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():
1488
return [self.bzrdir.open_branch()]
1489
except errors.NotBranchError:
1491
class Evaluator(object):
1494
self.first_call = True
1496
def __call__(self, bzrdir):
1497
# On the first call, the parameter is always the bzrdir
1498
# containing the current repo.
1499
if not self.first_call:
1501
repository = bzrdir.open_repository()
1502
except errors.NoRepositoryPresent:
1505
return False, (None, repository)
1506
self.first_call = False
1508
value = (bzrdir.open_branch(), None)
1509
except errors.NotBranchError:
1510
value = (None, None)
1514
for branch, repository in bzrdir.BzrDir.find_bzrdirs(
1515
self.bzrdir.root_transport, evaluate=Evaluator()):
1516
if branch is not None:
1517
branches.append(branch)
1518
if not using and repository is not None:
1519
branches.extend(repository.find_branches())
1523
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
1524
"""Return the revision ids that other has that this does not.
1526
These are returned in topological order.
1528
revision_id: only return revision ids included by revision_id.
1530
return InterRepository.get(other, self).search_missing_revision_ids(
1531
revision_id, find_ghosts)
1535
"""Open the repository rooted at base.
1537
For instance, if the repository is at URL/.bzr/repository,
1538
Repository.open(URL) -> a Repository instance.
1540
control = bzrdir.BzrDir.open(base)
1541
return control.open_repository()
1543
def copy_content_into(self, destination, revision_id=None):
1544
"""Make a complete copy of the content in self into destination.
1546
This is a destructive operation! Do not use it on existing
1549
return InterRepository.get(self, destination).copy_content(revision_id)
1551
def commit_write_group(self):
1552
"""Commit the contents accrued within the current write group.
1554
:seealso: start_write_group.
1556
:return: it may return an opaque hint that can be passed to 'pack'.
1558
if self._write_group is not self.get_transaction():
1559
# has an unlock or relock occured ?
1560
raise errors.BzrError('mismatched lock context %r and '
1562
(self.get_transaction(), self._write_group))
1563
result = self._commit_write_group()
1564
self._write_group = None
1567
def _commit_write_group(self):
1568
"""Template method for per-repository write group cleanup.
1570
This is called before the write group is considered to be
1571
finished and should ensure that all data handed to the repository
1572
for writing during the write group is safely committed (to the
1573
extent possible considering file system caching etc).
1576
def suspend_write_group(self):
1577
raise errors.UnsuspendableWriteGroup(self)
1579
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1580
"""Return the keys of missing inventory parents for revisions added in
1583
A revision is not complete if the inventory delta for that revision
1584
cannot be calculated. Therefore if the parent inventories of a
1585
revision are not present, the revision is incomplete, and e.g. cannot
1586
be streamed by a smart server. This method finds missing inventory
1587
parents for revisions added in this write group.
1589
if not self._format.supports_external_lookups:
1590
# This is only an issue for stacked repositories
1592
if not self.is_in_write_group():
1593
raise AssertionError('not in a write group')
1595
# XXX: We assume that every added revision already has its
1596
# corresponding inventory, so we only check for parent inventories that
1597
# might be missing, rather than all inventories.
1598
parents = set(self.revisions._index.get_missing_parents())
1599
parents.discard(_mod_revision.NULL_REVISION)
1600
unstacked_inventories = self.inventories._index
1601
present_inventories = unstacked_inventories.get_parent_map(
1602
key[-1:] for key in parents)
1603
parents.difference_update(present_inventories)
1604
if len(parents) == 0:
1605
# No missing parent inventories.
1607
if not check_for_missing_texts:
1608
return set(('inventories', rev_id) for (rev_id,) in parents)
1609
# Ok, now we have a list of missing inventories. But these only matter
1610
# if the inventories that reference them are missing some texts they
1611
# appear to introduce.
1612
# XXX: Texts referenced by all added inventories need to be present,
1613
# but at the moment we're only checking for texts referenced by
1614
# inventories at the graph's edge.
1615
key_deps = self.revisions._index._key_dependencies
1616
key_deps.satisfy_refs_for_keys(present_inventories)
1617
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1618
file_ids = self.fileids_altered_by_revision_ids(referrers)
1619
missing_texts = set()
1620
for file_id, version_ids in file_ids.iteritems():
1621
missing_texts.update(
1622
(file_id, version_id) for version_id in version_ids)
1623
present_texts = self.texts.get_parent_map(missing_texts)
1624
missing_texts.difference_update(present_texts)
1625
if not missing_texts:
1626
# No texts are missing, so all revisions and their deltas are
1629
# Alternatively the text versions could be returned as the missing
1630
# keys, but this is likely to be less data.
1631
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1634
def refresh_data(self):
1635
"""Re-read any data needed to to synchronise with disk.
1637
This method is intended to be called after another repository instance
1638
(such as one used by a smart server) has inserted data into the
1639
repository. It may not be called during a write group, but may be
1640
called at any other time.
1642
if self.is_in_write_group():
1643
raise errors.InternalBzrError(
1644
"May not refresh_data while in a write group.")
1645
self._refresh_data()
1647
def resume_write_group(self, tokens):
1648
if not self.is_write_locked():
1649
raise errors.NotWriteLocked(self)
1650
if self._write_group:
1651
raise errors.BzrError('already in a write group')
1652
self._resume_write_group(tokens)
1653
# so we can detect unlock/relock - the write group is now entered.
1654
self._write_group = self.get_transaction()
1656
def _resume_write_group(self, tokens):
1657
raise errors.UnsuspendableWriteGroup(self)
1659
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1661
"""Fetch the content required to construct revision_id from source.
1663
If revision_id is None and fetch_spec is None, then all content is
1666
fetch() may not be used when the repository is in a write group -
1667
either finish the current write group before using fetch, or use
1668
fetch before starting the write group.
1670
:param find_ghosts: Find and copy revisions in the source that are
1671
ghosts in the target (and not reachable directly by walking out to
1672
the first-present revision in target from revision_id).
1673
:param revision_id: If specified, all the content needed for this
1674
revision ID will be copied to the target. Fetch will determine for
1675
itself which content needs to be copied.
1676
:param fetch_spec: If specified, a SearchResult or
1677
PendingAncestryResult that describes which revisions to copy. This
1678
allows copying multiple heads at once. Mutually exclusive with
1681
if fetch_spec is not None and revision_id is not None:
1682
raise AssertionError(
1683
"fetch_spec and revision_id are mutually exclusive.")
1684
if self.is_in_write_group():
1685
raise errors.InternalBzrError(
1686
"May not fetch while in a write group.")
1687
# fast path same-url fetch operations
1688
# TODO: lift out to somewhere common with RemoteRepository
1689
# <https://bugs.edge.launchpad.net/bzr/+bug/401646>
1690
if (self.has_same_location(source)
1691
and fetch_spec is None
1692
and self._has_same_fallbacks(source)):
1693
# check that last_revision is in 'from' and then return a
1695
if (revision_id is not None and
1696
not _mod_revision.is_null(revision_id)):
1697
self.get_revision(revision_id)
1699
# if there is no specific appropriate InterRepository, this will get
1700
# the InterRepository base class, which raises an
1701
# IncompatibleRepositories when asked to fetch.
1702
inter = InterRepository.get(source, self)
1703
return inter.fetch(revision_id=revision_id, pb=pb,
1704
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1706
def create_bundle(self, target, base, fileobj, format=None):
1707
return serializer.write_bundle(self, target, base, fileobj, format)
1709
def get_commit_builder(self, branch, parents, config, timestamp=None,
1710
timezone=None, committer=None, revprops=None,
1712
"""Obtain a CommitBuilder for this repository.
1714
:param branch: Branch to commit to.
1715
:param parents: Revision ids of the parents of the new revision.
1716
:param config: Configuration to use.
1717
:param timestamp: Optional timestamp recorded for commit.
1718
:param timezone: Optional timezone for timestamp.
1719
:param committer: Optional committer to set for commit.
1720
:param revprops: Optional dictionary of revision properties.
1721
:param revision_id: Optional revision id.
1723
if self._fallback_repositories:
1724
raise errors.BzrError("Cannot commit from a lightweight checkout "
1725
"to a stacked branch. See "
1726
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1727
result = self._commit_builder_class(self, parents, config,
1728
timestamp, timezone, committer, revprops, revision_id)
1729
self.start_write_group()
1732
@only_raises(errors.LockNotHeld, errors.LockBroken)
1734
if (self.control_files._lock_count == 1 and
1735
self.control_files._lock_mode == 'w'):
1736
if self._write_group is not None:
1737
self.abort_write_group()
1738
self.control_files.unlock()
1739
raise errors.BzrError(
1740
'Must end write groups before releasing write locks.')
1741
self.control_files.unlock()
1742
if self.control_files._lock_count == 0:
1743
self._inventory_entry_cache.clear()
1744
for repo in self._fallback_repositories:
1748
def clone(self, a_bzrdir, revision_id=None):
1749
"""Clone this repository into a_bzrdir using the current format.
1751
Currently no check is made that the format of this repository and
1752
the bzrdir format are compatible. FIXME RBC 20060201.
1754
:return: The newly created destination repository.
1756
# TODO: deprecate after 0.16; cloning this with all its settings is
1757
# probably not very useful -- mbp 20070423
1758
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1759
self.copy_content_into(dest_repo, revision_id)
1762
def start_write_group(self):
1763
"""Start a write group in the repository.
1765
Write groups are used by repositories which do not have a 1:1 mapping
1766
between file ids and backend store to manage the insertion of data from
1767
both fetch and commit operations.
1769
A write lock is required around the start_write_group/commit_write_group
1770
for the support of lock-requiring repository formats.
1772
One can only insert data into a repository inside a write group.
1776
if not self.is_write_locked():
1777
raise errors.NotWriteLocked(self)
1778
if self._write_group:
1779
raise errors.BzrError('already in a write group')
1780
self._start_write_group()
1781
# so we can detect unlock/relock - the write group is now entered.
1782
self._write_group = self.get_transaction()
1784
def _start_write_group(self):
1785
"""Template method for per-repository write group startup.
1787
This is called before the write group is considered to be
1792
def sprout(self, to_bzrdir, revision_id=None):
1793
"""Create a descendent repository for new development.
1795
Unlike clone, this does not copy the settings of the repository.
1797
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1798
dest_repo.fetch(self, revision_id=revision_id)
1801
def _create_sprouting_repo(self, a_bzrdir, shared):
1802
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1803
# use target default format.
1804
dest_repo = a_bzrdir.create_repository()
1806
# Most control formats need the repository to be specifically
1807
# created, but on some old all-in-one formats it's not needed
1809
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1810
except errors.UninitializableFormat:
1811
dest_repo = a_bzrdir.open_repository()
1814
def _get_sink(self):
1815
"""Return a sink for streaming into this repository."""
1816
return StreamSink(self)
1818
def _get_source(self, to_format):
1819
"""Return a source for streaming from this repository."""
1820
return StreamSource(self, to_format)
1823
def has_revision(self, revision_id):
1824
"""True if this repository has a copy of the revision."""
1825
return revision_id in self.has_revisions((revision_id,))
1828
def has_revisions(self, revision_ids):
1829
"""Probe to find out the presence of multiple revisions.
1831
:param revision_ids: An iterable of revision_ids.
1832
:return: A set of the revision_ids that were present.
1834
parent_map = self.revisions.get_parent_map(
1835
[(rev_id,) for rev_id in revision_ids])
1837
if _mod_revision.NULL_REVISION in revision_ids:
1838
result.add(_mod_revision.NULL_REVISION)
1839
result.update([key[0] for key in parent_map])
1843
def get_revision(self, revision_id):
1844
"""Return the Revision object for a named revision."""
1845
return self.get_revisions([revision_id])[0]
1848
def get_revision_reconcile(self, revision_id):
1849
"""'reconcile' helper routine that allows access to a revision always.
1851
This variant of get_revision does not cross check the weave graph
1852
against the revision one as get_revision does: but it should only
1853
be used by reconcile, or reconcile-alike commands that are correcting
1854
or testing the revision graph.
1856
return self._get_revisions([revision_id])[0]
1859
def get_revisions(self, revision_ids):
1860
"""Get many revisions at once.
1862
Repositories that need to check data on every revision read should
1863
subclass this method.
1865
return self._get_revisions(revision_ids)
1868
def _get_revisions(self, revision_ids):
1869
"""Core work logic to get many revisions without sanity checks."""
1871
for revid, rev in self._iter_revisions(revision_ids):
1873
raise errors.NoSuchRevision(self, revid)
1875
return [revs[revid] for revid in revision_ids]
1877
def _iter_revisions(self, revision_ids):
1878
"""Iterate over revision objects.
1880
:param revision_ids: An iterable of revisions to examine. None may be
1881
passed to request all revisions known to the repository. Note that
1882
not all repositories can find unreferenced revisions; for those
1883
repositories only referenced ones will be returned.
1884
:return: An iterator of (revid, revision) tuples. Absent revisions (
1885
those asked for but not available) are returned as (revid, None).
