
670
(basis_inv.id2path(file_id), tree.id2path(file_id)),
672
(basis_entry.parent_id, basis_entry.parent_id),
673
(basis_entry.name, basis_entry.name),
674
(basis_entry.kind, basis_entry.kind),
675
(basis_entry.executable, basis_entry.executable))
676
changes[file_id] = (change, merged_ids[file_id])
677
# changes contains tuples with the change and a set of inventory
678
# candidates for the file.
680
# old_path, new_path, file_id, new_inventory_entry
681
seen_root = False # Is the root in the basis delta?
682
inv_delta = self._basis_delta
683
modified_rev = self._new_revision_id
684
for change, head_candidates in changes.values():
685
if change[3][1]: # versioned in target.
686
# Several things may be happening here:
687
# We may have a fork in the per-file graph
688
# - record a change with the content from tree
689
# We may have a change against < all trees
690
# - carry over the tree that hasn't changed
691
# We may have a change against all trees
692
# - record the change with the content from tree
695
entry = _entry_factory[kind](file_id, change[5][1],
697
head_set = self._heads(change[0], set(head_candidates))
700
for head_candidate in head_candidates:
701
if head_candidate in head_set:
702
heads.append(head_candidate)
703
head_set.remove(head_candidate)
706
# Could be a carry-over situation:
707
parent_entry_revs = parent_entries.get(file_id, None)
708
if parent_entry_revs:
709
parent_entry = parent_entry_revs.get(heads[0], None)
712
if parent_entry is None:
713
# The parent iter_changes was called against is the one
714
# that is the per-file head, so any change is relevant
715
# iter_changes is valid.
716
carry_over_possible = False
718
# could be a carry over situation
719
# A change against the basis may just indicate a merge,
720
# we need to check the content against the source of the
721
# merge to determine if it was changed after the merge
723
if (parent_entry.kind != entry.kind or
724
parent_entry.parent_id != entry.parent_id or
725
parent_entry.name != entry.name):
726
# Metadata common to all entries has changed
727
# against per-file parent
728
carry_over_possible = False
730
carry_over_possible = True
731
# per-type checks for changes against the parent_entry
734
# Cannot be a carry-over situation
735
carry_over_possible = False
736
# Populate the entry in the delta
738
# XXX: There is still a small race here: If someone reverts the content of a file
739
# after iter_changes examines and decides it has changed,
740
# we will unconditionally record a new version even if some
741
# other process reverts it while commit is running (with
742
# the revert happening after iter_changes did it's
745
entry.executable = True
747
entry.executable = False
748
if (carry_over_possible and
749
parent_entry.executable == entry.executable):
750
# Check the file length, content hash after reading
752
nostore_sha = parent_entry.text_sha1
755
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
757
lines = file_obj.readlines()
761
entry.text_sha1, entry.text_size = self._add_text_to_weave(
762
file_id, lines, heads, nostore_sha)
763
yield file_id, change[1][1], (entry.text_sha1, stat_value)
764
except errors.ExistingContent:
765
# No content change against a carry_over parent
766
# Perhaps this should also yield a fs hash update?
768
entry.text_size = parent_entry.text_size
769
entry.text_sha1 = parent_entry.text_sha1
770
elif kind == 'symlink':
772
entry.symlink_target = tree.get_symlink_target(file_id)
773
if (carry_over_possible and
774
parent_entry.symlink_target == entry.symlink_target):
777
self._add_text_to_weave(change[0], [], heads, None)
778
elif kind == 'directory':
779
if carry_over_possible:
782
# Nothing to set on the entry.
783
# XXX: split into the Root and nonRoot versions.
784
if change[1][1] != '' or self.repository.supports_rich_root():
785
self._add_text_to_weave(change[0], [], heads, None)
786
elif kind == 'tree-reference':
787
if not self.repository._format.supports_tree_reference:
788
# This isn't quite sane as an error, but we shouldn't
789
# ever see this code path in practice: tree's don't
790
# permit references when the repo doesn't support tree
792
raise errors.UnsupportedOperation(tree.add_reference,
794
entry.reference_revision = \
795
tree.get_reference_revision(change[0])
796
if (carry_over_possible and
797
parent_entry.reference_revision == reference_revision):
800
self._add_text_to_weave(change[0], [], heads, None)
802
raise AssertionError('unknown kind %r' % kind)
804
entry.revision = modified_rev
806
entry.revision = parent_entry.revision
809
new_path = change[1][1]
810
inv_delta.append((change[1][0], new_path, change[0], entry))
813
self.new_inventory = None
815
self._any_changes = True
817
# housekeeping root entry changes do not affect no-change commits.
818
self._require_root_change(tree)
819
self.basis_delta_revision = basis_revision_id
821
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
822
# Note: as we read the content directly from the tree, we know its not
823
# been turned into unicode or badly split - but a broken tree
824
# implementation could give us bad output from readlines() so this is
825
# not a guarantee of safety. What would be better is always checking
826
# the content during test suite execution. RBC 20070912
827
parent_keys = tuple((file_id, parent) for parent in parents)
828
return self.repository.texts.add_lines(
829
(file_id, self._new_revision_id), parent_keys, new_lines,
830
nostore_sha=nostore_sha, random_id=self.random_revid,
831
check_content=False)[0:2]
834
class RootCommitBuilder(CommitBuilder):
835
"""This commitbuilder actually records the root id"""
837
# the root entry gets versioned properly by this builder.
838
_versioned_root = True
840
def _check_root(self, ie, parent_invs, tree):
841
"""Helper for record_entry_contents.
843
:param ie: An entry being added.
844
:param parent_invs: The inventories of the parent revisions of the
846
:param tree: The tree that is being committed.
849
def _require_root_change(self, tree):
850
"""Enforce an appropriate root object change.
852
This is called once when record_iter_changes is called, if and only if
853
the root was not in the delta calculated by record_iter_changes.
855
:param tree: The tree which is being committed.
857
# versioned roots do not change unless the tree found a change.
860
######################################################################
863
class Repository(object):
864
"""Repository holding history for one or more branches.
866
The repository holds and retrieves historical information including
867
revisions and file history. It's normally accessed only by the Branch,
868
which views a particular line of development through that history.
870
The Repository builds on top of some byte storage facilies (the revisions,
871
signatures, inventories, texts and chk_bytes attributes) and a Transport,
872
which respectively provide byte storage and a means to access the (possibly
875
The byte storage facilities are addressed via tuples, which we refer to
876
as 'keys' throughout the code base. Revision_keys, inventory_keys and
877
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
878
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
879
byte string made up of a hash identifier and a hash value.
880
We use this interface because it allows low friction with the underlying
881
code that implements disk indices, network encoding and other parts of
884
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
885
the serialised revisions for the repository. This can be used to obtain
886
revision graph information or to access raw serialised revisions.
887
The result of trying to insert data into the repository via this store
888
is undefined: it should be considered read-only except for implementors
890
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
891
the serialised signatures for the repository. This can be used to
892
obtain access to raw serialised signatures. The result of trying to
893
insert data into the repository via this store is undefined: it should
894
be considered read-only except for implementors of repositories.
895
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
896
the serialised inventories for the repository. This can be used to
897
obtain unserialised inventories. The result of trying to insert data
898
into the repository via this store is undefined: it should be
899
considered read-only except for implementors of repositories.
900
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
901
texts of files and directories for the repository. This can be used to
902
obtain file texts or file graphs. Note that Repository.iter_file_bytes
903
is usually a better interface for accessing file texts.
904
The result of trying to insert data into the repository via this store
905
is undefined: it should be considered read-only except for implementors
907
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
908
any data the repository chooses to store or have indexed by its hash.
909
The result of trying to insert data into the repository via this store
910
is undefined: it should be considered read-only except for implementors
912
:ivar _transport: Transport for file access to repository, typically
913
pointing to .bzr/repository.
916
# What class to use for a CommitBuilder. Often its simpler to change this
917
# in a Repository class subclass rather than to override
918
# get_commit_builder.
919
_commit_builder_class = CommitBuilder
920
# The search regex used by xml based repositories to determine what things
921
# where changed in a single commit.
922
_file_ids_altered_regex = lazy_regex.lazy_compile(
923
r'file_id="(?P<file_id>[^"]+)"'
924
r'.* revision="(?P<revision_id>[^"]+)"'
927
def abort_write_group(self, suppress_errors=False):
928
"""Commit the contents accrued within the current write group.
930
:param suppress_errors: if true, abort_write_group will catch and log
931
unexpected errors that happen during the abort, rather than
932
allowing them to propagate. Defaults to False.
934
:seealso: start_write_group.
936
if self._write_group is not self.get_transaction():
937
# has an unlock or relock occured ?
938
raise errors.BzrError(
939
'mismatched lock context and write group. %r, %r' %
940
(self._write_group, self.get_transaction()))
942
self._abort_write_group()
943
except Exception, exc:
944
self._write_group = None
945
if not suppress_errors:
947
mutter('abort_write_group failed')
948
log_exception_quietly()
949
note('bzr: ERROR (ignored): %s', exc)
950
self._write_group = None
952
def _abort_write_group(self):
953
"""Template method for per-repository write group cleanup.
955
This is called during abort before the write group is considered to be
956
finished and should cleanup any internal state accrued during the write
957
group. There is no requirement that data handed to the repository be
958
*not* made available - this is not a rollback - but neither should any
959
attempt be made to ensure that data added is fully commited. Abort is
960
invoked when an error has occured so futher disk or network operations
961
may not be possible or may error and if possible should not be
965
def add_fallback_repository(self, repository):
966
"""Add a repository to use for looking up data not held locally.
968
:param repository: A repository.
970
if not self._format.supports_external_lookups:
971
raise errors.UnstackableRepositoryFormat(self._format, self.base)
973
# This repository will call fallback.unlock() when we transition to
974
# the unlocked state, so we make sure to increment the lock count
975
repository.lock_read()
976
self._check_fallback_repository(repository)
977
self._fallback_repositories.append(repository)
978
self.texts.add_fallback_versioned_files(repository.texts)
979
self.inventories.add_fallback_versioned_files(repository.inventories)
980
self.revisions.add_fallback_versioned_files(repository.revisions)
981
self.signatures.add_fallback_versioned_files(repository.signatures)
982
if self.chk_bytes is not None:
983
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
985
def _check_fallback_repository(self, repository):
986
"""Check that this repository can fallback to repository safely.
988
Raise an error if not.
990
:param repository: A repository to fallback to.
992
return InterRepository._assert_same_model(self, repository)
994
def add_inventory(self, revision_id, inv, parents):
995
"""Add the inventory inv to the repository as revision_id.
997
:param parents: The revision ids of the parents that revision_id
998
is known to have and are in the repository already.
1000
:returns: The validator(which is a sha1 digest, though what is sha'd is
1001
repository format specific) of the serialized inventory.
1003
if not self.is_in_write_group():
1004
raise AssertionError("%r not in write group" % (self,))
1005
_mod_revision.check_not_reserved_id(revision_id)
1006
if not (inv.revision_id is None or inv.revision_id == revision_id):
1007
raise AssertionError(
1008
"Mismatch between inventory revision"
1009
" id and insertion revid (%r, %r)"
1010
% (inv.revision_id, revision_id))
1011
if inv.root is None:
1012
raise AssertionError()
1013
return self._add_inventory_checked(revision_id, inv, parents)
1015
def _add_inventory_checked(self, revision_id, inv, parents):
1016
"""Add inv to the repository after checking the inputs.
1018
This function can be overridden to allow different inventory styles.
1020
:seealso: add_inventory, for the contract.
1022
inv_lines = self._serialise_inventory_to_lines(inv)
1023
return self._inventory_add_lines(revision_id, parents,
1024
inv_lines, check_content=False)
1026
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1027
parents, basis_inv=None, propagate_caches=False):
1028
"""Add a new inventory expressed as a delta against another revision.
1030
:param basis_revision_id: The inventory id the delta was created
1031
against. (This does not have to be a direct parent.)
1032
:param delta: The inventory delta (see Inventory.apply_delta for
1034
:param new_revision_id: The revision id that the inventory is being
1036
:param parents: The revision ids of the parents that revision_id is
1037
known to have and are in the repository already. These are supplied
1038
for repositories that depend on the inventory graph for revision
1039
graph access, as well as for those that pun ancestry with delta
1041
:param basis_inv: The basis inventory if it is already known,
1043
:param propagate_caches: If True, the caches for this inventory are
1044
copied to and updated for the result if possible.
1046
:returns: (validator, new_inv)
1047
The validator(which is a sha1 digest, though what is sha'd is
1048
repository format specific) of the serialized inventory, and the
1049
resulting inventory.
1051
if not self.is_in_write_group():
1052
raise AssertionError("%r not in write group" % (self,))
1053
_mod_revision.check_not_reserved_id(new_revision_id)
1054
basis_tree = self.revision_tree(basis_revision_id)
1055
basis_tree.lock_read()
1057
# Note that this mutates the inventory of basis_tree, which not all
1058
# inventory implementations may support: A better idiom would be to
1059
# return a new inventory, but as there is no revision tree cache in
1060
# repository this is safe for now - RBC 20081013
1061
if basis_inv is None:
1062
basis_inv = basis_tree.inventory
1063
basis_inv.apply_delta(delta)
1064
basis_inv.revision_id = new_revision_id
1065
return (self.add_inventory(new_revision_id, basis_inv, parents),
1070
def _inventory_add_lines(self, revision_id, parents, lines,
1071
check_content=True):
1072
"""Store lines in inv_vf and return the sha1 of the inventory."""
1073
parents = [(parent,) for parent in parents]
1074
return self.inventories.add_lines((revision_id,), parents, lines,
1075
check_content=check_content)[0]
1077
def add_revision(self, revision_id, rev, inv=None, config=None):
1078
"""Add rev to the revision store as revision_id.
1080
:param revision_id: the revision id to use.
1081
:param rev: The revision object.
1082
:param inv: The inventory for the revision. if None, it will be looked
1083
up in the inventory storer
1084
:param config: If None no digital signature will be created.
1085
If supplied its signature_needed method will be used
1086
to determine if a signature should be made.
1088
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1090
_mod_revision.check_not_reserved_id(revision_id)
1091
if config is not None and config.signature_needed():
1093
inv = self.get_inventory(revision_id)
1094
plaintext = Testament(rev, inv).as_short_text()
1095
self.store_revision_signature(
1096
gpg.GPGStrategy(config), plaintext, revision_id)
1097
# check inventory present
1098
if not self.inventories.get_parent_map([(revision_id,)]):
1100
raise errors.WeaveRevisionNotPresent(revision_id,
1103
# yes, this is not suitable for adding with ghosts.
1104
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1107
key = (revision_id,)
1108
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1109
self._add_revision(rev)
1111
def _add_revision(self, revision):
1112
text = self._serializer.write_revision_to_string(revision)
1113
key = (revision.revision_id,)
1114
parents = tuple((parent,) for parent in revision.parent_ids)
1115
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1117
def all_revision_ids(self):
1118
"""Returns a list of all the revision ids in the repository.
1120
This is conceptually deprecated because code should generally work on
1121
the graph reachable from a particular revision, and ignore any other
1122
revisions that might be present. There is no direct replacement
1125
if 'evil' in debug.debug_flags:
1126
mutter_callsite(2, "all_revision_ids is linear with history.")
1127
return self._all_revision_ids()
1129
def _all_revision_ids(self):
1130
"""Returns a list of all the revision ids in the repository.
1132
These are in as much topological order as the underlying store can
1135
raise NotImplementedError(self._all_revision_ids)
1137
def break_lock(self):
1138
"""Break a lock if one is present from another instance.
