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)
972
self._check_fallback_repository(repository)
973
self._fallback_repositories.append(repository)
974
self.texts.add_fallback_versioned_files(repository.texts)
975
self.inventories.add_fallback_versioned_files(repository.inventories)
976
self.revisions.add_fallback_versioned_files(repository.revisions)
977
self.signatures.add_fallback_versioned_files(repository.signatures)
978
if self.chk_bytes is not None:
979
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
981
def _check_fallback_repository(self, repository):
982
"""Check that this repository can fallback to repository safely.
984
Raise an error if not.
986
:param repository: A repository to fallback to.
988
return InterRepository._assert_same_model(self, repository)
990
def add_inventory(self, revision_id, inv, parents):
991
"""Add the inventory inv to the repository as revision_id.
993
:param parents: The revision ids of the parents that revision_id
994
is known to have and are in the repository already.
996
:returns: The validator(which is a sha1 digest, though what is sha'd is
997
repository format specific) of the serialized inventory.
999
if not self.is_in_write_group():
1000
raise AssertionError("%r not in write group" % (self,))
1001
_mod_revision.check_not_reserved_id(revision_id)
1002
if not (inv.revision_id is None or inv.revision_id == revision_id):
1003
raise AssertionError(
1004
"Mismatch between inventory revision"
1005
" id and insertion revid (%r, %r)"
1006
% (inv.revision_id, revision_id))
1007
if inv.root is None:
1008
raise AssertionError()
1009
return self._add_inventory_checked(revision_id, inv, parents)
1011
def _add_inventory_checked(self, revision_id, inv, parents):
1012
"""Add inv to the repository after checking the inputs.
1014
This function can be overridden to allow different inventory styles.
1016
:seealso: add_inventory, for the contract.
1018
inv_lines = self._serialise_inventory_to_lines(inv)
1019
return self._inventory_add_lines(revision_id, parents,
1020
inv_lines, check_content=False)
1022
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1023
parents, basis_inv=None, propagate_caches=False):
1024
"""Add a new inventory expressed as a delta against another revision.
1026
:param basis_revision_id: The inventory id the delta was created
1027
against. (This does not have to be a direct parent.)
1028
:param delta: The inventory delta (see Inventory.apply_delta for
1030
:param new_revision_id: The revision id that the inventory is being
1032
:param parents: The revision ids of the parents that revision_id is
1033
known to have and are in the repository already. These are supplied
1034
for repositories that depend on the inventory graph for revision
1035
graph access, as well as for those that pun ancestry with delta
1037
:param basis_inv: The basis inventory if it is already known,
1039
:param propagate_caches: If True, the caches for this inventory are
1040
copied to and updated for the result if possible.
1042
:returns: (validator, new_inv)
1043
The validator(which is a sha1 digest, though what is sha'd is
1044
repository format specific) of the serialized inventory, and the
1045
resulting inventory.
1047
if not self.is_in_write_group():
1048
raise AssertionError("%r not in write group" % (self,))
1049
_mod_revision.check_not_reserved_id(new_revision_id)
1050
basis_tree = self.revision_tree(basis_revision_id)
1051
basis_tree.lock_read()
1053
# Note that this mutates the inventory of basis_tree, which not all
1054
# inventory implementations may support: A better idiom would be to
1055
# return a new inventory, but as there is no revision tree cache in
1056
# repository this is safe for now - RBC 20081013
1057
if basis_inv is None:
1058
basis_inv = basis_tree.inventory
1059
basis_inv.apply_delta(delta)
1060
basis_inv.revision_id = new_revision_id
1061
return (self.add_inventory(new_revision_id, basis_inv, parents),
1066
def _inventory_add_lines(self, revision_id, parents, lines,
1067
check_content=True):
1068
"""Store lines in inv_vf and return the sha1 of the inventory."""
1069
parents = [(parent,) for parent in parents]
1070
return self.inventories.add_lines((revision_id,), parents, lines,
1071
check_content=check_content)[0]
1073
def add_revision(self, revision_id, rev, inv=None, config=None):
1074
"""Add rev to the revision store as revision_id.
1076
:param revision_id: the revision id to use.
1077
:param rev: The revision object.
1078
:param inv: The inventory for the revision. if None, it will be looked
1079
up in the inventory storer
1080
:param config: If None no digital signature will be created.
1081
If supplied its signature_needed method will be used
1082
to determine if a signature should be made.
1084
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1086
_mod_revision.check_not_reserved_id(revision_id)
1087
if config is not None and config.signature_needed():
1089
inv = self.get_inventory(revision_id)
1090
plaintext = Testament(rev, inv).as_short_text()
1091
self.store_revision_signature(
1092
gpg.GPGStrategy(config), plaintext, revision_id)
1093
# check inventory present
1094
if not self.inventories.get_parent_map([(revision_id,)]):
1096
raise errors.WeaveRevisionNotPresent(revision_id,
1099
# yes, this is not suitable for adding with ghosts.
1100
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1103
key = (revision_id,)
1104
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1105
self._add_revision(rev)
1107
def _add_revision(self, revision):
1108
text = self._serializer.write_revision_to_string(revision)
1109
key = (revision.revision_id,)
1110
parents = tuple((parent,) for parent in revision.parent_ids)
1111
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1113
def all_revision_ids(self):
1114
"""Returns a list of all the revision ids in the repository.
1116
This is conceptually deprecated because code should generally work on
1117
the graph reachable from a particular revision, and ignore any other
1118
revisions that might be present. There is no direct replacement
1121
if 'evil' in debug.debug_flags:
1122
mutter_callsite(2, "all_revision_ids is linear with history.")
1123
return self._all_revision_ids()
1125
def _all_revision_ids(self):
1126
"""Returns a list of all the revision ids in the repository.
1128
These are in as much topological order as the underlying store can
1131
raise NotImplementedError(self._all_revision_ids)
1133
def break_lock(self):
1134
"""Break a lock if one is present from another instance.
1136
Uses the ui factory to ask for confirmation if the lock may be from
1139
self.control_files.break_lock()
1142
def _eliminate_revisions_not_present(self, revision_ids):
1143
"""Check every revision id in revision_ids to see if we have it.
1145
Returns a set of the present revisions.
1148
graph = self.get_graph()
1149
parent_map = graph.get_parent_map(revision_ids)
1150
# The old API returned a list, should this actually be a set?
1151
return parent_map.keys()
1154
def create(a_bzrdir):
1155
"""Construct the current default format repository in a_bzrdir."""
1156
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1158
def __init__(self, _format, a_bzrdir, control_files):
1159
"""instantiate a Repository.
1161
:param _format: The format of the repository on disk.
1162
:param a_bzrdir: The BzrDir of the repository.
1164
In the future we will have a single api for all stores for
1165
getting file texts, inventories and revisions, then
1166
this construct will accept instances of those things.
1168
super(Repository, self).__init__()
1169
self._format = _format
1170
# the following are part of the public API for Repository:
1171
self.bzrdir = a_bzrdir
1172
self.control_files = control_files
1173
self._transport = control_files._transport
1174
self.base = self._transport.base
1176
self._reconcile_does_inventory_gc = True
1177
self._reconcile_fixes_text_parents = False
1178
self._reconcile_backsup_inventory = True
1179
# not right yet - should be more semantically clear ?
1181
# TODO: make sure to construct the right store classes, etc, depending
1182
# on whether escaping is required.
1183
self._warn_if_deprecated()
1184
self._write_group = None
1185
# Additional places to query for data.
1186
self._fallback_repositories = []
1187
# An InventoryEntry cache, used during deserialization
1188
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1191
return '%s(%r)' % (self.__class__.__name__,
1194
def has_same_location(self, other):
1195
"""Returns a boolean indicating if this repository is at the same
1196
location as another repository.
1198
This might return False even when two repository objects are accessing
1199
the same physical repository via different URLs.
1201
if self.__class__ is not other.__class__:
1203
return (self._transport.base == other._transport.base)
1205
def is_in_write_group(self):
1206
"""Return True if there is an open write group.
1208
:seealso: start_write_group.
1210
return self._write_group is not None
1212
def is_locked(self):
1213
return self.control_files.is_locked()
1215
def is_write_locked(self):
1216
"""Return True if this object is write locked."""
1217
return self.is_locked() and self.control_files._lock_mode == 'w'
1219
def lock_write(self, token=None):
1220
"""Lock this repository for writing.
1222
This causes caching within the repository obejct to start accumlating
1223
data during reads, and allows a 'write_group' to be obtained. Write
1224
groups must be used for actual data insertion.
1226
:param token: if this is already locked, then lock_write will fail
1227
unless the token matches the existing lock.
1228
:returns: a token if this instance supports tokens, otherwise None.
1229
:raises TokenLockingNotSupported: when a token is given but this
1230
instance doesn't support using token locks.
1231
:raises MismatchedToken: if the specified token doesn't match the token
1232
of the existing lock.
1233
:seealso: start_write_group.
1235
A token should be passed in if you know that you have locked the object
1236
some other way, and need to synchronise this object's state with that
1239
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1241
locked = self.is_locked()
1242
result = self.control_files.lock_write(token=token)
1243
for repo in self._fallback_repositories:
1244
# Writes don't affect fallback repos
1247
self._refresh_data()
1250
def lock_read(self):
1251
locked = self.is_locked()
1252
self.control_files.lock_read()
1253
for repo in self._fallback_repositories:
1256
self._refresh_data()
1258
def get_physical_lock_status(self):
1259
return self.control_files.get_physical_lock_status()
1261
def leave_lock_in_place(self):
1262
"""Tell this repository not to release the physical lock when this
1265
If lock_write doesn't return a token, then this method is not supported.
1267
self.control_files.leave_in_place()
1269
def dont_leave_lock_in_place(self):
1270
"""Tell this repository to release the physical lock when this
1271
object is unlocked, even if it didn't originally acquire it.
1273
If lock_write doesn't return a token, then this method is not supported.
1275
self.control_files.dont_leave_in_place()
1278
def gather_stats(self, revid=None, committers=None):
1279
"""Gather statistics from a revision id.
1281
:param revid: The revision id to gather statistics from, if None, then
1282
no revision specific statistics are gathered.
1283
:param committers: Optional parameter controlling whether to grab
1284
a count of committers from the revision specific statistics.
1285
:return: A dictionary of statistics. Currently this contains:
1286
committers: The number of committers if requested.
1287
firstrev: A tuple with timestamp, timezone for the penultimate left
1288
most ancestor of revid, if revid is not the NULL_REVISION.
1289
latestrev: A tuple with timestamp, timezone for revid, if revid is
1290
not the NULL_REVISION.
1291
revisions: The total revision count in the repository.
1292
size: An estimate disk size of the repository in bytes.
1295
if revid and committers:
1296
result['committers'] = 0
1297
if revid and revid != _mod_revision.NULL_REVISION:
1299
all_committers = set()
1300
revisions = self.get_ancestry(revid)
1301
# pop the leading None
1303
first_revision = None
1305
# ignore the revisions in the middle - just grab first and last
1306
revisions = revisions[0], revisions[-1]
1307
for revision in self.get_revisions(revisions):
1308
if not first_revision:
1309
first_revision = revision
1311
all_committers.add(revision.committer)
1312
last_revision = revision
1314
result['committers'] = len(all_committers)
1315
result['firstrev'] = (first_revision.timestamp,
1316
first_revision.timezone)
1317
result['latestrev'] = (last_revision.timestamp,
1318
last_revision.timezone)
1320
# now gather global repository information
1321
# XXX: This is available for many repos regardless of listability.
1322
if self.bzrdir.root_transport.listable():
1323
# XXX: do we want to __define len__() ?
1324
# Maybe the versionedfiles object should provide a different
1325
# method to get the number of keys.
1326
result['revisions'] = len(self.revisions.keys())
1327
# result['size'] = t
1330
def find_branches(self, using=False):
1331
"""Find branches underneath this repository.
1333
This will include branches inside other branches.
1335
:param using: If True, list only branches using this repository.
1337
if using and not self.is_shared():
1339
return [self.bzrdir.open_branch()]
1340
except errors.NotBranchError:
1342
class Evaluator(object):
1345
self.first_call = True
1347
def __call__(self, bzrdir):
1348
# On the first call, the parameter is always the bzrdir
1349
# containing the current repo.
1350
if not self.first_call:
1352
repository = bzrdir.open_repository()
1353
except errors.NoRepositoryPresent:
1356
return False, (None, repository)
1357
self.first_call = False
1359
value = (bzrdir.open_branch(), None)
1360
except errors.NotBranchError:
1361
value = (None, None)
1365
for branch, repository in bzrdir.BzrDir.find_bzrdirs(
1366
self.bzrdir.root_transport, evaluate=Evaluator()):
1367
if branch is not None:
1368
branches.append(branch)
1369
if not using and repository is not None:
1370
branches.extend(repository.find_branches())
1374
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
1375
"""Return the revision ids that other has that this does not.
1377
These are returned in topological order.
1379
revision_id: only return revision ids included by revision_id.
1381
return InterRepository.get(other, self).search_missing_revision_ids(
1382
revision_id, find_ghosts)
1386
"""Open the repository rooted at base.
1388
For instance, if the repository is at URL/.bzr/repository,
1389
Repository.open(URL) -> a Repository instance.
1391
control = bzrdir.BzrDir.open(base)
1392
return control.open_repository()
1394
def copy_content_into(self, destination, revision_id=None):
1395
"""Make a complete copy of the content in self into destination.
1397
This is a destructive operation! Do not use it on existing
1400
return InterRepository.get(self, destination).copy_content(revision_id)
1402
def commit_write_group(self):
1403
"""Commit the contents accrued within the current write group.
1405
:seealso: start_write_group.
1407
if self._write_group is not self.get_transaction():
1408
# has an unlock or relock occured ?
1409
raise errors.BzrError('mismatched lock context %r and '
1411
(self.get_transaction(), self._write_group))
1412
self._commit_write_group()
1413
self._write_group = None
1415
def _commit_write_group(self):
1416
"""Template method for per-repository write group cleanup.