1887
if revision_ids is None:
1888
revision_ids = self.all_revision_ids()
1890
for rev_id in revision_ids:
1891
if not rev_id or not isinstance(rev_id, basestring):
1892
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1893
keys = [(key,) for key in revision_ids]
1894
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1895
for record in stream:
1896
revid = record.key[0]
1897
if record.storage_kind == 'absent':
1900
text = record.get_bytes_as('fulltext')
1901
rev = self._serializer.read_revision_from_string(text)
1905
def get_revision_xml(self, revision_id):
1906
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1907
# would have already do it.
1908
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1909
# TODO: this can't just be replaced by:
1910
# return self._serializer.write_revision_to_string(
1911
# self.get_revision(revision_id))
1912
# as cStringIO preservers the encoding unlike write_revision_to_string
1913
# or some other call down the path.
1914
rev = self.get_revision(revision_id)
1915
rev_tmp = cStringIO.StringIO()
1916
# the current serializer..
1917
self._serializer.write_revision(rev, rev_tmp)
1919
return rev_tmp.getvalue()
1921
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1922
"""Produce a generator of revision deltas.
1924
Note that the input is a sequence of REVISIONS, not revision_ids.
1925
Trees will be held in memory until the generator exits.
1926
Each delta is relative to the revision's lefthand predecessor.
1928
:param specific_fileids: if not None, the result is filtered
1929
so that only those file-ids, their parents and their
1930
children are included.
1932
# Get the revision-ids of interest
1933
required_trees = set()
1934
for revision in revisions:
1935
required_trees.add(revision.revision_id)
1936
required_trees.update(revision.parent_ids[:1])
1938
# Get the matching filtered trees. Note that it's more
1939
# efficient to pass filtered trees to changes_from() rather
1940
# than doing the filtering afterwards. changes_from() could
1941
# arguably do the filtering itself but it's path-based, not
1942
# file-id based, so filtering before or afterwards is
1944
if specific_fileids is None:
1945
trees = dict((t.get_revision_id(), t) for
1946
t in self.revision_trees(required_trees))
1948
trees = dict((t.get_revision_id(), t) for
1949
t in self._filtered_revision_trees(required_trees,
1952
# Calculate the deltas
1953
for revision in revisions:
1954
if not revision.parent_ids:
1955
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1957
old_tree = trees[revision.parent_ids[0]]
1958
yield trees[revision.revision_id].changes_from(old_tree)
1961
def get_revision_delta(self, revision_id, specific_fileids=None):
1962
"""Return the delta for one revision.
1964
The delta is relative to the left-hand predecessor of the
1967
:param specific_fileids: if not None, the result is filtered
1968
so that only those file-ids, their parents and their
1969
children are included.
1971
r = self.get_revision(revision_id)
1972
return list(self.get_deltas_for_revisions([r],
1973
specific_fileids=specific_fileids))[0]
1976
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1977
signature = gpg_strategy.sign(plaintext)
1978
self.add_signature_text(revision_id, signature)
1981
def add_signature_text(self, revision_id, signature):
1982
self.signatures.add_lines((revision_id,), (),
1983
osutils.split_lines(signature))
1985
def find_text_key_references(self):
1986
"""Find the text key references within the repository.
1988
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1989
to whether they were referred to by the inventory of the
1990
revision_id that they contain. The inventory texts from all present
1991
revision ids are assessed to generate this report.
1993
revision_keys = self.revisions.keys()
1994
w = self.inventories
1995
pb = ui.ui_factory.nested_progress_bar()
1997
return self._find_text_key_references_from_xml_inventory_lines(
1998
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
2002
def _find_text_key_references_from_xml_inventory_lines(self,
2004
"""Core routine for extracting references to texts from inventories.
2006
This performs the translation of xml lines to revision ids.
2008
:param line_iterator: An iterator of lines, origin_version_id
2009
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2010
to whether they were referred to by the inventory of the
2011
revision_id that they contain. Note that if that revision_id was
2012
not part of the line_iterator's output then False will be given -
2013
even though it may actually refer to that key.
2015
if not self._serializer.support_altered_by_hack:
2016
raise AssertionError(
2017
"_find_text_key_references_from_xml_inventory_lines only "
2018
"supported for branches which store inventory as unnested xml"
2019
", not on %r" % self)
2022
# this code needs to read every new line in every inventory for the
2023
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2024
# not present in one of those inventories is unnecessary but not
2025
# harmful because we are filtering by the revision id marker in the
2026
# inventory lines : we only select file ids altered in one of those
2027
# revisions. We don't need to see all lines in the inventory because
2028
# only those added in an inventory in rev X can contain a revision=X
2030
unescape_revid_cache = {}
2031
unescape_fileid_cache = {}
2033
# jam 20061218 In a big fetch, this handles hundreds of thousands
2034
# of lines, so it has had a lot of inlining and optimizing done.
2035
# Sorry that it is a little bit messy.
2036
# Move several functions to be local variables, since this is a long
2038
search = self._file_ids_altered_regex.search
2039
unescape = _unescape_xml
2040
setdefault = result.setdefault
2041
for line, line_key in line_iterator:
2042
match = search(line)
2045
# One call to match.group() returning multiple items is quite a
2046
# bit faster than 2 calls to match.group() each returning 1
2047
file_id, revision_id = match.group('file_id', 'revision_id')
2049
# Inlining the cache lookups helps a lot when you make 170,000
2050
# lines and 350k ids, versus 8.4 unique ids.
2051
# Using a cache helps in 2 ways:
2052
# 1) Avoids unnecessary decoding calls
2053
# 2) Re-uses cached strings, which helps in future set and
2055
# (2) is enough that removing encoding entirely along with
2056
# the cache (so we are using plain strings) results in no
2057
# performance improvement.
2059
revision_id = unescape_revid_cache[revision_id]
2061
unescaped = unescape(revision_id)
2062
unescape_revid_cache[revision_id] = unescaped
2063
revision_id = unescaped
2065
# Note that unconditionally unescaping means that we deserialise
2066
# every fileid, which for general 'pull' is not great, but we don't
2067
# really want to have some many fulltexts that this matters anyway.
2070
file_id = unescape_fileid_cache[file_id]
2072
unescaped = unescape(file_id)
2073
unescape_fileid_cache[file_id] = unescaped
2076
key = (file_id, revision_id)
2077
setdefault(key, False)
2078
if revision_id == line_key[-1]:
2082
def _inventory_xml_lines_for_keys(self, keys):
2083
"""Get a line iterator of the sort needed for findind references.
2085
Not relevant for non-xml inventory repositories.
2087
Ghosts in revision_keys are ignored.
2089
:param revision_keys: The revision keys for the inventories to inspect.
2090
:return: An iterator over (inventory line, revid) for the fulltexts of
2091
all of the xml inventories specified by revision_keys.
2093
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2094
for record in stream:
2095
if record.storage_kind != 'absent':
2096
chunks = record.get_bytes_as('chunked')
2097
revid = record.key[-1]
2098
lines = osutils.chunks_to_lines(chunks)
2102
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2104
"""Helper routine for fileids_altered_by_revision_ids.
2106
This performs the translation of xml lines to revision ids.
2108
:param line_iterator: An iterator of lines, origin_version_id
2109
:param revision_keys: The revision ids to filter for. This should be a
2110
set or other type which supports efficient __contains__ lookups, as
2111
the revision key from each parsed line will be looked up in the
2112
revision_keys filter.
2113
:return: a dictionary mapping altered file-ids to an iterable of
2114
revision_ids. Each altered file-ids has the exact revision_ids that
2115
altered it listed explicitly.
2117
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2118
line_iterator).iterkeys())
2119
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2120
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2121
self._inventory_xml_lines_for_keys(parent_keys)))
2122
new_keys = seen - parent_seen
2124
setdefault = result.setdefault
2125
for key in new_keys:
2126
setdefault(key[0], set()).add(key[-1])
2129
def _find_parent_ids_of_revisions(self, revision_ids):
2130
"""Find all parent ids that are mentioned in the revision graph.
2132
:return: set of revisions that are parents of revision_ids which are
2133
not part of revision_ids themselves
2135
parent_map = self.get_parent_map(revision_ids)
2137
map(parent_ids.update, parent_map.itervalues())
2138
parent_ids.difference_update(revision_ids)
2139
parent_ids.discard(_mod_revision.NULL_REVISION)
2142
def _find_parent_keys_of_revisions(self, revision_keys):
2143
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2145
:param revision_keys: An iterable of revision_keys.
2146
:return: The parents of all revision_keys that are not already in
2149
parent_map = self.revisions.get_parent_map(revision_keys)
2151
map(parent_keys.update, parent_map.itervalues())
2152
parent_keys.difference_update(revision_keys)
2153
parent_keys.discard(_mod_revision.NULL_REVISION)
2156
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2157
"""Find the file ids and versions affected by revisions.
2159
:param revisions: an iterable containing revision ids.
2160
:param _inv_weave: The inventory weave from this repository or None.
2161
If None, the inventory weave will be opened automatically.
2162
:return: a dictionary mapping altered file-ids to an iterable of
2163
revision_ids. Each altered file-ids has the exact revision_ids that
2164
altered it listed explicitly.
2166
selected_keys = set((revid,) for revid in revision_ids)
2167
w = _inv_weave or self.inventories
2168
pb = ui.ui_factory.nested_progress_bar()
2170
return self._find_file_ids_from_xml_inventory_lines(
2171
w.iter_lines_added_or_present_in_keys(
2172
selected_keys, pb=pb),
2177
def iter_files_bytes(self, desired_files):
2178
"""Iterate through file versions.
2180
Files will not necessarily be returned in the order they occur in
2181
desired_files. No specific order is guaranteed.
2183
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2184
value supplied by the caller as part of desired_files. It should
2185
uniquely identify the file version in the caller's context. (Examples:
2186
an index number or a TreeTransform trans_id.)
2188
bytes_iterator is an iterable of bytestrings for the file. The
2189
kind of iterable and length of the bytestrings are unspecified, but for
2190
this implementation, it is a list of bytes produced by
2191
VersionedFile.get_record_stream().
2193
:param desired_files: a list of (file_id, revision_id, identifier)
2197
for file_id, revision_id, callable_data in desired_files:
2198
text_keys[(file_id, revision_id)] = callable_data
2199
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2200
if record.storage_kind == 'absent':
2201
raise errors.RevisionNotPresent(record.key, self)
2202
yield text_keys[record.key], record.get_bytes_as('chunked')
2204
def _generate_text_key_index(self, text_key_references=None,
2206
"""Generate a new text key index for the repository.
2208
This is an expensive function that will take considerable time to run.
2210
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2211
list of parents, also text keys. When a given key has no parents,
2212
the parents list will be [NULL_REVISION].
2214
# All revisions, to find inventory parents.
2215
if ancestors is None:
2216
graph = self.get_graph()
2217
ancestors = graph.get_parent_map(self.all_revision_ids())
2218
if text_key_references is None:
2219
text_key_references = self.find_text_key_references()
2220
pb = ui.ui_factory.nested_progress_bar()
2222
return self._do_generate_text_key_index(ancestors,
2223
text_key_references, pb)
2227
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2228
"""Helper for _generate_text_key_index to avoid deep nesting."""
2229
revision_order = tsort.topo_sort(ancestors)
2230
invalid_keys = set()
2232
for revision_id in revision_order:
2233
revision_keys[revision_id] = set()
2234
text_count = len(text_key_references)
2235
# a cache of the text keys to allow reuse; costs a dict of all the
2236
# keys, but saves a 2-tuple for every child of a given key.
2238
for text_key, valid in text_key_references.iteritems():
2240
invalid_keys.add(text_key)
2242
revision_keys[text_key[1]].add(text_key)
2243
text_key_cache[text_key] = text_key
2244
del text_key_references
2246
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2247
NULL_REVISION = _mod_revision.NULL_REVISION
2248
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2249
# too small for large or very branchy trees. However, for 55K path
2250
# trees, it would be easy to use too much memory trivially. Ideally we
2251
# could gauge this by looking at available real memory etc, but this is
2252
# always a tricky proposition.
2253
inventory_cache = lru_cache.LRUCache(10)
2254
batch_size = 10 # should be ~150MB on a 55K path tree
2255
batch_count = len(revision_order) / batch_size + 1
2257
pb.update("Calculating text parents", processed_texts, text_count)
2258
for offset in xrange(batch_count):
2259
to_query = revision_order[offset * batch_size:(offset + 1) *
2263
for revision_id in to_query:
2264
parent_ids = ancestors[revision_id]
2265
for text_key in revision_keys[revision_id]:
2266
pb.update("Calculating text parents", processed_texts)
2267
processed_texts += 1
2268
candidate_parents = []
2269
for parent_id in parent_ids:
2270
parent_text_key = (text_key[0], parent_id)
2272
check_parent = parent_text_key not in \
2273
revision_keys[parent_id]
2275
# the parent parent_id is a ghost:
2276
check_parent = False
2277
# truncate the derived graph against this ghost.
2278
parent_text_key = None
2280
# look at the parent commit details inventories to
2281
# determine possible candidates in the per file graph.
2284
inv = inventory_cache[parent_id]
2286
inv = self.revision_tree(parent_id).inventory
2287
inventory_cache[parent_id] = inv
2289
parent_entry = inv[text_key[0]]
2290
except (KeyError, errors.NoSuchId):
2292
if parent_entry is not None:
2294
text_key[0], parent_entry.revision)
2296
parent_text_key = None
2297
if parent_text_key is not None:
2298
candidate_parents.append(
2299
text_key_cache[parent_text_key])
2300
parent_heads = text_graph.heads(candidate_parents)
2301
new_parents = list(parent_heads)
2302
new_parents.sort(key=lambda x:candidate_parents.index(x))
2303
if new_parents == []:
2304
new_parents = [NULL_REVISION]
2305
text_index[text_key] = new_parents
2307
for text_key in invalid_keys:
2308
text_index[text_key] = [NULL_REVISION]
2311
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2312
"""Get an iterable listing the keys of all the data introduced by a set
2315
The keys will be ordered so that the corresponding items can be safely
2316
fetched and inserted in that order.
