1
# Copyright (C) 2005, 2006, 2007, 2008, 2009 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
from bzrlib.lazy_import import lazy_import
18
lazy_import(globals(), """
37
revision as _mod_revision,
43
from bzrlib.bundle import serializer
44
from bzrlib.revisiontree import RevisionTree
45
from bzrlib.store.versioned import VersionedFileStore
46
from bzrlib.testament import Testament
49
from bzrlib.decorators import needs_read_lock, needs_write_lock
50
from bzrlib.inter import InterObject
51
from bzrlib.inventory import (
57
from bzrlib import registry
58
from bzrlib.symbol_versioning import (
64
from bzrlib.trace import (
65
log_exception_quietly, note, mutter, mutter_callsite, warning)
68
# Old formats display a warning, but only once
69
_deprecation_warning_done = False
72
class CommitBuilder(object):
73
"""Provides an interface to build up a commit.
75
This allows describing a tree to be committed without needing to
76
know the internals of the format of the repository.
79
# all clients should supply tree roots.
80
record_root_entry = True
81
# the default CommitBuilder does not manage trees whose root is versioned.
82
_versioned_root = False
84
def __init__(self, repository, parents, config, timestamp=None,
85
timezone=None, committer=None, revprops=None,
87
"""Initiate a CommitBuilder.
89
:param repository: Repository to commit to.
90
:param parents: Revision ids of the parents of the new revision.
91
:param config: Configuration to use.
92
:param timestamp: Optional timestamp recorded for commit.
93
:param timezone: Optional timezone for timestamp.
94
:param committer: Optional committer to set for commit.
95
:param revprops: Optional dictionary of revision properties.
96
:param revision_id: Optional revision id.
100
if committer is None:
101
self._committer = self._config.username()
103
self._committer = committer
105
self.new_inventory = Inventory(None)
106
self._new_revision_id = revision_id
107
self.parents = parents
108
self.repository = repository
111
if revprops is not None:
112
self._validate_revprops(revprops)
113
self._revprops.update(revprops)
115
if timestamp is None:
116
timestamp = time.time()
117
# Restrict resolution to 1ms
118
self._timestamp = round(timestamp, 3)
121
self._timezone = osutils.local_time_offset()
123
self._timezone = int(timezone)
125
self._generate_revision_if_needed()
126
self.__heads = graph.HeadsCache(repository.get_graph()).heads
127
self._basis_delta = []
128
# API compatibility, older code that used CommitBuilder did not call
129
# .record_delete(), which means the delta that is computed would not be
130
# valid. Callers that will call record_delete() should call
131
# .will_record_deletes() to indicate that.
132
self._recording_deletes = False
133
# memo'd check for no-op commits.
134
self._any_changes = False
136
def any_changes(self):
137
"""Return True if any entries were changed.
139
This includes merge-only changes. It is the core for the --unchanged
142
:return: True if any changes have occured.
144
return self._any_changes
146
def _validate_unicode_text(self, text, context):
147
"""Verify things like commit messages don't have bogus characters."""
149
raise ValueError('Invalid value for %s: %r' % (context, text))
151
def _validate_revprops(self, revprops):
152
for key, value in revprops.iteritems():
153
# We know that the XML serializers do not round trip '\r'
154
# correctly, so refuse to accept them
155
if not isinstance(value, basestring):
156
raise ValueError('revision property (%s) is not a valid'
157
' (unicode) string: %r' % (key, value))
158
self._validate_unicode_text(value,
159
'revision property (%s)' % (key,))
161
def commit(self, message):
162
"""Make the actual commit.
164
:return: The revision id of the recorded revision.
166
self._validate_unicode_text(message, 'commit message')
167
rev = _mod_revision.Revision(
168
timestamp=self._timestamp,
169
timezone=self._timezone,
170
committer=self._committer,
172
inventory_sha1=self.inv_sha1,
173
revision_id=self._new_revision_id,
174
properties=self._revprops)
175
rev.parent_ids = self.parents
176
self.repository.add_revision(self._new_revision_id, rev,
177
self.new_inventory, self._config)
178
self.repository.commit_write_group()
179
return self._new_revision_id
182
"""Abort the commit that is being built.
184
self.repository.abort_write_group()
186
def revision_tree(self):
187
"""Return the tree that was just committed.
189
After calling commit() this can be called to get a RevisionTree
190
representing the newly committed tree. This is preferred to
191
calling Repository.revision_tree() because that may require
192
deserializing the inventory, while we already have a copy in
195
if self.new_inventory is None:
196
self.new_inventory = self.repository.get_inventory(
197
self._new_revision_id)
198
return RevisionTree(self.repository, self.new_inventory,
199
self._new_revision_id)
201
def finish_inventory(self):
202
"""Tell the builder that the inventory is finished.
204
:return: The inventory id in the repository, which can be used with
205
repository.get_inventory.
207
if self.new_inventory is None:
208
# an inventory delta was accumulated without creating a new
210
basis_id = self.basis_delta_revision
211
self.inv_sha1 = self.repository.add_inventory_by_delta(
212
basis_id, self._basis_delta, self._new_revision_id,
215
if self.new_inventory.root is None:
216
raise AssertionError('Root entry should be supplied to'
217
' record_entry_contents, as of bzr 0.10.')
218
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
219
self.new_inventory.revision_id = self._new_revision_id
220
self.inv_sha1 = self.repository.add_inventory(
221
self._new_revision_id,
225
return self._new_revision_id
227
def _gen_revision_id(self):
228
"""Return new revision-id."""
229
return generate_ids.gen_revision_id(self._config.username(),
232
def _generate_revision_if_needed(self):
233
"""Create a revision id if None was supplied.
235
If the repository can not support user-specified revision ids
236
they should override this function and raise CannotSetRevisionId
237
if _new_revision_id is not None.
239
:raises: CannotSetRevisionId
241
if self._new_revision_id is None:
242
self._new_revision_id = self._gen_revision_id()
243
self.random_revid = True
245
self.random_revid = False
247
def _heads(self, file_id, revision_ids):
248
"""Calculate the graph heads for revision_ids in the graph of file_id.
250
This can use either a per-file graph or a global revision graph as we
251
have an identity relationship between the two graphs.
253
return self.__heads(revision_ids)
255
def _check_root(self, ie, parent_invs, tree):
256
"""Helper for record_entry_contents.
258
:param ie: An entry being added.
259
:param parent_invs: The inventories of the parent revisions of the
261
:param tree: The tree that is being committed.
263
# In this revision format, root entries have no knit or weave When
264
# serializing out to disk and back in root.revision is always
266
ie.revision = self._new_revision_id
268
def _require_root_change(self, tree):
269
"""Enforce an appropriate root object change.
271
This is called once when record_iter_changes is called, if and only if
272
the root was not in the delta calculated by record_iter_changes.
274
:param tree: The tree which is being committed.
276
# NB: if there are no parents then this method is not called, so no
277
# need to guard on parents having length.
278
entry = entry_factory['directory'](tree.path2id(''), '',
280
entry.revision = self._new_revision_id
281
self._basis_delta.append(('', '', entry.file_id, entry))
283
def _get_delta(self, ie, basis_inv, path):
284
"""Get a delta against the basis inventory for ie."""
285
if ie.file_id not in basis_inv:
287
result = (None, path, ie.file_id, ie)
288
self._basis_delta.append(result)
290
elif ie != basis_inv[ie.file_id]:
292
# TODO: avoid tis id2path call.
293
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
294
self._basis_delta.append(result)
300
def get_basis_delta(self):
301
"""Return the complete inventory delta versus the basis inventory.
303
This has been built up with the calls to record_delete and
304
record_entry_contents. The client must have already called
305
will_record_deletes() to indicate that they will be generating a
308
:return: An inventory delta, suitable for use with apply_delta, or
309
Repository.add_inventory_by_delta, etc.
311
if not self._recording_deletes:
312
raise AssertionError("recording deletes not activated.")
313
return self._basis_delta
315
def record_delete(self, path, file_id):
316
"""Record that a delete occured against a basis tree.
318
This is an optional API - when used it adds items to the basis_delta
319
being accumulated by the commit builder. It cannot be called unless the
320
method will_record_deletes() has been called to inform the builder that
321
a delta is being supplied.
323
:param path: The path of the thing deleted.
324
:param file_id: The file id that was deleted.
326
if not self._recording_deletes:
327
raise AssertionError("recording deletes not activated.")
328
delta = (path, None, file_id, None)
329
self._basis_delta.append(delta)
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.
565
:param _entry_factory: Private method to bind entry_factory locally for
569
# Create an inventory delta based on deltas between all the parents and
570
# deltas between all the parent inventories. We use inventory delta's
571
# between the inventory objects because iter_changes masks
572
# last-changed-field only changes.
574
# file_id -> change map, change is fileid, paths, changed, versioneds,
575
# parents, names, kinds, executables
577
# {file_id -> revision_id -> inventory entry, for entries in parent
578
# trees that are not parents[0]
580
revtrees = list(self.repository.revision_trees(self.parents))
581
# The basis inventory from a repository
583
basis_inv = revtrees[0].inventory
585
basis_inv = self.repository.revision_tree(
586
_mod_revision.NULL_REVISION).inventory
587
if len(self.parents) > 0:
588
if basis_revision_id != self.parents[0]:
590
"arbitrary basis parents not yet supported with merges")
591
for revtree in revtrees[1:]:
592
for change in revtree.inventory._make_delta(basis_inv):
593
if change[1] is None:
594
# Not present in this parent.
596
if change[2] not in merged_ids:
597
if change[0] is not None:
598
merged_ids[change[2]] = [
599
basis_inv[change[2]].revision,
602
merged_ids[change[2]] = [change[3].revision]
603
parent_entries[change[2]] = {change[3].revision:change[3]}
605
merged_ids[change[2]].append(change[3].revision)
606
parent_entries[change[2]][change[3].revision] = change[3]
609
# Setup the changes from the tree:
610
# changes maps file_id -> (change, [parent revision_ids])
612
for change in iter_changes:
613
# This probably looks up in basis_inv way to much.
614
if change[1][0] is not None:
615
head_candidate = [basis_inv[change[0]].revision]
618
changes[change[0]] = change, merged_ids.get(change[0],
620
unchanged_merged = set(merged_ids) - set(changes)
621
# Extend the changes dict with synthetic changes to record merges of
623
for file_id in unchanged_merged:
624
# Record a merged version of these items that did not change vs the
625
# basis. This can be either identical parallel changes, or a revert
626
# of a specific file after a merge. The recorded content will be
627
# that of the current tree (which is the same as the basis), but
628
# the per-file graph will reflect a merge.
629
# NB:XXX: We are reconstructing path information we had, this
630
# should be preserved instead.
631
# inv delta change: (file_id, (path_in_source, path_in_target),
632
# changed_content, versioned, parent, name, kind,
634
basis_entry = basis_inv[file_id]
636
(basis_inv.id2path(file_id), tree.id2path(file_id)),
638
(basis_entry.parent_id, basis_entry.parent_id),
639
(basis_entry.name, basis_entry.name),
640
(basis_entry.kind, basis_entry.kind),
641
(basis_entry.executable, basis_entry.executable))
642
changes[file_id] = (change, merged_ids[file_id])
643
# changes contains tuples with the change and a set of inventory
644
# candidates for the file.
646
# old_path, new_path, file_id, new_inventory_entry
647
seen_root = False # Is the root in the basis delta?
648
inv_delta = self._basis_delta
649
modified_rev = self._new_revision_id
650
for change, head_candidates in changes.values():
651
if change[3][1]: # versioned in target.
652
# Several things may be happening here:
653
# We may have a fork in the per-file graph
654
# - record a change with the content from tree
655
# We may have a change against < all trees
656
# - carry over the tree that hasn't changed
657
# We may have a change against all trees
658
# - record the change with the content from tree
661
entry = _entry_factory[kind](file_id, change[5][1],
663
head_set = self._heads(change[0], set(head_candidates))
666
for head_candidate in head_candidates:
667
if head_candidate in head_set:
668
heads.append(head_candidate)
669
head_set.remove(head_candidate)
672
# Could be a carry-over situation:
673
parent_entry_revs = parent_entries.get(file_id, None)
674
if parent_entry_revs:
675
parent_entry = parent_entry_revs.get(heads[0], None)
678
if parent_entry is None:
679
# The parent iter_changes was called against is the one
680
# that is the per-file head, so any change is relevant
681
# iter_changes is valid.
682
carry_over_possible = False
684
# could be a carry over situation
685
# A change against the basis may just indicate a merge,
686
# we need to check the content against the source of the
687
# merge to determine if it was changed after the merge
689
if (parent_entry.kind != entry.kind or
690
parent_entry.parent_id != entry.parent_id or
691
parent_entry.name != entry.name):
692
# Metadata common to all entries has changed
693
# against per-file parent
694
carry_over_possible = False
696
carry_over_possible = True
697
# per-type checks for changes against the parent_entry
700
# Cannot be a carry-over situation
701
carry_over_possible = False
702
# Populate the entry in the delta
704
# XXX: There is still a small race here: If someone reverts the content of a file
705
# after iter_changes examines and decides it has changed,
706
# we will unconditionally record a new version even if some
707
# other process reverts it while commit is running (with
708
# the revert happening after iter_changes did it's
711
entry.executable = True
713
entry.executable = False
714
if (carry_over_possible and
715
parent_entry.executable == entry.executable):
716
# Check the file length, content hash after reading
718
nostore_sha = parent_entry.text_sha1
721
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
723
lines = file_obj.readlines()
727
entry.text_sha1, entry.text_size = self._add_text_to_weave(
728
file_id, lines, heads, nostore_sha)
729
except errors.ExistingContent:
730
# No content change against a carry_over parent
732
entry.text_size = parent_entry.text_size
733
entry.text_sha1 = parent_entry.text_sha1
734
elif kind == 'symlink':
736
entry.symlink_target = tree.get_symlink_target(file_id)
737
if (carry_over_possible and
738
parent_entry.symlink_target == entry.symlink_target):
741
self._add_text_to_weave(change[0], [], heads, None)
742
elif kind == 'directory':
743
if carry_over_possible:
746
# Nothing to set on the entry.
747
# XXX: split into the Root and nonRoot versions.
748
if change[1][1] != '' or self.repository.supports_rich_root():
749
self._add_text_to_weave(change[0], [], heads, None)
750
elif kind == 'tree-reference':
751
raise AssertionError('unknown kind %r' % kind)
753
raise AssertionError('unknown kind %r' % kind)
755
entry.revision = modified_rev
757
entry.revision = parent_entry.revision
760
new_path = change[1][1]
761
inv_delta.append((change[1][0], new_path, change[0], entry))
764
self.new_inventory = None
766
self._any_changes = True
768
# housekeeping root entry changes do not affect no-change commits.
769
self._require_root_change(tree)
770
self.basis_delta_revision = basis_revision_id
772
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
773
# Note: as we read the content directly from the tree, we know its not
774
# been turned into unicode or badly split - but a broken tree
775
# implementation could give us bad output from readlines() so this is
776
# not a guarantee of safety. What would be better is always checking
777
# the content during test suite execution. RBC 20070912
778
parent_keys = tuple((file_id, parent) for parent in parents)
779
return self.repository.texts.add_lines(
780
(file_id, self._new_revision_id), parent_keys, new_lines,
781
nostore_sha=nostore_sha, random_id=self.random_revid,
782
check_content=False)[0:2]
785
class RootCommitBuilder(CommitBuilder):
786
"""This commitbuilder actually records the root id"""
788
# the root entry gets versioned properly by this builder.
789
_versioned_root = True
791
def _check_root(self, ie, parent_invs, tree):
792
"""Helper for record_entry_contents.
