1
# Copyright (C) 2005-2010 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,
49
from bzrlib.bundle import serializer
50
from bzrlib.revisiontree import RevisionTree
51
from bzrlib.store.versioned import VersionedFileStore
52
from bzrlib.testament import Testament
55
from bzrlib.decorators import needs_read_lock, needs_write_lock, only_raises
56
from bzrlib.inter import InterObject
57
from bzrlib.inventory import (
63
from bzrlib.lock import _RelockDebugMixin
64
from bzrlib import registry
65
from bzrlib.trace import (
66
log_exception_quietly, note, mutter, mutter_callsite, warning)
69
# Old formats display a warning, but only once
70
_deprecation_warning_done = False
73
class CommitBuilder(object):
74
"""Provides an interface to build up a commit.
76
This allows describing a tree to be committed without needing to
77
know the internals of the format of the repository.
80
# all clients should supply tree roots.
81
record_root_entry = True
82
# the default CommitBuilder does not manage trees whose root is versioned.
83
_versioned_root = False
85
def __init__(self, repository, parents, config, timestamp=None,
86
timezone=None, committer=None, revprops=None,
88
"""Initiate a CommitBuilder.
90
:param repository: Repository to commit to.
91
:param parents: Revision ids of the parents of the new revision.
92
:param config: Configuration to use.
93
:param timestamp: Optional timestamp recorded for commit.
94
:param timezone: Optional timezone for timestamp.
95
:param committer: Optional committer to set for commit.
96
:param revprops: Optional dictionary of revision properties.
97
:param revision_id: Optional revision id.
101
if committer is None:
102
self._committer = self._config.username()
104
self._committer = committer
106
self.new_inventory = Inventory(None)
107
self._new_revision_id = revision_id
108
self.parents = parents
109
self.repository = repository
112
if revprops is not None:
113
self._validate_revprops(revprops)
114
self._revprops.update(revprops)
116
if timestamp is None:
117
timestamp = time.time()
118
# Restrict resolution to 1ms
119
self._timestamp = round(timestamp, 3)
122
self._timezone = osutils.local_time_offset()
124
self._timezone = int(timezone)
126
self._generate_revision_if_needed()
127
self.__heads = graph.HeadsCache(repository.get_graph()).heads
128
self._basis_delta = []
129
# API compatibility, older code that used CommitBuilder did not call
130
# .record_delete(), which means the delta that is computed would not be
131
# valid. Callers that will call record_delete() should call
132
# .will_record_deletes() to indicate that.
133
self._recording_deletes = False
134
# memo'd check for no-op commits.
135
self._any_changes = False
137
def any_changes(self):
138
"""Return True if any entries were changed.
140
This includes merge-only changes. It is the core for the --unchanged
143
:return: True if any changes have occured.
145
return self._any_changes
147
def _validate_unicode_text(self, text, context):
148
"""Verify things like commit messages don't have bogus characters."""
150
raise ValueError('Invalid value for %s: %r' % (context, text))
152
def _validate_revprops(self, revprops):
153
for key, value in revprops.iteritems():
154
# We know that the XML serializers do not round trip '\r'
155
# correctly, so refuse to accept them
156
if not isinstance(value, basestring):
157
raise ValueError('revision property (%s) is not a valid'
158
' (unicode) string: %r' % (key, value))
159
self._validate_unicode_text(value,
160
'revision property (%s)' % (key,))
162
def commit(self, message):
163
"""Make the actual commit.
165
:return: The revision id of the recorded revision.
167
self._validate_unicode_text(message, 'commit message')
168
rev = _mod_revision.Revision(
169
timestamp=self._timestamp,
170
timezone=self._timezone,
171
committer=self._committer,
173
inventory_sha1=self.inv_sha1,
174
revision_id=self._new_revision_id,
175
properties=self._revprops)
176
rev.parent_ids = self.parents
177
self.repository.add_revision(self._new_revision_id, rev,
178
self.new_inventory, self._config)
179
self.repository.commit_write_group()
180
return self._new_revision_id
183
"""Abort the commit that is being built.
185
self.repository.abort_write_group()
187
def revision_tree(self):
188
"""Return the tree that was just committed.
190
After calling commit() this can be called to get a RevisionTree
191
representing the newly committed tree. This is preferred to
192
calling Repository.revision_tree() because that may require
193
deserializing the inventory, while we already have a copy in
196
if self.new_inventory is None:
197
self.new_inventory = self.repository.get_inventory(
198
self._new_revision_id)
199
return RevisionTree(self.repository, self.new_inventory,
200
self._new_revision_id)
202
def finish_inventory(self):
203
"""Tell the builder that the inventory is finished.
205
:return: The inventory id in the repository, which can be used with
206
repository.get_inventory.
208
if self.new_inventory is None:
209
# an inventory delta was accumulated without creating a new
211
basis_id = self.basis_delta_revision
212
# We ignore the 'inventory' returned by add_inventory_by_delta
213
# because self.new_inventory is used to hint to the rest of the
214
# system what code path was taken
215
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
216
basis_id, self._basis_delta, self._new_revision_id,
219
if self.new_inventory.root is None:
220
raise AssertionError('Root entry should be supplied to'
221
' record_entry_contents, as of bzr 0.10.')
222
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
223
self.new_inventory.revision_id = self._new_revision_id
224
self.inv_sha1 = self.repository.add_inventory(
225
self._new_revision_id,
229
return self._new_revision_id
231
def _gen_revision_id(self):
232
"""Return new revision-id."""
233
return generate_ids.gen_revision_id(self._config.username(),
236
def _generate_revision_if_needed(self):
237
"""Create a revision id if None was supplied.
239
If the repository can not support user-specified revision ids
240
they should override this function and raise CannotSetRevisionId
241
if _new_revision_id is not None.
243
:raises: CannotSetRevisionId
245
if self._new_revision_id is None:
246
self._new_revision_id = self._gen_revision_id()
247
self.random_revid = True
249
self.random_revid = False
251
def _heads(self, file_id, revision_ids):
252
"""Calculate the graph heads for revision_ids in the graph of file_id.
254
This can use either a per-file graph or a global revision graph as we
255
have an identity relationship between the two graphs.
257
return self.__heads(revision_ids)
259
def _check_root(self, ie, parent_invs, tree):
260
"""Helper for record_entry_contents.
262
:param ie: An entry being added.
263
:param parent_invs: The inventories of the parent revisions of the
265
:param tree: The tree that is being committed.
267
# In this revision format, root entries have no knit or weave When
268
# serializing out to disk and back in root.revision is always
270
ie.revision = self._new_revision_id
272
def _require_root_change(self, tree):
273
"""Enforce an appropriate root object change.
275
This is called once when record_iter_changes is called, if and only if
276
the root was not in the delta calculated by record_iter_changes.
278
:param tree: The tree which is being committed.
280
# NB: if there are no parents then this method is not called, so no
281
# need to guard on parents having length.
282
entry = entry_factory['directory'](tree.path2id(''), '',
284
entry.revision = self._new_revision_id
285
self._basis_delta.append(('', '', entry.file_id, entry))
287
def _get_delta(self, ie, basis_inv, path):
288
"""Get a delta against the basis inventory for ie."""
289
if ie.file_id not in basis_inv:
291
result = (None, path, ie.file_id, ie)
292
self._basis_delta.append(result)
294
elif ie != basis_inv[ie.file_id]:
296
# TODO: avoid tis id2path call.
297
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
298
self._basis_delta.append(result)
304
def get_basis_delta(self):
305
"""Return the complete inventory delta versus the basis inventory.
307
This has been built up with the calls to record_delete and
308
record_entry_contents. The client must have already called
309
will_record_deletes() to indicate that they will be generating a
312
:return: An inventory delta, suitable for use with apply_delta, or
313
Repository.add_inventory_by_delta, etc.
315
if not self._recording_deletes:
316
raise AssertionError("recording deletes not activated.")
317
return self._basis_delta
319
def record_delete(self, path, file_id):
320
"""Record that a delete occured against a basis tree.
322
This is an optional API - when used it adds items to the basis_delta
323
being accumulated by the commit builder. It cannot be called unless the
324
method will_record_deletes() has been called to inform the builder that
325
a delta is being supplied.
327
:param path: The path of the thing deleted.
328
:param file_id: The file id that was deleted.
330
if not self._recording_deletes:
331
raise AssertionError("recording deletes not activated.")
332
delta = (path, None, file_id, None)
333
self._basis_delta.append(delta)
334
self._any_changes = True
337
def will_record_deletes(self):
338
"""Tell the commit builder that deletes are being notified.
340
This enables the accumulation of an inventory delta; for the resulting
341
commit to be valid, deletes against the basis MUST be recorded via
342
builder.record_delete().
344
self._recording_deletes = True
346
basis_id = self.parents[0]
348
basis_id = _mod_revision.NULL_REVISION
349
self.basis_delta_revision = basis_id
351
def record_entry_contents(self, ie, parent_invs, path, tree,
353
"""Record the content of ie from tree into the commit if needed.
355
Side effect: sets ie.revision when unchanged
357
:param ie: An inventory entry present in the commit.
358
:param parent_invs: The inventories of the parent revisions of the
360
:param path: The path the entry is at in the tree.
361
:param tree: The tree which contains this entry and should be used to
363
:param content_summary: Summary data from the tree about the paths
364
content - stat, length, exec, sha/link target. This is only
365
accessed when the entry has a revision of None - that is when it is
366
a candidate to commit.
367
:return: A tuple (change_delta, version_recorded, fs_hash).
368
change_delta is an inventory_delta change for this entry against
369
the basis tree of the commit, or None if no change occured against
371
version_recorded is True if a new version of the entry has been
372
recorded. For instance, committing a merge where a file was only
373
changed on the other side will return (delta, False).
374
fs_hash is either None, or the hash details for the path (currently
375
a tuple of the contents sha1 and the statvalue returned by
376
tree.get_file_with_stat()).
378
if self.new_inventory.root is None:
379
if ie.parent_id is not None:
380
raise errors.RootMissing()
381
self._check_root(ie, parent_invs, tree)
382
if ie.revision is None:
383
kind = content_summary[0]
385
# ie is carried over from a prior commit
387
# XXX: repository specific check for nested tree support goes here - if
388
# the repo doesn't want nested trees we skip it ?
389
if (kind == 'tree-reference' and
390
not self.repository._format.supports_tree_reference):
391
# mismatch between commit builder logic and repository:
392
# this needs the entry creation pushed down into the builder.
393
raise NotImplementedError('Missing repository subtree support.')
394
self.new_inventory.add(ie)
396
# TODO: slow, take it out of the inner loop.
398
basis_inv = parent_invs[0]
400
basis_inv = Inventory(root_id=None)
402
# ie.revision is always None if the InventoryEntry is considered
403
# for committing. We may record the previous parents revision if the
404
# content is actually unchanged against a sole head.
405
if ie.revision is not None:
406
if not self._versioned_root and path == '':
407
# repositories that do not version the root set the root's
408
# revision to the new commit even when no change occurs (more
409
# specifically, they do not record a revision on the root; and
410
# the rev id is assigned to the root during deserialisation -
411
# this masks when a change may have occurred against the basis.
412
# To match this we always issue a delta, because the revision
413
# of the root will always be changing.
414
if ie.file_id in basis_inv:
415
delta = (basis_inv.id2path(ie.file_id), path,
419
delta = (None, path, ie.file_id, ie)
420
self._basis_delta.append(delta)
421
return delta, False, None
423
# we don't need to commit this, because the caller already
424
# determined that an existing revision of this file is
425
# appropriate. If its not being considered for committing then
426
# it and all its parents to the root must be unaltered so
427
# no-change against the basis.
428
if ie.revision == self._new_revision_id:
429
raise AssertionError("Impossible situation, a skipped "
430
"inventory entry (%r) claims to be modified in this "
431
"commit (%r).", (ie, self._new_revision_id))
432
return None, False, None
433
# XXX: Friction: parent_candidates should return a list not a dict
434
# so that we don't have to walk the inventories again.
435
parent_candiate_entries = ie.parent_candidates(parent_invs)
436
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
438
for inv in parent_invs:
439
if ie.file_id in inv:
440
old_rev = inv[ie.file_id].revision
441
if old_rev in head_set:
442
heads.append(inv[ie.file_id].revision)
443
head_set.remove(inv[ie.file_id].revision)
446
# now we check to see if we need to write a new record to the
448
# We write a new entry unless there is one head to the ancestors, and
449
# the kind-derived content is unchanged.
451
# Cheapest check first: no ancestors, or more the one head in the
452
# ancestors, we write a new node.
456
# There is a single head, look it up for comparison
457
parent_entry = parent_candiate_entries[heads[0]]
458
# if the non-content specific data has changed, we'll be writing a
460
if (parent_entry.parent_id != ie.parent_id or
461
parent_entry.name != ie.name):
463
# now we need to do content specific checks:
465
# if the kind changed the content obviously has
466
if kind != parent_entry.kind:
468
# Stat cache fingerprint feedback for the caller - None as we usually
469
# don't generate one.
472
if content_summary[2] is None:
473
raise ValueError("Files must not have executable = None")
475
# We can't trust a check of the file length because of content
477
if (# if the exec bit has changed we have to store:
478
parent_entry.executable != content_summary[2]):
480
elif parent_entry.text_sha1 == content_summary[3]:
481
# all meta and content is unchanged (using a hash cache
482
# hit to check the sha)
483
ie.revision = parent_entry.revision
484
ie.text_size = parent_entry.text_size
485
ie.text_sha1 = parent_entry.text_sha1
486
ie.executable = parent_entry.executable
487
return self._get_delta(ie, basis_inv, path), False, None
489
# Either there is only a hash change(no hash cache entry,
490
# or same size content change), or there is no change on
492
# Provide the parent's hash to the store layer, so that the
493
# content is unchanged we will not store a new node.
494
nostore_sha = parent_entry.text_sha1
496
# We want to record a new node regardless of the presence or
497
# absence of a content change in the file.
499
ie.executable = content_summary[2]
500
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
502
text = file_obj.read()
506
ie.text_sha1, ie.text_size = self._add_text_to_weave(
507
ie.file_id, text, heads, nostore_sha)
508
# Let the caller know we generated a stat fingerprint.
509
fingerprint = (ie.text_sha1, stat_value)
510
except errors.ExistingContent:
511
# Turns out that the file content was unchanged, and we were
512
# only going to store a new node if it was changed. Carry over
514
ie.revision = parent_entry.revision
515
ie.text_size = parent_entry.text_size
516
ie.text_sha1 = parent_entry.text_sha1
517
ie.executable = parent_entry.executable
518
return self._get_delta(ie, basis_inv, path), False, None
519
elif kind == 'directory':
521
# all data is meta here, nothing specific to directory, so
523
ie.revision = parent_entry.revision
524
return self._get_delta(ie, basis_inv, path), False, None
525
self._add_text_to_weave(ie.file_id, '', heads, None)
526
elif kind == 'symlink':
527
current_link_target = content_summary[3]
529
# symlink target is not generic metadata, check if it has
531
if current_link_target != parent_entry.symlink_target:
534
# unchanged, carry over.
535
ie.revision = parent_entry.revision
536
ie.symlink_target = parent_entry.symlink_target
537
return self._get_delta(ie, basis_inv, path), False, None
538
ie.symlink_target = current_link_target
539
self._add_text_to_weave(ie.file_id, '', heads, None)
540
elif kind == 'tree-reference':
542
if content_summary[3] != parent_entry.reference_revision:
545
# unchanged, carry over.
546
ie.reference_revision = parent_entry.reference_revision
547
ie.revision = parent_entry.revision
548
return self._get_delta(ie, basis_inv, path), False, None
549
ie.reference_revision = content_summary[3]
550
if ie.reference_revision is None:
551
raise AssertionError("invalid content_summary for nested tree: %r"
552
% (content_summary,))
553
self._add_text_to_weave(ie.file_id, '', heads, None)
555
raise NotImplementedError('unknown kind')
556
ie.revision = self._new_revision_id
557
self._any_changes = True
558
return self._get_delta(ie, basis_inv, path), True, fingerprint
560
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
561
_entry_factory=entry_factory):
562
"""Record a new tree via iter_changes.
564
:param tree: The tree to obtain text contents from for changed objects.
565
:param basis_revision_id: The revision id of the tree the iter_changes
566
has been generated against. Currently assumed to be the same
567
as self.parents[0] - if it is not, errors may occur.
568
:param iter_changes: An iter_changes iterator with the changes to apply
569
to basis_revision_id. The iterator must not include any items with
570
a current kind of None - missing items must be either filtered out
571
or errored-on beefore record_iter_changes sees the item.
572
:param _entry_factory: Private method to bind entry_factory locally for
574
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
577
# Create an inventory delta based on deltas between all the parents and
578
# deltas between all the parent inventories. We use inventory delta's
579
# between the inventory objects because iter_changes masks
580
# last-changed-field only changes.
582
# file_id -> change map, change is fileid, paths, changed, versioneds,
583
# parents, names, kinds, executables
585
# {file_id -> revision_id -> inventory entry, for entries in parent
586
# trees that are not parents[0]
590
revtrees = list(self.repository.revision_trees(self.parents))
591
except errors.NoSuchRevision:
592
# one or more ghosts, slow path.
594
for revision_id in self.parents:
596
revtrees.append(self.repository.revision_tree(revision_id))
597
except errors.NoSuchRevision:
599
basis_revision_id = _mod_revision.NULL_REVISION
601
revtrees.append(self.repository.revision_tree(
602
_mod_revision.NULL_REVISION))
603
# The basis inventory from a repository
605
basis_inv = revtrees[0].inventory
607
basis_inv = self.repository.revision_tree(
608
_mod_revision.NULL_REVISION).inventory
609
if len(self.parents) > 0:
610
if basis_revision_id != self.parents[0] and not ghost_basis:
612
"arbitrary basis parents not yet supported with merges")
613
for revtree in revtrees[1:]:
614
for change in revtree.inventory._make_delta(basis_inv):
615
if change[1] is None:
616
# Not present in this parent.
618
if change[2] not in merged_ids:
619
if change[0] is not None:
620
basis_entry = basis_inv[change[2]]
621
merged_ids[change[2]] = [
623
basis_entry.revision,
626
parent_entries[change[2]] = {
628
basis_entry.revision:basis_entry,
630
change[3].revision:change[3],
633
merged_ids[change[2]] = [change[3].revision]
634
parent_entries[change[2]] = {change[3].revision:change[3]}
636
merged_ids[change[2]].append(change[3].revision)
637
parent_entries[change[2]][change[3].revision] = change[3]
640
# Setup the changes from the tree:
641
# changes maps file_id -> (change, [parent revision_ids])
643
for change in iter_changes:
644
# This probably looks up in basis_inv way to much.
645
if change[1][0] is not None:
646
head_candidate = [basis_inv[change[0]].revision]
649
changes[change[0]] = change, merged_ids.get(change[0],
651
unchanged_merged = set(merged_ids) - set(changes)
652
# Extend the changes dict with synthetic changes to record merges of
654
for file_id in unchanged_merged:
655
# Record a merged version of these items that did not change vs the
656
# basis. This can be either identical parallel changes, or a revert
657
# of a specific file after a merge. The recorded content will be
658
# that of the current tree (which is the same as the basis), but
659
# the per-file graph will reflect a merge.
660
# NB:XXX: We are reconstructing path information we had, this
661
# should be preserved instead.
662
# inv delta change: (file_id, (path_in_source, path_in_target),
663
# changed_content, versioned, parent, name, kind,
666
basis_entry = basis_inv[file_id]
667
except errors.NoSuchId:
668
# a change from basis->some_parents but file_id isn't in basis
669
# so was new in the merge, which means it must have changed
670
# from basis -> current, and as it hasn't the add was reverted
671
# by the user. So we discard this change.
675
(basis_inv.id2path(file_id), tree.id2path(file_id)),
677
(basis_entry.parent_id, basis_entry.parent_id),
678
(basis_entry.name, basis_entry.name),
679
(basis_entry.kind, basis_entry.kind),
680
(basis_entry.executable, basis_entry.executable))
681
changes[file_id] = (change, merged_ids[file_id])
682
# changes contains tuples with the change and a set of inventory
683
# candidates for the file.
685
# old_path, new_path, file_id, new_inventory_entry
686
seen_root = False # Is the root in the basis delta?