1418
This is called before the write group is considered to be
1419
finished and should ensure that all data handed to the repository
1420
for writing during the write group is safely committed (to the
1421
extent possible considering file system caching etc).
1424
def suspend_write_group(self):
1425
raise errors.UnsuspendableWriteGroup(self)
1427
def refresh_data(self):
1428
"""Re-read any data needed to to synchronise with disk.
1430
This method is intended to be called after another repository instance
1431
(such as one used by a smart server) has inserted data into the
1432
repository. It may not be called during a write group, but may be
1433
called at any other time.
1435
if self.is_in_write_group():
1436
raise errors.InternalBzrError(
1437
"May not refresh_data while in a write group.")
1438
self._refresh_data()
1440
def resume_write_group(self, tokens):
1441
if not self.is_write_locked():
1442
raise errors.NotWriteLocked(self)
1443
if self._write_group:
1444
raise errors.BzrError('already in a write group')
1445
self._resume_write_group(tokens)
1446
# so we can detect unlock/relock - the write group is now entered.
1447
self._write_group = self.get_transaction()
1449
def _resume_write_group(self, tokens):
1450
raise errors.UnsuspendableWriteGroup(self)
1452
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1454
"""Fetch the content required to construct revision_id from source.
1456
If revision_id is None and fetch_spec is None, then all content is
1459
fetch() may not be used when the repository is in a write group -
1460
either finish the current write group before using fetch, or use
1461
fetch before starting the write group.
1463
:param find_ghosts: Find and copy revisions in the source that are
1464
ghosts in the target (and not reachable directly by walking out to
1465
the first-present revision in target from revision_id).
1466
:param revision_id: If specified, all the content needed for this
1467
revision ID will be copied to the target. Fetch will determine for
1468
itself which content needs to be copied.
1469
:param fetch_spec: If specified, a SearchResult or
1470
PendingAncestryResult that describes which revisions to copy. This
1471
allows copying multiple heads at once. Mutually exclusive with
1474
if fetch_spec is not None and revision_id is not None:
1475
raise AssertionError(
1476
"fetch_spec and revision_id are mutually exclusive.")
1477
if self.is_in_write_group():
1478
raise errors.InternalBzrError(
1479
"May not fetch while in a write group.")
1480
# fast path same-url fetch operations
1481
if self.has_same_location(source) and fetch_spec is None:
1482
# check that last_revision is in 'from' and then return a
1484
if (revision_id is not None and
1485
not _mod_revision.is_null(revision_id)):
1486
self.get_revision(revision_id)
1488
# if there is no specific appropriate InterRepository, this will get
1489
# the InterRepository base class, which raises an
1490
# IncompatibleRepositories when asked to fetch.
1491
inter = InterRepository.get(source, self)
1492
return inter.fetch(revision_id=revision_id, pb=pb,
1493
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1495
def create_bundle(self, target, base, fileobj, format=None):
1496
return serializer.write_bundle(self, target, base, fileobj, format)
1498
def get_commit_builder(self, branch, parents, config, timestamp=None,
1499
timezone=None, committer=None, revprops=None,
1501
"""Obtain a CommitBuilder for this repository.
1503
:param branch: Branch to commit to.
1504
:param parents: Revision ids of the parents of the new revision.
1505
:param config: Configuration to use.
1506
:param timestamp: Optional timestamp recorded for commit.
1507
:param timezone: Optional timezone for timestamp.
1508
:param committer: Optional committer to set for commit.
1509
:param revprops: Optional dictionary of revision properties.
1510
:param revision_id: Optional revision id.
1512
result = self._commit_builder_class(self, parents, config,
1513
timestamp, timezone, committer, revprops, revision_id)
1514
self.start_write_group()
1518
if (self.control_files._lock_count == 1 and
1519
self.control_files._lock_mode == 'w'):
1520
if self._write_group is not None:
1521
self.abort_write_group()
1522
self.control_files.unlock()
1523
raise errors.BzrError(
1524
'Must end write groups before releasing write locks.')
1525
self.control_files.unlock()
1526
if self.control_files._lock_count == 0:
1527
self._inventory_entry_cache.clear()
1528
for repo in self._fallback_repositories:
1532
def clone(self, a_bzrdir, revision_id=None):
1533
"""Clone this repository into a_bzrdir using the current format.
1535
Currently no check is made that the format of this repository and
1536
the bzrdir format are compatible. FIXME RBC 20060201.
1538
:return: The newly created destination repository.
1540
# TODO: deprecate after 0.16; cloning this with all its settings is
1541
# probably not very useful -- mbp 20070423
1542
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1543
self.copy_content_into(dest_repo, revision_id)
1546
def start_write_group(self):
1547
"""Start a write group in the repository.
1549
Write groups are used by repositories which do not have a 1:1 mapping
1550
between file ids and backend store to manage the insertion of data from
1551
both fetch and commit operations.
1553
A write lock is required around the start_write_group/commit_write_group
1554
for the support of lock-requiring repository formats.
1556
One can only insert data into a repository inside a write group.
1560
if not self.is_write_locked():
1561
raise errors.NotWriteLocked(self)
1562
if self._write_group:
1563
raise errors.BzrError('already in a write group')
1564
self._start_write_group()
1565
# so we can detect unlock/relock - the write group is now entered.
1566
self._write_group = self.get_transaction()
1568
def _start_write_group(self):
1569
"""Template method for per-repository write group startup.
1571
This is called before the write group is considered to be
1576
def sprout(self, to_bzrdir, revision_id=None):
1577
"""Create a descendent repository for new development.
1579
Unlike clone, this does not copy the settings of the repository.
1581
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1582
dest_repo.fetch(self, revision_id=revision_id)
1585
def _create_sprouting_repo(self, a_bzrdir, shared):
1586
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1587
# use target default format.
1588
dest_repo = a_bzrdir.create_repository()
1590
# Most control formats need the repository to be specifically
1591
# created, but on some old all-in-one formats it's not needed
1593
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1594
except errors.UninitializableFormat:
1595
dest_repo = a_bzrdir.open_repository()
1598
def _get_sink(self):
1599
"""Return a sink for streaming into this repository."""
1600
return StreamSink(self)
1602
def _get_source(self, to_format):
1603
"""Return a source for streaming from this repository."""
1604
return StreamSource(self, to_format)
1607
def has_revision(self, revision_id):
1608
"""True if this repository has a copy of the revision."""
1609
return revision_id in self.has_revisions((revision_id,))
1612
def has_revisions(self, revision_ids):
1613
"""Probe to find out the presence of multiple revisions.
1615
:param revision_ids: An iterable of revision_ids.
1616
:return: A set of the revision_ids that were present.
1618
parent_map = self.revisions.get_parent_map(
1619
[(rev_id,) for rev_id in revision_ids])
1621
if _mod_revision.NULL_REVISION in revision_ids:
1622
result.add(_mod_revision.NULL_REVISION)
1623
result.update([key[0] for key in parent_map])
1627
def get_revision(self, revision_id):
1628
"""Return the Revision object for a named revision."""
1629
return self.get_revisions([revision_id])[0]
1632
def get_revision_reconcile(self, revision_id):
1633
"""'reconcile' helper routine that allows access to a revision always.
1635
This variant of get_revision does not cross check the weave graph
1636
against the revision one as get_revision does: but it should only
1637
be used by reconcile, or reconcile-alike commands that are correcting
1638
or testing the revision graph.
1640
return self._get_revisions([revision_id])[0]
1643
def get_revisions(self, revision_ids):
1644
"""Get many revisions at once."""
1645
return self._get_revisions(revision_ids)
1648
def _get_revisions(self, revision_ids):
1649
"""Core work logic to get many revisions without sanity checks."""
1650
for rev_id in revision_ids:
1651
if not rev_id or not isinstance(rev_id, basestring):
1652
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1653
keys = [(key,) for key in revision_ids]
1654
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1656
for record in stream:
1657
if record.storage_kind == 'absent':
1658
raise errors.NoSuchRevision(self, record.key[0])
1659
text = record.get_bytes_as('fulltext')
1660
rev = self._serializer.read_revision_from_string(text)
1661
revs[record.key[0]] = rev
1662
return [revs[revid] for revid in revision_ids]
1665
def get_revision_xml(self, revision_id):
1666
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1667
# would have already do it.
1668
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1669
# TODO: this can't just be replaced by:
1670
# return self._serializer.write_revision_to_string(
1671
# self.get_revision(revision_id))
1672
# as cStringIO preservers the encoding unlike write_revision_to_string
1673
# or some other call down the path.
1674
rev = self.get_revision(revision_id)
1675
rev_tmp = cStringIO.StringIO()
1676
# the current serializer..
1677
self._serializer.write_revision(rev, rev_tmp)
1679
return rev_tmp.getvalue()
1681
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1682
"""Produce a generator of revision deltas.
1684
Note that the input is a sequence of REVISIONS, not revision_ids.
1685
Trees will be held in memory until the generator exits.
1686
Each delta is relative to the revision's lefthand predecessor.
1688
:param specific_fileids: if not None, the result is filtered
1689
so that only those file-ids, their parents and their
1690
children are included.
1692
# Get the revision-ids of interest
1693
required_trees = set()
1694
for revision in revisions:
1695
required_trees.add(revision.revision_id)
1696
required_trees.update(revision.parent_ids[:1])
1698
# Get the matching filtered trees. Note that it's more
1699
# efficient to pass filtered trees to changes_from() rather
1700
# than doing the filtering afterwards. changes_from() could
1701
# arguably do the filtering itself but it's path-based, not
1702
# file-id based, so filtering before or afterwards is
1704
if specific_fileids is None:
1705
trees = dict((t.get_revision_id(), t) for
1706
t in self.revision_trees(required_trees))
1708
trees = dict((t.get_revision_id(), t) for
1709
t in self._filtered_revision_trees(required_trees,
1712
# Calculate the deltas
1713
for revision in revisions:
1714
if not revision.parent_ids:
1715
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1717
old_tree = trees[revision.parent_ids[0]]
1718
yield trees[revision.revision_id].changes_from(old_tree)
1721
def get_revision_delta(self, revision_id, specific_fileids=None):
1722
"""Return the delta for one revision.
1724
The delta is relative to the left-hand predecessor of the
1727
:param specific_fileids: if not None, the result is filtered
1728
so that only those file-ids, their parents and their
1729
children are included.
1731
r = self.get_revision(revision_id)
1732
return list(self.get_deltas_for_revisions([r],
1733
specific_fileids=specific_fileids))[0]
1736
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1737
signature = gpg_strategy.sign(plaintext)
1738
self.add_signature_text(revision_id, signature)
1741
def add_signature_text(self, revision_id, signature):
1742
self.signatures.add_lines((revision_id,), (),
1743
osutils.split_lines(signature))
1745
def find_text_key_references(self):
1746
"""Find the text key references within the repository.
1748
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1749
to whether they were referred to by the inventory of the
1750
revision_id that they contain. The inventory texts from all present
1751
revision ids are assessed to generate this report.
1753
revision_keys = self.revisions.keys()
1754
w = self.inventories
1755
pb = ui.ui_factory.nested_progress_bar()
1757
return self._find_text_key_references_from_xml_inventory_lines(
1758
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1762
def _find_text_key_references_from_xml_inventory_lines(self,
1764
"""Core routine for extracting references to texts from inventories.
1766
This performs the translation of xml lines to revision ids.
1768
:param line_iterator: An iterator of lines, origin_version_id
1769
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1770
to whether they were referred to by the inventory of the
1771
revision_id that they contain. Note that if that revision_id was
1772
not part of the line_iterator's output then False will be given -
1773
even though it may actually refer to that key.
1775
if not self._serializer.support_altered_by_hack:
1776
raise AssertionError(
1777
"_find_text_key_references_from_xml_inventory_lines only "
1778
"supported for branches which store inventory as unnested xml"
1779
", not on %r" % self)
1782
# this code needs to read every new line in every inventory for the
1783
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
1784
# not present in one of those inventories is unnecessary but not
1785
# harmful because we are filtering by the revision id marker in the
1786
# inventory lines : we only select file ids altered in one of those
1787
# revisions. We don't need to see all lines in the inventory because
1788
# only those added in an inventory in rev X can contain a revision=X
1790
unescape_revid_cache = {}
1791
unescape_fileid_cache = {}
1793
# jam 20061218 In a big fetch, this handles hundreds of thousands
1794
# of lines, so it has had a lot of inlining and optimizing done.
1795
# Sorry that it is a little bit messy.
1796
# Move several functions to be local variables, since this is a long
1798
search = self._file_ids_altered_regex.search
1799
unescape = _unescape_xml
1800
setdefault = result.setdefault
1801
for line, line_key in line_iterator:
1802
match = search(line)
1805
# One call to match.group() returning multiple items is quite a
1806
# bit faster than 2 calls to match.group() each returning 1
1807
file_id, revision_id = match.group('file_id', 'revision_id')
1809
# Inlining the cache lookups helps a lot when you make 170,000
1810
# lines and 350k ids, versus 8.4 unique ids.
1811
# Using a cache helps in 2 ways:
1812
# 1) Avoids unnecessary decoding calls
1813
# 2) Re-uses cached strings, which helps in future set and
1815
# (2) is enough that removing encoding entirely along with
1816
# the cache (so we are using plain strings) results in no
1817
# performance improvement.
1819
revision_id = unescape_revid_cache[revision_id]
1821
unescaped = unescape(revision_id)
1822
unescape_revid_cache[revision_id] = unescaped
1823
revision_id = unescaped
1825
# Note that unconditionally unescaping means that we deserialise
1826
# every fileid, which for general 'pull' is not great, but we don't
1827
# really want to have some many fulltexts that this matters anyway.
1830
file_id = unescape_fileid_cache[file_id]
1832
unescaped = unescape(file_id)
1833
unescape_fileid_cache[file_id] = unescaped
1836
key = (file_id, revision_id)
1837
setdefault(key, False)
1838
if revision_id == line_key[-1]:
1842
def _inventory_xml_lines_for_keys(self, keys):
1843
"""Get a line iterator of the sort needed for findind references.