794
:param ie: An entry being added.
795
:param parent_invs: The inventories of the parent revisions of the
797
:param tree: The tree that is being committed.
800
def _require_root_change(self, tree):
801
"""Enforce an appropriate root object change.
803
This is called once when record_iter_changes is called, if and only if
804
the root was not in the delta calculated by record_iter_changes.
806
:param tree: The tree which is being committed.
808
# versioned roots do not change unless the tree found a change.
811
######################################################################
814
class Repository(object):
815
"""Repository holding history for one or more branches.
817
The repository holds and retrieves historical information including
818
revisions and file history. It's normally accessed only by the Branch,
819
which views a particular line of development through that history.
821
The Repository builds on top of some byte storage facilies (the revisions,
822
signatures, inventories and texts attributes) and a Transport, which
823
respectively provide byte storage and a means to access the (possibly
826
The byte storage facilities are addressed via tuples, which we refer to
827
as 'keys' throughout the code base. Revision_keys, inventory_keys and
828
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
829
(file_id, revision_id). We use this interface because it allows low
830
friction with the underlying code that implements disk indices, network
831
encoding and other parts of bzrlib.
833
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
834
the serialised revisions for the repository. This can be used to obtain
835
revision graph information or to access raw serialised revisions.
836
The result of trying to insert data into the repository via this store
837
is undefined: it should be considered read-only except for implementors
839
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
840
the serialised signatures for the repository. This can be used to
841
obtain access to raw serialised signatures. The result of trying to
842
insert data into the repository via this store is undefined: it should
843
be considered read-only except for implementors of repositories.
844
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
845
the serialised inventories for the repository. This can be used to
846
obtain unserialised inventories. The result of trying to insert data
847
into the repository via this store is undefined: it should be
848
considered read-only except for implementors of repositories.
849
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
850
texts of files and directories for the repository. This can be used to
851
obtain file texts or file graphs. Note that Repository.iter_file_bytes
852
is usually a better interface for accessing file texts.
853
The result of trying to insert data into the repository via this store
854
is undefined: it should be considered read-only except for implementors
856
:ivar _transport: Transport for file access to repository, typically
857
pointing to .bzr/repository.
860
# What class to use for a CommitBuilder. Often its simpler to change this
861
# in a Repository class subclass rather than to override
862
# get_commit_builder.
863
_commit_builder_class = CommitBuilder
864
# The search regex used by xml based repositories to determine what things
865
# where changed in a single commit.
866
_file_ids_altered_regex = lazy_regex.lazy_compile(
867
r'file_id="(?P<file_id>[^"]+)"'
868
r'.* revision="(?P<revision_id>[^"]+)"'
871
def abort_write_group(self, suppress_errors=False):
872
"""Commit the contents accrued within the current write group.
874
:param suppress_errors: if true, abort_write_group will catch and log
875
unexpected errors that happen during the abort, rather than
876
allowing them to propagate. Defaults to False.
878
:seealso: start_write_group.
880
if self._write_group is not self.get_transaction():
881
# has an unlock or relock occured ?
882
raise errors.BzrError('mismatched lock context and write group.')
884
self._abort_write_group()
885
except Exception, exc:
886
self._write_group = None
887
if not suppress_errors:
889
mutter('abort_write_group failed')
890
log_exception_quietly()
891
note('bzr: ERROR (ignored): %s', exc)
892
self._write_group = None
894
def _abort_write_group(self):
895
"""Template method for per-repository write group cleanup.
897
This is called during abort before the write group is considered to be
898
finished and should cleanup any internal state accrued during the write
899
group. There is no requirement that data handed to the repository be
900
*not* made available - this is not a rollback - but neither should any
901
attempt be made to ensure that data added is fully commited. Abort is
902
invoked when an error has occured so futher disk or network operations
903
may not be possible or may error and if possible should not be
907
def add_fallback_repository(self, repository):
908
"""Add a repository to use for looking up data not held locally.
910
:param repository: A repository.
912
if not self._format.supports_external_lookups:
913
raise errors.UnstackableRepositoryFormat(self._format, self.base)
914
self._check_fallback_repository(repository)
915
self._fallback_repositories.append(repository)
916
self.texts.add_fallback_versioned_files(repository.texts)
917
self.inventories.add_fallback_versioned_files(repository.inventories)
918
self.revisions.add_fallback_versioned_files(repository.revisions)
919
self.signatures.add_fallback_versioned_files(repository.signatures)
921
def _check_fallback_repository(self, repository):
922
"""Check that this repository can fallback to repository safely.
924
Raise an error if not.
926
:param repository: A repository to fallback to.
928
return InterRepository._assert_same_model(self, repository)
930
def add_inventory(self, revision_id, inv, parents):
931
"""Add the inventory inv to the repository as revision_id.
933
:param parents: The revision ids of the parents that revision_id
934
is known to have and are in the repository already.
936
:returns: The validator(which is a sha1 digest, though what is sha'd is
937
repository format specific) of the serialized inventory.
939
if not self.is_in_write_group():
940
raise AssertionError("%r not in write group" % (self,))
941
_mod_revision.check_not_reserved_id(revision_id)
942
if not (inv.revision_id is None or inv.revision_id == revision_id):
943
raise AssertionError(
944
"Mismatch between inventory revision"
945
" id and insertion revid (%r, %r)"
946
% (inv.revision_id, revision_id))
948
raise AssertionError()
949
inv_lines = self._serialise_inventory_to_lines(inv)
950
return self._inventory_add_lines(revision_id, parents,
951
inv_lines, check_content=False)
953
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
955
"""Add a new inventory expressed as a delta against another revision.
957
:param basis_revision_id: The inventory id the delta was created
958
against. (This does not have to be a direct parent.)
959
:param delta: The inventory delta (see Inventory.apply_delta for
961
:param new_revision_id: The revision id that the inventory is being
963
:param parents: The revision ids of the parents that revision_id is
964
known to have and are in the repository already. These are supplied
965
for repositories that depend on the inventory graph for revision
966
graph access, as well as for those that pun ancestry with delta
969
:returns: (validator, new_inv)
970
The validator(which is a sha1 digest, though what is sha'd is
971
repository format specific) of the serialized inventory, and the
974
if not self.is_in_write_group():
975
raise AssertionError("%r not in write group" % (self,))
976
_mod_revision.check_not_reserved_id(new_revision_id)
977
basis_tree = self.revision_tree(basis_revision_id)
978
basis_tree.lock_read()
980
# Note that this mutates the inventory of basis_tree, which not all
981
# inventory implementations may support: A better idiom would be to
982
# return a new inventory, but as there is no revision tree cache in
983
# repository this is safe for now - RBC 20081013
984
basis_inv = basis_tree.inventory
985
basis_inv.apply_delta(delta)
986
basis_inv.revision_id = new_revision_id
987
return (self.add_inventory(new_revision_id, basis_inv, parents),
992
def _inventory_add_lines(self, revision_id, parents, lines,
994
"""Store lines in inv_vf and return the sha1 of the inventory."""
995
parents = [(parent,) for parent in parents]
996
return self.inventories.add_lines((revision_id,), parents, lines,
997
check_content=check_content)[0]
999
def add_revision(self, revision_id, rev, inv=None, config=None):
1000
"""Add rev to the revision store as revision_id.
1002
:param revision_id: the revision id to use.
1003
:param rev: The revision object.
1004
:param inv: The inventory for the revision. if None, it will be looked
1005
up in the inventory storer
1006
:param config: If None no digital signature will be created.
1007
If supplied its signature_needed method will be used
1008
to determine if a signature should be made.
1010
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1012
_mod_revision.check_not_reserved_id(revision_id)
1013
if config is not None and config.signature_needed():
1015
inv = self.get_inventory(revision_id)
1016
plaintext = Testament(rev, inv).as_short_text()
1017
self.store_revision_signature(
1018
gpg.GPGStrategy(config), plaintext, revision_id)
1019
# check inventory present
1020
if not self.inventories.get_parent_map([(revision_id,)]):
1022
raise errors.WeaveRevisionNotPresent(revision_id,
1025
# yes, this is not suitable for adding with ghosts.
1026
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1029
key = (revision_id,)
1030
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1031
self._add_revision(rev)
1033
def _add_revision(self, revision):
1034
text = self._serializer.write_revision_to_string(revision)
1035
key = (revision.revision_id,)
1036
parents = tuple((parent,) for parent in revision.parent_ids)
1037
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1039
def all_revision_ids(self):
1040
"""Returns a list of all the revision ids in the repository.
1042
This is conceptually deprecated because code should generally work on
1043
the graph reachable from a particular revision, and ignore any other
1044
revisions that might be present. There is no direct replacement
1047
if 'evil' in debug.debug_flags:
1048
mutter_callsite(2, "all_revision_ids is linear with history.")
1049
return self._all_revision_ids()
1051
def _all_revision_ids(self):
1052
"""Returns a list of all the revision ids in the repository.
1054
These are in as much topological order as the underlying store can
1057
raise NotImplementedError(self._all_revision_ids)
1059
def break_lock(self):
1060
"""Break a lock if one is present from another instance.
1062
Uses the ui factory to ask for confirmation if the lock may be from
1065
self.control_files.break_lock()
1068
def _eliminate_revisions_not_present(self, revision_ids):
1069
"""Check every revision id in revision_ids to see if we have it.
1071
Returns a set of the present revisions.
1074
graph = self.get_graph()
1075
parent_map = graph.get_parent_map(revision_ids)
1076
# The old API returned a list, should this actually be a set?
1077
return parent_map.keys()
1080
def create(a_bzrdir):
1081
"""Construct the current default format repository in a_bzrdir."""
1082
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1084
def __init__(self, _format, a_bzrdir, control_files):
1085
"""instantiate a Repository.
1087
:param _format: The format of the repository on disk.
1088
:param a_bzrdir: The BzrDir of the repository.
1090
In the future we will have a single api for all stores for
1091
getting file texts, inventories and revisions, then
1092
this construct will accept instances of those things.
1094
super(Repository, self).__init__()
1095
self._format = _format
1096
# the following are part of the public API for Repository:
1097
self.bzrdir = a_bzrdir
1098
self.control_files = control_files
1099
self._transport = control_files._transport
1100
self.base = self._transport.base
1102
self._reconcile_does_inventory_gc = True
1103
self._reconcile_fixes_text_parents = False
1104
self._reconcile_backsup_inventory = True
1105
# not right yet - should be more semantically clear ?
1107
# TODO: make sure to construct the right store classes, etc, depending
1108
# on whether escaping is required.
1109
self._warn_if_deprecated()
1110
self._write_group = None
1111
# Additional places to query for data.
1112
self._fallback_repositories = []
1113
# An InventoryEntry cache, used during deserialization
1114
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1117
return '%s(%r)' % (self.__class__.__name__,
1120
def has_same_location(self, other):
1121
"""Returns a boolean indicating if this repository is at the same
1122
location as another repository.
1124
This might return False even when two repository objects are accessing
1125
the same physical repository via different URLs.
1127
if self.__class__ is not other.__class__:
1129
return (self._transport.base == other._transport.base)
1131
def is_in_write_group(self):
1132
"""Return True if there is an open write group.
1134
:seealso: start_write_group.
1136
return self._write_group is not None
1138
def is_locked(self):
1139
return self.control_files.is_locked()
1141
def is_write_locked(self):
1142
"""Return True if this object is write locked."""
1143
return self.is_locked() and self.control_files._lock_mode == 'w'
1145
def lock_write(self, token=None):
1146
"""Lock this repository for writing.
1148
This causes caching within the repository obejct to start accumlating
1149
data during reads, and allows a 'write_group' to be obtained. Write
1150
groups must be used for actual data insertion.
1152
:param token: if this is already locked, then lock_write will fail
1153
unless the token matches the existing lock.
1154
:returns: a token if this instance supports tokens, otherwise None.
1155
:raises TokenLockingNotSupported: when a token is given but this
1156
instance doesn't support using token locks.
1157
:raises MismatchedToken: if the specified token doesn't match the token
1158
of the existing lock.
1159
:seealso: start_write_group.
1161
A token should be passed in if you know that you have locked the object
1162
some other way, and need to synchronise this object's state with that
1165
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1167
locked = self.is_locked()
1168
result = self.control_files.lock_write(token=token)
1169
for repo in self._fallback_repositories:
1170
# Writes don't affect fallback repos
1173
self._refresh_data()
1176
def lock_read(self):
1177
locked = self.is_locked()
1178
self.control_files.lock_read()
1179
for repo in self._fallback_repositories:
1182
self._refresh_data()
1184
def get_physical_lock_status(self):
1185
return self.control_files.get_physical_lock_status()
1187
def leave_lock_in_place(self):
1188
"""Tell this repository not to release the physical lock when this
1191
If lock_write doesn't return a token, then this method is not supported.
1193
self.control_files.leave_in_place()
1195
def dont_leave_lock_in_place(self):
1196
"""Tell this repository to release the physical lock when this
1197
object is unlocked, even if it didn't originally acquire it.
1199
If lock_write doesn't return a token, then this method is not supported.
1201
self.control_files.dont_leave_in_place()
1204
def gather_stats(self, revid=None, committers=None):
1205
"""Gather statistics from a revision id.
1207
:param revid: The revision id to gather statistics from, if None, then
1208
no revision specific statistics are gathered.
1209
:param committers: Optional parameter controlling whether to grab
1210
a count of committers from the revision specific statistics.
1211
:return: A dictionary of statistics. Currently this contains:
1212
committers: The number of committers if requested.
1213
firstrev: A tuple with timestamp, timezone for the penultimate left
1214
most ancestor of revid, if revid is not the NULL_REVISION.
1215
latestrev: A tuple with timestamp, timezone for revid, if revid is
1216
not the NULL_REVISION.
1217
revisions: The total revision count in the repository.
1218
size: An estimate disk size of the repository in bytes.
1221
if revid and committers:
1222
result['committers'] = 0
1223
if revid and revid != _mod_revision.NULL_REVISION:
1225
all_committers = set()
1226
revisions = self.get_ancestry(revid)
1227
# pop the leading None
1229
first_revision = None
1231
# ignore the revisions in the middle - just grab first and last
1232
revisions = revisions[0], revisions[-1]
1233
for revision in self.get_revisions(revisions):
1234
if not first_revision:
1235
first_revision = revision
1237
all_committers.add(revision.committer)
1238
last_revision = revision
1240
result['committers'] = len(all_committers)
1241
result['firstrev'] = (first_revision.timestamp,
1242
first_revision.timezone)
1243
result['latestrev'] = (last_revision.timestamp,
1244
last_revision.timezone)
1246
# now gather global repository information
1247
# XXX: This is available for many repos regardless of listability.
1248
if self.bzrdir.root_transport.listable():
1249
# XXX: do we want to __define len__() ?
1250
# Maybe the versionedfiles object should provide a different
1251
# method to get the number of keys.
1252
result['revisions'] = len(self.revisions.keys())
1253
# result['size'] = t
1256
def find_branches(self, using=False):
1257
"""Find branches underneath this repository.
