1
# Copyright (C) 2005, 2006, 2007, 2008, 2009 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
from bzrlib.lazy_import import lazy_import
18
lazy_import(globals(), """
40
revision as _mod_revision,
46
from bzrlib.bundle import serializer
47
from bzrlib.revisiontree import RevisionTree
48
from bzrlib.store.versioned import VersionedFileStore
49
from bzrlib.testament import Testament
52
from bzrlib.decorators import needs_read_lock, needs_write_lock, only_raises
53
from bzrlib.inter import InterObject
54
from bzrlib.inventory import (
60
from bzrlib.lock import _RelockDebugMixin
61
from bzrlib import registry
62
from bzrlib.trace import (
63
log_exception_quietly, note, mutter, mutter_callsite, warning)
66
# Old formats display a warning, but only once
67
_deprecation_warning_done = False
70
class CommitBuilder(object):
71
"""Provides an interface to build up a commit.
73
This allows describing a tree to be committed without needing to
74
know the internals of the format of the repository.
77
# all clients should supply tree roots.
78
record_root_entry = True
79
# the default CommitBuilder does not manage trees whose root is versioned.
80
_versioned_root = False
82
def __init__(self, repository, parents, config, timestamp=None,
83
timezone=None, committer=None, revprops=None,
85
"""Initiate a CommitBuilder.
87
:param repository: Repository to commit to.
88
:param parents: Revision ids of the parents of the new revision.
89
:param config: Configuration to use.
90
:param timestamp: Optional timestamp recorded for commit.
91
:param timezone: Optional timezone for timestamp.
92
:param committer: Optional committer to set for commit.
93
:param revprops: Optional dictionary of revision properties.
94
:param revision_id: Optional revision id.
99
self._committer = self._config.username()
101
self._committer = committer
103
self.new_inventory = Inventory(None)
104
self._new_revision_id = revision_id
105
self.parents = parents
106
self.repository = repository
109
if revprops is not None:
110
self._validate_revprops(revprops)
111
self._revprops.update(revprops)
113
if timestamp is None:
114
timestamp = time.time()
115
# Restrict resolution to 1ms
116
self._timestamp = round(timestamp, 3)
119
self._timezone = osutils.local_time_offset()
121
self._timezone = int(timezone)
123
self._generate_revision_if_needed()
124
self.__heads = graph.HeadsCache(repository.get_graph()).heads
125
self._basis_delta = []
126
# API compatibility, older code that used CommitBuilder did not call
127
# .record_delete(), which means the delta that is computed would not be
128
# valid. Callers that will call record_delete() should call
129
# .will_record_deletes() to indicate that.
130
self._recording_deletes = False
131
# memo'd check for no-op commits.
132
self._any_changes = False
134
def any_changes(self):
135
"""Return True if any entries were changed.
137
This includes merge-only changes. It is the core for the --unchanged
140
:return: True if any changes have occured.
142
return self._any_changes
144
def _validate_unicode_text(self, text, context):
145
"""Verify things like commit messages don't have bogus characters."""
147
raise ValueError('Invalid value for %s: %r' % (context, text))
149
def _validate_revprops(self, revprops):
150
for key, value in revprops.iteritems():
151
# We know that the XML serializers do not round trip '\r'
152
# correctly, so refuse to accept them
153
if not isinstance(value, basestring):
154
raise ValueError('revision property (%s) is not a valid'
155
' (unicode) string: %r' % (key, value))
156
self._validate_unicode_text(value,
157
'revision property (%s)' % (key,))
159
def commit(self, message):
160
"""Make the actual commit.
162
:return: The revision id of the recorded revision.
164
self._validate_unicode_text(message, 'commit message')
165
rev = _mod_revision.Revision(
166
timestamp=self._timestamp,
167
timezone=self._timezone,
168
committer=self._committer,
170
inventory_sha1=self.inv_sha1,
171
revision_id=self._new_revision_id,
172
properties=self._revprops)
173
rev.parent_ids = self.parents
174
self.repository.add_revision(self._new_revision_id, rev,
175
self.new_inventory, self._config)
176
self.repository.commit_write_group()
177
return self._new_revision_id
180
"""Abort the commit that is being built.
182
self.repository.abort_write_group()
184
def revision_tree(self):
185
"""Return the tree that was just committed.
187
After calling commit() this can be called to get a RevisionTree
188
representing the newly committed tree. This is preferred to
189
calling Repository.revision_tree() because that may require
190
deserializing the inventory, while we already have a copy in
193
if self.new_inventory is None:
194
self.new_inventory = self.repository.get_inventory(
195
self._new_revision_id)
196
return RevisionTree(self.repository, self.new_inventory,
197
self._new_revision_id)
199
def finish_inventory(self):
200
"""Tell the builder that the inventory is finished.
202
:return: The inventory id in the repository, which can be used with
203
repository.get_inventory.
205
if self.new_inventory is None:
206
# an inventory delta was accumulated without creating a new
208
basis_id = self.basis_delta_revision
209
# We ignore the 'inventory' returned by add_inventory_by_delta
210
# because self.new_inventory is used to hint to the rest of the
211
# system what code path was taken
212
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
213
basis_id, self._basis_delta, self._new_revision_id,
216
if self.new_inventory.root is None:
217
raise AssertionError('Root entry should be supplied to'
218
' record_entry_contents, as of bzr 0.10.')
219
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
220
self.new_inventory.revision_id = self._new_revision_id
221
self.inv_sha1 = self.repository.add_inventory(
222
self._new_revision_id,
226
return self._new_revision_id
228
def _gen_revision_id(self):
229
"""Return new revision-id."""
230
return generate_ids.gen_revision_id(self._config.username(),
233
def _generate_revision_if_needed(self):
234
"""Create a revision id if None was supplied.
236
If the repository can not support user-specified revision ids
237
they should override this function and raise CannotSetRevisionId
238
if _new_revision_id is not None.
240
:raises: CannotSetRevisionId
242
if self._new_revision_id is None:
243
self._new_revision_id = self._gen_revision_id()
244
self.random_revid = True
246
self.random_revid = False
248
def _heads(self, file_id, revision_ids):
249
"""Calculate the graph heads for revision_ids in the graph of file_id.
251
This can use either a per-file graph or a global revision graph as we
252
have an identity relationship between the two graphs.
254
return self.__heads(revision_ids)
256
def _check_root(self, ie, parent_invs, tree):
257
"""Helper for record_entry_contents.
259
:param ie: An entry being added.
260
:param parent_invs: The inventories of the parent revisions of the
262
:param tree: The tree that is being committed.
264
# In this revision format, root entries have no knit or weave When
265
# serializing out to disk and back in root.revision is always
267
ie.revision = self._new_revision_id
269
def _require_root_change(self, tree):
270
"""Enforce an appropriate root object change.
272
This is called once when record_iter_changes is called, if and only if
273
the root was not in the delta calculated by record_iter_changes.
275
:param tree: The tree which is being committed.
277
# NB: if there are no parents then this method is not called, so no
278
# need to guard on parents having length.
279
entry = entry_factory['directory'](tree.path2id(''), '',
281
entry.revision = self._new_revision_id
282
self._basis_delta.append(('', '', entry.file_id, entry))
284
def _get_delta(self, ie, basis_inv, path):
285
"""Get a delta against the basis inventory for ie."""
286
if ie.file_id not in basis_inv:
288
result = (None, path, ie.file_id, ie)
289
self._basis_delta.append(result)
291
elif ie != basis_inv[ie.file_id]:
293
# TODO: avoid tis id2path call.
294
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
295
self._basis_delta.append(result)
301
def get_basis_delta(self):
302
"""Return the complete inventory delta versus the basis inventory.
304
This has been built up with the calls to record_delete and
305
record_entry_contents. The client must have already called
306
will_record_deletes() to indicate that they will be generating a
309
:return: An inventory delta, suitable for use with apply_delta, or
310
Repository.add_inventory_by_delta, etc.
312
if not self._recording_deletes:
313
raise AssertionError("recording deletes not activated.")
314
return self._basis_delta
316
def record_delete(self, path, file_id):
317
"""Record that a delete occured against a basis tree.
319
This is an optional API - when used it adds items to the basis_delta
320
being accumulated by the commit builder. It cannot be called unless the
321
method will_record_deletes() has been called to inform the builder that
322
a delta is being supplied.
324
:param path: The path of the thing deleted.
325
:param file_id: The file id that was deleted.
327
if not self._recording_deletes:
328
raise AssertionError("recording deletes not activated.")
329
delta = (path, None, file_id, None)
330
self._basis_delta.append(delta)
331
self._any_changes = True
334
def will_record_deletes(self):
335
"""Tell the commit builder that deletes are being notified.
337
This enables the accumulation of an inventory delta; for the resulting
338
commit to be valid, deletes against the basis MUST be recorded via
339
builder.record_delete().
341
self._recording_deletes = True
343
basis_id = self.parents[0]
345
basis_id = _mod_revision.NULL_REVISION
346
self.basis_delta_revision = basis_id
348
def record_entry_contents(self, ie, parent_invs, path, tree,
350
"""Record the content of ie from tree into the commit if needed.
352
Side effect: sets ie.revision when unchanged
354
:param ie: An inventory entry present in the commit.
355
:param parent_invs: The inventories of the parent revisions of the
357
:param path: The path the entry is at in the tree.
358
:param tree: The tree which contains this entry and should be used to
360
:param content_summary: Summary data from the tree about the paths
361
content - stat, length, exec, sha/link target. This is only
362
accessed when the entry has a revision of None - that is when it is
363
a candidate to commit.
364
:return: A tuple (change_delta, version_recorded, fs_hash).
365
change_delta is an inventory_delta change for this entry against
366
the basis tree of the commit, or None if no change occured against
368
version_recorded is True if a new version of the entry has been
369
recorded. For instance, committing a merge where a file was only
370
changed on the other side will return (delta, False).
371
fs_hash is either None, or the hash details for the path (currently
372
a tuple of the contents sha1 and the statvalue returned by
373
tree.get_file_with_stat()).
375
if self.new_inventory.root is None:
376
if ie.parent_id is not None:
377
raise errors.RootMissing()
378
self._check_root(ie, parent_invs, tree)
379
if ie.revision is None:
380
kind = content_summary[0]
382
# ie is carried over from a prior commit
384
# XXX: repository specific check for nested tree support goes here - if
385
# the repo doesn't want nested trees we skip it ?
386
if (kind == 'tree-reference' and
387
not self.repository._format.supports_tree_reference):
388
# mismatch between commit builder logic and repository:
389
# this needs the entry creation pushed down into the builder.
390
raise NotImplementedError('Missing repository subtree support.')
391
self.new_inventory.add(ie)
393
# TODO: slow, take it out of the inner loop.
395
basis_inv = parent_invs[0]
397
basis_inv = Inventory(root_id=None)
399
# ie.revision is always None if the InventoryEntry is considered
400
# for committing. We may record the previous parents revision if the
401
# content is actually unchanged against a sole head.
402
if ie.revision is not None:
403
if not self._versioned_root and path == '':
404
# repositories that do not version the root set the root's
405
# revision to the new commit even when no change occurs (more
406
# specifically, they do not record a revision on the root; and
407
# the rev id is assigned to the root during deserialisation -
408
# this masks when a change may have occurred against the basis.
409
# To match this we always issue a delta, because the revision
410
# of the root will always be changing.
411
if ie.file_id in basis_inv:
412
delta = (basis_inv.id2path(ie.file_id), path,
416
delta = (None, path, ie.file_id, ie)
417
self._basis_delta.append(delta)
418
return delta, False, None
420
# we don't need to commit this, because the caller already
421
# determined that an existing revision of this file is
422
# appropriate. If its not being considered for committing then
423
# it and all its parents to the root must be unaltered so
424
# no-change against the basis.
425
if ie.revision == self._new_revision_id:
426
raise AssertionError("Impossible situation, a skipped "
427
"inventory entry (%r) claims to be modified in this "
428
"commit (%r).", (ie, self._new_revision_id))
429
return None, False, None
430
# XXX: Friction: parent_candidates should return a list not a dict
431
# so that we don't have to walk the inventories again.
432
parent_candiate_entries = ie.parent_candidates(parent_invs)
433
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
435
for inv in parent_invs:
436
if ie.file_id in inv:
437
old_rev = inv[ie.file_id].revision
438
if old_rev in head_set:
439
heads.append(inv[ie.file_id].revision)
440
head_set.remove(inv[ie.file_id].revision)
443
# now we check to see if we need to write a new record to the
445
# We write a new entry unless there is one head to the ancestors, and
446
# the kind-derived content is unchanged.
448
# Cheapest check first: no ancestors, or more the one head in the
449
# ancestors, we write a new node.
453
# There is a single head, look it up for comparison
454
parent_entry = parent_candiate_entries[heads[0]]
455
# if the non-content specific data has changed, we'll be writing a
457
if (parent_entry.parent_id != ie.parent_id or
458
parent_entry.name != ie.name):
460
# now we need to do content specific checks:
462
# if the kind changed the content obviously has
463
if kind != parent_entry.kind:
465
# Stat cache fingerprint feedback for the caller - None as we usually
466
# don't generate one.
469
if content_summary[2] is None:
470
raise ValueError("Files must not have executable = None")
472
# We can't trust a check of the file length because of content
474
if (# if the exec bit has changed we have to store:
475
parent_entry.executable != content_summary[2]):
477
elif parent_entry.text_sha1 == content_summary[3]:
478
# all meta and content is unchanged (using a hash cache
479
# hit to check the sha)
480
ie.revision = parent_entry.revision
481
ie.text_size = parent_entry.text_size
482
ie.text_sha1 = parent_entry.text_sha1
483
ie.executable = parent_entry.executable
484
return self._get_delta(ie, basis_inv, path), False, None
486
# Either there is only a hash change(no hash cache entry,
487
# or same size content change), or there is no change on
489
# Provide the parent's hash to the store layer, so that the
490
# content is unchanged we will not store a new node.
491
nostore_sha = parent_entry.text_sha1
493
# We want to record a new node regardless of the presence or
494
# absence of a content change in the file.
496
ie.executable = content_summary[2]
497
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
499
text = file_obj.read()
503
ie.text_sha1, ie.text_size = self._add_text_to_weave(
504
ie.file_id, text, heads, nostore_sha)
505
# Let the caller know we generated a stat fingerprint.
506
fingerprint = (ie.text_sha1, stat_value)
507
except errors.ExistingContent:
508
# Turns out that the file content was unchanged, and we were
509
# only going to store a new node if it was changed. Carry over
511
ie.revision = parent_entry.revision
512
ie.text_size = parent_entry.text_size
513
ie.text_sha1 = parent_entry.text_sha1
514
ie.executable = parent_entry.executable
515
return self._get_delta(ie, basis_inv, path), False, None
516
elif kind == 'directory':
518
# all data is meta here, nothing specific to directory, so
520
ie.revision = parent_entry.revision
521
return self._get_delta(ie, basis_inv, path), False, None
522
self._add_text_to_weave(ie.file_id, '', heads, None)
523
elif kind == 'symlink':
524
current_link_target = content_summary[3]
526
# symlink target is not generic metadata, check if it has
528
if current_link_target != parent_entry.symlink_target:
531
# unchanged, carry over.
532
ie.revision = parent_entry.revision
533
ie.symlink_target = parent_entry.symlink_target
534
return self._get_delta(ie, basis_inv, path), False, None
535
ie.symlink_target = current_link_target
536
self._add_text_to_weave(ie.file_id, '', heads, None)
537
elif kind == 'tree-reference':
539
if content_summary[3] != parent_entry.reference_revision:
542
# unchanged, carry over.
543
ie.reference_revision = parent_entry.reference_revision
544
ie.revision = parent_entry.revision
545
return self._get_delta(ie, basis_inv, path), False, None
546
ie.reference_revision = content_summary[3]
547
if ie.reference_revision is None:
548
raise AssertionError("invalid content_summary for nested tree: %r"
549
% (content_summary,))
550
self._add_text_to_weave(ie.file_id, '', heads, None)
552
raise NotImplementedError('unknown kind')
553
ie.revision = self._new_revision_id
554
self._any_changes = True
555
return self._get_delta(ie, basis_inv, path), True, fingerprint
557
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
558
_entry_factory=entry_factory):
559
"""Record a new tree via iter_changes.
561
:param tree: The tree to obtain text contents from for changed objects.
562
:param basis_revision_id: The revision id of the tree the iter_changes
563
has been generated against. Currently assumed to be the same
564
as self.parents[0] - if it is not, errors may occur.
565
:param iter_changes: An iter_changes iterator with the changes to apply
566
to basis_revision_id. The iterator must not include any items with
567
a current kind of None - missing items must be either filtered out
568
or errored-on beefore record_iter_changes sees the item.