1140
Uses the ui factory to ask for confirmation if the lock may be from
1143
self.control_files.break_lock()
1146
def _eliminate_revisions_not_present(self, revision_ids):
1147
"""Check every revision id in revision_ids to see if we have it.
1149
Returns a set of the present revisions.
1152
graph = self.get_graph()
1153
parent_map = graph.get_parent_map(revision_ids)
1154
# The old API returned a list, should this actually be a set?
1155
return parent_map.keys()
1158
def create(a_bzrdir):
1159
"""Construct the current default format repository in a_bzrdir."""
1160
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1162
def __init__(self, _format, a_bzrdir, control_files):
1163
"""instantiate a Repository.
1165
:param _format: The format of the repository on disk.
1166
:param a_bzrdir: The BzrDir of the repository.
1168
In the future we will have a single api for all stores for
1169
getting file texts, inventories and revisions, then
1170
this construct will accept instances of those things.
1172
super(Repository, self).__init__()
1173
self._format = _format
1174
# the following are part of the public API for Repository:
1175
self.bzrdir = a_bzrdir
1176
self.control_files = control_files
1177
self._transport = control_files._transport
1178
self.base = self._transport.base
1180
self._reconcile_does_inventory_gc = True
1181
self._reconcile_fixes_text_parents = False
1182
self._reconcile_backsup_inventory = True
1183
# not right yet - should be more semantically clear ?
1185
# TODO: make sure to construct the right store classes, etc, depending
1186
# on whether escaping is required.
1187
self._warn_if_deprecated()
1188
self._write_group = None
1189
# Additional places to query for data.
1190
self._fallback_repositories = []
1191
# An InventoryEntry cache, used during deserialization
1192
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1195
return '%s(%r)' % (self.__class__.__name__,
1198
def has_same_location(self, other):
1199
"""Returns a boolean indicating if this repository is at the same
1200
location as another repository.
1202
This might return False even when two repository objects are accessing
1203
the same physical repository via different URLs.
1205
if self.__class__ is not other.__class__:
1207
return (self._transport.base == other._transport.base)
1209
def is_in_write_group(self):
1210
"""Return True if there is an open write group.
1212
:seealso: start_write_group.
1214
return self._write_group is not None
1216
def is_locked(self):
1217
return self.control_files.is_locked()
1219
def is_write_locked(self):
1220
"""Return True if this object is write locked."""
1221
return self.is_locked() and self.control_files._lock_mode == 'w'
1223
def lock_write(self, token=None):
1224
"""Lock this repository for writing.
1226
This causes caching within the repository obejct to start accumlating
1227
data during reads, and allows a 'write_group' to be obtained. Write
1228
groups must be used for actual data insertion.
1230
:param token: if this is already locked, then lock_write will fail
1231
unless the token matches the existing lock.
1232
:returns: a token if this instance supports tokens, otherwise None.
1233
:raises TokenLockingNotSupported: when a token is given but this
1234
instance doesn't support using token locks.
1235
:raises MismatchedToken: if the specified token doesn't match the token
1236
of the existing lock.
1237
:seealso: start_write_group.
1239
A token should be passed in if you know that you have locked the object
1240
some other way, and need to synchronise this object's state with that
1243
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1245
locked = self.is_locked()
1246
result = self.control_files.lock_write(token=token)
1248
for repo in self._fallback_repositories:
1249
# Writes don't affect fallback repos
1251
self._refresh_data()
1254
def lock_read(self):
1255
locked = self.is_locked()
1256
self.control_files.lock_read()
1258
for repo in self._fallback_repositories:
1260
self._refresh_data()
1262
def get_physical_lock_status(self):
1263
return self.control_files.get_physical_lock_status()
1265
def leave_lock_in_place(self):
1266
"""Tell this repository not to release the physical lock when this
1269
If lock_write doesn't return a token, then this method is not supported.
1271
self.control_files.leave_in_place()
1273
def dont_leave_lock_in_place(self):
1274
"""Tell this repository to release the physical lock when this
1275
object is unlocked, even if it didn't originally acquire it.
1277
If lock_write doesn't return a token, then this method is not supported.
1279
self.control_files.dont_leave_in_place()
1282
def gather_stats(self, revid=None, committers=None):
1283
"""Gather statistics from a revision id.
1285
:param revid: The revision id to gather statistics from, if None, then
1286
no revision specific statistics are gathered.
1287
:param committers: Optional parameter controlling whether to grab
1288
a count of committers from the revision specific statistics.
1289
:return: A dictionary of statistics. Currently this contains:
1290
committers: The number of committers if requested.
1291
firstrev: A tuple with timestamp, timezone for the penultimate left
1292
most ancestor of revid, if revid is not the NULL_REVISION.
1293
latestrev: A tuple with timestamp, timezone for revid, if revid is
1294
not the NULL_REVISION.
1295
revisions: The total revision count in the repository.
1296
size: An estimate disk size of the repository in bytes.
1299
if revid and committers:
1300
result['committers'] = 0
1301
if revid and revid != _mod_revision.NULL_REVISION:
1303
all_committers = set()
1304
revisions = self.get_ancestry(revid)
1305
# pop the leading None
1307
first_revision = None
1309
# ignore the revisions in the middle - just grab first and last
1310
revisions = revisions[0], revisions[-1]
1311
for revision in self.get_revisions(revisions):
1312
if not first_revision:
1313
first_revision = revision
1315
all_committers.add(revision.committer)
1316
last_revision = revision
1318
result['committers'] = len(all_committers)
1319
result['firstrev'] = (first_revision.timestamp,
1320
first_revision.timezone)
1321
result['latestrev'] = (last_revision.timestamp,
1322
last_revision.timezone)
1324
# now gather global repository information
1325
# XXX: This is available for many repos regardless of listability.
1326
if self.bzrdir.root_transport.listable():
1327
# XXX: do we want to __define len__() ?
1328
# Maybe the versionedfiles object should provide a different
1329
# method to get the number of keys.
1330
result['revisions'] = len(self.revisions.keys())
1331
# result['size'] = t
1334
def find_branches(self, using=False):
1335
"""Find branches underneath this repository.
1337
This will include branches inside other branches.
1339
:param using: If True, list only branches using this repository.
1341
if using and not self.is_shared():
1343
return [self.bzrdir.open_branch()]
1344
except errors.NotBranchError:
1346
class Evaluator(object):
1349
self.first_call = True
1351
def __call__(self, bzrdir):
1352
# On the first call, the parameter is always the bzrdir
1353
# containing the current repo.
1354
if not self.first_call:
1356
repository = bzrdir.open_repository()
1357
except errors.NoRepositoryPresent:
1360
return False, (None, repository)
1361
self.first_call = False
1363
value = (bzrdir.open_branch(), None)
1364
except errors.NotBranchError:
1365
value = (None, None)
1369
for branch, repository in bzrdir.BzrDir.find_bzrdirs(
1370
self.bzrdir.root_transport, evaluate=Evaluator()):
1371
if branch is not None:
1372
branches.append(branch)
1373
if not using and repository is not None:
1374
branches.extend(repository.find_branches())
1378
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
1379
"""Return the revision ids that other has that this does not.
1381
These are returned in topological order.
1383
revision_id: only return revision ids included by revision_id.
1385
return InterRepository.get(other, self).search_missing_revision_ids(
1386
revision_id, find_ghosts)
1390
"""Open the repository rooted at base.
1392
For instance, if the repository is at URL/.bzr/repository,
1393
Repository.open(URL) -> a Repository instance.
1395
control = bzrdir.BzrDir.open(base)
1396
return control.open_repository()
1398
def copy_content_into(self, destination, revision_id=None):
1399
"""Make a complete copy of the content in self into destination.
1401
This is a destructive operation! Do not use it on existing
1404
return InterRepository.get(self, destination).copy_content(revision_id)
1406
def commit_write_group(self):
1407
"""Commit the contents accrued within the current write group.
1409
:seealso: start_write_group.
1411
if self._write_group is not self.get_transaction():
1412
# has an unlock or relock occured ?
1413
raise errors.BzrError('mismatched lock context %r and '
1415
(self.get_transaction(), self._write_group))
1416
self._commit_write_group()
1417
self._write_group = None
1419
def _commit_write_group(self):
1420
"""Template method for per-repository write group cleanup.
1422
This is called before the write group is considered to be
1423
finished and should ensure that all data handed to the repository
1424
for writing during the write group is safely committed (to the
1425
extent possible considering file system caching etc).
1428
def suspend_write_group(self):
1429
raise errors.UnsuspendableWriteGroup(self)
1431
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1432
"""Return the keys of missing inventory parents for revisions added in
1435
A revision is not complete if the inventory delta for that revision
1436
cannot be calculated. Therefore if the parent inventories of a
1437
revision are not present, the revision is incomplete, and e.g. cannot
1438
be streamed by a smart server. This method finds missing inventory
1439
parents for revisions added in this write group.
1441
if not self._format.supports_external_lookups:
1442
# This is only an issue for stacked repositories
1444
if not self.is_in_write_group():
1445
raise AssertionError('not in a write group')
1447
# XXX: We assume that every added revision already has its
1448
# corresponding inventory, so we only check for parent inventories that
1449
# might be missing, rather than all inventories.
1450
parents = set(self.revisions._index.get_missing_parents())
1451
parents.discard(_mod_revision.NULL_REVISION)
1452
unstacked_inventories = self.inventories._index
1453
present_inventories = unstacked_inventories.get_parent_map(
1454
key[-1:] for key in parents)
1455
parents.difference_update(present_inventories)
1456
if len(parents) == 0:
1457
# No missing parent inventories.
1459
if not check_for_missing_texts:
1460
return set(('inventories', rev_id) for (rev_id,) in parents)
1461
# Ok, now we have a list of missing inventories. But these only matter
1462
# if the inventories that reference them are missing some texts they
1463
# appear to introduce.
1464
# XXX: Texts referenced by all added inventories need to be present,
1465
# but at the moment we're only checking for texts referenced by
1466
# inventories at the graph's edge.
1467
key_deps = self.revisions._index._key_dependencies
1468
key_deps.add_keys(present_inventories)
1469
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1470
file_ids = self.fileids_altered_by_revision_ids(referrers)
1471
missing_texts = set()
1472
for file_id, version_ids in file_ids.iteritems():
1473
missing_texts.update(
1474
(file_id, version_id) for version_id in version_ids)
1475
present_texts = self.texts.get_parent_map(missing_texts)
1476
missing_texts.difference_update(present_texts)
1477
if not missing_texts:
1478
# No texts are missing, so all revisions and their deltas are
1481
# Alternatively the text versions could be returned as the missing
1482
# keys, but this is likely to be less data.
1483
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1486
def refresh_data(self):
1487
"""Re-read any data needed to to synchronise with disk.
1489
This method is intended to be called after another repository instance
1490
(such as one used by a smart server) has inserted data into the
1491
repository. It may not be called during a write group, but may be
1492
called at any other time.
1494
if self.is_in_write_group():
1495
raise errors.InternalBzrError(
1496
"May not refresh_data while in a write group.")
1497
self._refresh_data()
1499
def resume_write_group(self, tokens):
1500
if not self.is_write_locked():
1501
raise errors.NotWriteLocked(self)
1502
if self._write_group:
1503
raise errors.BzrError('already in a write group')
1504
self._resume_write_group(tokens)
1505
# so we can detect unlock/relock - the write group is now entered.
1506
self._write_group = self.get_transaction()
1508
def _resume_write_group(self, tokens):
1509
raise errors.UnsuspendableWriteGroup(self)
1511
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1513
"""Fetch the content required to construct revision_id from source.
1515
If revision_id is None and fetch_spec is None, then all content is
1518
fetch() may not be used when the repository is in a write group -
1519
either finish the current write group before using fetch, or use
1520
fetch before starting the write group.
1522
:param find_ghosts: Find and copy revisions in the source that are
1523
ghosts in the target (and not reachable directly by walking out to
1524
the first-present revision in target from revision_id).
1525
:param revision_id: If specified, all the content needed for this
1526
revision ID will be copied to the target. Fetch will determine for
1527
itself which content needs to be copied.
1528
:param fetch_spec: If specified, a SearchResult or
1529
PendingAncestryResult that describes which revisions to copy. This
1530
allows copying multiple heads at once. Mutually exclusive with
1533
if fetch_spec is not None and revision_id is not None:
1534
raise AssertionError(
1535
"fetch_spec and revision_id are mutually exclusive.")
1536
if self.is_in_write_group():
1537
raise errors.InternalBzrError(
1538
"May not fetch while in a write group.")
1539
# fast path same-url fetch operations
1540
if self.has_same_location(source) and fetch_spec is None:
1541
# check that last_revision is in 'from' and then return a
1543
if (revision_id is not None and
1544
not _mod_revision.is_null(revision_id)):
1545
self.get_revision(revision_id)
1547
# if there is no specific appropriate InterRepository, this will get
1548
# the InterRepository base class, which raises an
1549
# IncompatibleRepositories when asked to fetch.
1550
inter = InterRepository.get(source, self)
1551
return inter.fetch(revision_id=revision_id, pb=pb,
1552
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1554
def create_bundle(self, target, base, fileobj, format=None):
1555
return serializer.write_bundle(self, target, base, fileobj, format)
1557
def get_commit_builder(self, branch, parents, config, timestamp=None,
1558
timezone=None, committer=None, revprops=None,
1560
"""Obtain a CommitBuilder for this repository.
1562
:param branch: Branch to commit to.
1563
:param parents: Revision ids of the parents of the new revision.
1564
:param config: Configuration to use.
1565
:param timestamp: Optional timestamp recorded for commit.
1566
:param timezone: Optional timezone for timestamp.
1567
:param committer: Optional committer to set for commit.
1568
:param revprops: Optional dictionary of revision properties.
1569
:param revision_id: Optional revision id.
1571
result = self._commit_builder_class(self, parents, config,
1572
timestamp, timezone, committer, revprops, revision_id)
1573
self.start_write_group()
1577
if (self.control_files._lock_count == 1 and
1578
self.control_files._lock_mode == 'w'):
1579
if self._write_group is not None:
1580
self.abort_write_group()
1581
self.control_files.unlock()
1582
raise errors.BzrError(
1583
'Must end write groups before releasing write locks.')
1584
self.control_files.unlock()
1585
if self.control_files._lock_count == 0:
1586
self._inventory_entry_cache.clear()
1587
for repo in self._fallback_repositories:
1591
def clone(self, a_bzrdir, revision_id=None):
1592
"""Clone this repository into a_bzrdir using the current format.
1594
Currently no check is made that the format of this repository and
1595
the bzrdir format are compatible. FIXME RBC 20060201.
1597
:return: The newly created destination repository.
1599
# TODO: deprecate after 0.16; cloning this with all its settings is
1600
# probably not very useful -- mbp 20070423
1601
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1602
self.copy_content_into(dest_repo, revision_id)
1605
def start_write_group(self):
1606
"""Start a write group in the repository.
1608
Write groups are used by repositories which do not have a 1:1 mapping
1609
between file ids and backend store to manage the insertion of data from
1610
both fetch and commit operations.
1612
A write lock is required around the start_write_group/commit_write_group
1613
for the support of lock-requiring repository formats.
1615
One can only insert data into a repository inside a write group.
1619
if not self.is_write_locked():
1620
raise errors.NotWriteLocked(self)
1621
if self._write_group:
1622
raise errors.BzrError('already in a write group')
1623
self._start_write_group()
1624
# so we can detect unlock/relock - the write group is now entered.
1625
self._write_group = self.get_transaction()
1627
def _start_write_group(self):
1628
"""Template method for per-repository write group startup.