687
inv_delta = self._basis_delta
688
modified_rev = self._new_revision_id
689
for change, head_candidates in changes.values():
690
if change[3][1]: # versioned in target.
691
# Several things may be happening here:
692
# We may have a fork in the per-file graph
693
# - record a change with the content from tree
694
# We may have a change against < all trees
695
# - carry over the tree that hasn't changed
696
# We may have a change against all trees
697
# - record the change with the content from tree
700
entry = _entry_factory[kind](file_id, change[5][1],
702
head_set = self._heads(change[0], set(head_candidates))
705
for head_candidate in head_candidates:
706
if head_candidate in head_set:
707
heads.append(head_candidate)
708
head_set.remove(head_candidate)
711
# Could be a carry-over situation:
712
parent_entry_revs = parent_entries.get(file_id, None)
713
if parent_entry_revs:
714
parent_entry = parent_entry_revs.get(heads[0], None)
717
if parent_entry is None:
718
# The parent iter_changes was called against is the one
719
# that is the per-file head, so any change is relevant
720
# iter_changes is valid.
721
carry_over_possible = False
723
# could be a carry over situation
724
# A change against the basis may just indicate a merge,
725
# we need to check the content against the source of the
726
# merge to determine if it was changed after the merge
728
if (parent_entry.kind != entry.kind or
729
parent_entry.parent_id != entry.parent_id or
730
parent_entry.name != entry.name):
731
# Metadata common to all entries has changed
732
# against per-file parent
733
carry_over_possible = False
735
carry_over_possible = True
736
# per-type checks for changes against the parent_entry
739
# Cannot be a carry-over situation
740
carry_over_possible = False
741
# Populate the entry in the delta
743
# XXX: There is still a small race here: If someone reverts the content of a file
744
# after iter_changes examines and decides it has changed,
745
# we will unconditionally record a new version even if some
746
# other process reverts it while commit is running (with
747
# the revert happening after iter_changes did it's
750
entry.executable = True
752
entry.executable = False
753
if (carry_over_possible and
754
parent_entry.executable == entry.executable):
755
# Check the file length, content hash after reading
757
nostore_sha = parent_entry.text_sha1
760
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
762
text = file_obj.read()
766
entry.text_sha1, entry.text_size = self._add_text_to_weave(
767
file_id, text, heads, nostore_sha)
768
yield file_id, change[1][1], (entry.text_sha1, stat_value)
769
except errors.ExistingContent:
770
# No content change against a carry_over parent
771
# Perhaps this should also yield a fs hash update?
773
entry.text_size = parent_entry.text_size
774
entry.text_sha1 = parent_entry.text_sha1
775
elif kind == 'symlink':
777
entry.symlink_target = tree.get_symlink_target(file_id)
778
if (carry_over_possible and
779
parent_entry.symlink_target == entry.symlink_target):
782
self._add_text_to_weave(change[0], '', heads, None)
783
elif kind == 'directory':
784
if carry_over_possible:
787
# Nothing to set on the entry.
788
# XXX: split into the Root and nonRoot versions.
789
if change[1][1] != '' or self.repository.supports_rich_root():
790
self._add_text_to_weave(change[0], '', heads, None)
791
elif kind == 'tree-reference':
792
if not self.repository._format.supports_tree_reference:
793
# This isn't quite sane as an error, but we shouldn't
794
# ever see this code path in practice: tree's don't
795
# permit references when the repo doesn't support tree
797
raise errors.UnsupportedOperation(tree.add_reference,
799
reference_revision = tree.get_reference_revision(change[0])
800
entry.reference_revision = reference_revision
801
if (carry_over_possible and
802
parent_entry.reference_revision == reference_revision):
805
self._add_text_to_weave(change[0], '', heads, None)
807
raise AssertionError('unknown kind %r' % kind)
809
entry.revision = modified_rev
811
entry.revision = parent_entry.revision
814
new_path = change[1][1]
815
inv_delta.append((change[1][0], new_path, change[0], entry))
818
self.new_inventory = None
820
# This should perhaps be guarded by a check that the basis we
821
# commit against is the basis for the commit and if not do a delta
823
self._any_changes = True
825
# housekeeping root entry changes do not affect no-change commits.
826
self._require_root_change(tree)
827
self.basis_delta_revision = basis_revision_id
829
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
830
parent_keys = tuple([(file_id, parent) for parent in parents])
831
return self.repository.texts._add_text(
832
(file_id, self._new_revision_id), parent_keys, new_text,
833
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
836
class RootCommitBuilder(CommitBuilder):
837
"""This commitbuilder actually records the root id"""
839
# the root entry gets versioned properly by this builder.
840
_versioned_root = True
842
def _check_root(self, ie, parent_invs, tree):
843
"""Helper for record_entry_contents.
845
:param ie: An entry being added.
846
:param parent_invs: The inventories of the parent revisions of the
848
:param tree: The tree that is being committed.
851
def _require_root_change(self, tree):
852
"""Enforce an appropriate root object change.
854
This is called once when record_iter_changes is called, if and only if
855
the root was not in the delta calculated by record_iter_changes.
857
:param tree: The tree which is being committed.
859
# versioned roots do not change unless the tree found a change.
862
######################################################################
866
class Repository(_RelockDebugMixin):
867
"""Repository holding history for one or more branches.
869
The repository holds and retrieves historical information including
870
revisions and file history. It's normally accessed only by the Branch,
871
which views a particular line of development through that history.
873
The Repository builds on top of some byte storage facilies (the revisions,
874
signatures, inventories, texts and chk_bytes attributes) and a Transport,
875
which respectively provide byte storage and a means to access the (possibly
878
The byte storage facilities are addressed via tuples, which we refer to
879
as 'keys' throughout the code base. Revision_keys, inventory_keys and
880
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
881
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
882
byte string made up of a hash identifier and a hash value.
883
We use this interface because it allows low friction with the underlying
884
code that implements disk indices, network encoding and other parts of
887
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
888
the serialised revisions for the repository. This can be used to obtain
889
revision graph information or to access raw serialised revisions.
890
The result of trying to insert data into the repository via this store
891
is undefined: it should be considered read-only except for implementors
893
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
894
the serialised signatures for the repository. This can be used to
895
obtain access to raw serialised signatures. The result of trying to
896
insert data into the repository via this store is undefined: it should
897
be considered read-only except for implementors of repositories.
898
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
899
the serialised inventories for the repository. This can be used to
900
obtain unserialised inventories. The result of trying to insert data
901
into the repository via this store is undefined: it should be
902
considered read-only except for implementors of repositories.
903
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
904
texts of files and directories for the repository. This can be used to
905
obtain file texts or file graphs. Note that Repository.iter_file_bytes
906
is usually a better interface for accessing file texts.
907
The result of trying to insert data into the repository via this store
908
is undefined: it should be considered read-only except for implementors
910
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
911
any data the repository chooses to store or have indexed by its hash.
912
The result of trying to insert data into the repository via this store
913
is undefined: it should be considered read-only except for implementors
915
:ivar _transport: Transport for file access to repository, typically
916
pointing to .bzr/repository.
919
# What class to use for a CommitBuilder. Often its simpler to change this
920
# in a Repository class subclass rather than to override
921
# get_commit_builder.
922
_commit_builder_class = CommitBuilder
923
# The search regex used by xml based repositories to determine what things
924
# where changed in a single commit.
925
_file_ids_altered_regex = lazy_regex.lazy_compile(
926
r'file_id="(?P<file_id>[^"]+)"'
927
r'.* revision="(?P<revision_id>[^"]+)"'
930
def abort_write_group(self, suppress_errors=False):
931
"""Commit the contents accrued within the current write group.
933
:param suppress_errors: if true, abort_write_group will catch and log
934
unexpected errors that happen during the abort, rather than
935
allowing them to propagate. Defaults to False.
937
:seealso: start_write_group.
939
if self._write_group is not self.get_transaction():
940
# has an unlock or relock occured ?
943
'(suppressed) mismatched lock context and write group. %r, %r',
944
self._write_group, self.get_transaction())
946
raise errors.BzrError(
947
'mismatched lock context and write group. %r, %r' %
948
(self._write_group, self.get_transaction()))
950
self._abort_write_group()
951
except Exception, exc:
952
self._write_group = None
953
if not suppress_errors:
955
mutter('abort_write_group failed')
956
log_exception_quietly()
957
note('bzr: ERROR (ignored): %s', exc)
958
self._write_group = None
960
def _abort_write_group(self):
961
"""Template method for per-repository write group cleanup.
963
This is called during abort before the write group is considered to be
964
finished and should cleanup any internal state accrued during the write
965
group. There is no requirement that data handed to the repository be
966
*not* made available - this is not a rollback - but neither should any
967
attempt be made to ensure that data added is fully commited. Abort is
968
invoked when an error has occured so futher disk or network operations
969
may not be possible or may error and if possible should not be
973
def add_fallback_repository(self, repository):
974
"""Add a repository to use for looking up data not held locally.
976
:param repository: A repository.
978
if not self._format.supports_external_lookups:
979
raise errors.UnstackableRepositoryFormat(self._format, self.base)
981
# This repository will call fallback.unlock() when we transition to
982
# the unlocked state, so we make sure to increment the lock count
983
repository.lock_read()
984
self._check_fallback_repository(repository)
985
self._fallback_repositories.append(repository)
986
self.texts.add_fallback_versioned_files(repository.texts)
987
self.inventories.add_fallback_versioned_files(repository.inventories)
988
self.revisions.add_fallback_versioned_files(repository.revisions)
989
self.signatures.add_fallback_versioned_files(repository.signatures)
990
if self.chk_bytes is not None:
991
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
993
def _check_fallback_repository(self, repository):
994
"""Check that this repository can fallback to repository safely.
996
Raise an error if not.
998
:param repository: A repository to fallback to.
1000
return InterRepository._assert_same_model(self, repository)
1002
def add_inventory(self, revision_id, inv, parents):
1003
"""Add the inventory inv to the repository as revision_id.
1005
:param parents: The revision ids of the parents that revision_id
1006
is known to have and are in the repository already.
1008
:returns: The validator(which is a sha1 digest, though what is sha'd is
1009
repository format specific) of the serialized inventory.
1011
if not self.is_in_write_group():
1012
raise AssertionError("%r not in write group" % (self,))
1013
_mod_revision.check_not_reserved_id(revision_id)
1014
if not (inv.revision_id is None or inv.revision_id == revision_id):
1015
raise AssertionError(
1016
"Mismatch between inventory revision"
1017
" id and insertion revid (%r, %r)"
1018
% (inv.revision_id, revision_id))
1019
if inv.root is None:
1020
raise AssertionError()
1021
return self._add_inventory_checked(revision_id, inv, parents)
1023
def _add_inventory_checked(self, revision_id, inv, parents):
1024
"""Add inv to the repository after checking the inputs.
1026
This function can be overridden to allow different inventory styles.
1028
:seealso: add_inventory, for the contract.
1030
inv_lines = self._serialise_inventory_to_lines(inv)
1031
return self._inventory_add_lines(revision_id, parents,
1032
inv_lines, check_content=False)
1034
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1035
parents, basis_inv=None, propagate_caches=False):
1036
"""Add a new inventory expressed as a delta against another revision.
1038
See the inventory developers documentation for the theory behind
1041
:param basis_revision_id: The inventory id the delta was created
1042
against. (This does not have to be a direct parent.)
1043
:param delta: The inventory delta (see Inventory.apply_delta for
1045
:param new_revision_id: The revision id that the inventory is being
1047
:param parents: The revision ids of the parents that revision_id is
1048
known to have and are in the repository already. These are supplied
1049
for repositories that depend on the inventory graph for revision
1050
graph access, as well as for those that pun ancestry with delta
1052
:param basis_inv: The basis inventory if it is already known,
1054
:param propagate_caches: If True, the caches for this inventory are
1055
copied to and updated for the result if possible.
1057
:returns: (validator, new_inv)
1058
The validator(which is a sha1 digest, though what is sha'd is
1059
repository format specific) of the serialized inventory, and the
1060
resulting inventory.
1062
if not self.is_in_write_group():
1063
raise AssertionError("%r not in write group" % (self,))
1064
_mod_revision.check_not_reserved_id(new_revision_id)
1065
basis_tree = self.revision_tree(basis_revision_id)
1066
basis_tree.lock_read()
1068
# Note that this mutates the inventory of basis_tree, which not all
1069
# inventory implementations may support: A better idiom would be to
1070
# return a new inventory, but as there is no revision tree cache in
1071
# repository this is safe for now - RBC 20081013
1072
if basis_inv is None:
1073
basis_inv = basis_tree.inventory
1074
basis_inv.apply_delta(delta)
1075
basis_inv.revision_id = new_revision_id
1076
return (self.add_inventory(new_revision_id, basis_inv, parents),
1081
def _inventory_add_lines(self, revision_id, parents, lines,
1082
check_content=True):
1083
"""Store lines in inv_vf and return the sha1 of the inventory."""
1084
parents = [(parent,) for parent in parents]
1085
result = self.inventories.add_lines((revision_id,), parents, lines,
1086
check_content=check_content)[0]
1087
self.inventories._access.flush()
1090
def add_revision(self, revision_id, rev, inv=None, config=None):
1091
"""Add rev to the revision store as revision_id.
1093
:param revision_id: the revision id to use.
1094
:param rev: The revision object.
1095
:param inv: The inventory for the revision. if None, it will be looked
1096
up in the inventory storer
1097
:param config: If None no digital signature will be created.
1098
If supplied its signature_needed method will be used
1099
to determine if a signature should be made.
1101
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1103
_mod_revision.check_not_reserved_id(revision_id)
1104
if config is not None and config.signature_needed():
1106
inv = self.get_inventory(revision_id)
1107
plaintext = Testament(rev, inv).as_short_text()
1108
self.store_revision_signature(
1109
gpg.GPGStrategy(config), plaintext, revision_id)
1110
# check inventory present
1111
if not self.inventories.get_parent_map([(revision_id,)]):
1113
raise errors.WeaveRevisionNotPresent(revision_id,
1116
# yes, this is not suitable for adding with ghosts.
1117
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1120
key = (revision_id,)
1121
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1122
self._add_revision(rev)
1124
def _add_revision(self, revision):
1125
text = self._serializer.write_revision_to_string(revision)
1126
key = (revision.revision_id,)
1127
parents = tuple((parent,) for parent in revision.parent_ids)
1128
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1130
def all_revision_ids(self):
1131
"""Returns a list of all the revision ids in the repository.
1133
This is conceptually deprecated because code should generally work on
1134
the graph reachable from a particular revision, and ignore any other
1135
revisions that might be present. There is no direct replacement
1138
if 'evil' in debug.debug_flags:
1139
mutter_callsite(2, "all_revision_ids is linear with history.")
1140
return self._all_revision_ids()
1142
def _all_revision_ids(self):
1143
"""Returns a list of all the revision ids in the repository.
1145
These are in as much topological order as the underlying store can
1148
raise NotImplementedError(self._all_revision_ids)
1150
def break_lock(self):
1151
"""Break a lock if one is present from another instance.
1153
Uses the ui factory to ask for confirmation if the lock may be from
1156
self.control_files.break_lock()
1159
def _eliminate_revisions_not_present(self, revision_ids):
1160
"""Check every revision id in revision_ids to see if we have it.
1162
Returns a set of the present revisions.
1165
graph = self.get_graph()
1166
parent_map = graph.get_parent_map(revision_ids)
1167
# The old API returned a list, should this actually be a set?
1168
return parent_map.keys()
1170
def _check_inventories(self, checker):
1171
"""Check the inventories found from the revision scan.
1173
This is responsible for verifying the sha1 of inventories and
1174
creating a pending_keys set that covers data referenced by inventories.
1176
bar = ui.ui_factory.nested_progress_bar()
1178
self._do_check_inventories(checker, bar)
1182
def _do_check_inventories(self, checker, bar):
1183
"""Helper for _check_inventories."""
1185
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1186
kinds = ['chk_bytes', 'texts']
1187
count = len(checker.pending_keys)
1188
bar.update("inventories", 0, 2)
1189
current_keys = checker.pending_keys
1190
checker.pending_keys = {}
1191
# Accumulate current checks.
1192
for key in current_keys:
1193
if key[0] != 'inventories' and key[0] not in kinds:
1194
checker._report_items.append('unknown key type %r' % (key,))
1195
keys[key[0]].add(key[1:])
1196
if keys['inventories']:
1197
# NB: output order *should* be roughly sorted - topo or
1198
# inverse topo depending on repository - either way decent
1199
# to just delta against. However, pre-CHK formats didn't
1200
# try to optimise inventory layout on disk. As such the
1201
# pre-CHK code path does not use inventory deltas.
1203
for record in self.inventories.check(keys=keys['inventories']):
1204
if record.storage_kind == 'absent':
1205
checker._report_items.append(
1206
'Missing inventory {%s}' % (record.key,))
1208
last_object = self._check_record('inventories', record,
1209
checker, last_object,
1210
current_keys[('inventories',) + record.key])
1211
del keys['inventories']
1214
bar.update("texts", 1)
1215
while (checker.pending_keys or keys['chk_bytes']
1217
# Something to check.
1218
current_keys = checker.pending_keys
1219
checker.pending_keys = {}
1220
# Accumulate current checks.
1221
for key in current_keys:
1222
if key[0] not in kinds:
1223
checker._report_items.append('unknown key type %r' % (key,))
1224
keys[key[0]].add(key[1:])
1225
# Check the outermost kind only - inventories || chk_bytes || texts
1229
for record in getattr(self, kind).check(keys=keys[kind]):
1230
if record.storage_kind == 'absent':
1231
checker._report_items.append(
1232
'Missing %s {%s}' % (kind, record.key,))
1234
last_object = self._check_record(kind, record,
1235
checker, last_object, current_keys[(kind,) + record.key])
1239
def _check_record(self, kind, record, checker, last_object, item_data):
1240
"""Check a single text from this repository."""
