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(), """
42
revision as _mod_revision,
48
from bzrlib.bundle import serializer
49
from bzrlib.revisiontree import RevisionTree
50
from bzrlib.store.versioned import VersionedFileStore
51
from bzrlib.testament import Testament
54
from bzrlib.decorators import needs_read_lock, needs_write_lock, only_raises
55
from bzrlib.inter import InterObject
56
from bzrlib.inventory import (
62
from bzrlib.lock import _RelockDebugMixin
63
from bzrlib import registry
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
# We ignore the 'inventory' returned by add_inventory_by_delta
212
# because self.new_inventory is used to hint to the rest of the
213
# system what code path was taken
214
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
215
basis_id, self._basis_delta, self._new_revision_id,
218
if self.new_inventory.root is None:
219
raise AssertionError('Root entry should be supplied to'
220
' record_entry_contents, as of bzr 0.10.')
221
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
222
self.new_inventory.revision_id = self._new_revision_id
223
self.inv_sha1 = self.repository.add_inventory(
224
self._new_revision_id,
228
return self._new_revision_id
230
def _gen_revision_id(self):
231
"""Return new revision-id."""
232
return generate_ids.gen_revision_id(self._config.username(),
235
def _generate_revision_if_needed(self):
236
"""Create a revision id if None was supplied.
238
If the repository can not support user-specified revision ids
239
they should override this function and raise CannotSetRevisionId
240
if _new_revision_id is not None.
242
:raises: CannotSetRevisionId
244
if self._new_revision_id is None:
245
self._new_revision_id = self._gen_revision_id()
246
self.random_revid = True
248
self.random_revid = False
250
def _heads(self, file_id, revision_ids):
251
"""Calculate the graph heads for revision_ids in the graph of file_id.
253
This can use either a per-file graph or a global revision graph as we
254
have an identity relationship between the two graphs.
256
return self.__heads(revision_ids)
258
def _check_root(self, ie, parent_invs, tree):
259
"""Helper for record_entry_contents.
261
:param ie: An entry being added.
262
:param parent_invs: The inventories of the parent revisions of the
264
:param tree: The tree that is being committed.
266
# In this revision format, root entries have no knit or weave When
267
# serializing out to disk and back in root.revision is always
269
ie.revision = self._new_revision_id
271
def _require_root_change(self, tree):
272
"""Enforce an appropriate root object change.
274
This is called once when record_iter_changes is called, if and only if
275
the root was not in the delta calculated by record_iter_changes.
277
:param tree: The tree which is being committed.
279
# NB: if there are no parents then this method is not called, so no
280
# need to guard on parents having length.
281
entry = entry_factory['directory'](tree.path2id(''), '',
283
entry.revision = self._new_revision_id
284
self._basis_delta.append(('', '', entry.file_id, entry))
286
def _get_delta(self, ie, basis_inv, path):
287
"""Get a delta against the basis inventory for ie."""
288
if ie.file_id not in basis_inv:
290
result = (None, path, ie.file_id, ie)
291
self._basis_delta.append(result)
293
elif ie != basis_inv[ie.file_id]:
295
# TODO: avoid tis id2path call.
296
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
297
self._basis_delta.append(result)
303
def get_basis_delta(self):
304
"""Return the complete inventory delta versus the basis inventory.
306
This has been built up with the calls to record_delete and
307
record_entry_contents. The client must have already called
308
will_record_deletes() to indicate that they will be generating a
311
:return: An inventory delta, suitable for use with apply_delta, or
312
Repository.add_inventory_by_delta, etc.
314
if not self._recording_deletes:
315
raise AssertionError("recording deletes not activated.")
316
return self._basis_delta
318
def record_delete(self, path, file_id):
319
"""Record that a delete occured against a basis tree.
321
This is an optional API - when used it adds items to the basis_delta
322
being accumulated by the commit builder. It cannot be called unless the
323
method will_record_deletes() has been called to inform the builder that
324
a delta is being supplied.
326
:param path: The path of the thing deleted.
327
:param file_id: The file id that was deleted.
329
if not self._recording_deletes:
330
raise AssertionError("recording deletes not activated.")
331
delta = (path, None, file_id, None)
332
self._basis_delta.append(delta)
333
self._any_changes = True
336
def will_record_deletes(self):
337
"""Tell the commit builder that deletes are being notified.
339
This enables the accumulation of an inventory delta; for the resulting
340
commit to be valid, deletes against the basis MUST be recorded via
341
builder.record_delete().
343
self._recording_deletes = True
345
basis_id = self.parents[0]
347
basis_id = _mod_revision.NULL_REVISION
348
self.basis_delta_revision = basis_id
350
def record_entry_contents(self, ie, parent_invs, path, tree,
352
"""Record the content of ie from tree into the commit if needed.
354
Side effect: sets ie.revision when unchanged
356
:param ie: An inventory entry present in the commit.
357
:param parent_invs: The inventories of the parent revisions of the
359
:param path: The path the entry is at in the tree.
360
:param tree: The tree which contains this entry and should be used to
362
:param content_summary: Summary data from the tree about the paths
363
content - stat, length, exec, sha/link target. This is only
364
accessed when the entry has a revision of None - that is when it is
365
a candidate to commit.
366
:return: A tuple (change_delta, version_recorded, fs_hash).
367
change_delta is an inventory_delta change for this entry against
368
the basis tree of the commit, or None if no change occured against
370
version_recorded is True if a new version of the entry has been
371
recorded. For instance, committing a merge where a file was only
372
changed on the other side will return (delta, False).
373
fs_hash is either None, or the hash details for the path (currently
374
a tuple of the contents sha1 and the statvalue returned by
375
tree.get_file_with_stat()).
377
if self.new_inventory.root is None:
378
if ie.parent_id is not None:
379
raise errors.RootMissing()
380
self._check_root(ie, parent_invs, tree)
381
if ie.revision is None:
382
kind = content_summary[0]
384
# ie is carried over from a prior commit
386
# XXX: repository specific check for nested tree support goes here - if
387
# the repo doesn't want nested trees we skip it ?
388
if (kind == 'tree-reference' and
389
not self.repository._format.supports_tree_reference):
390
# mismatch between commit builder logic and repository:
391
# this needs the entry creation pushed down into the builder.
392
raise NotImplementedError('Missing repository subtree support.')
393
self.new_inventory.add(ie)
395
# TODO: slow, take it out of the inner loop.
397
basis_inv = parent_invs[0]
399
basis_inv = Inventory(root_id=None)
401
# ie.revision is always None if the InventoryEntry is considered
402
# for committing. We may record the previous parents revision if the
403
# content is actually unchanged against a sole head.
404
if ie.revision is not None:
405
if not self._versioned_root and path == '':
406
# repositories that do not version the root set the root's
407
# revision to the new commit even when no change occurs (more
408
# specifically, they do not record a revision on the root; and
409
# the rev id is assigned to the root during deserialisation -
410
# this masks when a change may have occurred against the basis.
411
# To match this we always issue a delta, because the revision
412
# of the root will always be changing.
413
if ie.file_id in basis_inv:
414
delta = (basis_inv.id2path(ie.file_id), path,
418
delta = (None, path, ie.file_id, ie)
419
self._basis_delta.append(delta)
420
return delta, False, None
422
# we don't need to commit this, because the caller already
423
# determined that an existing revision of this file is
424
# appropriate. If its not being considered for committing then
425
# it and all its parents to the root must be unaltered so
426
# no-change against the basis.
427
if ie.revision == self._new_revision_id:
428
raise AssertionError("Impossible situation, a skipped "
429
"inventory entry (%r) claims to be modified in this "
430
"commit (%r).", (ie, self._new_revision_id))
431
return None, False, None
432
# XXX: Friction: parent_candidates should return a list not a dict
433
# so that we don't have to walk the inventories again.
434
parent_candiate_entries = ie.parent_candidates(parent_invs)
435
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
437
for inv in parent_invs:
438
if ie.file_id in inv:
439
old_rev = inv[ie.file_id].revision
440
if old_rev in head_set:
441
heads.append(inv[ie.file_id].revision)
442
head_set.remove(inv[ie.file_id].revision)
445
# now we check to see if we need to write a new record to the
447
# We write a new entry unless there is one head to the ancestors, and
448
# the kind-derived content is unchanged.
450
# Cheapest check first: no ancestors, or more the one head in the
451
# ancestors, we write a new node.
455
# There is a single head, look it up for comparison
456
parent_entry = parent_candiate_entries[heads[0]]
457
# if the non-content specific data has changed, we'll be writing a
459
if (parent_entry.parent_id != ie.parent_id or
460
parent_entry.name != ie.name):
462
# now we need to do content specific checks:
464
# if the kind changed the content obviously has
465
if kind != parent_entry.kind:
467
# Stat cache fingerprint feedback for the caller - None as we usually
468
# don't generate one.
471
if content_summary[2] is None:
472
raise ValueError("Files must not have executable = None")
474
# We can't trust a check of the file length because of content
476
if (# if the exec bit has changed we have to store:
477
parent_entry.executable != content_summary[2]):
479
elif parent_entry.text_sha1 == content_summary[3]:
480
# all meta and content is unchanged (using a hash cache
481
# hit to check the sha)
482
ie.revision = parent_entry.revision
483
ie.text_size = parent_entry.text_size
484
ie.text_sha1 = parent_entry.text_sha1
485
ie.executable = parent_entry.executable
486
return self._get_delta(ie, basis_inv, path), False, None
488
# Either there is only a hash change(no hash cache entry,
489
# or same size content change), or there is no change on
491
# Provide the parent's hash to the store layer, so that the
492
# content is unchanged we will not store a new node.
493
nostore_sha = parent_entry.text_sha1
495
# We want to record a new node regardless of the presence or
496
# absence of a content change in the file.
498
ie.executable = content_summary[2]
499
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
501
text = file_obj.read()
505
ie.text_sha1, ie.text_size = self._add_text_to_weave(
506
ie.file_id, text, heads, nostore_sha)
507
# Let the caller know we generated a stat fingerprint.
508
fingerprint = (ie.text_sha1, stat_value)
509
except errors.ExistingContent:
510
# Turns out that the file content was unchanged, and we were
511
# only going to store a new node if it was changed. Carry over
513
ie.revision = parent_entry.revision
514
ie.text_size = parent_entry.text_size
515
ie.text_sha1 = parent_entry.text_sha1
516
ie.executable = parent_entry.executable
517
return self._get_delta(ie, basis_inv, path), False, None
518
elif kind == 'directory':
520
# all data is meta here, nothing specific to directory, so
522
ie.revision = parent_entry.revision
523
return self._get_delta(ie, basis_inv, path), False, None
524
self._add_text_to_weave(ie.file_id, '', heads, None)
525
elif kind == 'symlink':
526
current_link_target = content_summary[3]
528
# symlink target is not generic metadata, check if it has
530
if current_link_target != parent_entry.symlink_target:
533
# unchanged, carry over.
534
ie.revision = parent_entry.revision
535
ie.symlink_target = parent_entry.symlink_target
536
return self._get_delta(ie, basis_inv, path), False, None
537
ie.symlink_target = current_link_target
538
self._add_text_to_weave(ie.file_id, '', heads, None)
539
elif kind == 'tree-reference':
541
if content_summary[3] != parent_entry.reference_revision:
544
# unchanged, carry over.
545
ie.reference_revision = parent_entry.reference_revision
546
ie.revision = parent_entry.revision
547
return self._get_delta(ie, basis_inv, path), False, None
548
ie.reference_revision = content_summary[3]
549
if ie.reference_revision is None:
550
raise AssertionError("invalid content_summary for nested tree: %r"
551
% (content_summary,))
552
self._add_text_to_weave(ie.file_id, '', heads, None)
554
raise NotImplementedError('unknown kind')
555
ie.revision = self._new_revision_id
556
self._any_changes = True
557
return self._get_delta(ie, basis_inv, path), True, fingerprint
559
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
560
_entry_factory=entry_factory):
561
"""Record a new tree via iter_changes.
563
:param tree: The tree to obtain text contents from for changed objects.
564
:param basis_revision_id: The revision id of the tree the iter_changes
565
has been generated against. Currently assumed to be the same
566
as self.parents[0] - if it is not, errors may occur.
567
:param iter_changes: An iter_changes iterator with the changes to apply
568
to basis_revision_id. The iterator must not include any items with
569
a current kind of None - missing items must be either filtered out
570
or errored-on beefore record_iter_changes sees the item.
571
:param _entry_factory: Private method to bind entry_factory locally for
573
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
576
# Create an inventory delta based on deltas between all the parents and
577
# deltas between all the parent inventories. We use inventory delta's
578
# between the inventory objects because iter_changes masks
579
# last-changed-field only changes.
581
# file_id -> change map, change is fileid, paths, changed, versioneds,
582
# parents, names, kinds, executables
584
# {file_id -> revision_id -> inventory entry, for entries in parent
585
# trees that are not parents[0]
589
revtrees = list(self.repository.revision_trees(self.parents))
590
except errors.NoSuchRevision:
591
# one or more ghosts, slow path.
593
for revision_id in self.parents:
595
revtrees.append(self.repository.revision_tree(revision_id))
596
except errors.NoSuchRevision:
598
basis_revision_id = _mod_revision.NULL_REVISION
600
revtrees.append(self.repository.revision_tree(
601
_mod_revision.NULL_REVISION))
602
# The basis inventory from a repository
604
basis_inv = revtrees[0].inventory
606
basis_inv = self.repository.revision_tree(
607
_mod_revision.NULL_REVISION).inventory
608
if len(self.parents) > 0:
609
if basis_revision_id != self.parents[0] and not ghost_basis:
611
"arbitrary basis parents not yet supported with merges")
612
for revtree in revtrees[1:]:
613
for change in revtree.inventory._make_delta(basis_inv):
614
if change[1] is None:
615
# Not present in this parent.
617
if change[2] not in merged_ids:
618
if change[0] is not None:
619
basis_entry = basis_inv[change[2]]
620
merged_ids[change[2]] = [
622
basis_entry.revision,
625
parent_entries[change[2]] = {
627
basis_entry.revision:basis_entry,
629
change[3].revision:change[3],
632
merged_ids[change[2]] = [change[3].revision]
633
parent_entries[change[2]] = {change[3].revision:change[3]}
635
merged_ids[change[2]].append(change[3].revision)
636
parent_entries[change[2]][change[3].revision] = change[3]
639
# Setup the changes from the tree:
640
# changes maps file_id -> (change, [parent revision_ids])
642
for change in iter_changes:
643
# This probably looks up in basis_inv way to much.
644
if change[1][0] is not None:
645
head_candidate = [basis_inv[change[0]].revision]
648
changes[change[0]] = change, merged_ids.get(change[0],
650
unchanged_merged = set(merged_ids) - set(changes)
651
# Extend the changes dict with synthetic changes to record merges of
653
for file_id in unchanged_merged:
654
# Record a merged version of these items that did not change vs the
655
# basis. This can be either identical parallel changes, or a revert
656
# of a specific file after a merge. The recorded content will be
657
# that of the current tree (which is the same as the basis), but
658
# the per-file graph will reflect a merge.
659
# NB:XXX: We are reconstructing path information we had, this
660
# should be preserved instead.
661
# inv delta change: (file_id, (path_in_source, path_in_target),
662
# changed_content, versioned, parent, name, kind,
665
basis_entry = basis_inv[file_id]
666
except errors.NoSuchId:
667
# a change from basis->some_parents but file_id isn't in basis
668
# so was new in the merge, which means it must have changed
669
# from basis -> current, and as it hasn't the add was reverted
670
# by the user. So we discard this change.
674
(basis_inv.id2path(file_id), tree.id2path(file_id)),
676
(basis_entry.parent_id, basis_entry.parent_id),
677
(basis_entry.name, basis_entry.name),
678
(basis_entry.kind, basis_entry.kind),
679
(basis_entry.executable, basis_entry.executable))
680
changes[file_id] = (change, merged_ids[file_id])
681
# changes contains tuples with the change and a set of inventory
682
# candidates for the file.
684
# old_path, new_path, file_id, new_inventory_entry
685
seen_root = False # Is the root in the basis delta?
686
inv_delta = self._basis_delta
687
modified_rev = self._new_revision_id
688
for change, head_candidates in changes.values():
689
if change[3][1]: # versioned in target.
690
# Several things may be happening here:
691
# We may have a fork in the per-file graph
692
# - record a change with the content from tree
693
# We may have a change against < all trees
694
# - carry over the tree that hasn't changed
695
# We may have a change against all trees
696
# - record the change with the content from tree
699
entry = _entry_factory[kind](file_id, change[5][1],
701
head_set = self._heads(change[0], set(head_candidates))
704
for head_candidate in head_candidates:
705
if head_candidate in head_set:
706
heads.append(head_candidate)
707
head_set.remove(head_candidate)
710
# Could be a carry-over situation:
711
parent_entry_revs = parent_entries.get(file_id, None)
712
if parent_entry_revs:
713
parent_entry = parent_entry_revs.get(heads[0], None)
716
if parent_entry is None:
717
# The parent iter_changes was called against is the one
718
# that is the per-file head, so any change is relevant
719
# iter_changes is valid.
720
carry_over_possible = False
722
# could be a carry over situation
723
# A change against the basis may just indicate a merge,
724
# we need to check the content against the source of the
725
# merge to determine if it was changed after the merge
727
if (parent_entry.kind != entry.kind or
728
parent_entry.parent_id != entry.parent_id or
729
parent_entry.name != entry.name):
730
# Metadata common to all entries has changed
731
# against per-file parent
732
carry_over_possible = False
734
carry_over_possible = True
735
# per-type checks for changes against the parent_entry
738
# Cannot be a carry-over situation
739
carry_over_possible = False
740
# Populate the entry in the delta
742
# XXX: There is still a small race here: If someone reverts the content of a file
743
# after iter_changes examines and decides it has changed,
744
# we will unconditionally record a new version even if some
745
# other process reverts it while commit is running (with
746
# the revert happening after iter_changes did it's
749
entry.executable = True
751
entry.executable = False
752
if (carry_over_possible and
753
parent_entry.executable == entry.executable):
754
# Check the file length, content hash after reading
756
nostore_sha = parent_entry.text_sha1
759
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
761
text = file_obj.read()
765
entry.text_sha1, entry.text_size = self._add_text_to_weave(
766
file_id, text, heads, nostore_sha)
767
yield file_id, change[1][1], (entry.text_sha1, stat_value)
768
except errors.ExistingContent:
769
# No content change against a carry_over parent
770
# Perhaps this should also yield a fs hash update?
772
entry.text_size = parent_entry.text_size
773
entry.text_sha1 = parent_entry.text_sha1
774
elif kind == 'symlink':
776
entry.symlink_target = tree.get_symlink_target(file_id)
777
if (carry_over_possible and
778
parent_entry.symlink_target == entry.symlink_target):
781
self._add_text_to_weave(change[0], '', heads, None)
782
elif kind == 'directory':
783
if carry_over_possible:
786
# Nothing to set on the entry.
787
# XXX: split into the Root and nonRoot versions.
788
if change[1][1] != '' or self.repository.supports_rich_root():
789
self._add_text_to_weave(change[0], '', heads, None)
790
elif kind == 'tree-reference':
791
if not self.repository._format.supports_tree_reference:
792
# This isn't quite sane as an error, but we shouldn't
793
# ever see this code path in practice: tree's don't
794
# permit references when the repo doesn't support tree
796
raise errors.UnsupportedOperation(tree.add_reference,
798
reference_revision = tree.get_reference_revision(change[0])
799
entry.reference_revision = reference_revision
800
if (carry_over_possible and
801
parent_entry.reference_revision == reference_revision):
804
self._add_text_to_weave(change[0], '', heads, None)
806
raise AssertionError('unknown kind %r' % kind)
808
entry.revision = modified_rev
810
entry.revision = parent_entry.revision
813
new_path = change[1][1]
814
inv_delta.append((change[1][0], new_path, change[0], entry))
817
self.new_inventory = None
819
# This should perhaps be guarded by a check that the basis we
820
# commit against is the basis for the commit and if not do a delta
822
self._any_changes = True
824
# housekeeping root entry changes do not affect no-change commits.
825
self._require_root_change(tree)
826
self.basis_delta_revision = basis_revision_id
828
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
829
parent_keys = tuple([(file_id, parent) for parent in parents])
830
return self.repository.texts._add_text(
831
(file_id, self._new_revision_id), parent_keys, new_text,
832
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
835
class RootCommitBuilder(CommitBuilder):
836
"""This commitbuilder actually records the root id"""
838
# the root entry gets versioned properly by this builder.
839
_versioned_root = True
841
def _check_root(self, ie, parent_invs, tree):
842
"""Helper for record_entry_contents.