1259
This will include branches inside other branches.
1261
:param using: If True, list only branches using this repository.
1263
if using and not self.is_shared():
1265
return [self.bzrdir.open_branch()]
1266
except errors.NotBranchError:
1268
class Evaluator(object):
1271
self.first_call = True
1273
def __call__(self, bzrdir):
1274
# On the first call, the parameter is always the bzrdir
1275
# containing the current repo.
1276
if not self.first_call:
1278
repository = bzrdir.open_repository()
1279
except errors.NoRepositoryPresent:
1282
return False, (None, repository)
1283
self.first_call = False
1285
value = (bzrdir.open_branch(), None)
1286
except errors.NotBranchError:
1287
value = (None, None)
1291
for branch, repository in bzrdir.BzrDir.find_bzrdirs(
1292
self.bzrdir.root_transport, evaluate=Evaluator()):
1293
if branch is not None:
1294
branches.append(branch)
1295
if not using and repository is not None:
1296
branches.extend(repository.find_branches())
1300
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
1301
"""Return the revision ids that other has that this does not.
1303
These are returned in topological order.
1305
revision_id: only return revision ids included by revision_id.
1307
return InterRepository.get(other, self).search_missing_revision_ids(
1308
revision_id, find_ghosts)
1310
@deprecated_method(one_two)
1312
def missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
1313
"""Return the revision ids that other has that this does not.
1315
These are returned in topological order.
1317
revision_id: only return revision ids included by revision_id.
1319
keys = self.search_missing_revision_ids(
1320
other, revision_id, find_ghosts).get_keys()
1323
parents = other.get_graph().get_parent_map(keys)
1326
return tsort.topo_sort(parents)
1330
"""Open the repository rooted at base.
1332
For instance, if the repository is at URL/.bzr/repository,
1333
Repository.open(URL) -> a Repository instance.
1335
control = bzrdir.BzrDir.open(base)
1336
return control.open_repository()
1338
def copy_content_into(self, destination, revision_id=None):
1339
"""Make a complete copy of the content in self into destination.
1341
This is a destructive operation! Do not use it on existing
1344
return InterRepository.get(self, destination).copy_content(revision_id)
1346
def commit_write_group(self):
1347
"""Commit the contents accrued within the current write group.
1349
:seealso: start_write_group.
1351
if self._write_group is not self.get_transaction():
1352
# has an unlock or relock occured ?
1353
raise errors.BzrError('mismatched lock context %r and '
1355
(self.get_transaction(), self._write_group))
1356
self._commit_write_group()
1357
self._write_group = None
1359
def _commit_write_group(self):
1360
"""Template method for per-repository write group cleanup.
1362
This is called before the write group is considered to be
1363
finished and should ensure that all data handed to the repository
1364
for writing during the write group is safely committed (to the
1365
extent possible considering file system caching etc).
1368
def suspend_write_group(self):
1369
raise errors.UnsuspendableWriteGroup(self)
1371
def refresh_data(self):
1372
"""Re-read any data needed to to synchronise with disk.
1374
This method is intended to be called after another repository instance
1375
(such as one used by a smart server) has inserted data into the
1376
repository. It may not be called during a write group, but may be
1377
called at any other time.
1379
if self.is_in_write_group():
1380
raise errors.InternalBzrError(
1381
"May not refresh_data while in a write group.")
1382
self._refresh_data()
1384
def resume_write_group(self, tokens):
1385
if not self.is_write_locked():
1386
raise errors.NotWriteLocked(self)
1387
if self._write_group:
1388
raise errors.BzrError('already in a write group')
1389
self._resume_write_group(tokens)
1390
# so we can detect unlock/relock - the write group is now entered.
1391
self._write_group = self.get_transaction()
1393
def _resume_write_group(self, tokens):
1394
raise errors.UnsuspendableWriteGroup(self)
1396
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1398
"""Fetch the content required to construct revision_id from source.
1400
If revision_id is None and fetch_spec is None, then all content is
1403
fetch() may not be used when the repository is in a write group -
1404
either finish the current write group before using fetch, or use
1405
fetch before starting the write group.
1407
:param find_ghosts: Find and copy revisions in the source that are
1408
ghosts in the target (and not reachable directly by walking out to
1409
the first-present revision in target from revision_id).
1410
:param revision_id: If specified, all the content needed for this
1411
revision ID will be copied to the target. Fetch will determine for
1412
itself which content needs to be copied.
1413
:param fetch_spec: If specified, a SearchResult or
1414
PendingAncestryResult that describes which revisions to copy. This
1415
allows copying multiple heads at once. Mutually exclusive with
1418
if fetch_spec is not None and revision_id is not None:
1419
raise AssertionError(
1420
"fetch_spec and revision_id are mutually exclusive.")
1421
if self.is_in_write_group():
1422
raise errors.InternalBzrError(
1423
"May not fetch while in a write group.")
1424
# fast path same-url fetch operations
1425
if self.has_same_location(source) and fetch_spec is None:
1426
# check that last_revision is in 'from' and then return a
1428
if (revision_id is not None and
1429
not _mod_revision.is_null(revision_id)):
1430
self.get_revision(revision_id)
1432
# if there is no specific appropriate InterRepository, this will get
1433
# the InterRepository base class, which raises an
1434
# IncompatibleRepositories when asked to fetch.
1435
inter = InterRepository.get(source, self)
1436
return inter.fetch(revision_id=revision_id, pb=pb,
1437
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1439
def create_bundle(self, target, base, fileobj, format=None):
1440
return serializer.write_bundle(self, target, base, fileobj, format)
1442
def get_commit_builder(self, branch, parents, config, timestamp=None,
1443
timezone=None, committer=None, revprops=None,
1445
"""Obtain a CommitBuilder for this repository.
1447
:param branch: Branch to commit to.
1448
:param parents: Revision ids of the parents of the new revision.
1449
:param config: Configuration to use.
1450
:param timestamp: Optional timestamp recorded for commit.
1451
:param timezone: Optional timezone for timestamp.
1452
:param committer: Optional committer to set for commit.
1453
:param revprops: Optional dictionary of revision properties.
1454
:param revision_id: Optional revision id.
1456
result = self._commit_builder_class(self, parents, config,
1457
timestamp, timezone, committer, revprops, revision_id)
1458
self.start_write_group()
1462
if (self.control_files._lock_count == 1 and
1463
self.control_files._lock_mode == 'w'):
1464
if self._write_group is not None:
1465
self.abort_write_group()
1466
self.control_files.unlock()
1467
raise errors.BzrError(
1468
'Must end write groups before releasing write locks.')
1469
self.control_files.unlock()
1470
if self.control_files._lock_count == 0:
1471
self._inventory_entry_cache.clear()
1472
for repo in self._fallback_repositories:
1476
def clone(self, a_bzrdir, revision_id=None):
1477
"""Clone this repository into a_bzrdir using the current format.
1479
Currently no check is made that the format of this repository and
1480
the bzrdir format are compatible. FIXME RBC 20060201.
1482
:return: The newly created destination repository.
1484
# TODO: deprecate after 0.16; cloning this with all its settings is
1485
# probably not very useful -- mbp 20070423
1486
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1487
self.copy_content_into(dest_repo, revision_id)
1490
def start_write_group(self):
1491
"""Start a write group in the repository.
1493
Write groups are used by repositories which do not have a 1:1 mapping
1494
between file ids and backend store to manage the insertion of data from
1495
both fetch and commit operations.
1497
A write lock is required around the start_write_group/commit_write_group
1498
for the support of lock-requiring repository formats.
1500
One can only insert data into a repository inside a write group.
1504
if not self.is_write_locked():
1505
raise errors.NotWriteLocked(self)
1506
if self._write_group:
1507
raise errors.BzrError('already in a write group')
1508
self._start_write_group()
1509
# so we can detect unlock/relock - the write group is now entered.
1510
self._write_group = self.get_transaction()
1512
def _start_write_group(self):
1513
"""Template method for per-repository write group startup.
1515
This is called before the write group is considered to be
1520
def sprout(self, to_bzrdir, revision_id=None):
1521
"""Create a descendent repository for new development.
1523
Unlike clone, this does not copy the settings of the repository.
1525
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1526
dest_repo.fetch(self, revision_id=revision_id)
1529
def _create_sprouting_repo(self, a_bzrdir, shared):
1530
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1531
# use target default format.
1532
dest_repo = a_bzrdir.create_repository()
1534
# Most control formats need the repository to be specifically
1535
# created, but on some old all-in-one formats it's not needed
1537
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1538
except errors.UninitializableFormat:
1539
dest_repo = a_bzrdir.open_repository()
1542
def _get_sink(self):
1543
"""Return a sink for streaming into this repository."""
1544
return StreamSink(self)
1546
def _get_source(self, to_format):
1547
"""Return a source for streaming from this repository."""
1548
return StreamSource(self, to_format)
1551
def has_revision(self, revision_id):
1552
"""True if this repository has a copy of the revision."""
1553
return revision_id in self.has_revisions((revision_id,))
1556
def has_revisions(self, revision_ids):
1557
"""Probe to find out the presence of multiple revisions.
1559
:param revision_ids: An iterable of revision_ids.
1560
:return: A set of the revision_ids that were present.
1562
parent_map = self.revisions.get_parent_map(
1563
[(rev_id,) for rev_id in revision_ids])
1565
if _mod_revision.NULL_REVISION in revision_ids:
1566
result.add(_mod_revision.NULL_REVISION)
1567
result.update([key[0] for key in parent_map])
1571
def get_revision(self, revision_id):
1572
"""Return the Revision object for a named revision."""
1573
return self.get_revisions([revision_id])[0]
1576
def get_revision_reconcile(self, revision_id):
1577
"""'reconcile' helper routine that allows access to a revision always.
1579
This variant of get_revision does not cross check the weave graph
1580
against the revision one as get_revision does: but it should only
1581
be used by reconcile, or reconcile-alike commands that are correcting
1582
or testing the revision graph.
1584
return self._get_revisions([revision_id])[0]
1587
def get_revisions(self, revision_ids):
1588
"""Get many revisions at once."""
1589
return self._get_revisions(revision_ids)
1592
def _get_revisions(self, revision_ids):
1593
"""Core work logic to get many revisions without sanity checks."""
1594
for rev_id in revision_ids:
1595
if not rev_id or not isinstance(rev_id, basestring):
1596
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1597
keys = [(key,) for key in revision_ids]
1598
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1600
for record in stream:
1601
if record.storage_kind == 'absent':
1602
raise errors.NoSuchRevision(self, record.key[0])
1603
text = record.get_bytes_as('fulltext')
1604
rev = self._serializer.read_revision_from_string(text)
1605
revs[record.key[0]] = rev
1606
return [revs[revid] for revid in revision_ids]
1609
def get_revision_xml(self, revision_id):
1610
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1611
# would have already do it.
1612
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1613
rev = self.get_revision(revision_id)
1614
rev_tmp = cStringIO.StringIO()
1615
# the current serializer..
1616
self._serializer.write_revision(rev, rev_tmp)
1618
return rev_tmp.getvalue()
1620
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1621
"""Produce a generator of revision deltas.
1623
Note that the input is a sequence of REVISIONS, not revision_ids.
1624
Trees will be held in memory until the generator exits.
1625
Each delta is relative to the revision's lefthand predecessor.
1627
:param specific_fileids: if not None, the result is filtered
1628
so that only those file-ids, their parents and their
1629
children are included.
1631
# Get the revision-ids of interest
1632
required_trees = set()
1633
for revision in revisions:
1634
required_trees.add(revision.revision_id)
1635
required_trees.update(revision.parent_ids[:1])
1637
# Get the matching filtered trees. Note that it's more
1638
# efficient to pass filtered trees to changes_from() rather
1639
# than doing the filtering afterwards. changes_from() could
1640
# arguably do the filtering itself but it's path-based, not
1641
# file-id based, so filtering before or afterwards is
1643
if specific_fileids is None:
1644
trees = dict((t.get_revision_id(), t) for
1645
t in self.revision_trees(required_trees))
1647
trees = dict((t.get_revision_id(), t) for
1648
t in self._filtered_revision_trees(required_trees,
1651
# Calculate the deltas
1652
for revision in revisions:
1653
if not revision.parent_ids:
1654
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1656
old_tree = trees[revision.parent_ids[0]]
1657
yield trees[revision.revision_id].changes_from(old_tree)
1660
def get_revision_delta(self, revision_id, specific_fileids=None):
1661
"""Return the delta for one revision.
1663
The delta is relative to the left-hand predecessor of the
1666
:param specific_fileids: if not None, the result is filtered
1667
so that only those file-ids, their parents and their
1668
children are included.
1670
r = self.get_revision(revision_id)
1671
return list(self.get_deltas_for_revisions([r],
1672
specific_fileids=specific_fileids))[0]
1675
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1676
signature = gpg_strategy.sign(plaintext)
1677
self.add_signature_text(revision_id, signature)
1680
def add_signature_text(self, revision_id, signature):
1681
self.signatures.add_lines((revision_id,), (),
1682
osutils.split_lines(signature))
1684
def find_text_key_references(self):
1685
"""Find the text key references within the repository.
1687
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1688
to whether they were referred to by the inventory of the
1689
revision_id that they contain. The inventory texts from all present
1690
revision ids are assessed to generate this report.
1692
revision_keys = self.revisions.keys()
1693
w = self.inventories
1694
pb = ui.ui_factory.nested_progress_bar()
1696
return self._find_text_key_references_from_xml_inventory_lines(
1697
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1701
def _find_text_key_references_from_xml_inventory_lines(self,
1703
"""Core routine for extracting references to texts from inventories.
1705
This performs the translation of xml lines to revision ids.
1707
:param line_iterator: An iterator of lines, origin_version_id
1708
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1709
to whether they were referred to by the inventory of the
1710
revision_id that they contain. Note that if that revision_id was
1711
not part of the line_iterator's output then False will be given -
1712
even though it may actually refer to that key.