1630
This is called before the write group is considered to be
1635
def sprout(self, to_bzrdir, revision_id=None):
1636
"""Create a descendent repository for new development.
1638
Unlike clone, this does not copy the settings of the repository.
1640
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1641
dest_repo.fetch(self, revision_id=revision_id)
1644
def _create_sprouting_repo(self, a_bzrdir, shared):
1645
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1646
# use target default format.
1647
dest_repo = a_bzrdir.create_repository()
1649
# Most control formats need the repository to be specifically
1650
# created, but on some old all-in-one formats it's not needed
1652
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1653
except errors.UninitializableFormat:
1654
dest_repo = a_bzrdir.open_repository()
1657
def _get_sink(self):
1658
"""Return a sink for streaming into this repository."""
1659
return StreamSink(self)
1661
def _get_source(self, to_format):
1662
"""Return a source for streaming from this repository."""
1663
return StreamSource(self, to_format)
1666
def has_revision(self, revision_id):
1667
"""True if this repository has a copy of the revision."""
1668
return revision_id in self.has_revisions((revision_id,))
1671
def has_revisions(self, revision_ids):
1672
"""Probe to find out the presence of multiple revisions.
1674
:param revision_ids: An iterable of revision_ids.
1675
:return: A set of the revision_ids that were present.
1677
parent_map = self.revisions.get_parent_map(
1678
[(rev_id,) for rev_id in revision_ids])
1680
if _mod_revision.NULL_REVISION in revision_ids:
1681
result.add(_mod_revision.NULL_REVISION)
1682
result.update([key[0] for key in parent_map])
1686
def get_revision(self, revision_id):
1687
"""Return the Revision object for a named revision."""
1688
return self.get_revisions([revision_id])[0]
1691
def get_revision_reconcile(self, revision_id):
1692
"""'reconcile' helper routine that allows access to a revision always.
1694
This variant of get_revision does not cross check the weave graph
1695
against the revision one as get_revision does: but it should only
1696
be used by reconcile, or reconcile-alike commands that are correcting
1697
or testing the revision graph.
1699
return self._get_revisions([revision_id])[0]
1702
def get_revisions(self, revision_ids):
1703
"""Get many revisions at once."""
1704
return self._get_revisions(revision_ids)
1707
def _get_revisions(self, revision_ids):
1708
"""Core work logic to get many revisions without sanity checks."""
1709
for rev_id in revision_ids:
1710
if not rev_id or not isinstance(rev_id, basestring):
1711
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1712
keys = [(key,) for key in revision_ids]
1713
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1715
for record in stream:
1716
if record.storage_kind == 'absent':
1717
raise errors.NoSuchRevision(self, record.key[0])
1718
text = record.get_bytes_as('fulltext')
1719
rev = self._serializer.read_revision_from_string(text)
1720
revs[record.key[0]] = rev
1721
return [revs[revid] for revid in revision_ids]
1724
def get_revision_xml(self, revision_id):
1725
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1726
# would have already do it.
1727
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1728
# TODO: this can't just be replaced by:
1729
# return self._serializer.write_revision_to_string(
1730
# self.get_revision(revision_id))
1731
# as cStringIO preservers the encoding unlike write_revision_to_string
1732
# or some other call down the path.
1733
rev = self.get_revision(revision_id)
1734
rev_tmp = cStringIO.StringIO()
1735
# the current serializer..
1736
self._serializer.write_revision(rev, rev_tmp)
1738
return rev_tmp.getvalue()
1740
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1741
"""Produce a generator of revision deltas.
1743
Note that the input is a sequence of REVISIONS, not revision_ids.
1744
Trees will be held in memory until the generator exits.
1745
Each delta is relative to the revision's lefthand predecessor.
1747
:param specific_fileids: if not None, the result is filtered
1748
so that only those file-ids, their parents and their
1749
children are included.
1751
# Get the revision-ids of interest
1752
required_trees = set()
1753
for revision in revisions:
1754
required_trees.add(revision.revision_id)
1755
required_trees.update(revision.parent_ids[:1])
1757
# Get the matching filtered trees. Note that it's more
1758
# efficient to pass filtered trees to changes_from() rather
1759
# than doing the filtering afterwards. changes_from() could
1760
# arguably do the filtering itself but it's path-based, not
1761
# file-id based, so filtering before or afterwards is
1763
if specific_fileids is None:
1764
trees = dict((t.get_revision_id(), t) for
1765
t in self.revision_trees(required_trees))
1767
trees = dict((t.get_revision_id(), t) for
1768
t in self._filtered_revision_trees(required_trees,
1771
# Calculate the deltas
1772
for revision in revisions:
1773
if not revision.parent_ids:
1774
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1776
old_tree = trees[revision.parent_ids[0]]
1777
yield trees[revision.revision_id].changes_from(old_tree)
1780
def get_revision_delta(self, revision_id, specific_fileids=None):
1781
"""Return the delta for one revision.
1783
The delta is relative to the left-hand predecessor of the
1786
:param specific_fileids: if not None, the result is filtered
1787
so that only those file-ids, their parents and their
1788
children are included.
1790
r = self.get_revision(revision_id)
1791
return list(self.get_deltas_for_revisions([r],
1792
specific_fileids=specific_fileids))[0]
1795
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1796
signature = gpg_strategy.sign(plaintext)
1797
self.add_signature_text(revision_id, signature)
1800
def add_signature_text(self, revision_id, signature):
1801
self.signatures.add_lines((revision_id,), (),
1802
osutils.split_lines(signature))
1804
def find_text_key_references(self):
1805
"""Find the text key references within the repository.
1807
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1808
to whether they were referred to by the inventory of the
1809
revision_id that they contain. The inventory texts from all present
1810
revision ids are assessed to generate this report.
1812
revision_keys = self.revisions.keys()
1813
w = self.inventories
1814
pb = ui.ui_factory.nested_progress_bar()
1816
return self._find_text_key_references_from_xml_inventory_lines(
1817
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1821
def _find_text_key_references_from_xml_inventory_lines(self,
1823
"""Core routine for extracting references to texts from inventories.
1825
This performs the translation of xml lines to revision ids.
1827
:param line_iterator: An iterator of lines, origin_version_id
1828
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1829
to whether they were referred to by the inventory of the
1830
revision_id that they contain. Note that if that revision_id was
1831
not part of the line_iterator's output then False will be given -
1832
even though it may actually refer to that key.
1834
if not self._serializer.support_altered_by_hack:
1835
raise AssertionError(
1836
"_find_text_key_references_from_xml_inventory_lines only "
1837
"supported for branches which store inventory as unnested xml"
1838
", not on %r" % self)
1841
# this code needs to read every new line in every inventory for the
1842
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
1843
# not present in one of those inventories is unnecessary but not
1844
# harmful because we are filtering by the revision id marker in the
1845
# inventory lines : we only select file ids altered in one of those
1846
# revisions. We don't need to see all lines in the inventory because
1847
# only those added in an inventory in rev X can contain a revision=X
1849
unescape_revid_cache = {}
1850
unescape_fileid_cache = {}
1852
# jam 20061218 In a big fetch, this handles hundreds of thousands
1853
# of lines, so it has had a lot of inlining and optimizing done.
1854
# Sorry that it is a little bit messy.
1855
# Move several functions to be local variables, since this is a long
1857
search = self._file_ids_altered_regex.search
1858
unescape = _unescape_xml
1859
setdefault = result.setdefault
1860
for line, line_key in line_iterator:
1861
match = search(line)
1864
# One call to match.group() returning multiple items is quite a
1865
# bit faster than 2 calls to match.group() each returning 1
1866
file_id, revision_id = match.group('file_id', 'revision_id')
1868
# Inlining the cache lookups helps a lot when you make 170,000
1869
# lines and 350k ids, versus 8.4 unique ids.
1870
# Using a cache helps in 2 ways:
1871
# 1) Avoids unnecessary decoding calls
1872
# 2) Re-uses cached strings, which helps in future set and
1874
# (2) is enough that removing encoding entirely along with
1875
# the cache (so we are using plain strings) results in no
1876
# performance improvement.
1878
revision_id = unescape_revid_cache[revision_id]
1880
unescaped = unescape(revision_id)
1881
unescape_revid_cache[revision_id] = unescaped
1882
revision_id = unescaped
1884
# Note that unconditionally unescaping means that we deserialise
1885
# every fileid, which for general 'pull' is not great, but we don't
1886
# really want to have some many fulltexts that this matters anyway.
1889
file_id = unescape_fileid_cache[file_id]
1891
unescaped = unescape(file_id)
1892
unescape_fileid_cache[file_id] = unescaped
1895
key = (file_id, revision_id)
1896
setdefault(key, False)
1897
if revision_id == line_key[-1]:
1901
def _inventory_xml_lines_for_keys(self, keys):
1902
"""Get a line iterator of the sort needed for findind references.
1904
Not relevant for non-xml inventory repositories.
1906
Ghosts in revision_keys are ignored.
1908
:param revision_keys: The revision keys for the inventories to inspect.
1909
:return: An iterator over (inventory line, revid) for the fulltexts of
1910
all of the xml inventories specified by revision_keys.
1912
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1913
for record in stream:
1914
if record.storage_kind != 'absent':
1915
chunks = record.get_bytes_as('chunked')
1916
revid = record.key[-1]
1917
lines = osutils.chunks_to_lines(chunks)
1921
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1923
"""Helper routine for fileids_altered_by_revision_ids.
1925
This performs the translation of xml lines to revision ids.
1927
:param line_iterator: An iterator of lines, origin_version_id
1928
:param revision_ids: The revision ids to filter for. This should be a
1929
set or other type which supports efficient __contains__ lookups, as
1930
the revision id from each parsed line will be looked up in the
1931
revision_ids filter.
1932
:return: a dictionary mapping altered file-ids to an iterable of
1933
revision_ids. Each altered file-ids has the exact revision_ids that
1934
altered it listed explicitly.
1936
seen = set(self._find_text_key_references_from_xml_inventory_lines(
1937
line_iterator).iterkeys())
1938
# Note that revision_ids are revision keys.
1939
parent_maps = self.revisions.get_parent_map(revision_ids)
1941
map(parents.update, parent_maps.itervalues())
1942
parents.difference_update(revision_ids)
1943
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
1944
self._inventory_xml_lines_for_keys(parents)))
1945
new_keys = seen - parent_seen
1947
setdefault = result.setdefault
1948
for key in new_keys:
1949
setdefault(key[0], set()).add(key[-1])
1952
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
1953
"""Find the file ids and versions affected by revisions.
1955
:param revisions: an iterable containing revision ids.
1956
:param _inv_weave: The inventory weave from this repository or None.
1957
If None, the inventory weave will be opened automatically.
1958
:return: a dictionary mapping altered file-ids to an iterable of
1959
revision_ids. Each altered file-ids has the exact revision_ids that
1960
altered it listed explicitly.
1962
selected_keys = set((revid,) for revid in revision_ids)
1963
w = _inv_weave or self.inventories
1964
pb = ui.ui_factory.nested_progress_bar()
1966
return self._find_file_ids_from_xml_inventory_lines(
1967
w.iter_lines_added_or_present_in_keys(
1968
selected_keys, pb=pb),
1973
def iter_files_bytes(self, desired_files):
1974
"""Iterate through file versions.
1976
Files will not necessarily be returned in the order they occur in
1977
desired_files. No specific order is guaranteed.
1979
Yields pairs of identifier, bytes_iterator. identifier is an opaque
1980
value supplied by the caller as part of desired_files. It should
1981
uniquely identify the file version in the caller's context. (Examples:
1982
an index number or a TreeTransform trans_id.)
1984
bytes_iterator is an iterable of bytestrings for the file. The
1985
kind of iterable and length of the bytestrings are unspecified, but for
1986
this implementation, it is a list of bytes produced by
1987
VersionedFile.get_record_stream().
1989
:param desired_files: a list of (file_id, revision_id, identifier)
1993
for file_id, revision_id, callable_data in desired_files:
1994
text_keys[(file_id, revision_id)] = callable_data
1995
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
1996
if record.storage_kind == 'absent':
1997
raise errors.RevisionNotPresent(record.key, self)
1998
yield text_keys[record.key], record.get_bytes_as('chunked')
2000
def _generate_text_key_index(self, text_key_references=None,
2002
"""Generate a new text key index for the repository.
2004
This is an expensive function that will take considerable time to run.
2006
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2007
list of parents, also text keys. When a given key has no parents,
2008
the parents list will be [NULL_REVISION].
2010
# All revisions, to find inventory parents.
2011
if ancestors is None:
2012
graph = self.get_graph()
2013
ancestors = graph.get_parent_map(self.all_revision_ids())
2014
if text_key_references is None:
2015
text_key_references = self.find_text_key_references()
2016
pb = ui.ui_factory.nested_progress_bar()
2018
return self._do_generate_text_key_index(ancestors,
2019
text_key_references, pb)
2023
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2024
"""Helper for _generate_text_key_index to avoid deep nesting."""
2025
revision_order = tsort.topo_sort(ancestors)
2026
invalid_keys = set()
2028
for revision_id in revision_order:
2029
revision_keys[revision_id] = set()
2030
text_count = len(text_key_references)
2031
# a cache of the text keys to allow reuse; costs a dict of all the
2032
# keys, but saves a 2-tuple for every child of a given key.
2034
for text_key, valid in text_key_references.iteritems():
2036
invalid_keys.add(text_key)
2038
revision_keys[text_key[1]].add(text_key)
2039
text_key_cache[text_key] = text_key
2040
del text_key_references
2042
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2043
NULL_REVISION = _mod_revision.NULL_REVISION
2044
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2045
# too small for large or very branchy trees. However, for 55K path
2046
# trees, it would be easy to use too much memory trivially. Ideally we
2047
# could gauge this by looking at available real memory etc, but this is
2048
# always a tricky proposition.
2049
inventory_cache = lru_cache.LRUCache(10)
2050
batch_size = 10 # should be ~150MB on a 55K path tree
2051
batch_count = len(revision_order) / batch_size + 1
2053
pb.update("Calculating text parents", processed_texts, text_count)
2054
for offset in xrange(batch_count):
2055
to_query = revision_order[offset * batch_size:(offset + 1) *
2059
for rev_tree in self.revision_trees(to_query):
2060
revision_id = rev_tree.get_revision_id()
2061
parent_ids = ancestors[revision_id]
2062
for text_key in revision_keys[revision_id]:
2063
pb.update("Calculating text parents", processed_texts)
2064
processed_texts += 1
2065
candidate_parents = []
2066
for parent_id in parent_ids:
2067
parent_text_key = (text_key[0], parent_id)
2069
check_parent = parent_text_key not in \
2070
revision_keys[parent_id]
2072
# the parent parent_id is a ghost:
2073
check_parent = False
2074
# truncate the derived graph against this ghost.
2075
parent_text_key = None
2077
# look at the parent commit details inventories to
2078
# determine possible candidates in the per file graph.
2081
inv = inventory_cache[parent_id]
2083
inv = self.revision_tree(parent_id).inventory
2084
inventory_cache[parent_id] = inv
2086
parent_entry = inv[text_key[0]]
2087
except (KeyError, errors.NoSuchId):
2089
if parent_entry is not None:
2091
text_key[0], parent_entry.revision)
2093
parent_text_key = None
2094
if parent_text_key is not None:
2095
candidate_parents.append(
2096
text_key_cache[parent_text_key])
2097
parent_heads = text_graph.heads(candidate_parents)
2098
new_parents = list(parent_heads)
2099
new_parents.sort(key=lambda x:candidate_parents.index(x))
2100
if new_parents == []:
2101
new_parents = [NULL_REVISION]
2102
text_index[text_key] = new_parents
2104
for text_key in invalid_keys:
2105
text_index[text_key] = [NULL_REVISION]
2108
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2109
"""Get an iterable listing the keys of all the data introduced by a set
2112
The keys will be ordered so that the corresponding items can be safely
2113
fetched and inserted in that order.