844
:param ie: An entry being added.
845
:param parent_invs: The inventories of the parent revisions of the
847
:param tree: The tree that is being committed.
850
def _require_root_change(self, tree):
851
"""Enforce an appropriate root object change.
853
This is called once when record_iter_changes is called, if and only if
854
the root was not in the delta calculated by record_iter_changes.
856
:param tree: The tree which is being committed.
858
# versioned roots do not change unless the tree found a change.
861
######################################################################
865
class Repository(_RelockDebugMixin):
866
"""Repository holding history for one or more branches.
868
The repository holds and retrieves historical information including
869
revisions and file history. It's normally accessed only by the Branch,
870
which views a particular line of development through that history.
872
The Repository builds on top of some byte storage facilies (the revisions,
873
signatures, inventories, texts and chk_bytes attributes) and a Transport,
874
which respectively provide byte storage and a means to access the (possibly
877
The byte storage facilities are addressed via tuples, which we refer to
878
as 'keys' throughout the code base. Revision_keys, inventory_keys and
879
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
880
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
881
byte string made up of a hash identifier and a hash value.
882
We use this interface because it allows low friction with the underlying
883
code that implements disk indices, network encoding and other parts of
886
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
887
the serialised revisions for the repository. This can be used to obtain
888
revision graph information or to access raw serialised revisions.
889
The result of trying to insert data into the repository via this store
890
is undefined: it should be considered read-only except for implementors
892
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
893
the serialised signatures for the repository. This can be used to
894
obtain access to raw serialised signatures. The result of trying to
895
insert data into the repository via this store is undefined: it should
896
be considered read-only except for implementors of repositories.
897
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
898
the serialised inventories for the repository. This can be used to
899
obtain unserialised inventories. The result of trying to insert data
900
into the repository via this store is undefined: it should be
901
considered read-only except for implementors of repositories.
902
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
903
texts of files and directories for the repository. This can be used to
904
obtain file texts or file graphs. Note that Repository.iter_file_bytes
905
is usually a better interface for accessing file texts.
906
The result of trying to insert data into the repository via this store
907
is undefined: it should be considered read-only except for implementors
909
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
910
any data the repository chooses to store or have indexed by its hash.
911
The result of trying to insert data into the repository via this store
912
is undefined: it should be considered read-only except for implementors
914
:ivar _transport: Transport for file access to repository, typically
915
pointing to .bzr/repository.
918
# What class to use for a CommitBuilder. Often its simpler to change this
919
# in a Repository class subclass rather than to override
920
# get_commit_builder.
921
_commit_builder_class = CommitBuilder
922
# The search regex used by xml based repositories to determine what things
923
# where changed in a single commit.
924
_file_ids_altered_regex = lazy_regex.lazy_compile(
925
r'file_id="(?P<file_id>[^"]+)"'
926
r'.* revision="(?P<revision_id>[^"]+)"'
929
def abort_write_group(self, suppress_errors=False):
930
"""Commit the contents accrued within the current write group.
932
:param suppress_errors: if true, abort_write_group will catch and log
933
unexpected errors that happen during the abort, rather than
934
allowing them to propagate. Defaults to False.
936
:seealso: start_write_group.
938
if self._write_group is not self.get_transaction():
939
# has an unlock or relock occured ?
942
'(suppressed) mismatched lock context and write group. %r, %r',
943
self._write_group, self.get_transaction())
945
raise errors.BzrError(
946
'mismatched lock context and write group. %r, %r' %
947
(self._write_group, self.get_transaction()))
949
self._abort_write_group()
950
except Exception, exc:
951
self._write_group = None
952
if not suppress_errors:
954
mutter('abort_write_group failed')
955
log_exception_quietly()
956
note('bzr: ERROR (ignored): %s', exc)
957
self._write_group = None
959
def _abort_write_group(self):
960
"""Template method for per-repository write group cleanup.
962
This is called during abort before the write group is considered to be
963
finished and should cleanup any internal state accrued during the write
964
group. There is no requirement that data handed to the repository be
965
*not* made available - this is not a rollback - but neither should any
966
attempt be made to ensure that data added is fully commited. Abort is
967
invoked when an error has occured so futher disk or network operations
968
may not be possible or may error and if possible should not be
972
def add_fallback_repository(self, repository):
973
"""Add a repository to use for looking up data not held locally.
975
:param repository: A repository.
977
if not self._format.supports_external_lookups:
978
raise errors.UnstackableRepositoryFormat(self._format, self.base)
980
# This repository will call fallback.unlock() when we transition to
981
# the unlocked state, so we make sure to increment the lock count
982
repository.lock_read()
983
self._check_fallback_repository(repository)
984
self._fallback_repositories.append(repository)
985
self.texts.add_fallback_versioned_files(repository.texts)
986
self.inventories.add_fallback_versioned_files(repository.inventories)
987
self.revisions.add_fallback_versioned_files(repository.revisions)
988
self.signatures.add_fallback_versioned_files(repository.signatures)
989
if self.chk_bytes is not None:
990
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
992
def _check_fallback_repository(self, repository):
993
"""Check that this repository can fallback to repository safely.
995
Raise an error if not.
997
:param repository: A repository to fallback to.
999
return InterRepository._assert_same_model(self, repository)
1001
def add_inventory(self, revision_id, inv, parents):
1002
"""Add the inventory inv to the repository as revision_id.
1004
:param parents: The revision ids of the parents that revision_id
1005
is known to have and are in the repository already.
1007
:returns: The validator(which is a sha1 digest, though what is sha'd is
1008
repository format specific) of the serialized inventory.
1010
if not self.is_in_write_group():
1011
raise AssertionError("%r not in write group" % (self,))
1012
_mod_revision.check_not_reserved_id(revision_id)
1013
if not (inv.revision_id is None or inv.revision_id == revision_id):
1014
raise AssertionError(
1015
"Mismatch between inventory revision"
1016
" id and insertion revid (%r, %r)"
1017
% (inv.revision_id, revision_id))
1018
if inv.root is None:
1019
raise AssertionError()
1020
return self._add_inventory_checked(revision_id, inv, parents)
1022
def _add_inventory_checked(self, revision_id, inv, parents):
1023
"""Add inv to the repository after checking the inputs.
1025
This function can be overridden to allow different inventory styles.
1027
:seealso: add_inventory, for the contract.
1029
inv_lines = self._serialise_inventory_to_lines(inv)
1030
return self._inventory_add_lines(revision_id, parents,
1031
inv_lines, check_content=False)
1033
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1034
parents, basis_inv=None, propagate_caches=False):
1035
"""Add a new inventory expressed as a delta against another revision.
1037
See the inventory developers documentation for the theory behind
1040
:param basis_revision_id: The inventory id the delta was created
1041
against. (This does not have to be a direct parent.)
1042
:param delta: The inventory delta (see Inventory.apply_delta for
1044
:param new_revision_id: The revision id that the inventory is being
1046
:param parents: The revision ids of the parents that revision_id is
1047
known to have and are in the repository already. These are supplied
1048
for repositories that depend on the inventory graph for revision
1049
graph access, as well as for those that pun ancestry with delta
1051
:param basis_inv: The basis inventory if it is already known,
1053
:param propagate_caches: If True, the caches for this inventory are
1054
copied to and updated for the result if possible.
1056
:returns: (validator, new_inv)
1057
The validator(which is a sha1 digest, though what is sha'd is
1058
repository format specific) of the serialized inventory, and the
1059
resulting inventory.
1061
if not self.is_in_write_group():
1062
raise AssertionError("%r not in write group" % (self,))
1063
_mod_revision.check_not_reserved_id(new_revision_id)
1064
basis_tree = self.revision_tree(basis_revision_id)
1065
basis_tree.lock_read()
1067
# Note that this mutates the inventory of basis_tree, which not all
1068
# inventory implementations may support: A better idiom would be to
1069
# return a new inventory, but as there is no revision tree cache in
1070
# repository this is safe for now - RBC 20081013
1071
if basis_inv is None:
1072
basis_inv = basis_tree.inventory
1073
basis_inv.apply_delta(delta)
1074
basis_inv.revision_id = new_revision_id
1075
return (self.add_inventory(new_revision_id, basis_inv, parents),
1080
def _inventory_add_lines(self, revision_id, parents, lines,
1081
check_content=True):
1082
"""Store lines in inv_vf and return the sha1 of the inventory."""
1083
parents = [(parent,) for parent in parents]
1084
result = self.inventories.add_lines((revision_id,), parents, lines,
1085
check_content=check_content)[0]
1086
self.inventories._access.flush()
1089
def add_revision(self, revision_id, rev, inv=None, config=None):
1090
"""Add rev to the revision store as revision_id.
1092
:param revision_id: the revision id to use.
1093
:param rev: The revision object.
1094
:param inv: The inventory for the revision. if None, it will be looked
1095
up in the inventory storer
1096
:param config: If None no digital signature will be created.
1097
If supplied its signature_needed method will be used
1098
to determine if a signature should be made.
1100
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1102
_mod_revision.check_not_reserved_id(revision_id)
1103
if config is not None and config.signature_needed():
1105
inv = self.get_inventory(revision_id)
1106
plaintext = Testament(rev, inv).as_short_text()
1107
self.store_revision_signature(
1108
gpg.GPGStrategy(config), plaintext, revision_id)
1109
# check inventory present
1110
if not self.inventories.get_parent_map([(revision_id,)]):
1112
raise errors.WeaveRevisionNotPresent(revision_id,
1115
# yes, this is not suitable for adding with ghosts.
1116
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1119
key = (revision_id,)
1120
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1121
self._add_revision(rev)
1123
def _add_revision(self, revision):
1124
text = self._serializer.write_revision_to_string(revision)
1125
key = (revision.revision_id,)
1126
parents = tuple((parent,) for parent in revision.parent_ids)
1127
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1129
def all_revision_ids(self):
1130
"""Returns a list of all the revision ids in the repository.
1132
This is conceptually deprecated because code should generally work on
1133
the graph reachable from a particular revision, and ignore any other
1134
revisions that might be present. There is no direct replacement
1137
if 'evil' in debug.debug_flags:
1138
mutter_callsite(2, "all_revision_ids is linear with history.")
1139
return self._all_revision_ids()
1141
def _all_revision_ids(self):
1142
"""Returns a list of all the revision ids in the repository.
1144
These are in as much topological order as the underlying store can
1147
raise NotImplementedError(self._all_revision_ids)
1149
def break_lock(self):
1150
"""Break a lock if one is present from another instance.
1152
Uses the ui factory to ask for confirmation if the lock may be from
1155
self.control_files.break_lock()
1158
def _eliminate_revisions_not_present(self, revision_ids):
1159
"""Check every revision id in revision_ids to see if we have it.
1161
Returns a set of the present revisions.
1164
graph = self.get_graph()
1165
parent_map = graph.get_parent_map(revision_ids)
1166
# The old API returned a list, should this actually be a set?
1167
return parent_map.keys()
1169
def _check_inventories(self, checker):
1170
"""Check the inventories found from the revision scan.
1172
This is responsible for verifying the sha1 of inventories and
1173
creating a pending_keys set that covers data referenced by inventories.
1175
bar = ui.ui_factory.nested_progress_bar()
1177
self._do_check_inventories(checker, bar)
1181
def _do_check_inventories(self, checker, bar):
1182
"""Helper for _check_inventories."""
1184
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1185
kinds = ['chk_bytes', 'texts']
1186
count = len(checker.pending_keys)
1187
bar.update("inventories", 0, 2)
1188
current_keys = checker.pending_keys
1189
checker.pending_keys = {}
1190
# Accumulate current checks.
1191
for key in current_keys:
1192
if key[0] != 'inventories' and key[0] not in kinds:
1193
checker._report_items.append('unknown key type %r' % (key,))
1194
keys[key[0]].add(key[1:])
1195
if keys['inventories']:
1196
# NB: output order *should* be roughly sorted - topo or
1197
# inverse topo depending on repository - either way decent
1198
# to just delta against. However, pre-CHK formats didn't
1199
# try to optimise inventory layout on disk. As such the
1200
# pre-CHK code path does not use inventory deltas.
1202
for record in self.inventories.check(keys=keys['inventories']):
1203
if record.storage_kind == 'absent':
1204
checker._report_items.append(
1205
'Missing inventory {%s}' % (record.key,))
1207
last_object = self._check_record('inventories', record,
1208
checker, last_object,
1209
current_keys[('inventories',) + record.key])
1210
del keys['inventories']
1213
bar.update("texts", 1)
1214
while (checker.pending_keys or keys['chk_bytes']
1216
# Something to check.
1217
current_keys = checker.pending_keys
1218
checker.pending_keys = {}
1219
# Accumulate current checks.
1220
for key in current_keys:
1221
if key[0] not in kinds:
1222
checker._report_items.append('unknown key type %r' % (key,))
1223
keys[key[0]].add(key[1:])
1224
# Check the outermost kind only - inventories || chk_bytes || texts
1228
for record in getattr(self, kind).check(keys=keys[kind]):
1229
if record.storage_kind == 'absent':
1230
checker._report_items.append(
1231
'Missing %s {%s}' % (kind, record.key,))
1233
last_object = self._check_record(kind, record,
1234
checker, last_object, current_keys[(kind,) + record.key])
1238
def _check_record(self, kind, record, checker, last_object, item_data):
1239
"""Check a single text from this repository."""
1240
if kind == 'inventories':
1241
rev_id = record.key[0]
1242
inv = self.deserialise_inventory(rev_id,
1243
record.get_bytes_as('fulltext'))
1244
if last_object is not None:
1245
delta = inv._make_delta(last_object)
1246
for old_path, path, file_id, ie in delta:
1249
ie.check(checker, rev_id, inv)
1251
for path, ie in inv.iter_entries():
1252
ie.check(checker, rev_id, inv)
1253
if self._format.fast_deltas:
1255
elif kind == 'chk_bytes':
1256
# No code written to check chk_bytes for this repo format.
1257
checker._report_items.append(
1258
'unsupported key type chk_bytes for %s' % (record.key,))
1259
elif kind == 'texts':
1260
self._check_text(record, checker, item_data)
1262
checker._report_items.append(
1263
'unknown key type %s for %s' % (kind, record.key))
1265
def _check_text(self, record, checker, item_data):
1266
"""Check a single text."""
1267
# Check it is extractable.
1268
# TODO: check length.
1269
if record.storage_kind == 'chunked':
1270
chunks = record.get_bytes_as(record.storage_kind)
1271
sha1 = osutils.sha_strings(chunks)
1272
length = sum(map(len, chunks))
1274
content = record.get_bytes_as('fulltext')
1275
sha1 = osutils.sha_string(content)
1276
length = len(content)
1277
if item_data and sha1 != item_data[1]:
1278
checker._report_items.append(
1279
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1280
(record.key, sha1, item_data[1], item_data[2]))
1283
def create(a_bzrdir):
1284
"""Construct the current default format repository in a_bzrdir."""
1285
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1287
def __init__(self, _format, a_bzrdir, control_files):
1288
"""instantiate a Repository.
1290
:param _format: The format of the repository on disk.
1291
:param a_bzrdir: The BzrDir of the repository.
1293
In the future we will have a single api for all stores for
1294
getting file texts, inventories and revisions, then
1295
this construct will accept instances of those things.
1297
super(Repository, self).__init__()
1298
self._format = _format
1299
# the following are part of the public API for Repository:
1300
self.bzrdir = a_bzrdir
1301
self.control_files = control_files
1302
self._transport = control_files._transport
1303
self.base = self._transport.base
1305
self._reconcile_does_inventory_gc = True
1306
self._reconcile_fixes_text_parents = False
1307
self._reconcile_backsup_inventory = True
1308
self._write_group = None
1309
# Additional places to query for data.
1310
self._fallback_repositories = []
1311
# An InventoryEntry cache, used during deserialization
1312
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1315
if self._fallback_repositories:
1316
return '%s(%r, fallback_repositories=%r)' % (
1317
self.__class__.__name__,
1319
self._fallback_repositories)
1321
return '%s(%r)' % (self.__class__.__name__,
1324
def _has_same_fallbacks(self, other_repo):
1325
"""Returns true if the repositories have the same fallbacks."""
1326
my_fb = self._fallback_repositories
1327
other_fb = other_repo._fallback_repositories
1328
if len(my_fb) != len(other_fb):
1330
for f, g in zip(my_fb, other_fb):
1331
if not f.has_same_location(g):
1335
def has_same_location(self, other):
1336
"""Returns a boolean indicating if this repository is at the same
1337
location as another repository.