569
:param _entry_factory: Private method to bind entry_factory locally for
571
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
574
# Create an inventory delta based on deltas between all the parents and
575
# deltas between all the parent inventories. We use inventory delta's
576
# between the inventory objects because iter_changes masks
577
# last-changed-field only changes.
579
# file_id -> change map, change is fileid, paths, changed, versioneds,
580
# parents, names, kinds, executables
582
# {file_id -> revision_id -> inventory entry, for entries in parent
583
# trees that are not parents[0]
587
revtrees = list(self.repository.revision_trees(self.parents))
588
except errors.NoSuchRevision:
589
# one or more ghosts, slow path.
591
for revision_id in self.parents:
593
revtrees.append(self.repository.revision_tree(revision_id))
594
except errors.NoSuchRevision:
596
basis_revision_id = _mod_revision.NULL_REVISION
598
revtrees.append(self.repository.revision_tree(
599
_mod_revision.NULL_REVISION))
600
# The basis inventory from a repository
602
basis_inv = revtrees[0].inventory
604
basis_inv = self.repository.revision_tree(
605
_mod_revision.NULL_REVISION).inventory
606
if len(self.parents) > 0:
607
if basis_revision_id != self.parents[0] and not ghost_basis:
609
"arbitrary basis parents not yet supported with merges")
610
for revtree in revtrees[1:]:
611
for change in revtree.inventory._make_delta(basis_inv):
612
if change[1] is None:
613
# Not present in this parent.
615
if change[2] not in merged_ids:
616
if change[0] is not None:
617
basis_entry = basis_inv[change[2]]
618
merged_ids[change[2]] = [
620
basis_entry.revision,
623
parent_entries[change[2]] = {
625
basis_entry.revision:basis_entry,
627
change[3].revision:change[3],
630
merged_ids[change[2]] = [change[3].revision]
631
parent_entries[change[2]] = {change[3].revision:change[3]}
633
merged_ids[change[2]].append(change[3].revision)
634
parent_entries[change[2]][change[3].revision] = change[3]
637
# Setup the changes from the tree:
638
# changes maps file_id -> (change, [parent revision_ids])
640
for change in iter_changes:
641
# This probably looks up in basis_inv way to much.
642
if change[1][0] is not None:
643
head_candidate = [basis_inv[change[0]].revision]
646
changes[change[0]] = change, merged_ids.get(change[0],
648
unchanged_merged = set(merged_ids) - set(changes)
649
# Extend the changes dict with synthetic changes to record merges of
651
for file_id in unchanged_merged:
652
# Record a merged version of these items that did not change vs the
653
# basis. This can be either identical parallel changes, or a revert
654
# of a specific file after a merge. The recorded content will be
655
# that of the current tree (which is the same as the basis), but
656
# the per-file graph will reflect a merge.
657
# NB:XXX: We are reconstructing path information we had, this
658
# should be preserved instead.
659
# inv delta change: (file_id, (path_in_source, path_in_target),
660
# changed_content, versioned, parent, name, kind,
663
basis_entry = basis_inv[file_id]
664
except errors.NoSuchId:
665
# a change from basis->some_parents but file_id isn't in basis
666
# so was new in the merge, which means it must have changed
667
# from basis -> current, and as it hasn't the add was reverted
668
# by the user. So we discard this change.
672
(basis_inv.id2path(file_id), tree.id2path(file_id)),
674
(basis_entry.parent_id, basis_entry.parent_id),
675
(basis_entry.name, basis_entry.name),
676
(basis_entry.kind, basis_entry.kind),
677
(basis_entry.executable, basis_entry.executable))
678
changes[file_id] = (change, merged_ids[file_id])
679
# changes contains tuples with the change and a set of inventory
680
# candidates for the file.
682
# old_path, new_path, file_id, new_inventory_entry
683
seen_root = False # Is the root in the basis delta?
684
inv_delta = self._basis_delta
685
modified_rev = self._new_revision_id
686
for change, head_candidates in changes.values():
687
if change[3][1]: # versioned in target.
688
# Several things may be happening here:
689
# We may have a fork in the per-file graph
690
# - record a change with the content from tree
691
# We may have a change against < all trees
692
# - carry over the tree that hasn't changed
693
# We may have a change against all trees
694
# - record the change with the content from tree
697
entry = _entry_factory[kind](file_id, change[5][1],
699
head_set = self._heads(change[0], set(head_candidates))
702
for head_candidate in head_candidates:
703
if head_candidate in head_set:
704
heads.append(head_candidate)
705
head_set.remove(head_candidate)
708
# Could be a carry-over situation:
709
parent_entry_revs = parent_entries.get(file_id, None)
710
if parent_entry_revs:
711
parent_entry = parent_entry_revs.get(heads[0], None)
714
if parent_entry is None:
715
# The parent iter_changes was called against is the one
716
# that is the per-file head, so any change is relevant
717
# iter_changes is valid.
718
carry_over_possible = False
720
# could be a carry over situation
721
# A change against the basis may just indicate a merge,
722
# we need to check the content against the source of the
723
# merge to determine if it was changed after the merge
725
if (parent_entry.kind != entry.kind or
726
parent_entry.parent_id != entry.parent_id or
727
parent_entry.name != entry.name):
728
# Metadata common to all entries has changed
729
# against per-file parent
730
carry_over_possible = False
732
carry_over_possible = True
733
# per-type checks for changes against the parent_entry
736
# Cannot be a carry-over situation
737
carry_over_possible = False
738
# Populate the entry in the delta
740
# XXX: There is still a small race here: If someone reverts the content of a file
741
# after iter_changes examines and decides it has changed,
742
# we will unconditionally record a new version even if some
743
# other process reverts it while commit is running (with
744
# the revert happening after iter_changes did it's
747
entry.executable = True
749
entry.executable = False
750
if (carry_over_possible and
751
parent_entry.executable == entry.executable):
752
# Check the file length, content hash after reading
754
nostore_sha = parent_entry.text_sha1
757
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
759
text = file_obj.read()
763
entry.text_sha1, entry.text_size = self._add_text_to_weave(
764
file_id, text, heads, nostore_sha)
765
yield file_id, change[1][1], (entry.text_sha1, stat_value)
766
except errors.ExistingContent:
767
# No content change against a carry_over parent
768
# Perhaps this should also yield a fs hash update?
770
entry.text_size = parent_entry.text_size
771
entry.text_sha1 = parent_entry.text_sha1
772
elif kind == 'symlink':
774
entry.symlink_target = tree.get_symlink_target(file_id)
775
if (carry_over_possible and
776
parent_entry.symlink_target == entry.symlink_target):
779
self._add_text_to_weave(change[0], '', heads, None)
780
elif kind == 'directory':
781
if carry_over_possible:
784
# Nothing to set on the entry.
785
# XXX: split into the Root and nonRoot versions.
786
if change[1][1] != '' or self.repository.supports_rich_root():
787
self._add_text_to_weave(change[0], '', heads, None)
788
elif kind == 'tree-reference':
789
if not self.repository._format.supports_tree_reference:
790
# This isn't quite sane as an error, but we shouldn't
791
# ever see this code path in practice: tree's don't
792
# permit references when the repo doesn't support tree
794
raise errors.UnsupportedOperation(tree.add_reference,
796
reference_revision = tree.get_reference_revision(change[0])
797
entry.reference_revision = reference_revision
798
if (carry_over_possible and
799
parent_entry.reference_revision == reference_revision):
802
self._add_text_to_weave(change[0], '', heads, None)
804
raise AssertionError('unknown kind %r' % kind)
806
entry.revision = modified_rev
808
entry.revision = parent_entry.revision
811
new_path = change[1][1]
812
inv_delta.append((change[1][0], new_path, change[0], entry))
815
self.new_inventory = None
817
# This should perhaps be guarded by a check that the basis we
818
# commit against is the basis for the commit and if not do a delta
820
self._any_changes = True
822
# housekeeping root entry changes do not affect no-change commits.
823
self._require_root_change(tree)
824
self.basis_delta_revision = basis_revision_id
826
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
827
parent_keys = tuple([(file_id, parent) for parent in parents])
828
return self.repository.texts._add_text(
829
(file_id, self._new_revision_id), parent_keys, new_text,
830
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
833
class RootCommitBuilder(CommitBuilder):
834
"""This commitbuilder actually records the root id"""
836
# the root entry gets versioned properly by this builder.
837
_versioned_root = True
839
def _check_root(self, ie, parent_invs, tree):
840
"""Helper for record_entry_contents.
842
:param ie: An entry being added.
843
:param parent_invs: The inventories of the parent revisions of the
845
:param tree: The tree that is being committed.
848
def _require_root_change(self, tree):
849
"""Enforce an appropriate root object change.
851
This is called once when record_iter_changes is called, if and only if
852
the root was not in the delta calculated by record_iter_changes.
854
:param tree: The tree which is being committed.
856
# versioned roots do not change unless the tree found a change.
859
######################################################################
863
class Repository(_RelockDebugMixin):
864
"""Repository holding history for one or more branches.
866
The repository holds and retrieves historical information including
867
revisions and file history. It's normally accessed only by the Branch,
868
which views a particular line of development through that history.
870
The Repository builds on top of some byte storage facilies (the revisions,
871
signatures, inventories, texts and chk_bytes attributes) and a Transport,
872
which respectively provide byte storage and a means to access the (possibly
875
The byte storage facilities are addressed via tuples, which we refer to
876
as 'keys' throughout the code base. Revision_keys, inventory_keys and
877
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
878
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
879
byte string made up of a hash identifier and a hash value.
880
We use this interface because it allows low friction with the underlying
881
code that implements disk indices, network encoding and other parts of
884
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
885
the serialised revisions for the repository. This can be used to obtain
886
revision graph information or to access raw serialised revisions.
887
The result of trying to insert data into the repository via this store
888
is undefined: it should be considered read-only except for implementors
890
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
891
the serialised signatures for the repository. This can be used to
892
obtain access to raw serialised signatures. The result of trying to
893
insert data into the repository via this store is undefined: it should
894
be considered read-only except for implementors of repositories.
895
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
896
the serialised inventories for the repository. This can be used to
897
obtain unserialised inventories. The result of trying to insert data
898
into the repository via this store is undefined: it should be
899
considered read-only except for implementors of repositories.
900
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
901
texts of files and directories for the repository. This can be used to
902
obtain file texts or file graphs. Note that Repository.iter_file_bytes
903
is usually a better interface for accessing file texts.
904
The result of trying to insert data into the repository via this store
905
is undefined: it should be considered read-only except for implementors
907
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
908
any data the repository chooses to store or have indexed by its hash.
909
The result of trying to insert data into the repository via this store
910
is undefined: it should be considered read-only except for implementors
912
:ivar _transport: Transport for file access to repository, typically
913
pointing to .bzr/repository.
916
# What class to use for a CommitBuilder. Often its simpler to change this
917
# in a Repository class subclass rather than to override
918
# get_commit_builder.
919
_commit_builder_class = CommitBuilder
920
# The search regex used by xml based repositories to determine what things
921
# where changed in a single commit.
922
_file_ids_altered_regex = lazy_regex.lazy_compile(
923
r'file_id="(?P<file_id>[^"]+)"'
924
r'.* revision="(?P<revision_id>[^"]+)"'
927
def abort_write_group(self, suppress_errors=False):
928
"""Commit the contents accrued within the current write group.
930
:param suppress_errors: if true, abort_write_group will catch and log
931
unexpected errors that happen during the abort, rather than
932
allowing them to propagate. Defaults to False.
934
:seealso: start_write_group.
936
if self._write_group is not self.get_transaction():
937
# has an unlock or relock occured ?
940
'(suppressed) mismatched lock context and write group. %r, %r',
941
self._write_group, self.get_transaction())
943
raise errors.BzrError(
944
'mismatched lock context and write group. %r, %r' %
945
(self._write_group, self.get_transaction()))
947
self._abort_write_group()
948
except Exception, exc:
949
self._write_group = None
950
if not suppress_errors:
952
mutter('abort_write_group failed')
953
log_exception_quietly()
954
note('bzr: ERROR (ignored): %s', exc)
955
self._write_group = None
957
def _abort_write_group(self):
958
"""Template method for per-repository write group cleanup.
960
This is called during abort before the write group is considered to be
961
finished and should cleanup any internal state accrued during the write
962
group. There is no requirement that data handed to the repository be
963
*not* made available - this is not a rollback - but neither should any
964
attempt be made to ensure that data added is fully commited. Abort is
965
invoked when an error has occured so futher disk or network operations
966
may not be possible or may error and if possible should not be
970
def add_fallback_repository(self, repository):
971
"""Add a repository to use for looking up data not held locally.
973
:param repository: A repository.
975
if not self._format.supports_external_lookups:
976
raise errors.UnstackableRepositoryFormat(self._format, self.base)
978
# This repository will call fallback.unlock() when we transition to
979
# the unlocked state, so we make sure to increment the lock count
980
repository.lock_read()
981
self._check_fallback_repository(repository)
982
self._fallback_repositories.append(repository)
983
self.texts.add_fallback_versioned_files(repository.texts)
984
self.inventories.add_fallback_versioned_files(repository.inventories)
985
self.revisions.add_fallback_versioned_files(repository.revisions)
986
self.signatures.add_fallback_versioned_files(repository.signatures)
987
if self.chk_bytes is not None:
988
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
990
def _check_fallback_repository(self, repository):
991
"""Check that this repository can fallback to repository safely.
993
Raise an error if not.
995
:param repository: A repository to fallback to.
997
return InterRepository._assert_same_model(self, repository)
999
def add_inventory(self, revision_id, inv, parents):
1000
"""Add the inventory inv to the repository as revision_id.
1002
:param parents: The revision ids of the parents that revision_id
1003
is known to have and are in the repository already.
1005
:returns: The validator(which is a sha1 digest, though what is sha'd is
1006
repository format specific) of the serialized inventory.
1008
if not self.is_in_write_group():
1009
raise AssertionError("%r not in write group" % (self,))
1010
_mod_revision.check_not_reserved_id(revision_id)
1011
if not (inv.revision_id is None or inv.revision_id == revision_id):
1012
raise AssertionError(
1013
"Mismatch between inventory revision"
1014
" id and insertion revid (%r, %r)"
1015
% (inv.revision_id, revision_id))
1016
if inv.root is None:
1017
raise AssertionError()
1018
return self._add_inventory_checked(revision_id, inv, parents)
1020
def _add_inventory_checked(self, revision_id, inv, parents):
1021
"""Add inv to the repository after checking the inputs.
1023
This function can be overridden to allow different inventory styles.
1025
:seealso: add_inventory, for the contract.
1027
inv_lines = self._serialise_inventory_to_lines(inv)
1028
return self._inventory_add_lines(revision_id, parents,
1029
inv_lines, check_content=False)
1031
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1032
parents, basis_inv=None, propagate_caches=False):
1033
"""Add a new inventory expressed as a delta against another revision.
1035
See the inventory developers documentation for the theory behind
1038
:param basis_revision_id: The inventory id the delta was created
1039
against. (This does not have to be a direct parent.)
1040
:param delta: The inventory delta (see Inventory.apply_delta for
1042
:param new_revision_id: The revision id that the inventory is being
1044
:param parents: The revision ids of the parents that revision_id is
1045
known to have and are in the repository already. These are supplied
1046
for repositories that depend on the inventory graph for revision
1047
graph access, as well as for those that pun ancestry with delta
1049
:param basis_inv: The basis inventory if it is already known,
1051
:param propagate_caches: If True, the caches for this inventory are
1052
copied to and updated for the result if possible.
1054
:returns: (validator, new_inv)
1055
The validator(which is a sha1 digest, though what is sha'd is
1056
repository format specific) of the serialized inventory, and the
1057
resulting inventory.
1059
if not self.is_in_write_group():
1060
raise AssertionError("%r not in write group" % (self,))
1061
_mod_revision.check_not_reserved_id(new_revision_id)
1062
basis_tree = self.revision_tree(basis_revision_id)
1063
basis_tree.lock_read()
1065
# Note that this mutates the inventory of basis_tree, which not all
1066
# inventory implementations may support: A better idiom would be to
1067
# return a new inventory, but as there is no revision tree cache in
1068
# repository this is safe for now - RBC 20081013
1069
if basis_inv is None:
1070
basis_inv = basis_tree.inventory
1071
basis_inv.apply_delta(delta)
1072
basis_inv.revision_id = new_revision_id
1073
return (self.add_inventory(new_revision_id, basis_inv, parents),
1078
def _inventory_add_lines(self, revision_id, parents, lines,
1079
check_content=True):
1080
"""Store lines in inv_vf and return the sha1 of the inventory."""
1081
parents = [(parent,) for parent in parents]
1082
result = self.inventories.add_lines((revision_id,), parents, lines,
1083
check_content=check_content)[0]
1084
self.inventories._access.flush()
1087
def add_revision(self, revision_id, rev, inv=None, config=None):
1088
"""Add rev to the revision store as revision_id.
1090
:param revision_id: the revision id to use.
1091
:param rev: The revision object.
1092
:param inv: The inventory for the revision. if None, it will be looked
1093
up in the inventory storer
1094
:param config: If None no digital signature will be created.
1095
If supplied its signature_needed method will be used
1096
to determine if a signature should be made.
1098
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1100
_mod_revision.check_not_reserved_id(revision_id)
1101
if config is not None and config.signature_needed():
1103
inv = self.get_inventory(revision_id)
1104
plaintext = Testament(rev, inv).as_short_text()
1105
self.store_revision_signature(
1106
gpg.GPGStrategy(config), plaintext, revision_id)
1107
# check inventory present
1108
if not self.inventories.get_parent_map([(revision_id,)]):
1110
raise errors.WeaveRevisionNotPresent(revision_id,
1113
# yes, this is not suitable for adding with ghosts.
1114
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1117
key = (revision_id,)
1118
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1119
self._add_revision(rev)
1121
def _add_revision(self, revision):
1122
text = self._serializer.write_revision_to_string(revision)
1123
key = (revision.revision_id,)
1124
parents = tuple((parent,) for parent in revision.parent_ids)
1125
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1127
def all_revision_ids(self):
1128
"""Returns a list of all the revision ids in the repository.
1130
This is conceptually deprecated because code should generally work on
1131
the graph reachable from a particular revision, and ignore any other
1132
revisions that might be present. There is no direct replacement
1135
if 'evil' in debug.debug_flags:
1136
mutter_callsite(2, "all_revision_ids is linear with history.")
1137
return self._all_revision_ids()
1139
def _all_revision_ids(self):
1140
"""Returns a list of all the revision ids in the repository.
1142
These are in as much topological order as the underlying store can
1145
raise NotImplementedError(self._all_revision_ids)
1147
def break_lock(self):
1148
"""Break a lock if one is present from another instance.
1150
Uses the ui factory to ask for confirmation if the lock may be from
1153
self.control_files.break_lock()
1156
def _eliminate_revisions_not_present(self, revision_ids):
1157
"""Check every revision id in revision_ids to see if we have it.
1159
Returns a set of the present revisions.
1162
graph = self.get_graph()
1163
parent_map = graph.get_parent_map(revision_ids)
1164
# The old API returned a list, should this actually be a set?
1165
return parent_map.keys()
1167
def _check_inventories(self, checker):
1168
"""Check the inventories found from the revision scan.
1170
This is responsible for verifying the sha1 of inventories and
1171
creating a pending_keys set that covers data referenced by inventories.
1173
bar = ui.ui_factory.nested_progress_bar()
1175
self._do_check_inventories(checker, bar)
1179
def _do_check_inventories(self, checker, bar):
1180
"""Helper for _check_inventories."""