2115
:returns: An iterable producing tuples of (knit-kind, file-id,
2116
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2117
'revisions'. file-id is None unless knit-kind is 'file'.
2119
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2122
for result in self._find_non_file_keys_to_fetch(revision_ids):
2125
def _find_file_keys_to_fetch(self, revision_ids, pb):
2126
# XXX: it's a bit weird to control the inventory weave caching in this
2127
# generator. Ideally the caching would be done in fetch.py I think. Or
2128
# maybe this generator should explicitly have the contract that it
2129
# should not be iterated until the previously yielded item has been
2131
inv_w = self.inventories
2133
# file ids that changed
2134
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2136
num_file_ids = len(file_ids)
2137
for file_id, altered_versions in file_ids.iteritems():
2139
pb.update("fetch texts", count, num_file_ids)
2141
yield ("file", file_id, altered_versions)
2143
def _find_non_file_keys_to_fetch(self, revision_ids):
2145
yield ("inventory", None, revision_ids)
2148
# XXX: Note ATM no callers actually pay attention to this return
2149
# instead they just use the list of revision ids and ignore
2150
# missing sigs. Consider removing this work entirely
2151
revisions_with_signatures = set(self.signatures.get_parent_map(
2152
[(r,) for r in revision_ids]))
2153
revisions_with_signatures = set(
2154
[r for (r,) in revisions_with_signatures])
2155
revisions_with_signatures.intersection_update(revision_ids)
2156
yield ("signatures", None, revisions_with_signatures)
2159
yield ("revisions", None, revision_ids)
2162
def get_inventory(self, revision_id):
2163
"""Get Inventory object by revision id."""
2164
return self.iter_inventories([revision_id]).next()
2166
def iter_inventories(self, revision_ids):
2167
"""Get many inventories by revision_ids.
2169
This will buffer some or all of the texts used in constructing the
2170
inventories in memory, but will only parse a single inventory at a
2173
:param revision_ids: The expected revision ids of the inventories.
2174
:return: An iterator of inventories.
2176
if ((None in revision_ids)
2177
or (_mod_revision.NULL_REVISION in revision_ids)):
2178
raise ValueError('cannot get null revision inventory')
2179
return self._iter_inventories(revision_ids)
2181
def _iter_inventories(self, revision_ids):
2182
"""single-document based inventory iteration."""
2183
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2184
yield self.deserialise_inventory(revision_id, text)
2186
def _iter_inventory_xmls(self, revision_ids):
2187
keys = [(revision_id,) for revision_id in revision_ids]
2188
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2190
for record in stream:
2191
if record.storage_kind != 'absent':
2192
text_chunks[record.key] = record.get_bytes_as('chunked')
2194
raise errors.NoSuchRevision(self, record.key)
2196
chunks = text_chunks.pop(key)
2197
yield ''.join(chunks), key[-1]
2199
def deserialise_inventory(self, revision_id, xml):
2200
"""Transform the xml into an inventory object.
2202
:param revision_id: The expected revision id of the inventory.
2203
:param xml: A serialised inventory.
2205
result = self._serializer.read_inventory_from_string(xml, revision_id,
2206
entry_cache=self._inventory_entry_cache)
2207
if result.revision_id != revision_id:
2208
raise AssertionError('revision id mismatch %s != %s' % (
2209
result.revision_id, revision_id))
2212
def serialise_inventory(self, inv):
2213
return self._serializer.write_inventory_to_string(inv)
2215
def _serialise_inventory_to_lines(self, inv):
2216
return self._serializer.write_inventory_to_lines(inv)
2218
def get_serializer_format(self):
2219
return self._serializer.format_num
2222
def get_inventory_xml(self, revision_id):
2223
"""Get inventory XML as a file object."""
2224
texts = self._iter_inventory_xmls([revision_id])
2226
text, revision_id = texts.next()
2227
except StopIteration:
2228
raise errors.HistoryMissing(self, 'inventory', revision_id)
2232
def get_inventory_sha1(self, revision_id):
2233
"""Return the sha1 hash of the inventory entry
2235
return self.get_revision(revision_id).inventory_sha1
2237
def iter_reverse_revision_history(self, revision_id):
2238
"""Iterate backwards through revision ids in the lefthand history
2240
:param revision_id: The revision id to start with. All its lefthand
2241
ancestors will be traversed.
2243
graph = self.get_graph()
2244
next_id = revision_id
2246
if next_id in (None, _mod_revision.NULL_REVISION):
2249
parents = graph.get_parent_map([next_id])[next_id]
2251
raise errors.RevisionNotPresent(next_id, self)
2253
if len(parents) == 0:
2256
next_id = parents[0]
2259
def get_revision_inventory(self, revision_id):
2260
"""Return inventory of a past revision."""
2261
# TODO: Unify this with get_inventory()
2262
# bzr 0.0.6 and later imposes the constraint that the inventory_id
2263
# must be the same as its revision, so this is trivial.
2264
if revision_id is None:
2265
# This does not make sense: if there is no revision,
2266
# then it is the current tree inventory surely ?!
2267
# and thus get_root_id() is something that looks at the last
2268
# commit on the branch, and the get_root_id is an inventory check.
2269
raise NotImplementedError
2270
# return Inventory(self.get_root_id())
2272
return self.get_inventory(revision_id)
2274
def is_shared(self):
2275
"""Return True if this repository is flagged as a shared repository."""
2276
raise NotImplementedError(self.is_shared)
2279
def reconcile(self, other=None, thorough=False):
2280
"""Reconcile this repository."""
2281
from bzrlib.reconcile import RepoReconciler
2282
reconciler = RepoReconciler(self, thorough=thorough)
2283
reconciler.reconcile()
2286
def _refresh_data(self):
2287
"""Helper called from lock_* to ensure coherency with disk.
2289
The default implementation does nothing; it is however possible
2290
for repositories to maintain loaded indices across multiple locks
2291
by checking inside their implementation of this method to see
2292
whether their indices are still valid. This depends of course on
2293
the disk format being validatable in this manner. This method is
2294
also called by the refresh_data() public interface to cause a refresh
2295
to occur while in a write lock so that data inserted by a smart server
2296
push operation is visible on the client's instance of the physical
2301
def revision_tree(self, revision_id):
2302
"""Return Tree for a revision on this branch.
2304
`revision_id` may be NULL_REVISION for the empty tree revision.
2306
revision_id = _mod_revision.ensure_null(revision_id)
2307
# TODO: refactor this to use an existing revision object
2308
# so we don't need to read it in twice.
2309
if revision_id == _mod_revision.NULL_REVISION:
2310
return RevisionTree(self, Inventory(root_id=None),
2311
_mod_revision.NULL_REVISION)
2313
inv = self.get_revision_inventory(revision_id)
2314
return RevisionTree(self, inv, revision_id)
2316
def revision_trees(self, revision_ids):
2317
"""Return Trees for revisions in this repository.
2319
:param revision_ids: a sequence of revision-ids;
2320
a revision-id may not be None or 'null:'
2322
inventories = self.iter_inventories(revision_ids)
2323
for inv in inventories:
2324
yield RevisionTree(self, inv, inv.revision_id)
2326
def _filtered_revision_trees(self, revision_ids, file_ids):
2327
"""Return Tree for a revision on this branch with only some files.
2329
:param revision_ids: a sequence of revision-ids;
2330
a revision-id may not be None or 'null:'
2331
:param file_ids: if not None, the result is filtered
2332
so that only those file-ids, their parents and their
2333
children are included.
2335
inventories = self.iter_inventories(revision_ids)
2336
for inv in inventories:
2337
# Should we introduce a FilteredRevisionTree class rather
2338
# than pre-filter the inventory here?
2339
filtered_inv = inv.filter(file_ids)
2340
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2343
def get_ancestry(self, revision_id, topo_sorted=True):
2344
"""Return a list of revision-ids integrated by a revision.
2346
The first element of the list is always None, indicating the origin
2347
revision. This might change when we have history horizons, or
2348
perhaps we should have a new API.
2350
This is topologically sorted.
2352
if _mod_revision.is_null(revision_id):
2354
if not self.has_revision(revision_id):
2355
raise errors.NoSuchRevision(self, revision_id)
2356
graph = self.get_graph()
2358
search = graph._make_breadth_first_searcher([revision_id])
2361
found, ghosts = search.next_with_ghosts()
2362
except StopIteration:
2365
if _mod_revision.NULL_REVISION in keys:
2366
keys.remove(_mod_revision.NULL_REVISION)
2368
parent_map = graph.get_parent_map(keys)
2369
keys = tsort.topo_sort(parent_map)
2370
return [None] + list(keys)
2373
"""Compress the data within the repository.
2375
This operation only makes sense for some repository types. For other
2376
types it should be a no-op that just returns.
2378
This stub method does not require a lock, but subclasses should use
2379
@needs_write_lock as this is a long running call its reasonable to
2380
implicitly lock for the user.
2383
def get_transaction(self):
2384
return self.control_files.get_transaction()
2386
def get_parent_map(self, revision_ids):
2387
"""See graph.StackedParentsProvider.get_parent_map"""
2388
# revisions index works in keys; this just works in revisions
2389
# therefore wrap and unwrap
2392
for revision_id in revision_ids:
2393
if revision_id == _mod_revision.NULL_REVISION:
2394
result[revision_id] = ()
2395
elif revision_id is None:
2396
raise ValueError('get_parent_map(None) is not valid')
2398
query_keys.append((revision_id ,))
2399
for ((revision_id,), parent_keys) in \
2400
self.revisions.get_parent_map(query_keys).iteritems():
2402
result[revision_id] = tuple(parent_revid
2403
for (parent_revid,) in parent_keys)
2405
result[revision_id] = (_mod_revision.NULL_REVISION,)
2408
def _make_parents_provider(self):
2411
def get_graph(self, other_repository=None):
2412
"""Return the graph walker for this repository format"""
2413
parents_provider = self._make_parents_provider()
2414
if (other_repository is not None and
2415
not self.has_same_location(other_repository)):
2416
parents_provider = graph.StackedParentsProvider(
2417
[parents_provider, other_repository._make_parents_provider()])
2418
return graph.Graph(parents_provider)
2420
def _get_versioned_file_checker(self, text_key_references=None):
2421
"""Return an object suitable for checking versioned files.
2423
:param text_key_references: if non-None, an already built
2424
dictionary mapping text keys ((fileid, revision_id) tuples)
2425
to whether they were referred to by the inventory of the
2426
revision_id that they contain. If None, this will be
2429
return _VersionedFileChecker(self,
2430
text_key_references=text_key_references)
2432
def revision_ids_to_search_result(self, result_set):
2433
"""Convert a set of revision ids to a graph SearchResult."""
2434
result_parents = set()
2435
for parents in self.get_graph().get_parent_map(
2436
result_set).itervalues():
2437
result_parents.update(parents)
2438
included_keys = result_set.intersection(result_parents)
2439
start_keys = result_set.difference(included_keys)
2440
exclude_keys = result_parents.difference(result_set)
2441
result = graph.SearchResult(start_keys, exclude_keys,
2442
len(result_set), result_set)
2446
def set_make_working_trees(self, new_value):
2447
"""Set the policy flag for making working trees when creating branches.
2449
This only applies to branches that use this repository.
2451
The default is 'True'.
2452
:param new_value: True to restore the default, False to disable making
2455
raise NotImplementedError(self.set_make_working_trees)
2457
def make_working_trees(self):
2458
"""Returns the policy for making working trees on new branches."""
2459
raise NotImplementedError(self.make_working_trees)
2462
def sign_revision(self, revision_id, gpg_strategy):
2463
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2464
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2467
def has_signature_for_revision_id(self, revision_id):
2468
"""Query for a revision signature for revision_id in the repository."""
2469
if not self.has_revision(revision_id):
2470
raise errors.NoSuchRevision(self, revision_id)
2471
sig_present = (1 == len(
2472
self.signatures.get_parent_map([(revision_id,)])))
2476
def get_signature_text(self, revision_id):
2477
"""Return the text for a signature."""
2478
stream = self.signatures.get_record_stream([(revision_id,)],
2480
record = stream.next()
2481
if record.storage_kind == 'absent':
2482
raise errors.NoSuchRevision(self, revision_id)
2483
return record.get_bytes_as('fulltext')
2486
def check(self, revision_ids=None):
2487
"""Check consistency of all history of given revision_ids.
2489
Different repository implementations should override _check().
2491
:param revision_ids: A non-empty list of revision_ids whose ancestry
2492
will be checked. Typically the last revision_id of a branch.
2494
return self._check(revision_ids)
2496
def _check(self, revision_ids):
2497
result = check.Check(self)
2501
def _warn_if_deprecated(self):
2502
global _deprecation_warning_done
2503
if _deprecation_warning_done:
2505
_deprecation_warning_done = True
2506
warning("Format %s for %s is deprecated - please use 'bzr upgrade' to get better performance"
2507
% (self._format, self.bzrdir.transport.base))
2509
def supports_rich_root(self):
2510
return self._format.rich_root_data
2512
def _check_ascii_revisionid(self, revision_id, method):
2513
"""Private helper for ascii-only repositories."""
2514
# weave repositories refuse to store revisionids that are non-ascii.
2515
if revision_id is not None:
2516
# weaves require ascii revision ids.
2517
if isinstance(revision_id, unicode):
2519
revision_id.encode('ascii')
2520
except UnicodeEncodeError:
2521
raise errors.NonAsciiRevisionId(method, self)
2524
revision_id.decode('ascii')
2525
except UnicodeDecodeError:
2526
raise errors.NonAsciiRevisionId(method, self)
2528
def revision_graph_can_have_wrong_parents(self):
2529
"""Is it possible for this repository to have a revision graph with
2532
If True, then this repository must also implement
2533
_find_inconsistent_revision_parents so that check and reconcile can
2534
check for inconsistencies before proceeding with other checks that may
2535
depend on the revision index being consistent.
2537
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2540
# remove these delegates a while after bzr 0.15
2541
def __make_delegated(name, from_module):
2542
def _deprecated_repository_forwarder():
2543
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2544
% (name, from_module),
2547
m = __import__(from_module, globals(), locals(), [name])
2549
return getattr(m, name)
2550
except AttributeError:
2551
raise AttributeError('module %s has no name %s'
2553
globals()[name] = _deprecated_repository_forwarder
2556
'AllInOneRepository',
2557
'WeaveMetaDirRepository',
2558
'PreSplitOutRepositoryFormat',
2559
'RepositoryFormat4',
2560
'RepositoryFormat5',
2561
'RepositoryFormat6',
2562
'RepositoryFormat7',
2564
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2568
'RepositoryFormatKnit',
2569
'RepositoryFormatKnit1',
2571
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2574
def install_revision(repository, rev, revision_tree):
2575
"""Install all revision data into a repository."""
2576
install_revisions(repository, [(rev, revision_tree, None)])
2579
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2580
"""Install all revision data into a repository.
2582
Accepts an iterable of revision, tree, signature tuples. The signature
2585
repository.start_write_group()
2587
inventory_cache = lru_cache.LRUCache(10)
2588
for n, (revision, revision_tree, signature) in enumerate(iterable):
2589
_install_revision(repository, revision, revision_tree, signature,
2592
pb.update('Transferring revisions', n + 1, num_revisions)
2594
repository.abort_write_group()
2597
repository.commit_write_group()
2600
def _install_revision(repository, rev, revision_tree, signature,
2602
"""Install all revision data into a repository."""