1845
Not relevant for non-xml inventory repositories.
1847
Ghosts in revision_keys are ignored.
1849
:param revision_keys: The revision keys for the inventories to inspect.
1850
:return: An iterator over (inventory line, revid) for the fulltexts of
1851
all of the xml inventories specified by revision_keys.
1853
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1854
for record in stream:
1855
if record.storage_kind != 'absent':
1856
chunks = record.get_bytes_as('chunked')
1857
revid = record.key[-1]
1858
lines = osutils.chunks_to_lines(chunks)
1862
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1864
"""Helper routine for fileids_altered_by_revision_ids.
1866
This performs the translation of xml lines to revision ids.
1868
:param line_iterator: An iterator of lines, origin_version_id
1869
:param revision_ids: The revision ids to filter for. This should be a
1870
set or other type which supports efficient __contains__ lookups, as
1871
the revision id from each parsed line will be looked up in the
1872
revision_ids filter.
1873
:return: a dictionary mapping altered file-ids to an iterable of
1874
revision_ids. Each altered file-ids has the exact revision_ids that
1875
altered it listed explicitly.
1877
seen = set(self._find_text_key_references_from_xml_inventory_lines(
1878
line_iterator).iterkeys())
1879
# Note that revision_ids are revision keys.
1880
parent_maps = self.revisions.get_parent_map(revision_ids)
1882
map(parents.update, parent_maps.itervalues())
1883
parents.difference_update(revision_ids)
1884
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
1885
self._inventory_xml_lines_for_keys(parents)))
1886
new_keys = seen - parent_seen
1888
setdefault = result.setdefault
1889
for key in new_keys:
1890
setdefault(key[0], set()).add(key[-1])
1893
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
1894
"""Find the file ids and versions affected by revisions.
1896
:param revisions: an iterable containing revision ids.
1897
:param _inv_weave: The inventory weave from this repository or None.
1898
If None, the inventory weave will be opened automatically.
1899
:return: a dictionary mapping altered file-ids to an iterable of
1900
revision_ids. Each altered file-ids has the exact revision_ids that
1901
altered it listed explicitly.
1903
selected_keys = set((revid,) for revid in revision_ids)
1904
w = _inv_weave or self.inventories
1905
pb = ui.ui_factory.nested_progress_bar()
1907
return self._find_file_ids_from_xml_inventory_lines(
1908
w.iter_lines_added_or_present_in_keys(
1909
selected_keys, pb=pb),
1914
def iter_files_bytes(self, desired_files):
1915
"""Iterate through file versions.
1917
Files will not necessarily be returned in the order they occur in
1918
desired_files. No specific order is guaranteed.
1920
Yields pairs of identifier, bytes_iterator. identifier is an opaque
1921
value supplied by the caller as part of desired_files. It should
1922
uniquely identify the file version in the caller's context. (Examples:
1923
an index number or a TreeTransform trans_id.)
1925
bytes_iterator is an iterable of bytestrings for the file. The
1926
kind of iterable and length of the bytestrings are unspecified, but for
1927
this implementation, it is a list of bytes produced by
1928
VersionedFile.get_record_stream().
1930
:param desired_files: a list of (file_id, revision_id, identifier)
1934
for file_id, revision_id, callable_data in desired_files:
1935
text_keys[(file_id, revision_id)] = callable_data
1936
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
1937
if record.storage_kind == 'absent':
1938
raise errors.RevisionNotPresent(record.key, self)
1939
yield text_keys[record.key], record.get_bytes_as('chunked')
1941
def _generate_text_key_index(self, text_key_references=None,
1943
"""Generate a new text key index for the repository.
1945
This is an expensive function that will take considerable time to run.
1947
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
1948
list of parents, also text keys. When a given key has no parents,
1949
the parents list will be [NULL_REVISION].
1951
# All revisions, to find inventory parents.
1952
if ancestors is None:
1953
graph = self.get_graph()
1954
ancestors = graph.get_parent_map(self.all_revision_ids())
1955
if text_key_references is None:
1956
text_key_references = self.find_text_key_references()
1957
pb = ui.ui_factory.nested_progress_bar()
1959
return self._do_generate_text_key_index(ancestors,
1960
text_key_references, pb)
1964
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
1965
"""Helper for _generate_text_key_index to avoid deep nesting."""
1966
revision_order = tsort.topo_sort(ancestors)
1967
invalid_keys = set()
1969
for revision_id in revision_order:
1970
revision_keys[revision_id] = set()
1971
text_count = len(text_key_references)
1972
# a cache of the text keys to allow reuse; costs a dict of all the
1973
# keys, but saves a 2-tuple for every child of a given key.
1975
for text_key, valid in text_key_references.iteritems():
1977
invalid_keys.add(text_key)
1979
revision_keys[text_key[1]].add(text_key)
1980
text_key_cache[text_key] = text_key
1981
del text_key_references
1983
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
1984
NULL_REVISION = _mod_revision.NULL_REVISION
1985
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
1986
# too small for large or very branchy trees. However, for 55K path
1987
# trees, it would be easy to use too much memory trivially. Ideally we
1988
# could gauge this by looking at available real memory etc, but this is
1989
# always a tricky proposition.
1990
inventory_cache = lru_cache.LRUCache(10)
1991
batch_size = 10 # should be ~150MB on a 55K path tree
1992
batch_count = len(revision_order) / batch_size + 1
1994
pb.update("Calculating text parents", processed_texts, text_count)
1995
for offset in xrange(batch_count):
1996
to_query = revision_order[offset * batch_size:(offset + 1) *
2000
for rev_tree in self.revision_trees(to_query):
2001
revision_id = rev_tree.get_revision_id()
2002
parent_ids = ancestors[revision_id]
2003
for text_key in revision_keys[revision_id]:
2004
pb.update("Calculating text parents", processed_texts)
2005
processed_texts += 1
2006
candidate_parents = []
2007
for parent_id in parent_ids:
2008
parent_text_key = (text_key[0], parent_id)
2010
check_parent = parent_text_key not in \
2011
revision_keys[parent_id]
2013
# the parent parent_id is a ghost:
2014
check_parent = False
2015
# truncate the derived graph against this ghost.
2016
parent_text_key = None
2018
# look at the parent commit details inventories to
2019
# determine possible candidates in the per file graph.
2022
inv = inventory_cache[parent_id]
2024
inv = self.revision_tree(parent_id).inventory
2025
inventory_cache[parent_id] = inv
2027
parent_entry = inv[text_key[0]]
2028
except (KeyError, errors.NoSuchId):
2030
if parent_entry is not None:
2032
text_key[0], parent_entry.revision)
2034
parent_text_key = None
2035
if parent_text_key is not None:
2036
candidate_parents.append(
2037
text_key_cache[parent_text_key])
2038
parent_heads = text_graph.heads(candidate_parents)
2039
new_parents = list(parent_heads)
2040
new_parents.sort(key=lambda x:candidate_parents.index(x))
2041
if new_parents == []:
2042
new_parents = [NULL_REVISION]
2043
text_index[text_key] = new_parents
2045
for text_key in invalid_keys:
2046
text_index[text_key] = [NULL_REVISION]
2049
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2050
"""Get an iterable listing the keys of all the data introduced by a set
2053
The keys will be ordered so that the corresponding items can be safely
2054
fetched and inserted in that order.
2056
:returns: An iterable producing tuples of (knit-kind, file-id,
2057
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2058
'revisions'. file-id is None unless knit-kind is 'file'.
2060
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2063
for result in self._find_non_file_keys_to_fetch(revision_ids):
2066
def _find_file_keys_to_fetch(self, revision_ids, pb):
2067
# XXX: it's a bit weird to control the inventory weave caching in this
2068
# generator. Ideally the caching would be done in fetch.py I think. Or
2069
# maybe this generator should explicitly have the contract that it
2070
# should not be iterated until the previously yielded item has been
2072
inv_w = self.inventories
2074
# file ids that changed
2075
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2077
num_file_ids = len(file_ids)
2078
for file_id, altered_versions in file_ids.iteritems():
2080
pb.update("fetch texts", count, num_file_ids)
2082
yield ("file", file_id, altered_versions)
2084
def _find_non_file_keys_to_fetch(self, revision_ids):
2086
yield ("inventory", None, revision_ids)
2089
# XXX: Note ATM no callers actually pay attention to this return
2090
# instead they just use the list of revision ids and ignore
2091
# missing sigs. Consider removing this work entirely
2092
revisions_with_signatures = set(self.signatures.get_parent_map(
2093
[(r,) for r in revision_ids]))
2094
revisions_with_signatures = set(
2095
[r for (r,) in revisions_with_signatures])
2096
revisions_with_signatures.intersection_update(revision_ids)
2097
yield ("signatures", None, revisions_with_signatures)
2100
yield ("revisions", None, revision_ids)
2103
def get_inventory(self, revision_id):
2104
"""Get Inventory object by revision id."""
2105
return self.iter_inventories([revision_id]).next()
2107
def iter_inventories(self, revision_ids):
2108
"""Get many inventories by revision_ids.
2110
This will buffer some or all of the texts used in constructing the
2111
inventories in memory, but will only parse a single inventory at a
2114
:param revision_ids: The expected revision ids of the inventories.
2115
:return: An iterator of inventories.
2117
if ((None in revision_ids)
2118
or (_mod_revision.NULL_REVISION in revision_ids)):
2119
raise ValueError('cannot get null revision inventory')
2120
return self._iter_inventories(revision_ids)
2122
def _iter_inventories(self, revision_ids):
2123
"""single-document based inventory iteration."""
2124
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2125
yield self.deserialise_inventory(revision_id, text)
2127
def _iter_inventory_xmls(self, revision_ids):
2128
keys = [(revision_id,) for revision_id in revision_ids]
2129
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2131
for record in stream:
2132
if record.storage_kind != 'absent':
2133
text_chunks[record.key] = record.get_bytes_as('chunked')
2135
raise errors.NoSuchRevision(self, record.key)
2137
chunks = text_chunks.pop(key)
2138
yield ''.join(chunks), key[-1]
2140
def deserialise_inventory(self, revision_id, xml):
2141
"""Transform the xml into an inventory object.
2143
:param revision_id: The expected revision id of the inventory.
2144
:param xml: A serialised inventory.
2146
result = self._serializer.read_inventory_from_string(xml, revision_id,
2147
entry_cache=self._inventory_entry_cache)
2148
if result.revision_id != revision_id:
2149
raise AssertionError('revision id mismatch %s != %s' % (
2150
result.revision_id, revision_id))
2153
def serialise_inventory(self, inv):
2154
return self._serializer.write_inventory_to_string(inv)
2156
def _serialise_inventory_to_lines(self, inv):
2157
return self._serializer.write_inventory_to_lines(inv)
2159
def get_serializer_format(self):
2160
return self._serializer.format_num
2163
def get_inventory_xml(self, revision_id):
2164
"""Get inventory XML as a file object."""
2165
texts = self._iter_inventory_xmls([revision_id])
2167
text, revision_id = texts.next()
2168
except StopIteration:
2169
raise errors.HistoryMissing(self, 'inventory', revision_id)
2173
def get_inventory_sha1(self, revision_id):
2174
"""Return the sha1 hash of the inventory entry
2176
return self.get_revision(revision_id).inventory_sha1
2178
def iter_reverse_revision_history(self, revision_id):
2179
"""Iterate backwards through revision ids in the lefthand history
2181
:param revision_id: The revision id to start with. All its lefthand
2182
ancestors will be traversed.
2184
graph = self.get_graph()
2185
next_id = revision_id
2187
if next_id in (None, _mod_revision.NULL_REVISION):
2190
# Note: The following line may raise KeyError in the event of
2191
# truncated history. We decided not to have a try:except:raise
2192
# RevisionNotPresent here until we see a use for it, because of the
2193
# cost in an inner loop that is by its very nature O(history).
2194
# Robert Collins 20080326
2195
parents = graph.get_parent_map([next_id])[next_id]
2196
if len(parents) == 0:
2199
next_id = parents[0]
2202
def get_revision_inventory(self, revision_id):
2203
"""Return inventory of a past revision."""
2204
# TODO: Unify this with get_inventory()
2205
# bzr 0.0.6 and later imposes the constraint that the inventory_id
2206
# must be the same as its revision, so this is trivial.
2207
if revision_id is None:
2208
# This does not make sense: if there is no revision,
2209
# then it is the current tree inventory surely ?!
2210
# and thus get_root_id() is something that looks at the last
2211
# commit on the branch, and the get_root_id is an inventory check.
2212
raise NotImplementedError
2213
# return Inventory(self.get_root_id())
2215
return self.get_inventory(revision_id)
2217
def is_shared(self):
2218
"""Return True if this repository is flagged as a shared repository."""
2219
raise NotImplementedError(self.is_shared)
2222
def reconcile(self, other=None, thorough=False):
2223
"""Reconcile this repository."""
2224
from bzrlib.reconcile import RepoReconciler
2225
reconciler = RepoReconciler(self, thorough=thorough)
2226
reconciler.reconcile()
2229
def _refresh_data(self):
2230
"""Helper called from lock_* to ensure coherency with disk.
2232
The default implementation does nothing; it is however possible
2233
for repositories to maintain loaded indices across multiple locks
2234
by checking inside their implementation of this method to see
2235
whether their indices are still valid. This depends of course on
2236
the disk format being validatable in this manner. This method is
2237
also called by the refresh_data() public interface to cause a refresh
2238
to occur while in a write lock so that data inserted by a smart server
2239
push operation is visible on the client's instance of the physical
2244
def revision_tree(self, revision_id):
2245
"""Return Tree for a revision on this branch.