1241
if kind == 'inventories':
1242
rev_id = record.key[0]
1243
inv = self._deserialise_inventory(rev_id,
1244
record.get_bytes_as('fulltext'))
1245
if last_object is not None:
1246
delta = inv._make_delta(last_object)
1247
for old_path, path, file_id, ie in delta:
1250
ie.check(checker, rev_id, inv)
1252
for path, ie in inv.iter_entries():
1253
ie.check(checker, rev_id, inv)
1254
if self._format.fast_deltas:
1256
elif kind == 'chk_bytes':
1257
# No code written to check chk_bytes for this repo format.
1258
checker._report_items.append(
1259
'unsupported key type chk_bytes for %s' % (record.key,))
1260
elif kind == 'texts':
1261
self._check_text(record, checker, item_data)
1263
checker._report_items.append(
1264
'unknown key type %s for %s' % (kind, record.key))
1266
def _check_text(self, record, checker, item_data):
1267
"""Check a single text."""
1268
# Check it is extractable.
1269
# TODO: check length.
1270
if record.storage_kind == 'chunked':
1271
chunks = record.get_bytes_as(record.storage_kind)
1272
sha1 = osutils.sha_strings(chunks)
1273
length = sum(map(len, chunks))
1275
content = record.get_bytes_as('fulltext')
1276
sha1 = osutils.sha_string(content)
1277
length = len(content)
1278
if item_data and sha1 != item_data[1]:
1279
checker._report_items.append(
1280
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1281
(record.key, sha1, item_data[1], item_data[2]))
1284
def create(a_bzrdir):
1285
"""Construct the current default format repository in a_bzrdir."""
1286
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1288
def __init__(self, _format, a_bzrdir, control_files):
1289
"""instantiate a Repository.
1291
:param _format: The format of the repository on disk.
1292
:param a_bzrdir: The BzrDir of the repository.
1294
In the future we will have a single api for all stores for
1295
getting file texts, inventories and revisions, then
1296
this construct will accept instances of those things.
1298
super(Repository, self).__init__()
1299
self._format = _format
1300
# the following are part of the public API for Repository:
1301
self.bzrdir = a_bzrdir
1302
self.control_files = control_files
1303
self._transport = control_files._transport
1304
self.base = self._transport.base
1306
self._reconcile_does_inventory_gc = True
1307
self._reconcile_fixes_text_parents = False
1308
self._reconcile_backsup_inventory = True
1309
self._write_group = None
1310
# Additional places to query for data.
1311
self._fallback_repositories = []
1312
# An InventoryEntry cache, used during deserialization
1313
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1314
# Is it safe to return inventory entries directly from the entry cache,
1315
# rather copying them?
1316
self._safe_to_return_from_cache = False
1319
if self._fallback_repositories:
1320
return '%s(%r, fallback_repositories=%r)' % (
1321
self.__class__.__name__,
1323
self._fallback_repositories)
1325
return '%s(%r)' % (self.__class__.__name__,
1328
def _has_same_fallbacks(self, other_repo):
1329
"""Returns true if the repositories have the same fallbacks."""
1330
my_fb = self._fallback_repositories
1331
other_fb = other_repo._fallback_repositories
1332
if len(my_fb) != len(other_fb):
1334
for f, g in zip(my_fb, other_fb):
1335
if not f.has_same_location(g):
1339
def has_same_location(self, other):
1340
"""Returns a boolean indicating if this repository is at the same
1341
location as another repository.
1343
This might return False even when two repository objects are accessing
1344
the same physical repository via different URLs.
1346
if self.__class__ is not other.__class__:
1348
return (self._transport.base == other._transport.base)
1350
def is_in_write_group(self):
1351
"""Return True if there is an open write group.
1353
:seealso: start_write_group.
1355
return self._write_group is not None
1357
def is_locked(self):
1358
return self.control_files.is_locked()
1360
def is_write_locked(self):
1361
"""Return True if this object is write locked."""
1362
return self.is_locked() and self.control_files._lock_mode == 'w'
1364
def lock_write(self, token=None):
1365
"""Lock this repository for writing.
1367
This causes caching within the repository obejct to start accumlating
1368
data during reads, and allows a 'write_group' to be obtained. Write
1369
groups must be used for actual data insertion.
1371
:param token: if this is already locked, then lock_write will fail
1372
unless the token matches the existing lock.
1373
:returns: a token if this instance supports tokens, otherwise None.
1374
:raises TokenLockingNotSupported: when a token is given but this
1375
instance doesn't support using token locks.
1376
:raises MismatchedToken: if the specified token doesn't match the token
1377
of the existing lock.
1378
:seealso: start_write_group.
1380
A token should be passed in if you know that you have locked the object
1381
some other way, and need to synchronise this object's state with that
1384
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1386
locked = self.is_locked()
1387
result = self.control_files.lock_write(token=token)
1389
self._warn_if_deprecated()
1390
self._note_lock('w')
1391
for repo in self._fallback_repositories:
1392
# Writes don't affect fallback repos
1394
self._refresh_data()
1397
def lock_read(self):
1398
locked = self.is_locked()
1399
self.control_files.lock_read()
1401
self._warn_if_deprecated()
1402
self._note_lock('r')
1403
for repo in self._fallback_repositories:
1405
self._refresh_data()
1407
def get_physical_lock_status(self):
1408
return self.control_files.get_physical_lock_status()
1410
def leave_lock_in_place(self):
1411
"""Tell this repository not to release the physical lock when this
1414
If lock_write doesn't return a token, then this method is not supported.
1416
self.control_files.leave_in_place()
1418
def dont_leave_lock_in_place(self):
1419
"""Tell this repository to release the physical lock when this
1420
object is unlocked, even if it didn't originally acquire it.
1422
If lock_write doesn't return a token, then this method is not supported.
1424
self.control_files.dont_leave_in_place()
1427
def gather_stats(self, revid=None, committers=None):
1428
"""Gather statistics from a revision id.
1430
:param revid: The revision id to gather statistics from, if None, then
1431
no revision specific statistics are gathered.
1432
:param committers: Optional parameter controlling whether to grab
1433
a count of committers from the revision specific statistics.
1434
:return: A dictionary of statistics. Currently this contains:
1435
committers: The number of committers if requested.
1436
firstrev: A tuple with timestamp, timezone for the penultimate left
1437
most ancestor of revid, if revid is not the NULL_REVISION.
1438
latestrev: A tuple with timestamp, timezone for revid, if revid is
1439
not the NULL_REVISION.
1440
revisions: The total revision count in the repository.
1441
size: An estimate disk size of the repository in bytes.
1444
if revid and committers:
1445
result['committers'] = 0
1446
if revid and revid != _mod_revision.NULL_REVISION:
1448
all_committers = set()
1449
revisions = self.get_ancestry(revid)
1450
# pop the leading None
1452
first_revision = None
1454
# ignore the revisions in the middle - just grab first and last
1455
revisions = revisions[0], revisions[-1]
1456
for revision in self.get_revisions(revisions):
1457
if not first_revision:
1458
first_revision = revision
1460
all_committers.add(revision.committer)
1461
last_revision = revision
1463
result['committers'] = len(all_committers)
1464
result['firstrev'] = (first_revision.timestamp,
1465
first_revision.timezone)
1466
result['latestrev'] = (last_revision.timestamp,
1467
last_revision.timezone)
1469
# now gather global repository information
1470
# XXX: This is available for many repos regardless of listability.
1471
if self.bzrdir.root_transport.listable():
1472
# XXX: do we want to __define len__() ?
1473
# Maybe the versionedfiles object should provide a different
1474
# method to get the number of keys.
1475
result['revisions'] = len(self.revisions.keys())
1476
# result['size'] = t
1479
def find_branches(self, using=False):
1480
"""Find branches underneath this repository.
1482
This will include branches inside other branches.
1484
:param using: If True, list only branches using this repository.
1486
if using and not self.is_shared():
1487
return self.bzrdir.list_branches()
1488
class Evaluator(object):
1491
self.first_call = True
1493
def __call__(self, bzrdir):
1494
# On the first call, the parameter is always the bzrdir
1495
# containing the current repo.
1496
if not self.first_call:
1498
repository = bzrdir.open_repository()
1499
except errors.NoRepositoryPresent:
1502
return False, ([], repository)
1503
self.first_call = False
1504
value = (bzrdir.list_branches(), None)
1508
for branches, repository in bzrdir.BzrDir.find_bzrdirs(
1509
self.bzrdir.root_transport, evaluate=Evaluator()):
1510
if branches is not None:
1511
ret.extend(branches)
1512
if not using and repository is not None:
1513
ret.extend(repository.find_branches())
1517
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
1518
"""Return the revision ids that other has that this does not.
1520
These are returned in topological order.
1522
revision_id: only return revision ids included by revision_id.
1524
return InterRepository.get(other, self).search_missing_revision_ids(
1525
revision_id, find_ghosts)
1529
"""Open the repository rooted at base.
1531
For instance, if the repository is at URL/.bzr/repository,
1532
Repository.open(URL) -> a Repository instance.
1534
control = bzrdir.BzrDir.open(base)
1535
return control.open_repository()
1537
def copy_content_into(self, destination, revision_id=None):
1538
"""Make a complete copy of the content in self into destination.
1540
This is a destructive operation! Do not use it on existing
1543
return InterRepository.get(self, destination).copy_content(revision_id)
1545
def commit_write_group(self):
1546
"""Commit the contents accrued within the current write group.
1548
:seealso: start_write_group.
1550
:return: it may return an opaque hint that can be passed to 'pack'.
1552
if self._write_group is not self.get_transaction():
1553
# has an unlock or relock occured ?
1554
raise errors.BzrError('mismatched lock context %r and '
1556
(self.get_transaction(), self._write_group))
1557
result = self._commit_write_group()
1558
self._write_group = None
1561
def _commit_write_group(self):
1562
"""Template method for per-repository write group cleanup.
1564
This is called before the write group is considered to be
1565
finished and should ensure that all data handed to the repository
1566
for writing during the write group is safely committed (to the
1567
extent possible considering file system caching etc).
1570
def suspend_write_group(self):
1571
raise errors.UnsuspendableWriteGroup(self)
1573
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1574
"""Return the keys of missing inventory parents for revisions added in
1577
A revision is not complete if the inventory delta for that revision
1578
cannot be calculated. Therefore if the parent inventories of a
1579
revision are not present, the revision is incomplete, and e.g. cannot
1580
be streamed by a smart server. This method finds missing inventory
1581
parents for revisions added in this write group.
1583
if not self._format.supports_external_lookups:
1584
# This is only an issue for stacked repositories
1586
if not self.is_in_write_group():
1587
raise AssertionError('not in a write group')
1589
# XXX: We assume that every added revision already has its
1590
# corresponding inventory, so we only check for parent inventories that
1591
# might be missing, rather than all inventories.
1592
parents = set(self.revisions._index.get_missing_parents())
1593
parents.discard(_mod_revision.NULL_REVISION)
1594
unstacked_inventories = self.inventories._index
1595
present_inventories = unstacked_inventories.get_parent_map(
1596
key[-1:] for key in parents)
1597
parents.difference_update(present_inventories)
1598
if len(parents) == 0:
1599
# No missing parent inventories.
1601
if not check_for_missing_texts:
1602
return set(('inventories', rev_id) for (rev_id,) in parents)
1603
# Ok, now we have a list of missing inventories. But these only matter
1604
# if the inventories that reference them are missing some texts they
1605
# appear to introduce.
1606
# XXX: Texts referenced by all added inventories need to be present,
1607
# but at the moment we're only checking for texts referenced by
1608
# inventories at the graph's edge.
1609
key_deps = self.revisions._index._key_dependencies
1610
key_deps.satisfy_refs_for_keys(present_inventories)
1611
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1612
file_ids = self.fileids_altered_by_revision_ids(referrers)
1613
missing_texts = set()
1614
for file_id, version_ids in file_ids.iteritems():
1615
missing_texts.update(
1616
(file_id, version_id) for version_id in version_ids)
1617
present_texts = self.texts.get_parent_map(missing_texts)
1618
missing_texts.difference_update(present_texts)
1619
if not missing_texts:
1620
# No texts are missing, so all revisions and their deltas are
1623
# Alternatively the text versions could be returned as the missing
1624
# keys, but this is likely to be less data.
1625
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1628
def refresh_data(self):
1629
"""Re-read any data needed to to synchronise with disk.
1631
This method is intended to be called after another repository instance
1632
(such as one used by a smart server) has inserted data into the
1633
repository. It may not be called during a write group, but may be
1634
called at any other time.
1636
if self.is_in_write_group():
1637
raise errors.InternalBzrError(
1638
"May not refresh_data while in a write group.")
1639
self._refresh_data()
1641
def resume_write_group(self, tokens):
1642
if not self.is_write_locked():
1643
raise errors.NotWriteLocked(self)
1644
if self._write_group:
1645
raise errors.BzrError('already in a write group')
1646
self._resume_write_group(tokens)
1647
# so we can detect unlock/relock - the write group is now entered.
1648
self._write_group = self.get_transaction()
1650
def _resume_write_group(self, tokens):
1651
raise errors.UnsuspendableWriteGroup(self)
1653
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1655
"""Fetch the content required to construct revision_id from source.
1657
If revision_id is None and fetch_spec is None, then all content is
1660
fetch() may not be used when the repository is in a write group -
1661
either finish the current write group before using fetch, or use
1662
fetch before starting the write group.
1664
:param find_ghosts: Find and copy revisions in the source that are
1665
ghosts in the target (and not reachable directly by walking out to
1666
the first-present revision in target from revision_id).
1667
:param revision_id: If specified, all the content needed for this
1668
revision ID will be copied to the target. Fetch will determine for
1669
itself which content needs to be copied.
1670
:param fetch_spec: If specified, a SearchResult or
1671
PendingAncestryResult that describes which revisions to copy. This
1672
allows copying multiple heads at once. Mutually exclusive with
1675
if fetch_spec is not None and revision_id is not None:
1676
raise AssertionError(
1677
"fetch_spec and revision_id are mutually exclusive.")
1678
if self.is_in_write_group():
1679
raise errors.InternalBzrError(
1680
"May not fetch while in a write group.")
1681
# fast path same-url fetch operations
1682
# TODO: lift out to somewhere common with RemoteRepository
1683
# <https://bugs.edge.launchpad.net/bzr/+bug/401646>
1684
if (self.has_same_location(source)
1685
and fetch_spec is None
1686
and self._has_same_fallbacks(source)):
1687
# check that last_revision is in 'from' and then return a
1689
if (revision_id is not None and
1690
not _mod_revision.is_null(revision_id)):
1691
self.get_revision(revision_id)
1693
# if there is no specific appropriate InterRepository, this will get
1694
# the InterRepository base class, which raises an
1695
# IncompatibleRepositories when asked to fetch.
1696
inter = InterRepository.get(source, self)
1697
return inter.fetch(revision_id=revision_id, pb=pb,
1698
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1700
def create_bundle(self, target, base, fileobj, format=None):
1701
return serializer.write_bundle(self, target, base, fileobj, format)
1703
def get_commit_builder(self, branch, parents, config, timestamp=None,
1704
timezone=None, committer=None, revprops=None,
1706
"""Obtain a CommitBuilder for this repository.
1708
:param branch: Branch to commit to.
1709
:param parents: Revision ids of the parents of the new revision.
1710
:param config: Configuration to use.
1711
:param timestamp: Optional timestamp recorded for commit.
1712
:param timezone: Optional timezone for timestamp.
1713
:param committer: Optional committer to set for commit.
1714
:param revprops: Optional dictionary of revision properties.
1715
:param revision_id: Optional revision id.
1717
if self._fallback_repositories:
1718
raise errors.BzrError("Cannot commit from a lightweight checkout "
1719
"to a stacked branch. See "
1720
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1721
result = self._commit_builder_class(self, parents, config,
1722
timestamp, timezone, committer, revprops, revision_id)
1723
self.start_write_group()
1726
@only_raises(errors.LockNotHeld, errors.LockBroken)
1728
if (self.control_files._lock_count == 1 and
1729
self.control_files._lock_mode == 'w'):
1730
if self._write_group is not None:
1731
self.abort_write_group()
1732
self.control_files.unlock()
1733
raise errors.BzrError(
1734
'Must end write groups before releasing write locks.')
1735
self.control_files.unlock()
1736
if self.control_files._lock_count == 0:
1737
self._inventory_entry_cache.clear()
1738
for repo in self._fallback_repositories:
1742
def clone(self, a_bzrdir, revision_id=None):
1743
"""Clone this repository into a_bzrdir using the current format.
1745
Currently no check is made that the format of this repository and
1746
the bzrdir format are compatible. FIXME RBC 20060201.
1748
:return: The newly created destination repository.
1750
# TODO: deprecate after 0.16; cloning this with all its settings is
1751
# probably not very useful -- mbp 20070423
1752
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1753
self.copy_content_into(dest_repo, revision_id)
1756
def start_write_group(self):
1757
"""Start a write group in the repository.
1759
Write groups are used by repositories which do not have a 1:1 mapping
1760
between file ids and backend store to manage the insertion of data from
1761
both fetch and commit operations.
1763
A write lock is required around the start_write_group/commit_write_group
1764
for the support of lock-requiring repository formats.
1766
One can only insert data into a repository inside a write group.
1770
if not self.is_write_locked():
1771
raise errors.NotWriteLocked(self)
1772
if self._write_group:
1773
raise errors.BzrError('already in a write group')
1774
self._start_write_group()
1775
# so we can detect unlock/relock - the write group is now entered.
1776
self._write_group = self.get_transaction()
1778
def _start_write_group(self):
1779
"""Template method for per-repository write group startup.
1781
This is called before the write group is considered to be
1786
def sprout(self, to_bzrdir, revision_id=None):
1787
"""Create a descendent repository for new development.
1789
Unlike clone, this does not copy the settings of the repository.
1791
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1792
dest_repo.fetch(self, revision_id=revision_id)
1795
def _create_sprouting_repo(self, a_bzrdir, shared):
1796
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1797
# use target default format.
1798
dest_repo = a_bzrdir.create_repository()
1800
# Most control formats need the repository to be specifically
1801
# created, but on some old all-in-one formats it's not needed
1803
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1804
except errors.UninitializableFormat:
1805
dest_repo = a_bzrdir.open_repository()
1808
def _get_sink(self):
1809
"""Return a sink for streaming into this repository."""
1810
return StreamSink(self)
1812
def _get_source(self, to_format):
1813
"""Return a source for streaming from this repository."""
1814
return StreamSource(self, to_format)
1817
def has_revision(self, revision_id):
1818
"""True if this repository has a copy of the revision."""
1819
return revision_id in self.has_revisions((revision_id,))
1822
def has_revisions(self, revision_ids):
1823
"""Probe to find out the presence of multiple revisions.
1825
:param revision_ids: An iterable of revision_ids.
1826
:return: A set of the revision_ids that were present.
1828
parent_map = self.revisions.get_parent_map(
1829
[(rev_id,) for rev_id in revision_ids])
1831
if _mod_revision.NULL_REVISION in revision_ids:
1832
result.add(_mod_revision.NULL_REVISION)
1833
result.update([key[0] for key in parent_map])
1837
def get_revision(self, revision_id):
1838
"""Return the Revision object for a named revision."""