2603
present_parents = []
2605
for p_id in rev.parent_ids:
2606
if repository.has_revision(p_id):
2607
present_parents.append(p_id)
2608
parent_trees[p_id] = repository.revision_tree(p_id)
2610
parent_trees[p_id] = repository.revision_tree(
2611
_mod_revision.NULL_REVISION)
2613
inv = revision_tree.inventory
2614
entries = inv.iter_entries()
2615
# backwards compatibility hack: skip the root id.
2616
if not repository.supports_rich_root():
2617
path, root = entries.next()
2618
if root.revision != rev.revision_id:
2619
raise errors.IncompatibleRevision(repr(repository))
2621
for path, ie in entries:
2622
text_keys[(ie.file_id, ie.revision)] = ie
2623
text_parent_map = repository.texts.get_parent_map(text_keys)
2624
missing_texts = set(text_keys) - set(text_parent_map)
2625
# Add the texts that are not already present
2626
for text_key in missing_texts:
2627
ie = text_keys[text_key]
2629
# FIXME: TODO: The following loop overlaps/duplicates that done by
2630
# commit to determine parents. There is a latent/real bug here where
2631
# the parents inserted are not those commit would do - in particular
2632
# they are not filtered by heads(). RBC, AB
2633
for revision, tree in parent_trees.iteritems():
2634
if ie.file_id not in tree:
2636
parent_id = tree.inventory[ie.file_id].revision
2637
if parent_id in text_parents:
2639
text_parents.append((ie.file_id, parent_id))
2640
lines = revision_tree.get_file(ie.file_id).readlines()
2641
repository.texts.add_lines(text_key, text_parents, lines)
2643
# install the inventory
2644
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2645
# Cache this inventory
2646
inventory_cache[rev.revision_id] = inv
2648
basis_inv = inventory_cache[rev.parent_ids[0]]
2650
repository.add_inventory(rev.revision_id, inv, present_parents)
2652
delta = inv._make_delta(basis_inv)
2653
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2654
rev.revision_id, present_parents)
2656
repository.add_inventory(rev.revision_id, inv, present_parents)
2657
except errors.RevisionAlreadyPresent:
2659
if signature is not None:
2660
repository.add_signature_text(rev.revision_id, signature)
2661
repository.add_revision(rev.revision_id, rev, inv)
2664
class MetaDirRepository(Repository):
2665
"""Repositories in the new meta-dir layout.
2667
:ivar _transport: Transport for access to repository control files,
2668
typically pointing to .bzr/repository.
2671
def __init__(self, _format, a_bzrdir, control_files):
2672
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2673
self._transport = control_files._transport
2675
def is_shared(self):
2676
"""Return True if this repository is flagged as a shared repository."""
2677
return self._transport.has('shared-storage')
2680
def set_make_working_trees(self, new_value):
2681
"""Set the policy flag for making working trees when creating branches.
2683
This only applies to branches that use this repository.
2685
The default is 'True'.
2686
:param new_value: True to restore the default, False to disable making
2691
self._transport.delete('no-working-trees')
2692
except errors.NoSuchFile:
2695
self._transport.put_bytes('no-working-trees', '',
2696
mode=self.bzrdir._get_file_mode())
2698
def make_working_trees(self):
2699
"""Returns the policy for making working trees on new branches."""
2700
return not self._transport.has('no-working-trees')
2703
class MetaDirVersionedFileRepository(MetaDirRepository):
2704
"""Repositories in a meta-dir, that work via versioned file objects."""
2706
def __init__(self, _format, a_bzrdir, control_files):
2707
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2711
network_format_registry = registry.FormatRegistry()
2712
"""Registry of formats indexed by their network name.
2714
The network name for a repository format is an identifier that can be used when
2715
referring to formats with smart server operations. See
2716
RepositoryFormat.network_name() for more detail.
2720
format_registry = registry.FormatRegistry(network_format_registry)
2721
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
2723
This can contain either format instances themselves, or classes/factories that
2724
can be called to obtain one.
2728
#####################################################################
2729
# Repository Formats
2731
class RepositoryFormat(object):
2732
"""A repository format.
2734
Formats provide four things:
2735
* An initialization routine to construct repository data on disk.
2736
* a optional format string which is used when the BzrDir supports
2738
* an open routine which returns a Repository instance.
2739
* A network name for referring to the format in smart server RPC
2742
There is one and only one Format subclass for each on-disk format. But
2743
there can be one Repository subclass that is used for several different
2744
formats. The _format attribute on a Repository instance can be used to
2745
determine the disk format.
2747
Formats are placed in a registry by their format string for reference
2748
during opening. These should be subclasses of RepositoryFormat for
2751
Once a format is deprecated, just deprecate the initialize and open
2752
methods on the format class. Do not deprecate the object, as the
2753
object may be created even when a repository instance hasn't been
2756
Common instance attributes:
2757
_matchingbzrdir - the bzrdir format that the repository format was
2758
originally written to work with. This can be used if manually
2759
constructing a bzrdir and repository, or more commonly for test suite
2763
# Set to True or False in derived classes. True indicates that the format
2764
# supports ghosts gracefully.
2765
supports_ghosts = None
2766
# Can this repository be given external locations to lookup additional
2767
# data. Set to True or False in derived classes.
2768
supports_external_lookups = None
2769
# Does this format support CHK bytestring lookups. Set to True or False in
2771
supports_chks = None
2772
# Should commit add an inventory, or an inventory delta to the repository.
2773
_commit_inv_deltas = True
2774
# What order should fetch operations request streams in?
2775
# The default is unordered as that is the cheapest for an origin to
2777
_fetch_order = 'unordered'
2778
# Does this repository format use deltas that can be fetched as-deltas ?
2779
# (E.g. knits, where the knit deltas can be transplanted intact.
2780
# We default to False, which will ensure that enough data to get
2781
# a full text out of any fetch stream will be grabbed.
2782
_fetch_uses_deltas = False
2783
# Should fetch trigger a reconcile after the fetch? Only needed for
2784
# some repository formats that can suffer internal inconsistencies.
2785
_fetch_reconcile = False
2786
# Does this format have < O(tree_size) delta generation. Used to hint what
2787
# code path for commit, amongst other things.
2791
return "<%s>" % self.__class__.__name__
2793
def __eq__(self, other):
2794
# format objects are generally stateless
2795
return isinstance(other, self.__class__)
2797
def __ne__(self, other):
2798
return not self == other
2801
def find_format(klass, a_bzrdir):
2802
"""Return the format for the repository object in a_bzrdir.
2804
This is used by bzr native formats that have a "format" file in
2805
the repository. Other methods may be used by different types of
2809
transport = a_bzrdir.get_repository_transport(None)
2810
format_string = transport.get("format").read()
2811
return format_registry.get(format_string)
2812
except errors.NoSuchFile:
2813
raise errors.NoRepositoryPresent(a_bzrdir)
2815
raise errors.UnknownFormatError(format=format_string,
2819
def register_format(klass, format):
2820
format_registry.register(format.get_format_string(), format)
2823
def unregister_format(klass, format):
2824
format_registry.remove(format.get_format_string())
2827
def get_default_format(klass):
2828
"""Return the current default format."""
2829
from bzrlib import bzrdir
2830
return bzrdir.format_registry.make_bzrdir('default').repository_format
2832
def get_format_string(self):
2833
"""Return the ASCII format string that identifies this format.
2835
Note that in pre format ?? repositories the format string is
2836
not permitted nor written to disk.
2838
raise NotImplementedError(self.get_format_string)
2840
def get_format_description(self):
2841
"""Return the short description for this format."""
2842
raise NotImplementedError(self.get_format_description)
2844
# TODO: this shouldn't be in the base class, it's specific to things that
2845
# use weaves or knits -- mbp 20070207
2846
def _get_versioned_file_store(self,
2851
versionedfile_class=None,
2852
versionedfile_kwargs={},
2854
if versionedfile_class is None:
2855
versionedfile_class = self._versionedfile_class
2856
weave_transport = control_files._transport.clone(name)
2857
dir_mode = control_files._dir_mode
2858
file_mode = control_files._file_mode
2859
return VersionedFileStore(weave_transport, prefixed=prefixed,
2861
file_mode=file_mode,
2862
versionedfile_class=versionedfile_class,
2863
versionedfile_kwargs=versionedfile_kwargs,
2866
def initialize(self, a_bzrdir, shared=False):
2867
"""Initialize a repository of this format in a_bzrdir.
2869
:param a_bzrdir: The bzrdir to put the new repository in it.
2870
:param shared: The repository should be initialized as a sharable one.
2871
:returns: The new repository object.
2873
This may raise UninitializableFormat if shared repository are not
2874
compatible the a_bzrdir.
2876
raise NotImplementedError(self.initialize)
2878
def is_supported(self):
2879
"""Is this format supported?
2881
Supported formats must be initializable and openable.
2882
Unsupported formats may not support initialization or committing or
2883
some other features depending on the reason for not being supported.
2887
def network_name(self):
2888
"""A simple byte string uniquely identifying this format for RPC calls.
2890
MetaDir repository formats use their disk format string to identify the
2891
repository over the wire. All in one formats such as bzr < 0.8, and
2892
foreign formats like svn/git and hg should use some marker which is
2893
unique and immutable.
2895
raise NotImplementedError(self.network_name)
2897
def check_conversion_target(self, target_format):
2898
raise NotImplementedError(self.check_conversion_target)
2900
def open(self, a_bzrdir, _found=False):
2901
"""Return an instance of this format for the bzrdir a_bzrdir.
2903
_found is a private parameter, do not use it.
2905
raise NotImplementedError(self.open)
2908
class MetaDirRepositoryFormat(RepositoryFormat):
2909
"""Common base class for the new repositories using the metadir layout."""
2911
rich_root_data = False
2912
supports_tree_reference = False
2913
supports_external_lookups = False
2916
def _matchingbzrdir(self):
2917
matching = bzrdir.BzrDirMetaFormat1()
2918
matching.repository_format = self
2922
super(MetaDirRepositoryFormat, self).__init__()
2924
def _create_control_files(self, a_bzrdir):
2925
"""Create the required files and the initial control_files object."""
2926
# FIXME: RBC 20060125 don't peek under the covers
2927
# NB: no need to escape relative paths that are url safe.
2928
repository_transport = a_bzrdir.get_repository_transport(self)
2929
control_files = lockable_files.LockableFiles(repository_transport,
2930
'lock', lockdir.LockDir)
2931
control_files.create_lock()
2932
return control_files
2934
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
2935
"""Upload the initial blank content."""
2936
control_files = self._create_control_files(a_bzrdir)
2937
control_files.lock_write()
2938
transport = control_files._transport
2940
utf8_files += [('shared-storage', '')]
2942
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
2943
for (filename, content_stream) in files:
2944
transport.put_file(filename, content_stream,
2945
mode=a_bzrdir._get_file_mode())
2946
for (filename, content_bytes) in utf8_files:
2947
transport.put_bytes_non_atomic(filename, content_bytes,
2948
mode=a_bzrdir._get_file_mode())
2950
control_files.unlock()
2952
def network_name(self):
2953
"""Metadir formats have matching disk and network format strings."""
2954
return self.get_format_string()
2957
# Pre-0.8 formats that don't have a disk format string (because they are
2958
# versioned by the matching control directory). We use the control directories
2959
# disk format string as a key for the network_name because they meet the
2960
# constraints (simple string, unique, immutable).
2961
network_format_registry.register_lazy(
2962
"Bazaar-NG branch, format 5\n",
2963
'bzrlib.repofmt.weaverepo',
2964
'RepositoryFormat5',
2966
network_format_registry.register_lazy(
2967
"Bazaar-NG branch, format 6\n",
2968
'bzrlib.repofmt.weaverepo',
2969
'RepositoryFormat6',
2972
# formats which have no format string are not discoverable or independently
2973
# creatable on disk, so are not registered in format_registry. They're
2974
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
2975
# needed, it's constructed directly by the BzrDir. Non-native formats where
2976
# the repository is not separately opened are similar.
2978
format_registry.register_lazy(
2979
'Bazaar-NG Repository format 7',
2980
'bzrlib.repofmt.weaverepo',
2984
format_registry.register_lazy(
2985
'Bazaar-NG Knit Repository Format 1',
2986
'bzrlib.repofmt.knitrepo',
2987
'RepositoryFormatKnit1',
2990
format_registry.register_lazy(
2991
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
2992
'bzrlib.repofmt.knitrepo',
2993
'RepositoryFormatKnit3',
2996
format_registry.register_lazy(
2997
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
2998
'bzrlib.repofmt.knitrepo',
2999
'RepositoryFormatKnit4',
3002
# Pack-based formats. There is one format for pre-subtrees, and one for
3003
# post-subtrees to allow ease of testing.
3004
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
3005
format_registry.register_lazy(
3006
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
3007
'bzrlib.repofmt.pack_repo',
3008
'RepositoryFormatKnitPack1',
3010
format_registry.register_lazy(
3011
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
3012
'bzrlib.repofmt.pack_repo',
3013
'RepositoryFormatKnitPack3',
3015
format_registry.register_lazy(
3016
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
3017
'bzrlib.repofmt.pack_repo',
3018
'RepositoryFormatKnitPack4',
3020
format_registry.register_lazy(
3021
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
3022
'bzrlib.repofmt.pack_repo',
3023
'RepositoryFormatKnitPack5',
3025
format_registry.register_lazy(
3026
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
3027
'bzrlib.repofmt.pack_repo',
3028
'RepositoryFormatKnitPack5RichRoot',
3030
format_registry.register_lazy(
3031
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3032
'bzrlib.repofmt.pack_repo',
3033
'RepositoryFormatKnitPack5RichRootBroken',
3035
format_registry.register_lazy(
3036
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3037
'bzrlib.repofmt.pack_repo',
3038
'RepositoryFormatKnitPack6',
3040
format_registry.register_lazy(
3041
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3042
'bzrlib.repofmt.pack_repo',
3043
'RepositoryFormatKnitPack6RichRoot',
3046
# Development formats.
3047
# Obsolete but kept pending a CHK based subtree format.
3048
format_registry.register_lazy(
3049
("Bazaar development format 2 with subtree support "
3050
"(needs bzr.dev from before 1.8)\n"),
3051
'bzrlib.repofmt.pack_repo',
3052
'RepositoryFormatPackDevelopment2Subtree',
3055
# 1.14->1.16 go below here
3056
format_registry.register_lazy(
3057
'Bazaar development format - group compression and chk inventory'
3058
' (needs bzr.dev from 1.14)\n',
3059
'bzrlib.repofmt.groupcompress_repo',
3060
'RepositoryFormatCHK1',
3063
format_registry.register_lazy(
3064
'Bazaar development format - chk repository with bencode revision '
3065
'serialization (needs bzr.dev from 1.16)\n',
3066
'bzrlib.repofmt.groupcompress_repo',
3067
'RepositoryFormatCHK2',
3069
format_registry.register_lazy(
3070
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3071
'bzrlib.repofmt.groupcompress_repo',
3072
'RepositoryFormat2a',
3076
class InterRepository(InterObject):
3077
"""This class represents operations taking place between two repositories.
3079
Its instances have methods like copy_content and fetch, and contain
3080
references to the source and target repositories these operations can be
3083
Often we will provide convenience methods on 'repository' which carry out
3084
operations with another repository - they will always forward to
3085
InterRepository.get(other).method_name(parameters).
3088
_walk_to_common_revisions_batch_size = 50
3090
"""The available optimised InterRepository types."""
3093
def copy_content(self, revision_id=None):
3094
"""Make a complete copy of the content in self into destination.