2318
:returns: An iterable producing tuples of (knit-kind, file-id,
2319
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2320
'revisions'. file-id is None unless knit-kind is 'file'.
2322
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2325
for result in self._find_non_file_keys_to_fetch(revision_ids):
2328
def _find_file_keys_to_fetch(self, revision_ids, pb):
2329
# XXX: it's a bit weird to control the inventory weave caching in this
2330
# generator. Ideally the caching would be done in fetch.py I think. Or
2331
# maybe this generator should explicitly have the contract that it
2332
# should not be iterated until the previously yielded item has been
2334
inv_w = self.inventories
2336
# file ids that changed
2337
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2339
num_file_ids = len(file_ids)
2340
for file_id, altered_versions in file_ids.iteritems():
2342
pb.update("Fetch texts", count, num_file_ids)
2344
yield ("file", file_id, altered_versions)
2346
def _find_non_file_keys_to_fetch(self, revision_ids):
2348
yield ("inventory", None, revision_ids)
2351
# XXX: Note ATM no callers actually pay attention to this return
2352
# instead they just use the list of revision ids and ignore
2353
# missing sigs. Consider removing this work entirely
2354
revisions_with_signatures = set(self.signatures.get_parent_map(
2355
[(r,) for r in revision_ids]))
2356
revisions_with_signatures = set(
2357
[r for (r,) in revisions_with_signatures])
2358
revisions_with_signatures.intersection_update(revision_ids)
2359
yield ("signatures", None, revisions_with_signatures)
2362
yield ("revisions", None, revision_ids)
2365
def get_inventory(self, revision_id):
2366
"""Get Inventory object by revision id."""
2367
return self.iter_inventories([revision_id]).next()
2369
def iter_inventories(self, revision_ids, ordering=None):
2370
"""Get many inventories by revision_ids.
2372
This will buffer some or all of the texts used in constructing the
2373
inventories in memory, but will only parse a single inventory at a
2376
:param revision_ids: The expected revision ids of the inventories.
2377
:param ordering: optional ordering, e.g. 'topological'. If not
2378
specified, the order of revision_ids will be preserved (by
2379
buffering if necessary).
2380
:return: An iterator of inventories.
2382
if ((None in revision_ids)
2383
or (_mod_revision.NULL_REVISION in revision_ids)):
2384
raise ValueError('cannot get null revision inventory')
2385
return self._iter_inventories(revision_ids, ordering)
2387
def _iter_inventories(self, revision_ids, ordering):
2388
"""single-document based inventory iteration."""
2389
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2390
for text, revision_id in inv_xmls:
2391
yield self.deserialise_inventory(revision_id, text)
2393
def _iter_inventory_xmls(self, revision_ids, ordering):
2394
if ordering is None:
2395
order_as_requested = True
2396
ordering = 'unordered'
2398
order_as_requested = False
2399
keys = [(revision_id,) for revision_id in revision_ids]
2402
if order_as_requested:
2403
key_iter = iter(keys)
2404
next_key = key_iter.next()
2405
stream = self.inventories.get_record_stream(keys, ordering, True)
2407
for record in stream:
2408
if record.storage_kind != 'absent':
2409
chunks = record.get_bytes_as('chunked')
2410
if order_as_requested:
2411
text_chunks[record.key] = chunks
2413
yield ''.join(chunks), record.key[-1]
2415
raise errors.NoSuchRevision(self, record.key)
2416
if order_as_requested:
2417
# Yield as many results as we can while preserving order.
2418
while next_key in text_chunks:
2419
chunks = text_chunks.pop(next_key)
2420
yield ''.join(chunks), next_key[-1]
2422
next_key = key_iter.next()
2423
except StopIteration:
2424
# We still want to fully consume the get_record_stream,
2425
# just in case it is not actually finished at this point
2429
def deserialise_inventory(self, revision_id, xml):
2430
"""Transform the xml into an inventory object.
2432
:param revision_id: The expected revision id of the inventory.
2433
:param xml: A serialised inventory.
2435
result = self._serializer.read_inventory_from_string(xml, revision_id,
2436
entry_cache=self._inventory_entry_cache,
2437
return_from_cache=self._safe_to_return_from_cache)
2438
if result.revision_id != revision_id:
2439
raise AssertionError('revision id mismatch %s != %s' % (
2440
result.revision_id, revision_id))
2443
def serialise_inventory(self, inv):
2444
return self._serializer.write_inventory_to_string(inv)
2446
def _serialise_inventory_to_lines(self, inv):
2447
return self._serializer.write_inventory_to_lines(inv)
2449
def get_serializer_format(self):
2450
return self._serializer.format_num
2453
def get_inventory_xml(self, revision_id):
2454
"""Get inventory XML as a file object."""
2455
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2457
text, revision_id = texts.next()
2458
except StopIteration:
2459
raise errors.HistoryMissing(self, 'inventory', revision_id)
2463
def get_inventory_sha1(self, revision_id):
2464
"""Return the sha1 hash of the inventory entry
2466
return self.get_revision(revision_id).inventory_sha1
2468
def get_rev_id_for_revno(self, revno, known_pair):
2469
"""Return the revision id of a revno, given a later (revno, revid)
2470
pair in the same history.
2472
:return: if found (True, revid). If the available history ran out
2473
before reaching the revno, then this returns
2474
(False, (closest_revno, closest_revid)).
2476
known_revno, known_revid = known_pair
2477
partial_history = [known_revid]
2478
distance_from_known = known_revno - revno
2479
if distance_from_known < 0:
2481
'requested revno (%d) is later than given known revno (%d)'
2482
% (revno, known_revno))
2485
self, partial_history, stop_index=distance_from_known)
2486
except errors.RevisionNotPresent, err:
2487
if err.revision_id == known_revid:
2488
# The start revision (known_revid) wasn't found.
2490
# This is a stacked repository with no fallbacks, or a there's a
2491
# left-hand ghost. Either way, even though the revision named in
2492
# the error isn't in this repo, we know it's the next step in this
2493
# left-hand history.
2494
partial_history.append(err.revision_id)
2495
if len(partial_history) <= distance_from_known:
2496
# Didn't find enough history to get a revid for the revno.
2497
earliest_revno = known_revno - len(partial_history) + 1
2498
return (False, (earliest_revno, partial_history[-1]))
2499
if len(partial_history) - 1 > distance_from_known:
2500
raise AssertionError('_iter_for_revno returned too much history')
2501
return (True, partial_history[-1])
2503
def iter_reverse_revision_history(self, revision_id):
2504
"""Iterate backwards through revision ids in the lefthand history
2506
:param revision_id: The revision id to start with. All its lefthand
2507
ancestors will be traversed.
2509
graph = self.get_graph()
2510
next_id = revision_id
2512
if next_id in (None, _mod_revision.NULL_REVISION):
2515
parents = graph.get_parent_map([next_id])[next_id]
2517
raise errors.RevisionNotPresent(next_id, self)
2519
if len(parents) == 0:
2522
next_id = parents[0]
2525
def get_revision_inventory(self, revision_id):
2526
"""Return inventory of a past revision."""
2527
# TODO: Unify this with get_inventory()
2528
# bzr 0.0.6 and later imposes the constraint that the inventory_id
2529
# must be the same as its revision, so this is trivial.
2530
if revision_id is None:
2531
# This does not make sense: if there is no revision,
2532
# then it is the current tree inventory surely ?!
2533
# and thus get_root_id() is something that looks at the last
2534
# commit on the branch, and the get_root_id is an inventory check.
2535
raise NotImplementedError
2536
# return Inventory(self.get_root_id())
2538
return self.get_inventory(revision_id)
2540
def is_shared(self):
2541
"""Return True if this repository is flagged as a shared repository."""
2542
raise NotImplementedError(self.is_shared)
2545
def reconcile(self, other=None, thorough=False):
2546
"""Reconcile this repository."""
2547
from bzrlib.reconcile import RepoReconciler
2548
reconciler = RepoReconciler(self, thorough=thorough)
2549
reconciler.reconcile()
2552
def _refresh_data(self):
2553
"""Helper called from lock_* to ensure coherency with disk.
2555
The default implementation does nothing; it is however possible
2556
for repositories to maintain loaded indices across multiple locks
2557
by checking inside their implementation of this method to see
2558
whether their indices are still valid. This depends of course on
2559
the disk format being validatable in this manner. This method is
2560
also called by the refresh_data() public interface to cause a refresh
2561
to occur while in a write lock so that data inserted by a smart server
2562
push operation is visible on the client's instance of the physical
2567
def revision_tree(self, revision_id):
2568
"""Return Tree for a revision on this branch.
2570
`revision_id` may be NULL_REVISION for the empty tree revision.
2572
revision_id = _mod_revision.ensure_null(revision_id)
2573
# TODO: refactor this to use an existing revision object
2574
# so we don't need to read it in twice.
2575
if revision_id == _mod_revision.NULL_REVISION:
2576
return RevisionTree(self, Inventory(root_id=None),
2577
_mod_revision.NULL_REVISION)
2579
inv = self.get_revision_inventory(revision_id)
2580
return RevisionTree(self, inv, revision_id)
2582
def revision_trees(self, revision_ids):
2583
"""Return Trees for revisions in this repository.
2585
:param revision_ids: a sequence of revision-ids;
2586
a revision-id may not be None or 'null:'
2588
inventories = self.iter_inventories(revision_ids)
2589
for inv in inventories:
2590
yield RevisionTree(self, inv, inv.revision_id)
2592
def _filtered_revision_trees(self, revision_ids, file_ids):
2593
"""Return Tree for a revision on this branch with only some files.
2595
:param revision_ids: a sequence of revision-ids;
2596
a revision-id may not be None or 'null:'
2597
:param file_ids: if not None, the result is filtered
2598
so that only those file-ids, their parents and their
2599
children are included.
2601
inventories = self.iter_inventories(revision_ids)
2602
for inv in inventories:
2603
# Should we introduce a FilteredRevisionTree class rather
2604
# than pre-filter the inventory here?
2605
filtered_inv = inv.filter(file_ids)
2606
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2609
def get_ancestry(self, revision_id, topo_sorted=True):
2610
"""Return a list of revision-ids integrated by a revision.
2612
The first element of the list is always None, indicating the origin
2613
revision. This might change when we have history horizons, or
2614
perhaps we should have a new API.
2616
This is topologically sorted.
2618
if _mod_revision.is_null(revision_id):
2620
if not self.has_revision(revision_id):
2621
raise errors.NoSuchRevision(self, revision_id)
2622
graph = self.get_graph()
2624
search = graph._make_breadth_first_searcher([revision_id])
2627
found, ghosts = search.next_with_ghosts()
2628
except StopIteration:
2631
if _mod_revision.NULL_REVISION in keys:
2632
keys.remove(_mod_revision.NULL_REVISION)
2634
parent_map = graph.get_parent_map(keys)
2635
keys = tsort.topo_sort(parent_map)
2636
return [None] + list(keys)
2638
def pack(self, hint=None):
2639
"""Compress the data within the repository.
2641
This operation only makes sense for some repository types. For other
2642
types it should be a no-op that just returns.
2644
This stub method does not require a lock, but subclasses should use
2645
@needs_write_lock as this is a long running call its reasonable to
2646
implicitly lock for the user.
2648
:param hint: If not supplied, the whole repository is packed.
2649
If supplied, the repository may use the hint parameter as a
2650
hint for the parts of the repository to pack. A hint can be
2651
obtained from the result of commit_write_group(). Out of
2652
date hints are simply ignored, because concurrent operations
2653
can obsolete them rapidly.
2656
def get_transaction(self):
2657
return self.control_files.get_transaction()
2659
def get_parent_map(self, revision_ids):
2660
"""See graph.StackedParentsProvider.get_parent_map"""
2661
# revisions index works in keys; this just works in revisions
2662
# therefore wrap and unwrap
2665
for revision_id in revision_ids:
2666
if revision_id == _mod_revision.NULL_REVISION:
2667
result[revision_id] = ()
2668
elif revision_id is None:
2669
raise ValueError('get_parent_map(None) is not valid')
2671
query_keys.append((revision_id ,))
2672
for ((revision_id,), parent_keys) in \
2673
self.revisions.get_parent_map(query_keys).iteritems():
2675
result[revision_id] = tuple([parent_revid
2676
for (parent_revid,) in parent_keys])
2678
result[revision_id] = (_mod_revision.NULL_REVISION,)
2681
def _make_parents_provider(self):
2684
def get_graph(self, other_repository=None):
2685
"""Return the graph walker for this repository format"""
2686
parents_provider = self._make_parents_provider()
2687
if (other_repository is not None and
2688
not self.has_same_location(other_repository)):
2689
parents_provider = graph.StackedParentsProvider(
2690
[parents_provider, other_repository._make_parents_provider()])
2691
return graph.Graph(parents_provider)
2693
def _get_versioned_file_checker(self, text_key_references=None,
2695
"""Return an object suitable for checking versioned files.
2697
:param text_key_references: if non-None, an already built
2698
dictionary mapping text keys ((fileid, revision_id) tuples)
2699
to whether they were referred to by the inventory of the
2700
revision_id that they contain. If None, this will be
2702
:param ancestors: Optional result from
2703
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2706
return _VersionedFileChecker(self,
2707
text_key_references=text_key_references, ancestors=ancestors)
2709
def revision_ids_to_search_result(self, result_set):
2710
"""Convert a set of revision ids to a graph SearchResult."""