1839
return self.get_revisions([revision_id])[0]
1842
def get_revision_reconcile(self, revision_id):
1843
"""'reconcile' helper routine that allows access to a revision always.
1845
This variant of get_revision does not cross check the weave graph
1846
against the revision one as get_revision does: but it should only
1847
be used by reconcile, or reconcile-alike commands that are correcting
1848
or testing the revision graph.
1850
return self._get_revisions([revision_id])[0]
1853
def get_revisions(self, revision_ids):
1854
"""Get many revisions at once.
1856
Repositories that need to check data on every revision read should
1857
subclass this method.
1859
return self._get_revisions(revision_ids)
1862
def _get_revisions(self, revision_ids):
1863
"""Core work logic to get many revisions without sanity checks."""
1865
for revid, rev in self._iter_revisions(revision_ids):
1867
raise errors.NoSuchRevision(self, revid)
1869
return [revs[revid] for revid in revision_ids]
1871
def _iter_revisions(self, revision_ids):
1872
"""Iterate over revision objects.
1874
:param revision_ids: An iterable of revisions to examine. None may be
1875
passed to request all revisions known to the repository. Note that
1876
not all repositories can find unreferenced revisions; for those
1877
repositories only referenced ones will be returned.
1878
:return: An iterator of (revid, revision) tuples. Absent revisions (
1879
those asked for but not available) are returned as (revid, None).
1881
if revision_ids is None:
1882
revision_ids = self.all_revision_ids()
1884
for rev_id in revision_ids:
1885
if not rev_id or not isinstance(rev_id, basestring):
1886
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1887
keys = [(key,) for key in revision_ids]
1888
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1889
for record in stream:
1890
revid = record.key[0]
1891
if record.storage_kind == 'absent':
1894
text = record.get_bytes_as('fulltext')
1895
rev = self._serializer.read_revision_from_string(text)
1899
def get_revision_xml(self, revision_id):
1900
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1901
# would have already do it.
1902
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1903
# TODO: this can't just be replaced by:
1904
# return self._serializer.write_revision_to_string(
1905
# self.get_revision(revision_id))
1906
# as cStringIO preservers the encoding unlike write_revision_to_string
1907
# or some other call down the path.
1908
rev = self.get_revision(revision_id)
1909
rev_tmp = cStringIO.StringIO()
1910
# the current serializer..
1911
self._serializer.write_revision(rev, rev_tmp)
1913
return rev_tmp.getvalue()
1915
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1916
"""Produce a generator of revision deltas.
1918
Note that the input is a sequence of REVISIONS, not revision_ids.
1919
Trees will be held in memory until the generator exits.
1920
Each delta is relative to the revision's lefthand predecessor.
1922
:param specific_fileids: if not None, the result is filtered
1923
so that only those file-ids, their parents and their
1924
children are included.
1926
# Get the revision-ids of interest
1927
required_trees = set()
1928
for revision in revisions:
1929
required_trees.add(revision.revision_id)
1930
required_trees.update(revision.parent_ids[:1])
1932
# Get the matching filtered trees. Note that it's more
1933
# efficient to pass filtered trees to changes_from() rather
1934
# than doing the filtering afterwards. changes_from() could
1935
# arguably do the filtering itself but it's path-based, not
1936
# file-id based, so filtering before or afterwards is
1938
if specific_fileids is None:
1939
trees = dict((t.get_revision_id(), t) for
1940
t in self.revision_trees(required_trees))
1942
trees = dict((t.get_revision_id(), t) for
1943
t in self._filtered_revision_trees(required_trees,
1946
# Calculate the deltas
1947
for revision in revisions:
1948
if not revision.parent_ids:
1949
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1951
old_tree = trees[revision.parent_ids[0]]
1952
yield trees[revision.revision_id].changes_from(old_tree)
1955
def get_revision_delta(self, revision_id, specific_fileids=None):
1956
"""Return the delta for one revision.
1958
The delta is relative to the left-hand predecessor of the
1961
:param specific_fileids: if not None, the result is filtered
1962
so that only those file-ids, their parents and their
1963
children are included.
1965
r = self.get_revision(revision_id)
1966
return list(self.get_deltas_for_revisions([r],
1967
specific_fileids=specific_fileids))[0]
1970
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1971
signature = gpg_strategy.sign(plaintext)
1972
self.add_signature_text(revision_id, signature)
1975
def add_signature_text(self, revision_id, signature):
1976
self.signatures.add_lines((revision_id,), (),
1977
osutils.split_lines(signature))
1979
def find_text_key_references(self):
1980
"""Find the text key references within the repository.
1982
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1983
to whether they were referred to by the inventory of the
1984
revision_id that they contain. The inventory texts from all present
1985
revision ids are assessed to generate this report.
1987
revision_keys = self.revisions.keys()
1988
w = self.inventories
1989
pb = ui.ui_factory.nested_progress_bar()
1991
return self._find_text_key_references_from_xml_inventory_lines(
1992
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1996
def _find_text_key_references_from_xml_inventory_lines(self,
1998
"""Core routine for extracting references to texts from inventories.
2000
This performs the translation of xml lines to revision ids.
2002
:param line_iterator: An iterator of lines, origin_version_id
2003
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2004
to whether they were referred to by the inventory of the
2005
revision_id that they contain. Note that if that revision_id was
2006
not part of the line_iterator's output then False will be given -
2007
even though it may actually refer to that key.
2009
if not self._serializer.support_altered_by_hack:
2010
raise AssertionError(
2011
"_find_text_key_references_from_xml_inventory_lines only "
2012
"supported for branches which store inventory as unnested xml"
2013
", not on %r" % self)
2016
# this code needs to read every new line in every inventory for the
2017
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2018
# not present in one of those inventories is unnecessary but not
2019
# harmful because we are filtering by the revision id marker in the
2020
# inventory lines : we only select file ids altered in one of those
2021
# revisions. We don't need to see all lines in the inventory because
2022
# only those added in an inventory in rev X can contain a revision=X
2024
unescape_revid_cache = {}
2025
unescape_fileid_cache = {}
2027
# jam 20061218 In a big fetch, this handles hundreds of thousands
2028
# of lines, so it has had a lot of inlining and optimizing done.
2029
# Sorry that it is a little bit messy.
2030
# Move several functions to be local variables, since this is a long
2032
search = self._file_ids_altered_regex.search
2033
unescape = _unescape_xml
2034
setdefault = result.setdefault
2035
for line, line_key in line_iterator:
2036
match = search(line)
2039
# One call to match.group() returning multiple items is quite a
2040
# bit faster than 2 calls to match.group() each returning 1
2041
file_id, revision_id = match.group('file_id', 'revision_id')
2043
# Inlining the cache lookups helps a lot when you make 170,000
2044
# lines and 350k ids, versus 8.4 unique ids.
2045
# Using a cache helps in 2 ways:
2046
# 1) Avoids unnecessary decoding calls
2047
# 2) Re-uses cached strings, which helps in future set and
2049
# (2) is enough that removing encoding entirely along with
2050
# the cache (so we are using plain strings) results in no
2051
# performance improvement.
2053
revision_id = unescape_revid_cache[revision_id]
2055
unescaped = unescape(revision_id)
2056
unescape_revid_cache[revision_id] = unescaped
2057
revision_id = unescaped
2059
# Note that unconditionally unescaping means that we deserialise
2060
# every fileid, which for general 'pull' is not great, but we don't
2061
# really want to have some many fulltexts that this matters anyway.
2064
file_id = unescape_fileid_cache[file_id]
2066
unescaped = unescape(file_id)
2067
unescape_fileid_cache[file_id] = unescaped
2070
key = (file_id, revision_id)
2071
setdefault(key, False)
2072
if revision_id == line_key[-1]:
2076
def _inventory_xml_lines_for_keys(self, keys):
2077
"""Get a line iterator of the sort needed for findind references.
2079
Not relevant for non-xml inventory repositories.
2081
Ghosts in revision_keys are ignored.
2083
:param revision_keys: The revision keys for the inventories to inspect.
2084
:return: An iterator over (inventory line, revid) for the fulltexts of
2085
all of the xml inventories specified by revision_keys.
2087
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2088
for record in stream:
2089
if record.storage_kind != 'absent':
2090
chunks = record.get_bytes_as('chunked')
2091
revid = record.key[-1]
2092
lines = osutils.chunks_to_lines(chunks)
2096
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2098
"""Helper routine for fileids_altered_by_revision_ids.
2100
This performs the translation of xml lines to revision ids.
2102
:param line_iterator: An iterator of lines, origin_version_id
2103
:param revision_keys: The revision ids to filter for. This should be a
2104
set or other type which supports efficient __contains__ lookups, as
2105
the revision key from each parsed line will be looked up in the
2106
revision_keys filter.
2107
:return: a dictionary mapping altered file-ids to an iterable of
2108
revision_ids. Each altered file-ids has the exact revision_ids that
2109
altered it listed explicitly.
2111
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2112
line_iterator).iterkeys())
2113
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2114
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2115
self._inventory_xml_lines_for_keys(parent_keys)))
2116
new_keys = seen - parent_seen
2118
setdefault = result.setdefault
2119
for key in new_keys:
2120
setdefault(key[0], set()).add(key[-1])
2123
def _find_parent_ids_of_revisions(self, revision_ids):
2124
"""Find all parent ids that are mentioned in the revision graph.
2126
:return: set of revisions that are parents of revision_ids which are
2127
not part of revision_ids themselves
2129
parent_map = self.get_parent_map(revision_ids)
2131
map(parent_ids.update, parent_map.itervalues())
2132
parent_ids.difference_update(revision_ids)
2133
parent_ids.discard(_mod_revision.NULL_REVISION)
2136
def _find_parent_keys_of_revisions(self, revision_keys):
2137
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2139
:param revision_keys: An iterable of revision_keys.
2140
:return: The parents of all revision_keys that are not already in
2143
parent_map = self.revisions.get_parent_map(revision_keys)
2145
map(parent_keys.update, parent_map.itervalues())
2146
parent_keys.difference_update(revision_keys)
2147
parent_keys.discard(_mod_revision.NULL_REVISION)
2150
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2151
"""Find the file ids and versions affected by revisions.
2153
:param revisions: an iterable containing revision ids.
2154
:param _inv_weave: The inventory weave from this repository or None.
2155
If None, the inventory weave will be opened automatically.
2156
:return: a dictionary mapping altered file-ids to an iterable of
2157
revision_ids. Each altered file-ids has the exact revision_ids that
2158
altered it listed explicitly.
2160
selected_keys = set((revid,) for revid in revision_ids)
2161
w = _inv_weave or self.inventories
2162
return self._find_file_ids_from_xml_inventory_lines(
2163
w.iter_lines_added_or_present_in_keys(
2164
selected_keys, pb=None),
2167
def iter_files_bytes(self, desired_files):
2168
"""Iterate through file versions.
2170
Files will not necessarily be returned in the order they occur in
2171
desired_files. No specific order is guaranteed.
2173
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2174
value supplied by the caller as part of desired_files. It should
2175
uniquely identify the file version in the caller's context. (Examples:
2176
an index number or a TreeTransform trans_id.)
2178
bytes_iterator is an iterable of bytestrings for the file. The
2179
kind of iterable and length of the bytestrings are unspecified, but for
2180
this implementation, it is a list of bytes produced by
2181
VersionedFile.get_record_stream().
2183
:param desired_files: a list of (file_id, revision_id, identifier)
2187
for file_id, revision_id, callable_data in desired_files:
2188
text_keys[(file_id, revision_id)] = callable_data
2189
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2190
if record.storage_kind == 'absent':
2191
raise errors.RevisionNotPresent(record.key, self)
2192
yield text_keys[record.key], record.get_bytes_as('chunked')
2194
def _generate_text_key_index(self, text_key_references=None,
2196
"""Generate a new text key index for the repository.
2198
This is an expensive function that will take considerable time to run.
2200
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2201
list of parents, also text keys. When a given key has no parents,
2202
the parents list will be [NULL_REVISION].
2204
# All revisions, to find inventory parents.
2205
if ancestors is None:
2206
graph = self.get_graph()
2207
ancestors = graph.get_parent_map(self.all_revision_ids())
2208
if text_key_references is None:
2209
text_key_references = self.find_text_key_references()
2210
pb = ui.ui_factory.nested_progress_bar()
2212
return self._do_generate_text_key_index(ancestors,
2213
text_key_references, pb)
2217
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2218
"""Helper for _generate_text_key_index to avoid deep nesting."""
2219
revision_order = tsort.topo_sort(ancestors)
2220
invalid_keys = set()
2222
for revision_id in revision_order:
2223
revision_keys[revision_id] = set()
2224
text_count = len(text_key_references)
2225
# a cache of the text keys to allow reuse; costs a dict of all the
2226
# keys, but saves a 2-tuple for every child of a given key.
2228
for text_key, valid in text_key_references.iteritems():
2230
invalid_keys.add(text_key)
2232
revision_keys[text_key[1]].add(text_key)
2233
text_key_cache[text_key] = text_key
2234
del text_key_references
2236
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2237
NULL_REVISION = _mod_revision.NULL_REVISION
2238
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2239
# too small for large or very branchy trees. However, for 55K path
2240
# trees, it would be easy to use too much memory trivially. Ideally we
2241
# could gauge this by looking at available real memory etc, but this is
2242
# always a tricky proposition.
2243
inventory_cache = lru_cache.LRUCache(10)
2244
batch_size = 10 # should be ~150MB on a 55K path tree
2245
batch_count = len(revision_order) / batch_size + 1
2247
pb.update("Calculating text parents", processed_texts, text_count)
2248
for offset in xrange(batch_count):
2249
to_query = revision_order[offset * batch_size:(offset + 1) *
2253
for revision_id in to_query:
2254
parent_ids = ancestors[revision_id]
2255
for text_key in revision_keys[revision_id]:
2256
pb.update("Calculating text parents", processed_texts)
2257
processed_texts += 1
2258
candidate_parents = []
2259
for parent_id in parent_ids:
2260
parent_text_key = (text_key[0], parent_id)
2262
check_parent = parent_text_key not in \
2263
revision_keys[parent_id]
2265
# the parent parent_id is a ghost:
2266
check_parent = False
2267
# truncate the derived graph against this ghost.
2268
parent_text_key = None
2270
# look at the parent commit details inventories to
2271
# determine possible candidates in the per file graph.
2274
inv = inventory_cache[parent_id]
2276
inv = self.revision_tree(parent_id).inventory
2277
inventory_cache[parent_id] = inv
2279
parent_entry = inv[text_key[0]]
2280
except (KeyError, errors.NoSuchId):
2282
if parent_entry is not None:
2284
text_key[0], parent_entry.revision)
2286
parent_text_key = None
2287
if parent_text_key is not None:
2288
candidate_parents.append(
2289
text_key_cache[parent_text_key])
2290
parent_heads = text_graph.heads(candidate_parents)
2291
new_parents = list(parent_heads)
2292
new_parents.sort(key=lambda x:candidate_parents.index(x))
2293
if new_parents == []:
2294
new_parents = [NULL_REVISION]
2295
text_index[text_key] = new_parents
2297
for text_key in invalid_keys:
2298
text_index[text_key] = [NULL_REVISION]
2301
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2302
"""Get an iterable listing the keys of all the data introduced by a set
2305
The keys will be ordered so that the corresponding items can be safely
2306
fetched and inserted in that order.
2308
:returns: An iterable producing tuples of (knit-kind, file-id,
2309
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2310
'revisions'. file-id is None unless knit-kind is 'file'.
2312
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2315
for result in self._find_non_file_keys_to_fetch(revision_ids):
2318
def _find_file_keys_to_fetch(self, revision_ids, pb):
2319
# XXX: it's a bit weird to control the inventory weave caching in this
2320
# generator. Ideally the caching would be done in fetch.py I think. Or
2321
# maybe this generator should explicitly have the contract that it
2322
# should not be iterated until the previously yielded item has been
2324
inv_w = self.inventories
2326
# file ids that changed
2327
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2329
num_file_ids = len(file_ids)
2330
for file_id, altered_versions in file_ids.iteritems():
2332
pb.update("Fetch texts", count, num_file_ids)
2334
yield ("file", file_id, altered_versions)
2336
def _find_non_file_keys_to_fetch(self, revision_ids):
2338
yield ("inventory", None, revision_ids)
2341
# XXX: Note ATM no callers actually pay attention to this return
2342
# instead they just use the list of revision ids and ignore
2343
# missing sigs. Consider removing this work entirely
2344
revisions_with_signatures = set(self.signatures.get_parent_map(
2345
[(r,) for r in revision_ids]))
2346
revisions_with_signatures = set(
2347
[r for (r,) in revisions_with_signatures])
2348
revisions_with_signatures.intersection_update(revision_ids)
2349
yield ("signatures", None, revisions_with_signatures)
2352
yield ("revisions", None, revision_ids)
2355
def get_inventory(self, revision_id):
2356
"""Get Inventory object by revision id."""
2357
return self.iter_inventories([revision_id]).next()
2359
def iter_inventories(self, revision_ids, ordering=None):
2360
"""Get many inventories by revision_ids.
2362
This will buffer some or all of the texts used in constructing the
2363
inventories in memory, but will only parse a single inventory at a
2366
:param revision_ids: The expected revision ids of the inventories.
2367
:param ordering: optional ordering, e.g. 'topological'. If not
2368
specified, the order of revision_ids will be preserved (by
2369
buffering if necessary).
2370
:return: An iterator of inventories.
2372
if ((None in revision_ids)
2373
or (_mod_revision.NULL_REVISION in revision_ids)):
2374
raise ValueError('cannot get null revision inventory')
2375
return self._iter_inventories(revision_ids, ordering)
2377
def _iter_inventories(self, revision_ids, ordering):
2378
"""single-document based inventory iteration."""
2379
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2380
for text, revision_id in inv_xmls:
2381
yield self._deserialise_inventory(revision_id, text)
2383
def _iter_inventory_xmls(self, revision_ids, ordering):
2384
if ordering is None:
2385
order_as_requested = True
2386
ordering = 'unordered'
2388
order_as_requested = False
2389
keys = [(revision_id,) for revision_id in revision_ids]
2392
if order_as_requested:
2393
key_iter = iter(keys)
2394
next_key = key_iter.next()
2395
stream = self.inventories.get_record_stream(keys, ordering, True)
2397
for record in stream:
2398
if record.storage_kind != 'absent':
2399
chunks = record.get_bytes_as('chunked')
2400
if order_as_requested:
2401
text_chunks[record.key] = chunks
2403
yield ''.join(chunks), record.key[-1]
2405
raise errors.NoSuchRevision(self, record.key)
2406
if order_as_requested:
2407
# Yield as many results as we can while preserving order.
2408
while next_key in text_chunks:
2409
chunks = text_chunks.pop(next_key)
2410
yield ''.join(chunks), next_key[-1]
2412
next_key = key_iter.next()
2413
except StopIteration:
2414
# We still want to fully consume the get_record_stream,
2415
# just in case it is not actually finished at this point
2419
def _deserialise_inventory(self, revision_id, xml):
2420
"""Transform the xml into an inventory object.