1339
This might return False even when two repository objects are accessing
1340
the same physical repository via different URLs.
1342
if self.__class__ is not other.__class__:
1344
return (self._transport.base == other._transport.base)
1346
def is_in_write_group(self):
1347
"""Return True if there is an open write group.
1349
:seealso: start_write_group.
1351
return self._write_group is not None
1353
def is_locked(self):
1354
return self.control_files.is_locked()
1356
def is_write_locked(self):
1357
"""Return True if this object is write locked."""
1358
return self.is_locked() and self.control_files._lock_mode == 'w'
1360
def lock_write(self, token=None):
1361
"""Lock this repository for writing.
1363
This causes caching within the repository obejct to start accumlating
1364
data during reads, and allows a 'write_group' to be obtained. Write
1365
groups must be used for actual data insertion.
1367
:param token: if this is already locked, then lock_write will fail
1368
unless the token matches the existing lock.
1369
:returns: a token if this instance supports tokens, otherwise None.
1370
:raises TokenLockingNotSupported: when a token is given but this
1371
instance doesn't support using token locks.
1372
:raises MismatchedToken: if the specified token doesn't match the token
1373
of the existing lock.
1374
:seealso: start_write_group.
1376
A token should be passed in if you know that you have locked the object
1377
some other way, and need to synchronise this object's state with that
1380
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1382
locked = self.is_locked()
1383
result = self.control_files.lock_write(token=token)
1385
self._warn_if_deprecated()
1386
self._note_lock('w')
1387
for repo in self._fallback_repositories:
1388
# Writes don't affect fallback repos
1390
self._refresh_data()
1393
def lock_read(self):
1394
locked = self.is_locked()
1395
self.control_files.lock_read()
1397
self._warn_if_deprecated()
1398
self._note_lock('r')
1399
for repo in self._fallback_repositories:
1401
self._refresh_data()
1403
def get_physical_lock_status(self):
1404
return self.control_files.get_physical_lock_status()
1406
def leave_lock_in_place(self):
1407
"""Tell this repository not to release the physical lock when this
1410
If lock_write doesn't return a token, then this method is not supported.
1412
self.control_files.leave_in_place()
1414
def dont_leave_lock_in_place(self):
1415
"""Tell this repository to release the physical lock when this
1416
object is unlocked, even if it didn't originally acquire it.
1418
If lock_write doesn't return a token, then this method is not supported.
1420
self.control_files.dont_leave_in_place()
1423
def gather_stats(self, revid=None, committers=None):
1424
"""Gather statistics from a revision id.
1426
:param revid: The revision id to gather statistics from, if None, then
1427
no revision specific statistics are gathered.
1428
:param committers: Optional parameter controlling whether to grab
1429
a count of committers from the revision specific statistics.
1430
:return: A dictionary of statistics. Currently this contains:
1431
committers: The number of committers if requested.
1432
firstrev: A tuple with timestamp, timezone for the penultimate left
1433
most ancestor of revid, if revid is not the NULL_REVISION.
1434
latestrev: A tuple with timestamp, timezone for revid, if revid is
1435
not the NULL_REVISION.
1436
revisions: The total revision count in the repository.
1437
size: An estimate disk size of the repository in bytes.
1440
if revid and committers:
1441
result['committers'] = 0
1442
if revid and revid != _mod_revision.NULL_REVISION:
1444
all_committers = set()
1445
revisions = self.get_ancestry(revid)
1446
# pop the leading None
1448
first_revision = None
1450
# ignore the revisions in the middle - just grab first and last
1451
revisions = revisions[0], revisions[-1]
1452
for revision in self.get_revisions(revisions):
1453
if not first_revision:
1454
first_revision = revision
1456
all_committers.add(revision.committer)
1457
last_revision = revision
1459
result['committers'] = len(all_committers)
1460
result['firstrev'] = (first_revision.timestamp,
1461
first_revision.timezone)
1462
result['latestrev'] = (last_revision.timestamp,
1463
last_revision.timezone)
1465
# now gather global repository information
1466
# XXX: This is available for many repos regardless of listability.
1467
if self.bzrdir.root_transport.listable():
1468
# XXX: do we want to __define len__() ?
1469
# Maybe the versionedfiles object should provide a different
1470
# method to get the number of keys.
1471
result['revisions'] = len(self.revisions.keys())
1472
# result['size'] = t
1475
def find_branches(self, using=False):
1476
"""Find branches underneath this repository.
1478
This will include branches inside other branches.
1480
:param using: If True, list only branches using this repository.
1482
if using and not self.is_shared():
1484
return [self.bzrdir.open_branch()]
1485
except errors.NotBranchError:
1487
class Evaluator(object):
1490
self.first_call = True
1492
def __call__(self, bzrdir):
1493
# On the first call, the parameter is always the bzrdir
1494
# containing the current repo.
1495
if not self.first_call:
1497
repository = bzrdir.open_repository()
1498
except errors.NoRepositoryPresent:
1501
return False, (None, repository)
1502
self.first_call = False
1504
value = (bzrdir.open_branch(), None)
1505
except errors.NotBranchError:
1506
value = (None, None)
1510
for branch, repository in bzrdir.BzrDir.find_bzrdirs(
1511
self.bzrdir.root_transport, evaluate=Evaluator()):
1512
if branch is not None:
1513
branches.append(branch)
1514
if not using and repository is not None:
1515
branches.extend(repository.find_branches())
1519
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
1520
"""Return the revision ids that other has that this does not.
1522
These are returned in topological order.
1524
revision_id: only return revision ids included by revision_id.
1526
return InterRepository.get(other, self).search_missing_revision_ids(
1527
revision_id, find_ghosts)
1531
"""Open the repository rooted at base.
1533
For instance, if the repository is at URL/.bzr/repository,
1534
Repository.open(URL) -> a Repository instance.
1536
control = bzrdir.BzrDir.open(base)
1537
return control.open_repository()
1539
def copy_content_into(self, destination, revision_id=None):
1540
"""Make a complete copy of the content in self into destination.
1542
This is a destructive operation! Do not use it on existing
1545
return InterRepository.get(self, destination).copy_content(revision_id)
1547
def commit_write_group(self):
1548
"""Commit the contents accrued within the current write group.
1550
:seealso: start_write_group.
1552
:return: it may return an opaque hint that can be passed to 'pack'.
1554
if self._write_group is not self.get_transaction():
1555
# has an unlock or relock occured ?
1556
raise errors.BzrError('mismatched lock context %r and '
1558
(self.get_transaction(), self._write_group))
1559
result = self._commit_write_group()
1560
self._write_group = None
1563
def _commit_write_group(self):
1564
"""Template method for per-repository write group cleanup.
1566
This is called before the write group is considered to be
1567
finished and should ensure that all data handed to the repository
1568
for writing during the write group is safely committed (to the
1569
extent possible considering file system caching etc).
1572
def suspend_write_group(self):
1573
raise errors.UnsuspendableWriteGroup(self)
1575
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1576
"""Return the keys of missing inventory parents for revisions added in
1579
A revision is not complete if the inventory delta for that revision
1580
cannot be calculated. Therefore if the parent inventories of a
1581
revision are not present, the revision is incomplete, and e.g. cannot
1582
be streamed by a smart server. This method finds missing inventory
1583
parents for revisions added in this write group.
1585
if not self._format.supports_external_lookups:
1586
# This is only an issue for stacked repositories
1588
if not self.is_in_write_group():
1589
raise AssertionError('not in a write group')
1591
# XXX: We assume that every added revision already has its
1592
# corresponding inventory, so we only check for parent inventories that
1593
# might be missing, rather than all inventories.
1594
parents = set(self.revisions._index.get_missing_parents())
1595
parents.discard(_mod_revision.NULL_REVISION)
1596
unstacked_inventories = self.inventories._index
1597
present_inventories = unstacked_inventories.get_parent_map(
1598
key[-1:] for key in parents)
1599
parents.difference_update(present_inventories)
1600
if len(parents) == 0:
1601
# No missing parent inventories.
1603
if not check_for_missing_texts:
1604
return set(('inventories', rev_id) for (rev_id,) in parents)
1605
# Ok, now we have a list of missing inventories. But these only matter
1606
# if the inventories that reference them are missing some texts they
1607
# appear to introduce.
1608
# XXX: Texts referenced by all added inventories need to be present,
1609
# but at the moment we're only checking for texts referenced by
1610
# inventories at the graph's edge.
1611
key_deps = self.revisions._index._key_dependencies
1612
key_deps.satisfy_refs_for_keys(present_inventories)
1613
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1614
file_ids = self.fileids_altered_by_revision_ids(referrers)
1615
missing_texts = set()
1616
for file_id, version_ids in file_ids.iteritems():
1617
missing_texts.update(
1618
(file_id, version_id) for version_id in version_ids)
1619
present_texts = self.texts.get_parent_map(missing_texts)
1620
missing_texts.difference_update(present_texts)
1621
if not missing_texts:
1622
# No texts are missing, so all revisions and their deltas are
1625
# Alternatively the text versions could be returned as the missing
1626
# keys, but this is likely to be less data.
1627
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1630
def refresh_data(self):
1631
"""Re-read any data needed to to synchronise with disk.
1633
This method is intended to be called after another repository instance
1634
(such as one used by a smart server) has inserted data into the
1635
repository. It may not be called during a write group, but may be
1636
called at any other time.
1638
if self.is_in_write_group():
1639
raise errors.InternalBzrError(
1640
"May not refresh_data while in a write group.")
1641
self._refresh_data()
1643
def resume_write_group(self, tokens):
1644
if not self.is_write_locked():
1645
raise errors.NotWriteLocked(self)
1646
if self._write_group:
1647
raise errors.BzrError('already in a write group')
1648
self._resume_write_group(tokens)
1649
# so we can detect unlock/relock - the write group is now entered.
1650
self._write_group = self.get_transaction()
1652
def _resume_write_group(self, tokens):
1653
raise errors.UnsuspendableWriteGroup(self)
1655
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1657
"""Fetch the content required to construct revision_id from source.
1659
If revision_id is None and fetch_spec is None, then all content is
1662
fetch() may not be used when the repository is in a write group -
1663
either finish the current write group before using fetch, or use
1664
fetch before starting the write group.
1666
:param find_ghosts: Find and copy revisions in the source that are
1667
ghosts in the target (and not reachable directly by walking out to
1668
the first-present revision in target from revision_id).
1669
:param revision_id: If specified, all the content needed for this
1670
revision ID will be copied to the target. Fetch will determine for
1671
itself which content needs to be copied.
1672
:param fetch_spec: If specified, a SearchResult or
1673
PendingAncestryResult that describes which revisions to copy. This
1674
allows copying multiple heads at once. Mutually exclusive with
1677
if fetch_spec is not None and revision_id is not None:
1678
raise AssertionError(
1679
"fetch_spec and revision_id are mutually exclusive.")
1680
if self.is_in_write_group():
1681
raise errors.InternalBzrError(
1682
"May not fetch while in a write group.")
1683
# fast path same-url fetch operations
1684
# TODO: lift out to somewhere common with RemoteRepository
1685
# <https://bugs.edge.launchpad.net/bzr/+bug/401646>
1686
if (self.has_same_location(source)
1687
and fetch_spec is None
1688
and self._has_same_fallbacks(source)):
1689
# check that last_revision is in 'from' and then return a
1691
if (revision_id is not None and
1692
not _mod_revision.is_null(revision_id)):
1693
self.get_revision(revision_id)
1695
# if there is no specific appropriate InterRepository, this will get
1696
# the InterRepository base class, which raises an
1697
# IncompatibleRepositories when asked to fetch.
1698
inter = InterRepository.get(source, self)
1699
return inter.fetch(revision_id=revision_id, pb=pb,
1700
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1702
def create_bundle(self, target, base, fileobj, format=None):
1703
return serializer.write_bundle(self, target, base, fileobj, format)
1705
def get_commit_builder(self, branch, parents, config, timestamp=None,
1706
timezone=None, committer=None, revprops=None,
1708
"""Obtain a CommitBuilder for this repository.
1710
:param branch: Branch to commit to.
1711
:param parents: Revision ids of the parents of the new revision.
1712
:param config: Configuration to use.
1713
:param timestamp: Optional timestamp recorded for commit.
1714
:param timezone: Optional timezone for timestamp.
1715
:param committer: Optional committer to set for commit.
1716
:param revprops: Optional dictionary of revision properties.
1717
:param revision_id: Optional revision id.
1719
if self._fallback_repositories:
1720
raise errors.BzrError("Cannot commit from a lightweight checkout "
1721
"to a stacked branch. See "
1722
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1723
result = self._commit_builder_class(self, parents, config,
1724
timestamp, timezone, committer, revprops, revision_id)
1725
self.start_write_group()
1728
@only_raises(errors.LockNotHeld, errors.LockBroken)
1730
if (self.control_files._lock_count == 1 and
1731
self.control_files._lock_mode == 'w'):
1732
if self._write_group is not None:
1733
self.abort_write_group()
1734
self.control_files.unlock()
1735
raise errors.BzrError(
1736
'Must end write groups before releasing write locks.')
1737
self.control_files.unlock()
1738
if self.control_files._lock_count == 0:
1739
self._inventory_entry_cache.clear()
1740
for repo in self._fallback_repositories:
1744
def clone(self, a_bzrdir, revision_id=None):
1745
"""Clone this repository into a_bzrdir using the current format.
1747
Currently no check is made that the format of this repository and
1748
the bzrdir format are compatible. FIXME RBC 20060201.
1750
:return: The newly created destination repository.
1752
# TODO: deprecate after 0.16; cloning this with all its settings is
1753
# probably not very useful -- mbp 20070423
1754
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1755
self.copy_content_into(dest_repo, revision_id)
1758
def start_write_group(self):
1759
"""Start a write group in the repository.
1761
Write groups are used by repositories which do not have a 1:1 mapping
1762
between file ids and backend store to manage the insertion of data from
1763
both fetch and commit operations.
1765
A write lock is required around the start_write_group/commit_write_group
1766
for the support of lock-requiring repository formats.
1768
One can only insert data into a repository inside a write group.
1772
if not self.is_write_locked():
1773
raise errors.NotWriteLocked(self)
1774
if self._write_group:
1775
raise errors.BzrError('already in a write group')
1776
self._start_write_group()
1777
# so we can detect unlock/relock - the write group is now entered.
1778
self._write_group = self.get_transaction()
1780
def _start_write_group(self):
1781
"""Template method for per-repository write group startup.
1783
This is called before the write group is considered to be
1788
def sprout(self, to_bzrdir, revision_id=None):
1789
"""Create a descendent repository for new development.
1791
Unlike clone, this does not copy the settings of the repository.
1793
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1794
dest_repo.fetch(self, revision_id=revision_id)
1797
def _create_sprouting_repo(self, a_bzrdir, shared):
1798
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1799
# use target default format.
1800
dest_repo = a_bzrdir.create_repository()
1802
# Most control formats need the repository to be specifically
1803
# created, but on some old all-in-one formats it's not needed
1805
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1806
except errors.UninitializableFormat:
1807
dest_repo = a_bzrdir.open_repository()
1810
def _get_sink(self):
1811
"""Return a sink for streaming into this repository."""
1812
return StreamSink(self)
1814
def _get_source(self, to_format):
1815
"""Return a source for streaming from this repository."""
1816
return StreamSource(self, to_format)
1819
def has_revision(self, revision_id):
1820
"""True if this repository has a copy of the revision."""
1821
return revision_id in self.has_revisions((revision_id,))
1824
def has_revisions(self, revision_ids):
1825
"""Probe to find out the presence of multiple revisions.
1827
:param revision_ids: An iterable of revision_ids.
1828
:return: A set of the revision_ids that were present.
1830
parent_map = self.revisions.get_parent_map(
1831
[(rev_id,) for rev_id in revision_ids])
1833
if _mod_revision.NULL_REVISION in revision_ids:
1834
result.add(_mod_revision.NULL_REVISION)
1835
result.update([key[0] for key in parent_map])
1839
def get_revision(self, revision_id):
1840
"""Return the Revision object for a named revision."""
1841
return self.get_revisions([revision_id])[0]
1844
def get_revision_reconcile(self, revision_id):
1845
"""'reconcile' helper routine that allows access to a revision always.
1847
This variant of get_revision does not cross check the weave graph
1848
against the revision one as get_revision does: but it should only
1849
be used by reconcile, or reconcile-alike commands that are correcting
1850
or testing the revision graph.
1852
return self._get_revisions([revision_id])[0]
1855
def get_revisions(self, revision_ids):
1856
"""Get many revisions at once.
1858
Repositories that need to check data on every revision read should
1859
subclass this method.
1861
return self._get_revisions(revision_ids)
1864
def _get_revisions(self, revision_ids):
1865
"""Core work logic to get many revisions without sanity checks."""
1867
for revid, rev in self._iter_revisions(revision_ids):
1869
raise errors.NoSuchRevision(self, revid)
1871
return [revs[revid] for revid in revision_ids]
1873
def _iter_revisions(self, revision_ids):
1874
"""Iterate over revision objects.
1876
:param revision_ids: An iterable of revisions to examine. None may be
1877
passed to request all revisions known to the repository. Note that
1878
not all repositories can find unreferenced revisions; for those
1879
repositories only referenced ones will be returned.
1880
:return: An iterator of (revid, revision) tuples. Absent revisions (
1881
those asked for but not available) are returned as (revid, None).
1883
if revision_ids is None:
1884
revision_ids = self.all_revision_ids()
1886
for rev_id in revision_ids:
1887
if not rev_id or not isinstance(rev_id, basestring):
1888
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1889
keys = [(key,) for key in revision_ids]
1890
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1891
for record in stream:
1892
revid = record.key[0]
1893
if record.storage_kind == 'absent':
1896
text = record.get_bytes_as('fulltext')
1897
rev = self._serializer.read_revision_from_string(text)
1901
def get_revision_xml(self, revision_id):
1902
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1903
# would have already do it.
1904
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1905
# TODO: this can't just be replaced by:
1906
# return self._serializer.write_revision_to_string(
1907
# self.get_revision(revision_id))
1908
# as cStringIO preservers the encoding unlike write_revision_to_string
1909
# or some other call down the path.
1910
rev = self.get_revision(revision_id)
1911
rev_tmp = cStringIO.StringIO()
1912
# the current serializer..
1913
self._serializer.write_revision(rev, rev_tmp)
1915
return rev_tmp.getvalue()
1917
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1918
"""Produce a generator of revision deltas.