2247
`revision_id` may be NULL_REVISION for the empty tree revision.
2249
revision_id = _mod_revision.ensure_null(revision_id)
2250
# TODO: refactor this to use an existing revision object
2251
# so we don't need to read it in twice.
2252
if revision_id == _mod_revision.NULL_REVISION:
2253
return RevisionTree(self, Inventory(root_id=None),
2254
_mod_revision.NULL_REVISION)
2256
inv = self.get_revision_inventory(revision_id)
2257
return RevisionTree(self, inv, revision_id)
2259
def revision_trees(self, revision_ids):
2260
"""Return Trees for revisions in this repository.
2262
:param revision_ids: a sequence of revision-ids;
2263
a revision-id may not be None or 'null:'
2265
inventories = self.iter_inventories(revision_ids)
2266
for inv in inventories:
2267
yield RevisionTree(self, inv, inv.revision_id)
2269
def _filtered_revision_trees(self, revision_ids, file_ids):
2270
"""Return Tree for a revision on this branch with only some files.
2272
:param revision_ids: a sequence of revision-ids;
2273
a revision-id may not be None or 'null:'
2274
:param file_ids: if not None, the result is filtered
2275
so that only those file-ids, their parents and their
2276
children are included.
2278
inventories = self.iter_inventories(revision_ids)
2279
for inv in inventories:
2280
# Should we introduce a FilteredRevisionTree class rather
2281
# than pre-filter the inventory here?
2282
filtered_inv = inv.filter(file_ids)
2283
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2286
def get_ancestry(self, revision_id, topo_sorted=True):
2287
"""Return a list of revision-ids integrated by a revision.
2289
The first element of the list is always None, indicating the origin
2290
revision. This might change when we have history horizons, or
2291
perhaps we should have a new API.
2293
This is topologically sorted.
2295
if _mod_revision.is_null(revision_id):
2297
if not self.has_revision(revision_id):
2298
raise errors.NoSuchRevision(self, revision_id)
2299
graph = self.get_graph()
2301
search = graph._make_breadth_first_searcher([revision_id])
2304
found, ghosts = search.next_with_ghosts()
2305
except StopIteration:
2308
if _mod_revision.NULL_REVISION in keys:
2309
keys.remove(_mod_revision.NULL_REVISION)
2311
parent_map = graph.get_parent_map(keys)
2312
keys = tsort.topo_sort(parent_map)
2313
return [None] + list(keys)
2316
"""Compress the data within the repository.
2318
This operation only makes sense for some repository types. For other
2319
types it should be a no-op that just returns.
2321
This stub method does not require a lock, but subclasses should use
2322
@needs_write_lock as this is a long running call its reasonable to
2323
implicitly lock for the user.
2326
def get_transaction(self):
2327
return self.control_files.get_transaction()
2329
def get_parent_map(self, revision_ids):
2330
"""See graph._StackedParentsProvider.get_parent_map"""
2331
# revisions index works in keys; this just works in revisions
2332
# therefore wrap and unwrap
2335
for revision_id in revision_ids:
2336
if revision_id == _mod_revision.NULL_REVISION:
2337
result[revision_id] = ()
2338
elif revision_id is None:
2339
raise ValueError('get_parent_map(None) is not valid')
2341
query_keys.append((revision_id ,))
2342
for ((revision_id,), parent_keys) in \
2343
self.revisions.get_parent_map(query_keys).iteritems():
2345
result[revision_id] = tuple(parent_revid
2346
for (parent_revid,) in parent_keys)
2348
result[revision_id] = (_mod_revision.NULL_REVISION,)
2351
def _make_parents_provider(self):
2354
def get_graph(self, other_repository=None):
2355
"""Return the graph walker for this repository format"""
2356
parents_provider = self._make_parents_provider()
2357
if (other_repository is not None and
2358
not self.has_same_location(other_repository)):
2359
parents_provider = graph._StackedParentsProvider(
2360
[parents_provider, other_repository._make_parents_provider()])
2361
return graph.Graph(parents_provider)
2363
def _get_versioned_file_checker(self, text_key_references=None):
2364
"""Return an object suitable for checking versioned files.
2366
:param text_key_references: if non-None, an already built
2367
dictionary mapping text keys ((fileid, revision_id) tuples)
2368
to whether they were referred to by the inventory of the
2369
revision_id that they contain. If None, this will be
2372
return _VersionedFileChecker(self,
2373
text_key_references=text_key_references)
2375
def revision_ids_to_search_result(self, result_set):
2376
"""Convert a set of revision ids to a graph SearchResult."""
2377
result_parents = set()
2378
for parents in self.get_graph().get_parent_map(
2379
result_set).itervalues():
2380
result_parents.update(parents)
2381
included_keys = result_set.intersection(result_parents)
2382
start_keys = result_set.difference(included_keys)
2383
exclude_keys = result_parents.difference(result_set)
2384
result = graph.SearchResult(start_keys, exclude_keys,
2385
len(result_set), result_set)
2389
def set_make_working_trees(self, new_value):
2390
"""Set the policy flag for making working trees when creating branches.
2392
This only applies to branches that use this repository.
2394
The default is 'True'.
2395
:param new_value: True to restore the default, False to disable making
2398
raise NotImplementedError(self.set_make_working_trees)
2400
def make_working_trees(self):
2401
"""Returns the policy for making working trees on new branches."""
2402
raise NotImplementedError(self.make_working_trees)
2405
def sign_revision(self, revision_id, gpg_strategy):
2406
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2407
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2410
def has_signature_for_revision_id(self, revision_id):
2411
"""Query for a revision signature for revision_id in the repository."""
2412
if not self.has_revision(revision_id):
2413
raise errors.NoSuchRevision(self, revision_id)
2414
sig_present = (1 == len(
2415
self.signatures.get_parent_map([(revision_id,)])))
2419
def get_signature_text(self, revision_id):
2420
"""Return the text for a signature."""
2421
stream = self.signatures.get_record_stream([(revision_id,)],
2423
record = stream.next()
2424
if record.storage_kind == 'absent':
2425
raise errors.NoSuchRevision(self, revision_id)
2426
return record.get_bytes_as('fulltext')
2429
def check(self, revision_ids=None):
2430
"""Check consistency of all history of given revision_ids.
2432
Different repository implementations should override _check().
2434
:param revision_ids: A non-empty list of revision_ids whose ancestry
2435
will be checked. Typically the last revision_id of a branch.
2437
return self._check(revision_ids)
2439
def _check(self, revision_ids):
2440
result = check.Check(self)
2444
def _warn_if_deprecated(self):
2445
global _deprecation_warning_done
2446
if _deprecation_warning_done:
2448
_deprecation_warning_done = True
2449
warning("Format %s for %s is deprecated - please use 'bzr upgrade' to get better performance"
2450
% (self._format, self.bzrdir.transport.base))
2452
def supports_rich_root(self):
2453
return self._format.rich_root_data
2455
def _check_ascii_revisionid(self, revision_id, method):
2456
"""Private helper for ascii-only repositories."""
2457
# weave repositories refuse to store revisionids that are non-ascii.
2458
if revision_id is not None:
2459
# weaves require ascii revision ids.
2460
if isinstance(revision_id, unicode):
2462
revision_id.encode('ascii')
2463
except UnicodeEncodeError:
2464
raise errors.NonAsciiRevisionId(method, self)
2467
revision_id.decode('ascii')
2468
except UnicodeDecodeError:
2469
raise errors.NonAsciiRevisionId(method, self)
2471
def revision_graph_can_have_wrong_parents(self):
2472
"""Is it possible for this repository to have a revision graph with
2475
If True, then this repository must also implement
2476
_find_inconsistent_revision_parents so that check and reconcile can
2477
check for inconsistencies before proceeding with other checks that may
2478
depend on the revision index being consistent.
2480
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2483
# remove these delegates a while after bzr 0.15
2484
def __make_delegated(name, from_module):
2485
def _deprecated_repository_forwarder():
2486
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2487
% (name, from_module),
2490
m = __import__(from_module, globals(), locals(), [name])
2492
return getattr(m, name)
2493
except AttributeError:
2494
raise AttributeError('module %s has no name %s'
2496
globals()[name] = _deprecated_repository_forwarder
2499
'AllInOneRepository',
2500
'WeaveMetaDirRepository',
2501
'PreSplitOutRepositoryFormat',
2502
'RepositoryFormat4',
2503
'RepositoryFormat5',
2504
'RepositoryFormat6',
2505
'RepositoryFormat7',
2507
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2511
'RepositoryFormatKnit',
2512
'RepositoryFormatKnit1',
2514
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2517
def install_revision(repository, rev, revision_tree):
2518
"""Install all revision data into a repository."""
2519
install_revisions(repository, [(rev, revision_tree, None)])
2522
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2523
"""Install all revision data into a repository.
2525
Accepts an iterable of revision, tree, signature tuples. The signature
2528
repository.start_write_group()
2530
inventory_cache = lru_cache.LRUCache(10)
2531
for n, (revision, revision_tree, signature) in enumerate(iterable):
2532
_install_revision(repository, revision, revision_tree, signature,
2535
pb.update('Transferring revisions', n + 1, num_revisions)
2537
repository.abort_write_group()
2540
repository.commit_write_group()
2543
def _install_revision(repository, rev, revision_tree, signature,
2545
"""Install all revision data into a repository."""
2546
present_parents = []
2548
for p_id in rev.parent_ids:
2549
if repository.has_revision(p_id):
2550
present_parents.append(p_id)
2551
parent_trees[p_id] = repository.revision_tree(p_id)
2553
parent_trees[p_id] = repository.revision_tree(
2554
_mod_revision.NULL_REVISION)
2556
inv = revision_tree.inventory
2557
entries = inv.iter_entries()
2558
# backwards compatibility hack: skip the root id.
2559
if not repository.supports_rich_root():
2560
path, root = entries.next()
2561
if root.revision != rev.revision_id:
2562
raise errors.IncompatibleRevision(repr(repository))
2564
for path, ie in entries:
2565
text_keys[(ie.file_id, ie.revision)] = ie
2566
text_parent_map = repository.texts.get_parent_map(text_keys)
2567
missing_texts = set(text_keys) - set(text_parent_map)
2568
# Add the texts that are not already present
2569
for text_key in missing_texts:
2570
ie = text_keys[text_key]
2572
# FIXME: TODO: The following loop overlaps/duplicates that done by
2573
# commit to determine parents. There is a latent/real bug here where
2574
# the parents inserted are not those commit would do - in particular
2575
# they are not filtered by heads(). RBC, AB
2576
for revision, tree in parent_trees.iteritems():
2577
if ie.file_id not in tree:
2579
parent_id = tree.inventory[ie.file_id].revision
2580
if parent_id in text_parents:
2582
text_parents.append((ie.file_id, parent_id))
2583
lines = revision_tree.get_file(ie.file_id).readlines()
2584
repository.texts.add_lines(text_key, text_parents, lines)
2586
# install the inventory
2587
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2588
# Cache this inventory
2589
inventory_cache[rev.revision_id] = inv
2591
basis_inv = inventory_cache[rev.parent_ids[0]]
2593
repository.add_inventory(rev.revision_id, inv, present_parents)
2595
delta = inv._make_delta(basis_inv)
2596
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2597
rev.revision_id, present_parents)
2599
repository.add_inventory(rev.revision_id, inv, present_parents)
2600
except errors.RevisionAlreadyPresent:
2602
if signature is not None:
2603
repository.add_signature_text(rev.revision_id, signature)
2604
repository.add_revision(rev.revision_id, rev, inv)
2607
class MetaDirRepository(Repository):
2608
"""Repositories in the new meta-dir layout.
2610
:ivar _transport: Transport for access to repository control files,
2611
typically pointing to .bzr/repository.
2614
def __init__(self, _format, a_bzrdir, control_files):
2615
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2616
self._transport = control_files._transport
2618
def is_shared(self):
2619
"""Return True if this repository is flagged as a shared repository."""
2620
return self._transport.has('shared-storage')
2623
def set_make_working_trees(self, new_value):
2624
"""Set the policy flag for making working trees when creating branches.
2626
This only applies to branches that use this repository.
2628
The default is 'True'.
2629
:param new_value: True to restore the default, False to disable making
2634
self._transport.delete('no-working-trees')
2635
except errors.NoSuchFile:
2638
self._transport.put_bytes('no-working-trees', '',
2639
mode=self.bzrdir._get_file_mode())
2641
def make_working_trees(self):
2642
"""Returns the policy for making working trees on new branches."""
2643
return not self._transport.has('no-working-trees')
2646
class MetaDirVersionedFileRepository(MetaDirRepository):
2647
"""Repositories in a meta-dir, that work via versioned file objects."""
2649
def __init__(self, _format, a_bzrdir, control_files):
2650
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2654
network_format_registry = registry.FormatRegistry()
2655
"""Registry of formats indexed by their network name.
2657
The network name for a repository format is an identifier that can be used when
2658
referring to formats with smart server operations. See
2659
RepositoryFormat.network_name() for more detail.
2663
format_registry = registry.FormatRegistry(network_format_registry)
2664
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
2666
This can contain either format instances themselves, or classes/factories that
2667
can be called to obtain one.
2671
#####################################################################
2672
# Repository Formats
2674
class RepositoryFormat(object):
2675
"""A repository format.
2677
Formats provide four things:
2678
* An initialization routine to construct repository data on disk.
2679
* a optional format string which is used when the BzrDir supports
2681
* an open routine which returns a Repository instance.
2682
* A network name for referring to the format in smart server RPC
2685
There is one and only one Format subclass for each on-disk format. But
2686
there can be one Repository subclass that is used for several different
2687
formats. The _format attribute on a Repository instance can be used to
2688
determine the disk format.