2711
result_parents = set()
2712
for parents in self.get_graph().get_parent_map(
2713
result_set).itervalues():
2714
result_parents.update(parents)
2715
included_keys = result_set.intersection(result_parents)
2716
start_keys = result_set.difference(included_keys)
2717
exclude_keys = result_parents.difference(result_set)
2718
result = graph.SearchResult(start_keys, exclude_keys,
2719
len(result_set), result_set)
2723
def set_make_working_trees(self, new_value):
2724
"""Set the policy flag for making working trees when creating branches.
2726
This only applies to branches that use this repository.
2728
The default is 'True'.
2729
:param new_value: True to restore the default, False to disable making
2732
raise NotImplementedError(self.set_make_working_trees)
2734
def make_working_trees(self):
2735
"""Returns the policy for making working trees on new branches."""
2736
raise NotImplementedError(self.make_working_trees)
2739
def sign_revision(self, revision_id, gpg_strategy):
2740
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2741
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2744
def has_signature_for_revision_id(self, revision_id):
2745
"""Query for a revision signature for revision_id in the repository."""
2746
if not self.has_revision(revision_id):
2747
raise errors.NoSuchRevision(self, revision_id)
2748
sig_present = (1 == len(
2749
self.signatures.get_parent_map([(revision_id,)])))
2753
def get_signature_text(self, revision_id):
2754
"""Return the text for a signature."""
2755
stream = self.signatures.get_record_stream([(revision_id,)],
2757
record = stream.next()
2758
if record.storage_kind == 'absent':
2759
raise errors.NoSuchRevision(self, revision_id)
2760
return record.get_bytes_as('fulltext')
2763
def check(self, revision_ids=None, callback_refs=None, check_repo=True):
2764
"""Check consistency of all history of given revision_ids.
2766
Different repository implementations should override _check().
2768
:param revision_ids: A non-empty list of revision_ids whose ancestry
2769
will be checked. Typically the last revision_id of a branch.
2770
:param callback_refs: A dict of check-refs to resolve and callback
2771
the check/_check method on the items listed as wanting the ref.
2773
:param check_repo: If False do not check the repository contents, just
2774
calculate the data callback_refs requires and call them back.
2776
return self._check(revision_ids, callback_refs=callback_refs,
2777
check_repo=check_repo)
2779
def _check(self, revision_ids, callback_refs, check_repo):
2780
result = check.Check(self, check_repo=check_repo)
2781
result.check(callback_refs)
2784
def _warn_if_deprecated(self, branch=None):
2785
global _deprecation_warning_done
2786
if _deprecation_warning_done:
2790
conf = config.GlobalConfig()
2792
conf = branch.get_config()
2793
if conf.suppress_warning('format_deprecation'):
2795
warning("Format %s for %s is deprecated -"
2796
" please use 'bzr upgrade' to get better performance"
2797
% (self._format, self.bzrdir.transport.base))
2799
_deprecation_warning_done = True
2801
def supports_rich_root(self):
2802
return self._format.rich_root_data
2804
def _check_ascii_revisionid(self, revision_id, method):
2805
"""Private helper for ascii-only repositories."""
2806
# weave repositories refuse to store revisionids that are non-ascii.
2807
if revision_id is not None:
2808
# weaves require ascii revision ids.
2809
if isinstance(revision_id, unicode):
2811
revision_id.encode('ascii')
2812
except UnicodeEncodeError:
2813
raise errors.NonAsciiRevisionId(method, self)
2816
revision_id.decode('ascii')
2817
except UnicodeDecodeError:
2818
raise errors.NonAsciiRevisionId(method, self)
2820
def revision_graph_can_have_wrong_parents(self):
2821
"""Is it possible for this repository to have a revision graph with
2824
If True, then this repository must also implement
2825
_find_inconsistent_revision_parents so that check and reconcile can
2826
check for inconsistencies before proceeding with other checks that may
2827
depend on the revision index being consistent.
2829
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2832
# remove these delegates a while after bzr 0.15
2833
def __make_delegated(name, from_module):
2834
def _deprecated_repository_forwarder():
2835
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2836
% (name, from_module),
2839
m = __import__(from_module, globals(), locals(), [name])
2841
return getattr(m, name)
2842
except AttributeError:
2843
raise AttributeError('module %s has no name %s'
2845
globals()[name] = _deprecated_repository_forwarder
2848
'AllInOneRepository',
2849
'WeaveMetaDirRepository',
2850
'PreSplitOutRepositoryFormat',
2851
'RepositoryFormat4',
2852
'RepositoryFormat5',
2853
'RepositoryFormat6',
2854
'RepositoryFormat7',
2856
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2860
'RepositoryFormatKnit',
2861
'RepositoryFormatKnit1',
2863
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2866
def install_revision(repository, rev, revision_tree):
2867
"""Install all revision data into a repository."""
2868
install_revisions(repository, [(rev, revision_tree, None)])
2871
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2872
"""Install all revision data into a repository.
2874
Accepts an iterable of revision, tree, signature tuples. The signature
2877
repository.start_write_group()
2879
inventory_cache = lru_cache.LRUCache(10)
2880
for n, (revision, revision_tree, signature) in enumerate(iterable):
2881
_install_revision(repository, revision, revision_tree, signature,
2884
pb.update('Transferring revisions', n + 1, num_revisions)
2886
repository.abort_write_group()
2889
repository.commit_write_group()
2892
def _install_revision(repository, rev, revision_tree, signature,
2894
"""Install all revision data into a repository."""
2895
present_parents = []
2897
for p_id in rev.parent_ids:
2898
if repository.has_revision(p_id):
2899
present_parents.append(p_id)
2900
parent_trees[p_id] = repository.revision_tree(p_id)
2902
parent_trees[p_id] = repository.revision_tree(
2903
_mod_revision.NULL_REVISION)
2905
inv = revision_tree.inventory
2906
entries = inv.iter_entries()
2907
# backwards compatibility hack: skip the root id.
2908
if not repository.supports_rich_root():
2909
path, root = entries.next()
2910
if root.revision != rev.revision_id:
2911
raise errors.IncompatibleRevision(repr(repository))
2913
for path, ie in entries:
2914
text_keys[(ie.file_id, ie.revision)] = ie
2915
text_parent_map = repository.texts.get_parent_map(text_keys)
2916
missing_texts = set(text_keys) - set(text_parent_map)
2917
# Add the texts that are not already present
2918
for text_key in missing_texts:
2919
ie = text_keys[text_key]
2921
# FIXME: TODO: The following loop overlaps/duplicates that done by
2922
# commit to determine parents. There is a latent/real bug here where
2923
# the parents inserted are not those commit would do - in particular
2924
# they are not filtered by heads(). RBC, AB
2925
for revision, tree in parent_trees.iteritems():
2926
if ie.file_id not in tree:
2928
parent_id = tree.inventory[ie.file_id].revision
2929
if parent_id in text_parents:
2931
text_parents.append((ie.file_id, parent_id))
2932
lines = revision_tree.get_file(ie.file_id).readlines()
2933
repository.texts.add_lines(text_key, text_parents, lines)
2935
# install the inventory
2936
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2937
# Cache this inventory
2938
inventory_cache[rev.revision_id] = inv
2940
basis_inv = inventory_cache[rev.parent_ids[0]]
2942
repository.add_inventory(rev.revision_id, inv, present_parents)
2944
delta = inv._make_delta(basis_inv)
2945
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2946
rev.revision_id, present_parents)
2948
repository.add_inventory(rev.revision_id, inv, present_parents)
2949
except errors.RevisionAlreadyPresent:
2951
if signature is not None:
2952
repository.add_signature_text(rev.revision_id, signature)
2953
repository.add_revision(rev.revision_id, rev, inv)
2956
class MetaDirRepository(Repository):
2957
"""Repositories in the new meta-dir layout.
2959
:ivar _transport: Transport for access to repository control files,
2960
typically pointing to .bzr/repository.
2963
def __init__(self, _format, a_bzrdir, control_files):
2964
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2965
self._transport = control_files._transport
2967
def is_shared(self):
2968
"""Return True if this repository is flagged as a shared repository."""
2969
return self._transport.has('shared-storage')
2972
def set_make_working_trees(self, new_value):
2973
"""Set the policy flag for making working trees when creating branches.
2975
This only applies to branches that use this repository.
2977
The default is 'True'.
2978
:param new_value: True to restore the default, False to disable making
2983
self._transport.delete('no-working-trees')
2984
except errors.NoSuchFile:
2987
self._transport.put_bytes('no-working-trees', '',
2988
mode=self.bzrdir._get_file_mode())
2990
def make_working_trees(self):
2991
"""Returns the policy for making working trees on new branches."""
2992
return not self._transport.has('no-working-trees')
2995
class MetaDirVersionedFileRepository(MetaDirRepository):
2996
"""Repositories in a meta-dir, that work via versioned file objects."""
2998
def __init__(self, _format, a_bzrdir, control_files):
2999
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
3003
network_format_registry = registry.FormatRegistry()
3004
"""Registry of formats indexed by their network name.
3006
The network name for a repository format is an identifier that can be used when
3007
referring to formats with smart server operations. See
3008
RepositoryFormat.network_name() for more detail.
3012
format_registry = registry.FormatRegistry(network_format_registry)
3013
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
3015
This can contain either format instances themselves, or classes/factories that
3016
can be called to obtain one.
3020
#####################################################################
3021
# Repository Formats
3023
class RepositoryFormat(object):
3024
"""A repository format.
3026
Formats provide four things:
3027
* An initialization routine to construct repository data on disk.
3028
* a optional format string which is used when the BzrDir supports
3030
* an open routine which returns a Repository instance.
3031
* A network name for referring to the format in smart server RPC
3034
There is one and only one Format subclass for each on-disk format. But
3035
there can be one Repository subclass that is used for several different
3036
formats. The _format attribute on a Repository instance can be used to
3037
determine the disk format.
3039
Formats are placed in a registry by their format string for reference
3040
during opening. These should be subclasses of RepositoryFormat for
3043
Once a format is deprecated, just deprecate the initialize and open
3044
methods on the format class. Do not deprecate the object, as the
3045
object may be created even when a repository instance hasn't been
3048
Common instance attributes:
3049
_matchingbzrdir - the bzrdir format that the repository format was
3050
originally written to work with. This can be used if manually
3051
constructing a bzrdir and repository, or more commonly for test suite
3055
# Set to True or False in derived classes. True indicates that the format
3056
# supports ghosts gracefully.
3057
supports_ghosts = None
3058
# Can this repository be given external locations to lookup additional
3059
# data. Set to True or False in derived classes.
3060
supports_external_lookups = None
3061
# Does this format support CHK bytestring lookups. Set to True or False in
3063
supports_chks = None
3064
# Should commit add an inventory, or an inventory delta to the repository.
3065
_commit_inv_deltas = True
3066
# What order should fetch operations request streams in?
3067
# The default is unordered as that is the cheapest for an origin to
3069
_fetch_order = 'unordered'
3070
# Does this repository format use deltas that can be fetched as-deltas ?
3071
# (E.g. knits, where the knit deltas can be transplanted intact.
3072
# We default to False, which will ensure that enough data to get
3073
# a full text out of any fetch stream will be grabbed.
3074
_fetch_uses_deltas = False
3075
# Should fetch trigger a reconcile after the fetch? Only needed for
3076
# some repository formats that can suffer internal inconsistencies.
3077
_fetch_reconcile = False
3078
# Does this format have < O(tree_size) delta generation. Used to hint what
3079
# code path for commit, amongst other things.
3081
# Does doing a pack operation compress data? Useful for the pack UI command
3082
# (so if there is one pack, the operation can still proceed because it may
3083
# help), and for fetching when data won't have come from the same
3085
pack_compresses = False
3086
# Does the repository inventory storage understand references to trees?
3087
supports_tree_reference = None
3090
return "<%s>" % self.__class__.__name__
3092
def __eq__(self, other):
3093
# format objects are generally stateless
3094
return isinstance(other, self.__class__)
3096
def __ne__(self, other):
3097
return not self == other
3100
def find_format(klass, a_bzrdir):
3101
"""Return the format for the repository object in a_bzrdir.
3103
This is used by bzr native formats that have a "format" file in
3104
the repository. Other methods may be used by different types of
3108
transport = a_bzrdir.get_repository_transport(None)
3109
format_string = transport.get_bytes("format")
3110
return format_registry.get(format_string)
3111
except errors.NoSuchFile:
3112
raise errors.NoRepositoryPresent(a_bzrdir)
3114
raise errors.UnknownFormatError(format=format_string,
3118
def register_format(klass, format):
3119
format_registry.register(format.get_format_string(), format)
3122
def unregister_format(klass, format):
3123
format_registry.remove(format.get_format_string())
3126
def get_default_format(klass):
3127
"""Return the current default format."""
3128
from bzrlib import bzrdir
3129
return bzrdir.format_registry.make_bzrdir('default').repository_format
3131
def get_format_string(self):
3132
"""Return the ASCII format string that identifies this format.
3134
Note that in pre format ?? repositories the format string is
3135
not permitted nor written to disk.
3137
raise NotImplementedError(self.get_format_string)
3139
def get_format_description(self):
3140
"""Return the short description for this format."""