1182
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1183
kinds = ['chk_bytes', 'texts']
1184
count = len(checker.pending_keys)
1185
bar.update("inventories", 0, 2)
1186
current_keys = checker.pending_keys
1187
checker.pending_keys = {}
1188
# Accumulate current checks.
1189
for key in current_keys:
1190
if key[0] != 'inventories' and key[0] not in kinds:
1191
checker._report_items.append('unknown key type %r' % (key,))
1192
keys[key[0]].add(key[1:])
1193
if keys['inventories']:
1194
# NB: output order *should* be roughly sorted - topo or
1195
# inverse topo depending on repository - either way decent
1196
# to just delta against. However, pre-CHK formats didn't
1197
# try to optimise inventory layout on disk. As such the
1198
# pre-CHK code path does not use inventory deltas.
1200
for record in self.inventories.check(keys=keys['inventories']):
1201
if record.storage_kind == 'absent':
1202
checker._report_items.append(
1203
'Missing inventory {%s}' % (record.key,))
1205
last_object = self._check_record('inventories', record,
1206
checker, last_object,
1207
current_keys[('inventories',) + record.key])
1208
del keys['inventories']
1211
bar.update("texts", 1)
1212
while (checker.pending_keys or keys['chk_bytes']
1214
# Something to check.
1215
current_keys = checker.pending_keys
1216
checker.pending_keys = {}
1217
# Accumulate current checks.
1218
for key in current_keys:
1219
if key[0] not in kinds:
1220
checker._report_items.append('unknown key type %r' % (key,))
1221
keys[key[0]].add(key[1:])
1222
# Check the outermost kind only - inventories || chk_bytes || texts
1226
for record in getattr(self, kind).check(keys=keys[kind]):
1227
if record.storage_kind == 'absent':
1228
checker._report_items.append(
1229
'Missing %s {%s}' % (kind, record.key,))
1231
last_object = self._check_record(kind, record,
1232
checker, last_object, current_keys[(kind,) + record.key])
1236
def _check_record(self, kind, record, checker, last_object, item_data):
1237
"""Check a single text from this repository."""
1238
if kind == 'inventories':
1239
rev_id = record.key[0]
1240
inv = self.deserialise_inventory(rev_id,
1241
record.get_bytes_as('fulltext'))
1242
if last_object is not None:
1243
delta = inv._make_delta(last_object)
1244
for old_path, path, file_id, ie in delta:
1247
ie.check(checker, rev_id, inv)
1249
for path, ie in inv.iter_entries():
1250
ie.check(checker, rev_id, inv)
1251
if self._format.fast_deltas:
1253
elif kind == 'chk_bytes':
1254
# No code written to check chk_bytes for this repo format.
1255
checker._report_items.append(
1256
'unsupported key type chk_bytes for %s' % (record.key,))
1257
elif kind == 'texts':
1258
self._check_text(record, checker, item_data)
1260
checker._report_items.append(
1261
'unknown key type %s for %s' % (kind, record.key))
1263
def _check_text(self, record, checker, item_data):
1264
"""Check a single text."""
1265
# Check it is extractable.
1266
# TODO: check length.
1267
if record.storage_kind == 'chunked':
1268
chunks = record.get_bytes_as(record.storage_kind)
1269
sha1 = osutils.sha_strings(chunks)
1270
length = sum(map(len, chunks))
1272
content = record.get_bytes_as('fulltext')
1273
sha1 = osutils.sha_string(content)
1274
length = len(content)
1275
if item_data and sha1 != item_data[1]:
1276
checker._report_items.append(
1277
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1278
(record.key, sha1, item_data[1], item_data[2]))
1281
def create(a_bzrdir):
1282
"""Construct the current default format repository in a_bzrdir."""
1283
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1285
def __init__(self, _format, a_bzrdir, control_files):
1286
"""instantiate a Repository.
1288
:param _format: The format of the repository on disk.
1289
:param a_bzrdir: The BzrDir of the repository.
1291
In the future we will have a single api for all stores for
1292
getting file texts, inventories and revisions, then
1293
this construct will accept instances of those things.
1295
super(Repository, self).__init__()
1296
self._format = _format
1297
# the following are part of the public API for Repository:
1298
self.bzrdir = a_bzrdir
1299
self.control_files = control_files
1300
self._transport = control_files._transport
1301
self.base = self._transport.base
1303
self._reconcile_does_inventory_gc = True
1304
self._reconcile_fixes_text_parents = False
1305
self._reconcile_backsup_inventory = True
1306
# not right yet - should be more semantically clear ?
1308
# TODO: make sure to construct the right store classes, etc, depending
1309
# on whether escaping is required.
1310
self._warn_if_deprecated()
1311
self._write_group = None
1312
# Additional places to query for data.
1313
self._fallback_repositories = []
1314
# An InventoryEntry cache, used during deserialization
1315
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1318
if self._fallback_repositories:
1319
return '%s(%r, fallback_repositories=%r)' % (
1320
self.__class__.__name__,
1322
self._fallback_repositories)
1324
return '%s(%r)' % (self.__class__.__name__,
1327
def _has_same_fallbacks(self, other_repo):
1328
"""Returns true if the repositories have the same fallbacks."""
1329
my_fb = self._fallback_repositories
1330
other_fb = other_repo._fallback_repositories
1331
if len(my_fb) != len(other_fb):
1333
for f, g in zip(my_fb, other_fb):
1334
if not f.has_same_location(g):
1338
def has_same_location(self, other):
1339
"""Returns a boolean indicating if this repository is at the same
1340
location as another repository.
1342
This might return False even when two repository objects are accessing
1343
the same physical repository via different URLs.
1345
if self.__class__ is not other.__class__:
1347
return (self._transport.base == other._transport.base)
1349
def is_in_write_group(self):
1350
"""Return True if there is an open write group.
1352
:seealso: start_write_group.
1354
return self._write_group is not None
1356
def is_locked(self):
1357
return self.control_files.is_locked()
1359
def is_write_locked(self):
1360
"""Return True if this object is write locked."""
1361
return self.is_locked() and self.control_files._lock_mode == 'w'
1363
def lock_write(self, token=None):
1364
"""Lock this repository for writing.
1366
This causes caching within the repository obejct to start accumlating
1367
data during reads, and allows a 'write_group' to be obtained. Write
1368
groups must be used for actual data insertion.
1370
:param token: if this is already locked, then lock_write will fail
1371
unless the token matches the existing lock.
1372
:returns: a token if this instance supports tokens, otherwise None.
1373
:raises TokenLockingNotSupported: when a token is given but this
1374
instance doesn't support using token locks.
1375
:raises MismatchedToken: if the specified token doesn't match the token
1376
of the existing lock.
1377
:seealso: start_write_group.
1379
A token should be passed in if you know that you have locked the object
1380
some other way, and need to synchronise this object's state with that
1383
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1385
locked = self.is_locked()
1386
result = self.control_files.lock_write(token=token)
1388
self._note_lock('w')
1389
for repo in self._fallback_repositories:
1390
# Writes don't affect fallback repos
1392
self._refresh_data()
1395
def lock_read(self):
1396
locked = self.is_locked()
1397
self.control_files.lock_read()
1399
self._note_lock('r')
1400
for repo in self._fallback_repositories:
1402
self._refresh_data()
1404
def get_physical_lock_status(self):
1405
return self.control_files.get_physical_lock_status()
1407
def leave_lock_in_place(self):
1408
"""Tell this repository not to release the physical lock when this
1411
If lock_write doesn't return a token, then this method is not supported.
1413
self.control_files.leave_in_place()
1415
def dont_leave_lock_in_place(self):
1416
"""Tell this repository to release the physical lock when this
1417
object is unlocked, even if it didn't originally acquire it.
1419
If lock_write doesn't return a token, then this method is not supported.
1421
self.control_files.dont_leave_in_place()
1424
def gather_stats(self, revid=None, committers=None):
1425
"""Gather statistics from a revision id.
1427
:param revid: The revision id to gather statistics from, if None, then
1428
no revision specific statistics are gathered.
1429
:param committers: Optional parameter controlling whether to grab
1430
a count of committers from the revision specific statistics.
1431
:return: A dictionary of statistics. Currently this contains:
1432
committers: The number of committers if requested.
1433
firstrev: A tuple with timestamp, timezone for the penultimate left
1434
most ancestor of revid, if revid is not the NULL_REVISION.
1435
latestrev: A tuple with timestamp, timezone for revid, if revid is
1436
not the NULL_REVISION.
1437
revisions: The total revision count in the repository.
1438
size: An estimate disk size of the repository in bytes.
1441
if revid and committers:
1442
result['committers'] = 0
1443
if revid and revid != _mod_revision.NULL_REVISION:
1445
all_committers = set()
1446
revisions = self.get_ancestry(revid)
1447
# pop the leading None
1449
first_revision = None
1451
# ignore the revisions in the middle - just grab first and last
1452
revisions = revisions[0], revisions[-1]
1453
for revision in self.get_revisions(revisions):
1454
if not first_revision:
1455
first_revision = revision
1457
all_committers.add(revision.committer)
1458
last_revision = revision
1460
result['committers'] = len(all_committers)
1461
result['firstrev'] = (first_revision.timestamp,
1462
first_revision.timezone)
1463
result['latestrev'] = (last_revision.timestamp,
1464
last_revision.timezone)
1466
# now gather global repository information
1467
# XXX: This is available for many repos regardless of listability.
1468
if self.bzrdir.root_transport.listable():
1469
# XXX: do we want to __define len__() ?
1470
# Maybe the versionedfiles object should provide a different
1471
# method to get the number of keys.
1472
result['revisions'] = len(self.revisions.keys())
1473
# result['size'] = t
1476
def find_branches(self, using=False):
1477
"""Find branches underneath this repository.
1479
This will include branches inside other branches.
1481
:param using: If True, list only branches using this repository.
1483
if using and not self.is_shared():
1485
return [self.bzrdir.open_branch()]
1486
except errors.NotBranchError:
1488
class Evaluator(object):
1491
self.first_call = True
1493
def __call__(self, bzrdir):
1494
# On the first call, the parameter is always the bzrdir
1495
# containing the current repo.
1496
if not self.first_call:
1498
repository = bzrdir.open_repository()
1499
except errors.NoRepositoryPresent:
1502
return False, (None, repository)
1503
self.first_call = False
1505
value = (bzrdir.open_branch(), None)
1506
except errors.NotBranchError:
1507
value = (None, None)
1511
for branch, repository in bzrdir.BzrDir.find_bzrdirs(
1512
self.bzrdir.root_transport, evaluate=Evaluator()):
1513
if branch is not None:
1514
branches.append(branch)
1515
if not using and repository is not None:
1516
branches.extend(repository.find_branches())
1520
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
1521
"""Return the revision ids that other has that this does not.
1523
These are returned in topological order.
1525
revision_id: only return revision ids included by revision_id.
1527
return InterRepository.get(other, self).search_missing_revision_ids(
1528
revision_id, find_ghosts)
1532
"""Open the repository rooted at base.
1534
For instance, if the repository is at URL/.bzr/repository,
1535
Repository.open(URL) -> a Repository instance.
1537
control = bzrdir.BzrDir.open(base)
1538
return control.open_repository()
1540
def copy_content_into(self, destination, revision_id=None):
1541
"""Make a complete copy of the content in self into destination.
1543
This is a destructive operation! Do not use it on existing
1546
return InterRepository.get(self, destination).copy_content(revision_id)
1548
def commit_write_group(self):
1549
"""Commit the contents accrued within the current write group.
1551
:seealso: start_write_group.
1553
:return: it may return an opaque hint that can be passed to 'pack'.
1555
if self._write_group is not self.get_transaction():
1556
# has an unlock or relock occured ?
1557
raise errors.BzrError('mismatched lock context %r and '
1559
(self.get_transaction(), self._write_group))
1560
result = self._commit_write_group()
1561
self._write_group = None
1564
def _commit_write_group(self):
1565
"""Template method for per-repository write group cleanup.
1567
This is called before the write group is considered to be
1568
finished and should ensure that all data handed to the repository
1569
for writing during the write group is safely committed (to the
1570
extent possible considering file system caching etc).
1573
def suspend_write_group(self):
1574
raise errors.UnsuspendableWriteGroup(self)
1576
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1577
"""Return the keys of missing inventory parents for revisions added in
1580
A revision is not complete if the inventory delta for that revision
1581
cannot be calculated. Therefore if the parent inventories of a
1582
revision are not present, the revision is incomplete, and e.g. cannot
1583
be streamed by a smart server. This method finds missing inventory
1584
parents for revisions added in this write group.
1586
if not self._format.supports_external_lookups:
1587
# This is only an issue for stacked repositories
1589
if not self.is_in_write_group():
1590
raise AssertionError('not in a write group')
1592
# XXX: We assume that every added revision already has its
1593
# corresponding inventory, so we only check for parent inventories that
1594
# might be missing, rather than all inventories.
1595
parents = set(self.revisions._index.get_missing_parents())
1596
parents.discard(_mod_revision.NULL_REVISION)
1597
unstacked_inventories = self.inventories._index
1598
present_inventories = unstacked_inventories.get_parent_map(
1599
key[-1:] for key in parents)
1600
parents.difference_update(present_inventories)
1601
if len(parents) == 0:
1602
# No missing parent inventories.
1604
if not check_for_missing_texts:
1605
return set(('inventories', rev_id) for (rev_id,) in parents)
1606
# Ok, now we have a list of missing inventories. But these only matter
1607
# if the inventories that reference them are missing some texts they
1608
# appear to introduce.
1609
# XXX: Texts referenced by all added inventories need to be present,
1610
# but at the moment we're only checking for texts referenced by
1611
# inventories at the graph's edge.
1612
key_deps = self.revisions._index._key_dependencies
1613
key_deps.satisfy_refs_for_keys(present_inventories)
1614
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1615
file_ids = self.fileids_altered_by_revision_ids(referrers)
1616
missing_texts = set()
1617
for file_id, version_ids in file_ids.iteritems():
1618
missing_texts.update(
1619
(file_id, version_id) for version_id in version_ids)
1620
present_texts = self.texts.get_parent_map(missing_texts)
1621
missing_texts.difference_update(present_texts)
1622
if not missing_texts:
1623
# No texts are missing, so all revisions and their deltas are
1626
# Alternatively the text versions could be returned as the missing
1627
# keys, but this is likely to be less data.
1628
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1631
def refresh_data(self):
1632
"""Re-read any data needed to to synchronise with disk.
1634
This method is intended to be called after another repository instance
1635
(such as one used by a smart server) has inserted data into the
1636
repository. It may not be called during a write group, but may be
1637
called at any other time.
1639
if self.is_in_write_group():
1640
raise errors.InternalBzrError(
1641
"May not refresh_data while in a write group.")
1642
self._refresh_data()
1644
def resume_write_group(self, tokens):
1645
if not self.is_write_locked():
1646
raise errors.NotWriteLocked(self)
1647
if self._write_group:
1648
raise errors.BzrError('already in a write group')
1649
self._resume_write_group(tokens)
1650
# so we can detect unlock/relock - the write group is now entered.
1651
self._write_group = self.get_transaction()
1653
def _resume_write_group(self, tokens):
1654
raise errors.UnsuspendableWriteGroup(self)
1656
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1658
"""Fetch the content required to construct revision_id from source.
1660
If revision_id is None and fetch_spec is None, then all content is
1663
fetch() may not be used when the repository is in a write group -
1664
either finish the current write group before using fetch, or use
1665
fetch before starting the write group.
1667
:param find_ghosts: Find and copy revisions in the source that are
1668
ghosts in the target (and not reachable directly by walking out to
1669
the first-present revision in target from revision_id).
1670
:param revision_id: If specified, all the content needed for this
1671
revision ID will be copied to the target. Fetch will determine for
1672
itself which content needs to be copied.
1673
:param fetch_spec: If specified, a SearchResult or
1674
PendingAncestryResult that describes which revisions to copy. This
1675
allows copying multiple heads at once. Mutually exclusive with
1678
if fetch_spec is not None and revision_id is not None:
1679
raise AssertionError(
1680
"fetch_spec and revision_id are mutually exclusive.")
1681
if self.is_in_write_group():
1682
raise errors.InternalBzrError(
1683
"May not fetch while in a write group.")
1684
# fast path same-url fetch operations
1685
# TODO: lift out to somewhere common with RemoteRepository
1686
# <https://bugs.edge.launchpad.net/bzr/+bug/401646>
1687
if (self.has_same_location(source)
1688
and fetch_spec is None
1689
and self._has_same_fallbacks(source)):
1690
# check that last_revision is in 'from' and then return a
1692
if (revision_id is not None and
1693
not _mod_revision.is_null(revision_id)):
1694
self.get_revision(revision_id)
1696
# if there is no specific appropriate InterRepository, this will get
1697
# the InterRepository base class, which raises an
1698
# IncompatibleRepositories when asked to fetch.
1699
inter = InterRepository.get(source, self)
1700
return inter.fetch(revision_id=revision_id, pb=pb,
1701
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1703
def create_bundle(self, target, base, fileobj, format=None):
1704
return serializer.write_bundle(self, target, base, fileobj, format)
1706
def get_commit_builder(self, branch, parents, config, timestamp=None,
1707
timezone=None, committer=None, revprops=None,
1709
"""Obtain a CommitBuilder for this repository.
1711
:param branch: Branch to commit to.
1712
:param parents: Revision ids of the parents of the new revision.
1713
:param config: Configuration to use.
1714
:param timestamp: Optional timestamp recorded for commit.
1715
:param timezone: Optional timezone for timestamp.
1716
:param committer: Optional committer to set for commit.
1717
:param revprops: Optional dictionary of revision properties.
1718
:param revision_id: Optional revision id.
1720
if self._fallback_repositories:
1721
raise errors.BzrError("Cannot commit from a lightweight checkout "
1722
"to a stacked branch. See "
1723
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1724
result = self._commit_builder_class(self, parents, config,
1725
timestamp, timezone, committer, revprops, revision_id)
1726
self.start_write_group()
1729
@only_raises(errors.LockNotHeld, errors.LockBroken)
1731
if (self.control_files._lock_count == 1 and
1732
self.control_files._lock_mode == 'w'):
1733
if self._write_group is not None:
1734
self.abort_write_group()
1735
self.control_files.unlock()
1736
raise errors.BzrError(
1737
'Must end write groups before releasing write locks.')
1738
self.control_files.unlock()
1739
if self.control_files._lock_count == 0:
1740
self._inventory_entry_cache.clear()
1741
for repo in self._fallback_repositories:
1745
def clone(self, a_bzrdir, revision_id=None):
1746
"""Clone this repository into a_bzrdir using the current format.
1748
Currently no check is made that the format of this repository and
1749
the bzrdir format are compatible. FIXME RBC 20060201.
1751
:return: The newly created destination repository.