1920
Note that the input is a sequence of REVISIONS, not revision_ids.
1921
Trees will be held in memory until the generator exits.
1922
Each delta is relative to the revision's lefthand predecessor.
1924
:param specific_fileids: if not None, the result is filtered
1925
so that only those file-ids, their parents and their
1926
children are included.
1928
# Get the revision-ids of interest
1929
required_trees = set()
1930
for revision in revisions:
1931
required_trees.add(revision.revision_id)
1932
required_trees.update(revision.parent_ids[:1])
1934
# Get the matching filtered trees. Note that it's more
1935
# efficient to pass filtered trees to changes_from() rather
1936
# than doing the filtering afterwards. changes_from() could
1937
# arguably do the filtering itself but it's path-based, not
1938
# file-id based, so filtering before or afterwards is
1940
if specific_fileids is None:
1941
trees = dict((t.get_revision_id(), t) for
1942
t in self.revision_trees(required_trees))
1944
trees = dict((t.get_revision_id(), t) for
1945
t in self._filtered_revision_trees(required_trees,
1948
# Calculate the deltas
1949
for revision in revisions:
1950
if not revision.parent_ids:
1951
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1953
old_tree = trees[revision.parent_ids[0]]
1954
yield trees[revision.revision_id].changes_from(old_tree)
1957
def get_revision_delta(self, revision_id, specific_fileids=None):
1958
"""Return the delta for one revision.
1960
The delta is relative to the left-hand predecessor of the
1963
:param specific_fileids: if not None, the result is filtered
1964
so that only those file-ids, their parents and their
1965
children are included.
1967
r = self.get_revision(revision_id)
1968
return list(self.get_deltas_for_revisions([r],
1969
specific_fileids=specific_fileids))[0]
1972
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1973
signature = gpg_strategy.sign(plaintext)
1974
self.add_signature_text(revision_id, signature)
1977
def add_signature_text(self, revision_id, signature):
1978
self.signatures.add_lines((revision_id,), (),
1979
osutils.split_lines(signature))
1981
def find_text_key_references(self):
1982
"""Find the text key references within the repository.
1984
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1985
to whether they were referred to by the inventory of the
1986
revision_id that they contain. The inventory texts from all present
1987
revision ids are assessed to generate this report.
1989
revision_keys = self.revisions.keys()
1990
w = self.inventories
1991
pb = ui.ui_factory.nested_progress_bar()
1993
return self._find_text_key_references_from_xml_inventory_lines(
1994
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1998
def _find_text_key_references_from_xml_inventory_lines(self,
2000
"""Core routine for extracting references to texts from inventories.
2002
This performs the translation of xml lines to revision ids.
2004
:param line_iterator: An iterator of lines, origin_version_id
2005
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2006
to whether they were referred to by the inventory of the
2007
revision_id that they contain. Note that if that revision_id was
2008
not part of the line_iterator's output then False will be given -
2009
even though it may actually refer to that key.
2011
if not self._serializer.support_altered_by_hack:
2012
raise AssertionError(
2013
"_find_text_key_references_from_xml_inventory_lines only "
2014
"supported for branches which store inventory as unnested xml"
2015
", not on %r" % self)
2018
# this code needs to read every new line in every inventory for the
2019
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2020
# not present in one of those inventories is unnecessary but not
2021
# harmful because we are filtering by the revision id marker in the
2022
# inventory lines : we only select file ids altered in one of those
2023
# revisions. We don't need to see all lines in the inventory because
2024
# only those added in an inventory in rev X can contain a revision=X
2026
unescape_revid_cache = {}
2027
unescape_fileid_cache = {}
2029
# jam 20061218 In a big fetch, this handles hundreds of thousands
2030
# of lines, so it has had a lot of inlining and optimizing done.
2031
# Sorry that it is a little bit messy.
2032
# Move several functions to be local variables, since this is a long
2034
search = self._file_ids_altered_regex.search
2035
unescape = _unescape_xml
2036
setdefault = result.setdefault
2037
for line, line_key in line_iterator:
2038
match = search(line)
2041
# One call to match.group() returning multiple items is quite a
2042
# bit faster than 2 calls to match.group() each returning 1
2043
file_id, revision_id = match.group('file_id', 'revision_id')
2045
# Inlining the cache lookups helps a lot when you make 170,000
2046
# lines and 350k ids, versus 8.4 unique ids.
2047
# Using a cache helps in 2 ways:
2048
# 1) Avoids unnecessary decoding calls
2049
# 2) Re-uses cached strings, which helps in future set and
2051
# (2) is enough that removing encoding entirely along with
2052
# the cache (so we are using plain strings) results in no
2053
# performance improvement.
2055
revision_id = unescape_revid_cache[revision_id]
2057
unescaped = unescape(revision_id)
2058
unescape_revid_cache[revision_id] = unescaped
2059
revision_id = unescaped
2061
# Note that unconditionally unescaping means that we deserialise
2062
# every fileid, which for general 'pull' is not great, but we don't
2063
# really want to have some many fulltexts that this matters anyway.
2066
file_id = unescape_fileid_cache[file_id]
2068
unescaped = unescape(file_id)
2069
unescape_fileid_cache[file_id] = unescaped
2072
key = (file_id, revision_id)
2073
setdefault(key, False)
2074
if revision_id == line_key[-1]:
2078
def _inventory_xml_lines_for_keys(self, keys):
2079
"""Get a line iterator of the sort needed for findind references.
2081
Not relevant for non-xml inventory repositories.
2083
Ghosts in revision_keys are ignored.
2085
:param revision_keys: The revision keys for the inventories to inspect.
2086
:return: An iterator over (inventory line, revid) for the fulltexts of
2087
all of the xml inventories specified by revision_keys.
2089
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2090
for record in stream:
2091
if record.storage_kind != 'absent':
2092
chunks = record.get_bytes_as('chunked')
2093
revid = record.key[-1]
2094
lines = osutils.chunks_to_lines(chunks)
2098
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2100
"""Helper routine for fileids_altered_by_revision_ids.
2102
This performs the translation of xml lines to revision ids.
2104
:param line_iterator: An iterator of lines, origin_version_id
2105
:param revision_keys: The revision ids to filter for. This should be a
2106
set or other type which supports efficient __contains__ lookups, as
2107
the revision key from each parsed line will be looked up in the
2108
revision_keys filter.
2109
:return: a dictionary mapping altered file-ids to an iterable of
2110
revision_ids. Each altered file-ids has the exact revision_ids that
2111
altered it listed explicitly.
2113
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2114
line_iterator).iterkeys())
2115
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2116
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2117
self._inventory_xml_lines_for_keys(parent_keys)))
2118
new_keys = seen - parent_seen
2120
setdefault = result.setdefault
2121
for key in new_keys:
2122
setdefault(key[0], set()).add(key[-1])
2125
def _find_parent_ids_of_revisions(self, revision_ids):
2126
"""Find all parent ids that are mentioned in the revision graph.
2128
:return: set of revisions that are parents of revision_ids which are
2129
not part of revision_ids themselves
2131
parent_map = self.get_parent_map(revision_ids)
2133
map(parent_ids.update, parent_map.itervalues())
2134
parent_ids.difference_update(revision_ids)
2135
parent_ids.discard(_mod_revision.NULL_REVISION)
2138
def _find_parent_keys_of_revisions(self, revision_keys):
2139
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2141
:param revision_keys: An iterable of revision_keys.
2142
:return: The parents of all revision_keys that are not already in
2145
parent_map = self.revisions.get_parent_map(revision_keys)
2147
map(parent_keys.update, parent_map.itervalues())
2148
parent_keys.difference_update(revision_keys)
2149
parent_keys.discard(_mod_revision.NULL_REVISION)
2152
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2153
"""Find the file ids and versions affected by revisions.
2155
:param revisions: an iterable containing revision ids.
2156
:param _inv_weave: The inventory weave from this repository or None.
2157
If None, the inventory weave will be opened automatically.
2158
:return: a dictionary mapping altered file-ids to an iterable of
2159
revision_ids. Each altered file-ids has the exact revision_ids that
2160
altered it listed explicitly.
2162
selected_keys = set((revid,) for revid in revision_ids)
2163
w = _inv_weave or self.inventories
2164
pb = ui.ui_factory.nested_progress_bar()
2166
return self._find_file_ids_from_xml_inventory_lines(
2167
w.iter_lines_added_or_present_in_keys(
2168
selected_keys, pb=pb),
2173
def iter_files_bytes(self, desired_files):
2174
"""Iterate through file versions.
2176
Files will not necessarily be returned in the order they occur in
2177
desired_files. No specific order is guaranteed.
2179
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2180
value supplied by the caller as part of desired_files. It should
2181
uniquely identify the file version in the caller's context. (Examples:
2182
an index number or a TreeTransform trans_id.)
2184
bytes_iterator is an iterable of bytestrings for the file. The
2185
kind of iterable and length of the bytestrings are unspecified, but for
2186
this implementation, it is a list of bytes produced by
2187
VersionedFile.get_record_stream().
2189
:param desired_files: a list of (file_id, revision_id, identifier)
2193
for file_id, revision_id, callable_data in desired_files:
2194
text_keys[(file_id, revision_id)] = callable_data
2195
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2196
if record.storage_kind == 'absent':
2197
raise errors.RevisionNotPresent(record.key, self)
2198
yield text_keys[record.key], record.get_bytes_as('chunked')
2200
def _generate_text_key_index(self, text_key_references=None,
2202
"""Generate a new text key index for the repository.
2204
This is an expensive function that will take considerable time to run.
2206
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2207
list of parents, also text keys. When a given key has no parents,
2208
the parents list will be [NULL_REVISION].
2210
# All revisions, to find inventory parents.
2211
if ancestors is None:
2212
graph = self.get_graph()
2213
ancestors = graph.get_parent_map(self.all_revision_ids())
2214
if text_key_references is None:
2215
text_key_references = self.find_text_key_references()
2216
pb = ui.ui_factory.nested_progress_bar()
2218
return self._do_generate_text_key_index(ancestors,
2219
text_key_references, pb)
2223
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2224
"""Helper for _generate_text_key_index to avoid deep nesting."""
2225
revision_order = tsort.topo_sort(ancestors)
2226
invalid_keys = set()
2228
for revision_id in revision_order:
2229
revision_keys[revision_id] = set()
2230
text_count = len(text_key_references)
2231
# a cache of the text keys to allow reuse; costs a dict of all the
2232
# keys, but saves a 2-tuple for every child of a given key.
2234
for text_key, valid in text_key_references.iteritems():
2236
invalid_keys.add(text_key)
2238
revision_keys[text_key[1]].add(text_key)
2239
text_key_cache[text_key] = text_key
2240
del text_key_references
2242
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2243
NULL_REVISION = _mod_revision.NULL_REVISION
2244
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2245
# too small for large or very branchy trees. However, for 55K path
2246
# trees, it would be easy to use too much memory trivially. Ideally we
2247
# could gauge this by looking at available real memory etc, but this is
2248
# always a tricky proposition.
2249
inventory_cache = lru_cache.LRUCache(10)
2250
batch_size = 10 # should be ~150MB on a 55K path tree
2251
batch_count = len(revision_order) / batch_size + 1
2253
pb.update("Calculating text parents", processed_texts, text_count)
2254
for offset in xrange(batch_count):
2255
to_query = revision_order[offset * batch_size:(offset + 1) *
2259
for revision_id in to_query:
2260
parent_ids = ancestors[revision_id]
2261
for text_key in revision_keys[revision_id]:
2262
pb.update("Calculating text parents", processed_texts)
2263
processed_texts += 1
2264
candidate_parents = []
2265
for parent_id in parent_ids:
2266
parent_text_key = (text_key[0], parent_id)
2268
check_parent = parent_text_key not in \
2269
revision_keys[parent_id]
2271
# the parent parent_id is a ghost:
2272
check_parent = False
2273
# truncate the derived graph against this ghost.
2274
parent_text_key = None
2276
# look at the parent commit details inventories to
2277
# determine possible candidates in the per file graph.
2280
inv = inventory_cache[parent_id]
2282
inv = self.revision_tree(parent_id).inventory
2283
inventory_cache[parent_id] = inv
2285
parent_entry = inv[text_key[0]]
2286
except (KeyError, errors.NoSuchId):
2288
if parent_entry is not None:
2290
text_key[0], parent_entry.revision)
2292
parent_text_key = None
2293
if parent_text_key is not None:
2294
candidate_parents.append(
2295
text_key_cache[parent_text_key])
2296
parent_heads = text_graph.heads(candidate_parents)
2297
new_parents = list(parent_heads)
2298
new_parents.sort(key=lambda x:candidate_parents.index(x))
2299
if new_parents == []:
2300
new_parents = [NULL_REVISION]
2301
text_index[text_key] = new_parents
2303
for text_key in invalid_keys:
2304
text_index[text_key] = [NULL_REVISION]
2307
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2308
"""Get an iterable listing the keys of all the data introduced by a set
2311
The keys will be ordered so that the corresponding items can be safely
2312
fetched and inserted in that order.
2314
:returns: An iterable producing tuples of (knit-kind, file-id,
2315
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2316
'revisions'. file-id is None unless knit-kind is 'file'.
2318
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2321
for result in self._find_non_file_keys_to_fetch(revision_ids):
2324
def _find_file_keys_to_fetch(self, revision_ids, pb):
2325
# XXX: it's a bit weird to control the inventory weave caching in this
2326
# generator. Ideally the caching would be done in fetch.py I think. Or
2327
# maybe this generator should explicitly have the contract that it
2328
# should not be iterated until the previously yielded item has been
2330
inv_w = self.inventories
2332
# file ids that changed
2333
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2335
num_file_ids = len(file_ids)
2336
for file_id, altered_versions in file_ids.iteritems():
2338
pb.update("Fetch texts", count, num_file_ids)
2340
yield ("file", file_id, altered_versions)
2342
def _find_non_file_keys_to_fetch(self, revision_ids):
2344
yield ("inventory", None, revision_ids)
2347
# XXX: Note ATM no callers actually pay attention to this return
2348
# instead they just use the list of revision ids and ignore
2349
# missing sigs. Consider removing this work entirely
2350
revisions_with_signatures = set(self.signatures.get_parent_map(
2351
[(r,) for r in revision_ids]))
2352
revisions_with_signatures = set(
2353
[r for (r,) in revisions_with_signatures])
2354
revisions_with_signatures.intersection_update(revision_ids)
2355
yield ("signatures", None, revisions_with_signatures)
2358
yield ("revisions", None, revision_ids)
2361
def get_inventory(self, revision_id):
2362
"""Get Inventory object by revision id."""
2363
return self.iter_inventories([revision_id]).next()
2365
def iter_inventories(self, revision_ids, ordering=None):
2366
"""Get many inventories by revision_ids.
2368
This will buffer some or all of the texts used in constructing the
2369
inventories in memory, but will only parse a single inventory at a
2372
:param revision_ids: The expected revision ids of the inventories.
2373
:param ordering: optional ordering, e.g. 'topological'. If not
2374
specified, the order of revision_ids will be preserved (by
2375
buffering if necessary).
2376
:return: An iterator of inventories.
2378
if ((None in revision_ids)
2379
or (_mod_revision.NULL_REVISION in revision_ids)):
2380
raise ValueError('cannot get null revision inventory')
2381
return self._iter_inventories(revision_ids, ordering)
2383
def _iter_inventories(self, revision_ids, ordering):
2384
"""single-document based inventory iteration."""
2385
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2386
for text, revision_id in inv_xmls:
2387
yield self.deserialise_inventory(revision_id, text)
2389
def _iter_inventory_xmls(self, revision_ids, ordering):
2390
if ordering is None:
2391
order_as_requested = True
2392
ordering = 'unordered'
2394
order_as_requested = False
2395
keys = [(revision_id,) for revision_id in revision_ids]
2398
if order_as_requested:
2399
key_iter = iter(keys)
2400
next_key = key_iter.next()
2401
stream = self.inventories.get_record_stream(keys, ordering, True)
2403
for record in stream:
2404
if record.storage_kind != 'absent':
2405
chunks = record.get_bytes_as('chunked')
2406
if order_as_requested:
2407
text_chunks[record.key] = chunks
2409
yield ''.join(chunks), record.key[-1]
2411
raise errors.NoSuchRevision(self, record.key)
2412
if order_as_requested:
2413
# Yield as many results as we can while preserving order.
2414
while next_key in text_chunks:
2415
chunks = text_chunks.pop(next_key)
2416
yield ''.join(chunks), next_key[-1]
2418
next_key = key_iter.next()
2419
except StopIteration:
2420
# We still want to fully consume the get_record_stream,
2421
# just in case it is not actually finished at this point
2425
def deserialise_inventory(self, revision_id, xml):
2426
"""Transform the xml into an inventory object.
2428
:param revision_id: The expected revision id of the inventory.
2429
:param xml: A serialised inventory.
2431
result = self._serializer.read_inventory_from_string(xml, revision_id,
2432
entry_cache=self._inventory_entry_cache)
2433
if result.revision_id != revision_id:
2434
raise AssertionError('revision id mismatch %s != %s' % (
2435
result.revision_id, revision_id))
2438
def serialise_inventory(self, inv):
2439
return self._serializer.write_inventory_to_string(inv)
2441
def _serialise_inventory_to_lines(self, inv):
2442
return self._serializer.write_inventory_to_lines(inv)
2444
def get_serializer_format(self):
2445
return self._serializer.format_num
2448
def get_inventory_xml(self, revision_id):
2449
"""Get inventory XML as a file object."""
2450
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2452
text, revision_id = texts.next()
2453
except StopIteration:
2454
raise errors.HistoryMissing(self, 'inventory', revision_id)
2458
def get_inventory_sha1(self, revision_id):
2459
"""Return the sha1 hash of the inventory entry
2461
return self.get_revision(revision_id).inventory_sha1
2463
def get_rev_id_for_revno(self, revno, known_pair):
2464
"""Return the revision id of a revno, given a later (revno, revid)
2465
pair in the same history.
2467
:return: if found (True, revid). If the available history ran out
2468
before reaching the revno, then this returns
2469
(False, (closest_revno, closest_revid)).