3141
raise NotImplementedError(self.get_format_description)
3143
# TODO: this shouldn't be in the base class, it's specific to things that
3144
# use weaves or knits -- mbp 20070207
3145
def _get_versioned_file_store(self,
3150
versionedfile_class=None,
3151
versionedfile_kwargs={},
3153
if versionedfile_class is None:
3154
versionedfile_class = self._versionedfile_class
3155
weave_transport = control_files._transport.clone(name)
3156
dir_mode = control_files._dir_mode
3157
file_mode = control_files._file_mode
3158
return VersionedFileStore(weave_transport, prefixed=prefixed,
3160
file_mode=file_mode,
3161
versionedfile_class=versionedfile_class,
3162
versionedfile_kwargs=versionedfile_kwargs,
3165
def initialize(self, a_bzrdir, shared=False):
3166
"""Initialize a repository of this format in a_bzrdir.
3168
:param a_bzrdir: The bzrdir to put the new repository in it.
3169
:param shared: The repository should be initialized as a sharable one.
3170
:returns: The new repository object.
3172
This may raise UninitializableFormat if shared repository are not
3173
compatible the a_bzrdir.
3175
raise NotImplementedError(self.initialize)
3177
def is_supported(self):
3178
"""Is this format supported?
3180
Supported formats must be initializable and openable.
3181
Unsupported formats may not support initialization or committing or
3182
some other features depending on the reason for not being supported.
3186
def network_name(self):
3187
"""A simple byte string uniquely identifying this format for RPC calls.
3189
MetaDir repository formats use their disk format string to identify the
3190
repository over the wire. All in one formats such as bzr < 0.8, and
3191
foreign formats like svn/git and hg should use some marker which is
3192
unique and immutable.
3194
raise NotImplementedError(self.network_name)
3196
def check_conversion_target(self, target_format):
3197
if self.rich_root_data and not target_format.rich_root_data:
3198
raise errors.BadConversionTarget(
3199
'Does not support rich root data.', target_format,
3201
if (self.supports_tree_reference and
3202
not getattr(target_format, 'supports_tree_reference', False)):
3203
raise errors.BadConversionTarget(
3204
'Does not support nested trees', target_format,
3207
def open(self, a_bzrdir, _found=False):
3208
"""Return an instance of this format for the bzrdir a_bzrdir.
3210
_found is a private parameter, do not use it.
3212
raise NotImplementedError(self.open)
3215
class MetaDirRepositoryFormat(RepositoryFormat):
3216
"""Common base class for the new repositories using the metadir layout."""
3218
rich_root_data = False
3219
supports_tree_reference = False
3220
supports_external_lookups = False
3223
def _matchingbzrdir(self):
3224
matching = bzrdir.BzrDirMetaFormat1()
3225
matching.repository_format = self
3229
super(MetaDirRepositoryFormat, self).__init__()
3231
def _create_control_files(self, a_bzrdir):
3232
"""Create the required files and the initial control_files object."""
3233
# FIXME: RBC 20060125 don't peek under the covers
3234
# NB: no need to escape relative paths that are url safe.
3235
repository_transport = a_bzrdir.get_repository_transport(self)
3236
control_files = lockable_files.LockableFiles(repository_transport,
3237
'lock', lockdir.LockDir)
3238
control_files.create_lock()
3239
return control_files
3241
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
3242
"""Upload the initial blank content."""
3243
control_files = self._create_control_files(a_bzrdir)
3244
control_files.lock_write()
3245
transport = control_files._transport
3247
utf8_files += [('shared-storage', '')]
3249
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
3250
for (filename, content_stream) in files:
3251
transport.put_file(filename, content_stream,
3252
mode=a_bzrdir._get_file_mode())
3253
for (filename, content_bytes) in utf8_files:
3254
transport.put_bytes_non_atomic(filename, content_bytes,
3255
mode=a_bzrdir._get_file_mode())
3257
control_files.unlock()
3259
def network_name(self):
3260
"""Metadir formats have matching disk and network format strings."""
3261
return self.get_format_string()
3264
# Pre-0.8 formats that don't have a disk format string (because they are
3265
# versioned by the matching control directory). We use the control directories
3266
# disk format string as a key for the network_name because they meet the
3267
# constraints (simple string, unique, immutable).
3268
network_format_registry.register_lazy(
3269
"Bazaar-NG branch, format 5\n",
3270
'bzrlib.repofmt.weaverepo',
3271
'RepositoryFormat5',
3273
network_format_registry.register_lazy(
3274
"Bazaar-NG branch, format 6\n",
3275
'bzrlib.repofmt.weaverepo',
3276
'RepositoryFormat6',
3279
# formats which have no format string are not discoverable or independently
3280
# creatable on disk, so are not registered in format_registry. They're
3281
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
3282
# needed, it's constructed directly by the BzrDir. Non-native formats where
3283
# the repository is not separately opened are similar.
3285
format_registry.register_lazy(
3286
'Bazaar-NG Repository format 7',
3287
'bzrlib.repofmt.weaverepo',
3291
format_registry.register_lazy(
3292
'Bazaar-NG Knit Repository Format 1',
3293
'bzrlib.repofmt.knitrepo',
3294
'RepositoryFormatKnit1',
3297
format_registry.register_lazy(
3298
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
3299
'bzrlib.repofmt.knitrepo',
3300
'RepositoryFormatKnit3',
3303
format_registry.register_lazy(
3304
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
3305
'bzrlib.repofmt.knitrepo',
3306
'RepositoryFormatKnit4',
3309
# Pack-based formats. There is one format for pre-subtrees, and one for
3310
# post-subtrees to allow ease of testing.
3311
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
3312
format_registry.register_lazy(
3313
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
3314
'bzrlib.repofmt.pack_repo',
3315
'RepositoryFormatKnitPack1',
3317
format_registry.register_lazy(
3318
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
3319
'bzrlib.repofmt.pack_repo',
3320
'RepositoryFormatKnitPack3',
3322
format_registry.register_lazy(
3323
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
3324
'bzrlib.repofmt.pack_repo',
3325
'RepositoryFormatKnitPack4',
3327
format_registry.register_lazy(
3328
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
3329
'bzrlib.repofmt.pack_repo',
3330
'RepositoryFormatKnitPack5',
3332
format_registry.register_lazy(
3333
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
3334
'bzrlib.repofmt.pack_repo',
3335
'RepositoryFormatKnitPack5RichRoot',
3337
format_registry.register_lazy(
3338
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3339
'bzrlib.repofmt.pack_repo',
3340
'RepositoryFormatKnitPack5RichRootBroken',
3342
format_registry.register_lazy(
3343
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3344
'bzrlib.repofmt.pack_repo',
3345
'RepositoryFormatKnitPack6',
3347
format_registry.register_lazy(
3348
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3349
'bzrlib.repofmt.pack_repo',
3350
'RepositoryFormatKnitPack6RichRoot',
3353
# Development formats.
3354
# Obsolete but kept pending a CHK based subtree format.
3355
format_registry.register_lazy(
3356
("Bazaar development format 2 with subtree support "
3357
"(needs bzr.dev from before 1.8)\n"),
3358
'bzrlib.repofmt.pack_repo',
3359
'RepositoryFormatPackDevelopment2Subtree',
3362
# 1.14->1.16 go below here
3363
format_registry.register_lazy(
3364
'Bazaar development format - group compression and chk inventory'
3365
' (needs bzr.dev from 1.14)\n',
3366
'bzrlib.repofmt.groupcompress_repo',
3367
'RepositoryFormatCHK1',
3370
format_registry.register_lazy(
3371
'Bazaar development format - chk repository with bencode revision '
3372
'serialization (needs bzr.dev from 1.16)\n',
3373
'bzrlib.repofmt.groupcompress_repo',
3374
'RepositoryFormatCHK2',
3376
format_registry.register_lazy(
3377
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3378
'bzrlib.repofmt.groupcompress_repo',
3379
'RepositoryFormat2a',
3383
class InterRepository(InterObject):
3384
"""This class represents operations taking place between two repositories.
3386
Its instances have methods like copy_content and fetch, and contain
3387
references to the source and target repositories these operations can be
3390
Often we will provide convenience methods on 'repository' which carry out
3391
operations with another repository - they will always forward to
3392
InterRepository.get(other).method_name(parameters).
3395
_walk_to_common_revisions_batch_size = 50
3397
"""The available optimised InterRepository types."""
3400
def copy_content(self, revision_id=None):
3401
"""Make a complete copy of the content in self into destination.
3403
This is a destructive operation! Do not use it on existing
3406
:param revision_id: Only copy the content needed to construct
3407
revision_id and its parents.
3410
self.target.set_make_working_trees(self.source.make_working_trees())
3411
except NotImplementedError:
3413
self.target.fetch(self.source, revision_id=revision_id)
3416
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3418
"""Fetch the content required to construct revision_id.
3420
The content is copied from self.source to self.target.
3422
:param revision_id: if None all content is copied, if NULL_REVISION no
3424
:param pb: optional progress bar to use for progress reports. If not
3425
provided a default one will be created.
3428
f = _mod_fetch.RepoFetcher(to_repository=self.target,
3429
from_repository=self.source,
3430
last_revision=revision_id,
3431
fetch_spec=fetch_spec,
3432
pb=pb, find_ghosts=find_ghosts)
3434
def _walk_to_common_revisions(self, revision_ids):
3435
"""Walk out from revision_ids in source to revisions target has.
3437
:param revision_ids: The start point for the search.
3438
:return: A set of revision ids.
3440
target_graph = self.target.get_graph()
3441
revision_ids = frozenset(revision_ids)
3442
missing_revs = set()
3443
source_graph = self.source.get_graph()
3444
# ensure we don't pay silly lookup costs.
3445
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3446
null_set = frozenset([_mod_revision.NULL_REVISION])
3447
searcher_exhausted = False
3451
# Iterate the searcher until we have enough next_revs
3452
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3454
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3455
next_revs.update(next_revs_part)
3456
ghosts.update(ghosts_part)
3457
except StopIteration:
3458
searcher_exhausted = True
3460
# If there are ghosts in the source graph, and the caller asked for
3461
# them, make sure that they are present in the target.
3462
# We don't care about other ghosts as we can't fetch them and
3463
# haven't been asked to.
3464
ghosts_to_check = set(revision_ids.intersection(ghosts))
3465
revs_to_get = set(next_revs).union(ghosts_to_check)
3467
have_revs = set(target_graph.get_parent_map(revs_to_get))
3468
# we always have NULL_REVISION present.
3469
have_revs = have_revs.union(null_set)
3470
# Check if the target is missing any ghosts we need.
3471
ghosts_to_check.difference_update(have_revs)
3473
# One of the caller's revision_ids is a ghost in both the
3474
# source and the target.
3475
raise errors.NoSuchRevision(
3476
self.source, ghosts_to_check.pop())
3477
missing_revs.update(next_revs - have_revs)
3478
# Because we may have walked past the original stop point, make
3479
# sure everything is stopped
3480
stop_revs = searcher.find_seen_ancestors(have_revs)
3481
searcher.stop_searching_any(stop_revs)
3482
if searcher_exhausted:
3484
return searcher.get_result()
3487
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3488
"""Return the revision ids that source has that target does not.
3490
:param revision_id: only return revision ids included by this
3492
:param find_ghosts: If True find missing revisions in deep history
3493
rather than just finding the surface difference.
3494
:return: A bzrlib.graph.SearchResult.
3496
# stop searching at found target revisions.
3497
if not find_ghosts and revision_id is not None:
3498
return self._walk_to_common_revisions([revision_id])
3499
# generic, possibly worst case, slow code path.
3500
target_ids = set(self.target.all_revision_ids())
3501
if revision_id is not None:
3502
source_ids = self.source.get_ancestry(revision_id)
3503
if source_ids[0] is not None:
3504
raise AssertionError()
3507
source_ids = self.source.all_revision_ids()
3508
result_set = set(source_ids).difference(target_ids)
3509
return self.source.revision_ids_to_search_result(result_set)
3512
def _same_model(source, target):
3513
"""True if source and target have the same data representation.
3515
Note: this is always called on the base class; overriding it in a
3516
subclass will have no effect.
3519
InterRepository._assert_same_model(source, target)
3521
except errors.IncompatibleRepositories, e:
3525
def _assert_same_model(source, target):
3526
"""Raise an exception if two repositories do not use the same model.
3528
if source.supports_rich_root() != target.supports_rich_root():
3529
raise errors.IncompatibleRepositories(source, target,
3530
"different rich-root support")
3531
if source._serializer != target._serializer:
3532
raise errors.IncompatibleRepositories(source, target,
3533
"different serializers")
3536
class InterSameDataRepository(InterRepository):
3537
"""Code for converting between repositories that represent the same data.
3539
Data format and model must match for this to work.
3543
def _get_repo_format_to_test(self):
3544
"""Repository format for testing with.
3546
InterSameData can pull from subtree to subtree and from non-subtree to
3547
non-subtree, so we test this with the richest repository format.
3549
from bzrlib.repofmt import knitrepo
3550
return knitrepo.RepositoryFormatKnit3()
3553
def is_compatible(source, target):
3554
return InterRepository._same_model(source, target)
3557
class InterWeaveRepo(InterSameDataRepository):
3558
"""Optimised code paths between Weave based repositories.
3560
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3561
implemented lazy inter-object optimisation.
3565
def _get_repo_format_to_test(self):
3566
from bzrlib.repofmt import weaverepo
3567
return weaverepo.RepositoryFormat7()
3570
def is_compatible(source, target):
3571
"""Be compatible with known Weave formats.