1753
# TODO: deprecate after 0.16; cloning this with all its settings is
1754
# probably not very useful -- mbp 20070423
1755
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1756
self.copy_content_into(dest_repo, revision_id)
1759
def start_write_group(self):
1760
"""Start a write group in the repository.
1762
Write groups are used by repositories which do not have a 1:1 mapping
1763
between file ids and backend store to manage the insertion of data from
1764
both fetch and commit operations.
1766
A write lock is required around the start_write_group/commit_write_group
1767
for the support of lock-requiring repository formats.
1769
One can only insert data into a repository inside a write group.
1773
if not self.is_write_locked():
1774
raise errors.NotWriteLocked(self)
1775
if self._write_group:
1776
raise errors.BzrError('already in a write group')
1777
self._start_write_group()
1778
# so we can detect unlock/relock - the write group is now entered.
1779
self._write_group = self.get_transaction()
1781
def _start_write_group(self):
1782
"""Template method for per-repository write group startup.
1784
This is called before the write group is considered to be
1789
def sprout(self, to_bzrdir, revision_id=None):
1790
"""Create a descendent repository for new development.
1792
Unlike clone, this does not copy the settings of the repository.
1794
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1795
dest_repo.fetch(self, revision_id=revision_id)
1798
def _create_sprouting_repo(self, a_bzrdir, shared):
1799
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1800
# use target default format.
1801
dest_repo = a_bzrdir.create_repository()
1803
# Most control formats need the repository to be specifically
1804
# created, but on some old all-in-one formats it's not needed
1806
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1807
except errors.UninitializableFormat:
1808
dest_repo = a_bzrdir.open_repository()
1811
def _get_sink(self):
1812
"""Return a sink for streaming into this repository."""
1813
return StreamSink(self)
1815
def _get_source(self, to_format):
1816
"""Return a source for streaming from this repository."""
1817
return StreamSource(self, to_format)
1820
def has_revision(self, revision_id):
1821
"""True if this repository has a copy of the revision."""
1822
return revision_id in self.has_revisions((revision_id,))
1825
def has_revisions(self, revision_ids):
1826
"""Probe to find out the presence of multiple revisions.
1828
:param revision_ids: An iterable of revision_ids.
1829
:return: A set of the revision_ids that were present.
1831
parent_map = self.revisions.get_parent_map(
1832
[(rev_id,) for rev_id in revision_ids])
1834
if _mod_revision.NULL_REVISION in revision_ids:
1835
result.add(_mod_revision.NULL_REVISION)
1836
result.update([key[0] for key in parent_map])
1840
def get_revision(self, revision_id):
1841
"""Return the Revision object for a named revision."""
1842
return self.get_revisions([revision_id])[0]
1845
def get_revision_reconcile(self, revision_id):
1846
"""'reconcile' helper routine that allows access to a revision always.
1848
This variant of get_revision does not cross check the weave graph
1849
against the revision one as get_revision does: but it should only
1850
be used by reconcile, or reconcile-alike commands that are correcting
1851
or testing the revision graph.
1853
return self._get_revisions([revision_id])[0]
1856
def get_revisions(self, revision_ids):
1857
"""Get many revisions at once.
1859
Repositories that need to check data on every revision read should
1860
subclass this method.
1862
return self._get_revisions(revision_ids)
1865
def _get_revisions(self, revision_ids):
1866
"""Core work logic to get many revisions without sanity checks."""
1868
for revid, rev in self._iter_revisions(revision_ids):
1870
raise errors.NoSuchRevision(self, revid)
1872
return [revs[revid] for revid in revision_ids]
1874
def _iter_revisions(self, revision_ids):
1875
"""Iterate over revision objects.
1877
:param revision_ids: An iterable of revisions to examine. None may be
1878
passed to request all revisions known to the repository. Note that
1879
not all repositories can find unreferenced revisions; for those
1880
repositories only referenced ones will be returned.
1881
:return: An iterator of (revid, revision) tuples. Absent revisions (
1882
those asked for but not available) are returned as (revid, None).
1884
if revision_ids is None:
1885
revision_ids = self.all_revision_ids()
1887
for rev_id in revision_ids:
1888
if not rev_id or not isinstance(rev_id, basestring):
1889
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1890
keys = [(key,) for key in revision_ids]
1891
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1892
for record in stream:
1893
revid = record.key[0]
1894
if record.storage_kind == 'absent':
1897
text = record.get_bytes_as('fulltext')
1898
rev = self._serializer.read_revision_from_string(text)
1902
def get_revision_xml(self, revision_id):
1903
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1904
# would have already do it.
1905
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1906
# TODO: this can't just be replaced by:
1907
# return self._serializer.write_revision_to_string(
1908
# self.get_revision(revision_id))
1909
# as cStringIO preservers the encoding unlike write_revision_to_string
1910
# or some other call down the path.
1911
rev = self.get_revision(revision_id)
1912
rev_tmp = cStringIO.StringIO()
1913
# the current serializer..
1914
self._serializer.write_revision(rev, rev_tmp)
1916
return rev_tmp.getvalue()
1918
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1919
"""Produce a generator of revision deltas.
1921
Note that the input is a sequence of REVISIONS, not revision_ids.
1922
Trees will be held in memory until the generator exits.
1923
Each delta is relative to the revision's lefthand predecessor.
1925
:param specific_fileids: if not None, the result is filtered
1926
so that only those file-ids, their parents and their
1927
children are included.
1929
# Get the revision-ids of interest
1930
required_trees = set()
1931
for revision in revisions:
1932
required_trees.add(revision.revision_id)
1933
required_trees.update(revision.parent_ids[:1])
1935
# Get the matching filtered trees. Note that it's more
1936
# efficient to pass filtered trees to changes_from() rather
1937
# than doing the filtering afterwards. changes_from() could
1938
# arguably do the filtering itself but it's path-based, not
1939
# file-id based, so filtering before or afterwards is
1941
if specific_fileids is None:
1942
trees = dict((t.get_revision_id(), t) for
1943
t in self.revision_trees(required_trees))
1945
trees = dict((t.get_revision_id(), t) for
1946
t in self._filtered_revision_trees(required_trees,
1949
# Calculate the deltas
1950
for revision in revisions:
1951
if not revision.parent_ids:
1952
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1954
old_tree = trees[revision.parent_ids[0]]
1955
yield trees[revision.revision_id].changes_from(old_tree)
1958
def get_revision_delta(self, revision_id, specific_fileids=None):
1959
"""Return the delta for one revision.
1961
The delta is relative to the left-hand predecessor of the
1964
:param specific_fileids: if not None, the result is filtered
1965
so that only those file-ids, their parents and their
1966
children are included.
1968
r = self.get_revision(revision_id)
1969
return list(self.get_deltas_for_revisions([r],
1970
specific_fileids=specific_fileids))[0]
1973
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1974
signature = gpg_strategy.sign(plaintext)
1975
self.add_signature_text(revision_id, signature)
1978
def add_signature_text(self, revision_id, signature):
1979
self.signatures.add_lines((revision_id,), (),
1980
osutils.split_lines(signature))
1982
def find_text_key_references(self):
1983
"""Find the text key references within the repository.
1985
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1986
to whether they were referred to by the inventory of the
1987
revision_id that they contain. The inventory texts from all present
1988
revision ids are assessed to generate this report.
1990
revision_keys = self.revisions.keys()
1991
w = self.inventories
1992
pb = ui.ui_factory.nested_progress_bar()
1994
return self._find_text_key_references_from_xml_inventory_lines(
1995
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1999
def _find_text_key_references_from_xml_inventory_lines(self,
2001
"""Core routine for extracting references to texts from inventories.
2003
This performs the translation of xml lines to revision ids.
2005
:param line_iterator: An iterator of lines, origin_version_id
2006
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2007
to whether they were referred to by the inventory of the
2008
revision_id that they contain. Note that if that revision_id was
2009
not part of the line_iterator's output then False will be given -
2010
even though it may actually refer to that key.
2012
if not self._serializer.support_altered_by_hack:
2013
raise AssertionError(
2014
"_find_text_key_references_from_xml_inventory_lines only "
2015
"supported for branches which store inventory as unnested xml"
2016
", not on %r" % self)
2019
# this code needs to read every new line in every inventory for the
2020
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2021
# not present in one of those inventories is unnecessary but not
2022
# harmful because we are filtering by the revision id marker in the
2023
# inventory lines : we only select file ids altered in one of those
2024
# revisions. We don't need to see all lines in the inventory because
2025
# only those added in an inventory in rev X can contain a revision=X
2027
unescape_revid_cache = {}
2028
unescape_fileid_cache = {}
2030
# jam 20061218 In a big fetch, this handles hundreds of thousands
2031
# of lines, so it has had a lot of inlining and optimizing done.
2032
# Sorry that it is a little bit messy.
2033
# Move several functions to be local variables, since this is a long
2035
search = self._file_ids_altered_regex.search
2036
unescape = _unescape_xml
2037
setdefault = result.setdefault
2038
for line, line_key in line_iterator:
2039
match = search(line)
2042
# One call to match.group() returning multiple items is quite a
2043
# bit faster than 2 calls to match.group() each returning 1
2044
file_id, revision_id = match.group('file_id', 'revision_id')
2046
# Inlining the cache lookups helps a lot when you make 170,000
2047
# lines and 350k ids, versus 8.4 unique ids.
2048
# Using a cache helps in 2 ways:
2049
# 1) Avoids unnecessary decoding calls
2050
# 2) Re-uses cached strings, which helps in future set and
2052
# (2) is enough that removing encoding entirely along with
2053
# the cache (so we are using plain strings) results in no
2054
# performance improvement.
2056
revision_id = unescape_revid_cache[revision_id]
2058
unescaped = unescape(revision_id)
2059
unescape_revid_cache[revision_id] = unescaped
2060
revision_id = unescaped
2062
# Note that unconditionally unescaping means that we deserialise
2063
# every fileid, which for general 'pull' is not great, but we don't
2064
# really want to have some many fulltexts that this matters anyway.
2067
file_id = unescape_fileid_cache[file_id]
2069
unescaped = unescape(file_id)
2070
unescape_fileid_cache[file_id] = unescaped
2073
key = (file_id, revision_id)
2074
setdefault(key, False)
2075
if revision_id == line_key[-1]:
2079
def _inventory_xml_lines_for_keys(self, keys):
2080
"""Get a line iterator of the sort needed for findind references.
2082
Not relevant for non-xml inventory repositories.
2084
Ghosts in revision_keys are ignored.
2086
:param revision_keys: The revision keys for the inventories to inspect.
2087
:return: An iterator over (inventory line, revid) for the fulltexts of
2088
all of the xml inventories specified by revision_keys.
2090
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2091
for record in stream:
2092
if record.storage_kind != 'absent':
2093
chunks = record.get_bytes_as('chunked')
2094
revid = record.key[-1]
2095
lines = osutils.chunks_to_lines(chunks)
2099
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2101
"""Helper routine for fileids_altered_by_revision_ids.
2103
This performs the translation of xml lines to revision ids.
2105
:param line_iterator: An iterator of lines, origin_version_id
2106
:param revision_keys: The revision ids to filter for. This should be a
2107
set or other type which supports efficient __contains__ lookups, as
2108
the revision key from each parsed line will be looked up in the
2109
revision_keys filter.
2110
:return: a dictionary mapping altered file-ids to an iterable of
2111
revision_ids. Each altered file-ids has the exact revision_ids that
2112
altered it listed explicitly.
2114
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2115
line_iterator).iterkeys())
2116
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2117
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2118
self._inventory_xml_lines_for_keys(parent_keys)))
2119
new_keys = seen - parent_seen
2121
setdefault = result.setdefault
2122
for key in new_keys:
2123
setdefault(key[0], set()).add(key[-1])
2126
def _find_parent_ids_of_revisions(self, revision_ids):
2127
"""Find all parent ids that are mentioned in the revision graph.
2129
:return: set of revisions that are parents of revision_ids which are
2130
not part of revision_ids themselves
2132
parent_map = self.get_parent_map(revision_ids)
2134
map(parent_ids.update, parent_map.itervalues())
2135
parent_ids.difference_update(revision_ids)
2136
parent_ids.discard(_mod_revision.NULL_REVISION)
2139
def _find_parent_keys_of_revisions(self, revision_keys):
2140
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2142
:param revision_keys: An iterable of revision_keys.
2143
:return: The parents of all revision_keys that are not already in
2146
parent_map = self.revisions.get_parent_map(revision_keys)
2148
map(parent_keys.update, parent_map.itervalues())
2149
parent_keys.difference_update(revision_keys)
2150
parent_keys.discard(_mod_revision.NULL_REVISION)
2153
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2154
"""Find the file ids and versions affected by revisions.
2156
:param revisions: an iterable containing revision ids.
2157
:param _inv_weave: The inventory weave from this repository or None.
2158
If None, the inventory weave will be opened automatically.
2159
:return: a dictionary mapping altered file-ids to an iterable of
2160
revision_ids. Each altered file-ids has the exact revision_ids that
2161
altered it listed explicitly.
2163
selected_keys = set((revid,) for revid in revision_ids)
2164
w = _inv_weave or self.inventories
2165
pb = ui.ui_factory.nested_progress_bar()
2167
return self._find_file_ids_from_xml_inventory_lines(
2168
w.iter_lines_added_or_present_in_keys(
2169
selected_keys, pb=pb),
2174
def iter_files_bytes(self, desired_files):
2175
"""Iterate through file versions.
2177
Files will not necessarily be returned in the order they occur in
2178
desired_files. No specific order is guaranteed.
2180
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2181
value supplied by the caller as part of desired_files. It should
2182
uniquely identify the file version in the caller's context. (Examples:
2183
an index number or a TreeTransform trans_id.)
2185
bytes_iterator is an iterable of bytestrings for the file. The
2186
kind of iterable and length of the bytestrings are unspecified, but for
2187
this implementation, it is a list of bytes produced by
2188
VersionedFile.get_record_stream().
2190
:param desired_files: a list of (file_id, revision_id, identifier)
2194
for file_id, revision_id, callable_data in desired_files:
2195
text_keys[(file_id, revision_id)] = callable_data
2196
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2197
if record.storage_kind == 'absent':
2198
raise errors.RevisionNotPresent(record.key, self)
2199
yield text_keys[record.key], record.get_bytes_as('chunked')
2201
def _generate_text_key_index(self, text_key_references=None,
2203
"""Generate a new text key index for the repository.
2205
This is an expensive function that will take considerable time to run.
2207
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2208
list of parents, also text keys. When a given key has no parents,
2209
the parents list will be [NULL_REVISION].
2211
# All revisions, to find inventory parents.
2212
if ancestors is None:
2213
graph = self.get_graph()
2214
ancestors = graph.get_parent_map(self.all_revision_ids())
2215
if text_key_references is None:
2216
text_key_references = self.find_text_key_references()
2217
pb = ui.ui_factory.nested_progress_bar()
2219
return self._do_generate_text_key_index(ancestors,
2220
text_key_references, pb)
2224
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2225
"""Helper for _generate_text_key_index to avoid deep nesting."""
2226
revision_order = tsort.topo_sort(ancestors)
2227
invalid_keys = set()
2229
for revision_id in revision_order:
2230
revision_keys[revision_id] = set()
2231
text_count = len(text_key_references)
2232
# a cache of the text keys to allow reuse; costs a dict of all the
2233
# keys, but saves a 2-tuple for every child of a given key.
2235
for text_key, valid in text_key_references.iteritems():
2237
invalid_keys.add(text_key)
2239
revision_keys[text_key[1]].add(text_key)
2240
text_key_cache[text_key] = text_key
2241
del text_key_references
2243
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2244
NULL_REVISION = _mod_revision.NULL_REVISION
2245
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2246
# too small for large or very branchy trees. However, for 55K path
2247
# trees, it would be easy to use too much memory trivially. Ideally we
2248
# could gauge this by looking at available real memory etc, but this is
2249
# always a tricky proposition.
2250
inventory_cache = lru_cache.LRUCache(10)
2251
batch_size = 10 # should be ~150MB on a 55K path tree
2252
batch_count = len(revision_order) / batch_size + 1
2254
pb.update("Calculating text parents", processed_texts, text_count)
2255
for offset in xrange(batch_count):
2256
to_query = revision_order[offset * batch_size:(offset + 1) *
2260
for revision_id in to_query:
2261
parent_ids = ancestors[revision_id]
2262
for text_key in revision_keys[revision_id]:
2263
pb.update("Calculating text parents", processed_texts)
2264
processed_texts += 1
2265
candidate_parents = []
2266
for parent_id in parent_ids:
2267
parent_text_key = (text_key[0], parent_id)
2269
check_parent = parent_text_key not in \
2270
revision_keys[parent_id]
2272
# the parent parent_id is a ghost:
2273
check_parent = False
2274
# truncate the derived graph against this ghost.
2275
parent_text_key = None
2277
# look at the parent commit details inventories to
2278
# determine possible candidates in the per file graph.
2281
inv = inventory_cache[parent_id]
2283
inv = self.revision_tree(parent_id).inventory
2284
inventory_cache[parent_id] = inv
2286
parent_entry = inv[text_key[0]]
2287
except (KeyError, errors.NoSuchId):
2289
if parent_entry is not None:
2291
text_key[0], parent_entry.revision)
2293
parent_text_key = None
2294
if parent_text_key is not None:
2295
candidate_parents.append(
2296
text_key_cache[parent_text_key])
2297
parent_heads = text_graph.heads(candidate_parents)
2298
new_parents = list(parent_heads)
2299
new_parents.sort(key=lambda x:candidate_parents.index(x))
2300
if new_parents == []:
2301
new_parents = [NULL_REVISION]
2302
text_index[text_key] = new_parents
2304
for text_key in invalid_keys:
2305
text_index[text_key] = [NULL_REVISION]
2308
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2309
"""Get an iterable listing the keys of all the data introduced by a set
2312
The keys will be ordered so that the corresponding items can be safely
2313
fetched and inserted in that order.
2315
:returns: An iterable producing tuples of (knit-kind, file-id,
2316
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2317
'revisions'. file-id is None unless knit-kind is 'file'.