2471
known_revno, known_revid = known_pair
2472
partial_history = [known_revid]
2473
distance_from_known = known_revno - revno
2474
if distance_from_known < 0:
2476
'requested revno (%d) is later than given known revno (%d)'
2477
% (revno, known_revno))
2480
self, partial_history, stop_index=distance_from_known)
2481
except errors.RevisionNotPresent, err:
2482
if err.revision_id == known_revid:
2483
# The start revision (known_revid) wasn't found.
2485
# This is a stacked repository with no fallbacks, or a there's a
2486
# left-hand ghost. Either way, even though the revision named in
2487
# the error isn't in this repo, we know it's the next step in this
2488
# left-hand history.
2489
partial_history.append(err.revision_id)
2490
if len(partial_history) <= distance_from_known:
2491
# Didn't find enough history to get a revid for the revno.
2492
earliest_revno = known_revno - len(partial_history) + 1
2493
return (False, (earliest_revno, partial_history[-1]))
2494
if len(partial_history) - 1 > distance_from_known:
2495
raise AssertionError('_iter_for_revno returned too much history')
2496
return (True, partial_history[-1])
2498
def iter_reverse_revision_history(self, revision_id):
2499
"""Iterate backwards through revision ids in the lefthand history
2501
:param revision_id: The revision id to start with. All its lefthand
2502
ancestors will be traversed.
2504
graph = self.get_graph()
2505
next_id = revision_id
2507
if next_id in (None, _mod_revision.NULL_REVISION):
2510
parents = graph.get_parent_map([next_id])[next_id]
2512
raise errors.RevisionNotPresent(next_id, self)
2514
if len(parents) == 0:
2517
next_id = parents[0]
2520
def get_revision_inventory(self, revision_id):
2521
"""Return inventory of a past revision."""
2522
# TODO: Unify this with get_inventory()
2523
# bzr 0.0.6 and later imposes the constraint that the inventory_id
2524
# must be the same as its revision, so this is trivial.
2525
if revision_id is None:
2526
# This does not make sense: if there is no revision,
2527
# then it is the current tree inventory surely ?!
2528
# and thus get_root_id() is something that looks at the last
2529
# commit on the branch, and the get_root_id is an inventory check.
2530
raise NotImplementedError
2531
# return Inventory(self.get_root_id())
2533
return self.get_inventory(revision_id)
2535
def is_shared(self):
2536
"""Return True if this repository is flagged as a shared repository."""
2537
raise NotImplementedError(self.is_shared)
2540
def reconcile(self, other=None, thorough=False):
2541
"""Reconcile this repository."""
2542
from bzrlib.reconcile import RepoReconciler
2543
reconciler = RepoReconciler(self, thorough=thorough)
2544
reconciler.reconcile()
2547
def _refresh_data(self):
2548
"""Helper called from lock_* to ensure coherency with disk.
2550
The default implementation does nothing; it is however possible
2551
for repositories to maintain loaded indices across multiple locks
2552
by checking inside their implementation of this method to see
2553
whether their indices are still valid. This depends of course on
2554
the disk format being validatable in this manner. This method is
2555
also called by the refresh_data() public interface to cause a refresh
2556
to occur while in a write lock so that data inserted by a smart server
2557
push operation is visible on the client's instance of the physical
2562
def revision_tree(self, revision_id):
2563
"""Return Tree for a revision on this branch.
2565
`revision_id` may be NULL_REVISION for the empty tree revision.
2567
revision_id = _mod_revision.ensure_null(revision_id)
2568
# TODO: refactor this to use an existing revision object
2569
# so we don't need to read it in twice.
2570
if revision_id == _mod_revision.NULL_REVISION:
2571
return RevisionTree(self, Inventory(root_id=None),
2572
_mod_revision.NULL_REVISION)
2574
inv = self.get_revision_inventory(revision_id)
2575
return RevisionTree(self, inv, revision_id)
2577
def revision_trees(self, revision_ids):
2578
"""Return Trees for revisions in this repository.
2580
:param revision_ids: a sequence of revision-ids;
2581
a revision-id may not be None or 'null:'
2583
inventories = self.iter_inventories(revision_ids)
2584
for inv in inventories:
2585
yield RevisionTree(self, inv, inv.revision_id)
2587
def _filtered_revision_trees(self, revision_ids, file_ids):
2588
"""Return Tree for a revision on this branch with only some files.
2590
:param revision_ids: a sequence of revision-ids;
2591
a revision-id may not be None or 'null:'
2592
:param file_ids: if not None, the result is filtered
2593
so that only those file-ids, their parents and their
2594
children are included.
2596
inventories = self.iter_inventories(revision_ids)
2597
for inv in inventories:
2598
# Should we introduce a FilteredRevisionTree class rather
2599
# than pre-filter the inventory here?
2600
filtered_inv = inv.filter(file_ids)
2601
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2604
def get_ancestry(self, revision_id, topo_sorted=True):
2605
"""Return a list of revision-ids integrated by a revision.
2607
The first element of the list is always None, indicating the origin
2608
revision. This might change when we have history horizons, or
2609
perhaps we should have a new API.
2611
This is topologically sorted.
2613
if _mod_revision.is_null(revision_id):
2615
if not self.has_revision(revision_id):
2616
raise errors.NoSuchRevision(self, revision_id)
2617
graph = self.get_graph()
2619
search = graph._make_breadth_first_searcher([revision_id])
2622
found, ghosts = search.next_with_ghosts()
2623
except StopIteration:
2626
if _mod_revision.NULL_REVISION in keys:
2627
keys.remove(_mod_revision.NULL_REVISION)
2629
parent_map = graph.get_parent_map(keys)
2630
keys = tsort.topo_sort(parent_map)
2631
return [None] + list(keys)
2633
def pack(self, hint=None):
2634
"""Compress the data within the repository.
2636
This operation only makes sense for some repository types. For other
2637
types it should be a no-op that just returns.
2639
This stub method does not require a lock, but subclasses should use
2640
@needs_write_lock as this is a long running call its reasonable to
2641
implicitly lock for the user.
2643
:param hint: If not supplied, the whole repository is packed.
2644
If supplied, the repository may use the hint parameter as a
2645
hint for the parts of the repository to pack. A hint can be
2646
obtained from the result of commit_write_group(). Out of
2647
date hints are simply ignored, because concurrent operations
2648
can obsolete them rapidly.
2651
def get_transaction(self):
2652
return self.control_files.get_transaction()
2654
def get_parent_map(self, revision_ids):
2655
"""See graph.StackedParentsProvider.get_parent_map"""
2656
# revisions index works in keys; this just works in revisions
2657
# therefore wrap and unwrap
2660
for revision_id in revision_ids:
2661
if revision_id == _mod_revision.NULL_REVISION:
2662
result[revision_id] = ()
2663
elif revision_id is None:
2664
raise ValueError('get_parent_map(None) is not valid')
2666
query_keys.append((revision_id ,))
2667
for ((revision_id,), parent_keys) in \
2668
self.revisions.get_parent_map(query_keys).iteritems():
2670
result[revision_id] = tuple([parent_revid
2671
for (parent_revid,) in parent_keys])
2673
result[revision_id] = (_mod_revision.NULL_REVISION,)
2676
def _make_parents_provider(self):
2679
def get_graph(self, other_repository=None):
2680
"""Return the graph walker for this repository format"""
2681
parents_provider = self._make_parents_provider()
2682
if (other_repository is not None and
2683
not self.has_same_location(other_repository)):
2684
parents_provider = graph.StackedParentsProvider(
2685
[parents_provider, other_repository._make_parents_provider()])
2686
return graph.Graph(parents_provider)
2688
def _get_versioned_file_checker(self, text_key_references=None,
2690
"""Return an object suitable for checking versioned files.
2692
:param text_key_references: if non-None, an already built
2693
dictionary mapping text keys ((fileid, revision_id) tuples)
2694
to whether they were referred to by the inventory of the
2695
revision_id that they contain. If None, this will be
2697
:param ancestors: Optional result from
2698
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2701
return _VersionedFileChecker(self,
2702
text_key_references=text_key_references, ancestors=ancestors)
2704
def revision_ids_to_search_result(self, result_set):
2705
"""Convert a set of revision ids to a graph SearchResult."""
2706
result_parents = set()
2707
for parents in self.get_graph().get_parent_map(
2708
result_set).itervalues():
2709
result_parents.update(parents)
2710
included_keys = result_set.intersection(result_parents)
2711
start_keys = result_set.difference(included_keys)
2712
exclude_keys = result_parents.difference(result_set)
2713
result = graph.SearchResult(start_keys, exclude_keys,
2714
len(result_set), result_set)
2718
def set_make_working_trees(self, new_value):
2719
"""Set the policy flag for making working trees when creating branches.
2721
This only applies to branches that use this repository.
2723
The default is 'True'.
2724
:param new_value: True to restore the default, False to disable making
2727
raise NotImplementedError(self.set_make_working_trees)
2729
def make_working_trees(self):
2730
"""Returns the policy for making working trees on new branches."""
2731
raise NotImplementedError(self.make_working_trees)
2734
def sign_revision(self, revision_id, gpg_strategy):
2735
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2736
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2739
def has_signature_for_revision_id(self, revision_id):
2740
"""Query for a revision signature for revision_id in the repository."""
2741
if not self.has_revision(revision_id):
2742
raise errors.NoSuchRevision(self, revision_id)
2743
sig_present = (1 == len(
2744
self.signatures.get_parent_map([(revision_id,)])))
2748
def get_signature_text(self, revision_id):
2749
"""Return the text for a signature."""
2750
stream = self.signatures.get_record_stream([(revision_id,)],
2752
record = stream.next()
2753
if record.storage_kind == 'absent':
2754
raise errors.NoSuchRevision(self, revision_id)
2755
return record.get_bytes_as('fulltext')
2758
def check(self, revision_ids=None, callback_refs=None, check_repo=True):
2759
"""Check consistency of all history of given revision_ids.
2761
Different repository implementations should override _check().
2763
:param revision_ids: A non-empty list of revision_ids whose ancestry
2764
will be checked. Typically the last revision_id of a branch.
2765
:param callback_refs: A dict of check-refs to resolve and callback
2766
the check/_check method on the items listed as wanting the ref.
2768
:param check_repo: If False do not check the repository contents, just
2769
calculate the data callback_refs requires and call them back.
2771
return self._check(revision_ids, callback_refs=callback_refs,
2772
check_repo=check_repo)
2774
def _check(self, revision_ids, callback_refs, check_repo):
2775
result = check.Check(self, check_repo=check_repo)
2776
result.check(callback_refs)
2779
def _warn_if_deprecated(self, branch=None):
2780
global _deprecation_warning_done
2781
if _deprecation_warning_done:
2785
conf = config.GlobalConfig()
2787
conf = branch.get_config()
2788
if conf.suppress_warning('format_deprecation'):
2790
warning("Format %s for %s is deprecated -"
2791
" please use 'bzr upgrade' to get better performance"
2792
% (self._format, self.bzrdir.transport.base))
2794
_deprecation_warning_done = True
2796
def supports_rich_root(self):
2797
return self._format.rich_root_data
2799
def _check_ascii_revisionid(self, revision_id, method):
2800
"""Private helper for ascii-only repositories."""
2801
# weave repositories refuse to store revisionids that are non-ascii.
2802
if revision_id is not None:
2803
# weaves require ascii revision ids.
2804
if isinstance(revision_id, unicode):
2806
revision_id.encode('ascii')
2807
except UnicodeEncodeError:
2808
raise errors.NonAsciiRevisionId(method, self)
2811
revision_id.decode('ascii')
2812
except UnicodeDecodeError:
2813
raise errors.NonAsciiRevisionId(method, self)
2815
def revision_graph_can_have_wrong_parents(self):
2816
"""Is it possible for this repository to have a revision graph with
2819
If True, then this repository must also implement
2820
_find_inconsistent_revision_parents so that check and reconcile can
2821
check for inconsistencies before proceeding with other checks that may
2822
depend on the revision index being consistent.
2824
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2827
# remove these delegates a while after bzr 0.15
2828
def __make_delegated(name, from_module):
2829
def _deprecated_repository_forwarder():
2830
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2831
% (name, from_module),
2834
m = __import__(from_module, globals(), locals(), [name])
2836
return getattr(m, name)
2837
except AttributeError:
2838
raise AttributeError('module %s has no name %s'
2840
globals()[name] = _deprecated_repository_forwarder
2843
'AllInOneRepository',
2844
'WeaveMetaDirRepository',
2845
'PreSplitOutRepositoryFormat',
2846
'RepositoryFormat4',
2847
'RepositoryFormat5',
2848
'RepositoryFormat6',
2849
'RepositoryFormat7',
2851
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2855
'RepositoryFormatKnit',
2856
'RepositoryFormatKnit1',
2858
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2861
def install_revision(repository, rev, revision_tree):
2862
"""Install all revision data into a repository."""
2863
install_revisions(repository, [(rev, revision_tree, None)])
2866
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2867
"""Install all revision data into a repository.
2869
Accepts an iterable of revision, tree, signature tuples. The signature
2872
repository.start_write_group()
2874
inventory_cache = lru_cache.LRUCache(10)
2875
for n, (revision, revision_tree, signature) in enumerate(iterable):
2876
_install_revision(repository, revision, revision_tree, signature,
2879
pb.update('Transferring revisions', n + 1, num_revisions)
2881
repository.abort_write_group()
2884
repository.commit_write_group()
2887
def _install_revision(repository, rev, revision_tree, signature,
2889
"""Install all revision data into a repository."""
2890
present_parents = []
2892
for p_id in rev.parent_ids:
2893
if repository.has_revision(p_id):
2894
present_parents.append(p_id)
2895
parent_trees[p_id] = repository.revision_tree(p_id)
2897
parent_trees[p_id] = repository.revision_tree(
2898
_mod_revision.NULL_REVISION)
2900
inv = revision_tree.inventory
2901
entries = inv.iter_entries()
2902
# backwards compatibility hack: skip the root id.
2903
if not repository.supports_rich_root():
2904
path, root = entries.next()
2905
if root.revision != rev.revision_id:
2906
raise errors.IncompatibleRevision(repr(repository))
2908
for path, ie in entries:
2909
text_keys[(ie.file_id, ie.revision)] = ie
2910
text_parent_map = repository.texts.get_parent_map(text_keys)
2911
missing_texts = set(text_keys) - set(text_parent_map)
2912
# Add the texts that are not already present
2913
for text_key in missing_texts:
2914
ie = text_keys[text_key]
2916
# FIXME: TODO: The following loop overlaps/duplicates that done by
2917
# commit to determine parents. There is a latent/real bug here where
2918
# the parents inserted are not those commit would do - in particular
2919
# they are not filtered by heads(). RBC, AB
2920
for revision, tree in parent_trees.iteritems():
2921
if ie.file_id not in tree:
2923
parent_id = tree.inventory[ie.file_id].revision
2924
if parent_id in text_parents:
2926
text_parents.append((ie.file_id, parent_id))
2927
lines = revision_tree.get_file(ie.file_id).readlines()
2928
repository.texts.add_lines(text_key, text_parents, lines)
2930
# install the inventory
2931
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2932
# Cache this inventory
2933
inventory_cache[rev.revision_id] = inv
2935
basis_inv = inventory_cache[rev.parent_ids[0]]
2937
repository.add_inventory(rev.revision_id, inv, present_parents)
2939
delta = inv._make_delta(basis_inv)
2940
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2941
rev.revision_id, present_parents)
2943
repository.add_inventory(rev.revision_id, inv, present_parents)
2944
except errors.RevisionAlreadyPresent:
2946
if signature is not None:
2947
repository.add_signature_text(rev.revision_id, signature)
2948
repository.add_revision(rev.revision_id, rev, inv)
2951
class MetaDirRepository(Repository):
2952
"""Repositories in the new meta-dir layout.
2954
:ivar _transport: Transport for access to repository control files,
2955
typically pointing to .bzr/repository.
2958
def __init__(self, _format, a_bzrdir, control_files):
2959
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2960
self._transport = control_files._transport
2962
def is_shared(self):
2963
"""Return True if this repository is flagged as a shared repository."""
2964
return self._transport.has('shared-storage')
2967
def set_make_working_trees(self, new_value):
2968
"""Set the policy flag for making working trees when creating branches.
2970
This only applies to branches that use this repository.
2972
The default is 'True'.
2973
:param new_value: True to restore the default, False to disable making
2978
self._transport.delete('no-working-trees')
2979
except errors.NoSuchFile:
2982
self._transport.put_bytes('no-working-trees', '',
2983
mode=self.bzrdir._get_file_mode())
2985
def make_working_trees(self):
2986
"""Returns the policy for making working trees on new branches."""
2987
return not self._transport.has('no-working-trees')
2990
class MetaDirVersionedFileRepository(MetaDirRepository):
2991
"""Repositories in a meta-dir, that work via versioned file objects."""
2993
def __init__(self, _format, a_bzrdir, control_files):
2994
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2998
network_format_registry = registry.FormatRegistry()
2999
"""Registry of formats indexed by their network name.
3001
The network name for a repository format is an identifier that can be used when
3002
referring to formats with smart server operations. See
3003
RepositoryFormat.network_name() for more detail.
3007
format_registry = registry.FormatRegistry(network_format_registry)
3008
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
3010
This can contain either format instances themselves, or classes/factories that
3011
can be called to obtain one.
3015
#####################################################################
3016
# Repository Formats
3018
class RepositoryFormat(object):
3019
"""A repository format.
3021
Formats provide four things:
3022
* An initialization routine to construct repository data on disk.
3023
* a optional format string which is used when the BzrDir supports
3025
* an open routine which returns a Repository instance.
3026
* A network name for referring to the format in smart server RPC
3029
There is one and only one Format subclass for each on-disk format. But
3030
there can be one Repository subclass that is used for several different
3031
formats. The _format attribute on a Repository instance can be used to
3032
determine the disk format.