3573
We don't test for the stores being of specific types because that
3574
could lead to confusing results, and there is no need to be
3577
from bzrlib.repofmt.weaverepo import (
3583
return (isinstance(source._format, (RepositoryFormat5,
3585
RepositoryFormat7)) and
3586
isinstance(target._format, (RepositoryFormat5,
3588
RepositoryFormat7)))
3589
except AttributeError:
3593
def copy_content(self, revision_id=None):
3594
"""See InterRepository.copy_content()."""
3595
# weave specific optimised path:
3597
self.target.set_make_working_trees(self.source.make_working_trees())
3598
except (errors.RepositoryUpgradeRequired, NotImplemented):
3600
# FIXME do not peek!
3601
if self.source._transport.listable():
3602
pb = ui.ui_factory.nested_progress_bar()
3604
self.target.texts.insert_record_stream(
3605
self.source.texts.get_record_stream(
3606
self.source.texts.keys(), 'topological', False))
3607
pb.update('Copying inventory', 0, 1)
3608
self.target.inventories.insert_record_stream(
3609
self.source.inventories.get_record_stream(
3610
self.source.inventories.keys(), 'topological', False))
3611
self.target.signatures.insert_record_stream(
3612
self.source.signatures.get_record_stream(
3613
self.source.signatures.keys(),
3615
self.target.revisions.insert_record_stream(
3616
self.source.revisions.get_record_stream(
3617
self.source.revisions.keys(),
3618
'topological', True))
3622
self.target.fetch(self.source, revision_id=revision_id)
3625
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3626
"""See InterRepository.missing_revision_ids()."""
3627
# we want all revisions to satisfy revision_id in source.
3628
# but we don't want to stat every file here and there.
3629
# we want then, all revisions other needs to satisfy revision_id
3630
# checked, but not those that we have locally.
3631
# so the first thing is to get a subset of the revisions to
3632
# satisfy revision_id in source, and then eliminate those that
3633
# we do already have.
3634
# this is slow on high latency connection to self, but as this
3635
# disk format scales terribly for push anyway due to rewriting
3636
# inventory.weave, this is considered acceptable.
3638
if revision_id is not None:
3639
source_ids = self.source.get_ancestry(revision_id)
3640
if source_ids[0] is not None:
3641
raise AssertionError()
3644
source_ids = self.source._all_possible_ids()
3645
source_ids_set = set(source_ids)
3646
# source_ids is the worst possible case we may need to pull.
3647
# now we want to filter source_ids against what we actually
3648
# have in target, but don't try to check for existence where we know
3649
# we do not have a revision as that would be pointless.
3650
target_ids = set(self.target._all_possible_ids())
3651
possibly_present_revisions = target_ids.intersection(source_ids_set)
3652
actually_present_revisions = set(
3653
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3654
required_revisions = source_ids_set.difference(actually_present_revisions)
3655
if revision_id is not None:
3656
# we used get_ancestry to determine source_ids then we are assured all
3657
# revisions referenced are present as they are installed in topological order.
3658
# and the tip revision was validated by get_ancestry.
3659
result_set = required_revisions
3661
# if we just grabbed the possibly available ids, then
3662
# we only have an estimate of whats available and need to validate
3663
# that against the revision records.
3665
self.source._eliminate_revisions_not_present(required_revisions))
3666
return self.source.revision_ids_to_search_result(result_set)
3669
class InterKnitRepo(InterSameDataRepository):
3670
"""Optimised code paths between Knit based repositories."""
3673
def _get_repo_format_to_test(self):
3674
from bzrlib.repofmt import knitrepo
3675
return knitrepo.RepositoryFormatKnit1()
3678
def is_compatible(source, target):
3679
"""Be compatible with known Knit formats.
3681
We don't test for the stores being of specific types because that
3682
could lead to confusing results, and there is no need to be
3685
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3687
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3688
isinstance(target._format, RepositoryFormatKnit))
3689
except AttributeError:
3691
return are_knits and InterRepository._same_model(source, target)
3694
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3695
"""See InterRepository.missing_revision_ids()."""
3696
if revision_id is not None:
3697
source_ids = self.source.get_ancestry(revision_id)
3698
if source_ids[0] is not None:
3699
raise AssertionError()
3702
source_ids = self.source.all_revision_ids()
3703
source_ids_set = set(source_ids)
3704
# source_ids is the worst possible case we may need to pull.
3705
# now we want to filter source_ids against what we actually
3706
# have in target, but don't try to check for existence where we know
3707
# we do not have a revision as that would be pointless.
3708
target_ids = set(self.target.all_revision_ids())
3709
possibly_present_revisions = target_ids.intersection(source_ids_set)
3710
actually_present_revisions = set(
3711
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3712
required_revisions = source_ids_set.difference(actually_present_revisions)
3713
if revision_id is not None:
3714
# we used get_ancestry to determine source_ids then we are assured all
3715
# revisions referenced are present as they are installed in topological order.
3716
# and the tip revision was validated by get_ancestry.
3717
result_set = required_revisions
3719
# if we just grabbed the possibly available ids, then
3720
# we only have an estimate of whats available and need to validate
3721
# that against the revision records.
3723
self.source._eliminate_revisions_not_present(required_revisions))
3724
return self.source.revision_ids_to_search_result(result_set)
3727
class InterDifferingSerializer(InterRepository):
3730
def _get_repo_format_to_test(self):
3734
def is_compatible(source, target):
3735
"""Be compatible with Knit2 source and Knit3 target"""
3736
# This is redundant with format.check_conversion_target(), however that
3737
# raises an exception, and we just want to say "False" as in we won't
3738
# support converting between these formats.
3739
if 'IDS_never' in debug.debug_flags:
3741
if source.supports_rich_root() and not target.supports_rich_root():
3743
if (source._format.supports_tree_reference
3744
and not target._format.supports_tree_reference):
3746
if target._fallback_repositories and target._format.supports_chks:
3747
# IDS doesn't know how to copy CHKs for the parent inventories it
3748
# adds to stacked repos.
3750
if 'IDS_always' in debug.debug_flags:
3752
# Only use this code path for local source and target. IDS does far
3753
# too much IO (both bandwidth and roundtrips) over a network.
3754
if not source.bzrdir.transport.base.startswith('file:///'):
3756
if not target.bzrdir.transport.base.startswith('file:///'):
3760
def _get_trees(self, revision_ids, cache):
3762
for rev_id in revision_ids:
3764
possible_trees.append((rev_id, cache[rev_id]))
3766
# Not cached, but inventory might be present anyway.
3768
tree = self.source.revision_tree(rev_id)
3769
except errors.NoSuchRevision:
3770
# Nope, parent is ghost.
3773
cache[rev_id] = tree
3774
possible_trees.append((rev_id, tree))
3775
return possible_trees
3777
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3778
"""Get the best delta and base for this revision.
3780
:return: (basis_id, delta)
3783
# Generate deltas against each tree, to find the shortest.
3784
texts_possibly_new_in_tree = set()
3785
for basis_id, basis_tree in possible_trees:
3786
delta = tree.inventory._make_delta(basis_tree.inventory)
3787
for old_path, new_path, file_id, new_entry in delta:
3788
if new_path is None:
3789
# This file_id isn't present in the new rev, so we don't
3793
# Rich roots are handled elsewhere...
3795
kind = new_entry.kind
3796
if kind != 'directory' and kind != 'file':
3797
# No text record associated with this inventory entry.
3799
# This is a directory or file that has changed somehow.
3800
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3801
deltas.append((len(delta), basis_id, delta))
3803
return deltas[0][1:]
3805
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3806
"""Find all parent revisions that are absent, but for which the
3807
inventory is present, and copy those inventories.
3809
This is necessary to preserve correctness when the source is stacked
3810
without fallbacks configured. (Note that in cases like upgrade the
3811
source may be not have _fallback_repositories even though it is
3815
for parents in parent_map.values():
3816
parent_revs.update(parents)
3817
present_parents = self.source.get_parent_map(parent_revs)
3818
absent_parents = set(parent_revs).difference(present_parents)
3819
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3820
(rev_id,) for rev_id in absent_parents)
3821
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3822
for parent_tree in self.source.revision_trees(parent_inv_ids):
3823
current_revision_id = parent_tree.get_revision_id()
3824
parents_parents_keys = parent_invs_keys_for_stacking[
3825
(current_revision_id,)]
3826
parents_parents = [key[-1] for key in parents_parents_keys]
3827
basis_id = _mod_revision.NULL_REVISION
3828
basis_tree = self.source.revision_tree(basis_id)
3829
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3830
self.target.add_inventory_by_delta(
3831
basis_id, delta, current_revision_id, parents_parents)
3832
cache[current_revision_id] = parent_tree
3834
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3835
"""Fetch across a few revisions.
3837
:param revision_ids: The revisions to copy
3838
:param basis_id: The revision_id of a tree that must be in cache, used
3839
as a basis for delta when no other base is available
3840
:param cache: A cache of RevisionTrees that we can use.
3841
:param a_graph: A Graph object to determine the heads() of the
3842
rich-root data stream.
3843
:return: The revision_id of the last converted tree. The RevisionTree
3844
for it will be in cache
3846
# Walk though all revisions; get inventory deltas, copy referenced
3847
# texts that delta references, insert the delta, revision and
3849
root_keys_to_create = set()
3852
pending_revisions = []
3853
parent_map = self.source.get_parent_map(revision_ids)
3854
self._fetch_parent_invs_for_stacking(parent_map, cache)
3855
self.source._safe_to_return_from_cache = True
3856
for tree in self.source.revision_trees(revision_ids):
3857
# Find a inventory delta for this revision.
3858
# Find text entries that need to be copied, too.
3859
current_revision_id = tree.get_revision_id()
3860
parent_ids = parent_map.get(current_revision_id, ())
3861
parent_trees = self._get_trees(parent_ids, cache)
3862
possible_trees = list(parent_trees)
3863
if len(possible_trees) == 0:
3864
# There either aren't any parents, or the parents are ghosts,
3865
# so just use the last converted tree.
3866
possible_trees.append((basis_id, cache[basis_id]))
3867
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3869
revision = self.source.get_revision(current_revision_id)
3870
pending_deltas.append((basis_id, delta,
3871
current_revision_id, revision.parent_ids))
3872
if self._converting_to_rich_root:
3873
self._revision_id_to_root_id[current_revision_id] = \
3875
# Determine which texts are in present in this revision but not in
3876
# any of the available parents.
3877
texts_possibly_new_in_tree = set()
3878
for old_path, new_path, file_id, entry in delta:
3879
if new_path is None:
3880
# This file_id isn't present in the new rev
3884
if not self.target.supports_rich_root():
3885
# The target doesn't support rich root, so we don't
3888
if self._converting_to_rich_root:
3889
# This can't be copied normally, we have to insert
3891
root_keys_to_create.add((file_id, entry.revision))
3894
texts_possibly_new_in_tree.add((file_id, entry.revision))
3895
for basis_id, basis_tree in possible_trees:
3896
basis_inv = basis_tree.inventory
3897
for file_key in list(texts_possibly_new_in_tree):
3898
file_id, file_revision = file_key
3900
entry = basis_inv[file_id]
3901
except errors.NoSuchId:
3903
if entry.revision == file_revision:
3904
texts_possibly_new_in_tree.remove(file_key)
3905
text_keys.update(texts_possibly_new_in_tree)
3906
pending_revisions.append(revision)
3907
cache[current_revision_id] = tree
3908
basis_id = current_revision_id
3909
self.source._safe_to_return_from_cache = False
3911
from_texts = self.source.texts
3912
to_texts = self.target.texts
3913
if root_keys_to_create:
3914
root_stream = _mod_fetch._new_root_data_stream(
3915
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3916
self.source, graph=a_graph)
3917
to_texts.insert_record_stream(root_stream)
3918
to_texts.insert_record_stream(from_texts.get_record_stream(
3919
text_keys, self.target._format._fetch_order,
3920
not self.target._format._fetch_uses_deltas))
3921
# insert inventory deltas
3922
for delta in pending_deltas:
3923
self.target.add_inventory_by_delta(*delta)
3924
if self.target._fallback_repositories:
3925
# Make sure this stacked repository has all the parent inventories
3926
# for the new revisions that we are about to insert. We do this
3927
# before adding the revisions so that no revision is added until
3928
# all the inventories it may depend on are added.
3929
# Note that this is overzealous, as we may have fetched these in an
3932
revision_ids = set()
3933
for revision in pending_revisions:
3934
revision_ids.add(revision.revision_id)
3935
parent_ids.update(revision.parent_ids)
3936
parent_ids.difference_update(revision_ids)
3937
parent_ids.discard(_mod_revision.NULL_REVISION)
3938
parent_map = self.source.get_parent_map(parent_ids)
3939
# we iterate over parent_map and not parent_ids because we don't
3940
# want to try copying any revision which is a ghost
3941
for parent_tree in self.source.revision_trees(parent_map):
3942
current_revision_id = parent_tree.get_revision_id()
3943
parents_parents = parent_map[current_revision_id]
3944
possible_trees = self._get_trees(parents_parents, cache)
3945
if len(possible_trees) == 0:
3946
# There either aren't any parents, or the parents are
3947
# ghosts, so just use the last converted tree.
3948
possible_trees.append((basis_id, cache[basis_id]))
3949
basis_id, delta = self._get_delta_for_revision(parent_tree,
3950
parents_parents, possible_trees)
3951
self.target.add_inventory_by_delta(
3952
basis_id, delta, current_revision_id, parents_parents)
3953
# insert signatures and revisions
3954
for revision in pending_revisions:
3956
signature = self.source.get_signature_text(
3957
revision.revision_id)
3958
self.target.add_signature_text(revision.revision_id,
3960
except errors.NoSuchRevision:
3962
self.target.add_revision(revision.revision_id, revision)
3965
def _fetch_all_revisions(self, revision_ids, pb):
3966
"""Fetch everything for the list of revisions.