2319
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2322
for result in self._find_non_file_keys_to_fetch(revision_ids):
2325
def _find_file_keys_to_fetch(self, revision_ids, pb):
2326
# XXX: it's a bit weird to control the inventory weave caching in this
2327
# generator. Ideally the caching would be done in fetch.py I think. Or
2328
# maybe this generator should explicitly have the contract that it
2329
# should not be iterated until the previously yielded item has been
2331
inv_w = self.inventories
2333
# file ids that changed
2334
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2336
num_file_ids = len(file_ids)
2337
for file_id, altered_versions in file_ids.iteritems():
2339
pb.update("Fetch texts", count, num_file_ids)
2341
yield ("file", file_id, altered_versions)
2343
def _find_non_file_keys_to_fetch(self, revision_ids):
2345
yield ("inventory", None, revision_ids)
2348
# XXX: Note ATM no callers actually pay attention to this return
2349
# instead they just use the list of revision ids and ignore
2350
# missing sigs. Consider removing this work entirely
2351
revisions_with_signatures = set(self.signatures.get_parent_map(
2352
[(r,) for r in revision_ids]))
2353
revisions_with_signatures = set(
2354
[r for (r,) in revisions_with_signatures])
2355
revisions_with_signatures.intersection_update(revision_ids)
2356
yield ("signatures", None, revisions_with_signatures)
2359
yield ("revisions", None, revision_ids)
2362
def get_inventory(self, revision_id):
2363
"""Get Inventory object by revision id."""
2364
return self.iter_inventories([revision_id]).next()
2366
def iter_inventories(self, revision_ids, ordering=None):
2367
"""Get many inventories by revision_ids.
2369
This will buffer some or all of the texts used in constructing the
2370
inventories in memory, but will only parse a single inventory at a
2373
:param revision_ids: The expected revision ids of the inventories.
2374
:param ordering: optional ordering, e.g. 'topological'. If not
2375
specified, the order of revision_ids will be preserved (by
2376
buffering if necessary).
2377
:return: An iterator of inventories.
2379
if ((None in revision_ids)
2380
or (_mod_revision.NULL_REVISION in revision_ids)):
2381
raise ValueError('cannot get null revision inventory')
2382
return self._iter_inventories(revision_ids, ordering)
2384
def _iter_inventories(self, revision_ids, ordering):
2385
"""single-document based inventory iteration."""
2386
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2387
for text, revision_id in inv_xmls:
2388
yield self.deserialise_inventory(revision_id, text)
2390
def _iter_inventory_xmls(self, revision_ids, ordering):
2391
if ordering is None:
2392
order_as_requested = True
2393
ordering = 'unordered'
2395
order_as_requested = False
2396
keys = [(revision_id,) for revision_id in revision_ids]
2399
if order_as_requested:
2400
key_iter = iter(keys)
2401
next_key = key_iter.next()
2402
stream = self.inventories.get_record_stream(keys, ordering, True)
2404
for record in stream:
2405
if record.storage_kind != 'absent':
2406
chunks = record.get_bytes_as('chunked')
2407
if order_as_requested:
2408
text_chunks[record.key] = chunks
2410
yield ''.join(chunks), record.key[-1]
2412
raise errors.NoSuchRevision(self, record.key)
2413
if order_as_requested:
2414
# Yield as many results as we can while preserving order.
2415
while next_key in text_chunks:
2416
chunks = text_chunks.pop(next_key)
2417
yield ''.join(chunks), next_key[-1]
2419
next_key = key_iter.next()
2420
except StopIteration:
2421
# We still want to fully consume the get_record_stream,
2422
# just in case it is not actually finished at this point
2426
def deserialise_inventory(self, revision_id, xml):
2427
"""Transform the xml into an inventory object.
2429
:param revision_id: The expected revision id of the inventory.
2430
:param xml: A serialised inventory.
2432
result = self._serializer.read_inventory_from_string(xml, revision_id,
2433
entry_cache=self._inventory_entry_cache)
2434
if result.revision_id != revision_id:
2435
raise AssertionError('revision id mismatch %s != %s' % (
2436
result.revision_id, revision_id))
2439
def serialise_inventory(self, inv):
2440
return self._serializer.write_inventory_to_string(inv)
2442
def _serialise_inventory_to_lines(self, inv):
2443
return self._serializer.write_inventory_to_lines(inv)
2445
def get_serializer_format(self):
2446
return self._serializer.format_num
2449
def get_inventory_xml(self, revision_id):
2450
"""Get inventory XML as a file object."""
2451
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2453
text, revision_id = texts.next()
2454
except StopIteration:
2455
raise errors.HistoryMissing(self, 'inventory', revision_id)
2459
def get_inventory_sha1(self, revision_id):
2460
"""Return the sha1 hash of the inventory entry
2462
return self.get_revision(revision_id).inventory_sha1
2464
def get_rev_id_for_revno(self, revno, known_pair):
2465
"""Return the revision id of a revno, given a later (revno, revid)
2466
pair in the same history.
2468
:return: if found (True, revid). If the available history ran out
2469
before reaching the revno, then this returns
2470
(False, (closest_revno, closest_revid)).
2472
known_revno, known_revid = known_pair
2473
partial_history = [known_revid]
2474
distance_from_known = known_revno - revno
2475
if distance_from_known < 0:
2477
'requested revno (%d) is later than given known revno (%d)'
2478
% (revno, known_revno))
2481
self, partial_history, stop_index=distance_from_known)
2482
except errors.RevisionNotPresent, err:
2483
if err.revision_id == known_revid:
2484
# The start revision (known_revid) wasn't found.
2486
# This is a stacked repository with no fallbacks, or a there's a
2487
# left-hand ghost. Either way, even though the revision named in
2488
# the error isn't in this repo, we know it's the next step in this
2489
# left-hand history.
2490
partial_history.append(err.revision_id)
2491
if len(partial_history) <= distance_from_known:
2492
# Didn't find enough history to get a revid for the revno.
2493
earliest_revno = known_revno - len(partial_history) + 1
2494
return (False, (earliest_revno, partial_history[-1]))
2495
if len(partial_history) - 1 > distance_from_known:
2496
raise AssertionError('_iter_for_revno returned too much history')
2497
return (True, partial_history[-1])
2499
def iter_reverse_revision_history(self, revision_id):
2500
"""Iterate backwards through revision ids in the lefthand history
2502
:param revision_id: The revision id to start with. All its lefthand
2503
ancestors will be traversed.
2505
graph = self.get_graph()
2506
next_id = revision_id
2508
if next_id in (None, _mod_revision.NULL_REVISION):
2511
parents = graph.get_parent_map([next_id])[next_id]
2513
raise errors.RevisionNotPresent(next_id, self)
2515
if len(parents) == 0:
2518
next_id = parents[0]
2521
def get_revision_inventory(self, revision_id):
2522
"""Return inventory of a past revision."""
2523
# TODO: Unify this with get_inventory()
2524
# bzr 0.0.6 and later imposes the constraint that the inventory_id
2525
# must be the same as its revision, so this is trivial.
2526
if revision_id is None:
2527
# This does not make sense: if there is no revision,
2528
# then it is the current tree inventory surely ?!
2529
# and thus get_root_id() is something that looks at the last
2530
# commit on the branch, and the get_root_id is an inventory check.
2531
raise NotImplementedError
2532
# return Inventory(self.get_root_id())
2534
return self.get_inventory(revision_id)
2536
def is_shared(self):
2537
"""Return True if this repository is flagged as a shared repository."""
2538
raise NotImplementedError(self.is_shared)
2541
def reconcile(self, other=None, thorough=False):
2542
"""Reconcile this repository."""
2543
from bzrlib.reconcile import RepoReconciler
2544
reconciler = RepoReconciler(self, thorough=thorough)
2545
reconciler.reconcile()
2548
def _refresh_data(self):
2549
"""Helper called from lock_* to ensure coherency with disk.
2551
The default implementation does nothing; it is however possible
2552
for repositories to maintain loaded indices across multiple locks
2553
by checking inside their implementation of this method to see
2554
whether their indices are still valid. This depends of course on
2555
the disk format being validatable in this manner. This method is
2556
also called by the refresh_data() public interface to cause a refresh
2557
to occur while in a write lock so that data inserted by a smart server
2558
push operation is visible on the client's instance of the physical
2563
def revision_tree(self, revision_id):
2564
"""Return Tree for a revision on this branch.
2566
`revision_id` may be NULL_REVISION for the empty tree revision.
2568
revision_id = _mod_revision.ensure_null(revision_id)
2569
# TODO: refactor this to use an existing revision object
2570
# so we don't need to read it in twice.
2571
if revision_id == _mod_revision.NULL_REVISION:
2572
return RevisionTree(self, Inventory(root_id=None),
2573
_mod_revision.NULL_REVISION)
2575
inv = self.get_revision_inventory(revision_id)
2576
return RevisionTree(self, inv, revision_id)
2578
def revision_trees(self, revision_ids):
2579
"""Return Trees for revisions in this repository.
2581
:param revision_ids: a sequence of revision-ids;
2582
a revision-id may not be None or 'null:'
2584
inventories = self.iter_inventories(revision_ids)
2585
for inv in inventories:
2586
yield RevisionTree(self, inv, inv.revision_id)
2588
def _filtered_revision_trees(self, revision_ids, file_ids):
2589
"""Return Tree for a revision on this branch with only some files.
2591
:param revision_ids: a sequence of revision-ids;
2592
a revision-id may not be None or 'null:'
2593
:param file_ids: if not None, the result is filtered
2594
so that only those file-ids, their parents and their
2595
children are included.
2597
inventories = self.iter_inventories(revision_ids)
2598
for inv in inventories:
2599
# Should we introduce a FilteredRevisionTree class rather
2600
# than pre-filter the inventory here?
2601
filtered_inv = inv.filter(file_ids)
2602
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2605
def get_ancestry(self, revision_id, topo_sorted=True):
2606
"""Return a list of revision-ids integrated by a revision.
2608
The first element of the list is always None, indicating the origin
2609
revision. This might change when we have history horizons, or
2610
perhaps we should have a new API.
2612
This is topologically sorted.
2614
if _mod_revision.is_null(revision_id):
2616
if not self.has_revision(revision_id):
2617
raise errors.NoSuchRevision(self, revision_id)
2618
graph = self.get_graph()
2620
search = graph._make_breadth_first_searcher([revision_id])
2623
found, ghosts = search.next_with_ghosts()
2624
except StopIteration:
2627
if _mod_revision.NULL_REVISION in keys:
2628
keys.remove(_mod_revision.NULL_REVISION)
2630
parent_map = graph.get_parent_map(keys)
2631
keys = tsort.topo_sort(parent_map)
2632
return [None] + list(keys)
2634
def pack(self, hint=None):
2635
"""Compress the data within the repository.
2637
This operation only makes sense for some repository types. For other
2638
types it should be a no-op that just returns.
2640
This stub method does not require a lock, but subclasses should use
2641
@needs_write_lock as this is a long running call its reasonable to
2642
implicitly lock for the user.
2644
:param hint: If not supplied, the whole repository is packed.
2645
If supplied, the repository may use the hint parameter as a
2646
hint for the parts of the repository to pack. A hint can be
2647
obtained from the result of commit_write_group(). Out of
2648
date hints are simply ignored, because concurrent operations
2649
can obsolete them rapidly.
2652
def get_transaction(self):
2653
return self.control_files.get_transaction()
2655
def get_parent_map(self, revision_ids):
2656
"""See graph.StackedParentsProvider.get_parent_map"""
2657
# revisions index works in keys; this just works in revisions
2658
# therefore wrap and unwrap
2661
for revision_id in revision_ids:
2662
if revision_id == _mod_revision.NULL_REVISION:
2663
result[revision_id] = ()
2664
elif revision_id is None:
2665
raise ValueError('get_parent_map(None) is not valid')
2667
query_keys.append((revision_id ,))
2668
for ((revision_id,), parent_keys) in \
2669
self.revisions.get_parent_map(query_keys).iteritems():
2671
result[revision_id] = tuple(parent_revid
2672
for (parent_revid,) in parent_keys)
2674
result[revision_id] = (_mod_revision.NULL_REVISION,)
2677
def _make_parents_provider(self):
2680
def get_graph(self, other_repository=None):
2681
"""Return the graph walker for this repository format"""
2682
parents_provider = self._make_parents_provider()
2683
if (other_repository is not None and
2684
not self.has_same_location(other_repository)):
2685
parents_provider = graph.StackedParentsProvider(
2686
[parents_provider, other_repository._make_parents_provider()])
2687
return graph.Graph(parents_provider)
2689
def _get_versioned_file_checker(self, text_key_references=None,
2691
"""Return an object suitable for checking versioned files.
2693
:param text_key_references: if non-None, an already built
2694
dictionary mapping text keys ((fileid, revision_id) tuples)
2695
to whether they were referred to by the inventory of the
2696
revision_id that they contain. If None, this will be
2698
:param ancestors: Optional result from
2699
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2702
return _VersionedFileChecker(self,
2703
text_key_references=text_key_references, ancestors=ancestors)
2705
def revision_ids_to_search_result(self, result_set):
2706
"""Convert a set of revision ids to a graph SearchResult."""
2707
result_parents = set()
2708
for parents in self.get_graph().get_parent_map(
2709
result_set).itervalues():
2710
result_parents.update(parents)
2711
included_keys = result_set.intersection(result_parents)
2712
start_keys = result_set.difference(included_keys)
2713
exclude_keys = result_parents.difference(result_set)
2714
result = graph.SearchResult(start_keys, exclude_keys,
2715
len(result_set), result_set)
2719
def set_make_working_trees(self, new_value):
2720
"""Set the policy flag for making working trees when creating branches.
2722
This only applies to branches that use this repository.
2724
The default is 'True'.
2725
:param new_value: True to restore the default, False to disable making
2728
raise NotImplementedError(self.set_make_working_trees)
2730
def make_working_trees(self):
2731
"""Returns the policy for making working trees on new branches."""
2732
raise NotImplementedError(self.make_working_trees)
2735
def sign_revision(self, revision_id, gpg_strategy):
2736
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2737
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2740
def has_signature_for_revision_id(self, revision_id):
2741
"""Query for a revision signature for revision_id in the repository."""
2742
if not self.has_revision(revision_id):
2743
raise errors.NoSuchRevision(self, revision_id)
2744
sig_present = (1 == len(
2745
self.signatures.get_parent_map([(revision_id,)])))
2749
def get_signature_text(self, revision_id):
2750
"""Return the text for a signature."""
2751
stream = self.signatures.get_record_stream([(revision_id,)],
2753
record = stream.next()
2754
if record.storage_kind == 'absent':
2755
raise errors.NoSuchRevision(self, revision_id)
2756
return record.get_bytes_as('fulltext')
2759
def check(self, revision_ids=None, callback_refs=None, check_repo=True):
2760
"""Check consistency of all history of given revision_ids.
2762
Different repository implementations should override _check().
2764
:param revision_ids: A non-empty list of revision_ids whose ancestry
2765
will be checked. Typically the last revision_id of a branch.
2766
:param callback_refs: A dict of check-refs to resolve and callback
2767
the check/_check method on the items listed as wanting the ref.
2769
:param check_repo: If False do not check the repository contents, just
2770
calculate the data callback_refs requires and call them back.
2772
return self._check(revision_ids, callback_refs=callback_refs,
2773
check_repo=check_repo)
2775
def _check(self, revision_ids, callback_refs, check_repo):
2776
result = check.Check(self, check_repo=check_repo)
2777
result.check(callback_refs)
2780
def _warn_if_deprecated(self):
2781
global _deprecation_warning_done
2782
if _deprecation_warning_done:
2784
_deprecation_warning_done = True
2785
warning("Format %s for %s is deprecated - please use 'bzr upgrade' to get better performance"
2786
% (self._format, self.bzrdir.transport.base))
2788
def supports_rich_root(self):
2789
return self._format.rich_root_data
2791
def _check_ascii_revisionid(self, revision_id, method):
2792
"""Private helper for ascii-only repositories."""
2793
# weave repositories refuse to store revisionids that are non-ascii.
2794
if revision_id is not None:
2795
# weaves require ascii revision ids.
2796
if isinstance(revision_id, unicode):
2798
revision_id.encode('ascii')
2799
except UnicodeEncodeError:
2800
raise errors.NonAsciiRevisionId(method, self)
2803
revision_id.decode('ascii')
2804
except UnicodeDecodeError:
2805
raise errors.NonAsciiRevisionId(method, self)
2807
def revision_graph_can_have_wrong_parents(self):
2808
"""Is it possible for this repository to have a revision graph with
2811
If True, then this repository must also implement
2812
_find_inconsistent_revision_parents so that check and reconcile can
2813
check for inconsistencies before proceeding with other checks that may
2814
depend on the revision index being consistent.
2816
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2819
# remove these delegates a while after bzr 0.15
2820
def __make_delegated(name, from_module):
2821
def _deprecated_repository_forwarder():
2822
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2823
% (name, from_module),
2826
m = __import__(from_module, globals(), locals(), [name])
2828
return getattr(m, name)
2829
except AttributeError:
2830
raise AttributeError('module %s has no name %s'
2832
globals()[name] = _deprecated_repository_forwarder
2835
'AllInOneRepository',
2836
'WeaveMetaDirRepository',
2837
'PreSplitOutRepositoryFormat',
2838
'RepositoryFormat4',
2839
'RepositoryFormat5',
2840
'RepositoryFormat6',
2841
'RepositoryFormat7',
2843
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2847
'RepositoryFormatKnit',
2848
'RepositoryFormatKnit1',
2850
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2853
def install_revision(repository, rev, revision_tree):
2854
"""Install all revision data into a repository."""
2855
install_revisions(repository, [(rev, revision_tree, None)])
2858
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2859
"""Install all revision data into a repository.
2861
Accepts an iterable of revision, tree, signature tuples. The signature
2864
repository.start_write_group()
2866
inventory_cache = lru_cache.LRUCache(10)
2867
for n, (revision, revision_tree, signature) in enumerate(iterable):
2868
_install_revision(repository, revision, revision_tree, signature,
2871
pb.update('Transferring revisions', n + 1, num_revisions)
2873
repository.abort_write_group()
2876
repository.commit_write_group()
2879
def _install_revision(repository, rev, revision_tree, signature,
2881
"""Install all revision data into a repository."""