2422
:param revision_id: The expected revision id of the inventory.
2423
:param xml: A serialised inventory.
2425
result = self._serializer.read_inventory_from_string(xml, revision_id,
2426
entry_cache=self._inventory_entry_cache,
2427
return_from_cache=self._safe_to_return_from_cache)
2428
if result.revision_id != revision_id:
2429
raise AssertionError('revision id mismatch %s != %s' % (
2430
result.revision_id, revision_id))
2433
def _serialise_inventory(self, inv):
2434
return self._serializer.write_inventory_to_string(inv)
2436
def _serialise_inventory_to_lines(self, inv):
2437
return self._serializer.write_inventory_to_lines(inv)
2439
def get_serializer_format(self):
2440
return self._serializer.format_num
2443
def _get_inventory_xml(self, revision_id):
2444
"""Get serialized inventory as a string."""
2445
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2447
text, revision_id = texts.next()
2448
except StopIteration:
2449
raise errors.HistoryMissing(self, 'inventory', revision_id)
2452
def get_rev_id_for_revno(self, revno, known_pair):
2453
"""Return the revision id of a revno, given a later (revno, revid)
2454
pair in the same history.
2456
:return: if found (True, revid). If the available history ran out
2457
before reaching the revno, then this returns
2458
(False, (closest_revno, closest_revid)).
2460
known_revno, known_revid = known_pair
2461
partial_history = [known_revid]
2462
distance_from_known = known_revno - revno
2463
if distance_from_known < 0:
2465
'requested revno (%d) is later than given known revno (%d)'
2466
% (revno, known_revno))
2469
self, partial_history, stop_index=distance_from_known)
2470
except errors.RevisionNotPresent, err:
2471
if err.revision_id == known_revid:
2472
# The start revision (known_revid) wasn't found.
2474
# This is a stacked repository with no fallbacks, or a there's a
2475
# left-hand ghost. Either way, even though the revision named in
2476
# the error isn't in this repo, we know it's the next step in this
2477
# left-hand history.
2478
partial_history.append(err.revision_id)
2479
if len(partial_history) <= distance_from_known:
2480
# Didn't find enough history to get a revid for the revno.
2481
earliest_revno = known_revno - len(partial_history) + 1
2482
return (False, (earliest_revno, partial_history[-1]))
2483
if len(partial_history) - 1 > distance_from_known:
2484
raise AssertionError('_iter_for_revno returned too much history')
2485
return (True, partial_history[-1])
2487
def iter_reverse_revision_history(self, revision_id):
2488
"""Iterate backwards through revision ids in the lefthand history
2490
:param revision_id: The revision id to start with. All its lefthand
2491
ancestors will be traversed.
2493
graph = self.get_graph()
2494
next_id = revision_id
2496
if next_id in (None, _mod_revision.NULL_REVISION):
2499
parents = graph.get_parent_map([next_id])[next_id]
2501
raise errors.RevisionNotPresent(next_id, self)
2503
if len(parents) == 0:
2506
next_id = parents[0]
2509
def get_revision_inventory(self, revision_id):
2510
"""Return inventory of a past revision."""
2511
# TODO: Unify this with get_inventory()
2512
# bzr 0.0.6 and later imposes the constraint that the inventory_id
2513
# must be the same as its revision, so this is trivial.
2514
if revision_id is None:
2515
# This does not make sense: if there is no revision,
2516
# then it is the current tree inventory surely ?!
2517
# and thus get_root_id() is something that looks at the last
2518
# commit on the branch, and the get_root_id is an inventory check.
2519
raise NotImplementedError
2520
# return Inventory(self.get_root_id())
2522
return self.get_inventory(revision_id)
2524
def is_shared(self):
2525
"""Return True if this repository is flagged as a shared repository."""
2526
raise NotImplementedError(self.is_shared)
2529
def reconcile(self, other=None, thorough=False):
2530
"""Reconcile this repository."""
2531
from bzrlib.reconcile import RepoReconciler
2532
reconciler = RepoReconciler(self, thorough=thorough)
2533
reconciler.reconcile()
2536
def _refresh_data(self):
2537
"""Helper called from lock_* to ensure coherency with disk.
2539
The default implementation does nothing; it is however possible
2540
for repositories to maintain loaded indices across multiple locks
2541
by checking inside their implementation of this method to see
2542
whether their indices are still valid. This depends of course on
2543
the disk format being validatable in this manner. This method is
2544
also called by the refresh_data() public interface to cause a refresh
2545
to occur while in a write lock so that data inserted by a smart server
2546
push operation is visible on the client's instance of the physical
2551
def revision_tree(self, revision_id):
2552
"""Return Tree for a revision on this branch.
2554
`revision_id` may be NULL_REVISION for the empty tree revision.
2556
revision_id = _mod_revision.ensure_null(revision_id)
2557
# TODO: refactor this to use an existing revision object
2558
# so we don't need to read it in twice.
2559
if revision_id == _mod_revision.NULL_REVISION:
2560
return RevisionTree(self, Inventory(root_id=None),
2561
_mod_revision.NULL_REVISION)
2563
inv = self.get_revision_inventory(revision_id)
2564
return RevisionTree(self, inv, revision_id)
2566
def revision_trees(self, revision_ids):
2567
"""Return Trees for revisions in this repository.
2569
:param revision_ids: a sequence of revision-ids;
2570
a revision-id may not be None or 'null:'
2572
inventories = self.iter_inventories(revision_ids)
2573
for inv in inventories:
2574
yield RevisionTree(self, inv, inv.revision_id)
2576
def _filtered_revision_trees(self, revision_ids, file_ids):
2577
"""Return Tree for a revision on this branch with only some files.
2579
:param revision_ids: a sequence of revision-ids;
2580
a revision-id may not be None or 'null:'
2581
:param file_ids: if not None, the result is filtered
2582
so that only those file-ids, their parents and their
2583
children are included.
2585
inventories = self.iter_inventories(revision_ids)
2586
for inv in inventories:
2587
# Should we introduce a FilteredRevisionTree class rather
2588
# than pre-filter the inventory here?
2589
filtered_inv = inv.filter(file_ids)
2590
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2593
def get_ancestry(self, revision_id, topo_sorted=True):
2594
"""Return a list of revision-ids integrated by a revision.
2596
The first element of the list is always None, indicating the origin
2597
revision. This might change when we have history horizons, or
2598
perhaps we should have a new API.
2600
This is topologically sorted.
2602
if _mod_revision.is_null(revision_id):
2604
if not self.has_revision(revision_id):
2605
raise errors.NoSuchRevision(self, revision_id)
2606
graph = self.get_graph()
2608
search = graph._make_breadth_first_searcher([revision_id])
2611
found, ghosts = search.next_with_ghosts()
2612
except StopIteration:
2615
if _mod_revision.NULL_REVISION in keys:
2616
keys.remove(_mod_revision.NULL_REVISION)
2618
parent_map = graph.get_parent_map(keys)
2619
keys = tsort.topo_sort(parent_map)
2620
return [None] + list(keys)
2622
def pack(self, hint=None):
2623
"""Compress the data within the repository.
2625
This operation only makes sense for some repository types. For other
2626
types it should be a no-op that just returns.
2628
This stub method does not require a lock, but subclasses should use
2629
@needs_write_lock as this is a long running call its reasonable to
2630
implicitly lock for the user.
2632
:param hint: If not supplied, the whole repository is packed.
2633
If supplied, the repository may use the hint parameter as a
2634
hint for the parts of the repository to pack. A hint can be
2635
obtained from the result of commit_write_group(). Out of
2636
date hints are simply ignored, because concurrent operations
2637
can obsolete them rapidly.
2640
def get_transaction(self):
2641
return self.control_files.get_transaction()
2643
def get_parent_map(self, revision_ids):
2644
"""See graph.StackedParentsProvider.get_parent_map"""
2645
# revisions index works in keys; this just works in revisions
2646
# therefore wrap and unwrap
2649
for revision_id in revision_ids:
2650
if revision_id == _mod_revision.NULL_REVISION:
2651
result[revision_id] = ()
2652
elif revision_id is None:
2653
raise ValueError('get_parent_map(None) is not valid')
2655
query_keys.append((revision_id ,))
2656
for ((revision_id,), parent_keys) in \
2657
self.revisions.get_parent_map(query_keys).iteritems():
2659
result[revision_id] = tuple([parent_revid
2660
for (parent_revid,) in parent_keys])
2662
result[revision_id] = (_mod_revision.NULL_REVISION,)
2665
def _make_parents_provider(self):
2668
def get_graph(self, other_repository=None):
2669
"""Return the graph walker for this repository format"""
2670
parents_provider = self._make_parents_provider()
2671
if (other_repository is not None and
2672
not self.has_same_location(other_repository)):
2673
parents_provider = graph.StackedParentsProvider(
2674
[parents_provider, other_repository._make_parents_provider()])
2675
return graph.Graph(parents_provider)
2677
def _get_versioned_file_checker(self, text_key_references=None,
2679
"""Return an object suitable for checking versioned files.
2681
:param text_key_references: if non-None, an already built
2682
dictionary mapping text keys ((fileid, revision_id) tuples)
2683
to whether they were referred to by the inventory of the
2684
revision_id that they contain. If None, this will be
2686
:param ancestors: Optional result from
2687
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2690
return _VersionedFileChecker(self,
2691
text_key_references=text_key_references, ancestors=ancestors)
2693
def revision_ids_to_search_result(self, result_set):
2694
"""Convert a set of revision ids to a graph SearchResult."""
2695
result_parents = set()
2696
for parents in self.get_graph().get_parent_map(
2697
result_set).itervalues():
2698
result_parents.update(parents)
2699
included_keys = result_set.intersection(result_parents)
2700
start_keys = result_set.difference(included_keys)
2701
exclude_keys = result_parents.difference(result_set)
2702
result = graph.SearchResult(start_keys, exclude_keys,
2703
len(result_set), result_set)
2707
def set_make_working_trees(self, new_value):
2708
"""Set the policy flag for making working trees when creating branches.
2710
This only applies to branches that use this repository.
2712
The default is 'True'.
2713
:param new_value: True to restore the default, False to disable making
2716
raise NotImplementedError(self.set_make_working_trees)
2718
def make_working_trees(self):
2719
"""Returns the policy for making working trees on new branches."""
2720
raise NotImplementedError(self.make_working_trees)
2723
def sign_revision(self, revision_id, gpg_strategy):
2724
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2725
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2728
def has_signature_for_revision_id(self, revision_id):
2729
"""Query for a revision signature for revision_id in the repository."""
2730
if not self.has_revision(revision_id):
2731
raise errors.NoSuchRevision(self, revision_id)
2732
sig_present = (1 == len(
2733
self.signatures.get_parent_map([(revision_id,)])))
2737
def get_signature_text(self, revision_id):
2738
"""Return the text for a signature."""
2739
stream = self.signatures.get_record_stream([(revision_id,)],
2741
record = stream.next()
2742
if record.storage_kind == 'absent':
2743
raise errors.NoSuchRevision(self, revision_id)
2744
return record.get_bytes_as('fulltext')
2747
def check(self, revision_ids=None, callback_refs=None, check_repo=True):
2748
"""Check consistency of all history of given revision_ids.
2750
Different repository implementations should override _check().
2752
:param revision_ids: A non-empty list of revision_ids whose ancestry
2753
will be checked. Typically the last revision_id of a branch.
2754
:param callback_refs: A dict of check-refs to resolve and callback
2755
the check/_check method on the items listed as wanting the ref.
2757
:param check_repo: If False do not check the repository contents, just
2758
calculate the data callback_refs requires and call them back.
2760
return self._check(revision_ids, callback_refs=callback_refs,
2761
check_repo=check_repo)
2763
def _check(self, revision_ids, callback_refs, check_repo):
2764
result = check.Check(self, check_repo=check_repo)
2765
result.check(callback_refs)
2768
def _warn_if_deprecated(self, branch=None):
2769
global _deprecation_warning_done
2770
if _deprecation_warning_done:
2774
conf = config.GlobalConfig()
2776
conf = branch.get_config()
2777
if conf.suppress_warning('format_deprecation'):
2779
warning("Format %s for %s is deprecated -"
2780
" please use 'bzr upgrade' to get better performance"
2781
% (self._format, self.bzrdir.transport.base))
2783
_deprecation_warning_done = True
2785
def supports_rich_root(self):
2786
return self._format.rich_root_data
2788
def _check_ascii_revisionid(self, revision_id, method):
2789
"""Private helper for ascii-only repositories."""
2790
# weave repositories refuse to store revisionids that are non-ascii.
2791
if revision_id is not None:
2792
# weaves require ascii revision ids.
2793
if isinstance(revision_id, unicode):
2795
revision_id.encode('ascii')
2796
except UnicodeEncodeError:
2797
raise errors.NonAsciiRevisionId(method, self)
2800
revision_id.decode('ascii')
2801
except UnicodeDecodeError:
2802
raise errors.NonAsciiRevisionId(method, self)
2804
def revision_graph_can_have_wrong_parents(self):
2805
"""Is it possible for this repository to have a revision graph with
2808
If True, then this repository must also implement
2809
_find_inconsistent_revision_parents so that check and reconcile can
2810
check for inconsistencies before proceeding with other checks that may
2811
depend on the revision index being consistent.
2813
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2816
# remove these delegates a while after bzr 0.15
2817
def __make_delegated(name, from_module):
2818
def _deprecated_repository_forwarder():
2819
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2820
% (name, from_module),
2823
m = __import__(from_module, globals(), locals(), [name])
2825
return getattr(m, name)
2826
except AttributeError:
2827
raise AttributeError('module %s has no name %s'
2829
globals()[name] = _deprecated_repository_forwarder
2832
'AllInOneRepository',
2833
'WeaveMetaDirRepository',
2834
'PreSplitOutRepositoryFormat',
2835
'RepositoryFormat4',
2836
'RepositoryFormat5',
2837
'RepositoryFormat6',
2838
'RepositoryFormat7',
2840
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2844
'RepositoryFormatKnit',
2845
'RepositoryFormatKnit1',
2847
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2850
def install_revision(repository, rev, revision_tree):
2851
"""Install all revision data into a repository."""
2852
install_revisions(repository, [(rev, revision_tree, None)])
2855
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2856
"""Install all revision data into a repository.
2858
Accepts an iterable of revision, tree, signature tuples. The signature
2861
repository.start_write_group()
2863
inventory_cache = lru_cache.LRUCache(10)
2864
for n, (revision, revision_tree, signature) in enumerate(iterable):
2865
_install_revision(repository, revision, revision_tree, signature,
2868
pb.update('Transferring revisions', n + 1, num_revisions)
2870
repository.abort_write_group()
2873
repository.commit_write_group()
2876
def _install_revision(repository, rev, revision_tree, signature,
2878
"""Install all revision data into a repository."""
2879
present_parents = []
2881
for p_id in rev.parent_ids:
2882
if repository.has_revision(p_id):
2883
present_parents.append(p_id)
2884
parent_trees[p_id] = repository.revision_tree(p_id)
2886
parent_trees[p_id] = repository.revision_tree(
2887
_mod_revision.NULL_REVISION)
2889
inv = revision_tree.inventory
2890
entries = inv.iter_entries()
2891
# backwards compatibility hack: skip the root id.
2892
if not repository.supports_rich_root():
2893
path, root = entries.next()
2894
if root.revision != rev.revision_id:
2895
raise errors.IncompatibleRevision(repr(repository))
2897
for path, ie in entries:
2898
text_keys[(ie.file_id, ie.revision)] = ie
2899
text_parent_map = repository.texts.get_parent_map(text_keys)
2900
missing_texts = set(text_keys) - set(text_parent_map)
2901
# Add the texts that are not already present
2902
for text_key in missing_texts:
2903
ie = text_keys[text_key]
2905
# FIXME: TODO: The following loop overlaps/duplicates that done by
2906
# commit to determine parents. There is a latent/real bug here where
2907
# the parents inserted are not those commit would do - in particular
2908
# they are not filtered by heads(). RBC, AB
2909
for revision, tree in parent_trees.iteritems():
2910
if ie.file_id not in tree:
2912
parent_id = tree.inventory[ie.file_id].revision
2913
if parent_id in text_parents:
2915
text_parents.append((ie.file_id, parent_id))
2916
lines = revision_tree.get_file(ie.file_id).readlines()
2917
repository.texts.add_lines(text_key, text_parents, lines)
2919
# install the inventory
2920
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2921
# Cache this inventory
2922
inventory_cache[rev.revision_id] = inv
2924
basis_inv = inventory_cache[rev.parent_ids[0]]
2926
repository.add_inventory(rev.revision_id, inv, present_parents)
2928
delta = inv._make_delta(basis_inv)
2929
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2930
rev.revision_id, present_parents)
2932
repository.add_inventory(rev.revision_id, inv, present_parents)
2933
except errors.RevisionAlreadyPresent:
2935
if signature is not None:
2936
repository.add_signature_text(rev.revision_id, signature)
2937
repository.add_revision(rev.revision_id, rev, inv)
2940
class MetaDirRepository(Repository):
2941
"""Repositories in the new meta-dir layout.
2943
:ivar _transport: Transport for access to repository control files,
2944
typically pointing to .bzr/repository.
2947
def __init__(self, _format, a_bzrdir, control_files):
2948
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2949
self._transport = control_files._transport
2951
def is_shared(self):
2952
"""Return True if this repository is flagged as a shared repository."""
2953
return self._transport.has('shared-storage')
2956
def set_make_working_trees(self, new_value):
2957
"""Set the policy flag for making working trees when creating branches.
2959
This only applies to branches that use this repository.
2961
The default is 'True'.
2962
:param new_value: True to restore the default, False to disable making
2967
self._transport.delete('no-working-trees')
2968
except errors.NoSuchFile:
2971
self._transport.put_bytes('no-working-trees', '',
2972
mode=self.bzrdir._get_file_mode())
2974
def make_working_trees(self):
2975
"""Returns the policy for making working trees on new branches."""
2976
return not self._transport.has('no-working-trees')
2979
class MetaDirVersionedFileRepository(MetaDirRepository):
2980
"""Repositories in a meta-dir, that work via versioned file objects."""
2982
def __init__(self, _format, a_bzrdir, control_files):
2983
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2987
network_format_registry = registry.FormatRegistry()
2988
"""Registry of formats indexed by their network name.
2990
The network name for a repository format is an identifier that can be used when
2991
referring to formats with smart server operations. See
2992
RepositoryFormat.network_name() for more detail.
2996
format_registry = registry.FormatRegistry(network_format_registry)
2997
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
2999
This can contain either format instances themselves, or classes/factories that
3000
can be called to obtain one.
3004
#####################################################################
3005
# Repository Formats
3007
class RepositoryFormat(object):
3008
"""A repository format.
3010
Formats provide four things:
3011
* An initialization routine to construct repository data on disk.
3012
* a optional format string which is used when the BzrDir supports
3014
* an open routine which returns a Repository instance.
3015
* A network name for referring to the format in smart server RPC
3018
There is one and only one Format subclass for each on-disk format. But
3019
there can be one Repository subclass that is used for several different
3020
formats. The _format attribute on a Repository instance can be used to
3021
determine the disk format.