3968
:param revision_ids: The list of revisions to fetch. Must be in
3970
:param pb: A ProgressTask
3973
basis_id, basis_tree = self._get_basis(revision_ids[0])
3975
cache = lru_cache.LRUCache(100)
3976
cache[basis_id] = basis_tree
3977
del basis_tree # We don't want to hang on to it here
3979
if self._converting_to_rich_root and len(revision_ids) > 100:
3980
a_graph = _mod_fetch._get_rich_root_heads_graph(self.source,
3985
for offset in range(0, len(revision_ids), batch_size):
3986
self.target.start_write_group()
3988
pb.update('Transferring revisions', offset,
3990
batch = revision_ids[offset:offset+batch_size]
3991
basis_id = self._fetch_batch(batch, basis_id, cache,
3994
self.source._safe_to_return_from_cache = False
3995
self.target.abort_write_group()
3998
hint = self.target.commit_write_group()
4001
if hints and self.target._format.pack_compresses:
4002
self.target.pack(hint=hints)
4003
pb.update('Transferring revisions', len(revision_ids),
4007
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
4009
"""See InterRepository.fetch()."""
4010
if fetch_spec is not None:
4011
raise AssertionError("Not implemented yet...")
4012
# See <https://launchpad.net/bugs/456077> asking for a warning here
4014
# nb this is only active for local-local fetches; other things using
4016
trace.warning("Fetching between repositories with different formats\n"
4018
"This may take some time. Upgrade the branches to the same format \n"
4019
"for better results.\n"
4020
% (self.source._format, self.target._format))
4021
if (not self.source.supports_rich_root()
4022
and self.target.supports_rich_root()):
4023
self._converting_to_rich_root = True
4024
self._revision_id_to_root_id = {}
4026
self._converting_to_rich_root = False
4027
revision_ids = self.target.search_missing_revision_ids(self.source,
4028
revision_id, find_ghosts=find_ghosts).get_keys()
4029
if not revision_ids:
4031
revision_ids = tsort.topo_sort(
4032
self.source.get_graph().get_parent_map(revision_ids))
4033
if not revision_ids:
4035
# Walk though all revisions; get inventory deltas, copy referenced
4036
# texts that delta references, insert the delta, revision and
4039
my_pb = ui.ui_factory.nested_progress_bar()
4042
symbol_versioning.warn(
4043
symbol_versioning.deprecated_in((1, 14, 0))
4044
% "pb parameter to fetch()")
4047
self._fetch_all_revisions(revision_ids, pb)
4049
if my_pb is not None:
4051
return len(revision_ids), 0
4053
def _get_basis(self, first_revision_id):
4054
"""Get a revision and tree which exists in the target.
4056
This assumes that first_revision_id is selected for transmission
4057
because all other ancestors are already present. If we can't find an
4058
ancestor we fall back to NULL_REVISION since we know that is safe.
4060
:return: (basis_id, basis_tree)
4062
first_rev = self.source.get_revision(first_revision_id)
4064
basis_id = first_rev.parent_ids[0]
4065
# only valid as a basis if the target has it
4066
self.target.get_revision(basis_id)
4067
# Try to get a basis tree - if its a ghost it will hit the
4068
# NoSuchRevision case.
4069
basis_tree = self.source.revision_tree(basis_id)
4070
except (IndexError, errors.NoSuchRevision):
4071
basis_id = _mod_revision.NULL_REVISION
4072
basis_tree = self.source.revision_tree(basis_id)
4073
return basis_id, basis_tree
4076
InterRepository.register_optimiser(InterDifferingSerializer)
4077
InterRepository.register_optimiser(InterSameDataRepository)
4078
InterRepository.register_optimiser(InterWeaveRepo)
4079
InterRepository.register_optimiser(InterKnitRepo)
4082
class CopyConverter(object):
4083
"""A repository conversion tool which just performs a copy of the content.
4085
This is slow but quite reliable.
4088
def __init__(self, target_format):
4089
"""Create a CopyConverter.
4091
:param target_format: The format the resulting repository should be.
4093
self.target_format = target_format
4095
def convert(self, repo, pb):
4096
"""Perform the conversion of to_convert, giving feedback via pb.
4098
:param to_convert: The disk object to convert.
4099
:param pb: a progress bar to use for progress information.
4104
# this is only useful with metadir layouts - separated repo content.
4105
# trigger an assertion if not such
4106
repo._format.get_format_string()
4107
self.repo_dir = repo.bzrdir
4108
self.step('Moving repository to repository.backup')
4109
self.repo_dir.transport.move('repository', 'repository.backup')
4110
backup_transport = self.repo_dir.transport.clone('repository.backup')
4111
repo._format.check_conversion_target(self.target_format)
4112
self.source_repo = repo._format.open(self.repo_dir,
4114
_override_transport=backup_transport)
4115
self.step('Creating new repository')
4116
converted = self.target_format.initialize(self.repo_dir,
4117
self.source_repo.is_shared())
4118
converted.lock_write()
4120
self.step('Copying content')
4121
self.source_repo.copy_content_into(converted)
4124
self.step('Deleting old repository content')
4125
self.repo_dir.transport.delete_tree('repository.backup')
4126
ui.ui_factory.note('repository converted')
4128
def step(self, message):
4129
"""Update the pb by a step."""
4131
self.pb.update(message, self.count, self.total)
4143
def _unescaper(match, _map=_unescape_map):
4144
code = match.group(1)
4148
if not code.startswith('#'):
4150
return unichr(int(code[1:])).encode('utf8')
4156
def _unescape_xml(data):
4157
"""Unescape predefined XML entities in a string of data."""
4159
if _unescape_re is None:
4160
_unescape_re = re.compile('\&([^;]*);')
4161
return _unescape_re.sub(_unescaper, data)
4164
class _VersionedFileChecker(object):
4166
def __init__(self, repository, text_key_references=None, ancestors=None):
4167
self.repository = repository
4168
self.text_index = self.repository._generate_text_key_index(
4169
text_key_references=text_key_references, ancestors=ancestors)
4171
def calculate_file_version_parents(self, text_key):
4172
"""Calculate the correct parents for a file version according to
4175
parent_keys = self.text_index[text_key]
4176
if parent_keys == [_mod_revision.NULL_REVISION]:
4178
return tuple(parent_keys)
4180
def check_file_version_parents(self, texts, progress_bar=None):
4181
"""Check the parents stored in a versioned file are correct.
4183
It also detects file versions that are not referenced by their
4184
corresponding revision's inventory.
4186
:returns: A tuple of (wrong_parents, dangling_file_versions).
4187
wrong_parents is a dict mapping {revision_id: (stored_parents,
4188
correct_parents)} for each revision_id where the stored parents
4189
are not correct. dangling_file_versions is a set of (file_id,
4190
revision_id) tuples for versions that are present in this versioned
4191
file, but not used by the corresponding inventory.
4193
local_progress = None
4194
if progress_bar is None:
4195
local_progress = ui.ui_factory.nested_progress_bar()
4196
progress_bar = local_progress
4198
return self._check_file_version_parents(texts, progress_bar)
4201
local_progress.finished()
4203
def _check_file_version_parents(self, texts, progress_bar):
4204
"""See check_file_version_parents."""
4206
self.file_ids = set([file_id for file_id, _ in
4207
self.text_index.iterkeys()])
4208
# text keys is now grouped by file_id
4209
n_versions = len(self.text_index)
4210
progress_bar.update('loading text store', 0, n_versions)
4211
parent_map = self.repository.texts.get_parent_map(self.text_index)
4212
# On unlistable transports this could well be empty/error...
4213
text_keys = self.repository.texts.keys()
4214
unused_keys = frozenset(text_keys) - set(self.text_index)
4215
for num, key in enumerate(self.text_index.iterkeys()):
4216
progress_bar.update('checking text graph', num, n_versions)
4217
correct_parents = self.calculate_file_version_parents(key)
4219
knit_parents = parent_map[key]
4220
except errors.RevisionNotPresent:
4223
if correct_parents != knit_parents:
4224
wrong_parents[key] = (knit_parents, correct_parents)
4225
return wrong_parents, unused_keys
4228
def _old_get_graph(repository, revision_id):
4229
"""DO NOT USE. That is all. I'm serious."""
4230
graph = repository.get_graph()
4231
revision_graph = dict(((key, value) for key, value in
4232
graph.iter_ancestry([revision_id]) if value is not None))
4233
return _strip_NULL_ghosts(revision_graph)
4236
def _strip_NULL_ghosts(revision_graph):
4237
"""Also don't use this. more compatibility code for unmigrated clients."""
4238
# Filter ghosts, and null:
4239
if _mod_revision.NULL_REVISION in revision_graph:
4240
del revision_graph[_mod_revision.NULL_REVISION]
4241
for key, parents in revision_graph.items():
4242
revision_graph[key] = tuple(parent for parent in parents if parent
4244
return revision_graph
4247
class StreamSink(object):
4248
"""An object that can insert a stream into a repository.
4250
This interface handles the complexity of reserialising inventories and
4251
revisions from different formats, and allows unidirectional insertion into
4252
stacked repositories without looking for the missing basis parents
4256
def __init__(self, target_repo):
4257
self.target_repo = target_repo
4259
def insert_stream(self, stream, src_format, resume_tokens):
4260
"""Insert a stream's content into the target repository.
4262
:param src_format: a bzr repository format.
4264
:return: a list of resume tokens and an iterable of keys additional
4265
items required before the insertion can be completed.
4267
self.target_repo.lock_write()
4270
self.target_repo.resume_write_group(resume_tokens)
4273
self.target_repo.start_write_group()
4276
# locked_insert_stream performs a commit|suspend.
4277
return self._locked_insert_stream(stream, src_format, is_resume)
4279
self.target_repo.abort_write_group(suppress_errors=True)
4282
self.target_repo.unlock()
4284
def _locked_insert_stream(self, stream, src_format, is_resume):
4285
to_serializer = self.target_repo._format._serializer
4286
src_serializer = src_format._serializer
4288
if to_serializer == src_serializer:
4289
# If serializers match and the target is a pack repository, set the
4290
# write cache size on the new pack. This avoids poor performance
4291
# on transports where append is unbuffered (such as
4292
# RemoteTransport). This is safe to do because nothing should read
4293
# back from the target repository while a stream with matching
4294
# serialization is being inserted.
4295
# The exception is that a delta record from the source that should
4296
# be a fulltext may need to be expanded by the target (see
4297
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4298
# explicitly flush any buffered writes first in that rare case.
4300
new_pack = self.target_repo._pack_collection._new_pack
4301
except AttributeError:
4302
# Not a pack repository
4305
new_pack.set_write_cache_size(1024*1024)
4306
for substream_type, substream in stream:
4307
if 'stream' in debug.debug_flags:
4308
mutter('inserting substream: %s', substream_type)
4309
if substream_type == 'texts':
4310
self.target_repo.texts.insert_record_stream(substream)
4311
elif substream_type == 'inventories':
4312
if src_serializer == to_serializer:
4313
self.target_repo.inventories.insert_record_stream(
4316
self._extract_and_insert_inventories(
4317
substream, src_serializer)
4318
elif substream_type == 'inventory-deltas':
4319
self._extract_and_insert_inventory_deltas(
4320
substream, src_serializer)
4321
elif substream_type == 'chk_bytes':
4322
# XXX: This doesn't support conversions, as it assumes the
4323
# conversion was done in the fetch code.
4324
self.target_repo.chk_bytes.insert_record_stream(substream)
4325
elif substream_type == 'revisions':
4326
# This may fallback to extract-and-insert more often than
4327
# required if the serializers are different only in terms of
4329
if src_serializer == to_serializer:
4330
self.target_repo.revisions.insert_record_stream(
4333
self._extract_and_insert_revisions(substream,
4335
elif substream_type == 'signatures':
4336
self.target_repo.signatures.insert_record_stream(substream)
4338
raise AssertionError('kaboom! %s' % (substream_type,))
4339
# Done inserting data, and the missing_keys calculations will try to
4340
# read back from the inserted data, so flush the writes to the new pack
4341
# (if this is pack format).
4342
if new_pack is not None:
4343
new_pack._write_data('', flush=True)
4344
# Find all the new revisions (including ones from resume_tokens)
4345
missing_keys = self.target_repo.get_missing_parent_inventories(
4346
check_for_missing_texts=is_resume)
4348
for prefix, versioned_file in (
4349
('texts', self.target_repo.texts),
4350
('inventories', self.target_repo.inventories),
4351
('revisions', self.target_repo.revisions),
4352
('signatures', self.target_repo.signatures),
4353
('chk_bytes', self.target_repo.chk_bytes),
4355
if versioned_file is None:
4357
# TODO: key is often going to be a StaticTuple object
4358
# I don't believe we can define a method by which
4359
# (prefix,) + StaticTuple will work, though we could
4360
# define a StaticTuple.sq_concat that would allow you to
4361
# pass in either a tuple or a StaticTuple as the second
4362
# object, so instead we could have:
4363
# StaticTuple(prefix) + key here...
4364
missing_keys.update((prefix,) + key for key in
4365
versioned_file.get_missing_compression_parent_keys())
4366
except NotImplementedError:
4367
# cannot even attempt suspending, and missing would have failed
4368
# during stream insertion.
4369
missing_keys = set()
4372
# suspend the write group and tell the caller what we is
4373
# missing. We know we can suspend or else we would not have
4374
# entered this code path. (All repositories that can handle
4375
# missing keys can handle suspending a write group).
4376
write_group_tokens = self.target_repo.suspend_write_group()
4377
return write_group_tokens, missing_keys
4378
hint = self.target_repo.commit_write_group()
4379
if (to_serializer != src_serializer and
4380
self.target_repo._format.pack_compresses):
4381
self.target_repo.pack(hint=hint)
4384
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4385
target_rich_root = self.target_repo._format.rich_root_data
4386
target_tree_refs = self.target_repo._format.supports_tree_reference
4387
for record in substream:
4388
# Insert the delta directly
4389
inventory_delta_bytes = record.get_bytes_as('fulltext')
4390
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4392
parse_result = deserialiser.parse_text_bytes(
4393
inventory_delta_bytes)
4394
except inventory_delta.IncompatibleInventoryDelta, err:
4395
trace.mutter("Incompatible delta: %s", err.msg)
4396
raise errors.IncompatibleRevision(self.target_repo._format)
4397
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4398
revision_id = new_id
4399
parents = [key[0] for key in record.parents]
4400
self.target_repo.add_inventory_by_delta(
4401
basis_id, inv_delta, revision_id, parents)
4403
def _extract_and_insert_inventories(self, substream, serializer,
4405
"""Generate a new inventory versionedfile in target, converting data.