2690
Formats are placed in a registry by their format string for reference
2691
during opening. These should be subclasses of RepositoryFormat for
2694
Once a format is deprecated, just deprecate the initialize and open
2695
methods on the format class. Do not deprecate the object, as the
2696
object may be created even when a repository instance hasn't been
2699
Common instance attributes:
2700
_matchingbzrdir - the bzrdir format that the repository format was
2701
originally written to work with. This can be used if manually
2702
constructing a bzrdir and repository, or more commonly for test suite
2706
# Set to True or False in derived classes. True indicates that the format
2707
# supports ghosts gracefully.
2708
supports_ghosts = None
2709
# Can this repository be given external locations to lookup additional
2710
# data. Set to True or False in derived classes.
2711
supports_external_lookups = None
2712
# Does this format support CHK bytestring lookups. Set to True or False in
2714
supports_chks = None
2715
# Should commit add an inventory, or an inventory delta to the repository.
2716
_commit_inv_deltas = True
2717
# What order should fetch operations request streams in?
2718
# The default is unordered as that is the cheapest for an origin to
2720
_fetch_order = 'unordered'
2721
# Does this repository format use deltas that can be fetched as-deltas ?
2722
# (E.g. knits, where the knit deltas can be transplanted intact.
2723
# We default to False, which will ensure that enough data to get
2724
# a full text out of any fetch stream will be grabbed.
2725
_fetch_uses_deltas = False
2726
# Should fetch trigger a reconcile after the fetch? Only needed for
2727
# some repository formats that can suffer internal inconsistencies.
2728
_fetch_reconcile = False
2729
# Does this format have < O(tree_size) delta generation. Used to hint what
2730
# code path for commit, amongst other things.
2734
return "<%s>" % self.__class__.__name__
2736
def __eq__(self, other):
2737
# format objects are generally stateless
2738
return isinstance(other, self.__class__)
2740
def __ne__(self, other):
2741
return not self == other
2744
def find_format(klass, a_bzrdir):
2745
"""Return the format for the repository object in a_bzrdir.
2747
This is used by bzr native formats that have a "format" file in
2748
the repository. Other methods may be used by different types of
2752
transport = a_bzrdir.get_repository_transport(None)
2753
format_string = transport.get("format").read()
2754
return format_registry.get(format_string)
2755
except errors.NoSuchFile:
2756
raise errors.NoRepositoryPresent(a_bzrdir)
2758
raise errors.UnknownFormatError(format=format_string,
2762
def register_format(klass, format):
2763
format_registry.register(format.get_format_string(), format)
2766
def unregister_format(klass, format):
2767
format_registry.remove(format.get_format_string())
2770
def get_default_format(klass):
2771
"""Return the current default format."""
2772
from bzrlib import bzrdir
2773
return bzrdir.format_registry.make_bzrdir('default').repository_format
2775
def get_format_string(self):
2776
"""Return the ASCII format string that identifies this format.
2778
Note that in pre format ?? repositories the format string is
2779
not permitted nor written to disk.
2781
raise NotImplementedError(self.get_format_string)
2783
def get_format_description(self):
2784
"""Return the short description for this format."""
2785
raise NotImplementedError(self.get_format_description)
2787
# TODO: this shouldn't be in the base class, it's specific to things that
2788
# use weaves or knits -- mbp 20070207
2789
def _get_versioned_file_store(self,
2794
versionedfile_class=None,
2795
versionedfile_kwargs={},
2797
if versionedfile_class is None:
2798
versionedfile_class = self._versionedfile_class
2799
weave_transport = control_files._transport.clone(name)
2800
dir_mode = control_files._dir_mode
2801
file_mode = control_files._file_mode
2802
return VersionedFileStore(weave_transport, prefixed=prefixed,
2804
file_mode=file_mode,
2805
versionedfile_class=versionedfile_class,
2806
versionedfile_kwargs=versionedfile_kwargs,
2809
def initialize(self, a_bzrdir, shared=False):
2810
"""Initialize a repository of this format in a_bzrdir.
2812
:param a_bzrdir: The bzrdir to put the new repository in it.
2813
:param shared: The repository should be initialized as a sharable one.
2814
:returns: The new repository object.
2816
This may raise UninitializableFormat if shared repository are not
2817
compatible the a_bzrdir.
2819
raise NotImplementedError(self.initialize)
2821
def is_supported(self):
2822
"""Is this format supported?
2824
Supported formats must be initializable and openable.
2825
Unsupported formats may not support initialization or committing or
2826
some other features depending on the reason for not being supported.
2830
def network_name(self):
2831
"""A simple byte string uniquely identifying this format for RPC calls.
2833
MetaDir repository formats use their disk format string to identify the
2834
repository over the wire. All in one formats such as bzr < 0.8, and
2835
foreign formats like svn/git and hg should use some marker which is
2836
unique and immutable.
2838
raise NotImplementedError(self.network_name)
2840
def check_conversion_target(self, target_format):
2841
raise NotImplementedError(self.check_conversion_target)
2843
def open(self, a_bzrdir, _found=False):
2844
"""Return an instance of this format for the bzrdir a_bzrdir.
2846
_found is a private parameter, do not use it.
2848
raise NotImplementedError(self.open)
2851
class MetaDirRepositoryFormat(RepositoryFormat):
2852
"""Common base class for the new repositories using the metadir layout."""
2854
rich_root_data = False
2855
supports_tree_reference = False
2856
supports_external_lookups = False
2859
def _matchingbzrdir(self):
2860
matching = bzrdir.BzrDirMetaFormat1()
2861
matching.repository_format = self
2865
super(MetaDirRepositoryFormat, self).__init__()
2867
def _create_control_files(self, a_bzrdir):
2868
"""Create the required files and the initial control_files object."""
2869
# FIXME: RBC 20060125 don't peek under the covers
2870
# NB: no need to escape relative paths that are url safe.
2871
repository_transport = a_bzrdir.get_repository_transport(self)
2872
control_files = lockable_files.LockableFiles(repository_transport,
2873
'lock', lockdir.LockDir)
2874
control_files.create_lock()
2875
return control_files
2877
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
2878
"""Upload the initial blank content."""
2879
control_files = self._create_control_files(a_bzrdir)
2880
control_files.lock_write()
2881
transport = control_files._transport
2883
utf8_files += [('shared-storage', '')]
2885
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
2886
for (filename, content_stream) in files:
2887
transport.put_file(filename, content_stream,
2888
mode=a_bzrdir._get_file_mode())
2889
for (filename, content_bytes) in utf8_files:
2890
transport.put_bytes_non_atomic(filename, content_bytes,
2891
mode=a_bzrdir._get_file_mode())
2893
control_files.unlock()
2895
def network_name(self):
2896
"""Metadir formats have matching disk and network format strings."""
2897
return self.get_format_string()
2900
# Pre-0.8 formats that don't have a disk format string (because they are
2901
# versioned by the matching control directory). We use the control directories
2902
# disk format string as a key for the network_name because they meet the
2903
# constraints (simple string, unique, immutable).
2904
network_format_registry.register_lazy(
2905
"Bazaar-NG branch, format 5\n",
2906
'bzrlib.repofmt.weaverepo',
2907
'RepositoryFormat5',
2909
network_format_registry.register_lazy(
2910
"Bazaar-NG branch, format 6\n",
2911
'bzrlib.repofmt.weaverepo',
2912
'RepositoryFormat6',
2915
# formats which have no format string are not discoverable or independently
2916
# creatable on disk, so are not registered in format_registry. They're
2917
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
2918
# needed, it's constructed directly by the BzrDir. Non-native formats where
2919
# the repository is not separately opened are similar.
2921
format_registry.register_lazy(
2922
'Bazaar-NG Repository format 7',
2923
'bzrlib.repofmt.weaverepo',
2927
format_registry.register_lazy(
2928
'Bazaar-NG Knit Repository Format 1',
2929
'bzrlib.repofmt.knitrepo',
2930
'RepositoryFormatKnit1',
2933
format_registry.register_lazy(
2934
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
2935
'bzrlib.repofmt.knitrepo',
2936
'RepositoryFormatKnit3',
2939
format_registry.register_lazy(
2940
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
2941
'bzrlib.repofmt.knitrepo',
2942
'RepositoryFormatKnit4',
2945
# Pack-based formats. There is one format for pre-subtrees, and one for
2946
# post-subtrees to allow ease of testing.
2947
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
2948
format_registry.register_lazy(
2949
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
2950
'bzrlib.repofmt.pack_repo',
2951
'RepositoryFormatKnitPack1',
2953
format_registry.register_lazy(
2954
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
2955
'bzrlib.repofmt.pack_repo',
2956
'RepositoryFormatKnitPack3',
2958
format_registry.register_lazy(
2959
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
2960
'bzrlib.repofmt.pack_repo',
2961
'RepositoryFormatKnitPack4',
2963
format_registry.register_lazy(
2964
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
2965
'bzrlib.repofmt.pack_repo',
2966
'RepositoryFormatKnitPack5',
2968
format_registry.register_lazy(
2969
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
2970
'bzrlib.repofmt.pack_repo',
2971
'RepositoryFormatKnitPack5RichRoot',
2973
format_registry.register_lazy(
2974
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
2975
'bzrlib.repofmt.pack_repo',
2976
'RepositoryFormatKnitPack5RichRootBroken',
2978
format_registry.register_lazy(
2979
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
2980
'bzrlib.repofmt.pack_repo',
2981
'RepositoryFormatKnitPack6',
2983
format_registry.register_lazy(
2984
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
2985
'bzrlib.repofmt.pack_repo',
2986
'RepositoryFormatKnitPack6RichRoot',
2989
# Development formats.
2990
# Obsolete but kept pending a CHK based subtree format.
2991
format_registry.register_lazy(
2992
("Bazaar development format 2 with subtree support "
2993
"(needs bzr.dev from before 1.8)\n"),
2994
'bzrlib.repofmt.pack_repo',
2995
'RepositoryFormatPackDevelopment2Subtree',
2998
# 1.14->1.16 go below here
2999
format_registry.register_lazy(
3000
'Bazaar development format - group compression and chk inventory'
3001
' (needs bzr.dev from 1.14)\n',
3002
'bzrlib.repofmt.groupcompress_repo',
3003
'RepositoryFormatCHK1',
3007
class InterRepository(InterObject):
3008
"""This class represents operations taking place between two repositories.
3010
Its instances have methods like copy_content and fetch, and contain
3011
references to the source and target repositories these operations can be
3014
Often we will provide convenience methods on 'repository' which carry out
3015
operations with another repository - they will always forward to
3016
InterRepository.get(other).method_name(parameters).
3019
_walk_to_common_revisions_batch_size = 50
3021
"""The available optimised InterRepository types."""
3024
def copy_content(self, revision_id=None):
3025
"""Make a complete copy of the content in self into destination.
3027
This is a destructive operation! Do not use it on existing
3030
:param revision_id: Only copy the content needed to construct
3031
revision_id and its parents.
3034
self.target.set_make_working_trees(self.source.make_working_trees())
3035
except NotImplementedError:
3037
self.target.fetch(self.source, revision_id=revision_id)
3040
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3042
"""Fetch the content required to construct revision_id.
3044
The content is copied from self.source to self.target.
3046
:param revision_id: if None all content is copied, if NULL_REVISION no
3048
:param pb: optional progress bar to use for progress reports. If not
3049
provided a default one will be created.
3052
from bzrlib.fetch import RepoFetcher
3053
f = RepoFetcher(to_repository=self.target,
3054
from_repository=self.source,
3055
last_revision=revision_id,
3056
fetch_spec=fetch_spec,
3057
pb=pb, find_ghosts=find_ghosts)
3059
def _walk_to_common_revisions(self, revision_ids):
3060
"""Walk out from revision_ids in source to revisions target has.
3062
:param revision_ids: The start point for the search.
3063
:return: A set of revision ids.
3065
target_graph = self.target.get_graph()
3066
revision_ids = frozenset(revision_ids)
3067
# Fast path for the case where all the revisions are already in the
3069
# (Although this does incur an extra round trip for the
3070
# fairly common case where the target doesn't already have the revision
3072
if set(target_graph.get_parent_map(revision_ids)) == revision_ids:
3073
return graph.SearchResult(revision_ids, set(), 0, set())
3074
missing_revs = set()
3075
source_graph = self.source.get_graph()
3076
# ensure we don't pay silly lookup costs.
3077
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3078
null_set = frozenset([_mod_revision.NULL_REVISION])
3079
searcher_exhausted = False
3083
# Iterate the searcher until we have enough next_revs
3084
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3086
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3087
next_revs.update(next_revs_part)
3088
ghosts.update(ghosts_part)
3089
except StopIteration:
3090
searcher_exhausted = True
3092
# If there are ghosts in the source graph, and the caller asked for
3093
# them, make sure that they are present in the target.
3094
# We don't care about other ghosts as we can't fetch them and
3095
# haven't been asked to.
3096
ghosts_to_check = set(revision_ids.intersection(ghosts))
3097
revs_to_get = set(next_revs).union(ghosts_to_check)
3099
have_revs = set(target_graph.get_parent_map(revs_to_get))
3100
# we always have NULL_REVISION present.
3101
have_revs = have_revs.union(null_set)
3102
# Check if the target is missing any ghosts we need.
3103
ghosts_to_check.difference_update(have_revs)
3105
# One of the caller's revision_ids is a ghost in both the
3106
# source and the target.
3107
raise errors.NoSuchRevision(
3108
self.source, ghosts_to_check.pop())
3109
missing_revs.update(next_revs - have_revs)
3110
# Because we may have walked past the original stop point, make
3111
# sure everything is stopped
3112
stop_revs = searcher.find_seen_ancestors(have_revs)
3113
searcher.stop_searching_any(stop_revs)
3114
if searcher_exhausted:
3116
return searcher.get_result()
3119
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3120
"""Return the revision ids that source has that target does not.
3122
:param revision_id: only return revision ids included by this
3124
:param find_ghosts: If True find missing revisions in deep history
3125
rather than just finding the surface difference.