3023
Formats are placed in a registry by their format string for reference
3024
during opening. These should be subclasses of RepositoryFormat for
3027
Once a format is deprecated, just deprecate the initialize and open
3028
methods on the format class. Do not deprecate the object, as the
3029
object may be created even when a repository instance hasn't been
3032
Common instance attributes:
3033
_matchingbzrdir - the bzrdir format that the repository format was
3034
originally written to work with. This can be used if manually
3035
constructing a bzrdir and repository, or more commonly for test suite
3039
# Set to True or False in derived classes. True indicates that the format
3040
# supports ghosts gracefully.
3041
supports_ghosts = None
3042
# Can this repository be given external locations to lookup additional
3043
# data. Set to True or False in derived classes.
3044
supports_external_lookups = None
3045
# Does this format support CHK bytestring lookups. Set to True or False in
3047
supports_chks = None
3048
# Should commit add an inventory, or an inventory delta to the repository.
3049
_commit_inv_deltas = True
3050
# What order should fetch operations request streams in?
3051
# The default is unordered as that is the cheapest for an origin to
3053
_fetch_order = 'unordered'
3054
# Does this repository format use deltas that can be fetched as-deltas ?
3055
# (E.g. knits, where the knit deltas can be transplanted intact.
3056
# We default to False, which will ensure that enough data to get
3057
# a full text out of any fetch stream will be grabbed.
3058
_fetch_uses_deltas = False
3059
# Should fetch trigger a reconcile after the fetch? Only needed for
3060
# some repository formats that can suffer internal inconsistencies.
3061
_fetch_reconcile = False
3062
# Does this format have < O(tree_size) delta generation. Used to hint what
3063
# code path for commit, amongst other things.
3065
# Does doing a pack operation compress data? Useful for the pack UI command
3066
# (so if there is one pack, the operation can still proceed because it may
3067
# help), and for fetching when data won't have come from the same
3069
pack_compresses = False
3070
# Does the repository inventory storage understand references to trees?
3071
supports_tree_reference = None
3072
# Is the format experimental ?
3073
experimental = False
3076
return "<%s>" % self.__class__.__name__
3078
def __eq__(self, other):
3079
# format objects are generally stateless
3080
return isinstance(other, self.__class__)
3082
def __ne__(self, other):
3083
return not self == other
3086
def find_format(klass, a_bzrdir):
3087
"""Return the format for the repository object in a_bzrdir.
3089
This is used by bzr native formats that have a "format" file in
3090
the repository. Other methods may be used by different types of
3094
transport = a_bzrdir.get_repository_transport(None)
3095
format_string = transport.get_bytes("format")
3096
return format_registry.get(format_string)
3097
except errors.NoSuchFile:
3098
raise errors.NoRepositoryPresent(a_bzrdir)
3100
raise errors.UnknownFormatError(format=format_string,
3104
def register_format(klass, format):
3105
format_registry.register(format.get_format_string(), format)
3108
def unregister_format(klass, format):
3109
format_registry.remove(format.get_format_string())
3112
def get_default_format(klass):
3113
"""Return the current default format."""
3114
from bzrlib import bzrdir
3115
return bzrdir.format_registry.make_bzrdir('default').repository_format
3117
def get_format_string(self):
3118
"""Return the ASCII format string that identifies this format.
3120
Note that in pre format ?? repositories the format string is
3121
not permitted nor written to disk.
3123
raise NotImplementedError(self.get_format_string)
3125
def get_format_description(self):
3126
"""Return the short description for this format."""
3127
raise NotImplementedError(self.get_format_description)
3129
# TODO: this shouldn't be in the base class, it's specific to things that
3130
# use weaves or knits -- mbp 20070207
3131
def _get_versioned_file_store(self,
3136
versionedfile_class=None,
3137
versionedfile_kwargs={},
3139
if versionedfile_class is None:
3140
versionedfile_class = self._versionedfile_class
3141
weave_transport = control_files._transport.clone(name)
3142
dir_mode = control_files._dir_mode
3143
file_mode = control_files._file_mode
3144
return VersionedFileStore(weave_transport, prefixed=prefixed,
3146
file_mode=file_mode,
3147
versionedfile_class=versionedfile_class,
3148
versionedfile_kwargs=versionedfile_kwargs,
3151
def initialize(self, a_bzrdir, shared=False):
3152
"""Initialize a repository of this format in a_bzrdir.
3154
:param a_bzrdir: The bzrdir to put the new repository in it.
3155
:param shared: The repository should be initialized as a sharable one.
3156
:returns: The new repository object.
3158
This may raise UninitializableFormat if shared repository are not
3159
compatible the a_bzrdir.
3161
raise NotImplementedError(self.initialize)
3163
def is_supported(self):
3164
"""Is this format supported?
3166
Supported formats must be initializable and openable.
3167
Unsupported formats may not support initialization or committing or
3168
some other features depending on the reason for not being supported.
3172
def network_name(self):
3173
"""A simple byte string uniquely identifying this format for RPC calls.
3175
MetaDir repository formats use their disk format string to identify the
3176
repository over the wire. All in one formats such as bzr < 0.8, and
3177
foreign formats like svn/git and hg should use some marker which is
3178
unique and immutable.
3180
raise NotImplementedError(self.network_name)
3182
def check_conversion_target(self, target_format):
3183
if self.rich_root_data and not target_format.rich_root_data:
3184
raise errors.BadConversionTarget(
3185
'Does not support rich root data.', target_format,
3187
if (self.supports_tree_reference and
3188
not getattr(target_format, 'supports_tree_reference', False)):
3189
raise errors.BadConversionTarget(
3190
'Does not support nested trees', target_format,
3193
def open(self, a_bzrdir, _found=False):
3194
"""Return an instance of this format for the bzrdir a_bzrdir.
3196
_found is a private parameter, do not use it.
3198
raise NotImplementedError(self.open)
3201
class MetaDirRepositoryFormat(RepositoryFormat):
3202
"""Common base class for the new repositories using the metadir layout."""
3204
rich_root_data = False
3205
supports_tree_reference = False
3206
supports_external_lookups = False
3209
def _matchingbzrdir(self):
3210
matching = bzrdir.BzrDirMetaFormat1()
3211
matching.repository_format = self
3215
super(MetaDirRepositoryFormat, self).__init__()
3217
def _create_control_files(self, a_bzrdir):
3218
"""Create the required files and the initial control_files object."""
3219
# FIXME: RBC 20060125 don't peek under the covers
3220
# NB: no need to escape relative paths that are url safe.
3221
repository_transport = a_bzrdir.get_repository_transport(self)
3222
control_files = lockable_files.LockableFiles(repository_transport,
3223
'lock', lockdir.LockDir)
3224
control_files.create_lock()
3225
return control_files
3227
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
3228
"""Upload the initial blank content."""
3229
control_files = self._create_control_files(a_bzrdir)
3230
control_files.lock_write()
3231
transport = control_files._transport
3233
utf8_files += [('shared-storage', '')]
3235
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
3236
for (filename, content_stream) in files:
3237
transport.put_file(filename, content_stream,
3238
mode=a_bzrdir._get_file_mode())
3239
for (filename, content_bytes) in utf8_files:
3240
transport.put_bytes_non_atomic(filename, content_bytes,
3241
mode=a_bzrdir._get_file_mode())
3243
control_files.unlock()
3245
def network_name(self):
3246
"""Metadir formats have matching disk and network format strings."""
3247
return self.get_format_string()
3250
# Pre-0.8 formats that don't have a disk format string (because they are
3251
# versioned by the matching control directory). We use the control directories
3252
# disk format string as a key for the network_name because they meet the
3253
# constraints (simple string, unique, immutable).
3254
network_format_registry.register_lazy(
3255
"Bazaar-NG branch, format 5\n",
3256
'bzrlib.repofmt.weaverepo',
3257
'RepositoryFormat5',
3259
network_format_registry.register_lazy(
3260
"Bazaar-NG branch, format 6\n",
3261
'bzrlib.repofmt.weaverepo',
3262
'RepositoryFormat6',
3265
# formats which have no format string are not discoverable or independently
3266
# creatable on disk, so are not registered in format_registry. They're
3267
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
3268
# needed, it's constructed directly by the BzrDir. Non-native formats where
3269
# the repository is not separately opened are similar.
3271
format_registry.register_lazy(
3272
'Bazaar-NG Repository format 7',
3273
'bzrlib.repofmt.weaverepo',
3277
format_registry.register_lazy(
3278
'Bazaar-NG Knit Repository Format 1',
3279
'bzrlib.repofmt.knitrepo',
3280
'RepositoryFormatKnit1',
3283
format_registry.register_lazy(
3284
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
3285
'bzrlib.repofmt.knitrepo',
3286
'RepositoryFormatKnit3',
3289
format_registry.register_lazy(
3290
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
3291
'bzrlib.repofmt.knitrepo',
3292
'RepositoryFormatKnit4',
3295
# Pack-based formats. There is one format for pre-subtrees, and one for
3296
# post-subtrees to allow ease of testing.
3297
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
3298
format_registry.register_lazy(
3299
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
3300
'bzrlib.repofmt.pack_repo',
3301
'RepositoryFormatKnitPack1',
3303
format_registry.register_lazy(
3304
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
3305
'bzrlib.repofmt.pack_repo',
3306
'RepositoryFormatKnitPack3',
3308
format_registry.register_lazy(
3309
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
3310
'bzrlib.repofmt.pack_repo',
3311
'RepositoryFormatKnitPack4',
3313
format_registry.register_lazy(
3314
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
3315
'bzrlib.repofmt.pack_repo',
3316
'RepositoryFormatKnitPack5',
3318
format_registry.register_lazy(
3319
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
3320
'bzrlib.repofmt.pack_repo',
3321
'RepositoryFormatKnitPack5RichRoot',
3323
format_registry.register_lazy(
3324
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3325
'bzrlib.repofmt.pack_repo',
3326
'RepositoryFormatKnitPack5RichRootBroken',
3328
format_registry.register_lazy(
3329
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3330
'bzrlib.repofmt.pack_repo',
3331
'RepositoryFormatKnitPack6',
3333
format_registry.register_lazy(
3334
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3335
'bzrlib.repofmt.pack_repo',
3336
'RepositoryFormatKnitPack6RichRoot',
3339
# Development formats.
3340
# Obsolete but kept pending a CHK based subtree format.
3341
format_registry.register_lazy(
3342
("Bazaar development format 2 with subtree support "
3343
"(needs bzr.dev from before 1.8)\n"),
3344
'bzrlib.repofmt.pack_repo',
3345
'RepositoryFormatPackDevelopment2Subtree',
3348
# 1.14->1.16 go below here
3349
format_registry.register_lazy(
3350
'Bazaar development format - group compression and chk inventory'
3351
' (needs bzr.dev from 1.14)\n',
3352
'bzrlib.repofmt.groupcompress_repo',
3353
'RepositoryFormatCHK1',
3356
format_registry.register_lazy(
3357
'Bazaar development format - chk repository with bencode revision '
3358
'serialization (needs bzr.dev from 1.16)\n',
3359
'bzrlib.repofmt.groupcompress_repo',
3360
'RepositoryFormatCHK2',
3362
format_registry.register_lazy(
3363
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3364
'bzrlib.repofmt.groupcompress_repo',
3365
'RepositoryFormat2a',
3369
class InterRepository(InterObject):
3370
"""This class represents operations taking place between two repositories.
3372
Its instances have methods like copy_content and fetch, and contain
3373
references to the source and target repositories these operations can be
3376
Often we will provide convenience methods on 'repository' which carry out
3377
operations with another repository - they will always forward to
3378
InterRepository.get(other).method_name(parameters).
3381
_walk_to_common_revisions_batch_size = 50
3383
"""The available optimised InterRepository types."""
3386
def copy_content(self, revision_id=None):
3387
"""Make a complete copy of the content in self into destination.
3389
This is a destructive operation! Do not use it on existing
3392
:param revision_id: Only copy the content needed to construct
3393
revision_id and its parents.
3396
self.target.set_make_working_trees(self.source.make_working_trees())
3397
except NotImplementedError:
3399
self.target.fetch(self.source, revision_id=revision_id)
3402
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3404
"""Fetch the content required to construct revision_id.
3406
The content is copied from self.source to self.target.
3408
:param revision_id: if None all content is copied, if NULL_REVISION no
3413
ui.ui_factory.warn_experimental_format_fetch(self)
3414
f = _mod_fetch.RepoFetcher(to_repository=self.target,
3415
from_repository=self.source,
3416
last_revision=revision_id,
3417
fetch_spec=fetch_spec,
3418
find_ghosts=find_ghosts)
3420
def _walk_to_common_revisions(self, revision_ids):
3421
"""Walk out from revision_ids in source to revisions target has.
3423
:param revision_ids: The start point for the search.
3424
:return: A set of revision ids.
3426
target_graph = self.target.get_graph()
3427
revision_ids = frozenset(revision_ids)
3428
missing_revs = set()
3429
source_graph = self.source.get_graph()
3430
# ensure we don't pay silly lookup costs.
3431
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3432
null_set = frozenset([_mod_revision.NULL_REVISION])
3433
searcher_exhausted = False
3437
# Iterate the searcher until we have enough next_revs
3438
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3440
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3441
next_revs.update(next_revs_part)
3442
ghosts.update(ghosts_part)
3443
except StopIteration:
3444
searcher_exhausted = True
3446
# If there are ghosts in the source graph, and the caller asked for
3447
# them, make sure that they are present in the target.
3448
# We don't care about other ghosts as we can't fetch them and
3449
# haven't been asked to.
3450
ghosts_to_check = set(revision_ids.intersection(ghosts))
3451
revs_to_get = set(next_revs).union(ghosts_to_check)
3453
have_revs = set(target_graph.get_parent_map(revs_to_get))
3454
# we always have NULL_REVISION present.
3455
have_revs = have_revs.union(null_set)
3456
# Check if the target is missing any ghosts we need.
3457
ghosts_to_check.difference_update(have_revs)
3459
# One of the caller's revision_ids is a ghost in both the
3460
# source and the target.
3461
raise errors.NoSuchRevision(
3462
self.source, ghosts_to_check.pop())
3463
missing_revs.update(next_revs - have_revs)
3464
# Because we may have walked past the original stop point, make
3465
# sure everything is stopped
3466
stop_revs = searcher.find_seen_ancestors(have_revs)
3467
searcher.stop_searching_any(stop_revs)
3468
if searcher_exhausted:
3470
return searcher.get_result()
3473
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3474
"""Return the revision ids that source has that target does not.
3476
:param revision_id: only return revision ids included by this
3478
:param find_ghosts: If True find missing revisions in deep history
3479
rather than just finding the surface difference.
3480
:return: A bzrlib.graph.SearchResult.
3482
# stop searching at found target revisions.
3483
if not find_ghosts and revision_id is not None:
3484
return self._walk_to_common_revisions([revision_id])
3485
# generic, possibly worst case, slow code path.
3486
target_ids = set(self.target.all_revision_ids())
3487
if revision_id is not None:
3488
source_ids = self.source.get_ancestry(revision_id)
3489
if source_ids[0] is not None:
3490
raise AssertionError()
3493
source_ids = self.source.all_revision_ids()
3494
result_set = set(source_ids).difference(target_ids)
3495
return self.source.revision_ids_to_search_result(result_set)
3498
def _same_model(source, target):
3499
"""True if source and target have the same data representation.
3501
Note: this is always called on the base class; overriding it in a
3502
subclass will have no effect.
3505
InterRepository._assert_same_model(source, target)
3507
except errors.IncompatibleRepositories, e:
3511
def _assert_same_model(source, target):
3512
"""Raise an exception if two repositories do not use the same model.
3514
if source.supports_rich_root() != target.supports_rich_root():
3515
raise errors.IncompatibleRepositories(source, target,
3516
"different rich-root support")
3517
if source._serializer != target._serializer:
3518
raise errors.IncompatibleRepositories(source, target,
3519
"different serializers")
3522
class InterSameDataRepository(InterRepository):
3523
"""Code for converting between repositories that represent the same data.
3525
Data format and model must match for this to work.
3529
def _get_repo_format_to_test(self):
3530
"""Repository format for testing with.
3532
InterSameData can pull from subtree to subtree and from non-subtree to
3533
non-subtree, so we test this with the richest repository format.
3535
from bzrlib.repofmt import knitrepo
3536
return knitrepo.RepositoryFormatKnit3()
3539
def is_compatible(source, target):
3540
return InterRepository._same_model(source, target)
3543
class InterWeaveRepo(InterSameDataRepository):
3544
"""Optimised code paths between Weave based repositories.
3546
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3547
implemented lazy inter-object optimisation.
3551
def _get_repo_format_to_test(self):
3552
from bzrlib.repofmt import weaverepo
3553
return weaverepo.RepositoryFormat7()
3556
def is_compatible(source, target):
3557
"""Be compatible with known Weave formats.
3559
We don't test for the stores being of specific types because that
3560
could lead to confusing results, and there is no need to be
3563
from bzrlib.repofmt.weaverepo import (
3569
return (isinstance(source._format, (RepositoryFormat5,
3571
RepositoryFormat7)) and
3572
isinstance(target._format, (RepositoryFormat5,
3574
RepositoryFormat7)))
3575
except AttributeError:
3579
def copy_content(self, revision_id=None):
3580
"""See InterRepository.copy_content()."""
3581
# weave specific optimised path:
3583
self.target.set_make_working_trees(self.source.make_working_trees())
3584
except (errors.RepositoryUpgradeRequired, NotImplemented):
3586
# FIXME do not peek!
3587
if self.source._transport.listable():
3588
pb = ui.ui_factory.nested_progress_bar()
3590
self.target.texts.insert_record_stream(
3591
self.source.texts.get_record_stream(
3592
self.source.texts.keys(), 'topological', False))
3593
pb.update('Copying inventory', 0, 1)
3594
self.target.inventories.insert_record_stream(
3595
self.source.inventories.get_record_stream(
3596
self.source.inventories.keys(), 'topological', False))
3597
self.target.signatures.insert_record_stream(
3598
self.source.signatures.get_record_stream(
3599
self.source.signatures.keys(),
3601
self.target.revisions.insert_record_stream(
3602
self.source.revisions.get_record_stream(
3603
self.source.revisions.keys(),
3604
'topological', True))
3608
self.target.fetch(self.source, revision_id=revision_id)
3611
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3612
"""See InterRepository.missing_revision_ids()."""