4407
The inventory is retrieved from the source, (deserializing it), and
4408
stored in the target (reserializing it in a different format).
4410
target_rich_root = self.target_repo._format.rich_root_data
4411
target_tree_refs = self.target_repo._format.supports_tree_reference
4412
for record in substream:
4413
# It's not a delta, so it must be a fulltext in the source
4414
# serializer's format.
4415
bytes = record.get_bytes_as('fulltext')
4416
revision_id = record.key[0]
4417
inv = serializer.read_inventory_from_string(bytes, revision_id)
4418
parents = [key[0] for key in record.parents]
4419
self.target_repo.add_inventory(revision_id, inv, parents)
4420
# No need to keep holding this full inv in memory when the rest of
4421
# the substream is likely to be all deltas.
4424
def _extract_and_insert_revisions(self, substream, serializer):
4425
for record in substream:
4426
bytes = record.get_bytes_as('fulltext')
4427
revision_id = record.key[0]
4428
rev = serializer.read_revision_from_string(bytes)
4429
if rev.revision_id != revision_id:
4430
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4431
self.target_repo.add_revision(revision_id, rev)
4434
if self.target_repo._format._fetch_reconcile:
4435
self.target_repo.reconcile()
4438
class StreamSource(object):
4439
"""A source of a stream for fetching between repositories."""
4441
def __init__(self, from_repository, to_format):
4442
"""Create a StreamSource streaming from from_repository."""
4443
self.from_repository = from_repository
4444
self.to_format = to_format
4446
def delta_on_metadata(self):
4447
"""Return True if delta's are permitted on metadata streams.
4449
That is on revisions and signatures.
4451
src_serializer = self.from_repository._format._serializer
4452
target_serializer = self.to_format._serializer
4453
return (self.to_format._fetch_uses_deltas and
4454
src_serializer == target_serializer)
4456
def _fetch_revision_texts(self, revs):
4457
# fetch signatures first and then the revision texts
4458
# may need to be a InterRevisionStore call here.
4459
from_sf = self.from_repository.signatures
4460
# A missing signature is just skipped.
4461
keys = [(rev_id,) for rev_id in revs]
4462
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4464
self.to_format._fetch_order,
4465
not self.to_format._fetch_uses_deltas))
4466
# If a revision has a delta, this is actually expanded inside the
4467
# insert_record_stream code now, which is an alternate fix for
4469
from_rf = self.from_repository.revisions
4470
revisions = from_rf.get_record_stream(
4472
self.to_format._fetch_order,
4473
not self.delta_on_metadata())
4474
return [('signatures', signatures), ('revisions', revisions)]
4476
def _generate_root_texts(self, revs):
4477
"""This will be called by get_stream between fetching weave texts and
4478
fetching the inventory weave.
4480
if self._rich_root_upgrade():
4481
return _mod_fetch.Inter1and2Helper(
4482
self.from_repository).generate_root_texts(revs)
4486
def get_stream(self, search):
4488
revs = search.get_keys()
4489
graph = self.from_repository.get_graph()
4490
revs = tsort.topo_sort(graph.get_parent_map(revs))
4491
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4493
for knit_kind, file_id, revisions in data_to_fetch:
4494
if knit_kind != phase:
4496
# Make a new progress bar for this phase
4497
if knit_kind == "file":
4498
# Accumulate file texts
4499
text_keys.extend([(file_id, revision) for revision in
4501
elif knit_kind == "inventory":
4502
# Now copy the file texts.
4503
from_texts = self.from_repository.texts
4504
yield ('texts', from_texts.get_record_stream(
4505
text_keys, self.to_format._fetch_order,
4506
not self.to_format._fetch_uses_deltas))
4507
# Cause an error if a text occurs after we have done the
4510
# Before we process the inventory we generate the root
4511
# texts (if necessary) so that the inventories references
4513
for _ in self._generate_root_texts(revs):
4515
# we fetch only the referenced inventories because we do not
4516
# know for unselected inventories whether all their required
4517
# texts are present in the other repository - it could be
4519
for info in self._get_inventory_stream(revs):
4521
elif knit_kind == "signatures":
4522
# Nothing to do here; this will be taken care of when
4523
# _fetch_revision_texts happens.
4525
elif knit_kind == "revisions":
4526
for record in self._fetch_revision_texts(revs):
4529
raise AssertionError("Unknown knit kind %r" % knit_kind)
4531
def get_stream_for_missing_keys(self, missing_keys):
4532
# missing keys can only occur when we are byte copying and not
4533
# translating (because translation means we don't send
4534
# unreconstructable deltas ever).
4536
keys['texts'] = set()
4537
keys['revisions'] = set()
4538
keys['inventories'] = set()
4539
keys['chk_bytes'] = set()
4540
keys['signatures'] = set()
4541
for key in missing_keys:
4542
keys[key[0]].add(key[1:])
4543
if len(keys['revisions']):
4544
# If we allowed copying revisions at this point, we could end up
4545
# copying a revision without copying its required texts: a
4546
# violation of the requirements for repository integrity.
4547
raise AssertionError(
4548
'cannot copy revisions to fill in missing deltas %s' % (
4549
keys['revisions'],))
4550
for substream_kind, keys in keys.iteritems():
4551
vf = getattr(self.from_repository, substream_kind)
4552
if vf is None and keys:
4553
raise AssertionError(
4554
"cannot fill in keys for a versioned file we don't"
4555
" have: %s needs %s" % (substream_kind, keys))
4557
# No need to stream something we don't have
4559
if substream_kind == 'inventories':
4560
# Some missing keys are genuinely ghosts, filter those out.
4561
present = self.from_repository.inventories.get_parent_map(keys)
4562
revs = [key[0] for key in present]
4563
# Get the inventory stream more-or-less as we do for the
4564
# original stream; there's no reason to assume that records
4565
# direct from the source will be suitable for the sink. (Think
4566
# e.g. 2a -> 1.9-rich-root).
4567
for info in self._get_inventory_stream(revs, missing=True):
4571
# Ask for full texts always so that we don't need more round trips
4572
# after this stream.
4573
# Some of the missing keys are genuinely ghosts, so filter absent
4574
# records. The Sink is responsible for doing another check to
4575
# ensure that ghosts don't introduce missing data for future
4577
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4578
self.to_format._fetch_order, True))
4579
yield substream_kind, stream
4581
def inventory_fetch_order(self):
4582
if self._rich_root_upgrade():
4583
return 'topological'
4585
return self.to_format._fetch_order
4587
def _rich_root_upgrade(self):
4588
return (not self.from_repository._format.rich_root_data and
4589
self.to_format.rich_root_data)
4591
def _get_inventory_stream(self, revision_ids, missing=False):
4592
from_format = self.from_repository._format
4593
if (from_format.supports_chks and self.to_format.supports_chks and
4594
from_format.network_name() == self.to_format.network_name()):
4595
raise AssertionError(
4596
"this case should be handled by GroupCHKStreamSource")
4597
elif 'forceinvdeltas' in debug.debug_flags:
4598
return self._get_convertable_inventory_stream(revision_ids,
4599
delta_versus_null=missing)
4600
elif from_format.network_name() == self.to_format.network_name():
4602
return self._get_simple_inventory_stream(revision_ids,
4604
elif (not from_format.supports_chks and not self.to_format.supports_chks
4605
and from_format._serializer == self.to_format._serializer):
4606
# Essentially the same format.
4607
return self._get_simple_inventory_stream(revision_ids,
4610
# Any time we switch serializations, we want to use an
4611
# inventory-delta based approach.
4612
return self._get_convertable_inventory_stream(revision_ids,
4613
delta_versus_null=missing)
4615
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4616
# NB: This currently reopens the inventory weave in source;
4617
# using a single stream interface instead would avoid this.
4618
from_weave = self.from_repository.inventories
4620
delta_closure = True
4622
delta_closure = not self.delta_on_metadata()
4623
yield ('inventories', from_weave.get_record_stream(
4624
[(rev_id,) for rev_id in revision_ids],
4625
self.inventory_fetch_order(), delta_closure))
4627
def _get_convertable_inventory_stream(self, revision_ids,
4628
delta_versus_null=False):
4629
# The source is using CHKs, but the target either doesn't or it has a
4630
# different serializer. The StreamSink code expects to be able to
4631
# convert on the target, so we need to put bytes-on-the-wire that can
4632
# be converted. That means inventory deltas (if the remote is <1.19,
4633
# RemoteStreamSink will fallback to VFS to insert the deltas).
4634
yield ('inventory-deltas',
4635
self._stream_invs_as_deltas(revision_ids,
4636
delta_versus_null=delta_versus_null))
4638
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4639
"""Return a stream of inventory-deltas for the given rev ids.
4641
:param revision_ids: The list of inventories to transmit
4642
:param delta_versus_null: Don't try to find a minimal delta for this
4643
entry, instead compute the delta versus the NULL_REVISION. This
4644
effectively streams a complete inventory. Used for stuff like
4645
filling in missing parents, etc.
4647
from_repo = self.from_repository
4648
revision_keys = [(rev_id,) for rev_id in revision_ids]
4649
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4650
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4652
inventories = self.from_repository.iter_inventories(
4653
revision_ids, 'topological')
4654
format = from_repo._format
4655
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4656
inventory_cache = lru_cache.LRUCache(50)
4657
null_inventory = from_repo.revision_tree(
4658
_mod_revision.NULL_REVISION).inventory
4659
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4660
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4661
# repo back into a non-rich-root repo ought to be allowed)
4662
serializer = inventory_delta.InventoryDeltaSerializer(
4663
versioned_root=format.rich_root_data,
4664
tree_references=format.supports_tree_reference)
4665
for inv in inventories:
4666
key = (inv.revision_id,)
4667
parent_keys = parent_map.get(key, ())
4669
if not delta_versus_null and parent_keys:
4670
# The caller did not ask for complete inventories and we have
4671
# some parents that we can delta against. Make a delta against
4672
# each parent so that we can find the smallest.
4673
parent_ids = [parent_key[0] for parent_key in parent_keys]
4674
for parent_id in parent_ids:
4675
if parent_id not in invs_sent_so_far:
4676
# We don't know that the remote side has this basis, so
4679
if parent_id == _mod_revision.NULL_REVISION:
4680
parent_inv = null_inventory
4682
parent_inv = inventory_cache.get(parent_id, None)
4683
if parent_inv is None:
4684
parent_inv = from_repo.get_inventory(parent_id)
4685
candidate_delta = inv._make_delta(parent_inv)
4686
if (delta is None or
4687
len(delta) > len(candidate_delta)):
4688
delta = candidate_delta
4689
basis_id = parent_id
4691
# Either none of the parents ended up being suitable, or we
4692
# were asked to delta against NULL
4693
basis_id = _mod_revision.NULL_REVISION
4694
delta = inv._make_delta(null_inventory)
4695
invs_sent_so_far.add(inv.revision_id)
4696
inventory_cache[inv.revision_id] = inv
4697
delta_serialized = ''.join(
4698
serializer.delta_to_lines(basis_id, key[-1], delta))
4699
yield versionedfile.FulltextContentFactory(
4700
key, parent_keys, None, delta_serialized)
4703
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4704
stop_revision=None):
4705
"""Extend the partial history to include a given index
4707
If a stop_index is supplied, stop when that index has been reached.
4708
If a stop_revision is supplied, stop when that revision is
4709
encountered. Otherwise, stop when the beginning of history is
4712
:param stop_index: The index which should be present. When it is
4713
present, history extension will stop.
4714
:param stop_revision: The revision id which should be present. When
4715
it is encountered, history extension will stop.
4717
start_revision = partial_history_cache[-1]
4718
iterator = repo.iter_reverse_revision_history(start_revision)
4720
#skip the last revision in the list
4723
if (stop_index is not None and
4724
len(partial_history_cache) > stop_index):
4726
if partial_history_cache[-1] == stop_revision:
4728
revision_id = iterator.next()
4729
partial_history_cache.append(revision_id)
4730
except StopIteration:
added
removed
bzrlib/repository.py