3126
:return: A bzrlib.graph.SearchResult.
3128
# stop searching at found target revisions.
3129
if not find_ghosts and revision_id is not None:
3130
return self._walk_to_common_revisions([revision_id])
3131
# generic, possibly worst case, slow code path.
3132
target_ids = set(self.target.all_revision_ids())
3133
if revision_id is not None:
3134
source_ids = self.source.get_ancestry(revision_id)
3135
if source_ids[0] is not None:
3136
raise AssertionError()
3139
source_ids = self.source.all_revision_ids()
3140
result_set = set(source_ids).difference(target_ids)
3141
return self.source.revision_ids_to_search_result(result_set)
3144
def _same_model(source, target):
3145
"""True if source and target have the same data representation.
3147
Note: this is always called on the base class; overriding it in a
3148
subclass will have no effect.
3151
InterRepository._assert_same_model(source, target)
3153
except errors.IncompatibleRepositories, e:
3157
def _assert_same_model(source, target):
3158
"""Raise an exception if two repositories do not use the same model.
3160
if source.supports_rich_root() != target.supports_rich_root():
3161
raise errors.IncompatibleRepositories(source, target,
3162
"different rich-root support")
3163
if source._serializer != target._serializer:
3164
raise errors.IncompatibleRepositories(source, target,
3165
"different serializers")
3168
class InterSameDataRepository(InterRepository):
3169
"""Code for converting between repositories that represent the same data.
3171
Data format and model must match for this to work.
3175
def _get_repo_format_to_test(self):
3176
"""Repository format for testing with.
3178
InterSameData can pull from subtree to subtree and from non-subtree to
3179
non-subtree, so we test this with the richest repository format.
3181
from bzrlib.repofmt import knitrepo
3182
return knitrepo.RepositoryFormatKnit3()
3185
def is_compatible(source, target):
3186
return InterRepository._same_model(source, target)
3189
class InterWeaveRepo(InterSameDataRepository):
3190
"""Optimised code paths between Weave based repositories.
3192
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3193
implemented lazy inter-object optimisation.
3197
def _get_repo_format_to_test(self):
3198
from bzrlib.repofmt import weaverepo
3199
return weaverepo.RepositoryFormat7()
3202
def is_compatible(source, target):
3203
"""Be compatible with known Weave formats.
3205
We don't test for the stores being of specific types because that
3206
could lead to confusing results, and there is no need to be
3209
from bzrlib.repofmt.weaverepo import (
3215
return (isinstance(source._format, (RepositoryFormat5,
3217
RepositoryFormat7)) and
3218
isinstance(target._format, (RepositoryFormat5,
3220
RepositoryFormat7)))
3221
except AttributeError:
3225
def copy_content(self, revision_id=None):
3226
"""See InterRepository.copy_content()."""
3227
# weave specific optimised path:
3229
self.target.set_make_working_trees(self.source.make_working_trees())
3230
except (errors.RepositoryUpgradeRequired, NotImplemented):
3232
# FIXME do not peek!
3233
if self.source._transport.listable():
3234
pb = ui.ui_factory.nested_progress_bar()
3236
self.target.texts.insert_record_stream(
3237
self.source.texts.get_record_stream(
3238
self.source.texts.keys(), 'topological', False))
3239
pb.update('copying inventory', 0, 1)
3240
self.target.inventories.insert_record_stream(
3241
self.source.inventories.get_record_stream(
3242
self.source.inventories.keys(), 'topological', False))
3243
self.target.signatures.insert_record_stream(
3244
self.source.signatures.get_record_stream(
3245
self.source.signatures.keys(),
3247
self.target.revisions.insert_record_stream(
3248
self.source.revisions.get_record_stream(
3249
self.source.revisions.keys(),
3250
'topological', True))
3254
self.target.fetch(self.source, revision_id=revision_id)
3257
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3258
"""See InterRepository.missing_revision_ids()."""
3259
# we want all revisions to satisfy revision_id in source.
3260
# but we don't want to stat every file here and there.
3261
# we want then, all revisions other needs to satisfy revision_id
3262
# checked, but not those that we have locally.
3263
# so the first thing is to get a subset of the revisions to
3264
# satisfy revision_id in source, and then eliminate those that
3265
# we do already have.
3266
# this is slow on high latency connection to self, but as this
3267
# disk format scales terribly for push anyway due to rewriting
3268
# inventory.weave, this is considered acceptable.
3270
if revision_id is not None:
3271
source_ids = self.source.get_ancestry(revision_id)
3272
if source_ids[0] is not None:
3273
raise AssertionError()
3276
source_ids = self.source._all_possible_ids()
3277
source_ids_set = set(source_ids)
3278
# source_ids is the worst possible case we may need to pull.
3279
# now we want to filter source_ids against what we actually
3280
# have in target, but don't try to check for existence where we know
3281
# we do not have a revision as that would be pointless.
3282
target_ids = set(self.target._all_possible_ids())
3283
possibly_present_revisions = target_ids.intersection(source_ids_set)
3284
actually_present_revisions = set(
3285
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3286
required_revisions = source_ids_set.difference(actually_present_revisions)
3287
if revision_id is not None:
3288
# we used get_ancestry to determine source_ids then we are assured all
3289
# revisions referenced are present as they are installed in topological order.
3290
# and the tip revision was validated by get_ancestry.
3291
result_set = required_revisions
3293
# if we just grabbed the possibly available ids, then
3294
# we only have an estimate of whats available and need to validate
3295
# that against the revision records.
3297
self.source._eliminate_revisions_not_present(required_revisions))
3298
return self.source.revision_ids_to_search_result(result_set)
3301
class InterKnitRepo(InterSameDataRepository):
3302
"""Optimised code paths between Knit based repositories."""
3305
def _get_repo_format_to_test(self):
3306
from bzrlib.repofmt import knitrepo
3307
return knitrepo.RepositoryFormatKnit1()
3310
def is_compatible(source, target):
3311
"""Be compatible with known Knit formats.
3313
We don't test for the stores being of specific types because that
3314
could lead to confusing results, and there is no need to be
3317
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3319
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3320
isinstance(target._format, RepositoryFormatKnit))
3321
except AttributeError:
3323
return are_knits and InterRepository._same_model(source, target)
3326
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3327
"""See InterRepository.missing_revision_ids()."""
3328
if revision_id is not None:
3329
source_ids = self.source.get_ancestry(revision_id)
3330
if source_ids[0] is not None:
3331
raise AssertionError()
3334
source_ids = self.source.all_revision_ids()
3335
source_ids_set = set(source_ids)
3336
# source_ids is the worst possible case we may need to pull.
3337
# now we want to filter source_ids against what we actually
3338
# have in target, but don't try to check for existence where we know
3339
# we do not have a revision as that would be pointless.
3340
target_ids = set(self.target.all_revision_ids())
3341
possibly_present_revisions = target_ids.intersection(source_ids_set)
3342
actually_present_revisions = set(
3343
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3344
required_revisions = source_ids_set.difference(actually_present_revisions)
3345
if revision_id is not None:
3346
# we used get_ancestry to determine source_ids then we are assured all
3347
# revisions referenced are present as they are installed in topological order.
3348
# and the tip revision was validated by get_ancestry.
3349
result_set = required_revisions
3351
# if we just grabbed the possibly available ids, then
3352
# we only have an estimate of whats available and need to validate
3353
# that against the revision records.
3355
self.source._eliminate_revisions_not_present(required_revisions))
3356
return self.source.revision_ids_to_search_result(result_set)
3359
class InterPackRepo(InterSameDataRepository):
3360
"""Optimised code paths between Pack based repositories."""
3363
def _get_repo_format_to_test(self):
3364
from bzrlib.repofmt import pack_repo
3365
return pack_repo.RepositoryFormatKnitPack1()
3368
def is_compatible(source, target):
3369
"""Be compatible with known Pack formats.
3371
We don't test for the stores being of specific types because that
3372
could lead to confusing results, and there is no need to be
3375
InterPackRepo does not support CHK based repositories.
3377
from bzrlib.repofmt.pack_repo import RepositoryFormatPack
3378
from bzrlib.repofmt.groupcompress_repo import RepositoryFormatCHK1
3380
are_packs = (isinstance(source._format, RepositoryFormatPack) and
3381
isinstance(target._format, RepositoryFormatPack))
3382
not_packs = (isinstance(source._format, RepositoryFormatCHK1) or
3383
isinstance(target._format, RepositoryFormatCHK1))
3384
except AttributeError:
3386
if not_packs or not are_packs:
3388
return InterRepository._same_model(source, target)
3391
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3393
"""See InterRepository.fetch()."""
3394
if (len(self.source._fallback_repositories) > 0 or
3395
len(self.target._fallback_repositories) > 0):
3396
# The pack layer is not aware of fallback repositories, so when
3397
# fetching from a stacked repository or into a stacked repository
3398
# we use the generic fetch logic which uses the VersionedFiles
3399
# attributes on repository.
3400
from bzrlib.fetch import RepoFetcher
3401
fetcher = RepoFetcher(self.target, self.source, revision_id,
3402
pb, find_ghosts, fetch_spec=fetch_spec)
3403
if fetch_spec is not None:
3404
if len(list(fetch_spec.heads)) != 1:
3405
raise AssertionError(
3406
"InterPackRepo.fetch doesn't support "
3407
"fetching multiple heads yet.")
3408
revision_id = list(fetch_spec.heads)[0]
3410
if revision_id is None:
3412
# everything to do - use pack logic
3413
# to fetch from all packs to one without
3414
# inventory parsing etc, IFF nothing to be copied is in the target.
3416
source_revision_ids = frozenset(self.source.all_revision_ids())
3417
revision_ids = source_revision_ids - \
3418
frozenset(self.target.get_parent_map(source_revision_ids))
3419
revision_keys = [(revid,) for revid in revision_ids]
3420
index = self.target._pack_collection.revision_index.combined_index
3421
present_revision_ids = set(item[1][0] for item in
3422
index.iter_entries(revision_keys))
3423
revision_ids = set(revision_ids) - present_revision_ids
3424
# implementing the TODO will involve:
3425
# - detecting when all of a pack is selected
3426
# - avoiding as much as possible pre-selection, so the
3427
# more-core routines such as create_pack_from_packs can filter in
3428
# a just-in-time fashion. (though having a HEADS list on a
3429
# repository might make this a lot easier, because we could
3430
# sensibly detect 'new revisions' without doing a full index scan.
3431
elif _mod_revision.is_null(revision_id):
3436
revision_ids = self.search_missing_revision_ids(revision_id,
3437
find_ghosts=find_ghosts).get_keys()
3438
except errors.NoSuchRevision:
3439
raise errors.InstallFailed([revision_id])
3440
if len(revision_ids) == 0:
3442
return self._pack(self.source, self.target, revision_ids)
3444
def _pack(self, source, target, revision_ids):
3445
from bzrlib.repofmt.pack_repo import Packer
3446
packs = source._pack_collection.all_packs()
3447
pack = Packer(self.target._pack_collection, packs, '.fetch',
3448
revision_ids).pack()
3449
if pack is not None:
3450
self.target._pack_collection._save_pack_names()
3451
copied_revs = pack.get_revision_count()
3452
# Trigger an autopack. This may duplicate effort as we've just done
3453
# a pack creation, but for now it is simpler to think about as
3454
# 'upload data, then repack if needed'.
3455
self.target._pack_collection.autopack()
3456
return (copied_revs, [])
3461
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3462
"""See InterRepository.missing_revision_ids().
3464
:param find_ghosts: Find ghosts throughout the ancestry of
3467
if not find_ghosts and revision_id is not None:
3468
return self._walk_to_common_revisions([revision_id])
3469
elif revision_id is not None:
3470
# Find ghosts: search for revisions pointing from one repository to
3471
# the other, and vice versa, anywhere in the history of revision_id.
3472
graph = self.target.get_graph(other_repository=self.source)
3473
searcher = graph._make_breadth_first_searcher([revision_id])
3477
next_revs, ghosts = searcher.next_with_ghosts()
3478
except StopIteration:
3480
if revision_id in ghosts:
3481
raise errors.NoSuchRevision(self.source, revision_id)
3482
found_ids.update(next_revs)
3483
found_ids.update(ghosts)
3484
found_ids = frozenset(found_ids)
3485
# Double query here: should be able to avoid this by changing the
3486
# graph api further.
3487
result_set = found_ids - frozenset(
3488
self.target.get_parent_map(found_ids))
3490
source_ids = self.source.all_revision_ids()
3491
# source_ids is the worst possible case we may need to pull.
3492
# now we want to filter source_ids against what we actually
3493
# have in target, but don't try to check for existence where we know
3494
# we do not have a revision as that would be pointless.
3495
target_ids = set(self.target.all_revision_ids())
3496
result_set = set(source_ids).difference(target_ids)
3497
return self.source.revision_ids_to_search_result(result_set)
3500
class InterDifferingSerializer(InterRepository):
3503
def _get_repo_format_to_test(self):
3507
def is_compatible(source, target):
3508
"""Be compatible with Knit2 source and Knit3 target"""
3509
# This is redundant with format.check_conversion_target(), however that
3510
# raises an exception, and we just want to say "False" as in we won't
3511
# support converting between these formats.
3512
if source.supports_rich_root() and not target.supports_rich_root():
3514
if (source._format.supports_tree_reference
3515
and not target._format.supports_tree_reference):
3519
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3520
"""Get the best delta and base for this revision.
3522
:return: (basis_id, delta)
3524
possible_trees = [(parent_id, cache[parent_id])
3525
for parent_id in parent_ids
3526
if parent_id in cache]
3527
if len(possible_trees) == 0:
3528
# There either aren't any parents, or the parents aren't in the
3529
# cache, so just use the last converted tree
3530
possible_trees.append((basis_id, cache[basis_id]))
3532
for basis_id, basis_tree in possible_trees:
3533
delta = tree.inventory._make_delta(basis_tree.inventory)
3534
deltas.append((len(delta), basis_id, delta))
3536
return deltas[0][1:]
3538
def _get_parent_keys(self, root_key, parent_map):
3539
"""Get the parent keys for a given root id."""