3096
This is a destructive operation! Do not use it on existing
3099
:param revision_id: Only copy the content needed to construct
3100
revision_id and its parents.
3103
self.target.set_make_working_trees(self.source.make_working_trees())
3104
except NotImplementedError:
3106
self.target.fetch(self.source, revision_id=revision_id)
3109
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3111
"""Fetch the content required to construct revision_id.
3113
The content is copied from self.source to self.target.
3115
:param revision_id: if None all content is copied, if NULL_REVISION no
3117
:param pb: optional progress bar to use for progress reports. If not
3118
provided a default one will be created.
3121
from bzrlib.fetch import RepoFetcher
3122
f = RepoFetcher(to_repository=self.target,
3123
from_repository=self.source,
3124
last_revision=revision_id,
3125
fetch_spec=fetch_spec,
3126
pb=pb, find_ghosts=find_ghosts)
3128
def _walk_to_common_revisions(self, revision_ids):
3129
"""Walk out from revision_ids in source to revisions target has.
3131
:param revision_ids: The start point for the search.
3132
:return: A set of revision ids.
3134
target_graph = self.target.get_graph()
3135
revision_ids = frozenset(revision_ids)
3136
missing_revs = set()
3137
source_graph = self.source.get_graph()
3138
# ensure we don't pay silly lookup costs.
3139
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3140
null_set = frozenset([_mod_revision.NULL_REVISION])
3141
searcher_exhausted = False
3145
# Iterate the searcher until we have enough next_revs
3146
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3148
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3149
next_revs.update(next_revs_part)
3150
ghosts.update(ghosts_part)
3151
except StopIteration:
3152
searcher_exhausted = True
3154
# If there are ghosts in the source graph, and the caller asked for
3155
# them, make sure that they are present in the target.
3156
# We don't care about other ghosts as we can't fetch them and
3157
# haven't been asked to.
3158
ghosts_to_check = set(revision_ids.intersection(ghosts))
3159
revs_to_get = set(next_revs).union(ghosts_to_check)
3161
have_revs = set(target_graph.get_parent_map(revs_to_get))
3162
# we always have NULL_REVISION present.
3163
have_revs = have_revs.union(null_set)
3164
# Check if the target is missing any ghosts we need.
3165
ghosts_to_check.difference_update(have_revs)
3167
# One of the caller's revision_ids is a ghost in both the
3168
# source and the target.
3169
raise errors.NoSuchRevision(
3170
self.source, ghosts_to_check.pop())
3171
missing_revs.update(next_revs - have_revs)
3172
# Because we may have walked past the original stop point, make
3173
# sure everything is stopped
3174
stop_revs = searcher.find_seen_ancestors(have_revs)
3175
searcher.stop_searching_any(stop_revs)
3176
if searcher_exhausted:
3178
return searcher.get_result()
3181
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3182
"""Return the revision ids that source has that target does not.
3184
:param revision_id: only return revision ids included by this
3186
:param find_ghosts: If True find missing revisions in deep history
3187
rather than just finding the surface difference.
3188
:return: A bzrlib.graph.SearchResult.
3190
# stop searching at found target revisions.
3191
if not find_ghosts and revision_id is not None:
3192
return self._walk_to_common_revisions([revision_id])
3193
# generic, possibly worst case, slow code path.
3194
target_ids = set(self.target.all_revision_ids())
3195
if revision_id is not None:
3196
source_ids = self.source.get_ancestry(revision_id)
3197
if source_ids[0] is not None:
3198
raise AssertionError()
3201
source_ids = self.source.all_revision_ids()
3202
result_set = set(source_ids).difference(target_ids)
3203
return self.source.revision_ids_to_search_result(result_set)
3206
def _same_model(source, target):
3207
"""True if source and target have the same data representation.
3209
Note: this is always called on the base class; overriding it in a
3210
subclass will have no effect.
3213
InterRepository._assert_same_model(source, target)
3215
except errors.IncompatibleRepositories, e:
3219
def _assert_same_model(source, target):
3220
"""Raise an exception if two repositories do not use the same model.
3222
if source.supports_rich_root() != target.supports_rich_root():
3223
raise errors.IncompatibleRepositories(source, target,
3224
"different rich-root support")
3225
if source._serializer != target._serializer:
3226
raise errors.IncompatibleRepositories(source, target,
3227
"different serializers")
3230
class InterSameDataRepository(InterRepository):
3231
"""Code for converting between repositories that represent the same data.
3233
Data format and model must match for this to work.
3237
def _get_repo_format_to_test(self):
3238
"""Repository format for testing with.
3240
InterSameData can pull from subtree to subtree and from non-subtree to
3241
non-subtree, so we test this with the richest repository format.
3243
from bzrlib.repofmt import knitrepo
3244
return knitrepo.RepositoryFormatKnit3()
3247
def is_compatible(source, target):
3248
return InterRepository._same_model(source, target)
3251
class InterWeaveRepo(InterSameDataRepository):
3252
"""Optimised code paths between Weave based repositories.
3254
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3255
implemented lazy inter-object optimisation.
3259
def _get_repo_format_to_test(self):
3260
from bzrlib.repofmt import weaverepo
3261
return weaverepo.RepositoryFormat7()
3264
def is_compatible(source, target):
3265
"""Be compatible with known Weave formats.
3267
We don't test for the stores being of specific types because that
3268
could lead to confusing results, and there is no need to be
3271
from bzrlib.repofmt.weaverepo import (
3277
return (isinstance(source._format, (RepositoryFormat5,
3279
RepositoryFormat7)) and
3280
isinstance(target._format, (RepositoryFormat5,
3282
RepositoryFormat7)))
3283
except AttributeError:
3287
def copy_content(self, revision_id=None):
3288
"""See InterRepository.copy_content()."""
3289
# weave specific optimised path:
3291
self.target.set_make_working_trees(self.source.make_working_trees())
3292
except (errors.RepositoryUpgradeRequired, NotImplemented):
3294
# FIXME do not peek!
3295
if self.source._transport.listable():
3296
pb = ui.ui_factory.nested_progress_bar()
3298
self.target.texts.insert_record_stream(
3299
self.source.texts.get_record_stream(
3300
self.source.texts.keys(), 'topological', False))
3301
pb.update('copying inventory', 0, 1)
3302
self.target.inventories.insert_record_stream(
3303
self.source.inventories.get_record_stream(
3304
self.source.inventories.keys(), 'topological', False))
3305
self.target.signatures.insert_record_stream(
3306
self.source.signatures.get_record_stream(
3307
self.source.signatures.keys(),
3309
self.target.revisions.insert_record_stream(
3310
self.source.revisions.get_record_stream(
3311
self.source.revisions.keys(),
3312
'topological', True))
3316
self.target.fetch(self.source, revision_id=revision_id)
3319
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3320
"""See InterRepository.missing_revision_ids()."""
3321
# we want all revisions to satisfy revision_id in source.
3322
# but we don't want to stat every file here and there.
3323
# we want then, all revisions other needs to satisfy revision_id
3324
# checked, but not those that we have locally.
3325
# so the first thing is to get a subset of the revisions to
3326
# satisfy revision_id in source, and then eliminate those that
3327
# we do already have.
3328
# this is slow on high latency connection to self, but as this
3329
# disk format scales terribly for push anyway due to rewriting
3330
# inventory.weave, this is considered acceptable.
3332
if revision_id is not None:
3333
source_ids = self.source.get_ancestry(revision_id)
3334
if source_ids[0] is not None:
3335
raise AssertionError()
3338
source_ids = self.source._all_possible_ids()
3339
source_ids_set = set(source_ids)
3340
# source_ids is the worst possible case we may need to pull.
3341
# now we want to filter source_ids against what we actually
3342
# have in target, but don't try to check for existence where we know
3343
# we do not have a revision as that would be pointless.
3344
target_ids = set(self.target._all_possible_ids())
3345
possibly_present_revisions = target_ids.intersection(source_ids_set)
3346
actually_present_revisions = set(
3347
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3348
required_revisions = source_ids_set.difference(actually_present_revisions)
3349
if revision_id is not None:
3350
# we used get_ancestry to determine source_ids then we are assured all
3351
# revisions referenced are present as they are installed in topological order.
3352
# and the tip revision was validated by get_ancestry.
3353
result_set = required_revisions
3355
# if we just grabbed the possibly available ids, then
3356
# we only have an estimate of whats available and need to validate
3357
# that against the revision records.
3359
self.source._eliminate_revisions_not_present(required_revisions))
3360
return self.source.revision_ids_to_search_result(result_set)
3363
class InterKnitRepo(InterSameDataRepository):
3364
"""Optimised code paths between Knit based repositories."""
3367
def _get_repo_format_to_test(self):
3368
from bzrlib.repofmt import knitrepo
3369
return knitrepo.RepositoryFormatKnit1()
3372
def is_compatible(source, target):
3373
"""Be compatible with known Knit formats.
3375
We don't test for the stores being of specific types because that
3376
could lead to confusing results, and there is no need to be
3379
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3381
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3382
isinstance(target._format, RepositoryFormatKnit))
3383
except AttributeError:
3385
return are_knits and InterRepository._same_model(source, target)
3388
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3389
"""See InterRepository.missing_revision_ids()."""
3390
if revision_id is not None:
3391
source_ids = self.source.get_ancestry(revision_id)
3392
if source_ids[0] is not None:
3393
raise AssertionError()
3396
source_ids = self.source.all_revision_ids()
3397
source_ids_set = set(source_ids)
3398
# source_ids is the worst possible case we may need to pull.
3399
# now we want to filter source_ids against what we actually
3400
# have in target, but don't try to check for existence where we know
3401
# we do not have a revision as that would be pointless.
3402
target_ids = set(self.target.all_revision_ids())
3403
possibly_present_revisions = target_ids.intersection(source_ids_set)
3404
actually_present_revisions = set(
3405
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3406
required_revisions = source_ids_set.difference(actually_present_revisions)
3407
if revision_id is not None:
3408
# we used get_ancestry to determine source_ids then we are assured all
3409
# revisions referenced are present as they are installed in topological order.
3410
# and the tip revision was validated by get_ancestry.
3411
result_set = required_revisions
3413
# if we just grabbed the possibly available ids, then
3414
# we only have an estimate of whats available and need to validate
3415
# that against the revision records.
3417
self.source._eliminate_revisions_not_present(required_revisions))
3418
return self.source.revision_ids_to_search_result(result_set)
3421
class InterPackRepo(InterSameDataRepository):
3422
"""Optimised code paths between Pack based repositories."""
3425
def _get_repo_format_to_test(self):
3426
from bzrlib.repofmt import pack_repo
3427
return pack_repo.RepositoryFormatKnitPack6RichRoot()
3430
def is_compatible(source, target):
3431
"""Be compatible with known Pack formats.
3433
We don't test for the stores being of specific types because that
3434
could lead to confusing results, and there is no need to be
3437
InterPackRepo does not support CHK based repositories.
3439
from bzrlib.repofmt.pack_repo import RepositoryFormatPack
3440
from bzrlib.repofmt.groupcompress_repo import RepositoryFormatCHK1
3442
are_packs = (isinstance(source._format, RepositoryFormatPack) and
3443
isinstance(target._format, RepositoryFormatPack))
3444
not_packs = (isinstance(source._format, RepositoryFormatCHK1) or
3445
isinstance(target._format, RepositoryFormatCHK1))
3446
except AttributeError:
3448
if not_packs or not are_packs:
3450
return InterRepository._same_model(source, target)
3453
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3455
"""See InterRepository.fetch()."""
3456
if (len(self.source._fallback_repositories) > 0 or
3457
len(self.target._fallback_repositories) > 0):
3458
# The pack layer is not aware of fallback repositories, so when
3459
# fetching from a stacked repository or into a stacked repository
3460
# we use the generic fetch logic which uses the VersionedFiles
3461
# attributes on repository.
3462
from bzrlib.fetch import RepoFetcher
3463
fetcher = RepoFetcher(self.target, self.source, revision_id,
3464
pb, find_ghosts, fetch_spec=fetch_spec)
3465
if fetch_spec is not None:
3466
if len(list(fetch_spec.heads)) != 1:
3467
raise AssertionError(
3468
"InterPackRepo.fetch doesn't support "
3469
"fetching multiple heads yet.")
3470
revision_id = list(fetch_spec.heads)[0]
3472
if revision_id is None:
3474
# everything to do - use pack logic
3475
# to fetch from all packs to one without
3476
# inventory parsing etc, IFF nothing to be copied is in the target.
3478
source_revision_ids = frozenset(self.source.all_revision_ids())
3479
revision_ids = source_revision_ids - \
3480
frozenset(self.target.get_parent_map(source_revision_ids))
3481
revision_keys = [(revid,) for revid in revision_ids]
3482
index = self.target._pack_collection.revision_index.combined_index
3483
present_revision_ids = set(item[1][0] for item in
3484
index.iter_entries(revision_keys))
3485
revision_ids = set(revision_ids) - present_revision_ids
3486
# implementing the TODO will involve:
3487
# - detecting when all of a pack is selected
3488
# - avoiding as much as possible pre-selection, so the
3489
# more-core routines such as create_pack_from_packs can filter in
3490
# a just-in-time fashion. (though having a HEADS list on a
3491
# repository might make this a lot easier, because we could
3492
# sensibly detect 'new revisions' without doing a full index scan.
3493
elif _mod_revision.is_null(revision_id):
3497
revision_ids = self.search_missing_revision_ids(revision_id,
3498
find_ghosts=find_ghosts).get_keys()
3499
if len(revision_ids) == 0:
3501
return self._pack(self.source, self.target, revision_ids)
3503
def _pack(self, source, target, revision_ids):
3504
from bzrlib.repofmt.pack_repo import Packer
3505
packs = source._pack_collection.all_packs()
3506
pack = Packer(self.target._pack_collection, packs, '.fetch',
3507
revision_ids).pack()
3508
if pack is not None:
3509
self.target._pack_collection._save_pack_names()
3510
copied_revs = pack.get_revision_count()
3511
# Trigger an autopack. This may duplicate effort as we've just done
3512
# a pack creation, but for now it is simpler to think about as
3513
# 'upload data, then repack if needed'.
3514
self.target._pack_collection.autopack()
3515
return (copied_revs, [])
3520
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3521
"""See InterRepository.missing_revision_ids().
3523
:param find_ghosts: Find ghosts throughout the ancestry of
3526
if not find_ghosts and revision_id is not None:
3527
return self._walk_to_common_revisions([revision_id])
3528
elif revision_id is not None:
3529
# Find ghosts: search for revisions pointing from one repository to
3530
# the other, and vice versa, anywhere in the history of revision_id.
3531
graph = self.target.get_graph(other_repository=self.source)
3532
searcher = graph._make_breadth_first_searcher([revision_id])
3536
next_revs, ghosts = searcher.next_with_ghosts()
3537
except StopIteration:
3539
if revision_id in ghosts:
3540
raise errors.NoSuchRevision(self.source, revision_id)
3541
found_ids.update(next_revs)
3542
found_ids.update(ghosts)
3543
found_ids = frozenset(found_ids)
3544
# Double query here: should be able to avoid this by changing the
3545
# graph api further.
3546
result_set = found_ids - frozenset(
3547
self.target.get_parent_map(found_ids))
3549
source_ids = self.source.all_revision_ids()
3550
# source_ids is the worst possible case we may need to pull.
3551
# now we want to filter source_ids against what we actually
3552
# have in target, but don't try to check for existence where we know
3553
# we do not have a revision as that would be pointless.
3554
target_ids = set(self.target.all_revision_ids())
3555
result_set = set(source_ids).difference(target_ids)
3556
return self.source.revision_ids_to_search_result(result_set)
3559
class InterDifferingSerializer(InterRepository):
3562
def _get_repo_format_to_test(self):
3566
def is_compatible(source, target):
3567
"""Be compatible with Knit2 source and Knit3 target"""
3568
# This is redundant with format.check_conversion_target(), however that
3569
# raises an exception, and we just want to say "False" as in we won't
3570
# support converting between these formats.
3571
if source.supports_rich_root() and not target.supports_rich_root():
3573
if (source._format.supports_tree_reference
3574
and not target._format.supports_tree_reference):
3578
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3579
"""Get the best delta and base for this revision.