2882
present_parents = []
2884
for p_id in rev.parent_ids:
2885
if repository.has_revision(p_id):
2886
present_parents.append(p_id)
2887
parent_trees[p_id] = repository.revision_tree(p_id)
2889
parent_trees[p_id] = repository.revision_tree(
2890
_mod_revision.NULL_REVISION)
2892
inv = revision_tree.inventory
2893
entries = inv.iter_entries()
2894
# backwards compatibility hack: skip the root id.
2895
if not repository.supports_rich_root():
2896
path, root = entries.next()
2897
if root.revision != rev.revision_id:
2898
raise errors.IncompatibleRevision(repr(repository))
2900
for path, ie in entries:
2901
text_keys[(ie.file_id, ie.revision)] = ie
2902
text_parent_map = repository.texts.get_parent_map(text_keys)
2903
missing_texts = set(text_keys) - set(text_parent_map)
2904
# Add the texts that are not already present
2905
for text_key in missing_texts:
2906
ie = text_keys[text_key]
2908
# FIXME: TODO: The following loop overlaps/duplicates that done by
2909
# commit to determine parents. There is a latent/real bug here where
2910
# the parents inserted are not those commit would do - in particular
2911
# they are not filtered by heads(). RBC, AB
2912
for revision, tree in parent_trees.iteritems():
2913
if ie.file_id not in tree:
2915
parent_id = tree.inventory[ie.file_id].revision
2916
if parent_id in text_parents:
2918
text_parents.append((ie.file_id, parent_id))
2919
lines = revision_tree.get_file(ie.file_id).readlines()
2920
repository.texts.add_lines(text_key, text_parents, lines)
2922
# install the inventory
2923
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2924
# Cache this inventory
2925
inventory_cache[rev.revision_id] = inv
2927
basis_inv = inventory_cache[rev.parent_ids[0]]
2929
repository.add_inventory(rev.revision_id, inv, present_parents)
2931
delta = inv._make_delta(basis_inv)
2932
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2933
rev.revision_id, present_parents)
2935
repository.add_inventory(rev.revision_id, inv, present_parents)
2936
except errors.RevisionAlreadyPresent:
2938
if signature is not None:
2939
repository.add_signature_text(rev.revision_id, signature)
2940
repository.add_revision(rev.revision_id, rev, inv)
2943
class MetaDirRepository(Repository):
2944
"""Repositories in the new meta-dir layout.
2946
:ivar _transport: Transport for access to repository control files,
2947
typically pointing to .bzr/repository.
2950
def __init__(self, _format, a_bzrdir, control_files):
2951
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2952
self._transport = control_files._transport
2954
def is_shared(self):
2955
"""Return True if this repository is flagged as a shared repository."""
2956
return self._transport.has('shared-storage')
2959
def set_make_working_trees(self, new_value):
2960
"""Set the policy flag for making working trees when creating branches.
2962
This only applies to branches that use this repository.
2964
The default is 'True'.
2965
:param new_value: True to restore the default, False to disable making
2970
self._transport.delete('no-working-trees')
2971
except errors.NoSuchFile:
2974
self._transport.put_bytes('no-working-trees', '',
2975
mode=self.bzrdir._get_file_mode())
2977
def make_working_trees(self):
2978
"""Returns the policy for making working trees on new branches."""
2979
return not self._transport.has('no-working-trees')
2982
class MetaDirVersionedFileRepository(MetaDirRepository):
2983
"""Repositories in a meta-dir, that work via versioned file objects."""
2985
def __init__(self, _format, a_bzrdir, control_files):
2986
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2990
network_format_registry = registry.FormatRegistry()
2991
"""Registry of formats indexed by their network name.
2993
The network name for a repository format is an identifier that can be used when
2994
referring to formats with smart server operations. See
2995
RepositoryFormat.network_name() for more detail.
2999
format_registry = registry.FormatRegistry(network_format_registry)
3000
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
3002
This can contain either format instances themselves, or classes/factories that
3003
can be called to obtain one.
3007
#####################################################################
3008
# Repository Formats
3010
class RepositoryFormat(object):
3011
"""A repository format.
3013
Formats provide four things:
3014
* An initialization routine to construct repository data on disk.
3015
* a optional format string which is used when the BzrDir supports
3017
* an open routine which returns a Repository instance.
3018
* A network name for referring to the format in smart server RPC
3021
There is one and only one Format subclass for each on-disk format. But
3022
there can be one Repository subclass that is used for several different
3023
formats. The _format attribute on a Repository instance can be used to
3024
determine the disk format.
3026
Formats are placed in a registry by their format string for reference
3027
during opening. These should be subclasses of RepositoryFormat for
3030
Once a format is deprecated, just deprecate the initialize and open
3031
methods on the format class. Do not deprecate the object, as the
3032
object may be created even when a repository instance hasn't been
3035
Common instance attributes:
3036
_matchingbzrdir - the bzrdir format that the repository format was
3037
originally written to work with. This can be used if manually
3038
constructing a bzrdir and repository, or more commonly for test suite
3042
# Set to True or False in derived classes. True indicates that the format
3043
# supports ghosts gracefully.
3044
supports_ghosts = None
3045
# Can this repository be given external locations to lookup additional
3046
# data. Set to True or False in derived classes.
3047
supports_external_lookups = None
3048
# Does this format support CHK bytestring lookups. Set to True or False in
3050
supports_chks = None
3051
# Should commit add an inventory, or an inventory delta to the repository.
3052
_commit_inv_deltas = True
3053
# What order should fetch operations request streams in?
3054
# The default is unordered as that is the cheapest for an origin to
3056
_fetch_order = 'unordered'
3057
# Does this repository format use deltas that can be fetched as-deltas ?
3058
# (E.g. knits, where the knit deltas can be transplanted intact.
3059
# We default to False, which will ensure that enough data to get
3060
# a full text out of any fetch stream will be grabbed.
3061
_fetch_uses_deltas = False
3062
# Should fetch trigger a reconcile after the fetch? Only needed for
3063
# some repository formats that can suffer internal inconsistencies.
3064
_fetch_reconcile = False
3065
# Does this format have < O(tree_size) delta generation. Used to hint what
3066
# code path for commit, amongst other things.
3068
# Does doing a pack operation compress data? Useful for the pack UI command
3069
# (so if there is one pack, the operation can still proceed because it may
3070
# help), and for fetching when data won't have come from the same
3072
pack_compresses = False
3073
# Does the repository inventory storage understand references to trees?
3074
supports_tree_reference = None
3077
return "<%s>" % self.__class__.__name__
3079
def __eq__(self, other):
3080
# format objects are generally stateless
3081
return isinstance(other, self.__class__)
3083
def __ne__(self, other):
3084
return not self == other
3087
def find_format(klass, a_bzrdir):
3088
"""Return the format for the repository object in a_bzrdir.
3090
This is used by bzr native formats that have a "format" file in
3091
the repository. Other methods may be used by different types of
3095
transport = a_bzrdir.get_repository_transport(None)
3096
format_string = transport.get("format").read()
3097
return format_registry.get(format_string)
3098
except errors.NoSuchFile:
3099
raise errors.NoRepositoryPresent(a_bzrdir)
3101
raise errors.UnknownFormatError(format=format_string,
3105
def register_format(klass, format):
3106
format_registry.register(format.get_format_string(), format)
3109
def unregister_format(klass, format):
3110
format_registry.remove(format.get_format_string())
3113
def get_default_format(klass):
3114
"""Return the current default format."""
3115
from bzrlib import bzrdir
3116
return bzrdir.format_registry.make_bzrdir('default').repository_format
3118
def get_format_string(self):
3119
"""Return the ASCII format string that identifies this format.
3121
Note that in pre format ?? repositories the format string is
3122
not permitted nor written to disk.
3124
raise NotImplementedError(self.get_format_string)
3126
def get_format_description(self):
3127
"""Return the short description for this format."""
3128
raise NotImplementedError(self.get_format_description)
3130
# TODO: this shouldn't be in the base class, it's specific to things that
3131
# use weaves or knits -- mbp 20070207
3132
def _get_versioned_file_store(self,
3137
versionedfile_class=None,
3138
versionedfile_kwargs={},
3140
if versionedfile_class is None:
3141
versionedfile_class = self._versionedfile_class
3142
weave_transport = control_files._transport.clone(name)
3143
dir_mode = control_files._dir_mode
3144
file_mode = control_files._file_mode
3145
return VersionedFileStore(weave_transport, prefixed=prefixed,
3147
file_mode=file_mode,
3148
versionedfile_class=versionedfile_class,
3149
versionedfile_kwargs=versionedfile_kwargs,
3152
def initialize(self, a_bzrdir, shared=False):
3153
"""Initialize a repository of this format in a_bzrdir.
3155
:param a_bzrdir: The bzrdir to put the new repository in it.
3156
:param shared: The repository should be initialized as a sharable one.
3157
:returns: The new repository object.
3159
This may raise UninitializableFormat if shared repository are not
3160
compatible the a_bzrdir.
3162
raise NotImplementedError(self.initialize)
3164
def is_supported(self):
3165
"""Is this format supported?
3167
Supported formats must be initializable and openable.
3168
Unsupported formats may not support initialization or committing or
3169
some other features depending on the reason for not being supported.
3173
def network_name(self):
3174
"""A simple byte string uniquely identifying this format for RPC calls.
3176
MetaDir repository formats use their disk format string to identify the
3177
repository over the wire. All in one formats such as bzr < 0.8, and
3178
foreign formats like svn/git and hg should use some marker which is
3179
unique and immutable.
3181
raise NotImplementedError(self.network_name)
3183
def check_conversion_target(self, target_format):
3184
if self.rich_root_data and not target_format.rich_root_data:
3185
raise errors.BadConversionTarget(
3186
'Does not support rich root data.', target_format,
3188
if (self.supports_tree_reference and
3189
not getattr(target_format, 'supports_tree_reference', False)):
3190
raise errors.BadConversionTarget(
3191
'Does not support nested trees', target_format,
3194
def open(self, a_bzrdir, _found=False):
3195
"""Return an instance of this format for the bzrdir a_bzrdir.
3197
_found is a private parameter, do not use it.
3199
raise NotImplementedError(self.open)
3202
class MetaDirRepositoryFormat(RepositoryFormat):
3203
"""Common base class for the new repositories using the metadir layout."""
3205
rich_root_data = False
3206
supports_tree_reference = False
3207
supports_external_lookups = False
3210
def _matchingbzrdir(self):
3211
matching = bzrdir.BzrDirMetaFormat1()
3212
matching.repository_format = self
3216
super(MetaDirRepositoryFormat, self).__init__()
3218
def _create_control_files(self, a_bzrdir):
3219
"""Create the required files and the initial control_files object."""
3220
# FIXME: RBC 20060125 don't peek under the covers
3221
# NB: no need to escape relative paths that are url safe.
3222
repository_transport = a_bzrdir.get_repository_transport(self)
3223
control_files = lockable_files.LockableFiles(repository_transport,
3224
'lock', lockdir.LockDir)
3225
control_files.create_lock()
3226
return control_files
3228
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
3229
"""Upload the initial blank content."""
3230
control_files = self._create_control_files(a_bzrdir)
3231
control_files.lock_write()
3232
transport = control_files._transport
3234
utf8_files += [('shared-storage', '')]
3236
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
3237
for (filename, content_stream) in files:
3238
transport.put_file(filename, content_stream,
3239
mode=a_bzrdir._get_file_mode())
3240
for (filename, content_bytes) in utf8_files:
3241
transport.put_bytes_non_atomic(filename, content_bytes,
3242
mode=a_bzrdir._get_file_mode())
3244
control_files.unlock()
3246
def network_name(self):
3247
"""Metadir formats have matching disk and network format strings."""
3248
return self.get_format_string()
3251
# Pre-0.8 formats that don't have a disk format string (because they are
3252
# versioned by the matching control directory). We use the control directories
3253
# disk format string as a key for the network_name because they meet the
3254
# constraints (simple string, unique, immutable).
3255
network_format_registry.register_lazy(
3256
"Bazaar-NG branch, format 5\n",
3257
'bzrlib.repofmt.weaverepo',
3258
'RepositoryFormat5',
3260
network_format_registry.register_lazy(
3261
"Bazaar-NG branch, format 6\n",
3262
'bzrlib.repofmt.weaverepo',
3263
'RepositoryFormat6',
3266
# formats which have no format string are not discoverable or independently
3267
# creatable on disk, so are not registered in format_registry. They're
3268
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
3269
# needed, it's constructed directly by the BzrDir. Non-native formats where
3270
# the repository is not separately opened are similar.
3272
format_registry.register_lazy(
3273
'Bazaar-NG Repository format 7',
3274
'bzrlib.repofmt.weaverepo',
3278
format_registry.register_lazy(
3279
'Bazaar-NG Knit Repository Format 1',
3280
'bzrlib.repofmt.knitrepo',
3281
'RepositoryFormatKnit1',
3284
format_registry.register_lazy(
3285
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
3286
'bzrlib.repofmt.knitrepo',
3287
'RepositoryFormatKnit3',
3290
format_registry.register_lazy(
3291
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
3292
'bzrlib.repofmt.knitrepo',
3293
'RepositoryFormatKnit4',
3296
# Pack-based formats. There is one format for pre-subtrees, and one for
3297
# post-subtrees to allow ease of testing.
3298
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
3299
format_registry.register_lazy(
3300
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
3301
'bzrlib.repofmt.pack_repo',
3302
'RepositoryFormatKnitPack1',
3304
format_registry.register_lazy(
3305
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
3306
'bzrlib.repofmt.pack_repo',
3307
'RepositoryFormatKnitPack3',
3309
format_registry.register_lazy(
3310
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
3311
'bzrlib.repofmt.pack_repo',
3312
'RepositoryFormatKnitPack4',
3314
format_registry.register_lazy(
3315
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
3316
'bzrlib.repofmt.pack_repo',
3317
'RepositoryFormatKnitPack5',
3319
format_registry.register_lazy(
3320
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
3321
'bzrlib.repofmt.pack_repo',
3322
'RepositoryFormatKnitPack5RichRoot',
3324
format_registry.register_lazy(
3325
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3326
'bzrlib.repofmt.pack_repo',
3327
'RepositoryFormatKnitPack5RichRootBroken',
3329
format_registry.register_lazy(
3330
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3331
'bzrlib.repofmt.pack_repo',
3332
'RepositoryFormatKnitPack6',
3334
format_registry.register_lazy(
3335
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3336
'bzrlib.repofmt.pack_repo',
3337
'RepositoryFormatKnitPack6RichRoot',
3340
# Development formats.
3341
# Obsolete but kept pending a CHK based subtree format.
3342
format_registry.register_lazy(
3343
("Bazaar development format 2 with subtree support "
3344
"(needs bzr.dev from before 1.8)\n"),
3345
'bzrlib.repofmt.pack_repo',
3346
'RepositoryFormatPackDevelopment2Subtree',
3349
# 1.14->1.16 go below here
3350
format_registry.register_lazy(
3351
'Bazaar development format - group compression and chk inventory'
3352
' (needs bzr.dev from 1.14)\n',
3353
'bzrlib.repofmt.groupcompress_repo',
3354
'RepositoryFormatCHK1',
3357
format_registry.register_lazy(
3358
'Bazaar development format - chk repository with bencode revision '
3359
'serialization (needs bzr.dev from 1.16)\n',
3360
'bzrlib.repofmt.groupcompress_repo',
3361
'RepositoryFormatCHK2',
3363
format_registry.register_lazy(
3364
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3365
'bzrlib.repofmt.groupcompress_repo',
3366
'RepositoryFormat2a',
3370
class InterRepository(InterObject):
3371
"""This class represents operations taking place between two repositories.
3373
Its instances have methods like copy_content and fetch, and contain
3374
references to the source and target repositories these operations can be
3377
Often we will provide convenience methods on 'repository' which carry out
3378
operations with another repository - they will always forward to
3379
InterRepository.get(other).method_name(parameters).
3382
_walk_to_common_revisions_batch_size = 50
3384
"""The available optimised InterRepository types."""
3387
def copy_content(self, revision_id=None):
3388
"""Make a complete copy of the content in self into destination.
3390
This is a destructive operation! Do not use it on existing
3393
:param revision_id: Only copy the content needed to construct
3394
revision_id and its parents.
3397
self.target.set_make_working_trees(self.source.make_working_trees())
3398
except NotImplementedError:
3400
self.target.fetch(self.source, revision_id=revision_id)
3403
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3405
"""Fetch the content required to construct revision_id.
3407
The content is copied from self.source to self.target.
3409
:param revision_id: if None all content is copied, if NULL_REVISION no
3411
:param pb: optional progress bar to use for progress reports. If not
3412
provided a default one will be created.
3415
from bzrlib.fetch import RepoFetcher
3416
f = RepoFetcher(to_repository=self.target,
3417
from_repository=self.source,
3418
last_revision=revision_id,
3419
fetch_spec=fetch_spec,
3420
pb=pb, find_ghosts=find_ghosts)
3422
def _walk_to_common_revisions(self, revision_ids):
3423
"""Walk out from revision_ids in source to revisions target has.
3425
:param revision_ids: The start point for the search.
3426
:return: A set of revision ids.
3428
target_graph = self.target.get_graph()
3429
revision_ids = frozenset(revision_ids)
3430
missing_revs = set()
3431
source_graph = self.source.get_graph()
3432
# ensure we don't pay silly lookup costs.
3433
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3434
null_set = frozenset([_mod_revision.NULL_REVISION])
3435
searcher_exhausted = False
3439
# Iterate the searcher until we have enough next_revs
3440
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3442
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3443
next_revs.update(next_revs_part)
3444
ghosts.update(ghosts_part)
3445
except StopIteration:
3446
searcher_exhausted = True
3448
# If there are ghosts in the source graph, and the caller asked for
3449
# them, make sure that they are present in the target.
3450
# We don't care about other ghosts as we can't fetch them and
3451
# haven't been asked to.
3452
ghosts_to_check = set(revision_ids.intersection(ghosts))
3453
revs_to_get = set(next_revs).union(ghosts_to_check)
3455
have_revs = set(target_graph.get_parent_map(revs_to_get))
3456
# we always have NULL_REVISION present.
3457
have_revs = have_revs.union(null_set)
3458
# Check if the target is missing any ghosts we need.
3459
ghosts_to_check.difference_update(have_revs)
3461
# One of the caller's revision_ids is a ghost in both the
3462
# source and the target.
3463
raise errors.NoSuchRevision(
3464
self.source, ghosts_to_check.pop())
3465
missing_revs.update(next_revs - have_revs)
3466
# Because we may have walked past the original stop point, make
3467
# sure everything is stopped
3468
stop_revs = searcher.find_seen_ancestors(have_revs)
3469
searcher.stop_searching_any(stop_revs)
3470
if searcher_exhausted:
3472
return searcher.get_result()
3475
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3476
"""Return the revision ids that source has that target does not.