3613
# we want all revisions to satisfy revision_id in source.
3614
# but we don't want to stat every file here and there.
3615
# we want then, all revisions other needs to satisfy revision_id
3616
# checked, but not those that we have locally.
3617
# so the first thing is to get a subset of the revisions to
3618
# satisfy revision_id in source, and then eliminate those that
3619
# we do already have.
3620
# this is slow on high latency connection to self, but as this
3621
# disk format scales terribly for push anyway due to rewriting
3622
# inventory.weave, this is considered acceptable.
3624
if revision_id is not None:
3625
source_ids = self.source.get_ancestry(revision_id)
3626
if source_ids[0] is not None:
3627
raise AssertionError()
3630
source_ids = self.source._all_possible_ids()
3631
source_ids_set = set(source_ids)
3632
# source_ids is the worst possible case we may need to pull.
3633
# now we want to filter source_ids against what we actually
3634
# have in target, but don't try to check for existence where we know
3635
# we do not have a revision as that would be pointless.
3636
target_ids = set(self.target._all_possible_ids())
3637
possibly_present_revisions = target_ids.intersection(source_ids_set)
3638
actually_present_revisions = set(
3639
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3640
required_revisions = source_ids_set.difference(actually_present_revisions)
3641
if revision_id is not None:
3642
# we used get_ancestry to determine source_ids then we are assured all
3643
# revisions referenced are present as they are installed in topological order.
3644
# and the tip revision was validated by get_ancestry.
3645
result_set = required_revisions
3647
# if we just grabbed the possibly available ids, then
3648
# we only have an estimate of whats available and need to validate
3649
# that against the revision records.
3651
self.source._eliminate_revisions_not_present(required_revisions))
3652
return self.source.revision_ids_to_search_result(result_set)
3655
class InterKnitRepo(InterSameDataRepository):
3656
"""Optimised code paths between Knit based repositories."""
3659
def _get_repo_format_to_test(self):
3660
from bzrlib.repofmt import knitrepo
3661
return knitrepo.RepositoryFormatKnit1()
3664
def is_compatible(source, target):
3665
"""Be compatible with known Knit formats.
3667
We don't test for the stores being of specific types because that
3668
could lead to confusing results, and there is no need to be
3671
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3673
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3674
isinstance(target._format, RepositoryFormatKnit))
3675
except AttributeError:
3677
return are_knits and InterRepository._same_model(source, target)
3680
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3681
"""See InterRepository.missing_revision_ids()."""
3682
if revision_id is not None:
3683
source_ids = self.source.get_ancestry(revision_id)
3684
if source_ids[0] is not None:
3685
raise AssertionError()
3688
source_ids = self.source.all_revision_ids()
3689
source_ids_set = set(source_ids)
3690
# source_ids is the worst possible case we may need to pull.
3691
# now we want to filter source_ids against what we actually
3692
# have in target, but don't try to check for existence where we know
3693
# we do not have a revision as that would be pointless.
3694
target_ids = set(self.target.all_revision_ids())
3695
possibly_present_revisions = target_ids.intersection(source_ids_set)
3696
actually_present_revisions = set(
3697
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3698
required_revisions = source_ids_set.difference(actually_present_revisions)
3699
if revision_id is not None:
3700
# we used get_ancestry to determine source_ids then we are assured all
3701
# revisions referenced are present as they are installed in topological order.
3702
# and the tip revision was validated by get_ancestry.
3703
result_set = required_revisions
3705
# if we just grabbed the possibly available ids, then
3706
# we only have an estimate of whats available and need to validate
3707
# that against the revision records.
3709
self.source._eliminate_revisions_not_present(required_revisions))
3710
return self.source.revision_ids_to_search_result(result_set)
3713
class InterDifferingSerializer(InterRepository):
3716
def _get_repo_format_to_test(self):
3720
def is_compatible(source, target):
3721
"""Be compatible with Knit2 source and Knit3 target"""
3722
# This is redundant with format.check_conversion_target(), however that
3723
# raises an exception, and we just want to say "False" as in we won't
3724
# support converting between these formats.
3725
if 'IDS_never' in debug.debug_flags:
3727
if source.supports_rich_root() and not target.supports_rich_root():
3729
if (source._format.supports_tree_reference
3730
and not target._format.supports_tree_reference):
3732
if target._fallback_repositories and target._format.supports_chks:
3733
# IDS doesn't know how to copy CHKs for the parent inventories it
3734
# adds to stacked repos.
3736
if 'IDS_always' in debug.debug_flags:
3738
# Only use this code path for local source and target. IDS does far
3739
# too much IO (both bandwidth and roundtrips) over a network.
3740
if not source.bzrdir.transport.base.startswith('file:///'):
3742
if not target.bzrdir.transport.base.startswith('file:///'):
3746
def _get_trees(self, revision_ids, cache):
3748
for rev_id in revision_ids:
3750
possible_trees.append((rev_id, cache[rev_id]))
3752
# Not cached, but inventory might be present anyway.
3754
tree = self.source.revision_tree(rev_id)
3755
except errors.NoSuchRevision:
3756
# Nope, parent is ghost.
3759
cache[rev_id] = tree
3760
possible_trees.append((rev_id, tree))
3761
return possible_trees
3763
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3764
"""Get the best delta and base for this revision.
3766
:return: (basis_id, delta)
3769
# Generate deltas against each tree, to find the shortest.
3770
texts_possibly_new_in_tree = set()
3771
for basis_id, basis_tree in possible_trees:
3772
delta = tree.inventory._make_delta(basis_tree.inventory)
3773
for old_path, new_path, file_id, new_entry in delta:
3774
if new_path is None:
3775
# This file_id isn't present in the new rev, so we don't
3779
# Rich roots are handled elsewhere...
3781
kind = new_entry.kind
3782
if kind != 'directory' and kind != 'file':
3783
# No text record associated with this inventory entry.
3785
# This is a directory or file that has changed somehow.
3786
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3787
deltas.append((len(delta), basis_id, delta))
3789
return deltas[0][1:]
3791
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3792
"""Find all parent revisions that are absent, but for which the
3793
inventory is present, and copy those inventories.
3795
This is necessary to preserve correctness when the source is stacked
3796
without fallbacks configured. (Note that in cases like upgrade the
3797
source may be not have _fallback_repositories even though it is
3801
for parents in parent_map.values():
3802
parent_revs.update(parents)
3803
present_parents = self.source.get_parent_map(parent_revs)
3804
absent_parents = set(parent_revs).difference(present_parents)
3805
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3806
(rev_id,) for rev_id in absent_parents)
3807
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3808
for parent_tree in self.source.revision_trees(parent_inv_ids):
3809
current_revision_id = parent_tree.get_revision_id()
3810
parents_parents_keys = parent_invs_keys_for_stacking[
3811
(current_revision_id,)]
3812
parents_parents = [key[-1] for key in parents_parents_keys]
3813
basis_id = _mod_revision.NULL_REVISION
3814
basis_tree = self.source.revision_tree(basis_id)
3815
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3816
self.target.add_inventory_by_delta(
3817
basis_id, delta, current_revision_id, parents_parents)
3818
cache[current_revision_id] = parent_tree
3820
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3821
"""Fetch across a few revisions.
3823
:param revision_ids: The revisions to copy
3824
:param basis_id: The revision_id of a tree that must be in cache, used
3825
as a basis for delta when no other base is available
3826
:param cache: A cache of RevisionTrees that we can use.
3827
:param a_graph: A Graph object to determine the heads() of the
3828
rich-root data stream.
3829
:return: The revision_id of the last converted tree. The RevisionTree
3830
for it will be in cache
3832
# Walk though all revisions; get inventory deltas, copy referenced
3833
# texts that delta references, insert the delta, revision and
3835
root_keys_to_create = set()
3838
pending_revisions = []
3839
parent_map = self.source.get_parent_map(revision_ids)
3840
self._fetch_parent_invs_for_stacking(parent_map, cache)
3841
self.source._safe_to_return_from_cache = True
3842
for tree in self.source.revision_trees(revision_ids):
3843
# Find a inventory delta for this revision.
3844
# Find text entries that need to be copied, too.
3845
current_revision_id = tree.get_revision_id()
3846
parent_ids = parent_map.get(current_revision_id, ())
3847
parent_trees = self._get_trees(parent_ids, cache)
3848
possible_trees = list(parent_trees)
3849
if len(possible_trees) == 0:
3850
# There either aren't any parents, or the parents are ghosts,
3851
# so just use the last converted tree.
3852
possible_trees.append((basis_id, cache[basis_id]))
3853
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3855
revision = self.source.get_revision(current_revision_id)
3856
pending_deltas.append((basis_id, delta,
3857
current_revision_id, revision.parent_ids))
3858
if self._converting_to_rich_root:
3859
self._revision_id_to_root_id[current_revision_id] = \
3861
# Determine which texts are in present in this revision but not in
3862
# any of the available parents.
3863
texts_possibly_new_in_tree = set()
3864
for old_path, new_path, file_id, entry in delta:
3865
if new_path is None:
3866
# This file_id isn't present in the new rev
3870
if not self.target.supports_rich_root():
3871
# The target doesn't support rich root, so we don't
3874
if self._converting_to_rich_root:
3875
# This can't be copied normally, we have to insert
3877
root_keys_to_create.add((file_id, entry.revision))
3880
texts_possibly_new_in_tree.add((file_id, entry.revision))
3881
for basis_id, basis_tree in possible_trees:
3882
basis_inv = basis_tree.inventory
3883
for file_key in list(texts_possibly_new_in_tree):
3884
file_id, file_revision = file_key
3886
entry = basis_inv[file_id]
3887
except errors.NoSuchId:
3889
if entry.revision == file_revision:
3890
texts_possibly_new_in_tree.remove(file_key)
3891
text_keys.update(texts_possibly_new_in_tree)
3892
pending_revisions.append(revision)
3893
cache[current_revision_id] = tree
3894
basis_id = current_revision_id
3895
self.source._safe_to_return_from_cache = False
3897
from_texts = self.source.texts
3898
to_texts = self.target.texts
3899
if root_keys_to_create:
3900
root_stream = _mod_fetch._new_root_data_stream(
3901
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3902
self.source, graph=a_graph)
3903
to_texts.insert_record_stream(root_stream)
3904
to_texts.insert_record_stream(from_texts.get_record_stream(
3905
text_keys, self.target._format._fetch_order,
3906
not self.target._format._fetch_uses_deltas))
3907
# insert inventory deltas
3908
for delta in pending_deltas:
3909
self.target.add_inventory_by_delta(*delta)
3910
if self.target._fallback_repositories:
3911
# Make sure this stacked repository has all the parent inventories
3912
# for the new revisions that we are about to insert. We do this
3913
# before adding the revisions so that no revision is added until
3914
# all the inventories it may depend on are added.
3915
# Note that this is overzealous, as we may have fetched these in an
3918
revision_ids = set()
3919
for revision in pending_revisions:
3920
revision_ids.add(revision.revision_id)
3921
parent_ids.update(revision.parent_ids)
3922
parent_ids.difference_update(revision_ids)
3923
parent_ids.discard(_mod_revision.NULL_REVISION)
3924
parent_map = self.source.get_parent_map(parent_ids)
3925
# we iterate over parent_map and not parent_ids because we don't
3926
# want to try copying any revision which is a ghost
3927
for parent_tree in self.source.revision_trees(parent_map):
3928
current_revision_id = parent_tree.get_revision_id()
3929
parents_parents = parent_map[current_revision_id]
3930
possible_trees = self._get_trees(parents_parents, cache)
3931
if len(possible_trees) == 0:
3932
# There either aren't any parents, or the parents are
3933
# ghosts, so just use the last converted tree.
3934
possible_trees.append((basis_id, cache[basis_id]))
3935
basis_id, delta = self._get_delta_for_revision(parent_tree,
3936
parents_parents, possible_trees)
3937
self.target.add_inventory_by_delta(
3938
basis_id, delta, current_revision_id, parents_parents)
3939
# insert signatures and revisions
3940
for revision in pending_revisions:
3942
signature = self.source.get_signature_text(
3943
revision.revision_id)
3944
self.target.add_signature_text(revision.revision_id,
3946
except errors.NoSuchRevision:
3948
self.target.add_revision(revision.revision_id, revision)
3951
def _fetch_all_revisions(self, revision_ids, pb):
3952
"""Fetch everything for the list of revisions.
3954
:param revision_ids: The list of revisions to fetch. Must be in
3956
:param pb: A ProgressTask
3959
basis_id, basis_tree = self._get_basis(revision_ids[0])
3961
cache = lru_cache.LRUCache(100)
3962
cache[basis_id] = basis_tree
3963
del basis_tree # We don't want to hang on to it here
3965
if self._converting_to_rich_root and len(revision_ids) > 100:
3966
a_graph = _mod_fetch._get_rich_root_heads_graph(self.source,
3971
for offset in range(0, len(revision_ids), batch_size):
3972
self.target.start_write_group()
3974
pb.update('Transferring revisions', offset,
3976
batch = revision_ids[offset:offset+batch_size]
3977
basis_id = self._fetch_batch(batch, basis_id, cache,
3980
self.source._safe_to_return_from_cache = False
3981
self.target.abort_write_group()
3984
hint = self.target.commit_write_group()
3987
if hints and self.target._format.pack_compresses:
3988
self.target.pack(hint=hints)
3989
pb.update('Transferring revisions', len(revision_ids),
3993
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3995
"""See InterRepository.fetch()."""
3996
if fetch_spec is not None:
3997
raise AssertionError("Not implemented yet...")
3998
# See <https://launchpad.net/bugs/456077> asking for a warning here
4000
# nb this is only active for local-local fetches; other things using
4002
ui.ui_factory.warn_cross_format_fetch(self.source._format,
4003
self.target._format)
4004
ui.ui_factory.warn_experimental_format_fetch(self)
4005
if (not self.source.supports_rich_root()
4006
and self.target.supports_rich_root()):
4007
self._converting_to_rich_root = True
4008
self._revision_id_to_root_id = {}
4010
self._converting_to_rich_root = False
4011
revision_ids = self.target.search_missing_revision_ids(self.source,
4012
revision_id, find_ghosts=find_ghosts).get_keys()
4013
if not revision_ids:
4015
revision_ids = tsort.topo_sort(
4016
self.source.get_graph().get_parent_map(revision_ids))
4017
if not revision_ids:
4019
# Walk though all revisions; get inventory deltas, copy referenced
4020
# texts that delta references, insert the delta, revision and
4023
my_pb = ui.ui_factory.nested_progress_bar()
4026
symbol_versioning.warn(
4027
symbol_versioning.deprecated_in((1, 14, 0))
4028
% "pb parameter to fetch()")
4031
self._fetch_all_revisions(revision_ids, pb)
4033
if my_pb is not None:
4035
return len(revision_ids), 0
4037
def _get_basis(self, first_revision_id):
4038
"""Get a revision and tree which exists in the target.
4040
This assumes that first_revision_id is selected for transmission
4041
because all other ancestors are already present. If we can't find an
4042
ancestor we fall back to NULL_REVISION since we know that is safe.
4044
:return: (basis_id, basis_tree)
4046
first_rev = self.source.get_revision(first_revision_id)
4048
basis_id = first_rev.parent_ids[0]
4049
# only valid as a basis if the target has it
4050
self.target.get_revision(basis_id)
4051
# Try to get a basis tree - if its a ghost it will hit the
4052
# NoSuchRevision case.
4053
basis_tree = self.source.revision_tree(basis_id)
4054
except (IndexError, errors.NoSuchRevision):
4055
basis_id = _mod_revision.NULL_REVISION
4056
basis_tree = self.source.revision_tree(basis_id)
4057
return basis_id, basis_tree
4060
InterRepository.register_optimiser(InterDifferingSerializer)
4061
InterRepository.register_optimiser(InterSameDataRepository)
4062
InterRepository.register_optimiser(InterWeaveRepo)
4063
InterRepository.register_optimiser(InterKnitRepo)
4066
class CopyConverter(object):
4067
"""A repository conversion tool which just performs a copy of the content.
4069
This is slow but quite reliable.
4072
def __init__(self, target_format):
4073
"""Create a CopyConverter.
4075
:param target_format: The format the resulting repository should be.
4077
self.target_format = target_format
4079
def convert(self, repo, pb):
4080
"""Perform the conversion of to_convert, giving feedback via pb.
4082
:param to_convert: The disk object to convert.
4083
:param pb: a progress bar to use for progress information.
4085
pb = ui.ui_factory.nested_progress_bar()
4088
# this is only useful with metadir layouts - separated repo content.
4089
# trigger an assertion if not such
4090
repo._format.get_format_string()
4091
self.repo_dir = repo.bzrdir
4092
pb.update('Moving repository to repository.backup')
4093
self.repo_dir.transport.move('repository', 'repository.backup')
4094
backup_transport = self.repo_dir.transport.clone('repository.backup')
4095
repo._format.check_conversion_target(self.target_format)
4096
self.source_repo = repo._format.open(self.repo_dir,
4098
_override_transport=backup_transport)
4099
pb.update('Creating new repository')
4100
converted = self.target_format.initialize(self.repo_dir,
4101
self.source_repo.is_shared())
4102
converted.lock_write()
4104
pb.update('Copying content')
4105
self.source_repo.copy_content_into(converted)
4108
pb.update('Deleting old repository content')
4109
self.repo_dir.transport.delete_tree('repository.backup')
4110
ui.ui_factory.note('repository converted')
4123
def _unescaper(match, _map=_unescape_map):
4124
code = match.group(1)
4128
if not code.startswith('#'):
4130
return unichr(int(code[1:])).encode('utf8')
4136
def _unescape_xml(data):
4137
"""Unescape predefined XML entities in a string of data."""
4139
if _unescape_re is None:
4140
_unescape_re = re.compile('\&([^;]*);')
4141
return _unescape_re.sub(_unescaper, data)
4144
class _VersionedFileChecker(object):
4146
def __init__(self, repository, text_key_references=None, ancestors=None):
4147
self.repository = repository
4148
self.text_index = self.repository._generate_text_key_index(
4149
text_key_references=text_key_references, ancestors=ancestors)
4151
def calculate_file_version_parents(self, text_key):
4152
"""Calculate the correct parents for a file version according to
4155
parent_keys = self.text_index[text_key]
4156
if parent_keys == [_mod_revision.NULL_REVISION]:
4158
return tuple(parent_keys)
4160
def check_file_version_parents(self, texts, progress_bar=None):
4161
"""Check the parents stored in a versioned file are correct.
4163
It also detects file versions that are not referenced by their
4164
corresponding revision's inventory.
4166
:returns: A tuple of (wrong_parents, dangling_file_versions).
4167
wrong_parents is a dict mapping {revision_id: (stored_parents,
4168
correct_parents)} for each revision_id where the stored parents
4169
are not correct. dangling_file_versions is a set of (file_id,
4170
revision_id) tuples for versions that are present in this versioned
4171
file, but not used by the corresponding inventory.
4173
local_progress = None
4174
if progress_bar is None:
4175
local_progress = ui.ui_factory.nested_progress_bar()
4176
progress_bar = local_progress
4178
return self._check_file_version_parents(texts, progress_bar)
4181
local_progress.finished()
4183
def _check_file_version_parents(self, texts, progress_bar):
4184
"""See check_file_version_parents."""