3581
:return: (basis_id, delta)
3583
possible_trees = [(parent_id, cache[parent_id])
3584
for parent_id in parent_ids
3585
if parent_id in cache]
3586
if len(possible_trees) == 0:
3587
# There either aren't any parents, or the parents aren't in the
3588
# cache, so just use the last converted tree
3589
possible_trees.append((basis_id, cache[basis_id]))
3591
for basis_id, basis_tree in possible_trees:
3592
delta = tree.inventory._make_delta(basis_tree.inventory)
3593
deltas.append((len(delta), basis_id, delta))
3595
return deltas[0][1:]
3597
def _get_parent_keys(self, root_key, parent_map):
3598
"""Get the parent keys for a given root id."""
3599
root_id, rev_id = root_key
3600
# Include direct parents of the revision, but only if they used
3601
# the same root_id and are heads.
3603
for parent_id in parent_map[rev_id]:
3604
if parent_id == _mod_revision.NULL_REVISION:
3606
if parent_id not in self._revision_id_to_root_id:
3607
# We probably didn't read this revision, go spend the
3608
# extra effort to actually check
3610
tree = self.source.revision_tree(parent_id)
3611
except errors.NoSuchRevision:
3612
# Ghost, fill out _revision_id_to_root_id in case we
3613
# encounter this again.
3614
# But set parent_root_id to None since we don't really know
3615
parent_root_id = None
3617
parent_root_id = tree.get_root_id()
3618
self._revision_id_to_root_id[parent_id] = None
3620
parent_root_id = self._revision_id_to_root_id[parent_id]
3621
if root_id == parent_root_id:
3622
# With stacking we _might_ want to refer to a non-local
3623
# revision, but this code path only applies when we have the
3624
# full content available, so ghosts really are ghosts, not just
3625
# the edge of local data.
3626
parent_keys.append((parent_id,))
3628
# root_id may be in the parent anyway.
3630
tree = self.source.revision_tree(parent_id)
3631
except errors.NoSuchRevision:
3632
# ghost, can't refer to it.
3636
parent_keys.append((tree.inventory[root_id].revision,))
3637
except errors.NoSuchId:
3640
g = graph.Graph(self.source.revisions)
3641
heads = g.heads(parent_keys)
3643
for key in parent_keys:
3644
if key in heads and key not in selected_keys:
3645
selected_keys.append(key)
3646
return tuple([(root_id,)+ key for key in selected_keys])
3648
def _new_root_data_stream(self, root_keys_to_create, parent_map):
3649
for root_key in root_keys_to_create:
3650
parent_keys = self._get_parent_keys(root_key, parent_map)
3651
yield versionedfile.FulltextContentFactory(root_key,
3652
parent_keys, None, '')
3654
def _fetch_batch(self, revision_ids, basis_id, cache):
3655
"""Fetch across a few revisions.
3657
:param revision_ids: The revisions to copy
3658
:param basis_id: The revision_id of a tree that must be in cache, used
3659
as a basis for delta when no other base is available
3660
:param cache: A cache of RevisionTrees that we can use.
3661
:return: The revision_id of the last converted tree. The RevisionTree
3662
for it will be in cache
3664
# Walk though all revisions; get inventory deltas, copy referenced
3665
# texts that delta references, insert the delta, revision and
3667
root_keys_to_create = set()
3670
pending_revisions = []
3671
parent_map = self.source.get_parent_map(revision_ids)
3672
for tree in self.source.revision_trees(revision_ids):
3673
current_revision_id = tree.get_revision_id()
3674
parent_ids = parent_map.get(current_revision_id, ())
3675
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3677
if self._converting_to_rich_root:
3678
self._revision_id_to_root_id[current_revision_id] = \
3680
# Find text entries that need to be copied
3681
for old_path, new_path, file_id, entry in delta:
3682
if new_path is not None:
3685
if not self.target.supports_rich_root():
3686
# The target doesn't support rich root, so we don't
3689
if self._converting_to_rich_root:
3690
# This can't be copied normally, we have to insert
3692
root_keys_to_create.add((file_id, entry.revision))
3694
text_keys.add((file_id, entry.revision))
3695
revision = self.source.get_revision(current_revision_id)
3696
pending_deltas.append((basis_id, delta,
3697
current_revision_id, revision.parent_ids))
3698
pending_revisions.append(revision)
3699
cache[current_revision_id] = tree
3700
basis_id = current_revision_id
3702
from_texts = self.source.texts
3703
to_texts = self.target.texts
3704
if root_keys_to_create:
3705
root_stream = self._new_root_data_stream(root_keys_to_create,
3707
to_texts.insert_record_stream(root_stream)
3708
to_texts.insert_record_stream(from_texts.get_record_stream(
3709
text_keys, self.target._format._fetch_order,
3710
not self.target._format._fetch_uses_deltas))
3711
# insert inventory deltas
3712
for delta in pending_deltas:
3713
self.target.add_inventory_by_delta(*delta)
3714
if self.target._fallback_repositories:
3715
# Make sure this stacked repository has all the parent inventories
3716
# for the new revisions that we are about to insert. We do this
3717
# before adding the revisions so that no revision is added until
3718
# all the inventories it may depend on are added.
3720
revision_ids = set()
3721
for revision in pending_revisions:
3722
revision_ids.add(revision.revision_id)
3723
parent_ids.update(revision.parent_ids)
3724
parent_ids.difference_update(revision_ids)
3725
parent_ids.discard(_mod_revision.NULL_REVISION)
3726
parent_map = self.source.get_parent_map(parent_ids)
3727
for parent_tree in self.source.revision_trees(parent_ids):
3728
basis_id, delta = self._get_delta_for_revision(tree, parent_ids, basis_id, cache)
3729
current_revision_id = parent_tree.get_revision_id()
3730
parents_parents = parent_map[current_revision_id]
3731
self.target.add_inventory_by_delta(
3732
basis_id, delta, current_revision_id, parents_parents)
3733
# insert signatures and revisions
3734
for revision in pending_revisions:
3736
signature = self.source.get_signature_text(
3737
revision.revision_id)
3738
self.target.add_signature_text(revision.revision_id,
3740
except errors.NoSuchRevision:
3742
self.target.add_revision(revision.revision_id, revision)
3745
def _fetch_all_revisions(self, revision_ids, pb):
3746
"""Fetch everything for the list of revisions.
3748
:param revision_ids: The list of revisions to fetch. Must be in
3750
:param pb: A ProgressBar
3753
basis_id, basis_tree = self._get_basis(revision_ids[0])
3755
cache = lru_cache.LRUCache(100)
3756
cache[basis_id] = basis_tree
3757
del basis_tree # We don't want to hang on to it here
3758
for offset in range(0, len(revision_ids), batch_size):
3759
self.target.start_write_group()
3761
pb.update('Transferring revisions', offset,
3763
batch = revision_ids[offset:offset+batch_size]
3764
basis_id = self._fetch_batch(batch, basis_id, cache)
3766
self.target.abort_write_group()
3769
self.target.commit_write_group()
3770
pb.update('Transferring revisions', len(revision_ids),
3774
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3776
"""See InterRepository.fetch()."""
3777
if fetch_spec is not None:
3778
raise AssertionError("Not implemented yet...")
3779
if (not self.source.supports_rich_root()
3780
and self.target.supports_rich_root()):
3781
self._converting_to_rich_root = True
3782
self._revision_id_to_root_id = {}
3784
self._converting_to_rich_root = False
3785
revision_ids = self.target.search_missing_revision_ids(self.source,
3786
revision_id, find_ghosts=find_ghosts).get_keys()
3787
if not revision_ids:
3789
revision_ids = tsort.topo_sort(
3790
self.source.get_graph().get_parent_map(revision_ids))
3791
if not revision_ids:
3793
# Walk though all revisions; get inventory deltas, copy referenced
3794
# texts that delta references, insert the delta, revision and
3796
first_rev = self.source.get_revision(revision_ids[0])
3798
my_pb = ui.ui_factory.nested_progress_bar()
3801
symbol_versioning.warn(
3802
symbol_versioning.deprecated_in((1, 14, 0))
3803
% "pb parameter to fetch()")
3806
self._fetch_all_revisions(revision_ids, pb)
3808
if my_pb is not None:
3810
return len(revision_ids), 0
3812
def _get_basis(self, first_revision_id):
3813
"""Get a revision and tree which exists in the target.
3815
This assumes that first_revision_id is selected for transmission
3816
because all other ancestors are already present. If we can't find an
3817
ancestor we fall back to NULL_REVISION since we know that is safe.
3819
:return: (basis_id, basis_tree)
3821
first_rev = self.source.get_revision(first_revision_id)
3823
basis_id = first_rev.parent_ids[0]
3824
# only valid as a basis if the target has it
3825
self.target.get_revision(basis_id)
3826
# Try to get a basis tree - if its a ghost it will hit the
3827
# NoSuchRevision case.
3828
basis_tree = self.source.revision_tree(basis_id)
3829
except (IndexError, errors.NoSuchRevision):
3830
basis_id = _mod_revision.NULL_REVISION
3831
basis_tree = self.source.revision_tree(basis_id)
3832
return basis_id, basis_tree
3835
InterRepository.register_optimiser(InterDifferingSerializer)
3836
InterRepository.register_optimiser(InterSameDataRepository)
3837
InterRepository.register_optimiser(InterWeaveRepo)
3838
InterRepository.register_optimiser(InterKnitRepo)
3839
InterRepository.register_optimiser(InterPackRepo)
3842
class CopyConverter(object):
3843
"""A repository conversion tool which just performs a copy of the content.
3845
This is slow but quite reliable.
3848
def __init__(self, target_format):
3849
"""Create a CopyConverter.
3851
:param target_format: The format the resulting repository should be.
3853
self.target_format = target_format
3855
def convert(self, repo, pb):
3856
"""Perform the conversion of to_convert, giving feedback via pb.
3858
:param to_convert: The disk object to convert.
3859
:param pb: a progress bar to use for progress information.
3864
# this is only useful with metadir layouts - separated repo content.
3865
# trigger an assertion if not such
3866
repo._format.get_format_string()
3867
self.repo_dir = repo.bzrdir
3868
self.step('Moving repository to repository.backup')
3869
self.repo_dir.transport.move('repository', 'repository.backup')
3870
backup_transport = self.repo_dir.transport.clone('repository.backup')
3871
repo._format.check_conversion_target(self.target_format)
3872
self.source_repo = repo._format.open(self.repo_dir,
3874
_override_transport=backup_transport)
3875
self.step('Creating new repository')
3876
converted = self.target_format.initialize(self.repo_dir,
3877
self.source_repo.is_shared())
3878
converted.lock_write()
3880
self.step('Copying content into repository.')
3881
self.source_repo.copy_content_into(converted)
3884
self.step('Deleting old repository content.')
3885
self.repo_dir.transport.delete_tree('repository.backup')
3886
self.pb.note('repository converted')
3888
def step(self, message):
3889
"""Update the pb by a step."""
3891
self.pb.update(message, self.count, self.total)
3903
def _unescaper(match, _map=_unescape_map):
3904
code = match.group(1)
3908
if not code.startswith('#'):
3910
return unichr(int(code[1:])).encode('utf8')
3916
def _unescape_xml(data):
3917
"""Unescape predefined XML entities in a string of data."""
3919
if _unescape_re is None:
3920
_unescape_re = re.compile('\&([^;]*);')
3921
return _unescape_re.sub(_unescaper, data)
3924
class _VersionedFileChecker(object):
3926
def __init__(self, repository, text_key_references=None):
3927
self.repository = repository
3928
self.text_index = self.repository._generate_text_key_index(
3929
text_key_references=text_key_references)
3931
def calculate_file_version_parents(self, text_key):
3932
"""Calculate the correct parents for a file version according to
3935
parent_keys = self.text_index[text_key]
3936
if parent_keys == [_mod_revision.NULL_REVISION]:
3938
return tuple(parent_keys)
3940
def check_file_version_parents(self, texts, progress_bar=None):
3941
"""Check the parents stored in a versioned file are correct.
3943
It also detects file versions that are not referenced by their
3944
corresponding revision's inventory.
3946
:returns: A tuple of (wrong_parents, dangling_file_versions).
3947
wrong_parents is a dict mapping {revision_id: (stored_parents,
3948
correct_parents)} for each revision_id where the stored parents
3949
are not correct. dangling_file_versions is a set of (file_id,
3950
revision_id) tuples for versions that are present in this versioned
3951
file, but not used by the corresponding inventory.
3954
self.file_ids = set([file_id for file_id, _ in
3955
self.text_index.iterkeys()])
3956
# text keys is now grouped by file_id
3957
n_weaves = len(self.file_ids)
3958
files_in_revisions = {}
3959
revisions_of_files = {}
3960
n_versions = len(self.text_index)
3961
progress_bar.update('loading text store', 0, n_versions)
3962
parent_map = self.repository.texts.get_parent_map(self.text_index)
3963
# On unlistable transports this could well be empty/error...
3964
text_keys = self.repository.texts.keys()
3965
unused_keys = frozenset(text_keys) - set(self.text_index)
3966
for num, key in enumerate(self.text_index.iterkeys()):
3967
if progress_bar is not None:
3968
progress_bar.update('checking text graph', num, n_versions)
3969
correct_parents = self.calculate_file_version_parents(key)
3971
knit_parents = parent_map[key]
3972
except errors.RevisionNotPresent:
3975
if correct_parents != knit_parents:
3976
wrong_parents[key] = (knit_parents, correct_parents)
3977
return wrong_parents, unused_keys
3980
def _old_get_graph(repository, revision_id):
3981
"""DO NOT USE. That is all. I'm serious."""
3982
graph = repository.get_graph()
3983
revision_graph = dict(((key, value) for key, value in
3984
graph.iter_ancestry([revision_id]) if value is not None))
3985
return _strip_NULL_ghosts(revision_graph)
3988
def _strip_NULL_ghosts(revision_graph):
3989
"""Also don't use this. more compatibility code for unmigrated clients."""
3990
# Filter ghosts, and null:
3991
if _mod_revision.NULL_REVISION in revision_graph:
3992
del revision_graph[_mod_revision.NULL_REVISION]
3993
for key, parents in revision_graph.items():
3994
revision_graph[key] = tuple(parent for parent in parents if parent
3996
return revision_graph
3999
class StreamSink(object):
4000
"""An object that can insert a stream into a repository.
4002
This interface handles the complexity of reserialising inventories and
4003
revisions from different formats, and allows unidirectional insertion into
4004
stacked repositories without looking for the missing basis parents
4008
def __init__(self, target_repo):
4009
self.target_repo = target_repo
4011
def insert_stream(self, stream, src_format, resume_tokens):
4012
"""Insert a stream's content into the target repository.
4014
:param src_format: a bzr repository format.
4016
:return: a list of resume tokens and an iterable of keys additional
4017
items required before the insertion can be completed.