1714
if not self._serializer.support_altered_by_hack:
1715
raise AssertionError(
1716
"_find_text_key_references_from_xml_inventory_lines only "
1717
"supported for branches which store inventory as unnested xml"
1718
", not on %r" % self)
1721
# this code needs to read every new line in every inventory for the
1722
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
1723
# not present in one of those inventories is unnecessary but not
1724
# harmful because we are filtering by the revision id marker in the
1725
# inventory lines : we only select file ids altered in one of those
1726
# revisions. We don't need to see all lines in the inventory because
1727
# only those added in an inventory in rev X can contain a revision=X
1729
unescape_revid_cache = {}
1730
unescape_fileid_cache = {}
1732
# jam 20061218 In a big fetch, this handles hundreds of thousands
1733
# of lines, so it has had a lot of inlining and optimizing done.
1734
# Sorry that it is a little bit messy.
1735
# Move several functions to be local variables, since this is a long
1737
search = self._file_ids_altered_regex.search
1738
unescape = _unescape_xml
1739
setdefault = result.setdefault
1740
for line, line_key in line_iterator:
1741
match = search(line)
1744
# One call to match.group() returning multiple items is quite a
1745
# bit faster than 2 calls to match.group() each returning 1
1746
file_id, revision_id = match.group('file_id', 'revision_id')
1748
# Inlining the cache lookups helps a lot when you make 170,000
1749
# lines and 350k ids, versus 8.4 unique ids.
1750
# Using a cache helps in 2 ways:
1751
# 1) Avoids unnecessary decoding calls
1752
# 2) Re-uses cached strings, which helps in future set and
1754
# (2) is enough that removing encoding entirely along with
1755
# the cache (so we are using plain strings) results in no
1756
# performance improvement.
1758
revision_id = unescape_revid_cache[revision_id]
1760
unescaped = unescape(revision_id)
1761
unescape_revid_cache[revision_id] = unescaped
1762
revision_id = unescaped
1764
# Note that unconditionally unescaping means that we deserialise
1765
# every fileid, which for general 'pull' is not great, but we don't
1766
# really want to have some many fulltexts that this matters anyway.
1769
file_id = unescape_fileid_cache[file_id]
1771
unescaped = unescape(file_id)
1772
unescape_fileid_cache[file_id] = unescaped
1775
key = (file_id, revision_id)
1776
setdefault(key, False)
1777
if revision_id == line_key[-1]:
1781
def _inventory_xml_lines_for_keys(self, keys):
1782
"""Get a line iterator of the sort needed for findind references.
1784
Not relevant for non-xml inventory repositories.
1786
Ghosts in revision_keys are ignored.
1788
:param revision_keys: The revision keys for the inventories to inspect.
1789
:return: An iterator over (inventory line, revid) for the fulltexts of
1790
all of the xml inventories specified by revision_keys.
1792
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1793
for record in stream:
1794
if record.storage_kind != 'absent':
1795
chunks = record.get_bytes_as('chunked')
1796
revid = record.key[-1]
1797
lines = osutils.chunks_to_lines(chunks)
1801
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1803
"""Helper routine for fileids_altered_by_revision_ids.
1805
This performs the translation of xml lines to revision ids.
1807
:param line_iterator: An iterator of lines, origin_version_id
1808
:param revision_ids: The revision ids to filter for. This should be a
1809
set or other type which supports efficient __contains__ lookups, as
1810
the revision id from each parsed line will be looked up in the
1811
revision_ids filter.
1812
:return: a dictionary mapping altered file-ids to an iterable of
1813
revision_ids. Each altered file-ids has the exact revision_ids that
1814
altered it listed explicitly.
1816
seen = set(self._find_text_key_references_from_xml_inventory_lines(
1817
line_iterator).iterkeys())
1818
# Note that revision_ids are revision keys.
1819
parent_maps = self.revisions.get_parent_map(revision_ids)
1821
map(parents.update, parent_maps.itervalues())
1822
parents.difference_update(revision_ids)
1823
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
1824
self._inventory_xml_lines_for_keys(parents)))
1825
new_keys = seen - parent_seen
1827
setdefault = result.setdefault
1828
for key in new_keys:
1829
setdefault(key[0], set()).add(key[-1])
1832
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
1833
"""Find the file ids and versions affected by revisions.
1835
:param revisions: an iterable containing revision ids.
1836
:param _inv_weave: The inventory weave from this repository or None.
1837
If None, the inventory weave will be opened automatically.
1838
:return: a dictionary mapping altered file-ids to an iterable of
1839
revision_ids. Each altered file-ids has the exact revision_ids that
1840
altered it listed explicitly.
1842
selected_keys = set((revid,) for revid in revision_ids)
1843
w = _inv_weave or self.inventories
1844
pb = ui.ui_factory.nested_progress_bar()
1846
return self._find_file_ids_from_xml_inventory_lines(
1847
w.iter_lines_added_or_present_in_keys(
1848
selected_keys, pb=pb),
1853
def iter_files_bytes(self, desired_files):
1854
"""Iterate through file versions.
1856
Files will not necessarily be returned in the order they occur in
1857
desired_files. No specific order is guaranteed.
1859
Yields pairs of identifier, bytes_iterator. identifier is an opaque
1860
value supplied by the caller as part of desired_files. It should
1861
uniquely identify the file version in the caller's context. (Examples:
1862
an index number or a TreeTransform trans_id.)
1864
bytes_iterator is an iterable of bytestrings for the file. The
1865
kind of iterable and length of the bytestrings are unspecified, but for
1866
this implementation, it is a list of bytes produced by
1867
VersionedFile.get_record_stream().
1869
:param desired_files: a list of (file_id, revision_id, identifier)
1873
for file_id, revision_id, callable_data in desired_files:
1874
text_keys[(file_id, revision_id)] = callable_data
1875
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
1876
if record.storage_kind == 'absent':
1877
raise errors.RevisionNotPresent(record.key, self)
1878
yield text_keys[record.key], record.get_bytes_as('fulltext')
1880
def _generate_text_key_index(self, text_key_references=None,
1882
"""Generate a new text key index for the repository.
1884
This is an expensive function that will take considerable time to run.
1886
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
1887
list of parents, also text keys. When a given key has no parents,
1888
the parents list will be [NULL_REVISION].
1890
# All revisions, to find inventory parents.
1891
if ancestors is None:
1892
graph = self.get_graph()
1893
ancestors = graph.get_parent_map(self.all_revision_ids())
1894
if text_key_references is None:
1895
text_key_references = self.find_text_key_references()
1896
pb = ui.ui_factory.nested_progress_bar()
1898
return self._do_generate_text_key_index(ancestors,
1899
text_key_references, pb)
1903
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
1904
"""Helper for _generate_text_key_index to avoid deep nesting."""
1905
revision_order = tsort.topo_sort(ancestors)
1906
invalid_keys = set()
1908
for revision_id in revision_order:
1909
revision_keys[revision_id] = set()
1910
text_count = len(text_key_references)
1911
# a cache of the text keys to allow reuse; costs a dict of all the
1912
# keys, but saves a 2-tuple for every child of a given key.
1914
for text_key, valid in text_key_references.iteritems():
1916
invalid_keys.add(text_key)
1918
revision_keys[text_key[1]].add(text_key)
1919
text_key_cache[text_key] = text_key
1920
del text_key_references
1922
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
1923
NULL_REVISION = _mod_revision.NULL_REVISION
1924
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
1925
# too small for large or very branchy trees. However, for 55K path
1926
# trees, it would be easy to use too much memory trivially. Ideally we
1927
# could gauge this by looking at available real memory etc, but this is
1928
# always a tricky proposition.
1929
inventory_cache = lru_cache.LRUCache(10)
1930
batch_size = 10 # should be ~150MB on a 55K path tree
1931
batch_count = len(revision_order) / batch_size + 1
1933
pb.update("Calculating text parents", processed_texts, text_count)
1934
for offset in xrange(batch_count):
1935
to_query = revision_order[offset * batch_size:(offset + 1) *
1939
for rev_tree in self.revision_trees(to_query):
1940
revision_id = rev_tree.get_revision_id()
1941
parent_ids = ancestors[revision_id]
1942
for text_key in revision_keys[revision_id]:
1943
pb.update("Calculating text parents", processed_texts)
1944
processed_texts += 1
1945
candidate_parents = []
1946
for parent_id in parent_ids:
1947
parent_text_key = (text_key[0], parent_id)
1949
check_parent = parent_text_key not in \
1950
revision_keys[parent_id]
1952
# the parent parent_id is a ghost:
1953
check_parent = False
1954
# truncate the derived graph against this ghost.
1955
parent_text_key = None
1957
# look at the parent commit details inventories to
1958
# determine possible candidates in the per file graph.
1961
inv = inventory_cache[parent_id]
1963
inv = self.revision_tree(parent_id).inventory
1964
inventory_cache[parent_id] = inv
1965
parent_entry = inv._byid.get(text_key[0], None)
1966
if parent_entry is not None:
1968
text_key[0], parent_entry.revision)
1970
parent_text_key = None
1971
if parent_text_key is not None:
1972
candidate_parents.append(
1973
text_key_cache[parent_text_key])
1974
parent_heads = text_graph.heads(candidate_parents)
1975
new_parents = list(parent_heads)
1976
new_parents.sort(key=lambda x:candidate_parents.index(x))
1977
if new_parents == []:
1978
new_parents = [NULL_REVISION]
1979
text_index[text_key] = new_parents
1981
for text_key in invalid_keys:
1982
text_index[text_key] = [NULL_REVISION]
1985
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
1986
"""Get an iterable listing the keys of all the data introduced by a set
1989
The keys will be ordered so that the corresponding items can be safely
1990
fetched and inserted in that order.
1992
:returns: An iterable producing tuples of (knit-kind, file-id,
1993
versions). knit-kind is one of 'file', 'inventory', 'signatures',
1994
'revisions'. file-id is None unless knit-kind is 'file'.
1996
# XXX: it's a bit weird to control the inventory weave caching in this
1997
# generator. Ideally the caching would be done in fetch.py I think. Or
1998
# maybe this generator should explicitly have the contract that it
1999
# should not be iterated until the previously yielded item has been
2001
inv_w = self.inventories
2003
# file ids that changed
2004
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2006
num_file_ids = len(file_ids)
2007
for file_id, altered_versions in file_ids.iteritems():
2008
if _files_pb is not None:
2009
_files_pb.update("fetch texts", count, num_file_ids)
2011
yield ("file", file_id, altered_versions)
2012
# We're done with the files_pb. Note that it finished by the caller,
2013
# just as it was created by the caller.
2017
yield ("inventory", None, revision_ids)
2020
# XXX: Note ATM no callers actually pay attention to this return
2021
# instead they just use the list of revision ids and ignore
2022
# missing sigs. Consider removing this work entirely
2023
revisions_with_signatures = set(self.signatures.get_parent_map(
2024
[(r,) for r in revision_ids]))
2025
revisions_with_signatures = set(
2026
[r for (r,) in revisions_with_signatures])
2027
revisions_with_signatures.intersection_update(revision_ids)
2028
yield ("signatures", None, revisions_with_signatures)
2031
yield ("revisions", None, revision_ids)
2034
def get_inventory(self, revision_id):
2035
"""Get Inventory object by revision id."""
2036
return self.iter_inventories([revision_id]).next()
2038
def iter_inventories(self, revision_ids):
2039
"""Get many inventories by revision_ids.
2041
This will buffer some or all of the texts used in constructing the
2042
inventories in memory, but will only parse a single inventory at a
2045
:return: An iterator of inventories.
2047
if ((None in revision_ids)
2048
or (_mod_revision.NULL_REVISION in revision_ids)):
2049
raise ValueError('cannot get null revision inventory')
2050
return self._iter_inventories(revision_ids)
2052
def _iter_inventories(self, revision_ids):
2053
"""single-document based inventory iteration."""
2054
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2055
yield self.deserialise_inventory(revision_id, text)
2057
def _iter_inventory_xmls(self, revision_ids):
2058
keys = [(revision_id,) for revision_id in revision_ids]
2059
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2061
for record in stream:
2062
if record.storage_kind != 'absent':
2063
text_chunks[record.key] = record.get_bytes_as('chunked')
2065
raise errors.NoSuchRevision(self, record.key)
2067
chunks = text_chunks.pop(key)
2068
yield ''.join(chunks), key[-1]
2070
def deserialise_inventory(self, revision_id, xml):
2071
"""Transform the xml into an inventory object.
2073
:param revision_id: The expected revision id of the inventory.
2074
:param xml: A serialised inventory.
2076
result = self._serializer.read_inventory_from_string(xml, revision_id,
2077
entry_cache=self._inventory_entry_cache)
2078
if result.revision_id != revision_id:
2079
raise AssertionError('revision id mismatch %s != %s' % (
2080
result.revision_id, revision_id))
2083
def serialise_inventory(self, inv):
2084
return self._serializer.write_inventory_to_string(inv)
2086
def _serialise_inventory_to_lines(self, inv):
2087
return self._serializer.write_inventory_to_lines(inv)
2089
def get_serializer_format(self):
2090
return self._serializer.format_num
2093
def get_inventory_xml(self, revision_id):
2094
"""Get inventory XML as a file object."""
2095
texts = self._iter_inventory_xmls([revision_id])
2097
text, revision_id = texts.next()
2098
except StopIteration:
2099
raise errors.HistoryMissing(self, 'inventory', revision_id)
2103
def get_inventory_sha1(self, revision_id):
2104
"""Return the sha1 hash of the inventory entry
2106
return self.get_revision(revision_id).inventory_sha1
2108
def iter_reverse_revision_history(self, revision_id):
2109
"""Iterate backwards through revision ids in the lefthand history
2111
:param revision_id: The revision id to start with. All its lefthand
2112
ancestors will be traversed.
2114
graph = self.get_graph()
2115
next_id = revision_id
2117
if next_id in (None, _mod_revision.NULL_REVISION):
2120
# Note: The following line may raise KeyError in the event of
2121
# truncated history. We decided not to have a try:except:raise
2122
# RevisionNotPresent here until we see a use for it, because of the
2123
# cost in an inner loop that is by its very nature O(history).
2124
# Robert Collins 20080326
2125
parents = graph.get_parent_map([next_id])[next_id]
2126
if len(parents) == 0:
2129
next_id = parents[0]
2132
def get_revision_inventory(self, revision_id):
2133
"""Return inventory of a past revision."""
2134
# TODO: Unify this with get_inventory()
2135
# bzr 0.0.6 and later imposes the constraint that the inventory_id
2136
# must be the same as its revision, so this is trivial.
2137
if revision_id is None:
2138
# This does not make sense: if there is no revision,
2139
# then it is the current tree inventory surely ?!
2140
# and thus get_root_id() is something that looks at the last
2141
# commit on the branch, and the get_root_id is an inventory check.
2142
raise NotImplementedError
2143
# return Inventory(self.get_root_id())
2145
return self.get_inventory(revision_id)
2147
def is_shared(self):
2148
"""Return True if this repository is flagged as a shared repository."""
2149
raise NotImplementedError(self.is_shared)
2152
def reconcile(self, other=None, thorough=False):
2153
"""Reconcile this repository."""
2154
from bzrlib.reconcile import RepoReconciler
2155
reconciler = RepoReconciler(self, thorough=thorough)
2156
reconciler.reconcile()
2159
def _refresh_data(self):
2160
"""Helper called from lock_* to ensure coherency with disk.
2162
The default implementation does nothing; it is however possible
2163
for repositories to maintain loaded indices across multiple locks
2164
by checking inside their implementation of this method to see
2165
whether their indices are still valid. This depends of course on
2166
the disk format being validatable in this manner. This method is
2167
also called by the refresh_data() public interface to cause a refresh
2168
to occur while in a write lock so that data inserted by a smart server
2169
push operation is visible on the client's instance of the physical
2174
def revision_tree(self, revision_id):
2175
"""Return Tree for a revision on this branch.