3034
Formats are placed in a registry by their format string for reference
3035
during opening. These should be subclasses of RepositoryFormat for
3038
Once a format is deprecated, just deprecate the initialize and open
3039
methods on the format class. Do not deprecate the object, as the
3040
object may be created even when a repository instance hasn't been
3043
Common instance attributes:
3044
_matchingbzrdir - the bzrdir format that the repository format was
3045
originally written to work with. This can be used if manually
3046
constructing a bzrdir and repository, or more commonly for test suite
3050
# Set to True or False in derived classes. True indicates that the format
3051
# supports ghosts gracefully.
3052
supports_ghosts = None
3053
# Can this repository be given external locations to lookup additional
3054
# data. Set to True or False in derived classes.
3055
supports_external_lookups = None
3056
# Does this format support CHK bytestring lookups. Set to True or False in
3058
supports_chks = None
3059
# Should commit add an inventory, or an inventory delta to the repository.
3060
_commit_inv_deltas = True
3061
# What order should fetch operations request streams in?
3062
# The default is unordered as that is the cheapest for an origin to
3064
_fetch_order = 'unordered'
3065
# Does this repository format use deltas that can be fetched as-deltas ?
3066
# (E.g. knits, where the knit deltas can be transplanted intact.
3067
# We default to False, which will ensure that enough data to get
3068
# a full text out of any fetch stream will be grabbed.
3069
_fetch_uses_deltas = False
3070
# Should fetch trigger a reconcile after the fetch? Only needed for
3071
# some repository formats that can suffer internal inconsistencies.
3072
_fetch_reconcile = False
3073
# Does this format have < O(tree_size) delta generation. Used to hint what
3074
# code path for commit, amongst other things.
3076
# Does doing a pack operation compress data? Useful for the pack UI command
3077
# (so if there is one pack, the operation can still proceed because it may
3078
# help), and for fetching when data won't have come from the same
3080
pack_compresses = False
3081
# Does the repository inventory storage understand references to trees?
3082
supports_tree_reference = None
3085
return "<%s>" % self.__class__.__name__
3087
def __eq__(self, other):
3088
# format objects are generally stateless
3089
return isinstance(other, self.__class__)
3091
def __ne__(self, other):
3092
return not self == other
3095
def find_format(klass, a_bzrdir):
3096
"""Return the format for the repository object in a_bzrdir.
3098
This is used by bzr native formats that have a "format" file in
3099
the repository. Other methods may be used by different types of
3103
transport = a_bzrdir.get_repository_transport(None)
3104
format_string = transport.get("format").read()
3105
return format_registry.get(format_string)
3106
except errors.NoSuchFile:
3107
raise errors.NoRepositoryPresent(a_bzrdir)
3109
raise errors.UnknownFormatError(format=format_string,
3113
def register_format(klass, format):
3114
format_registry.register(format.get_format_string(), format)
3117
def unregister_format(klass, format):
3118
format_registry.remove(format.get_format_string())
3121
def get_default_format(klass):
3122
"""Return the current default format."""
3123
from bzrlib import bzrdir
3124
return bzrdir.format_registry.make_bzrdir('default').repository_format
3126
def get_format_string(self):
3127
"""Return the ASCII format string that identifies this format.
3129
Note that in pre format ?? repositories the format string is
3130
not permitted nor written to disk.
3132
raise NotImplementedError(self.get_format_string)
3134
def get_format_description(self):
3135
"""Return the short description for this format."""
3136
raise NotImplementedError(self.get_format_description)
3138
# TODO: this shouldn't be in the base class, it's specific to things that
3139
# use weaves or knits -- mbp 20070207
3140
def _get_versioned_file_store(self,
3145
versionedfile_class=None,
3146
versionedfile_kwargs={},
3148
if versionedfile_class is None:
3149
versionedfile_class = self._versionedfile_class
3150
weave_transport = control_files._transport.clone(name)
3151
dir_mode = control_files._dir_mode
3152
file_mode = control_files._file_mode
3153
return VersionedFileStore(weave_transport, prefixed=prefixed,
3155
file_mode=file_mode,
3156
versionedfile_class=versionedfile_class,
3157
versionedfile_kwargs=versionedfile_kwargs,
3160
def initialize(self, a_bzrdir, shared=False):
3161
"""Initialize a repository of this format in a_bzrdir.
3163
:param a_bzrdir: The bzrdir to put the new repository in it.
3164
:param shared: The repository should be initialized as a sharable one.
3165
:returns: The new repository object.
3167
This may raise UninitializableFormat if shared repository are not
3168
compatible the a_bzrdir.
3170
raise NotImplementedError(self.initialize)
3172
def is_supported(self):
3173
"""Is this format supported?
3175
Supported formats must be initializable and openable.
3176
Unsupported formats may not support initialization or committing or
3177
some other features depending on the reason for not being supported.
3181
def network_name(self):
3182
"""A simple byte string uniquely identifying this format for RPC calls.
3184
MetaDir repository formats use their disk format string to identify the
3185
repository over the wire. All in one formats such as bzr < 0.8, and
3186
foreign formats like svn/git and hg should use some marker which is
3187
unique and immutable.
3189
raise NotImplementedError(self.network_name)
3191
def check_conversion_target(self, target_format):
3192
if self.rich_root_data and not target_format.rich_root_data:
3193
raise errors.BadConversionTarget(
3194
'Does not support rich root data.', target_format,
3196
if (self.supports_tree_reference and
3197
not getattr(target_format, 'supports_tree_reference', False)):
3198
raise errors.BadConversionTarget(
3199
'Does not support nested trees', target_format,
3202
def open(self, a_bzrdir, _found=False):
3203
"""Return an instance of this format for the bzrdir a_bzrdir.
3205
_found is a private parameter, do not use it.
3207
raise NotImplementedError(self.open)
3210
class MetaDirRepositoryFormat(RepositoryFormat):
3211
"""Common base class for the new repositories using the metadir layout."""
3213
rich_root_data = False
3214
supports_tree_reference = False
3215
supports_external_lookups = False
3218
def _matchingbzrdir(self):
3219
matching = bzrdir.BzrDirMetaFormat1()
3220
matching.repository_format = self
3224
super(MetaDirRepositoryFormat, self).__init__()
3226
def _create_control_files(self, a_bzrdir):
3227
"""Create the required files and the initial control_files object."""
3228
# FIXME: RBC 20060125 don't peek under the covers
3229
# NB: no need to escape relative paths that are url safe.
3230
repository_transport = a_bzrdir.get_repository_transport(self)
3231
control_files = lockable_files.LockableFiles(repository_transport,
3232
'lock', lockdir.LockDir)
3233
control_files.create_lock()
3234
return control_files
3236
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
3237
"""Upload the initial blank content."""
3238
control_files = self._create_control_files(a_bzrdir)
3239
control_files.lock_write()
3240
transport = control_files._transport
3242
utf8_files += [('shared-storage', '')]
3244
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
3245
for (filename, content_stream) in files:
3246
transport.put_file(filename, content_stream,
3247
mode=a_bzrdir._get_file_mode())
3248
for (filename, content_bytes) in utf8_files:
3249
transport.put_bytes_non_atomic(filename, content_bytes,
3250
mode=a_bzrdir._get_file_mode())
3252
control_files.unlock()
3254
def network_name(self):
3255
"""Metadir formats have matching disk and network format strings."""
3256
return self.get_format_string()
3259
# Pre-0.8 formats that don't have a disk format string (because they are
3260
# versioned by the matching control directory). We use the control directories
3261
# disk format string as a key for the network_name because they meet the
3262
# constraints (simple string, unique, immutable).
3263
network_format_registry.register_lazy(
3264
"Bazaar-NG branch, format 5\n",
3265
'bzrlib.repofmt.weaverepo',
3266
'RepositoryFormat5',
3268
network_format_registry.register_lazy(
3269
"Bazaar-NG branch, format 6\n",
3270
'bzrlib.repofmt.weaverepo',
3271
'RepositoryFormat6',
3274
# formats which have no format string are not discoverable or independently
3275
# creatable on disk, so are not registered in format_registry. They're
3276
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
3277
# needed, it's constructed directly by the BzrDir. Non-native formats where
3278
# the repository is not separately opened are similar.
3280
format_registry.register_lazy(
3281
'Bazaar-NG Repository format 7',
3282
'bzrlib.repofmt.weaverepo',
3286
format_registry.register_lazy(
3287
'Bazaar-NG Knit Repository Format 1',
3288
'bzrlib.repofmt.knitrepo',
3289
'RepositoryFormatKnit1',
3292
format_registry.register_lazy(
3293
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
3294
'bzrlib.repofmt.knitrepo',
3295
'RepositoryFormatKnit3',
3298
format_registry.register_lazy(
3299
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
3300
'bzrlib.repofmt.knitrepo',
3301
'RepositoryFormatKnit4',
3304
# Pack-based formats. There is one format for pre-subtrees, and one for
3305
# post-subtrees to allow ease of testing.
3306
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
3307
format_registry.register_lazy(
3308
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
3309
'bzrlib.repofmt.pack_repo',
3310
'RepositoryFormatKnitPack1',
3312
format_registry.register_lazy(
3313
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
3314
'bzrlib.repofmt.pack_repo',
3315
'RepositoryFormatKnitPack3',
3317
format_registry.register_lazy(
3318
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
3319
'bzrlib.repofmt.pack_repo',
3320
'RepositoryFormatKnitPack4',
3322
format_registry.register_lazy(
3323
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
3324
'bzrlib.repofmt.pack_repo',
3325
'RepositoryFormatKnitPack5',
3327
format_registry.register_lazy(
3328
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
3329
'bzrlib.repofmt.pack_repo',
3330
'RepositoryFormatKnitPack5RichRoot',
3332
format_registry.register_lazy(
3333
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3334
'bzrlib.repofmt.pack_repo',
3335
'RepositoryFormatKnitPack5RichRootBroken',
3337
format_registry.register_lazy(
3338
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3339
'bzrlib.repofmt.pack_repo',
3340
'RepositoryFormatKnitPack6',
3342
format_registry.register_lazy(
3343
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3344
'bzrlib.repofmt.pack_repo',
3345
'RepositoryFormatKnitPack6RichRoot',
3348
# Development formats.
3349
# Obsolete but kept pending a CHK based subtree format.
3350
format_registry.register_lazy(
3351
("Bazaar development format 2 with subtree support "
3352
"(needs bzr.dev from before 1.8)\n"),
3353
'bzrlib.repofmt.pack_repo',
3354
'RepositoryFormatPackDevelopment2Subtree',
3357
# 1.14->1.16 go below here
3358
format_registry.register_lazy(
3359
'Bazaar development format - group compression and chk inventory'
3360
' (needs bzr.dev from 1.14)\n',
3361
'bzrlib.repofmt.groupcompress_repo',
3362
'RepositoryFormatCHK1',
3365
format_registry.register_lazy(
3366
'Bazaar development format - chk repository with bencode revision '
3367
'serialization (needs bzr.dev from 1.16)\n',
3368
'bzrlib.repofmt.groupcompress_repo',
3369
'RepositoryFormatCHK2',
3371
format_registry.register_lazy(
3372
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3373
'bzrlib.repofmt.groupcompress_repo',
3374
'RepositoryFormat2a',
3378
class InterRepository(InterObject):
3379
"""This class represents operations taking place between two repositories.
3381
Its instances have methods like copy_content and fetch, and contain
3382
references to the source and target repositories these operations can be
3385
Often we will provide convenience methods on 'repository' which carry out
3386
operations with another repository - they will always forward to
3387
InterRepository.get(other).method_name(parameters).
3390
_walk_to_common_revisions_batch_size = 50
3392
"""The available optimised InterRepository types."""
3395
def copy_content(self, revision_id=None):
3396
"""Make a complete copy of the content in self into destination.
3398
This is a destructive operation! Do not use it on existing
3401
:param revision_id: Only copy the content needed to construct
3402
revision_id and its parents.
3405
self.target.set_make_working_trees(self.source.make_working_trees())
3406
except NotImplementedError:
3408
self.target.fetch(self.source, revision_id=revision_id)
3411
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3413
"""Fetch the content required to construct revision_id.
3415
The content is copied from self.source to self.target.
3417
:param revision_id: if None all content is copied, if NULL_REVISION no
3419
:param pb: optional progress bar to use for progress reports. If not
3420
provided a default one will be created.
3423
f = _mod_fetch.RepoFetcher(to_repository=self.target,
3424
from_repository=self.source,
3425
last_revision=revision_id,
3426
fetch_spec=fetch_spec,
3427
pb=pb, find_ghosts=find_ghosts)
3429
def _walk_to_common_revisions(self, revision_ids):
3430
"""Walk out from revision_ids in source to revisions target has.
3432
:param revision_ids: The start point for the search.
3433
:return: A set of revision ids.
3435
target_graph = self.target.get_graph()
3436
revision_ids = frozenset(revision_ids)
3437
missing_revs = set()
3438
source_graph = self.source.get_graph()
3439
# ensure we don't pay silly lookup costs.
3440
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3441
null_set = frozenset([_mod_revision.NULL_REVISION])
3442
searcher_exhausted = False
3446
# Iterate the searcher until we have enough next_revs
3447
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3449
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3450
next_revs.update(next_revs_part)
3451
ghosts.update(ghosts_part)
3452
except StopIteration:
3453
searcher_exhausted = True
3455
# If there are ghosts in the source graph, and the caller asked for
3456
# them, make sure that they are present in the target.
3457
# We don't care about other ghosts as we can't fetch them and
3458
# haven't been asked to.
3459
ghosts_to_check = set(revision_ids.intersection(ghosts))
3460
revs_to_get = set(next_revs).union(ghosts_to_check)
3462
have_revs = set(target_graph.get_parent_map(revs_to_get))
3463
# we always have NULL_REVISION present.
3464
have_revs = have_revs.union(null_set)
3465
# Check if the target is missing any ghosts we need.
3466
ghosts_to_check.difference_update(have_revs)
3468
# One of the caller's revision_ids is a ghost in both the
3469
# source and the target.
3470
raise errors.NoSuchRevision(
3471
self.source, ghosts_to_check.pop())
3472
missing_revs.update(next_revs - have_revs)
3473
# Because we may have walked past the original stop point, make
3474
# sure everything is stopped
3475
stop_revs = searcher.find_seen_ancestors(have_revs)
3476
searcher.stop_searching_any(stop_revs)
3477
if searcher_exhausted:
3479
return searcher.get_result()
3482
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3483
"""Return the revision ids that source has that target does not.
3485
:param revision_id: only return revision ids included by this
3487
:param find_ghosts: If True find missing revisions in deep history
3488
rather than just finding the surface difference.
3489
:return: A bzrlib.graph.SearchResult.
3491
# stop searching at found target revisions.
3492
if not find_ghosts and revision_id is not None:
3493
return self._walk_to_common_revisions([revision_id])
3494
# generic, possibly worst case, slow code path.
3495
target_ids = set(self.target.all_revision_ids())
3496
if revision_id is not None:
3497
source_ids = self.source.get_ancestry(revision_id)
3498
if source_ids[0] is not None:
3499
raise AssertionError()
3502
source_ids = self.source.all_revision_ids()
3503
result_set = set(source_ids).difference(target_ids)
3504
return self.source.revision_ids_to_search_result(result_set)
3507
def _same_model(source, target):
3508
"""True if source and target have the same data representation.
3510
Note: this is always called on the base class; overriding it in a
3511
subclass will have no effect.
3514
InterRepository._assert_same_model(source, target)
3516
except errors.IncompatibleRepositories, e:
3520
def _assert_same_model(source, target):
3521
"""Raise an exception if two repositories do not use the same model.
3523
if source.supports_rich_root() != target.supports_rich_root():
3524
raise errors.IncompatibleRepositories(source, target,
3525
"different rich-root support")
3526
if source._serializer != target._serializer:
3527
raise errors.IncompatibleRepositories(source, target,
3528
"different serializers")
3531
class InterSameDataRepository(InterRepository):
3532
"""Code for converting between repositories that represent the same data.
3534
Data format and model must match for this to work.
3538
def _get_repo_format_to_test(self):
3539
"""Repository format for testing with.
3541
InterSameData can pull from subtree to subtree and from non-subtree to
3542
non-subtree, so we test this with the richest repository format.
3544
from bzrlib.repofmt import knitrepo
3545
return knitrepo.RepositoryFormatKnit3()
3548
def is_compatible(source, target):
3549
return InterRepository._same_model(source, target)
3552
class InterWeaveRepo(InterSameDataRepository):
3553
"""Optimised code paths between Weave based repositories.
3555
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3556
implemented lazy inter-object optimisation.
3560
def _get_repo_format_to_test(self):
3561
from bzrlib.repofmt import weaverepo
3562
return weaverepo.RepositoryFormat7()
3565
def is_compatible(source, target):
3566
"""Be compatible with known Weave formats.
3568
We don't test for the stores being of specific types because that
3569
could lead to confusing results, and there is no need to be
3572
from bzrlib.repofmt.weaverepo import (
3578
return (isinstance(source._format, (RepositoryFormat5,
3580
RepositoryFormat7)) and
3581
isinstance(target._format, (RepositoryFormat5,
3583
RepositoryFormat7)))
3584
except AttributeError:
3588
def copy_content(self, revision_id=None):
3589
"""See InterRepository.copy_content()."""
3590
# weave specific optimised path:
3592
self.target.set_make_working_trees(self.source.make_working_trees())
3593
except (errors.RepositoryUpgradeRequired, NotImplemented):
3595
# FIXME do not peek!
3596
if self.source._transport.listable():
3597
pb = ui.ui_factory.nested_progress_bar()
3599
self.target.texts.insert_record_stream(
3600
self.source.texts.get_record_stream(
3601
self.source.texts.keys(), 'topological', False))
3602
pb.update('Copying inventory', 0, 1)
3603
self.target.inventories.insert_record_stream(
3604
self.source.inventories.get_record_stream(
3605
self.source.inventories.keys(), 'topological', False))
3606
self.target.signatures.insert_record_stream(
3607
self.source.signatures.get_record_stream(
3608
self.source.signatures.keys(),
3610
self.target.revisions.insert_record_stream(
3611
self.source.revisions.get_record_stream(
3612
self.source.revisions.keys(),
3613
'topological', True))
3617
self.target.fetch(self.source, revision_id=revision_id)
3620
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3621
"""See InterRepository.missing_revision_ids()."""
3622
# we want all revisions to satisfy revision_id in source.
3623
# but we don't want to stat every file here and there.
3624
# we want then, all revisions other needs to satisfy revision_id
3625
# checked, but not those that we have locally.
3626
# so the first thing is to get a subset of the revisions to
3627
# satisfy revision_id in source, and then eliminate those that
3628
# we do already have.
3629
# this is slow on high latency connection to self, but as this
3630
# disk format scales terribly for push anyway due to rewriting
3631
# inventory.weave, this is considered acceptable.
3633
if revision_id is not None:
3634
source_ids = self.source.get_ancestry(revision_id)
3635
if source_ids[0] is not None:
3636
raise AssertionError()
3639
source_ids = self.source._all_possible_ids()
3640
source_ids_set = set(source_ids)
3641
# source_ids is the worst possible case we may need to pull.
3642
# now we want to filter source_ids against what we actually
3643
# have in target, but don't try to check for existence where we know
3644
# we do not have a revision as that would be pointless.
3645
target_ids = set(self.target._all_possible_ids())
3646
possibly_present_revisions = target_ids.intersection(source_ids_set)
3647
actually_present_revisions = set(
3648
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3649
required_revisions = source_ids_set.difference(actually_present_revisions)
3650
if revision_id is not None:
3651
# we used get_ancestry to determine source_ids then we are assured all
3652
# revisions referenced are present as they are installed in topological order.
3653
# and the tip revision was validated by get_ancestry.
3654
result_set = required_revisions
3656
# if we just grabbed the possibly available ids, then
3657
# we only have an estimate of whats available and need to validate
3658
# that against the revision records.
3660
self.source._eliminate_revisions_not_present(required_revisions))
3661
return self.source.revision_ids_to_search_result(result_set)
3664
class InterKnitRepo(InterSameDataRepository):
3665
"""Optimised code paths between Knit based repositories."""