3540
root_id, rev_id = root_key
3541
# Include direct parents of the revision, but only if they used
3544
for parent_id in parent_map[rev_id]:
3545
if parent_id == _mod_revision.NULL_REVISION:
3547
if parent_id not in self._revision_id_to_root_id:
3548
# We probably didn't read this revision, go spend the
3549
# extra effort to actually check
3551
tree = self.source.revision_tree(parent_id)
3552
except errors.NoSuchRevision:
3553
# Ghost, fill out _revision_id_to_root_id in case we
3554
# encounter this again.
3555
# But set parent_root_id to None since we don't really know
3556
parent_root_id = None
3558
parent_root_id = tree.get_root_id()
3559
self._revision_id_to_root_id[parent_id] = None
3561
parent_root_id = self._revision_id_to_root_id[parent_id]
3562
if root_id == parent_root_id or parent_root_id is None:
3563
parent_keys.append((root_id, parent_id))
3564
return tuple(parent_keys)
3566
def _new_root_data_stream(self, root_keys_to_create, parent_map):
3567
for root_key in root_keys_to_create:
3568
parent_keys = self._get_parent_keys(root_key, parent_map)
3569
yield versionedfile.FulltextContentFactory(root_key,
3570
parent_keys, None, '')
3572
def _fetch_batch(self, revision_ids, basis_id, cache):
3573
"""Fetch across a few revisions.
3575
:param revision_ids: The revisions to copy
3576
:param basis_id: The revision_id of a tree that must be in cache, used
3577
as a basis for delta when no other base is available
3578
:param cache: A cache of RevisionTrees that we can use.
3579
:return: The revision_id of the last converted tree. The RevisionTree
3580
for it will be in cache
3582
# Walk though all revisions; get inventory deltas, copy referenced
3583
# texts that delta references, insert the delta, revision and
3585
root_keys_to_create = set()
3588
pending_revisions = []
3589
parent_map = self.source.get_parent_map(revision_ids)
3590
for tree in self.source.revision_trees(revision_ids):
3591
current_revision_id = tree.get_revision_id()
3592
parent_ids = parent_map.get(current_revision_id, ())
3593
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3595
if self._converting_to_rich_root:
3596
self._revision_id_to_root_id[current_revision_id] = \
3598
# Find text entries that need to be copied
3599
for old_path, new_path, file_id, entry in delta:
3600
if new_path is not None:
3603
if not self.target.supports_rich_root():
3604
# The target doesn't support rich root, so we don't
3607
if self._converting_to_rich_root:
3608
# This can't be copied normally, we have to insert
3610
root_keys_to_create.add((file_id, entry.revision))
3612
text_keys.add((file_id, entry.revision))
3613
revision = self.source.get_revision(current_revision_id)
3614
pending_deltas.append((basis_id, delta,
3615
current_revision_id, revision.parent_ids))
3616
pending_revisions.append(revision)
3617
cache[current_revision_id] = tree
3618
basis_id = current_revision_id
3620
from_texts = self.source.texts
3621
to_texts = self.target.texts
3622
if root_keys_to_create:
3623
root_stream = self._new_root_data_stream(root_keys_to_create,
3625
to_texts.insert_record_stream(root_stream)
3626
to_texts.insert_record_stream(from_texts.get_record_stream(
3627
text_keys, self.target._format._fetch_order,
3628
not self.target._format._fetch_uses_deltas))
3630
for delta in pending_deltas:
3631
self.target.add_inventory_by_delta(*delta)
3632
# insert signatures and revisions
3633
for revision in pending_revisions:
3635
signature = self.source.get_signature_text(
3636
revision.revision_id)
3637
self.target.add_signature_text(revision.revision_id,
3639
except errors.NoSuchRevision:
3641
self.target.add_revision(revision.revision_id, revision)
3644
def _fetch_all_revisions(self, revision_ids, pb):
3645
"""Fetch everything for the list of revisions.
3647
:param revision_ids: The list of revisions to fetch. Must be in
3649
:param pb: A ProgressBar
3652
basis_id, basis_tree = self._get_basis(revision_ids[0])
3654
cache = lru_cache.LRUCache(100)
3655
cache[basis_id] = basis_tree
3656
del basis_tree # We don't want to hang on to it here
3657
for offset in range(0, len(revision_ids), batch_size):
3658
self.target.start_write_group()
3660
pb.update('Transferring revisions', offset,
3662
batch = revision_ids[offset:offset+batch_size]
3663
basis_id = self._fetch_batch(batch, basis_id, cache)
3665
self.target.abort_write_group()
3668
self.target.commit_write_group()
3669
pb.update('Transferring revisions', len(revision_ids),
3673
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3675
"""See InterRepository.fetch()."""
3676
if fetch_spec is not None:
3677
raise AssertionError("Not implemented yet...")
3678
if (not self.source.supports_rich_root()
3679
and self.target.supports_rich_root()):
3680
self._converting_to_rich_root = True
3681
self._revision_id_to_root_id = {}
3683
self._converting_to_rich_root = False
3684
revision_ids = self.target.search_missing_revision_ids(self.source,
3685
revision_id, find_ghosts=find_ghosts).get_keys()
3686
if not revision_ids:
3688
revision_ids = tsort.topo_sort(
3689
self.source.get_graph().get_parent_map(revision_ids))
3690
if not revision_ids:
3692
# Walk though all revisions; get inventory deltas, copy referenced
3693
# texts that delta references, insert the delta, revision and
3695
first_rev = self.source.get_revision(revision_ids[0])
3697
my_pb = ui.ui_factory.nested_progress_bar()
3700
symbol_versioning.warn(
3701
symbol_versioning.deprecated_in((1, 14, 0))
3702
% "pb parameter to fetch()")
3705
self._fetch_all_revisions(revision_ids, pb)
3707
if my_pb is not None:
3709
return len(revision_ids), 0
3711
def _get_basis(self, first_revision_id):
3712
"""Get a revision and tree which exists in the target.
3714
This assumes that first_revision_id is selected for transmission
3715
because all other ancestors are already present. If we can't find an
3716
ancestor we fall back to NULL_REVISION since we know that is safe.
3718
:return: (basis_id, basis_tree)
3720
first_rev = self.source.get_revision(first_revision_id)
3722
basis_id = first_rev.parent_ids[0]
3723
# only valid as a basis if the target has it
3724
self.target.get_revision(basis_id)
3725
# Try to get a basis tree - if its a ghost it will hit the
3726
# NoSuchRevision case.
3727
basis_tree = self.source.revision_tree(basis_id)
3728
except (IndexError, errors.NoSuchRevision):
3729
basis_id = _mod_revision.NULL_REVISION
3730
basis_tree = self.source.revision_tree(basis_id)
3731
return basis_id, basis_tree
3734
InterRepository.register_optimiser(InterDifferingSerializer)
3735
InterRepository.register_optimiser(InterSameDataRepository)
3736
InterRepository.register_optimiser(InterWeaveRepo)
3737
InterRepository.register_optimiser(InterKnitRepo)
3738
InterRepository.register_optimiser(InterPackRepo)
3741
class CopyConverter(object):
3742
"""A repository conversion tool which just performs a copy of the content.
3744
This is slow but quite reliable.
3747
def __init__(self, target_format):
3748
"""Create a CopyConverter.
3750
:param target_format: The format the resulting repository should be.
3752
self.target_format = target_format
3754
def convert(self, repo, pb):
3755
"""Perform the conversion of to_convert, giving feedback via pb.
3757
:param to_convert: The disk object to convert.
3758
:param pb: a progress bar to use for progress information.
3763
# this is only useful with metadir layouts - separated repo content.
3764
# trigger an assertion if not such
3765
repo._format.get_format_string()
3766
self.repo_dir = repo.bzrdir
3767
self.step('Moving repository to repository.backup')
3768
self.repo_dir.transport.move('repository', 'repository.backup')
3769
backup_transport = self.repo_dir.transport.clone('repository.backup')
3770
repo._format.check_conversion_target(self.target_format)
3771
self.source_repo = repo._format.open(self.repo_dir,
3773
_override_transport=backup_transport)
3774
self.step('Creating new repository')
3775
converted = self.target_format.initialize(self.repo_dir,
3776
self.source_repo.is_shared())
3777
converted.lock_write()
3779
self.step('Copying content into repository.')
3780
self.source_repo.copy_content_into(converted)
3783
self.step('Deleting old repository content.')
3784
self.repo_dir.transport.delete_tree('repository.backup')
3785
self.pb.note('repository converted')
3787
def step(self, message):
3788
"""Update the pb by a step."""
3790
self.pb.update(message, self.count, self.total)
3802
def _unescaper(match, _map=_unescape_map):
3803
code = match.group(1)
3807
if not code.startswith('#'):
3809
return unichr(int(code[1:])).encode('utf8')
3815
def _unescape_xml(data):
3816
"""Unescape predefined XML entities in a string of data."""
3818
if _unescape_re is None:
3819
_unescape_re = re.compile('\&([^;]*);')
3820
return _unescape_re.sub(_unescaper, data)
3823
class _VersionedFileChecker(object):
3825
def __init__(self, repository, text_key_references=None):
3826
self.repository = repository
3827
self.text_index = self.repository._generate_text_key_index(
3828
text_key_references=text_key_references)
3830
def calculate_file_version_parents(self, text_key):
3831
"""Calculate the correct parents for a file version according to
3834
parent_keys = self.text_index[text_key]
3835
if parent_keys == [_mod_revision.NULL_REVISION]:
3837
return tuple(parent_keys)
3839
def check_file_version_parents(self, texts, progress_bar=None):
3840
"""Check the parents stored in a versioned file are correct.
3842
It also detects file versions that are not referenced by their
3843
corresponding revision's inventory.
3845
:returns: A tuple of (wrong_parents, dangling_file_versions).
3846
wrong_parents is a dict mapping {revision_id: (stored_parents,
3847
correct_parents)} for each revision_id where the stored parents
3848
are not correct. dangling_file_versions is a set of (file_id,
3849
revision_id) tuples for versions that are present in this versioned
3850
file, but not used by the corresponding inventory.
3853
self.file_ids = set([file_id for file_id, _ in
3854
self.text_index.iterkeys()])
3855
# text keys is now grouped by file_id
3856
n_weaves = len(self.file_ids)
3857
files_in_revisions = {}
3858
revisions_of_files = {}
3859
n_versions = len(self.text_index)
3860
progress_bar.update('loading text store', 0, n_versions)
3861
parent_map = self.repository.texts.get_parent_map(self.text_index)
3862
# On unlistable transports this could well be empty/error...
3863
text_keys = self.repository.texts.keys()
3864
unused_keys = frozenset(text_keys) - set(self.text_index)
3865
for num, key in enumerate(self.text_index.iterkeys()):
3866
if progress_bar is not None:
3867
progress_bar.update('checking text graph', num, n_versions)
3868
correct_parents = self.calculate_file_version_parents(key)
3870
knit_parents = parent_map[key]
3871
except errors.RevisionNotPresent:
3874
if correct_parents != knit_parents:
3875
wrong_parents[key] = (knit_parents, correct_parents)
3876
return wrong_parents, unused_keys
3879
def _old_get_graph(repository, revision_id):
3880
"""DO NOT USE. That is all. I'm serious."""
3881
graph = repository.get_graph()
3882
revision_graph = dict(((key, value) for key, value in
3883
graph.iter_ancestry([revision_id]) if value is not None))
3884
return _strip_NULL_ghosts(revision_graph)
3887
def _strip_NULL_ghosts(revision_graph):
3888
"""Also don't use this. more compatibility code for unmigrated clients."""
3889
# Filter ghosts, and null:
3890
if _mod_revision.NULL_REVISION in revision_graph:
3891
del revision_graph[_mod_revision.NULL_REVISION]
3892
for key, parents in revision_graph.items():
3893
revision_graph[key] = tuple(parent for parent in parents if parent
3895
return revision_graph
3898
class StreamSink(object):
3899
"""An object that can insert a stream into a repository.
3901
This interface handles the complexity of reserialising inventories and
3902
revisions from different formats, and allows unidirectional insertion into
3903
stacked repositories without looking for the missing basis parents
3907
def __init__(self, target_repo):
3908
self.target_repo = target_repo
3910
def insert_stream(self, stream, src_format, resume_tokens):
3911
"""Insert a stream's content into the target repository.
3913
:param src_format: a bzr repository format.
3915
:return: a list of resume tokens and an iterable of keys additional
3916
items required before the insertion can be completed.