2177
`revision_id` may be NULL_REVISION for the empty tree revision.
2179
revision_id = _mod_revision.ensure_null(revision_id)
2180
# TODO: refactor this to use an existing revision object
2181
# so we don't need to read it in twice.
2182
if revision_id == _mod_revision.NULL_REVISION:
2183
return RevisionTree(self, Inventory(root_id=None),
2184
_mod_revision.NULL_REVISION)
2186
inv = self.get_revision_inventory(revision_id)
2187
return RevisionTree(self, inv, revision_id)
2189
def revision_trees(self, revision_ids):
2190
"""Return Trees for revisions in this repository.
2192
:param revision_ids: a sequence of revision-ids;
2193
a revision-id may not be None or 'null:'
2195
inventories = self.iter_inventories(revision_ids)
2196
for inv in inventories:
2197
yield RevisionTree(self, inv, inv.revision_id)
2199
def _filtered_revision_trees(self, revision_ids, file_ids):
2200
"""Return Tree for a revision on this branch with only some files.
2202
:param revision_ids: a sequence of revision-ids;
2203
a revision-id may not be None or 'null:'
2204
:param file_ids: if not None, the result is filtered
2205
so that only those file-ids, their parents and their
2206
children are included.
2208
inventories = self.iter_inventories(revision_ids)
2209
for inv in inventories:
2210
# Should we introduce a FilteredRevisionTree class rather
2211
# than pre-filter the inventory here?
2212
filtered_inv = inv.filter(file_ids)
2213
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2216
def get_ancestry(self, revision_id, topo_sorted=True):
2217
"""Return a list of revision-ids integrated by a revision.
2219
The first element of the list is always None, indicating the origin
2220
revision. This might change when we have history horizons, or
2221
perhaps we should have a new API.
2223
This is topologically sorted.
2225
if _mod_revision.is_null(revision_id):
2227
if not self.has_revision(revision_id):
2228
raise errors.NoSuchRevision(self, revision_id)
2229
graph = self.get_graph()
2231
search = graph._make_breadth_first_searcher([revision_id])
2234
found, ghosts = search.next_with_ghosts()
2235
except StopIteration:
2238
if _mod_revision.NULL_REVISION in keys:
2239
keys.remove(_mod_revision.NULL_REVISION)
2241
parent_map = graph.get_parent_map(keys)
2242
keys = tsort.topo_sort(parent_map)
2243
return [None] + list(keys)
2246
"""Compress the data within the repository.
2248
This operation only makes sense for some repository types. For other
2249
types it should be a no-op that just returns.
2251
This stub method does not require a lock, but subclasses should use
2252
@needs_write_lock as this is a long running call its reasonable to
2253
implicitly lock for the user.
2257
@deprecated_method(one_six)
2258
def print_file(self, file, revision_id):
2259
"""Print `file` to stdout.
2261
FIXME RBC 20060125 as John Meinel points out this is a bad api
2262
- it writes to stdout, it assumes that that is valid etc. Fix
2263
by creating a new more flexible convenience function.
2265
tree = self.revision_tree(revision_id)
2266
# use inventory as it was in that revision
2267
file_id = tree.inventory.path2id(file)
2269
# TODO: jam 20060427 Write a test for this code path
2270
# it had a bug in it, and was raising the wrong
2272
raise errors.BzrError("%r is not present in revision %s" % (file, revision_id))
2273
tree.print_file(file_id)
2275
def get_transaction(self):
2276
return self.control_files.get_transaction()
2278
@deprecated_method(one_one)
2279
def get_parents(self, revision_ids):
2280
"""See StackedParentsProvider.get_parents"""
2281
parent_map = self.get_parent_map(revision_ids)
2282
return [parent_map.get(r, None) for r in revision_ids]
2284
def get_parent_map(self, revision_ids):
2285
"""See graph._StackedParentsProvider.get_parent_map"""
2286
# revisions index works in keys; this just works in revisions
2287
# therefore wrap and unwrap
2290
for revision_id in revision_ids:
2291
if revision_id == _mod_revision.NULL_REVISION:
2292
result[revision_id] = ()
2293
elif revision_id is None:
2294
raise ValueError('get_parent_map(None) is not valid')
2296
query_keys.append((revision_id ,))
2297
for ((revision_id,), parent_keys) in \
2298
self.revisions.get_parent_map(query_keys).iteritems():
2300
result[revision_id] = tuple(parent_revid
2301
for (parent_revid,) in parent_keys)
2303
result[revision_id] = (_mod_revision.NULL_REVISION,)
2306
def _make_parents_provider(self):
2309
def get_graph(self, other_repository=None):
2310
"""Return the graph walker for this repository format"""
2311
parents_provider = self._make_parents_provider()
2312
if (other_repository is not None and
2313
not self.has_same_location(other_repository)):
2314
parents_provider = graph._StackedParentsProvider(
2315
[parents_provider, other_repository._make_parents_provider()])
2316
return graph.Graph(parents_provider)
2318
def _get_versioned_file_checker(self, text_key_references=None):
2319
"""Return an object suitable for checking versioned files.
2321
:param text_key_references: if non-None, an already built
2322
dictionary mapping text keys ((fileid, revision_id) tuples)
2323
to whether they were referred to by the inventory of the
2324
revision_id that they contain. If None, this will be
2327
return _VersionedFileChecker(self,
2328
text_key_references=text_key_references)
2330
def revision_ids_to_search_result(self, result_set):
2331
"""Convert a set of revision ids to a graph SearchResult."""
2332
result_parents = set()
2333
for parents in self.get_graph().get_parent_map(
2334
result_set).itervalues():
2335
result_parents.update(parents)
2336
included_keys = result_set.intersection(result_parents)
2337
start_keys = result_set.difference(included_keys)
2338
exclude_keys = result_parents.difference(result_set)
2339
result = graph.SearchResult(start_keys, exclude_keys,
2340
len(result_set), result_set)
2344
def set_make_working_trees(self, new_value):
2345
"""Set the policy flag for making working trees when creating branches.
2347
This only applies to branches that use this repository.
2349
The default is 'True'.
2350
:param new_value: True to restore the default, False to disable making
2353
raise NotImplementedError(self.set_make_working_trees)
2355
def make_working_trees(self):
2356
"""Returns the policy for making working trees on new branches."""
2357
raise NotImplementedError(self.make_working_trees)
2360
def sign_revision(self, revision_id, gpg_strategy):
2361
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2362
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2365
def has_signature_for_revision_id(self, revision_id):
2366
"""Query for a revision signature for revision_id in the repository."""
2367
if not self.has_revision(revision_id):
2368
raise errors.NoSuchRevision(self, revision_id)
2369
sig_present = (1 == len(
2370
self.signatures.get_parent_map([(revision_id,)])))
2374
def get_signature_text(self, revision_id):
2375
"""Return the text for a signature."""
2376
stream = self.signatures.get_record_stream([(revision_id,)],
2378
record = stream.next()
2379
if record.storage_kind == 'absent':
2380
raise errors.NoSuchRevision(self, revision_id)
2381
return record.get_bytes_as('fulltext')
2384
def check(self, revision_ids=None):
2385
"""Check consistency of all history of given revision_ids.
2387
Different repository implementations should override _check().
2389
:param revision_ids: A non-empty list of revision_ids whose ancestry
2390
will be checked. Typically the last revision_id of a branch.
2392
return self._check(revision_ids)
2394
def _check(self, revision_ids):
2395
result = check.Check(self)
2399
def _warn_if_deprecated(self):
2400
global _deprecation_warning_done
2401
if _deprecation_warning_done:
2403
_deprecation_warning_done = True
2404
warning("Format %s for %s is deprecated - please use 'bzr upgrade' to get better performance"
2405
% (self._format, self.bzrdir.transport.base))
2407
def supports_rich_root(self):
2408
return self._format.rich_root_data
2410
def _check_ascii_revisionid(self, revision_id, method):
2411
"""Private helper for ascii-only repositories."""
2412
# weave repositories refuse to store revisionids that are non-ascii.
2413
if revision_id is not None:
2414
# weaves require ascii revision ids.
2415
if isinstance(revision_id, unicode):
2417
revision_id.encode('ascii')
2418
except UnicodeEncodeError:
2419
raise errors.NonAsciiRevisionId(method, self)
2422
revision_id.decode('ascii')
2423
except UnicodeDecodeError:
2424
raise errors.NonAsciiRevisionId(method, self)
2426
def revision_graph_can_have_wrong_parents(self):
2427
"""Is it possible for this repository to have a revision graph with
2430
If True, then this repository must also implement
2431
_find_inconsistent_revision_parents so that check and reconcile can
2432
check for inconsistencies before proceeding with other checks that may
2433
depend on the revision index being consistent.
2435
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2438
# remove these delegates a while after bzr 0.15
2439
def __make_delegated(name, from_module):
2440
def _deprecated_repository_forwarder():
2441
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2442
% (name, from_module),
2445
m = __import__(from_module, globals(), locals(), [name])
2447
return getattr(m, name)
2448
except AttributeError:
2449
raise AttributeError('module %s has no name %s'
2451
globals()[name] = _deprecated_repository_forwarder
2454
'AllInOneRepository',
2455
'WeaveMetaDirRepository',
2456
'PreSplitOutRepositoryFormat',
2457
'RepositoryFormat4',
2458
'RepositoryFormat5',
2459
'RepositoryFormat6',
2460
'RepositoryFormat7',
2462
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2466
'RepositoryFormatKnit',
2467
'RepositoryFormatKnit1',
2469
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2472
def install_revision(repository, rev, revision_tree):
2473
"""Install all revision data into a repository."""
2474
install_revisions(repository, [(rev, revision_tree, None)])
2477
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2478
"""Install all revision data into a repository.
2480
Accepts an iterable of revision, tree, signature tuples. The signature
2483
repository.start_write_group()
2485
for n, (revision, revision_tree, signature) in enumerate(iterable):
2486
_install_revision(repository, revision, revision_tree, signature)
2488
pb.update('Transferring revisions', n + 1, num_revisions)
2490
repository.abort_write_group()
2493
repository.commit_write_group()
2496
def _install_revision(repository, rev, revision_tree, signature):
2497
"""Install all revision data into a repository."""
2498
present_parents = []
2500
for p_id in rev.parent_ids:
2501
if repository.has_revision(p_id):
2502
present_parents.append(p_id)
2503
parent_trees[p_id] = repository.revision_tree(p_id)
2505
parent_trees[p_id] = repository.revision_tree(
2506
_mod_revision.NULL_REVISION)
2508
inv = revision_tree.inventory
2509
entries = inv.iter_entries()
2510
# backwards compatibility hack: skip the root id.
2511
if not repository.supports_rich_root():
2512
path, root = entries.next()
2513
if root.revision != rev.revision_id:
2514
raise errors.IncompatibleRevision(repr(repository))
2516
for path, ie in entries:
2517
text_keys[(ie.file_id, ie.revision)] = ie
2518
text_parent_map = repository.texts.get_parent_map(text_keys)
2519
missing_texts = set(text_keys) - set(text_parent_map)
2520
# Add the texts that are not already present
2521
for text_key in missing_texts:
2522
ie = text_keys[text_key]
2524
# FIXME: TODO: The following loop overlaps/duplicates that done by
2525
# commit to determine parents. There is a latent/real bug here where
2526
# the parents inserted are not those commit would do - in particular
2527
# they are not filtered by heads(). RBC, AB
2528
for revision, tree in parent_trees.iteritems():
2529
if ie.file_id not in tree:
2531
parent_id = tree.inventory[ie.file_id].revision
2532
if parent_id in text_parents:
2534
text_parents.append((ie.file_id, parent_id))
2535
lines = revision_tree.get_file(ie.file_id).readlines()
2536
repository.texts.add_lines(text_key, text_parents, lines)
2538
# install the inventory
2539
repository.add_inventory(rev.revision_id, inv, present_parents)
2540
except errors.RevisionAlreadyPresent:
2542
if signature is not None:
2543
repository.add_signature_text(rev.revision_id, signature)
2544
repository.add_revision(rev.revision_id, rev, inv)
2547
class MetaDirRepository(Repository):
2548
"""Repositories in the new meta-dir layout.
2550
:ivar _transport: Transport for access to repository control files,
2551
typically pointing to .bzr/repository.
2554
def __init__(self, _format, a_bzrdir, control_files):
2555
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2556
self._transport = control_files._transport
2558
def is_shared(self):
2559
"""Return True if this repository is flagged as a shared repository."""
2560
return self._transport.has('shared-storage')
2563
def set_make_working_trees(self, new_value):
2564
"""Set the policy flag for making working trees when creating branches.
2566
This only applies to branches that use this repository.
2568
The default is 'True'.
2569
:param new_value: True to restore the default, False to disable making
2574
self._transport.delete('no-working-trees')
2575
except errors.NoSuchFile:
2578
self._transport.put_bytes('no-working-trees', '',
2579
mode=self.bzrdir._get_file_mode())
2581
def make_working_trees(self):
2582
"""Returns the policy for making working trees on new branches."""
2583
return not self._transport.has('no-working-trees')
2586
class MetaDirVersionedFileRepository(MetaDirRepository):
2587
"""Repositories in a meta-dir, that work via versioned file objects."""
2589
def __init__(self, _format, a_bzrdir, control_files):
2590
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2594
network_format_registry = registry.FormatRegistry()
2595
"""Registry of formats indexed by their network name.
2597
The network name for a repository format is an identifier that can be used when
2598
referring to formats with smart server operations. See
2599
RepositoryFormat.network_name() for more detail.
2603
format_registry = registry.FormatRegistry(network_format_registry)
2604
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
2606
This can contain either format instances themselves, or classes/factories that
2607
can be called to obtain one.
2611
#####################################################################
2612
# Repository Formats
2614
class RepositoryFormat(object):
2615
"""A repository format.
2617
Formats provide four things:
2618
* An initialization routine to construct repository data on disk.
2619
* a optional format string which is used when the BzrDir supports
2621
* an open routine which returns a Repository instance.
2622
* A network name for referring to the format in smart server RPC
2625
There is one and only one Format subclass for each on-disk format. But
2626
there can be one Repository subclass that is used for several different
2627
formats. The _format attribute on a Repository instance can be used to
2628
determine the disk format.