3668
def _get_repo_format_to_test(self):
3669
from bzrlib.repofmt import knitrepo
3670
return knitrepo.RepositoryFormatKnit1()
3673
def is_compatible(source, target):
3674
"""Be compatible with known Knit formats.
3676
We don't test for the stores being of specific types because that
3677
could lead to confusing results, and there is no need to be
3680
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3682
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3683
isinstance(target._format, RepositoryFormatKnit))
3684
except AttributeError:
3686
return are_knits and InterRepository._same_model(source, target)
3689
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3690
"""See InterRepository.missing_revision_ids()."""
3691
if revision_id is not None:
3692
source_ids = self.source.get_ancestry(revision_id)
3693
if source_ids[0] is not None:
3694
raise AssertionError()
3697
source_ids = self.source.all_revision_ids()
3698
source_ids_set = set(source_ids)
3699
# source_ids is the worst possible case we may need to pull.
3700
# now we want to filter source_ids against what we actually
3701
# have in target, but don't try to check for existence where we know
3702
# we do not have a revision as that would be pointless.
3703
target_ids = set(self.target.all_revision_ids())
3704
possibly_present_revisions = target_ids.intersection(source_ids_set)
3705
actually_present_revisions = set(
3706
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3707
required_revisions = source_ids_set.difference(actually_present_revisions)
3708
if revision_id is not None:
3709
# we used get_ancestry to determine source_ids then we are assured all
3710
# revisions referenced are present as they are installed in topological order.
3711
# and the tip revision was validated by get_ancestry.
3712
result_set = required_revisions
3714
# if we just grabbed the possibly available ids, then
3715
# we only have an estimate of whats available and need to validate
3716
# that against the revision records.
3718
self.source._eliminate_revisions_not_present(required_revisions))
3719
return self.source.revision_ids_to_search_result(result_set)
3722
class InterDifferingSerializer(InterRepository):
3725
def _get_repo_format_to_test(self):
3729
def is_compatible(source, target):
3730
"""Be compatible with Knit2 source and Knit3 target"""
3731
# This is redundant with format.check_conversion_target(), however that
3732
# raises an exception, and we just want to say "False" as in we won't
3733
# support converting between these formats.
3734
if 'IDS_never' in debug.debug_flags:
3736
if source.supports_rich_root() and not target.supports_rich_root():
3738
if (source._format.supports_tree_reference
3739
and not target._format.supports_tree_reference):
3741
if target._fallback_repositories and target._format.supports_chks:
3742
# IDS doesn't know how to copy CHKs for the parent inventories it
3743
# adds to stacked repos.
3745
if 'IDS_always' in debug.debug_flags:
3747
# Only use this code path for local source and target. IDS does far
3748
# too much IO (both bandwidth and roundtrips) over a network.
3749
if not source.bzrdir.transport.base.startswith('file:///'):
3751
if not target.bzrdir.transport.base.startswith('file:///'):
3755
def _get_trees(self, revision_ids, cache):
3757
for rev_id in revision_ids:
3759
possible_trees.append((rev_id, cache[rev_id]))
3761
# Not cached, but inventory might be present anyway.
3763
tree = self.source.revision_tree(rev_id)
3764
except errors.NoSuchRevision:
3765
# Nope, parent is ghost.
3768
cache[rev_id] = tree
3769
possible_trees.append((rev_id, tree))
3770
return possible_trees
3772
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3773
"""Get the best delta and base for this revision.
3775
:return: (basis_id, delta)
3778
# Generate deltas against each tree, to find the shortest.
3779
texts_possibly_new_in_tree = set()
3780
for basis_id, basis_tree in possible_trees:
3781
delta = tree.inventory._make_delta(basis_tree.inventory)
3782
for old_path, new_path, file_id, new_entry in delta:
3783
if new_path is None:
3784
# This file_id isn't present in the new rev, so we don't
3788
# Rich roots are handled elsewhere...
3790
kind = new_entry.kind
3791
if kind != 'directory' and kind != 'file':
3792
# No text record associated with this inventory entry.
3794
# This is a directory or file that has changed somehow.
3795
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3796
deltas.append((len(delta), basis_id, delta))
3798
return deltas[0][1:]
3800
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3801
"""Find all parent revisions that are absent, but for which the
3802
inventory is present, and copy those inventories.
3804
This is necessary to preserve correctness when the source is stacked
3805
without fallbacks configured. (Note that in cases like upgrade the
3806
source may be not have _fallback_repositories even though it is
3810
for parents in parent_map.values():
3811
parent_revs.update(parents)
3812
present_parents = self.source.get_parent_map(parent_revs)
3813
absent_parents = set(parent_revs).difference(present_parents)
3814
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3815
(rev_id,) for rev_id in absent_parents)
3816
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3817
for parent_tree in self.source.revision_trees(parent_inv_ids):
3818
current_revision_id = parent_tree.get_revision_id()
3819
parents_parents_keys = parent_invs_keys_for_stacking[
3820
(current_revision_id,)]
3821
parents_parents = [key[-1] for key in parents_parents_keys]
3822
basis_id = _mod_revision.NULL_REVISION
3823
basis_tree = self.source.revision_tree(basis_id)
3824
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3825
self.target.add_inventory_by_delta(
3826
basis_id, delta, current_revision_id, parents_parents)
3827
cache[current_revision_id] = parent_tree
3829
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3830
"""Fetch across a few revisions.
3832
:param revision_ids: The revisions to copy
3833
:param basis_id: The revision_id of a tree that must be in cache, used
3834
as a basis for delta when no other base is available
3835
:param cache: A cache of RevisionTrees that we can use.
3836
:param a_graph: A Graph object to determine the heads() of the
3837
rich-root data stream.
3838
:return: The revision_id of the last converted tree. The RevisionTree
3839
for it will be in cache
3841
# Walk though all revisions; get inventory deltas, copy referenced
3842
# texts that delta references, insert the delta, revision and
3844
root_keys_to_create = set()
3847
pending_revisions = []
3848
parent_map = self.source.get_parent_map(revision_ids)
3849
self._fetch_parent_invs_for_stacking(parent_map, cache)
3850
for tree in self.source.revision_trees(revision_ids):
3851
# Find a inventory delta for this revision.
3852
# Find text entries that need to be copied, too.
3853
current_revision_id = tree.get_revision_id()
3854
parent_ids = parent_map.get(current_revision_id, ())
3855
parent_trees = self._get_trees(parent_ids, cache)
3856
possible_trees = list(parent_trees)
3857
if len(possible_trees) == 0:
3858
# There either aren't any parents, or the parents are ghosts,
3859
# so just use the last converted tree.
3860
possible_trees.append((basis_id, cache[basis_id]))
3861
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3863
revision = self.source.get_revision(current_revision_id)
3864
pending_deltas.append((basis_id, delta,
3865
current_revision_id, revision.parent_ids))
3866
if self._converting_to_rich_root:
3867
self._revision_id_to_root_id[current_revision_id] = \
3869
# Determine which texts are in present in this revision but not in
3870
# any of the available parents.
3871
texts_possibly_new_in_tree = set()
3872
for old_path, new_path, file_id, entry in delta:
3873
if new_path is None:
3874
# This file_id isn't present in the new rev
3878
if not self.target.supports_rich_root():
3879
# The target doesn't support rich root, so we don't
3882
if self._converting_to_rich_root:
3883
# This can't be copied normally, we have to insert
3885
root_keys_to_create.add((file_id, entry.revision))
3888
texts_possibly_new_in_tree.add((file_id, entry.revision))
3889
for basis_id, basis_tree in possible_trees:
3890
basis_inv = basis_tree.inventory
3891
for file_key in list(texts_possibly_new_in_tree):
3892
file_id, file_revision = file_key
3894
entry = basis_inv[file_id]
3895
except errors.NoSuchId:
3897
if entry.revision == file_revision:
3898
texts_possibly_new_in_tree.remove(file_key)
3899
text_keys.update(texts_possibly_new_in_tree)
3900
pending_revisions.append(revision)
3901
cache[current_revision_id] = tree
3902
basis_id = current_revision_id
3904
from_texts = self.source.texts
3905
to_texts = self.target.texts
3906
if root_keys_to_create:
3907
root_stream = _mod_fetch._new_root_data_stream(
3908
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3909
self.source, graph=a_graph)
3910
to_texts.insert_record_stream(root_stream)
3911
to_texts.insert_record_stream(from_texts.get_record_stream(
3912
text_keys, self.target._format._fetch_order,
3913
not self.target._format._fetch_uses_deltas))
3914
# insert inventory deltas
3915
for delta in pending_deltas:
3916
self.target.add_inventory_by_delta(*delta)
3917
if self.target._fallback_repositories:
3918
# Make sure this stacked repository has all the parent inventories
3919
# for the new revisions that we are about to insert. We do this
3920
# before adding the revisions so that no revision is added until
3921
# all the inventories it may depend on are added.
3922
# Note that this is overzealous, as we may have fetched these in an
3925
revision_ids = set()
3926
for revision in pending_revisions:
3927
revision_ids.add(revision.revision_id)
3928
parent_ids.update(revision.parent_ids)
3929
parent_ids.difference_update(revision_ids)
3930
parent_ids.discard(_mod_revision.NULL_REVISION)
3931
parent_map = self.source.get_parent_map(parent_ids)
3932
# we iterate over parent_map and not parent_ids because we don't
3933
# want to try copying any revision which is a ghost
3934
for parent_tree in self.source.revision_trees(parent_map):
3935
current_revision_id = parent_tree.get_revision_id()
3936
parents_parents = parent_map[current_revision_id]
3937
possible_trees = self._get_trees(parents_parents, cache)
3938
if len(possible_trees) == 0:
3939
# There either aren't any parents, or the parents are
3940
# ghosts, so just use the last converted tree.
3941
possible_trees.append((basis_id, cache[basis_id]))
3942
basis_id, delta = self._get_delta_for_revision(parent_tree,
3943
parents_parents, possible_trees)
3944
self.target.add_inventory_by_delta(
3945
basis_id, delta, current_revision_id, parents_parents)
3946
# insert signatures and revisions
3947
for revision in pending_revisions:
3949
signature = self.source.get_signature_text(
3950
revision.revision_id)
3951
self.target.add_signature_text(revision.revision_id,
3953
except errors.NoSuchRevision:
3955
self.target.add_revision(revision.revision_id, revision)
3958
def _fetch_all_revisions(self, revision_ids, pb):
3959
"""Fetch everything for the list of revisions.
3961
:param revision_ids: The list of revisions to fetch. Must be in
3963
:param pb: A ProgressTask
3966
basis_id, basis_tree = self._get_basis(revision_ids[0])
3968
cache = lru_cache.LRUCache(100)
3969
cache[basis_id] = basis_tree
3970
del basis_tree # We don't want to hang on to it here
3972
if self._converting_to_rich_root and len(revision_ids) > 100:
3973
a_graph = _mod_fetch._get_rich_root_heads_graph(self.source,
3978
for offset in range(0, len(revision_ids), batch_size):
3979
self.target.start_write_group()
3981
pb.update('Transferring revisions', offset,
3983
batch = revision_ids[offset:offset+batch_size]
3984
basis_id = self._fetch_batch(batch, basis_id, cache,
3987
self.target.abort_write_group()
3990
hint = self.target.commit_write_group()
3993
if hints and self.target._format.pack_compresses:
3994
self.target.pack(hint=hints)
3995
pb.update('Transferring revisions', len(revision_ids),
3999
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
4001
"""See InterRepository.fetch()."""
4002
if fetch_spec is not None:
4003
raise AssertionError("Not implemented yet...")
4004
if (not self.source.supports_rich_root()
4005
and self.target.supports_rich_root()):
4006
self._converting_to_rich_root = True
4007
self._revision_id_to_root_id = {}
4009
self._converting_to_rich_root = False
4010
revision_ids = self.target.search_missing_revision_ids(self.source,
4011
revision_id, find_ghosts=find_ghosts).get_keys()
4012
if not revision_ids:
4014
revision_ids = tsort.topo_sort(
4015
self.source.get_graph().get_parent_map(revision_ids))
4016
if not revision_ids:
4018
# Walk though all revisions; get inventory deltas, copy referenced
4019
# texts that delta references, insert the delta, revision and
4022
my_pb = ui.ui_factory.nested_progress_bar()
4025
symbol_versioning.warn(
4026
symbol_versioning.deprecated_in((1, 14, 0))
4027
% "pb parameter to fetch()")
4030
self._fetch_all_revisions(revision_ids, pb)
4032
if my_pb is not None:
4034
return len(revision_ids), 0
4036
def _get_basis(self, first_revision_id):
4037
"""Get a revision and tree which exists in the target.
4039
This assumes that first_revision_id is selected for transmission
4040
because all other ancestors are already present. If we can't find an
4041
ancestor we fall back to NULL_REVISION since we know that is safe.
4043
:return: (basis_id, basis_tree)
4045
first_rev = self.source.get_revision(first_revision_id)
4047
basis_id = first_rev.parent_ids[0]
4048
# only valid as a basis if the target has it
4049
self.target.get_revision(basis_id)
4050
# Try to get a basis tree - if its a ghost it will hit the
4051
# NoSuchRevision case.
4052
basis_tree = self.source.revision_tree(basis_id)
4053
except (IndexError, errors.NoSuchRevision):
4054
basis_id = _mod_revision.NULL_REVISION
4055
basis_tree = self.source.revision_tree(basis_id)
4056
return basis_id, basis_tree
4059
InterRepository.register_optimiser(InterDifferingSerializer)
4060
InterRepository.register_optimiser(InterSameDataRepository)
4061
InterRepository.register_optimiser(InterWeaveRepo)
4062
InterRepository.register_optimiser(InterKnitRepo)
4065
class CopyConverter(object):
4066
"""A repository conversion tool which just performs a copy of the content.
4068
This is slow but quite reliable.
4071
def __init__(self, target_format):
4072
"""Create a CopyConverter.
4074
:param target_format: The format the resulting repository should be.
4076
self.target_format = target_format
4078
def convert(self, repo, pb):
4079
"""Perform the conversion of to_convert, giving feedback via pb.
4081
:param to_convert: The disk object to convert.
4082
:param pb: a progress bar to use for progress information.
4087
# this is only useful with metadir layouts - separated repo content.
4088
# trigger an assertion if not such
4089
repo._format.get_format_string()
4090
self.repo_dir = repo.bzrdir
4091
self.step('Moving repository to repository.backup')
4092
self.repo_dir.transport.move('repository', 'repository.backup')
4093
backup_transport = self.repo_dir.transport.clone('repository.backup')
4094
repo._format.check_conversion_target(self.target_format)
4095
self.source_repo = repo._format.open(self.repo_dir,
4097
_override_transport=backup_transport)
4098
self.step('Creating new repository')
4099
converted = self.target_format.initialize(self.repo_dir,
4100
self.source_repo.is_shared())
4101
converted.lock_write()
4103
self.step('Copying content')
4104
self.source_repo.copy_content_into(converted)
4107
self.step('Deleting old repository content')
4108
self.repo_dir.transport.delete_tree('repository.backup')
4109
ui.ui_factory.note('repository converted')
4111
def step(self, message):
4112
"""Update the pb by a step."""
4114
self.pb.update(message, self.count, self.total)
4126
def _unescaper(match, _map=_unescape_map):
4127
code = match.group(1)
4131
if not code.startswith('#'):
4133
return unichr(int(code[1:])).encode('utf8')
4139
def _unescape_xml(data):
4140
"""Unescape predefined XML entities in a string of data."""
4142
if _unescape_re is None:
4143
_unescape_re = re.compile('\&([^;]*);')
4144
return _unescape_re.sub(_unescaper, data)
4147
class _VersionedFileChecker(object):
4149
def __init__(self, repository, text_key_references=None, ancestors=None):
4150
self.repository = repository
4151
self.text_index = self.repository._generate_text_key_index(
4152
text_key_references=text_key_references, ancestors=ancestors)
4154
def calculate_file_version_parents(self, text_key):
4155
"""Calculate the correct parents for a file version according to
4158
parent_keys = self.text_index[text_key]
4159
if parent_keys == [_mod_revision.NULL_REVISION]:
4161
return tuple(parent_keys)
4163
def check_file_version_parents(self, texts, progress_bar=None):
4164
"""Check the parents stored in a versioned file are correct.
4166
It also detects file versions that are not referenced by their
4167
corresponding revision's inventory.
4169
:returns: A tuple of (wrong_parents, dangling_file_versions).
4170
wrong_parents is a dict mapping {revision_id: (stored_parents,
4171
correct_parents)} for each revision_id where the stored parents
4172
are not correct. dangling_file_versions is a set of (file_id,
4173
revision_id) tuples for versions that are present in this versioned
4174
file, but not used by the corresponding inventory.
4176
local_progress = None
4177
if progress_bar is None:
4178
local_progress = ui.ui_factory.nested_progress_bar()
4179
progress_bar = local_progress
4181
return self._check_file_version_parents(texts, progress_bar)
4184
local_progress.finished()
4186
def _check_file_version_parents(self, texts, progress_bar):
4187
"""See check_file_version_parents."""
4189
self.file_ids = set([file_id for file_id, _ in
4190
self.text_index.iterkeys()])
4191
# text keys is now grouped by file_id
4192
n_versions = len(self.text_index)
4193
progress_bar.update('loading text store', 0, n_versions)
4194
parent_map = self.repository.texts.get_parent_map(self.text_index)
4195
# On unlistable transports this could well be empty/error...
4196
text_keys = self.repository.texts.keys()
4197
unused_keys = frozenset(text_keys) - set(self.text_index)
4198
for num, key in enumerate(self.text_index.iterkeys()):
4199
progress_bar.update('checking text graph', num, n_versions)
4200
correct_parents = self.calculate_file_version_parents(key)
4202
knit_parents = parent_map[key]
4203
except errors.RevisionNotPresent:
4206
if correct_parents != knit_parents:
4207
wrong_parents[key] = (knit_parents, correct_parents)
4208
return wrong_parents, unused_keys
4211
def _old_get_graph(repository, revision_id):
4212
"""DO NOT USE. That is all. I'm serious."""