4186
self.file_ids = set([file_id for file_id, _ in
4187
self.text_index.iterkeys()])
4188
# text keys is now grouped by file_id
4189
n_versions = len(self.text_index)
4190
progress_bar.update('loading text store', 0, n_versions)
4191
parent_map = self.repository.texts.get_parent_map(self.text_index)
4192
# On unlistable transports this could well be empty/error...
4193
text_keys = self.repository.texts.keys()
4194
unused_keys = frozenset(text_keys) - set(self.text_index)
4195
for num, key in enumerate(self.text_index.iterkeys()):
4196
progress_bar.update('checking text graph', num, n_versions)
4197
correct_parents = self.calculate_file_version_parents(key)
4199
knit_parents = parent_map[key]
4200
except errors.RevisionNotPresent:
4203
if correct_parents != knit_parents:
4204
wrong_parents[key] = (knit_parents, correct_parents)
4205
return wrong_parents, unused_keys
4208
def _old_get_graph(repository, revision_id):
4209
"""DO NOT USE. That is all. I'm serious."""
4210
graph = repository.get_graph()
4211
revision_graph = dict(((key, value) for key, value in
4212
graph.iter_ancestry([revision_id]) if value is not None))
4213
return _strip_NULL_ghosts(revision_graph)
4216
def _strip_NULL_ghosts(revision_graph):
4217
"""Also don't use this. more compatibility code for unmigrated clients."""
4218
# Filter ghosts, and null:
4219
if _mod_revision.NULL_REVISION in revision_graph:
4220
del revision_graph[_mod_revision.NULL_REVISION]
4221
for key, parents in revision_graph.items():
4222
revision_graph[key] = tuple(parent for parent in parents if parent
4224
return revision_graph
4227
class StreamSink(object):
4228
"""An object that can insert a stream into a repository.
4230
This interface handles the complexity of reserialising inventories and
4231
revisions from different formats, and allows unidirectional insertion into
4232
stacked repositories without looking for the missing basis parents
4236
def __init__(self, target_repo):
4237
self.target_repo = target_repo
4239
def insert_stream(self, stream, src_format, resume_tokens):
4240
"""Insert a stream's content into the target repository.
4242
:param src_format: a bzr repository format.
4244
:return: a list of resume tokens and an iterable of keys additional
4245
items required before the insertion can be completed.
4247
self.target_repo.lock_write()
4250
self.target_repo.resume_write_group(resume_tokens)
4253
self.target_repo.start_write_group()
4256
# locked_insert_stream performs a commit|suspend.
4257
return self._locked_insert_stream(stream, src_format, is_resume)
4259
self.target_repo.abort_write_group(suppress_errors=True)
4262
self.target_repo.unlock()
4264
def _locked_insert_stream(self, stream, src_format, is_resume):
4265
to_serializer = self.target_repo._format._serializer
4266
src_serializer = src_format._serializer
4268
if to_serializer == src_serializer:
4269
# If serializers match and the target is a pack repository, set the
4270
# write cache size on the new pack. This avoids poor performance
4271
# on transports where append is unbuffered (such as
4272
# RemoteTransport). This is safe to do because nothing should read
4273
# back from the target repository while a stream with matching
4274
# serialization is being inserted.
4275
# The exception is that a delta record from the source that should
4276
# be a fulltext may need to be expanded by the target (see
4277
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4278
# explicitly flush any buffered writes first in that rare case.
4280
new_pack = self.target_repo._pack_collection._new_pack
4281
except AttributeError:
4282
# Not a pack repository
4285
new_pack.set_write_cache_size(1024*1024)
4286
for substream_type, substream in stream:
4287
if 'stream' in debug.debug_flags:
4288
mutter('inserting substream: %s', substream_type)
4289
if substream_type == 'texts':
4290
self.target_repo.texts.insert_record_stream(substream)
4291
elif substream_type == 'inventories':
4292
if src_serializer == to_serializer:
4293
self.target_repo.inventories.insert_record_stream(
4296
self._extract_and_insert_inventories(
4297
substream, src_serializer)
4298
elif substream_type == 'inventory-deltas':
4299
ui.ui_factory.warn_cross_format_fetch(src_format,
4300
self.target_repo._format)
4301
self._extract_and_insert_inventory_deltas(
4302
substream, src_serializer)
4303
elif substream_type == 'chk_bytes':
4304
# XXX: This doesn't support conversions, as it assumes the
4305
# conversion was done in the fetch code.
4306
self.target_repo.chk_bytes.insert_record_stream(substream)
4307
elif substream_type == 'revisions':
4308
# This may fallback to extract-and-insert more often than
4309
# required if the serializers are different only in terms of
4311
if src_serializer == to_serializer:
4312
self.target_repo.revisions.insert_record_stream(
4315
self._extract_and_insert_revisions(substream,
4317
elif substream_type == 'signatures':
4318
self.target_repo.signatures.insert_record_stream(substream)
4320
raise AssertionError('kaboom! %s' % (substream_type,))
4321
# Done inserting data, and the missing_keys calculations will try to
4322
# read back from the inserted data, so flush the writes to the new pack
4323
# (if this is pack format).
4324
if new_pack is not None:
4325
new_pack._write_data('', flush=True)
4326
# Find all the new revisions (including ones from resume_tokens)
4327
missing_keys = self.target_repo.get_missing_parent_inventories(
4328
check_for_missing_texts=is_resume)
4330
for prefix, versioned_file in (
4331
('texts', self.target_repo.texts),
4332
('inventories', self.target_repo.inventories),
4333
('revisions', self.target_repo.revisions),
4334
('signatures', self.target_repo.signatures),
4335
('chk_bytes', self.target_repo.chk_bytes),
4337
if versioned_file is None:
4339
# TODO: key is often going to be a StaticTuple object
4340
# I don't believe we can define a method by which
4341
# (prefix,) + StaticTuple will work, though we could
4342
# define a StaticTuple.sq_concat that would allow you to
4343
# pass in either a tuple or a StaticTuple as the second
4344
# object, so instead we could have:
4345
# StaticTuple(prefix) + key here...
4346
missing_keys.update((prefix,) + key for key in
4347
versioned_file.get_missing_compression_parent_keys())
4348
except NotImplementedError:
4349
# cannot even attempt suspending, and missing would have failed
4350
# during stream insertion.
4351
missing_keys = set()
4354
# suspend the write group and tell the caller what we is
4355
# missing. We know we can suspend or else we would not have
4356
# entered this code path. (All repositories that can handle
4357
# missing keys can handle suspending a write group).
4358
write_group_tokens = self.target_repo.suspend_write_group()
4359
return write_group_tokens, missing_keys
4360
hint = self.target_repo.commit_write_group()
4361
if (to_serializer != src_serializer and
4362
self.target_repo._format.pack_compresses):
4363
self.target_repo.pack(hint=hint)
4366
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4367
target_rich_root = self.target_repo._format.rich_root_data
4368
target_tree_refs = self.target_repo._format.supports_tree_reference
4369
for record in substream:
4370
# Insert the delta directly
4371
inventory_delta_bytes = record.get_bytes_as('fulltext')
4372
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4374
parse_result = deserialiser.parse_text_bytes(
4375
inventory_delta_bytes)
4376
except inventory_delta.IncompatibleInventoryDelta, err:
4377
trace.mutter("Incompatible delta: %s", err.msg)
4378
raise errors.IncompatibleRevision(self.target_repo._format)
4379
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4380
revision_id = new_id
4381
parents = [key[0] for key in record.parents]
4382
self.target_repo.add_inventory_by_delta(
4383
basis_id, inv_delta, revision_id, parents)
4385
def _extract_and_insert_inventories(self, substream, serializer,
4387
"""Generate a new inventory versionedfile in target, converting data.
4389
The inventory is retrieved from the source, (deserializing it), and
4390
stored in the target (reserializing it in a different format).
4392
target_rich_root = self.target_repo._format.rich_root_data
4393
target_tree_refs = self.target_repo._format.supports_tree_reference
4394
for record in substream:
4395
# It's not a delta, so it must be a fulltext in the source
4396
# serializer's format.
4397
bytes = record.get_bytes_as('fulltext')
4398
revision_id = record.key[0]
4399
inv = serializer.read_inventory_from_string(bytes, revision_id)
4400
parents = [key[0] for key in record.parents]
4401
self.target_repo.add_inventory(revision_id, inv, parents)
4402
# No need to keep holding this full inv in memory when the rest of
4403
# the substream is likely to be all deltas.
4406
def _extract_and_insert_revisions(self, substream, serializer):
4407
for record in substream:
4408
bytes = record.get_bytes_as('fulltext')
4409
revision_id = record.key[0]
4410
rev = serializer.read_revision_from_string(bytes)
4411
if rev.revision_id != revision_id:
4412
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4413
self.target_repo.add_revision(revision_id, rev)
4416
if self.target_repo._format._fetch_reconcile:
4417
self.target_repo.reconcile()
4420
class StreamSource(object):
4421
"""A source of a stream for fetching between repositories."""
4423
def __init__(self, from_repository, to_format):
4424
"""Create a StreamSource streaming from from_repository."""
4425
self.from_repository = from_repository
4426
self.to_format = to_format
4428
def delta_on_metadata(self):
4429
"""Return True if delta's are permitted on metadata streams.
4431
That is on revisions and signatures.
4433
src_serializer = self.from_repository._format._serializer
4434
target_serializer = self.to_format._serializer
4435
return (self.to_format._fetch_uses_deltas and
4436
src_serializer == target_serializer)
4438
def _fetch_revision_texts(self, revs):
4439
# fetch signatures first and then the revision texts
4440
# may need to be a InterRevisionStore call here.
4441
from_sf = self.from_repository.signatures
4442
# A missing signature is just skipped.
4443
keys = [(rev_id,) for rev_id in revs]
4444
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4446
self.to_format._fetch_order,
4447
not self.to_format._fetch_uses_deltas))
4448
# If a revision has a delta, this is actually expanded inside the
4449
# insert_record_stream code now, which is an alternate fix for
4451
from_rf = self.from_repository.revisions
4452
revisions = from_rf.get_record_stream(
4454
self.to_format._fetch_order,
4455
not self.delta_on_metadata())
4456
return [('signatures', signatures), ('revisions', revisions)]
4458
def _generate_root_texts(self, revs):
4459
"""This will be called by get_stream between fetching weave texts and
4460
fetching the inventory weave.
4462
if self._rich_root_upgrade():
4463
return _mod_fetch.Inter1and2Helper(
4464
self.from_repository).generate_root_texts(revs)
4468
def get_stream(self, search):
4470
revs = search.get_keys()
4471
graph = self.from_repository.get_graph()
4472
revs = tsort.topo_sort(graph.get_parent_map(revs))
4473
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4475
for knit_kind, file_id, revisions in data_to_fetch:
4476
if knit_kind != phase:
4478
# Make a new progress bar for this phase
4479
if knit_kind == "file":
4480
# Accumulate file texts
4481
text_keys.extend([(file_id, revision) for revision in
4483
elif knit_kind == "inventory":
4484
# Now copy the file texts.
4485
from_texts = self.from_repository.texts
4486
yield ('texts', from_texts.get_record_stream(
4487
text_keys, self.to_format._fetch_order,
4488
not self.to_format._fetch_uses_deltas))
4489
# Cause an error if a text occurs after we have done the
4492
# Before we process the inventory we generate the root
4493
# texts (if necessary) so that the inventories references
4495
for _ in self._generate_root_texts(revs):
4497
# we fetch only the referenced inventories because we do not
4498
# know for unselected inventories whether all their required
4499
# texts are present in the other repository - it could be
4501
for info in self._get_inventory_stream(revs):
4503
elif knit_kind == "signatures":
4504
# Nothing to do here; this will be taken care of when
4505
# _fetch_revision_texts happens.
4507
elif knit_kind == "revisions":
4508
for record in self._fetch_revision_texts(revs):
4511
raise AssertionError("Unknown knit kind %r" % knit_kind)
4513
def get_stream_for_missing_keys(self, missing_keys):
4514
# missing keys can only occur when we are byte copying and not
4515
# translating (because translation means we don't send
4516
# unreconstructable deltas ever).
4518
keys['texts'] = set()
4519
keys['revisions'] = set()
4520
keys['inventories'] = set()
4521
keys['chk_bytes'] = set()
4522
keys['signatures'] = set()
4523
for key in missing_keys:
4524
keys[key[0]].add(key[1:])
4525
if len(keys['revisions']):
4526
# If we allowed copying revisions at this point, we could end up
4527
# copying a revision without copying its required texts: a
4528
# violation of the requirements for repository integrity.
4529
raise AssertionError(
4530
'cannot copy revisions to fill in missing deltas %s' % (
4531
keys['revisions'],))
4532
for substream_kind, keys in keys.iteritems():
4533
vf = getattr(self.from_repository, substream_kind)
4534
if vf is None and keys:
4535
raise AssertionError(
4536
"cannot fill in keys for a versioned file we don't"
4537
" have: %s needs %s" % (substream_kind, keys))
4539
# No need to stream something we don't have
4541
if substream_kind == 'inventories':
4542
# Some missing keys are genuinely ghosts, filter those out.
4543
present = self.from_repository.inventories.get_parent_map(keys)
4544
revs = [key[0] for key in present]
4545
# Get the inventory stream more-or-less as we do for the
4546
# original stream; there's no reason to assume that records
4547
# direct from the source will be suitable for the sink. (Think
4548
# e.g. 2a -> 1.9-rich-root).
4549
for info in self._get_inventory_stream(revs, missing=True):
4553
# Ask for full texts always so that we don't need more round trips
4554
# after this stream.
4555
# Some of the missing keys are genuinely ghosts, so filter absent
4556
# records. The Sink is responsible for doing another check to
4557
# ensure that ghosts don't introduce missing data for future
4559
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4560
self.to_format._fetch_order, True))
4561
yield substream_kind, stream
4563
def inventory_fetch_order(self):
4564
if self._rich_root_upgrade():
4565
return 'topological'
4567
return self.to_format._fetch_order
4569
def _rich_root_upgrade(self):
4570
return (not self.from_repository._format.rich_root_data and
4571
self.to_format.rich_root_data)
4573
def _get_inventory_stream(self, revision_ids, missing=False):
4574
from_format = self.from_repository._format
4575
if (from_format.supports_chks and self.to_format.supports_chks and
4576
from_format.network_name() == self.to_format.network_name()):
4577
raise AssertionError(
4578
"this case should be handled by GroupCHKStreamSource")
4579
elif 'forceinvdeltas' in debug.debug_flags:
4580
return self._get_convertable_inventory_stream(revision_ids,
4581
delta_versus_null=missing)
4582
elif from_format.network_name() == self.to_format.network_name():
4584
return self._get_simple_inventory_stream(revision_ids,
4586
elif (not from_format.supports_chks and not self.to_format.supports_chks
4587
and from_format._serializer == self.to_format._serializer):
4588
# Essentially the same format.
4589
return self._get_simple_inventory_stream(revision_ids,
4592
# Any time we switch serializations, we want to use an
4593
# inventory-delta based approach.
4594
return self._get_convertable_inventory_stream(revision_ids,
4595
delta_versus_null=missing)
4597
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4598
# NB: This currently reopens the inventory weave in source;
4599
# using a single stream interface instead would avoid this.
4600
from_weave = self.from_repository.inventories
4602
delta_closure = True
4604
delta_closure = not self.delta_on_metadata()
4605
yield ('inventories', from_weave.get_record_stream(
4606
[(rev_id,) for rev_id in revision_ids],
4607
self.inventory_fetch_order(), delta_closure))
4609
def _get_convertable_inventory_stream(self, revision_ids,
4610
delta_versus_null=False):
4611
# The two formats are sufficiently different that there is no fast
4612
# path, so we need to send just inventorydeltas, which any
4613
# sufficiently modern client can insert into any repository.
4614
# The StreamSink code expects to be able to
4615
# convert on the target, so we need to put bytes-on-the-wire that can
4616
# be converted. That means inventory deltas (if the remote is <1.19,
4617
# RemoteStreamSink will fallback to VFS to insert the deltas).
4618
yield ('inventory-deltas',
4619
self._stream_invs_as_deltas(revision_ids,
4620
delta_versus_null=delta_versus_null))
4622
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4623
"""Return a stream of inventory-deltas for the given rev ids.
4625
:param revision_ids: The list of inventories to transmit
4626
:param delta_versus_null: Don't try to find a minimal delta for this
4627
entry, instead compute the delta versus the NULL_REVISION. This
4628
effectively streams a complete inventory. Used for stuff like
4629
filling in missing parents, etc.
4631
from_repo = self.from_repository
4632
revision_keys = [(rev_id,) for rev_id in revision_ids]
4633
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4634
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4636
inventories = self.from_repository.iter_inventories(
4637
revision_ids, 'topological')
4638
format = from_repo._format
4639
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4640
inventory_cache = lru_cache.LRUCache(50)
4641
null_inventory = from_repo.revision_tree(
4642
_mod_revision.NULL_REVISION).inventory
4643
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4644
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4645
# repo back into a non-rich-root repo ought to be allowed)
4646
serializer = inventory_delta.InventoryDeltaSerializer(
4647
versioned_root=format.rich_root_data,
4648
tree_references=format.supports_tree_reference)
4649
for inv in inventories:
4650
key = (inv.revision_id,)
4651
parent_keys = parent_map.get(key, ())
4653
if not delta_versus_null and parent_keys:
4654
# The caller did not ask for complete inventories and we have
4655
# some parents that we can delta against. Make a delta against
4656
# each parent so that we can find the smallest.
4657
parent_ids = [parent_key[0] for parent_key in parent_keys]
4658
for parent_id in parent_ids:
4659
if parent_id not in invs_sent_so_far:
4660
# We don't know that the remote side has this basis, so
4663
if parent_id == _mod_revision.NULL_REVISION:
4664
parent_inv = null_inventory
4666
parent_inv = inventory_cache.get(parent_id, None)
4667
if parent_inv is None:
4668
parent_inv = from_repo.get_inventory(parent_id)
4669
candidate_delta = inv._make_delta(parent_inv)
4670
if (delta is None or
4671
len(delta) > len(candidate_delta)):
4672
delta = candidate_delta
4673
basis_id = parent_id
4675
# Either none of the parents ended up being suitable, or we
4676
# were asked to delta against NULL
4677
basis_id = _mod_revision.NULL_REVISION
4678
delta = inv._make_delta(null_inventory)
4679
invs_sent_so_far.add(inv.revision_id)
4680
inventory_cache[inv.revision_id] = inv
4681
delta_serialized = ''.join(
4682
serializer.delta_to_lines(basis_id, key[-1], delta))
4683
yield versionedfile.FulltextContentFactory(
4684
key, parent_keys, None, delta_serialized)
4687
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4688
stop_revision=None):
4689
"""Extend the partial history to include a given index
4691
If a stop_index is supplied, stop when that index has been reached.
4692
If a stop_revision is supplied, stop when that revision is
4693
encountered. Otherwise, stop when the beginning of history is
4696
:param stop_index: The index which should be present. When it is
4697
present, history extension will stop.
4698
:param stop_revision: The revision id which should be present. When
4699
it is encountered, history extension will stop.
4701
start_revision = partial_history_cache[-1]
4702
iterator = repo.iter_reverse_revision_history(start_revision)
4704
#skip the last revision in the list
4707
if (stop_index is not None and
4708
len(partial_history_cache) > stop_index):
4710
if partial_history_cache[-1] == stop_revision:
4712
revision_id = iterator.next()
4713
partial_history_cache.append(revision_id)
4714
except StopIteration:
added
removed
bzrlib/repository.py