2630
Formats are placed in a registry by their format string for reference
2631
during opening. These should be subclasses of RepositoryFormat for
2634
Once a format is deprecated, just deprecate the initialize and open
2635
methods on the format class. Do not deprecate the object, as the
2636
object may be created even when a repository instnace hasn't been
2639
Common instance attributes:
2640
_matchingbzrdir - the bzrdir format that the repository format was
2641
originally written to work with. This can be used if manually
2642
constructing a bzrdir and repository, or more commonly for test suite
2646
# Set to True or False in derived classes. True indicates that the format
2647
# supports ghosts gracefully.
2648
supports_ghosts = None
2649
# Can this repository be given external locations to lookup additional
2650
# data. Set to True or False in derived classes.
2651
supports_external_lookups = None
2652
# What order should fetch operations request streams in?
2653
# The default is unordered as that is the cheapest for an origin to
2655
_fetch_order = 'unordered'
2656
# Does this repository format use deltas that can be fetched as-deltas ?
2657
# (E.g. knits, where the knit deltas can be transplanted intact.
2658
# We default to False, which will ensure that enough data to get
2659
# a full text out of any fetch stream will be grabbed.
2660
_fetch_uses_deltas = False
2661
# Should fetch trigger a reconcile after the fetch? Only needed for
2662
# some repository formats that can suffer internal inconsistencies.
2663
_fetch_reconcile = False
2666
return "<%s>" % self.__class__.__name__
2668
def __eq__(self, other):
2669
# format objects are generally stateless
2670
return isinstance(other, self.__class__)
2672
def __ne__(self, other):
2673
return not self == other
2676
def find_format(klass, a_bzrdir):
2677
"""Return the format for the repository object in a_bzrdir.
2679
This is used by bzr native formats that have a "format" file in
2680
the repository. Other methods may be used by different types of
2684
transport = a_bzrdir.get_repository_transport(None)
2685
format_string = transport.get("format").read()
2686
return format_registry.get(format_string)
2687
except errors.NoSuchFile:
2688
raise errors.NoRepositoryPresent(a_bzrdir)
2690
raise errors.UnknownFormatError(format=format_string,
2694
def register_format(klass, format):
2695
format_registry.register(format.get_format_string(), format)
2698
def unregister_format(klass, format):
2699
format_registry.remove(format.get_format_string())
2702
def get_default_format(klass):
2703
"""Return the current default format."""
2704
from bzrlib import bzrdir
2705
return bzrdir.format_registry.make_bzrdir('default').repository_format
2707
def get_format_string(self):
2708
"""Return the ASCII format string that identifies this format.
2710
Note that in pre format ?? repositories the format string is
2711
not permitted nor written to disk.
2713
raise NotImplementedError(self.get_format_string)
2715
def get_format_description(self):
2716
"""Return the short description for this format."""
2717
raise NotImplementedError(self.get_format_description)
2719
# TODO: this shouldn't be in the base class, it's specific to things that
2720
# use weaves or knits -- mbp 20070207
2721
def _get_versioned_file_store(self,
2726
versionedfile_class=None,
2727
versionedfile_kwargs={},
2729
if versionedfile_class is None:
2730
versionedfile_class = self._versionedfile_class
2731
weave_transport = control_files._transport.clone(name)
2732
dir_mode = control_files._dir_mode
2733
file_mode = control_files._file_mode
2734
return VersionedFileStore(weave_transport, prefixed=prefixed,
2736
file_mode=file_mode,
2737
versionedfile_class=versionedfile_class,
2738
versionedfile_kwargs=versionedfile_kwargs,
2741
def initialize(self, a_bzrdir, shared=False):
2742
"""Initialize a repository of this format in a_bzrdir.
2744
:param a_bzrdir: The bzrdir to put the new repository in it.
2745
:param shared: The repository should be initialized as a sharable one.
2746
:returns: The new repository object.
2748
This may raise UninitializableFormat if shared repository are not
2749
compatible the a_bzrdir.
2751
raise NotImplementedError(self.initialize)
2753
def is_supported(self):
2754
"""Is this format supported?
2756
Supported formats must be initializable and openable.
2757
Unsupported formats may not support initialization or committing or
2758
some other features depending on the reason for not being supported.
2762
def network_name(self):
2763
"""A simple byte string uniquely identifying this format for RPC calls.
2765
MetaDir repository formats use their disk format string to identify the
2766
repository over the wire. All in one formats such as bzr < 0.8, and
2767
foreign formats like svn/git and hg should use some marker which is
2768
unique and immutable.
2770
raise NotImplementedError(self.network_name)
2772
def check_conversion_target(self, target_format):
2773
raise NotImplementedError(self.check_conversion_target)
2775
def open(self, a_bzrdir, _found=False):
2776
"""Return an instance of this format for the bzrdir a_bzrdir.
2778
_found is a private parameter, do not use it.
2780
raise NotImplementedError(self.open)
2783
class MetaDirRepositoryFormat(RepositoryFormat):
2784
"""Common base class for the new repositories using the metadir layout."""
2786
rich_root_data = False
2787
supports_tree_reference = False
2788
supports_external_lookups = False
2791
def _matchingbzrdir(self):
2792
matching = bzrdir.BzrDirMetaFormat1()
2793
matching.repository_format = self
2797
super(MetaDirRepositoryFormat, self).__init__()
2799
def _create_control_files(self, a_bzrdir):
2800
"""Create the required files and the initial control_files object."""
2801
# FIXME: RBC 20060125 don't peek under the covers
2802
# NB: no need to escape relative paths that are url safe.
2803
repository_transport = a_bzrdir.get_repository_transport(self)
2804
control_files = lockable_files.LockableFiles(repository_transport,
2805
'lock', lockdir.LockDir)
2806
control_files.create_lock()
2807
return control_files
2809
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
2810
"""Upload the initial blank content."""
2811
control_files = self._create_control_files(a_bzrdir)
2812
control_files.lock_write()
2813
transport = control_files._transport
2815
utf8_files += [('shared-storage', '')]
2817
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
2818
for (filename, content_stream) in files:
2819
transport.put_file(filename, content_stream,
2820
mode=a_bzrdir._get_file_mode())
2821
for (filename, content_bytes) in utf8_files:
2822
transport.put_bytes_non_atomic(filename, content_bytes,
2823
mode=a_bzrdir._get_file_mode())
2825
control_files.unlock()
2827
def network_name(self):
2828
"""Metadir formats have matching disk and network format strings."""
2829
return self.get_format_string()
2832
# Pre-0.8 formats that don't have a disk format string (because they are
2833
# versioned by the matching control directory). We use the control directories
2834
# disk format string as a key for the network_name because they meet the
2835
# constraints (simple string, unique, immmutable).
2836
network_format_registry.register_lazy(
2837
"Bazaar-NG branch, format 5\n",
2838
'bzrlib.repofmt.weaverepo',
2839
'RepositoryFormat5',
2841
network_format_registry.register_lazy(
2842
"Bazaar-NG branch, format 6\n",
2843
'bzrlib.repofmt.weaverepo',
2844
'RepositoryFormat6',
2847
# formats which have no format string are not discoverable or independently
2848
# creatable on disk, so are not registered in format_registry. They're
2849
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
2850
# needed, it's constructed directly by the BzrDir. Non-native formats where
2851
# the repository is not separately opened are similar.
2853
format_registry.register_lazy(
2854
'Bazaar-NG Repository format 7',
2855
'bzrlib.repofmt.weaverepo',
2859
format_registry.register_lazy(
2860
'Bazaar-NG Knit Repository Format 1',
2861
'bzrlib.repofmt.knitrepo',
2862
'RepositoryFormatKnit1',
2865
format_registry.register_lazy(
2866
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
2867
'bzrlib.repofmt.knitrepo',
2868
'RepositoryFormatKnit3',
2871
format_registry.register_lazy(
2872
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
2873
'bzrlib.repofmt.knitrepo',
2874
'RepositoryFormatKnit4',
2877
# Pack-based formats. There is one format for pre-subtrees, and one for
2878
# post-subtrees to allow ease of testing.
2879
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
2880
format_registry.register_lazy(
2881
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
2882
'bzrlib.repofmt.pack_repo',
2883
'RepositoryFormatKnitPack1',
2885
format_registry.register_lazy(
2886
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
2887
'bzrlib.repofmt.pack_repo',
2888
'RepositoryFormatKnitPack3',
2890
format_registry.register_lazy(
2891
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
2892
'bzrlib.repofmt.pack_repo',
2893
'RepositoryFormatKnitPack4',
2895
format_registry.register_lazy(
2896
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
2897
'bzrlib.repofmt.pack_repo',
2898
'RepositoryFormatKnitPack5',
2900
format_registry.register_lazy(
2901
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
2902
'bzrlib.repofmt.pack_repo',
2903
'RepositoryFormatKnitPack5RichRoot',
2905
format_registry.register_lazy(
2906
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
2907
'bzrlib.repofmt.pack_repo',
2908
'RepositoryFormatKnitPack5RichRootBroken',
2910
format_registry.register_lazy(
2911
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
2912
'bzrlib.repofmt.pack_repo',
2913
'RepositoryFormatKnitPack6',
2915
format_registry.register_lazy(
2916
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
2917
'bzrlib.repofmt.pack_repo',
2918
'RepositoryFormatKnitPack6RichRoot',
2921
# Development formats.
2922
# 1.7->1.8 go below here
2923
format_registry.register_lazy(
2924
"Bazaar development format 2 (needs bzr.dev from before 1.8)\n",
2925
'bzrlib.repofmt.pack_repo',
2926
'RepositoryFormatPackDevelopment2',
2928
format_registry.register_lazy(
2929
("Bazaar development format 2 with subtree support "
2930
"(needs bzr.dev from before 1.8)\n"),
2931
'bzrlib.repofmt.pack_repo',
2932
'RepositoryFormatPackDevelopment2Subtree',
2936
class InterRepository(InterObject):
2937
"""This class represents operations taking place between two repositories.
2939
Its instances have methods like copy_content and fetch, and contain
2940
references to the source and target repositories these operations can be
2943
Often we will provide convenience methods on 'repository' which carry out
2944
operations with another repository - they will always forward to
2945
InterRepository.get(other).method_name(parameters).
2948
_walk_to_common_revisions_batch_size = 50
2950
"""The available optimised InterRepository types."""
2953
def copy_content(self, revision_id=None):
2954
"""Make a complete copy of the content in self into destination.
2956
This is a destructive operation! Do not use it on existing
2959
:param revision_id: Only copy the content needed to construct
2960
revision_id and its parents.
2963
self.target.set_make_working_trees(self.source.make_working_trees())
2964
except NotImplementedError:
2966
self.target.fetch(self.source, revision_id=revision_id)
2969
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
2971
"""Fetch the content required to construct revision_id.
2973
The content is copied from self.source to self.target.
2975
:param revision_id: if None all content is copied, if NULL_REVISION no
2977
:param pb: optional progress bar to use for progress reports. If not
2978
provided a default one will be created.
2981
from bzrlib.fetch import RepoFetcher
2982
f = RepoFetcher(to_repository=self.target,
2983
from_repository=self.source,
2984
last_revision=revision_id,
2985
fetch_spec=fetch_spec,
2986
pb=pb, find_ghosts=find_ghosts)
2988
def _walk_to_common_revisions(self, revision_ids):
2989
"""Walk out from revision_ids in source to revisions target has.
2991
:param revision_ids: The start point for the search.
2992
:return: A set of revision ids.
2994
target_graph = self.target.get_graph()
2995
revision_ids = frozenset(revision_ids)
2996
# Fast path for the case where all the revisions are already in the
2998
# (Although this does incur an extra round trip for the
2999
# fairly common case where the target doesn't already have the revision
3001
if set(target_graph.get_parent_map(revision_ids)) == revision_ids:
3002
return graph.SearchResult(revision_ids, set(), 0, set())
3003
missing_revs = set()
3004
source_graph = self.source.get_graph()
3005
# ensure we don't pay silly lookup costs.
3006
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3007
null_set = frozenset([_mod_revision.NULL_REVISION])
3008
searcher_exhausted = False
3012
# Iterate the searcher until we have enough next_revs
3013
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3015
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3016
next_revs.update(next_revs_part)
3017
ghosts.update(ghosts_part)
3018
except StopIteration:
3019
searcher_exhausted = True
3021
# If there are ghosts in the source graph, and the caller asked for
3022
# them, make sure that they are present in the target.
3023
# We don't care about other ghosts as we can't fetch them and
3024
# haven't been asked to.
3025
ghosts_to_check = set(revision_ids.intersection(ghosts))
3026
revs_to_get = set(next_revs).union(ghosts_to_check)
3028
have_revs = set(target_graph.get_parent_map(revs_to_get))
3029
# we always have NULL_REVISION present.
3030
have_revs = have_revs.union(null_set)
3031
# Check if the target is missing any ghosts we need.
3032
ghosts_to_check.difference_update(have_revs)
3034
# One of the caller's revision_ids is a ghost in both the
3035
# source and the target.
3036
raise errors.NoSuchRevision(
3037
self.source, ghosts_to_check.pop())
3038
missing_revs.update(next_revs - have_revs)
3039
# Because we may have walked past the original stop point, make
3040
# sure everything is stopped
3041
stop_revs = searcher.find_seen_ancestors(have_revs)
3042
searcher.stop_searching_any(stop_revs)
3043
if searcher_exhausted:
3045
return searcher.get_result()
3047
@deprecated_method(one_two)
3049
def missing_revision_ids(self, revision_id=None, find_ghosts=True):
3050
"""Return the revision ids that source has that target does not.
3052
These are returned in topological order.
3054
:param revision_id: only return revision ids included by this
3056
:param find_ghosts: If True find missing revisions in deep history
3057
rather than just finding the surface difference.
3059
return list(self.search_missing_revision_ids(
3060
revision_id, find_ghosts).get_keys())
3063
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3064
"""Return the revision ids that source has that target does not.
3066
:param revision_id: only return revision ids included by this
3068
:param find_ghosts: If True find missing revisions in deep history
3069
rather than just finding the surface difference.
3070
:return: A bzrlib.graph.SearchResult.
3072
# stop searching at found target revisions.
3073
if not find_ghosts and revision_id is not None:
3074
return self._walk_to_common_revisions([revision_id])
3075
# generic, possibly worst case, slow code path.
3076
target_ids = set(self.target.all_revision_ids())
3077
if revision_id is not None:
3078
source_ids = self.source.get_ancestry(revision_id)
3079
if source_ids[0] is not None:
3080
raise AssertionError()
3083
source_ids = self.source.all_revision_ids()
3084
result_set = set(source_ids).difference(target_ids)
3085
return self.source.revision_ids_to_search_result(result_set)
3088
def _same_model(source, target):
3089
"""True if source and target have the same data representation.
3091
Note: this is always called on the base class; overriding it in a
3092
subclass will have no effect.
3095
InterRepository._assert_same_model(source, target)
3097
except errors.IncompatibleRepositories, e:
3101
def _assert_same_model(source, target):
3102
"""Raise an exception if two repositories do not use the same model.
3104
if source.supports_rich_root() != target.supports_rich_root():
3105
raise errors.IncompatibleRepositories(source, target,
3106
"different rich-root support")
3107
if source._serializer != target._serializer:
3108
raise errors.IncompatibleRepositories(source, target,
3109
"different serializers")
3112
class InterSameDataRepository(InterRepository):
3113
"""Code for converting between repositories that represent the same data.
3115
Data format and model must match for this to work.
3119
def _get_repo_format_to_test(self):
3120
"""Repository format for testing with.
3122
InterSameData can pull from subtree to subtree and from non-subtree to
3123
non-subtree, so we test this with the richest repository format.
3125
from bzrlib.repofmt import knitrepo
3126
return knitrepo.RepositoryFormatKnit3()
3129
def is_compatible(source, target):
3130
return InterRepository._same_model(source, target)
3133
class InterWeaveRepo(InterSameDataRepository):
3134
"""Optimised code paths between Weave based repositories.