3918
self.target_repo.lock_write()
3921
self.target_repo.resume_write_group(resume_tokens)
3923
self.target_repo.start_write_group()
3925
# locked_insert_stream performs a commit|suspend.
3926
return self._locked_insert_stream(stream, src_format)
3928
self.target_repo.abort_write_group(suppress_errors=True)
3931
self.target_repo.unlock()
3933
def _locked_insert_stream(self, stream, src_format):
3934
to_serializer = self.target_repo._format._serializer
3935
src_serializer = src_format._serializer
3936
if to_serializer == src_serializer:
3937
# If serializers match and the target is a pack repository, set the
3938
# write cache size on the new pack. This avoids poor performance
3939
# on transports where append is unbuffered (such as
3940
# RemoteTransport). This is safe to do because nothing should read
3941
# back from the target repository while a stream with matching
3942
# serialization is being inserted.
3943
# The exception is that a delta record from the source that should
3944
# be a fulltext may need to be expanded by the target (see
3945
# test_fetch_revisions_with_deltas_into_pack); but we take care to
3946
# explicitly flush any buffered writes first in that rare case.
3948
new_pack = self.target_repo._pack_collection._new_pack
3949
except AttributeError:
3950
# Not a pack repository
3953
new_pack.set_write_cache_size(1024*1024)
3954
for substream_type, substream in stream:
3955
if substream_type == 'texts':
3956
self.target_repo.texts.insert_record_stream(substream)
3957
elif substream_type == 'inventories':
3958
if src_serializer == to_serializer:
3959
self.target_repo.inventories.insert_record_stream(
3962
self._extract_and_insert_inventories(
3963
substream, src_serializer)
3964
elif substream_type == 'chk_bytes':
3965
# XXX: This doesn't support conversions, as it assumes the
3966
# conversion was done in the fetch code.
3967
self.target_repo.chk_bytes.insert_record_stream(substream)
3968
elif substream_type == 'revisions':
3969
# This may fallback to extract-and-insert more often than
3970
# required if the serializers are different only in terms of
3972
if src_serializer == to_serializer:
3973
self.target_repo.revisions.insert_record_stream(
3976
self._extract_and_insert_revisions(substream,
3978
elif substream_type == 'signatures':
3979
self.target_repo.signatures.insert_record_stream(substream)
3981
raise AssertionError('kaboom! %s' % (substream_type,))
3983
missing_keys = set()
3984
for prefix, versioned_file in (
3985
('texts', self.target_repo.texts),
3986
('inventories', self.target_repo.inventories),
3987
('revisions', self.target_repo.revisions),
3988
('signatures', self.target_repo.signatures),
3990
missing_keys.update((prefix,) + key for key in
3991
versioned_file.get_missing_compression_parent_keys())
3992
except NotImplementedError:
3993
# cannot even attempt suspending, and missing would have failed
3994
# during stream insertion.
3995
missing_keys = set()
3998
# suspend the write group and tell the caller what we is
3999
# missing. We know we can suspend or else we would not have
4000
# entered this code path. (All repositories that can handle
4001
# missing keys can handle suspending a write group).
4002
write_group_tokens = self.target_repo.suspend_write_group()
4003
return write_group_tokens, missing_keys
4004
self.target_repo.commit_write_group()
4007
def _extract_and_insert_inventories(self, substream, serializer):
4008
"""Generate a new inventory versionedfile in target, converting data.
4010
The inventory is retrieved from the source, (deserializing it), and
4011
stored in the target (reserializing it in a different format).
4013
for record in substream:
4014
bytes = record.get_bytes_as('fulltext')
4015
revision_id = record.key[0]
4016
inv = serializer.read_inventory_from_string(bytes, revision_id)
4017
parents = [key[0] for key in record.parents]
4018
self.target_repo.add_inventory(revision_id, inv, parents)
4020
def _extract_and_insert_revisions(self, substream, serializer):
4021
for record in substream:
4022
bytes = record.get_bytes_as('fulltext')
4023
revision_id = record.key[0]
4024
rev = serializer.read_revision_from_string(bytes)
4025
if rev.revision_id != revision_id:
4026
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4027
self.target_repo.add_revision(revision_id, rev)
4030
if self.target_repo._format._fetch_reconcile:
4031
self.target_repo.reconcile()
4034
class StreamSource(object):
4035
"""A source of a stream for fetching between repositories."""
4037
def __init__(self, from_repository, to_format):
4038
"""Create a StreamSource streaming from from_repository."""
4039
self.from_repository = from_repository
4040
self.to_format = to_format
4042
def delta_on_metadata(self):
4043
"""Return True if delta's are permitted on metadata streams.
4045
That is on revisions and signatures.
4047
src_serializer = self.from_repository._format._serializer
4048
target_serializer = self.to_format._serializer
4049
return (self.to_format._fetch_uses_deltas and
4050
src_serializer == target_serializer)
4052
def _fetch_revision_texts(self, revs):
4053
# fetch signatures first and then the revision texts
4054
# may need to be a InterRevisionStore call here.
4055
from_sf = self.from_repository.signatures
4056
# A missing signature is just skipped.
4057
keys = [(rev_id,) for rev_id in revs]
4058
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4060
self.to_format._fetch_order,
4061
not self.to_format._fetch_uses_deltas))
4062
# If a revision has a delta, this is actually expanded inside the
4063
# insert_record_stream code now, which is an alternate fix for
4065
from_rf = self.from_repository.revisions
4066
revisions = from_rf.get_record_stream(
4068
self.to_format._fetch_order,
4069
not self.delta_on_metadata())
4070
return [('signatures', signatures), ('revisions', revisions)]
4072
def _generate_root_texts(self, revs):
4073
"""This will be called by __fetch between fetching weave texts and
4074
fetching the inventory weave.
4076
Subclasses should override this if they need to generate root texts
4077
after fetching weave texts.
4079
if self._rich_root_upgrade():
4081
return bzrlib.fetch.Inter1and2Helper(
4082
self.from_repository).generate_root_texts(revs)
4086
def get_stream(self, search):
4088
revs = search.get_keys()
4089
graph = self.from_repository.get_graph()
4090
revs = list(graph.iter_topo_order(revs))
4091
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4093
for knit_kind, file_id, revisions in data_to_fetch:
4094
if knit_kind != phase:
4096
# Make a new progress bar for this phase
4097
if knit_kind == "file":
4098
# Accumulate file texts
4099
text_keys.extend([(file_id, revision) for revision in
4101
elif knit_kind == "inventory":
4102
# Now copy the file texts.
4103
from_texts = self.from_repository.texts
4104
yield ('texts', from_texts.get_record_stream(
4105
text_keys, self.to_format._fetch_order,
4106
not self.to_format._fetch_uses_deltas))
4107
# Cause an error if a text occurs after we have done the
4110
# Before we process the inventory we generate the root
4111
# texts (if necessary) so that the inventories references
4113
for _ in self._generate_root_texts(revs):
4115
# NB: This currently reopens the inventory weave in source;
4116
# using a single stream interface instead would avoid this.
4117
from_weave = self.from_repository.inventories
4118
# we fetch only the referenced inventories because we do not
4119
# know for unselected inventories whether all their required
4120
# texts are present in the other repository - it could be
4122
for info in self._get_inventory_stream(revs):
4124
elif knit_kind == "signatures":
4125
# Nothing to do here; this will be taken care of when
4126
# _fetch_revision_texts happens.
4128
elif knit_kind == "revisions":
4129
for record in self._fetch_revision_texts(revs):
4132
raise AssertionError("Unknown knit kind %r" % knit_kind)
4134
def get_stream_for_missing_keys(self, missing_keys):
4135
# missing keys can only occur when we are byte copying and not
4136
# translating (because translation means we don't send
4137
# unreconstructable deltas ever).
4139
keys['texts'] = set()
4140
keys['revisions'] = set()
4141
keys['inventories'] = set()
4142
keys['signatures'] = set()
4143
for key in missing_keys:
4144
keys[key[0]].add(key[1:])
4145
if len(keys['revisions']):
4146
# If we allowed copying revisions at this point, we could end up
4147
# copying a revision without copying its required texts: a
4148
# violation of the requirements for repository integrity.
4149
raise AssertionError(
4150
'cannot copy revisions to fill in missing deltas %s' % (
4151
keys['revisions'],))
4152
for substream_kind, keys in keys.iteritems():
4153
vf = getattr(self.from_repository, substream_kind)
4154
# Ask for full texts always so that we don't need more round trips
4155
# after this stream.
4156
stream = vf.get_record_stream(keys,
4157
self.to_format._fetch_order, True)
4158
yield substream_kind, stream
4160
def inventory_fetch_order(self):
4161
if self._rich_root_upgrade():
4162
return 'topological'
4164
return self.to_format._fetch_order
4166
def _rich_root_upgrade(self):
4167
return (not self.from_repository._format.rich_root_data and
4168
self.to_format.rich_root_data)
4170
def _get_inventory_stream(self, revision_ids):
4171
from_format = self.from_repository._format
4172
if (from_format.supports_chks and self.to_format.supports_chks
4173
and (from_format._serializer == self.to_format._serializer)):
4174
# Both sides support chks, and they use the same serializer, so it
4175
# is safe to transmit the chk pages and inventory pages across
4177
return self._get_chk_inventory_stream(revision_ids)
4178
elif (not from_format.supports_chks):
4179
# Source repository doesn't support chks. So we can transmit the
4180
# inventories 'as-is' and either they are just accepted on the
4181
# target, or the Sink will properly convert it.
4182
return self._get_simple_inventory_stream(revision_ids)
4184
# XXX: Hack to make not-chk->chk fetch: copy the inventories as
4185
# inventories. Note that this should probably be done somehow
4186
# as part of bzrlib.repository.StreamSink. Except JAM couldn't
4187
# figure out how a non-chk repository could possibly handle
4188
# deserializing an inventory stream from a chk repo, as it
4189
# doesn't have a way to understand individual pages.
4190
return self._get_convertable_inventory_stream(revision_ids)
4192
def _get_simple_inventory_stream(self, revision_ids):
4193
from_weave = self.from_repository.inventories
4194
yield ('inventories', from_weave.get_record_stream(
4195
[(rev_id,) for rev_id in revision_ids],
4196
self.inventory_fetch_order(),
4197
not self.delta_on_metadata()))
4199
def _get_chk_inventory_stream(self, revision_ids):
4200
"""Fetch the inventory texts, along with the associated chk maps."""
4201
# We want an inventory outside of the search set, so that we can filter
4202
# out uninteresting chk pages. For now we use
4203
# _find_revision_outside_set, but if we had a Search with cut_revs, we
4204
# could use that instead.
4205
start_rev_id = self.from_repository._find_revision_outside_set(
4207
start_rev_key = (start_rev_id,)
4208
inv_keys_to_fetch = [(rev_id,) for rev_id in revision_ids]
4209
if start_rev_id != _mod_revision.NULL_REVISION:
4210
inv_keys_to_fetch.append((start_rev_id,))
4211
# Any repo that supports chk_bytes must also support out-of-order
4212
# insertion. At least, that is how we expect it to work
4213
# We use get_record_stream instead of iter_inventories because we want
4214
# to be able to insert the stream as well. We could instead fetch
4215
# allowing deltas, and then iter_inventories, but we don't know whether
4216
# source or target is more 'local' anway.
4217
inv_stream = self.from_repository.inventories.get_record_stream(
4218
inv_keys_to_fetch, 'unordered',
4219
True) # We need them as full-texts so we can find their references
4220
uninteresting_chk_roots = set()
4221
interesting_chk_roots = set()
4222
def filter_inv_stream(inv_stream):
4223
for idx, record in enumerate(inv_stream):
4224
### child_pb.update('fetch inv', idx, len(inv_keys_to_fetch))
4225
bytes = record.get_bytes_as('fulltext')
4226
chk_inv = inventory.CHKInventory.deserialise(
4227
self.from_repository.chk_bytes, bytes, record.key)
4228
if record.key == start_rev_key:
4229
uninteresting_chk_roots.add(chk_inv.id_to_entry.key())
4230
p_id_map = chk_inv.parent_id_basename_to_file_id
4231
if p_id_map is not None:
4232
uninteresting_chk_roots.add(p_id_map.key())
4235
interesting_chk_roots.add(chk_inv.id_to_entry.key())
4236
p_id_map = chk_inv.parent_id_basename_to_file_id
4237
if p_id_map is not None:
4238
interesting_chk_roots.add(p_id_map.key())
4239
### pb.update('fetch inventory', 0, 2)
4240
yield ('inventories', filter_inv_stream(inv_stream))
4241
# Now that we have worked out all of the interesting root nodes, grab
4242
# all of the interesting pages and insert them
4243
### pb.update('fetch inventory', 1, 2)
4244
interesting = chk_map.iter_interesting_nodes(
4245
self.from_repository.chk_bytes, interesting_chk_roots,
4246
uninteresting_chk_roots)
4247
def to_stream_adapter():
4248
"""Adapt the iter_interesting_nodes result to a single stream.
4250
iter_interesting_nodes returns records as it processes them, along
4251
with keys. However, we only want to return the records themselves.
4253
for record, items in interesting:
4254
if record is not None:
4256
# XXX: We could instead call get_record_stream(records.keys())
4257
# ATM, this will always insert the records as fulltexts, and
4258
# requires that you can hang on to records once you have gone
4259
# on to the next one. Further, it causes the target to
4260
# recompress the data. Testing shows it to be faster than
4261
# requesting the records again, though.
4262
yield ('chk_bytes', to_stream_adapter())
4263
### pb.update('fetch inventory', 2, 2)
4265
def _get_convertable_inventory_stream(self, revision_ids):
4266
# XXX: One of source or target is using chks, and they don't have
4267
# compatible serializations. The StreamSink code expects to be
4268
# able to convert on the target, so we need to put
4269
# bytes-on-the-wire that can be converted
4270
yield ('inventories', self._stream_invs_as_fulltexts(revision_ids))
4272
def _stream_invs_as_fulltexts(self, revision_ids):
4273
from_repo = self.from_repository
4274
from_serializer = from_repo._format._serializer
4275
revision_keys = [(rev_id,) for rev_id in revision_ids]
4276
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4277
for inv in self.from_repository.iter_inventories(revision_ids):
4278
# XXX: This is a bit hackish, but it works. Basically,
4279
# CHKSerializer 'accidentally' supports
4280
# read/write_inventory_to_string, even though that is never
4281
# the format that is stored on disk. It *does* give us a
4282
# single string representation for an inventory, so live with
4284
# This would be far better if we had a 'serialized inventory
4285
# delta' form. Then we could use 'inventory._make_delta', and
4286
# transmit that. This would both be faster to generate, and
4287
# result in fewer bytes-on-the-wire.
4288
as_bytes = from_serializer.write_inventory_to_string(inv)
4289
key = (inv.revision_id,)
4290
parent_keys = parent_map.get(key, ())
4291
yield versionedfile.FulltextContentFactory(
4292
key, parent_keys, None, as_bytes)
added
removed
bzrlib/repository.py