3478
:param revision_id: only return revision ids included by this
3480
:param find_ghosts: If True find missing revisions in deep history
3481
rather than just finding the surface difference.
3482
:return: A bzrlib.graph.SearchResult.
3484
# stop searching at found target revisions.
3485
if not find_ghosts and revision_id is not None:
3486
return self._walk_to_common_revisions([revision_id])
3487
# generic, possibly worst case, slow code path.
3488
target_ids = set(self.target.all_revision_ids())
3489
if revision_id is not None:
3490
source_ids = self.source.get_ancestry(revision_id)
3491
if source_ids[0] is not None:
3492
raise AssertionError()
3495
source_ids = self.source.all_revision_ids()
3496
result_set = set(source_ids).difference(target_ids)
3497
return self.source.revision_ids_to_search_result(result_set)
3500
def _same_model(source, target):
3501
"""True if source and target have the same data representation.
3503
Note: this is always called on the base class; overriding it in a
3504
subclass will have no effect.
3507
InterRepository._assert_same_model(source, target)
3509
except errors.IncompatibleRepositories, e:
3513
def _assert_same_model(source, target):
3514
"""Raise an exception if two repositories do not use the same model.
3516
if source.supports_rich_root() != target.supports_rich_root():
3517
raise errors.IncompatibleRepositories(source, target,
3518
"different rich-root support")
3519
if source._serializer != target._serializer:
3520
raise errors.IncompatibleRepositories(source, target,
3521
"different serializers")
3524
class InterSameDataRepository(InterRepository):
3525
"""Code for converting between repositories that represent the same data.
3527
Data format and model must match for this to work.
3531
def _get_repo_format_to_test(self):
3532
"""Repository format for testing with.
3534
InterSameData can pull from subtree to subtree and from non-subtree to
3535
non-subtree, so we test this with the richest repository format.
3537
from bzrlib.repofmt import knitrepo
3538
return knitrepo.RepositoryFormatKnit3()
3541
def is_compatible(source, target):
3542
return InterRepository._same_model(source, target)
3545
class InterWeaveRepo(InterSameDataRepository):
3546
"""Optimised code paths between Weave based repositories.
3548
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3549
implemented lazy inter-object optimisation.
3553
def _get_repo_format_to_test(self):
3554
from bzrlib.repofmt import weaverepo
3555
return weaverepo.RepositoryFormat7()
3558
def is_compatible(source, target):
3559
"""Be compatible with known Weave formats.
3561
We don't test for the stores being of specific types because that
3562
could lead to confusing results, and there is no need to be
3565
from bzrlib.repofmt.weaverepo import (
3571
return (isinstance(source._format, (RepositoryFormat5,
3573
RepositoryFormat7)) and
3574
isinstance(target._format, (RepositoryFormat5,
3576
RepositoryFormat7)))
3577
except AttributeError:
3581
def copy_content(self, revision_id=None):
3582
"""See InterRepository.copy_content()."""
3583
# weave specific optimised path:
3585
self.target.set_make_working_trees(self.source.make_working_trees())
3586
except (errors.RepositoryUpgradeRequired, NotImplemented):
3588
# FIXME do not peek!
3589
if self.source._transport.listable():
3590
pb = ui.ui_factory.nested_progress_bar()
3592
self.target.texts.insert_record_stream(
3593
self.source.texts.get_record_stream(
3594
self.source.texts.keys(), 'topological', False))
3595
pb.update('Copying inventory', 0, 1)
3596
self.target.inventories.insert_record_stream(
3597
self.source.inventories.get_record_stream(
3598
self.source.inventories.keys(), 'topological', False))
3599
self.target.signatures.insert_record_stream(
3600
self.source.signatures.get_record_stream(
3601
self.source.signatures.keys(),
3603
self.target.revisions.insert_record_stream(
3604
self.source.revisions.get_record_stream(
3605
self.source.revisions.keys(),
3606
'topological', True))
3610
self.target.fetch(self.source, revision_id=revision_id)
3613
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3614
"""See InterRepository.missing_revision_ids()."""
3615
# we want all revisions to satisfy revision_id in source.
3616
# but we don't want to stat every file here and there.
3617
# we want then, all revisions other needs to satisfy revision_id
3618
# checked, but not those that we have locally.
3619
# so the first thing is to get a subset of the revisions to
3620
# satisfy revision_id in source, and then eliminate those that
3621
# we do already have.
3622
# this is slow on high latency connection to self, but as this
3623
# disk format scales terribly for push anyway due to rewriting
3624
# inventory.weave, this is considered acceptable.
3626
if revision_id is not None:
3627
source_ids = self.source.get_ancestry(revision_id)
3628
if source_ids[0] is not None:
3629
raise AssertionError()
3632
source_ids = self.source._all_possible_ids()
3633
source_ids_set = set(source_ids)
3634
# source_ids is the worst possible case we may need to pull.
3635
# now we want to filter source_ids against what we actually
3636
# have in target, but don't try to check for existence where we know
3637
# we do not have a revision as that would be pointless.
3638
target_ids = set(self.target._all_possible_ids())
3639
possibly_present_revisions = target_ids.intersection(source_ids_set)
3640
actually_present_revisions = set(
3641
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3642
required_revisions = source_ids_set.difference(actually_present_revisions)
3643
if revision_id is not None:
3644
# we used get_ancestry to determine source_ids then we are assured all
3645
# revisions referenced are present as they are installed in topological order.
3646
# and the tip revision was validated by get_ancestry.
3647
result_set = required_revisions
3649
# if we just grabbed the possibly available ids, then
3650
# we only have an estimate of whats available and need to validate
3651
# that against the revision records.
3653
self.source._eliminate_revisions_not_present(required_revisions))
3654
return self.source.revision_ids_to_search_result(result_set)
3657
class InterKnitRepo(InterSameDataRepository):
3658
"""Optimised code paths between Knit based repositories."""
3661
def _get_repo_format_to_test(self):
3662
from bzrlib.repofmt import knitrepo
3663
return knitrepo.RepositoryFormatKnit1()
3666
def is_compatible(source, target):
3667
"""Be compatible with known Knit formats.
3669
We don't test for the stores being of specific types because that
3670
could lead to confusing results, and there is no need to be
3673
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3675
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3676
isinstance(target._format, RepositoryFormatKnit))
3677
except AttributeError:
3679
return are_knits and InterRepository._same_model(source, target)
3682
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3683
"""See InterRepository.missing_revision_ids()."""
3684
if revision_id is not None:
3685
source_ids = self.source.get_ancestry(revision_id)
3686
if source_ids[0] is not None:
3687
raise AssertionError()
3690
source_ids = self.source.all_revision_ids()
3691
source_ids_set = set(source_ids)
3692
# source_ids is the worst possible case we may need to pull.
3693
# now we want to filter source_ids against what we actually
3694
# have in target, but don't try to check for existence where we know
3695
# we do not have a revision as that would be pointless.
3696
target_ids = set(self.target.all_revision_ids())
3697
possibly_present_revisions = target_ids.intersection(source_ids_set)
3698
actually_present_revisions = set(
3699
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3700
required_revisions = source_ids_set.difference(actually_present_revisions)
3701
if revision_id is not None:
3702
# we used get_ancestry to determine source_ids then we are assured all
3703
# revisions referenced are present as they are installed in topological order.
3704
# and the tip revision was validated by get_ancestry.
3705
result_set = required_revisions
3707
# if we just grabbed the possibly available ids, then
3708
# we only have an estimate of whats available and need to validate
3709
# that against the revision records.
3711
self.source._eliminate_revisions_not_present(required_revisions))
3712
return self.source.revision_ids_to_search_result(result_set)
3715
class InterDifferingSerializer(InterRepository):
3718
def _get_repo_format_to_test(self):
3722
def is_compatible(source, target):
3723
"""Be compatible with Knit2 source and Knit3 target"""
3724
# This is redundant with format.check_conversion_target(), however that
3725
# raises an exception, and we just want to say "False" as in we won't
3726
# support converting between these formats.
3727
if 'IDS_never' in debug.debug_flags:
3729
if source.supports_rich_root() and not target.supports_rich_root():
3731
if (source._format.supports_tree_reference
3732
and not target._format.supports_tree_reference):
3734
if target._fallback_repositories and target._format.supports_chks:
3735
# IDS doesn't know how to copy CHKs for the parent inventories it
3736
# adds to stacked repos.
3738
if 'IDS_always' in debug.debug_flags:
3740
# Only use this code path for local source and target. IDS does far
3741
# too much IO (both bandwidth and roundtrips) over a network.
3742
if not source.bzrdir.transport.base.startswith('file:///'):
3744
if not target.bzrdir.transport.base.startswith('file:///'):
3748
def _get_trees(self, revision_ids, cache):
3750
for rev_id in revision_ids:
3752
possible_trees.append((rev_id, cache[rev_id]))
3754
# Not cached, but inventory might be present anyway.
3756
tree = self.source.revision_tree(rev_id)
3757
except errors.NoSuchRevision:
3758
# Nope, parent is ghost.
3761
cache[rev_id] = tree
3762
possible_trees.append((rev_id, tree))
3763
return possible_trees
3765
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3766
"""Get the best delta and base for this revision.
3768
:return: (basis_id, delta)
3771
# Generate deltas against each tree, to find the shortest.
3772
texts_possibly_new_in_tree = set()
3773
for basis_id, basis_tree in possible_trees:
3774
delta = tree.inventory._make_delta(basis_tree.inventory)
3775
for old_path, new_path, file_id, new_entry in delta:
3776
if new_path is None:
3777
# This file_id isn't present in the new rev, so we don't
3781
# Rich roots are handled elsewhere...
3783
kind = new_entry.kind
3784
if kind != 'directory' and kind != 'file':
3785
# No text record associated with this inventory entry.
3787
# This is a directory or file that has changed somehow.
3788
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3789
deltas.append((len(delta), basis_id, delta))
3791
return deltas[0][1:]
3793
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3794
"""Find all parent revisions that are absent, but for which the
3795
inventory is present, and copy those inventories.
3797
This is necessary to preserve correctness when the source is stacked
3798
without fallbacks configured. (Note that in cases like upgrade the
3799
source may be not have _fallback_repositories even though it is
3803
for parents in parent_map.values():
3804
parent_revs.update(parents)
3805
present_parents = self.source.get_parent_map(parent_revs)
3806
absent_parents = set(parent_revs).difference(present_parents)
3807
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3808
(rev_id,) for rev_id in absent_parents)
3809
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3810
for parent_tree in self.source.revision_trees(parent_inv_ids):
3811
current_revision_id = parent_tree.get_revision_id()
3812
parents_parents_keys = parent_invs_keys_for_stacking[
3813
(current_revision_id,)]
3814
parents_parents = [key[-1] for key in parents_parents_keys]
3815
basis_id = _mod_revision.NULL_REVISION
3816
basis_tree = self.source.revision_tree(basis_id)
3817
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3818
self.target.add_inventory_by_delta(
3819
basis_id, delta, current_revision_id, parents_parents)
3820
cache[current_revision_id] = parent_tree
3822
def _fetch_batch(self, revision_ids, basis_id, cache):
3823
"""Fetch across a few revisions.
3825
:param revision_ids: The revisions to copy
3826
:param basis_id: The revision_id of a tree that must be in cache, used
3827
as a basis for delta when no other base is available
3828
:param cache: A cache of RevisionTrees that we can use.
3829
:return: The revision_id of the last converted tree. The RevisionTree
3830
for it will be in cache
3832
# Walk though all revisions; get inventory deltas, copy referenced
3833
# texts that delta references, insert the delta, revision and
3835
root_keys_to_create = set()
3838
pending_revisions = []
3839
parent_map = self.source.get_parent_map(revision_ids)
3840
self._fetch_parent_invs_for_stacking(parent_map, cache)
3841
for tree in self.source.revision_trees(revision_ids):
3842
# Find a inventory delta for this revision.
3843
# Find text entries that need to be copied, too.
3844
current_revision_id = tree.get_revision_id()
3845
parent_ids = parent_map.get(current_revision_id, ())
3846
parent_trees = self._get_trees(parent_ids, cache)
3847
possible_trees = list(parent_trees)
3848
if len(possible_trees) == 0:
3849
# There either aren't any parents, or the parents are ghosts,
3850
# so just use the last converted tree.
3851
possible_trees.append((basis_id, cache[basis_id]))
3852
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3854
revision = self.source.get_revision(current_revision_id)
3855
pending_deltas.append((basis_id, delta,
3856
current_revision_id, revision.parent_ids))
3857
if self._converting_to_rich_root:
3858
self._revision_id_to_root_id[current_revision_id] = \
3860
# Determine which texts are in present in this revision but not in
3861
# any of the available parents.
3862
texts_possibly_new_in_tree = set()
3863
for old_path, new_path, file_id, entry in delta:
3864
if new_path is None:
3865
# This file_id isn't present in the new rev
3869
if not self.target.supports_rich_root():
3870
# The target doesn't support rich root, so we don't
3873
if self._converting_to_rich_root:
3874
# This can't be copied normally, we have to insert
3876
root_keys_to_create.add((file_id, entry.revision))
3879
texts_possibly_new_in_tree.add((file_id, entry.revision))
3880
for basis_id, basis_tree in possible_trees:
3881
basis_inv = basis_tree.inventory
3882
for file_key in list(texts_possibly_new_in_tree):
3883
file_id, file_revision = file_key
3885
entry = basis_inv[file_id]
3886
except errors.NoSuchId:
3888
if entry.revision == file_revision:
3889
texts_possibly_new_in_tree.remove(file_key)
3890
text_keys.update(texts_possibly_new_in_tree)
3891
pending_revisions.append(revision)
3892
cache[current_revision_id] = tree
3893
basis_id = current_revision_id
3895
from_texts = self.source.texts
3896
to_texts = self.target.texts
3897
if root_keys_to_create:
3898
from bzrlib.fetch import _new_root_data_stream
3899
root_stream = _new_root_data_stream(
3900
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3902
to_texts.insert_record_stream(root_stream)
3903
to_texts.insert_record_stream(from_texts.get_record_stream(
3904
text_keys, self.target._format._fetch_order,
3905
not self.target._format._fetch_uses_deltas))
3906
# insert inventory deltas
3907
for delta in pending_deltas:
3908
self.target.add_inventory_by_delta(*delta)
3909
if self.target._fallback_repositories:
3910
# Make sure this stacked repository has all the parent inventories
3911
# for the new revisions that we are about to insert. We do this
3912
# before adding the revisions so that no revision is added until
3913
# all the inventories it may depend on are added.
3914
# Note that this is overzealous, as we may have fetched these in an
3917
revision_ids = set()
3918
for revision in pending_revisions:
3919
revision_ids.add(revision.revision_id)
3920
parent_ids.update(revision.parent_ids)
3921
parent_ids.difference_update(revision_ids)
3922
parent_ids.discard(_mod_revision.NULL_REVISION)
3923
parent_map = self.source.get_parent_map(parent_ids)
3924
# we iterate over parent_map and not parent_ids because we don't
3925
# want to try copying any revision which is a ghost
3926
for parent_tree in self.source.revision_trees(parent_map):
3927
current_revision_id = parent_tree.get_revision_id()
3928
parents_parents = parent_map[current_revision_id]
3929
possible_trees = self._get_trees(parents_parents, cache)
3930
if len(possible_trees) == 0:
3931
# There either aren't any parents, or the parents are
3932
# ghosts, so just use the last converted tree.
3933
possible_trees.append((basis_id, cache[basis_id]))
3934
basis_id, delta = self._get_delta_for_revision(parent_tree,
3935
parents_parents, possible_trees)
3936
self.target.add_inventory_by_delta(
3937
basis_id, delta, current_revision_id, parents_parents)
3938
# insert signatures and revisions
3939
for revision in pending_revisions:
3941
signature = self.source.get_signature_text(
3942
revision.revision_id)
3943
self.target.add_signature_text(revision.revision_id,
3945
except errors.NoSuchRevision:
3947
self.target.add_revision(revision.revision_id, revision)
3950
def _fetch_all_revisions(self, revision_ids, pb):
3951
"""Fetch everything for the list of revisions.
3953
:param revision_ids: The list of revisions to fetch. Must be in
3955
:param pb: A ProgressTask
3958
basis_id, basis_tree = self._get_basis(revision_ids[0])
3960
cache = lru_cache.LRUCache(100)
3961
cache[basis_id] = basis_tree
3962
del basis_tree # We don't want to hang on to it here
3964
for offset in range(0, len(revision_ids), batch_size):
3965
self.target.start_write_group()
3967
pb.update('Transferring revisions', offset,
3969
batch = revision_ids[offset:offset+batch_size]
3970
basis_id = self._fetch_batch(batch, basis_id, cache)
3972
self.target.abort_write_group()
3975
hint = self.target.commit_write_group()
3978
if hints and self.target._format.pack_compresses:
3979
self.target.pack(hint=hints)
3980
pb.update('Transferring revisions', len(revision_ids),
3984
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3986
"""See InterRepository.fetch()."""
3987
if fetch_spec is not None:
3988
raise AssertionError("Not implemented yet...")
3989
if (not self.source.supports_rich_root()
3990
and self.target.supports_rich_root()):
3991
self._converting_to_rich_root = True
3992
self._revision_id_to_root_id = {}
3994
self._converting_to_rich_root = False
3995
revision_ids = self.target.search_missing_revision_ids(self.source,
3996
revision_id, find_ghosts=find_ghosts).get_keys()
3997
if not revision_ids:
3999
revision_ids = tsort.topo_sort(
4000
self.source.get_graph().get_parent_map(revision_ids))
4001
if not revision_ids:
4003
# Walk though all revisions; get inventory deltas, copy referenced
4004
# texts that delta references, insert the delta, revision and
4007
my_pb = ui.ui_factory.nested_progress_bar()
4010
symbol_versioning.warn(
4011
symbol_versioning.deprecated_in((1, 14, 0))
4012
% "pb parameter to fetch()")
4015
self._fetch_all_revisions(revision_ids, pb)
4017
if my_pb is not None:
4019
return len(revision_ids), 0
4021
def _get_basis(self, first_revision_id):
4022
"""Get a revision and tree which exists in the target.
4024
This assumes that first_revision_id is selected for transmission
4025
because all other ancestors are already present. If we can't find an
4026
ancestor we fall back to NULL_REVISION since we know that is safe.