3136
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3137
implemented lazy inter-object optimisation.
3141
def _get_repo_format_to_test(self):
3142
from bzrlib.repofmt import weaverepo
3143
return weaverepo.RepositoryFormat7()
3146
def is_compatible(source, target):
3147
"""Be compatible with known Weave formats.
3149
We don't test for the stores being of specific types because that
3150
could lead to confusing results, and there is no need to be
3153
from bzrlib.repofmt.weaverepo import (
3159
return (isinstance(source._format, (RepositoryFormat5,
3161
RepositoryFormat7)) and
3162
isinstance(target._format, (RepositoryFormat5,
3164
RepositoryFormat7)))
3165
except AttributeError:
3169
def copy_content(self, revision_id=None):
3170
"""See InterRepository.copy_content()."""
3171
# weave specific optimised path:
3173
self.target.set_make_working_trees(self.source.make_working_trees())
3174
except (errors.RepositoryUpgradeRequired, NotImplemented):
3176
# FIXME do not peek!
3177
if self.source._transport.listable():
3178
pb = ui.ui_factory.nested_progress_bar()
3180
self.target.texts.insert_record_stream(
3181
self.source.texts.get_record_stream(
3182
self.source.texts.keys(), 'topological', False))
3183
pb.update('copying inventory', 0, 1)
3184
self.target.inventories.insert_record_stream(
3185
self.source.inventories.get_record_stream(
3186
self.source.inventories.keys(), 'topological', False))
3187
self.target.signatures.insert_record_stream(
3188
self.source.signatures.get_record_stream(
3189
self.source.signatures.keys(),
3191
self.target.revisions.insert_record_stream(
3192
self.source.revisions.get_record_stream(
3193
self.source.revisions.keys(),
3194
'topological', True))
3198
self.target.fetch(self.source, revision_id=revision_id)
3201
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3202
"""See InterRepository.missing_revision_ids()."""
3203
# we want all revisions to satisfy revision_id in source.
3204
# but we don't want to stat every file here and there.
3205
# we want then, all revisions other needs to satisfy revision_id
3206
# checked, but not those that we have locally.
3207
# so the first thing is to get a subset of the revisions to
3208
# satisfy revision_id in source, and then eliminate those that
3209
# we do already have.
3210
# this is slow on high latency connection to self, but as as this
3211
# disk format scales terribly for push anyway due to rewriting
3212
# inventory.weave, this is considered acceptable.
3214
if revision_id is not None:
3215
source_ids = self.source.get_ancestry(revision_id)
3216
if source_ids[0] is not None:
3217
raise AssertionError()
3220
source_ids = self.source._all_possible_ids()
3221
source_ids_set = set(source_ids)
3222
# source_ids is the worst possible case we may need to pull.
3223
# now we want to filter source_ids against what we actually
3224
# have in target, but don't try to check for existence where we know
3225
# we do not have a revision as that would be pointless.
3226
target_ids = set(self.target._all_possible_ids())
3227
possibly_present_revisions = target_ids.intersection(source_ids_set)
3228
actually_present_revisions = set(
3229
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3230
required_revisions = source_ids_set.difference(actually_present_revisions)
3231
if revision_id is not None:
3232
# we used get_ancestry to determine source_ids then we are assured all
3233
# revisions referenced are present as they are installed in topological order.
3234
# and the tip revision was validated by get_ancestry.
3235
result_set = required_revisions
3237
# if we just grabbed the possibly available ids, then
3238
# we only have an estimate of whats available and need to validate
3239
# that against the revision records.
3241
self.source._eliminate_revisions_not_present(required_revisions))
3242
return self.source.revision_ids_to_search_result(result_set)
3245
class InterKnitRepo(InterSameDataRepository):
3246
"""Optimised code paths between Knit based repositories."""
3249
def _get_repo_format_to_test(self):
3250
from bzrlib.repofmt import knitrepo
3251
return knitrepo.RepositoryFormatKnit1()
3254
def is_compatible(source, target):
3255
"""Be compatible with known Knit formats.
3257
We don't test for the stores being of specific types because that
3258
could lead to confusing results, and there is no need to be
3261
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3263
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3264
isinstance(target._format, RepositoryFormatKnit))
3265
except AttributeError:
3267
return are_knits and InterRepository._same_model(source, target)
3270
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3271
"""See InterRepository.missing_revision_ids()."""
3272
if revision_id is not None:
3273
source_ids = self.source.get_ancestry(revision_id)
3274
if source_ids[0] is not None:
3275
raise AssertionError()
3278
source_ids = self.source.all_revision_ids()
3279
source_ids_set = set(source_ids)
3280
# source_ids is the worst possible case we may need to pull.
3281
# now we want to filter source_ids against what we actually
3282
# have in target, but don't try to check for existence where we know
3283
# we do not have a revision as that would be pointless.
3284
target_ids = set(self.target.all_revision_ids())
3285
possibly_present_revisions = target_ids.intersection(source_ids_set)
3286
actually_present_revisions = set(
3287
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3288
required_revisions = source_ids_set.difference(actually_present_revisions)
3289
if revision_id is not None:
3290
# we used get_ancestry to determine source_ids then we are assured all
3291
# revisions referenced are present as they are installed in topological order.
3292
# and the tip revision was validated by get_ancestry.
3293
result_set = required_revisions
3295
# if we just grabbed the possibly available ids, then
3296
# we only have an estimate of whats available and need to validate
3297
# that against the revision records.
3299
self.source._eliminate_revisions_not_present(required_revisions))
3300
return self.source.revision_ids_to_search_result(result_set)
3303
class InterPackRepo(InterSameDataRepository):
3304
"""Optimised code paths between Pack based repositories."""
3307
def _get_repo_format_to_test(self):
3308
from bzrlib.repofmt import pack_repo
3309
return pack_repo.RepositoryFormatKnitPack1()
3312
def is_compatible(source, target):
3313
"""Be compatible with known Pack formats.
3315
We don't test for the stores being of specific types because that
3316
could lead to confusing results, and there is no need to be
3319
from bzrlib.repofmt.pack_repo import RepositoryFormatPack
3321
are_packs = (isinstance(source._format, RepositoryFormatPack) and
3322
isinstance(target._format, RepositoryFormatPack))
3323
except AttributeError:
3325
return are_packs and InterRepository._same_model(source, target)
3328
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3330
"""See InterRepository.fetch()."""
3331
if (len(self.source._fallback_repositories) > 0 or
3332
len(self.target._fallback_repositories) > 0):
3333
# The pack layer is not aware of fallback repositories, so when
3334
# fetching from a stacked repository or into a stacked repository
3335
# we use the generic fetch logic which uses the VersionedFiles
3336
# attributes on repository.
3337
from bzrlib.fetch import RepoFetcher
3338
fetcher = RepoFetcher(self.target, self.source, revision_id,
3339
pb, find_ghosts, fetch_spec=fetch_spec)
3340
if fetch_spec is not None:
3341
if len(list(fetch_spec.heads)) != 1:
3342
raise AssertionError(
3343
"InterPackRepo.fetch doesn't support "
3344
"fetching multiple heads yet.")
3345
revision_id = list(fetch_spec.heads)[0]
3347
if revision_id is None:
3349
# everything to do - use pack logic
3350
# to fetch from all packs to one without
3351
# inventory parsing etc, IFF nothing to be copied is in the target.
3353
source_revision_ids = frozenset(self.source.all_revision_ids())
3354
revision_ids = source_revision_ids - \
3355
frozenset(self.target.get_parent_map(source_revision_ids))
3356
revision_keys = [(revid,) for revid in revision_ids]
3357
index = self.target._pack_collection.revision_index.combined_index
3358
present_revision_ids = set(item[1][0] for item in
3359
index.iter_entries(revision_keys))
3360
revision_ids = set(revision_ids) - present_revision_ids
3361
# implementing the TODO will involve:
3362
# - detecting when all of a pack is selected
3363
# - avoiding as much as possible pre-selection, so the
3364
# more-core routines such as create_pack_from_packs can filter in
3365
# a just-in-time fashion. (though having a HEADS list on a
3366
# repository might make this a lot easier, because we could
3367
# sensibly detect 'new revisions' without doing a full index scan.
3368
elif _mod_revision.is_null(revision_id):
3373
revision_ids = self.search_missing_revision_ids(revision_id,
3374
find_ghosts=find_ghosts).get_keys()
3375
except errors.NoSuchRevision:
3376
raise errors.InstallFailed([revision_id])
3377
if len(revision_ids) == 0:
3379
return self._pack(self.source, self.target, revision_ids)
3381
def _pack(self, source, target, revision_ids):
3382
from bzrlib.repofmt.pack_repo import Packer
3383
packs = source._pack_collection.all_packs()
3384
pack = Packer(self.target._pack_collection, packs, '.fetch',
3385
revision_ids).pack()
3386
if pack is not None:
3387
self.target._pack_collection._save_pack_names()
3388
copied_revs = pack.get_revision_count()
3389
# Trigger an autopack. This may duplicate effort as we've just done
3390
# a pack creation, but for now it is simpler to think about as
3391
# 'upload data, then repack if needed'.
3392
self.target._pack_collection.autopack()
3393
return (copied_revs, [])
3398
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3399
"""See InterRepository.missing_revision_ids().
3401
:param find_ghosts: Find ghosts throughout the ancestry of
3404
if not find_ghosts and revision_id is not None:
3405
return self._walk_to_common_revisions([revision_id])
3406
elif revision_id is not None:
3407
# Find ghosts: search for revisions pointing from one repository to
3408
# the other, and vice versa, anywhere in the history of revision_id.
3409
graph = self.target.get_graph(other_repository=self.source)
3410
searcher = graph._make_breadth_first_searcher([revision_id])
3414
next_revs, ghosts = searcher.next_with_ghosts()
3415
except StopIteration:
3417
if revision_id in ghosts:
3418
raise errors.NoSuchRevision(self.source, revision_id)
3419
found_ids.update(next_revs)
3420
found_ids.update(ghosts)
3421
found_ids = frozenset(found_ids)
3422
# Double query here: should be able to avoid this by changing the
3423
# graph api further.
3424
result_set = found_ids - frozenset(
3425
self.target.get_parent_map(found_ids))
3427
source_ids = self.source.all_revision_ids()
3428
# source_ids is the worst possible case we may need to pull.
3429
# now we want to filter source_ids against what we actually
3430
# have in target, but don't try to check for existence where we know
3431
# we do not have a revision as that would be pointless.
3432
target_ids = set(self.target.all_revision_ids())
3433
result_set = set(source_ids).difference(target_ids)
3434
return self.source.revision_ids_to_search_result(result_set)
3437
class InterDifferingSerializer(InterKnitRepo):
3440
def _get_repo_format_to_test(self):
3444
def is_compatible(source, target):
3445
"""Be compatible with Knit2 source and Knit3 target"""
3446
if source.supports_rich_root() != target.supports_rich_root():
3448
# Ideally, we'd support fetching if the source had no tree references
3449
# even if it supported them...
3450
if (getattr(source, '_format.supports_tree_reference', False) and
3451
not getattr(target, '_format.supports_tree_reference', False)):
3455
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3456
"""Get the best delta and base for this revision.
3458
:return: (basis_id, delta)
3460
possible_trees = [(parent_id, cache[parent_id])
3461
for parent_id in parent_ids
3462
if parent_id in cache]
3463
if len(possible_trees) == 0:
3464
# There either aren't any parents, or the parents aren't in the
3465
# cache, so just use the last converted tree
3466
possible_trees.append((basis_id, cache[basis_id]))
3468
for basis_id, basis_tree in possible_trees:
3469
delta = tree.inventory._make_delta(basis_tree.inventory)
3470
deltas.append((len(delta), basis_id, delta))
3472
return deltas[0][1:]
3474
def _fetch_batch(self, revision_ids, basis_id, cache):
3475
"""Fetch across a few revisions.
3477
:param revision_ids: The revisions to copy
3478
:param basis_id: The revision_id of a tree that must be in cache, used
3479
as a basis for delta when no other base is available
3480
:param cache: A cache of RevisionTrees that we can use.
3481
:return: The revision_id of the last converted tree. The RevisionTree
3482
for it will be in cache
3484
# Walk though all revisions; get inventory deltas, copy referenced
3485
# texts that delta references, insert the delta, revision and
3489
pending_revisions = []
3490
parent_map = self.source.get_parent_map(revision_ids)
3491
for tree in self.source.revision_trees(revision_ids):
3492
current_revision_id = tree.get_revision_id()
3493
parent_ids = parent_map.get(current_revision_id, ())
3494
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3496
# Find text entries that need to be copied
3497
for old_path, new_path, file_id, entry in delta:
3498
if new_path is not None:
3499
if not (new_path or self.target.supports_rich_root()):
3500
# We don't copy the text for the root node unless the
3501
# target supports_rich_root.
3503
text_keys.add((file_id, entry.revision))
3504
revision = self.source.get_revision(current_revision_id)
3505
pending_deltas.append((basis_id, delta,
3506
current_revision_id, revision.parent_ids))
3507
pending_revisions.append(revision)
3508
cache[current_revision_id] = tree
3509
basis_id = current_revision_id
3511
from_texts = self.source.texts
3512
to_texts = self.target.texts
3513
to_texts.insert_record_stream(from_texts.get_record_stream(
3514
text_keys, self.target._format._fetch_order,
3515
not self.target._format._fetch_uses_deltas))
3517
for delta in pending_deltas:
3518
self.target.add_inventory_by_delta(*delta)
3519
# insert signatures and revisions
3520
for revision in pending_revisions:
3522
signature = self.source.get_signature_text(
3523
revision.revision_id)
3524
self.target.add_signature_text(revision.revision_id,
3526
except errors.NoSuchRevision:
3528
self.target.add_revision(revision.revision_id, revision)
3531
def _fetch_all_revisions(self, revision_ids, pb):
3532
"""Fetch everything for the list of revisions.
3534
:param revision_ids: The list of revisions to fetch. Must be in
3536
:param pb: A ProgressBar
3539
basis_id, basis_tree = self._get_basis(revision_ids[0])
3541
cache = lru_cache.LRUCache(100)
3542
cache[basis_id] = basis_tree
3543
del basis_tree # We don't want to hang on to it here
3544
for offset in range(0, len(revision_ids), batch_size):
3545
self.target.start_write_group()
3547
pb.update('Transferring revisions', offset,
3549
batch = revision_ids[offset:offset+batch_size]
3550
basis_id = self._fetch_batch(batch, basis_id, cache)
3552
self.target.abort_write_group()
3555
self.target.commit_write_group()
3556
pb.update('Transferring revisions', len(revision_ids),
3560
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3562
"""See InterRepository.fetch()."""
3563
if fetch_spec is not None:
3564
raise AssertionError("Not implemented yet...")