4213
graph = repository.get_graph()
4214
revision_graph = dict(((key, value) for key, value in
4215
graph.iter_ancestry([revision_id]) if value is not None))
4216
return _strip_NULL_ghosts(revision_graph)
4219
def _strip_NULL_ghosts(revision_graph):
4220
"""Also don't use this. more compatibility code for unmigrated clients."""
4221
# Filter ghosts, and null:
4222
if _mod_revision.NULL_REVISION in revision_graph:
4223
del revision_graph[_mod_revision.NULL_REVISION]
4224
for key, parents in revision_graph.items():
4225
revision_graph[key] = tuple(parent for parent in parents if parent
4227
return revision_graph
4230
class StreamSink(object):
4231
"""An object that can insert a stream into a repository.
4233
This interface handles the complexity of reserialising inventories and
4234
revisions from different formats, and allows unidirectional insertion into
4235
stacked repositories without looking for the missing basis parents
4239
def __init__(self, target_repo):
4240
self.target_repo = target_repo
4242
def insert_stream(self, stream, src_format, resume_tokens):
4243
"""Insert a stream's content into the target repository.
4245
:param src_format: a bzr repository format.
4247
:return: a list of resume tokens and an iterable of keys additional
4248
items required before the insertion can be completed.
4250
self.target_repo.lock_write()
4253
self.target_repo.resume_write_group(resume_tokens)
4256
self.target_repo.start_write_group()
4259
# locked_insert_stream performs a commit|suspend.
4260
return self._locked_insert_stream(stream, src_format, is_resume)
4262
self.target_repo.abort_write_group(suppress_errors=True)
4265
self.target_repo.unlock()
4267
def _locked_insert_stream(self, stream, src_format, is_resume):
4268
to_serializer = self.target_repo._format._serializer
4269
src_serializer = src_format._serializer
4271
if to_serializer == src_serializer:
4272
# If serializers match and the target is a pack repository, set the
4273
# write cache size on the new pack. This avoids poor performance
4274
# on transports where append is unbuffered (such as
4275
# RemoteTransport). This is safe to do because nothing should read
4276
# back from the target repository while a stream with matching
4277
# serialization is being inserted.
4278
# The exception is that a delta record from the source that should
4279
# be a fulltext may need to be expanded by the target (see
4280
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4281
# explicitly flush any buffered writes first in that rare case.
4283
new_pack = self.target_repo._pack_collection._new_pack
4284
except AttributeError:
4285
# Not a pack repository
4288
new_pack.set_write_cache_size(1024*1024)
4289
for substream_type, substream in stream:
4290
if 'stream' in debug.debug_flags:
4291
mutter('inserting substream: %s', substream_type)
4292
if substream_type == 'texts':
4293
self.target_repo.texts.insert_record_stream(substream)
4294
elif substream_type == 'inventories':
4295
if src_serializer == to_serializer:
4296
self.target_repo.inventories.insert_record_stream(
4299
self._extract_and_insert_inventories(
4300
substream, src_serializer)
4301
elif substream_type == 'inventory-deltas':
4302
self._extract_and_insert_inventory_deltas(
4303
substream, src_serializer)
4304
elif substream_type == 'chk_bytes':
4305
# XXX: This doesn't support conversions, as it assumes the
4306
# conversion was done in the fetch code.
4307
self.target_repo.chk_bytes.insert_record_stream(substream)
4308
elif substream_type == 'revisions':
4309
# This may fallback to extract-and-insert more often than
4310
# required if the serializers are different only in terms of
4312
if src_serializer == to_serializer:
4313
self.target_repo.revisions.insert_record_stream(
4316
self._extract_and_insert_revisions(substream,
4318
elif substream_type == 'signatures':
4319
self.target_repo.signatures.insert_record_stream(substream)
4321
raise AssertionError('kaboom! %s' % (substream_type,))
4322
# Done inserting data, and the missing_keys calculations will try to
4323
# read back from the inserted data, so flush the writes to the new pack
4324
# (if this is pack format).
4325
if new_pack is not None:
4326
new_pack._write_data('', flush=True)
4327
# Find all the new revisions (including ones from resume_tokens)
4328
missing_keys = self.target_repo.get_missing_parent_inventories(
4329
check_for_missing_texts=is_resume)
4331
for prefix, versioned_file in (
4332
('texts', self.target_repo.texts),
4333
('inventories', self.target_repo.inventories),
4334
('revisions', self.target_repo.revisions),
4335
('signatures', self.target_repo.signatures),
4336
('chk_bytes', self.target_repo.chk_bytes),
4338
if versioned_file is None:
4340
# TODO: key is often going to be a StaticTuple object
4341
# I don't believe we can define a method by which
4342
# (prefix,) + StaticTuple will work, though we could
4343
# define a StaticTuple.sq_concat that would allow you to
4344
# pass in either a tuple or a StaticTuple as the second
4345
# object, so instead we could have:
4346
# StaticTuple(prefix) + key here...
4347
missing_keys.update((prefix,) + key for key in
4348
versioned_file.get_missing_compression_parent_keys())
4349
except NotImplementedError:
4350
# cannot even attempt suspending, and missing would have failed
4351
# during stream insertion.
4352
missing_keys = set()
4355
# suspend the write group and tell the caller what we is
4356
# missing. We know we can suspend or else we would not have
4357
# entered this code path. (All repositories that can handle
4358
# missing keys can handle suspending a write group).
4359
write_group_tokens = self.target_repo.suspend_write_group()
4360
return write_group_tokens, missing_keys
4361
hint = self.target_repo.commit_write_group()
4362
if (to_serializer != src_serializer and
4363
self.target_repo._format.pack_compresses):
4364
self.target_repo.pack(hint=hint)
4367
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4368
target_rich_root = self.target_repo._format.rich_root_data
4369
target_tree_refs = self.target_repo._format.supports_tree_reference
4370
for record in substream:
4371
# Insert the delta directly
4372
inventory_delta_bytes = record.get_bytes_as('fulltext')
4373
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4375
parse_result = deserialiser.parse_text_bytes(
4376
inventory_delta_bytes)
4377
except inventory_delta.IncompatibleInventoryDelta, err:
4378
trace.mutter("Incompatible delta: %s", err.msg)
4379
raise errors.IncompatibleRevision(self.target_repo._format)
4380
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4381
revision_id = new_id
4382
parents = [key[0] for key in record.parents]
4383
self.target_repo.add_inventory_by_delta(
4384
basis_id, inv_delta, revision_id, parents)
4386
def _extract_and_insert_inventories(self, substream, serializer,
4388
"""Generate a new inventory versionedfile in target, converting data.
4390
The inventory is retrieved from the source, (deserializing it), and
4391
stored in the target (reserializing it in a different format).
4393
target_rich_root = self.target_repo._format.rich_root_data
4394
target_tree_refs = self.target_repo._format.supports_tree_reference
4395
for record in substream:
4396
# It's not a delta, so it must be a fulltext in the source
4397
# serializer's format.
4398
bytes = record.get_bytes_as('fulltext')
4399
revision_id = record.key[0]
4400
inv = serializer.read_inventory_from_string(bytes, revision_id)
4401
parents = [key[0] for key in record.parents]
4402
self.target_repo.add_inventory(revision_id, inv, parents)
4403
# No need to keep holding this full inv in memory when the rest of
4404
# the substream is likely to be all deltas.
4407
def _extract_and_insert_revisions(self, substream, serializer):
4408
for record in substream:
4409
bytes = record.get_bytes_as('fulltext')
4410
revision_id = record.key[0]
4411
rev = serializer.read_revision_from_string(bytes)
4412
if rev.revision_id != revision_id:
4413
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4414
self.target_repo.add_revision(revision_id, rev)
4417
if self.target_repo._format._fetch_reconcile:
4418
self.target_repo.reconcile()
4421
class StreamSource(object):
4422
"""A source of a stream for fetching between repositories."""
4424
def __init__(self, from_repository, to_format):
4425
"""Create a StreamSource streaming from from_repository."""
4426
self.from_repository = from_repository
4427
self.to_format = to_format
4429
def delta_on_metadata(self):
4430
"""Return True if delta's are permitted on metadata streams.
4432
That is on revisions and signatures.
4434
src_serializer = self.from_repository._format._serializer
4435
target_serializer = self.to_format._serializer
4436
return (self.to_format._fetch_uses_deltas and
4437
src_serializer == target_serializer)
4439
def _fetch_revision_texts(self, revs):
4440
# fetch signatures first and then the revision texts
4441
# may need to be a InterRevisionStore call here.
4442
from_sf = self.from_repository.signatures
4443
# A missing signature is just skipped.
4444
keys = [(rev_id,) for rev_id in revs]
4445
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4447
self.to_format._fetch_order,
4448
not self.to_format._fetch_uses_deltas))
4449
# If a revision has a delta, this is actually expanded inside the
4450
# insert_record_stream code now, which is an alternate fix for
4452
from_rf = self.from_repository.revisions
4453
revisions = from_rf.get_record_stream(
4455
self.to_format._fetch_order,
4456
not self.delta_on_metadata())
4457
return [('signatures', signatures), ('revisions', revisions)]
4459
def _generate_root_texts(self, revs):
4460
"""This will be called by get_stream between fetching weave texts and
4461
fetching the inventory weave.
4463
if self._rich_root_upgrade():
4464
return _mod_fetch.Inter1and2Helper(
4465
self.from_repository).generate_root_texts(revs)
4469
def get_stream(self, search):
4471
revs = search.get_keys()
4472
graph = self.from_repository.get_graph()
4473
revs = tsort.topo_sort(graph.get_parent_map(revs))
4474
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4476
for knit_kind, file_id, revisions in data_to_fetch:
4477
if knit_kind != phase:
4479
# Make a new progress bar for this phase
4480
if knit_kind == "file":
4481
# Accumulate file texts
4482
text_keys.extend([(file_id, revision) for revision in
4484
elif knit_kind == "inventory":
4485
# Now copy the file texts.
4486
from_texts = self.from_repository.texts
4487
yield ('texts', from_texts.get_record_stream(
4488
text_keys, self.to_format._fetch_order,
4489
not self.to_format._fetch_uses_deltas))
4490
# Cause an error if a text occurs after we have done the
4493
# Before we process the inventory we generate the root
4494
# texts (if necessary) so that the inventories references
4496
for _ in self._generate_root_texts(revs):
4498
# we fetch only the referenced inventories because we do not
4499
# know for unselected inventories whether all their required
4500
# texts are present in the other repository - it could be
4502
for info in self._get_inventory_stream(revs):
4504
elif knit_kind == "signatures":
4505
# Nothing to do here; this will be taken care of when
4506
# _fetch_revision_texts happens.
4508
elif knit_kind == "revisions":
4509
for record in self._fetch_revision_texts(revs):
4512
raise AssertionError("Unknown knit kind %r" % knit_kind)
4514
def get_stream_for_missing_keys(self, missing_keys):
4515
# missing keys can only occur when we are byte copying and not
4516
# translating (because translation means we don't send
4517
# unreconstructable deltas ever).
4519
keys['texts'] = set()
4520
keys['revisions'] = set()
4521
keys['inventories'] = set()
4522
keys['chk_bytes'] = set()
4523
keys['signatures'] = set()
4524
for key in missing_keys:
4525
keys[key[0]].add(key[1:])
4526
if len(keys['revisions']):
4527
# If we allowed copying revisions at this point, we could end up
4528
# copying a revision without copying its required texts: a
4529
# violation of the requirements for repository integrity.
4530
raise AssertionError(
4531
'cannot copy revisions to fill in missing deltas %s' % (
4532
keys['revisions'],))
4533
for substream_kind, keys in keys.iteritems():
4534
vf = getattr(self.from_repository, substream_kind)
4535
if vf is None and keys:
4536
raise AssertionError(
4537
"cannot fill in keys for a versioned file we don't"
4538
" have: %s needs %s" % (substream_kind, keys))
4540
# No need to stream something we don't have
4542
if substream_kind == 'inventories':
4543
# Some missing keys are genuinely ghosts, filter those out.
4544
present = self.from_repository.inventories.get_parent_map(keys)
4545
revs = [key[0] for key in present]
4546
# Get the inventory stream more-or-less as we do for the
4547
# original stream; there's no reason to assume that records
4548
# direct from the source will be suitable for the sink. (Think
4549
# e.g. 2a -> 1.9-rich-root).
4550
for info in self._get_inventory_stream(revs, missing=True):
4554
# Ask for full texts always so that we don't need more round trips
4555
# after this stream.
4556
# Some of the missing keys are genuinely ghosts, so filter absent
4557
# records. The Sink is responsible for doing another check to
4558
# ensure that ghosts don't introduce missing data for future
4560
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4561
self.to_format._fetch_order, True))
4562
yield substream_kind, stream
4564
def inventory_fetch_order(self):
4565
if self._rich_root_upgrade():
4566
return 'topological'
4568
return self.to_format._fetch_order
4570
def _rich_root_upgrade(self):
4571
return (not self.from_repository._format.rich_root_data and
4572
self.to_format.rich_root_data)
4574
def _get_inventory_stream(self, revision_ids, missing=False):
4575
from_format = self.from_repository._format
4576
if (from_format.supports_chks and self.to_format.supports_chks and
4577
from_format.network_name() == self.to_format.network_name()):
4578
raise AssertionError(
4579
"this case should be handled by GroupCHKStreamSource")
4580
elif 'forceinvdeltas' in debug.debug_flags:
4581
return self._get_convertable_inventory_stream(revision_ids,
4582
delta_versus_null=missing)
4583
elif from_format.network_name() == self.to_format.network_name():
4585
return self._get_simple_inventory_stream(revision_ids,
4587
elif (not from_format.supports_chks and not self.to_format.supports_chks
4588
and from_format._serializer == self.to_format._serializer):
4589
# Essentially the same format.
4590
return self._get_simple_inventory_stream(revision_ids,
4593
# Any time we switch serializations, we want to use an
4594
# inventory-delta based approach.
4595
return self._get_convertable_inventory_stream(revision_ids,
4596
delta_versus_null=missing)
4598
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4599
# NB: This currently reopens the inventory weave in source;
4600
# using a single stream interface instead would avoid this.
4601
from_weave = self.from_repository.inventories
4603
delta_closure = True
4605
delta_closure = not self.delta_on_metadata()
4606
yield ('inventories', from_weave.get_record_stream(
4607
[(rev_id,) for rev_id in revision_ids],
4608
self.inventory_fetch_order(), delta_closure))
4610
def _get_convertable_inventory_stream(self, revision_ids,
4611
delta_versus_null=False):
4612
# The source is using CHKs, but the target either doesn't or it has a
4613
# different serializer. The StreamSink code expects to be able to
4614
# convert on the target, so we need to put bytes-on-the-wire that can
4615
# be converted. That means inventory deltas (if the remote is <1.19,
4616
# RemoteStreamSink will fallback to VFS to insert the deltas).
4617
yield ('inventory-deltas',
4618
self._stream_invs_as_deltas(revision_ids,
4619
delta_versus_null=delta_versus_null))
4621
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4622
"""Return a stream of inventory-deltas for the given rev ids.
4624
:param revision_ids: The list of inventories to transmit
4625
:param delta_versus_null: Don't try to find a minimal delta for this
4626
entry, instead compute the delta versus the NULL_REVISION. This
4627
effectively streams a complete inventory. Used for stuff like
4628
filling in missing parents, etc.
4630
from_repo = self.from_repository
4631
revision_keys = [(rev_id,) for rev_id in revision_ids]
4632
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4633
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4635
inventories = self.from_repository.iter_inventories(
4636
revision_ids, 'topological')
4637
format = from_repo._format
4638
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4639
inventory_cache = lru_cache.LRUCache(50)
4640
null_inventory = from_repo.revision_tree(
4641
_mod_revision.NULL_REVISION).inventory
4642
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4643
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4644
# repo back into a non-rich-root repo ought to be allowed)
4645
serializer = inventory_delta.InventoryDeltaSerializer(
4646
versioned_root=format.rich_root_data,
4647
tree_references=format.supports_tree_reference)
4648
for inv in inventories:
4649
key = (inv.revision_id,)
4650
parent_keys = parent_map.get(key, ())
4652
if not delta_versus_null and parent_keys:
4653
# The caller did not ask for complete inventories and we have
4654
# some parents that we can delta against. Make a delta against
4655
# each parent so that we can find the smallest.
4656
parent_ids = [parent_key[0] for parent_key in parent_keys]
4657
for parent_id in parent_ids:
4658
if parent_id not in invs_sent_so_far:
4659
# We don't know that the remote side has this basis, so
4662
if parent_id == _mod_revision.NULL_REVISION:
4663
parent_inv = null_inventory
4665
parent_inv = inventory_cache.get(parent_id, None)
4666
if parent_inv is None:
4667
parent_inv = from_repo.get_inventory(parent_id)
4668
candidate_delta = inv._make_delta(parent_inv)
4669
if (delta is None or
4670
len(delta) > len(candidate_delta)):
4671
delta = candidate_delta
4672
basis_id = parent_id
4674
# Either none of the parents ended up being suitable, or we
4675
# were asked to delta against NULL
4676
basis_id = _mod_revision.NULL_REVISION
4677
delta = inv._make_delta(null_inventory)
4678
invs_sent_so_far.add(inv.revision_id)
4679
inventory_cache[inv.revision_id] = inv
4680
delta_serialized = ''.join(
4681
serializer.delta_to_lines(basis_id, key[-1], delta))
4682
yield versionedfile.FulltextContentFactory(
4683
key, parent_keys, None, delta_serialized)
4686
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4687
stop_revision=None):
4688
"""Extend the partial history to include a given index
4690
If a stop_index is supplied, stop when that index has been reached.
4691
If a stop_revision is supplied, stop when that revision is
4692
encountered. Otherwise, stop when the beginning of history is
4695
:param stop_index: The index which should be present. When it is
4696
present, history extension will stop.
4697
:param stop_revision: The revision id which should be present. When
4698
it is encountered, history extension will stop.
4700
start_revision = partial_history_cache[-1]
4701
iterator = repo.iter_reverse_revision_history(start_revision)
4703
#skip the last revision in the list
4706
if (stop_index is not None and
4707
len(partial_history_cache) > stop_index):
4709
if partial_history_cache[-1] == stop_revision:
4711
revision_id = iterator.next()
4712
partial_history_cache.append(revision_id)
4713
except StopIteration:
added
removed
bzrlib/repository.py