4019
self.target_repo.lock_write()
4022
self.target_repo.resume_write_group(resume_tokens)
4025
self.target_repo.start_write_group()
4028
# locked_insert_stream performs a commit|suspend.
4029
return self._locked_insert_stream(stream, src_format, is_resume)
4031
self.target_repo.abort_write_group(suppress_errors=True)
4034
self.target_repo.unlock()
4036
def _locked_insert_stream(self, stream, src_format, is_resume):
4037
to_serializer = self.target_repo._format._serializer
4038
src_serializer = src_format._serializer
4040
if to_serializer == src_serializer:
4041
# If serializers match and the target is a pack repository, set the
4042
# write cache size on the new pack. This avoids poor performance
4043
# on transports where append is unbuffered (such as
4044
# RemoteTransport). This is safe to do because nothing should read
4045
# back from the target repository while a stream with matching
4046
# serialization is being inserted.
4047
# The exception is that a delta record from the source that should
4048
# be a fulltext may need to be expanded by the target (see
4049
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4050
# explicitly flush any buffered writes first in that rare case.
4052
new_pack = self.target_repo._pack_collection._new_pack
4053
except AttributeError:
4054
# Not a pack repository
4057
new_pack.set_write_cache_size(1024*1024)
4058
for substream_type, substream in stream:
4059
if substream_type == 'texts':
4060
self.target_repo.texts.insert_record_stream(substream)
4061
elif substream_type == 'inventories':
4062
if src_serializer == to_serializer:
4063
self.target_repo.inventories.insert_record_stream(
4066
self._extract_and_insert_inventories(
4067
substream, src_serializer)
4068
elif substream_type == 'chk_bytes':
4069
# XXX: This doesn't support conversions, as it assumes the
4070
# conversion was done in the fetch code.
4071
self.target_repo.chk_bytes.insert_record_stream(substream)
4072
elif substream_type == 'revisions':
4073
# This may fallback to extract-and-insert more often than
4074
# required if the serializers are different only in terms of
4076
if src_serializer == to_serializer:
4077
self.target_repo.revisions.insert_record_stream(
4080
self._extract_and_insert_revisions(substream,
4082
elif substream_type == 'signatures':
4083
self.target_repo.signatures.insert_record_stream(substream)
4085
raise AssertionError('kaboom! %s' % (substream_type,))
4086
# Done inserting data, and the missing_keys calculations will try to
4087
# read back from the inserted data, so flush the writes to the new pack
4088
# (if this is pack format).
4089
if new_pack is not None:
4090
new_pack._write_data('', flush=True)
4091
# Find all the new revisions (including ones from resume_tokens)
4092
missing_keys = self.target_repo.get_missing_parent_inventories(
4093
check_for_missing_texts=is_resume)
4095
for prefix, versioned_file in (
4096
('texts', self.target_repo.texts),
4097
('inventories', self.target_repo.inventories),
4098
('revisions', self.target_repo.revisions),
4099
('signatures', self.target_repo.signatures),
4100
('chk_bytes', self.target_repo.chk_bytes),
4102
if versioned_file is None:
4104
missing_keys.update((prefix,) + key for key in
4105
versioned_file.get_missing_compression_parent_keys())
4106
except NotImplementedError:
4107
# cannot even attempt suspending, and missing would have failed
4108
# during stream insertion.
4109
missing_keys = set()
4112
# suspend the write group and tell the caller what we is
4113
# missing. We know we can suspend or else we would not have
4114
# entered this code path. (All repositories that can handle
4115
# missing keys can handle suspending a write group).
4116
write_group_tokens = self.target_repo.suspend_write_group()
4117
return write_group_tokens, missing_keys
4118
self.target_repo.commit_write_group()
4121
def _extract_and_insert_inventories(self, substream, serializer):
4122
"""Generate a new inventory versionedfile in target, converting data.
4124
The inventory is retrieved from the source, (deserializing it), and
4125
stored in the target (reserializing it in a different format).
4127
for record in substream:
4128
bytes = record.get_bytes_as('fulltext')
4129
revision_id = record.key[0]
4130
inv = serializer.read_inventory_from_string(bytes, revision_id)
4131
parents = [key[0] for key in record.parents]
4132
self.target_repo.add_inventory(revision_id, inv, parents)
4134
def _extract_and_insert_revisions(self, substream, serializer):
4135
for record in substream:
4136
bytes = record.get_bytes_as('fulltext')
4137
revision_id = record.key[0]
4138
rev = serializer.read_revision_from_string(bytes)
4139
if rev.revision_id != revision_id:
4140
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4141
self.target_repo.add_revision(revision_id, rev)
4144
if self.target_repo._format._fetch_reconcile:
4145
self.target_repo.reconcile()
4148
class StreamSource(object):
4149
"""A source of a stream for fetching between repositories."""
4151
def __init__(self, from_repository, to_format):
4152
"""Create a StreamSource streaming from from_repository."""
4153
self.from_repository = from_repository
4154
self.to_format = to_format
4156
def delta_on_metadata(self):
4157
"""Return True if delta's are permitted on metadata streams.
4159
That is on revisions and signatures.
4161
src_serializer = self.from_repository._format._serializer
4162
target_serializer = self.to_format._serializer
4163
return (self.to_format._fetch_uses_deltas and
4164
src_serializer == target_serializer)
4166
def _fetch_revision_texts(self, revs):
4167
# fetch signatures first and then the revision texts
4168
# may need to be a InterRevisionStore call here.
4169
from_sf = self.from_repository.signatures
4170
# A missing signature is just skipped.
4171
keys = [(rev_id,) for rev_id in revs]
4172
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4174
self.to_format._fetch_order,
4175
not self.to_format._fetch_uses_deltas))
4176
# If a revision has a delta, this is actually expanded inside the
4177
# insert_record_stream code now, which is an alternate fix for
4179
from_rf = self.from_repository.revisions
4180
revisions = from_rf.get_record_stream(
4182
self.to_format._fetch_order,
4183
not self.delta_on_metadata())
4184
return [('signatures', signatures), ('revisions', revisions)]
4186
def _generate_root_texts(self, revs):
4187
"""This will be called by __fetch between fetching weave texts and
4188
fetching the inventory weave.
4190
Subclasses should override this if they need to generate root texts
4191
after fetching weave texts.
4193
if self._rich_root_upgrade():
4195
return bzrlib.fetch.Inter1and2Helper(
4196
self.from_repository).generate_root_texts(revs)
4200
def get_stream(self, search):
4202
revs = search.get_keys()
4203
graph = self.from_repository.get_graph()
4204
revs = list(graph.iter_topo_order(revs))
4205
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4207
for knit_kind, file_id, revisions in data_to_fetch:
4208
if knit_kind != phase:
4210
# Make a new progress bar for this phase
4211
if knit_kind == "file":
4212
# Accumulate file texts
4213
text_keys.extend([(file_id, revision) for revision in
4215
elif knit_kind == "inventory":
4216
# Now copy the file texts.
4217
from_texts = self.from_repository.texts
4218
yield ('texts', from_texts.get_record_stream(
4219
text_keys, self.to_format._fetch_order,
4220
not self.to_format._fetch_uses_deltas))
4221
# Cause an error if a text occurs after we have done the
4224
# Before we process the inventory we generate the root
4225
# texts (if necessary) so that the inventories references
4227
for _ in self._generate_root_texts(revs):
4229
# NB: This currently reopens the inventory weave in source;
4230
# using a single stream interface instead would avoid this.
4231
from_weave = self.from_repository.inventories
4232
# we fetch only the referenced inventories because we do not
4233
# know for unselected inventories whether all their required
4234
# texts are present in the other repository - it could be
4236
for info in self._get_inventory_stream(revs):
4238
elif knit_kind == "signatures":
4239
# Nothing to do here; this will be taken care of when
4240
# _fetch_revision_texts happens.
4242
elif knit_kind == "revisions":
4243
for record in self._fetch_revision_texts(revs):
4246
raise AssertionError("Unknown knit kind %r" % knit_kind)
4248
def get_stream_for_missing_keys(self, missing_keys):
4249
# missing keys can only occur when we are byte copying and not
4250
# translating (because translation means we don't send
4251
# unreconstructable deltas ever).
4253
keys['texts'] = set()
4254
keys['revisions'] = set()
4255
keys['inventories'] = set()
4256
keys['chk_bytes'] = set()
4257
keys['signatures'] = set()
4258
for key in missing_keys:
4259
keys[key[0]].add(key[1:])
4260
if len(keys['revisions']):
4261
# If we allowed copying revisions at this point, we could end up
4262
# copying a revision without copying its required texts: a
4263
# violation of the requirements for repository integrity.
4264
raise AssertionError(
4265
'cannot copy revisions to fill in missing deltas %s' % (
4266
keys['revisions'],))
4267
for substream_kind, keys in keys.iteritems():
4268
vf = getattr(self.from_repository, substream_kind)
4269
if vf is None and keys:
4270
raise AssertionError(
4271
"cannot fill in keys for a versioned file we don't"
4272
" have: %s needs %s" % (substream_kind, keys))
4274
# No need to stream something we don't have
4276
# Ask for full texts always so that we don't need more round trips
4277
# after this stream.
4278
# Some of the missing keys are genuinely ghosts, so filter absent
4279
# records. The Sink is responsible for doing another check to
4280
# ensure that ghosts don't introduce missing data for future
4282
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4283
self.to_format._fetch_order, True))
4284
yield substream_kind, stream
4286
def inventory_fetch_order(self):
4287
if self._rich_root_upgrade():
4288
return 'topological'
4290
return self.to_format._fetch_order
4292
def _rich_root_upgrade(self):
4293
return (not self.from_repository._format.rich_root_data and
4294
self.to_format.rich_root_data)
4296
def _get_inventory_stream(self, revision_ids):
4297
from_format = self.from_repository._format
4298
if (from_format.supports_chks and self.to_format.supports_chks
4299
and (from_format._serializer == self.to_format._serializer)):
4300
# Both sides support chks, and they use the same serializer, so it
4301
# is safe to transmit the chk pages and inventory pages across
4303
return self._get_chk_inventory_stream(revision_ids)
4304
elif (not from_format.supports_chks):
4305
# Source repository doesn't support chks. So we can transmit the
4306
# inventories 'as-is' and either they are just accepted on the
4307
# target, or the Sink will properly convert it.
4308
return self._get_simple_inventory_stream(revision_ids)
4310
# XXX: Hack to make not-chk->chk fetch: copy the inventories as
4311
# inventories. Note that this should probably be done somehow
4312
# as part of bzrlib.repository.StreamSink. Except JAM couldn't
4313
# figure out how a non-chk repository could possibly handle
4314
# deserializing an inventory stream from a chk repo, as it
4315
# doesn't have a way to understand individual pages.
4316
return self._get_convertable_inventory_stream(revision_ids)
4318
def _get_simple_inventory_stream(self, revision_ids):
4319
from_weave = self.from_repository.inventories
4320
yield ('inventories', from_weave.get_record_stream(
4321
[(rev_id,) for rev_id in revision_ids],
4322
self.inventory_fetch_order(),
4323
not self.delta_on_metadata()))
4325
def _get_chk_inventory_stream(self, revision_ids):
4326
"""Fetch the inventory texts, along with the associated chk maps."""
4327
# We want an inventory outside of the search set, so that we can filter
4328
# out uninteresting chk pages. For now we use
4329
# _find_revision_outside_set, but if we had a Search with cut_revs, we
4330
# could use that instead.
4331
start_rev_id = self.from_repository._find_revision_outside_set(
4333
start_rev_key = (start_rev_id,)
4334
inv_keys_to_fetch = [(rev_id,) for rev_id in revision_ids]
4335
if start_rev_id != _mod_revision.NULL_REVISION:
4336
inv_keys_to_fetch.append((start_rev_id,))
4337
# Any repo that supports chk_bytes must also support out-of-order
4338
# insertion. At least, that is how we expect it to work
4339
# We use get_record_stream instead of iter_inventories because we want
4340
# to be able to insert the stream as well. We could instead fetch
4341
# allowing deltas, and then iter_inventories, but we don't know whether
4342
# source or target is more 'local' anway.
4343
inv_stream = self.from_repository.inventories.get_record_stream(
4344
inv_keys_to_fetch, 'unordered',
4345
True) # We need them as full-texts so we can find their references
4346
uninteresting_chk_roots = set()
4347
interesting_chk_roots = set()
4348
def filter_inv_stream(inv_stream):
4349
for idx, record in enumerate(inv_stream):
4350
### child_pb.update('fetch inv', idx, len(inv_keys_to_fetch))
4351
bytes = record.get_bytes_as('fulltext')
4352
chk_inv = inventory.CHKInventory.deserialise(
4353
self.from_repository.chk_bytes, bytes, record.key)
4354
if record.key == start_rev_key:
4355
uninteresting_chk_roots.add(chk_inv.id_to_entry.key())
4356
p_id_map = chk_inv.parent_id_basename_to_file_id
4357
if p_id_map is not None:
4358
uninteresting_chk_roots.add(p_id_map.key())
4361
interesting_chk_roots.add(chk_inv.id_to_entry.key())
4362
p_id_map = chk_inv.parent_id_basename_to_file_id
4363
if p_id_map is not None:
4364
interesting_chk_roots.add(p_id_map.key())
4365
### pb.update('fetch inventory', 0, 2)
4366
yield ('inventories', filter_inv_stream(inv_stream))
4367
# Now that we have worked out all of the interesting root nodes, grab
4368
# all of the interesting pages and insert them
4369
### pb.update('fetch inventory', 1, 2)
4370
interesting = chk_map.iter_interesting_nodes(
4371
self.from_repository.chk_bytes, interesting_chk_roots,
4372
uninteresting_chk_roots)
4373
def to_stream_adapter():
4374
"""Adapt the iter_interesting_nodes result to a single stream.
4376
iter_interesting_nodes returns records as it processes them, along
4377
with keys. However, we only want to return the records themselves.
4379
for record, items in interesting:
4380
if record is not None:
4382
# XXX: We could instead call get_record_stream(records.keys())
4383
# ATM, this will always insert the records as fulltexts, and
4384
# requires that you can hang on to records once you have gone
4385
# on to the next one. Further, it causes the target to
4386
# recompress the data. Testing shows it to be faster than
4387
# requesting the records again, though.
4388
yield ('chk_bytes', to_stream_adapter())
4389
### pb.update('fetch inventory', 2, 2)
4391
def _get_convertable_inventory_stream(self, revision_ids):
4392
# XXX: One of source or target is using chks, and they don't have
4393
# compatible serializations. The StreamSink code expects to be
4394
# able to convert on the target, so we need to put
4395
# bytes-on-the-wire that can be converted
4396
yield ('inventories', self._stream_invs_as_fulltexts(revision_ids))
4398
def _stream_invs_as_fulltexts(self, revision_ids):
4399
from_repo = self.from_repository
4400
from_serializer = from_repo._format._serializer
4401
revision_keys = [(rev_id,) for rev_id in revision_ids]
4402
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4403
for inv in self.from_repository.iter_inventories(revision_ids):
4404
# XXX: This is a bit hackish, but it works. Basically,
4405
# CHKSerializer 'accidentally' supports
4406
# read/write_inventory_to_string, even though that is never
4407
# the format that is stored on disk. It *does* give us a
4408
# single string representation for an inventory, so live with
4410
# This would be far better if we had a 'serialized inventory
4411
# delta' form. Then we could use 'inventory._make_delta', and
4412
# transmit that. This would both be faster to generate, and
4413
# result in fewer bytes-on-the-wire.
4414
as_bytes = from_serializer.write_inventory_to_string(inv)
4415
key = (inv.revision_id,)
4416
parent_keys = parent_map.get(key, ())
4417
yield versionedfile.FulltextContentFactory(
4418
key, parent_keys, None, as_bytes)
added
removed
bzrlib/repository.py