3565
revision_ids = self.target.search_missing_revision_ids(self.source,
3566
revision_id, find_ghosts=find_ghosts).get_keys()
3567
if not revision_ids:
3569
revision_ids = tsort.topo_sort(
3570
self.source.get_graph().get_parent_map(revision_ids))
3572
my_pb = ui.ui_factory.nested_progress_bar()
3575
symbol_versioning.warn(
3576
symbol_versioning.deprecated_in((1, 14, 0))
3577
% "pb parameter to fetch()")
3580
self._fetch_all_revisions(revision_ids, pb)
3582
if my_pb is not None:
3584
return len(revision_ids), 0
3586
def _get_basis(self, first_revision_id):
3587
"""Get a revision and tree which exists in the target.
3589
This assumes that first_revision_id is selected for transmission
3590
because all other ancestors are already present. If we can't find an
3591
ancestor we fall back to NULL_REVISION since we know that is safe.
3593
:return: (basis_id, basis_tree)
3595
first_rev = self.source.get_revision(first_revision_id)
3597
basis_id = first_rev.parent_ids[0]
3598
# only valid as a basis if the target has it
3599
self.target.get_revision(basis_id)
3600
# Try to get a basis tree - if its a ghost it will hit the
3601
# NoSuchRevision case.
3602
basis_tree = self.source.revision_tree(basis_id)
3603
except (IndexError, errors.NoSuchRevision):
3604
basis_id = _mod_revision.NULL_REVISION
3605
basis_tree = self.source.revision_tree(basis_id)
3606
return basis_id, basis_tree
3609
InterRepository.register_optimiser(InterDifferingSerializer)
3610
InterRepository.register_optimiser(InterSameDataRepository)
3611
InterRepository.register_optimiser(InterWeaveRepo)
3612
InterRepository.register_optimiser(InterKnitRepo)
3613
InterRepository.register_optimiser(InterPackRepo)
3616
class CopyConverter(object):
3617
"""A repository conversion tool which just performs a copy of the content.
3619
This is slow but quite reliable.
3622
def __init__(self, target_format):
3623
"""Create a CopyConverter.
3625
:param target_format: The format the resulting repository should be.
3627
self.target_format = target_format
3629
def convert(self, repo, pb):
3630
"""Perform the conversion of to_convert, giving feedback via pb.
3632
:param to_convert: The disk object to convert.
3633
:param pb: a progress bar to use for progress information.
3638
# this is only useful with metadir layouts - separated repo content.
3639
# trigger an assertion if not such
3640
repo._format.get_format_string()
3641
self.repo_dir = repo.bzrdir
3642
self.step('Moving repository to repository.backup')
3643
self.repo_dir.transport.move('repository', 'repository.backup')
3644
backup_transport = self.repo_dir.transport.clone('repository.backup')
3645
repo._format.check_conversion_target(self.target_format)
3646
self.source_repo = repo._format.open(self.repo_dir,
3648
_override_transport=backup_transport)
3649
self.step('Creating new repository')
3650
converted = self.target_format.initialize(self.repo_dir,
3651
self.source_repo.is_shared())
3652
converted.lock_write()
3654
self.step('Copying content into repository.')
3655
self.source_repo.copy_content_into(converted)
3658
self.step('Deleting old repository content.')
3659
self.repo_dir.transport.delete_tree('repository.backup')
3660
self.pb.note('repository converted')
3662
def step(self, message):
3663
"""Update the pb by a step."""
3665
self.pb.update(message, self.count, self.total)
3677
def _unescaper(match, _map=_unescape_map):
3678
code = match.group(1)
3682
if not code.startswith('#'):
3684
return unichr(int(code[1:])).encode('utf8')
3690
def _unescape_xml(data):
3691
"""Unescape predefined XML entities in a string of data."""
3693
if _unescape_re is None:
3694
_unescape_re = re.compile('\&([^;]*);')
3695
return _unescape_re.sub(_unescaper, data)
3698
class _VersionedFileChecker(object):
3700
def __init__(self, repository, text_key_references=None):
3701
self.repository = repository
3702
self.text_index = self.repository._generate_text_key_index(
3703
text_key_references=text_key_references)
3705
def calculate_file_version_parents(self, text_key):
3706
"""Calculate the correct parents for a file version according to
3709
parent_keys = self.text_index[text_key]
3710
if parent_keys == [_mod_revision.NULL_REVISION]:
3712
return tuple(parent_keys)
3714
def check_file_version_parents(self, texts, progress_bar=None):
3715
"""Check the parents stored in a versioned file are correct.
3717
It also detects file versions that are not referenced by their
3718
corresponding revision's inventory.
3720
:returns: A tuple of (wrong_parents, dangling_file_versions).
3721
wrong_parents is a dict mapping {revision_id: (stored_parents,
3722
correct_parents)} for each revision_id where the stored parents
3723
are not correct. dangling_file_versions is a set of (file_id,
3724
revision_id) tuples for versions that are present in this versioned
3725
file, but not used by the corresponding inventory.
3728
self.file_ids = set([file_id for file_id, _ in
3729
self.text_index.iterkeys()])
3730
# text keys is now grouped by file_id
3731
n_weaves = len(self.file_ids)
3732
files_in_revisions = {}
3733
revisions_of_files = {}
3734
n_versions = len(self.text_index)
3735
progress_bar.update('loading text store', 0, n_versions)
3736
parent_map = self.repository.texts.get_parent_map(self.text_index)
3737
# On unlistable transports this could well be empty/error...
3738
text_keys = self.repository.texts.keys()
3739
unused_keys = frozenset(text_keys) - set(self.text_index)
3740
for num, key in enumerate(self.text_index.iterkeys()):
3741
if progress_bar is not None:
3742
progress_bar.update('checking text graph', num, n_versions)
3743
correct_parents = self.calculate_file_version_parents(key)
3745
knit_parents = parent_map[key]
3746
except errors.RevisionNotPresent:
3749
if correct_parents != knit_parents:
3750
wrong_parents[key] = (knit_parents, correct_parents)
3751
return wrong_parents, unused_keys
3754
def _old_get_graph(repository, revision_id):
3755
"""DO NOT USE. That is all. I'm serious."""
3756
graph = repository.get_graph()
3757
revision_graph = dict(((key, value) for key, value in
3758
graph.iter_ancestry([revision_id]) if value is not None))
3759
return _strip_NULL_ghosts(revision_graph)
3762
def _strip_NULL_ghosts(revision_graph):
3763
"""Also don't use this. more compatibility code for unmigrated clients."""
3764
# Filter ghosts, and null:
3765
if _mod_revision.NULL_REVISION in revision_graph:
3766
del revision_graph[_mod_revision.NULL_REVISION]
3767
for key, parents in revision_graph.items():
3768
revision_graph[key] = tuple(parent for parent in parents if parent
3770
return revision_graph
3773
class StreamSink(object):
3774
"""An object that can insert a stream into a repository.
3776
This interface handles the complexity of reserialising inventories and
3777
revisions from different formats, and allows unidirectional insertion into
3778
stacked repositories without looking for the missing basis parents
3782
def __init__(self, target_repo):
3783
self.target_repo = target_repo
3785
def insert_stream(self, stream, src_format, resume_tokens):
3786
"""Insert a stream's content into the target repository.
3788
:param src_format: a bzr repository format.
3790
:return: a list of resume tokens and an iterable of keys additional
3791
items required before the insertion can be completed.
3793
self.target_repo.lock_write()
3796
self.target_repo.resume_write_group(resume_tokens)
3798
self.target_repo.start_write_group()
3800
# locked_insert_stream performs a commit|suspend.
3801
return self._locked_insert_stream(stream, src_format)
3803
self.target_repo.abort_write_group(suppress_errors=True)
3806
self.target_repo.unlock()
3808
def _locked_insert_stream(self, stream, src_format):
3809
to_serializer = self.target_repo._format._serializer
3810
src_serializer = src_format._serializer
3811
for substream_type, substream in stream:
3812
if substream_type == 'texts':
3813
self.target_repo.texts.insert_record_stream(substream)
3814
elif substream_type == 'inventories':
3815
if src_serializer == to_serializer:
3816
self.target_repo.inventories.insert_record_stream(
3819
self._extract_and_insert_inventories(
3820
substream, src_serializer)
3821
elif substream_type == 'revisions':
3822
# This may fallback to extract-and-insert more often than
3823
# required if the serializers are different only in terms of
3825
if src_serializer == to_serializer:
3826
self.target_repo.revisions.insert_record_stream(
3829
self._extract_and_insert_revisions(substream,
3831
elif substream_type == 'signatures':
3832
self.target_repo.signatures.insert_record_stream(substream)
3834
raise AssertionError('kaboom! %s' % (substream_type,))
3836
missing_keys = set()
3837
for prefix, versioned_file in (
3838
('texts', self.target_repo.texts),
3839
('inventories', self.target_repo.inventories),
3840
('revisions', self.target_repo.revisions),
3841
('signatures', self.target_repo.signatures),
3843
missing_keys.update((prefix,) + key for key in
3844
versioned_file.get_missing_compression_parent_keys())
3845
except NotImplementedError:
3846
# cannot even attempt suspending, and missing would have failed
3847
# during stream insertion.
3848
missing_keys = set()
3851
# suspend the write group and tell the caller what we is
3852
# missing. We know we can suspend or else we would not have
3853
# entered this code path. (All repositories that can handle
3854
# missing keys can handle suspending a write group).
3855
write_group_tokens = self.target_repo.suspend_write_group()
3856
return write_group_tokens, missing_keys
3857
self.target_repo.commit_write_group()
3860
def _extract_and_insert_inventories(self, substream, serializer):
3861
"""Generate a new inventory versionedfile in target, converting data.
3863
The inventory is retrieved from the source, (deserializing it), and
3864
stored in the target (reserializing it in a different format).
3866
for record in substream:
3867
bytes = record.get_bytes_as('fulltext')
3868
revision_id = record.key[0]
3869
inv = serializer.read_inventory_from_string(bytes, revision_id)
3870
parents = [key[0] for key in record.parents]
3871
self.target_repo.add_inventory(revision_id, inv, parents)
3873
def _extract_and_insert_revisions(self, substream, serializer):
3874
for record in substream:
3875
bytes = record.get_bytes_as('fulltext')
3876
revision_id = record.key[0]
3877
rev = serializer.read_revision_from_string(bytes)
3878
if rev.revision_id != revision_id:
3879
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
3880
self.target_repo.add_revision(revision_id, rev)
3883
if self.target_repo._format._fetch_reconcile:
3884
self.target_repo.reconcile()
3887
class StreamSource(object):
3888
"""A source of a stream for fetching between repositories."""
3890
def __init__(self, from_repository, to_format):
3891
"""Create a StreamSource streaming from from_repository."""
3892
self.from_repository = from_repository
3893
self.to_format = to_format
3895
def delta_on_metadata(self):
3896
"""Return True if delta's are permitted on metadata streams.
3898
That is on revisions and signatures.
3900
src_serializer = self.from_repository._format._serializer
3901
target_serializer = self.to_format._serializer
3902
return (self.to_format._fetch_uses_deltas and
3903
src_serializer == target_serializer)
3905
def _fetch_revision_texts(self, revs):
3906
# fetch signatures first and then the revision texts
3907
# may need to be a InterRevisionStore call here.
3908
from_sf = self.from_repository.signatures
3909
# A missing signature is just skipped.
3910
keys = [(rev_id,) for rev_id in revs]
3911
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
3913
self.to_format._fetch_order,
3914
not self.to_format._fetch_uses_deltas))
3915
# If a revision has a delta, this is actually expanded inside the
3916
# insert_record_stream code now, which is an alternate fix for
3918
from_rf = self.from_repository.revisions
3919
revisions = from_rf.get_record_stream(
3921
self.to_format._fetch_order,
3922
not self.delta_on_metadata())
3923
return [('signatures', signatures), ('revisions', revisions)]
3925
def _generate_root_texts(self, revs):
3926
"""This will be called by __fetch between fetching weave texts and
3927
fetching the inventory weave.
3929
Subclasses should override this if they need to generate root texts
3930
after fetching weave texts.
3932
if self._rich_root_upgrade():
3934
return bzrlib.fetch.Inter1and2Helper(
3935
self.from_repository).generate_root_texts(revs)
3939
def get_stream(self, search):
3941
revs = search.get_keys()
3942
graph = self.from_repository.get_graph()
3943
revs = list(graph.iter_topo_order(revs))
3944
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
3946
for knit_kind, file_id, revisions in data_to_fetch:
3947
if knit_kind != phase:
3949
# Make a new progress bar for this phase
3950
if knit_kind == "file":
3951
# Accumulate file texts
3952
text_keys.extend([(file_id, revision) for revision in
3954
elif knit_kind == "inventory":
3955
# Now copy the file texts.
3956
from_texts = self.from_repository.texts
3957
yield ('texts', from_texts.get_record_stream(
3958
text_keys, self.to_format._fetch_order,
3959
not self.to_format._fetch_uses_deltas))
3960
# Cause an error if a text occurs after we have done the
3963
# Before we process the inventory we generate the root
3964
# texts (if necessary) so that the inventories references
3966
for _ in self._generate_root_texts(revs):
3968
# NB: This currently reopens the inventory weave in source;
3969
# using a single stream interface instead would avoid this.
3970
from_weave = self.from_repository.inventories
3971
# we fetch only the referenced inventories because we do not
3972
# know for unselected inventories whether all their required
3973
# texts are present in the other repository - it could be
3975
yield ('inventories', from_weave.get_record_stream(
3976
[(rev_id,) for rev_id in revs],
3977
self.inventory_fetch_order(),
3978
not self.delta_on_metadata()))
3979
elif knit_kind == "signatures":
3980
# Nothing to do here; this will be taken care of when
3981
# _fetch_revision_texts happens.
3983
elif knit_kind == "revisions":
3984
for record in self._fetch_revision_texts(revs):
3987
raise AssertionError("Unknown knit kind %r" % knit_kind)
3989
def get_stream_for_missing_keys(self, missing_keys):
3990
# missing keys can only occur when we are byte copying and not
3991
# translating (because translation means we don't send
3992
# unreconstructable deltas ever).
3994
keys['texts'] = set()
3995
keys['revisions'] = set()
3996
keys['inventories'] = set()
3997
keys['signatures'] = set()
3998
for key in missing_keys:
3999
keys[key[0]].add(key[1:])
4000
if len(keys['revisions']):
4001
# If we allowed copying revisions at this point, we could end up
4002
# copying a revision without copying its required texts: a
4003
# violation of the requirements for repository integrity.
4004
raise AssertionError(
4005
'cannot copy revisions to fill in missing deltas %s' % (
4006
keys['revisions'],))
4007
for substream_kind, keys in keys.iteritems():
4008
vf = getattr(self.from_repository, substream_kind)
4009
# Ask for full texts always so that we don't need more round trips
4010
# after this stream.
4011
stream = vf.get_record_stream(keys,
4012
self.to_format._fetch_order, True)
4013
yield substream_kind, stream
4015
def inventory_fetch_order(self):
4016
if self._rich_root_upgrade():
4017
return 'topological'
4019
return self.to_format._fetch_order
4021
def _rich_root_upgrade(self):
4022
return (not self.from_repository._format.rich_root_data and
4023
self.to_format.rich_root_data)
added
removed
bzrlib/repository.py