4028
:return: (basis_id, basis_tree)
4030
first_rev = self.source.get_revision(first_revision_id)
4032
basis_id = first_rev.parent_ids[0]
4033
# only valid as a basis if the target has it
4034
self.target.get_revision(basis_id)
4035
# Try to get a basis tree - if its a ghost it will hit the
4036
# NoSuchRevision case.
4037
basis_tree = self.source.revision_tree(basis_id)
4038
except (IndexError, errors.NoSuchRevision):
4039
basis_id = _mod_revision.NULL_REVISION
4040
basis_tree = self.source.revision_tree(basis_id)
4041
return basis_id, basis_tree
4044
InterRepository.register_optimiser(InterDifferingSerializer)
4045
InterRepository.register_optimiser(InterSameDataRepository)
4046
InterRepository.register_optimiser(InterWeaveRepo)
4047
InterRepository.register_optimiser(InterKnitRepo)
4050
class CopyConverter(object):
4051
"""A repository conversion tool which just performs a copy of the content.
4053
This is slow but quite reliable.
4056
def __init__(self, target_format):
4057
"""Create a CopyConverter.
4059
:param target_format: The format the resulting repository should be.
4061
self.target_format = target_format
4063
def convert(self, repo, pb):
4064
"""Perform the conversion of to_convert, giving feedback via pb.
4066
:param to_convert: The disk object to convert.
4067
:param pb: a progress bar to use for progress information.
4072
# this is only useful with metadir layouts - separated repo content.
4073
# trigger an assertion if not such
4074
repo._format.get_format_string()
4075
self.repo_dir = repo.bzrdir
4076
self.step('Moving repository to repository.backup')
4077
self.repo_dir.transport.move('repository', 'repository.backup')
4078
backup_transport = self.repo_dir.transport.clone('repository.backup')
4079
repo._format.check_conversion_target(self.target_format)
4080
self.source_repo = repo._format.open(self.repo_dir,
4082
_override_transport=backup_transport)
4083
self.step('Creating new repository')
4084
converted = self.target_format.initialize(self.repo_dir,
4085
self.source_repo.is_shared())
4086
converted.lock_write()
4088
self.step('Copying content')
4089
self.source_repo.copy_content_into(converted)
4092
self.step('Deleting old repository content')
4093
self.repo_dir.transport.delete_tree('repository.backup')
4094
ui.ui_factory.note('repository converted')
4096
def step(self, message):
4097
"""Update the pb by a step."""
4099
self.pb.update(message, self.count, self.total)
4111
def _unescaper(match, _map=_unescape_map):
4112
code = match.group(1)
4116
if not code.startswith('#'):
4118
return unichr(int(code[1:])).encode('utf8')
4124
def _unescape_xml(data):
4125
"""Unescape predefined XML entities in a string of data."""
4127
if _unescape_re is None:
4128
_unescape_re = re.compile('\&([^;]*);')
4129
return _unescape_re.sub(_unescaper, data)
4132
class _VersionedFileChecker(object):
4134
def __init__(self, repository, text_key_references=None, ancestors=None):
4135
self.repository = repository
4136
self.text_index = self.repository._generate_text_key_index(
4137
text_key_references=text_key_references, ancestors=ancestors)
4139
def calculate_file_version_parents(self, text_key):
4140
"""Calculate the correct parents for a file version according to
4143
parent_keys = self.text_index[text_key]
4144
if parent_keys == [_mod_revision.NULL_REVISION]:
4146
return tuple(parent_keys)
4148
def check_file_version_parents(self, texts, progress_bar=None):
4149
"""Check the parents stored in a versioned file are correct.
4151
It also detects file versions that are not referenced by their
4152
corresponding revision's inventory.
4154
:returns: A tuple of (wrong_parents, dangling_file_versions).
4155
wrong_parents is a dict mapping {revision_id: (stored_parents,
4156
correct_parents)} for each revision_id where the stored parents
4157
are not correct. dangling_file_versions is a set of (file_id,
4158
revision_id) tuples for versions that are present in this versioned
4159
file, but not used by the corresponding inventory.
4161
local_progress = None
4162
if progress_bar is None:
4163
local_progress = ui.ui_factory.nested_progress_bar()
4164
progress_bar = local_progress
4166
return self._check_file_version_parents(texts, progress_bar)
4169
local_progress.finished()
4171
def _check_file_version_parents(self, texts, progress_bar):
4172
"""See check_file_version_parents."""
4174
self.file_ids = set([file_id for file_id, _ in
4175
self.text_index.iterkeys()])
4176
# text keys is now grouped by file_id
4177
n_versions = len(self.text_index)
4178
progress_bar.update('loading text store', 0, n_versions)
4179
parent_map = self.repository.texts.get_parent_map(self.text_index)
4180
# On unlistable transports this could well be empty/error...
4181
text_keys = self.repository.texts.keys()
4182
unused_keys = frozenset(text_keys) - set(self.text_index)
4183
for num, key in enumerate(self.text_index.iterkeys()):
4184
progress_bar.update('checking text graph', num, n_versions)
4185
correct_parents = self.calculate_file_version_parents(key)
4187
knit_parents = parent_map[key]
4188
except errors.RevisionNotPresent:
4191
if correct_parents != knit_parents:
4192
wrong_parents[key] = (knit_parents, correct_parents)
4193
return wrong_parents, unused_keys
4196
def _old_get_graph(repository, revision_id):
4197
"""DO NOT USE. That is all. I'm serious."""
4198
graph = repository.get_graph()
4199
revision_graph = dict(((key, value) for key, value in
4200
graph.iter_ancestry([revision_id]) if value is not None))
4201
return _strip_NULL_ghosts(revision_graph)
4204
def _strip_NULL_ghosts(revision_graph):
4205
"""Also don't use this. more compatibility code for unmigrated clients."""
4206
# Filter ghosts, and null:
4207
if _mod_revision.NULL_REVISION in revision_graph:
4208
del revision_graph[_mod_revision.NULL_REVISION]
4209
for key, parents in revision_graph.items():
4210
revision_graph[key] = tuple(parent for parent in parents if parent
4212
return revision_graph
4215
class StreamSink(object):
4216
"""An object that can insert a stream into a repository.
4218
This interface handles the complexity of reserialising inventories and
4219
revisions from different formats, and allows unidirectional insertion into
4220
stacked repositories without looking for the missing basis parents
4224
def __init__(self, target_repo):
4225
self.target_repo = target_repo
4227
def insert_stream(self, stream, src_format, resume_tokens):
4228
"""Insert a stream's content into the target repository.
4230
:param src_format: a bzr repository format.
4232
:return: a list of resume tokens and an iterable of keys additional
4233
items required before the insertion can be completed.
4235
self.target_repo.lock_write()
4238
self.target_repo.resume_write_group(resume_tokens)
4241
self.target_repo.start_write_group()
4244
# locked_insert_stream performs a commit|suspend.
4245
return self._locked_insert_stream(stream, src_format, is_resume)
4247
self.target_repo.abort_write_group(suppress_errors=True)
4250
self.target_repo.unlock()
4252
def _locked_insert_stream(self, stream, src_format, is_resume):
4253
to_serializer = self.target_repo._format._serializer
4254
src_serializer = src_format._serializer
4256
if to_serializer == src_serializer:
4257
# If serializers match and the target is a pack repository, set the
4258
# write cache size on the new pack. This avoids poor performance
4259
# on transports where append is unbuffered (such as
4260
# RemoteTransport). This is safe to do because nothing should read
4261
# back from the target repository while a stream with matching
4262
# serialization is being inserted.
4263
# The exception is that a delta record from the source that should
4264
# be a fulltext may need to be expanded by the target (see
4265
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4266
# explicitly flush any buffered writes first in that rare case.
4268
new_pack = self.target_repo._pack_collection._new_pack
4269
except AttributeError:
4270
# Not a pack repository
4273
new_pack.set_write_cache_size(1024*1024)
4274
for substream_type, substream in stream:
4275
if 'stream' in debug.debug_flags:
4276
mutter('inserting substream: %s', substream_type)
4277
if substream_type == 'texts':
4278
self.target_repo.texts.insert_record_stream(substream)
4279
elif substream_type == 'inventories':
4280
if src_serializer == to_serializer:
4281
self.target_repo.inventories.insert_record_stream(
4284
self._extract_and_insert_inventories(
4285
substream, src_serializer)
4286
elif substream_type == 'inventory-deltas':
4287
self._extract_and_insert_inventory_deltas(
4288
substream, src_serializer)
4289
elif substream_type == 'chk_bytes':
4290
# XXX: This doesn't support conversions, as it assumes the
4291
# conversion was done in the fetch code.
4292
self.target_repo.chk_bytes.insert_record_stream(substream)
4293
elif substream_type == 'revisions':
4294
# This may fallback to extract-and-insert more often than
4295
# required if the serializers are different only in terms of
4297
if src_serializer == to_serializer:
4298
self.target_repo.revisions.insert_record_stream(
4301
self._extract_and_insert_revisions(substream,
4303
elif substream_type == 'signatures':
4304
self.target_repo.signatures.insert_record_stream(substream)
4306
raise AssertionError('kaboom! %s' % (substream_type,))
4307
# Done inserting data, and the missing_keys calculations will try to
4308
# read back from the inserted data, so flush the writes to the new pack
4309
# (if this is pack format).
4310
if new_pack is not None:
4311
new_pack._write_data('', flush=True)
4312
# Find all the new revisions (including ones from resume_tokens)
4313
missing_keys = self.target_repo.get_missing_parent_inventories(
4314
check_for_missing_texts=is_resume)
4316
for prefix, versioned_file in (
4317
('texts', self.target_repo.texts),
4318
('inventories', self.target_repo.inventories),
4319
('revisions', self.target_repo.revisions),
4320
('signatures', self.target_repo.signatures),
4321
('chk_bytes', self.target_repo.chk_bytes),
4323
if versioned_file is None:
4325
# TODO: key is often going to be a StaticTuple object
4326
# I don't believe we can define a method by which
4327
# (prefix,) + StaticTuple will work, though we could
4328
# define a StaticTuple.sq_concat that would allow you to
4329
# pass in either a tuple or a StaticTuple as the second
4330
# object, so instead we could have:
4331
# StaticTuple(prefix) + key here...
4332
missing_keys.update((prefix,) + key for key in
4333
versioned_file.get_missing_compression_parent_keys())
4334
except NotImplementedError:
4335
# cannot even attempt suspending, and missing would have failed
4336
# during stream insertion.
4337
missing_keys = set()
4340
# suspend the write group and tell the caller what we is
4341
# missing. We know we can suspend or else we would not have
4342
# entered this code path. (All repositories that can handle
4343
# missing keys can handle suspending a write group).
4344
write_group_tokens = self.target_repo.suspend_write_group()
4345
return write_group_tokens, missing_keys
4346
hint = self.target_repo.commit_write_group()
4347
if (to_serializer != src_serializer and
4348
self.target_repo._format.pack_compresses):
4349
self.target_repo.pack(hint=hint)
4352
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4353
target_rich_root = self.target_repo._format.rich_root_data
4354
target_tree_refs = self.target_repo._format.supports_tree_reference
4355
for record in substream:
4356
# Insert the delta directly
4357
inventory_delta_bytes = record.get_bytes_as('fulltext')
4358
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4360
parse_result = deserialiser.parse_text_bytes(
4361
inventory_delta_bytes)
4362
except inventory_delta.IncompatibleInventoryDelta, err:
4363
trace.mutter("Incompatible delta: %s", err.msg)
4364
raise errors.IncompatibleRevision(self.target_repo._format)
4365
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4366
revision_id = new_id
4367
parents = [key[0] for key in record.parents]
4368
self.target_repo.add_inventory_by_delta(
4369
basis_id, inv_delta, revision_id, parents)
4371
def _extract_and_insert_inventories(self, substream, serializer,
4373
"""Generate a new inventory versionedfile in target, converting data.
4375
The inventory is retrieved from the source, (deserializing it), and
4376
stored in the target (reserializing it in a different format).
4378
target_rich_root = self.target_repo._format.rich_root_data
4379
target_tree_refs = self.target_repo._format.supports_tree_reference
4380
for record in substream:
4381
# It's not a delta, so it must be a fulltext in the source
4382
# serializer's format.
4383
bytes = record.get_bytes_as('fulltext')
4384
revision_id = record.key[0]
4385
inv = serializer.read_inventory_from_string(bytes, revision_id)
4386
parents = [key[0] for key in record.parents]
4387
self.target_repo.add_inventory(revision_id, inv, parents)
4388
# No need to keep holding this full inv in memory when the rest of
4389
# the substream is likely to be all deltas.
4392
def _extract_and_insert_revisions(self, substream, serializer):
4393
for record in substream:
4394
bytes = record.get_bytes_as('fulltext')
4395
revision_id = record.key[0]
4396
rev = serializer.read_revision_from_string(bytes)
4397
if rev.revision_id != revision_id:
4398
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4399
self.target_repo.add_revision(revision_id, rev)
4402
if self.target_repo._format._fetch_reconcile:
4403
self.target_repo.reconcile()
4406
class StreamSource(object):
4407
"""A source of a stream for fetching between repositories."""
4409
def __init__(self, from_repository, to_format):
4410
"""Create a StreamSource streaming from from_repository."""
4411
self.from_repository = from_repository
4412
self.to_format = to_format
4414
def delta_on_metadata(self):
4415
"""Return True if delta's are permitted on metadata streams.
4417
That is on revisions and signatures.
4419
src_serializer = self.from_repository._format._serializer
4420
target_serializer = self.to_format._serializer
4421
return (self.to_format._fetch_uses_deltas and
4422
src_serializer == target_serializer)
4424
def _fetch_revision_texts(self, revs):
4425
# fetch signatures first and then the revision texts
4426
# may need to be a InterRevisionStore call here.
4427
from_sf = self.from_repository.signatures
4428
# A missing signature is just skipped.
4429
keys = [(rev_id,) for rev_id in revs]
4430
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4432
self.to_format._fetch_order,
4433
not self.to_format._fetch_uses_deltas))
4434
# If a revision has a delta, this is actually expanded inside the
4435
# insert_record_stream code now, which is an alternate fix for
4437
from_rf = self.from_repository.revisions
4438
revisions = from_rf.get_record_stream(
4440
self.to_format._fetch_order,
4441
not self.delta_on_metadata())
4442
return [('signatures', signatures), ('revisions', revisions)]
4444
def _generate_root_texts(self, revs):
4445
"""This will be called by get_stream between fetching weave texts and
4446
fetching the inventory weave.
4448
if self._rich_root_upgrade():
4450
return bzrlib.fetch.Inter1and2Helper(
4451
self.from_repository).generate_root_texts(revs)
4455
def get_stream(self, search):
4457
revs = search.get_keys()
4458
graph = self.from_repository.get_graph()
4459
revs = tsort.topo_sort(graph.get_parent_map(revs))
4460
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4462
for knit_kind, file_id, revisions in data_to_fetch:
4463
if knit_kind != phase:
4465
# Make a new progress bar for this phase
4466
if knit_kind == "file":
4467
# Accumulate file texts
4468
text_keys.extend([(file_id, revision) for revision in
4470
elif knit_kind == "inventory":
4471
# Now copy the file texts.
4472
from_texts = self.from_repository.texts
4473
yield ('texts', from_texts.get_record_stream(
4474
text_keys, self.to_format._fetch_order,
4475
not self.to_format._fetch_uses_deltas))
4476
# Cause an error if a text occurs after we have done the
4479
# Before we process the inventory we generate the root
4480
# texts (if necessary) so that the inventories references
4482
for _ in self._generate_root_texts(revs):
4484
# we fetch only the referenced inventories because we do not
4485
# know for unselected inventories whether all their required
4486
# texts are present in the other repository - it could be
4488
for info in self._get_inventory_stream(revs):
4490
elif knit_kind == "signatures":
4491
# Nothing to do here; this will be taken care of when
4492
# _fetch_revision_texts happens.
4494
elif knit_kind == "revisions":
4495
for record in self._fetch_revision_texts(revs):
4498
raise AssertionError("Unknown knit kind %r" % knit_kind)
4500
def get_stream_for_missing_keys(self, missing_keys):
4501
# missing keys can only occur when we are byte copying and not
4502
# translating (because translation means we don't send
4503
# unreconstructable deltas ever).
4505
keys['texts'] = set()
4506
keys['revisions'] = set()
4507
keys['inventories'] = set()
4508
keys['chk_bytes'] = set()
4509
keys['signatures'] = set()
4510
for key in missing_keys:
4511
keys[key[0]].add(key[1:])
4512
if len(keys['revisions']):
4513
# If we allowed copying revisions at this point, we could end up
4514
# copying a revision without copying its required texts: a
4515
# violation of the requirements for repository integrity.
4516
raise AssertionError(
4517
'cannot copy revisions to fill in missing deltas %s' % (
4518
keys['revisions'],))
4519
for substream_kind, keys in keys.iteritems():
4520
vf = getattr(self.from_repository, substream_kind)
4521
if vf is None and keys:
4522
raise AssertionError(
4523
"cannot fill in keys for a versioned file we don't"
4524
" have: %s needs %s" % (substream_kind, keys))
4526
# No need to stream something we don't have
4528
if substream_kind == 'inventories':
4529
# Some missing keys are genuinely ghosts, filter those out.
4530
present = self.from_repository.inventories.get_parent_map(keys)
4531
revs = [key[0] for key in present]
4532
# Get the inventory stream more-or-less as we do for the
4533
# original stream; there's no reason to assume that records
4534
# direct from the source will be suitable for the sink. (Think
4535
# e.g. 2a -> 1.9-rich-root).
4536
for info in self._get_inventory_stream(revs, missing=True):
4540
# Ask for full texts always so that we don't need more round trips
4541
# after this stream.
4542
# Some of the missing keys are genuinely ghosts, so filter absent
4543
# records. The Sink is responsible for doing another check to
4544
# ensure that ghosts don't introduce missing data for future
4546
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4547
self.to_format._fetch_order, True))
4548
yield substream_kind, stream
4550
def inventory_fetch_order(self):
4551
if self._rich_root_upgrade():
4552
return 'topological'
4554
return self.to_format._fetch_order
4556
def _rich_root_upgrade(self):
4557
return (not self.from_repository._format.rich_root_data and
4558
self.to_format.rich_root_data)
4560
def _get_inventory_stream(self, revision_ids, missing=False):
4561
from_format = self.from_repository._format
4562
if (from_format.supports_chks and self.to_format.supports_chks and
4563
from_format.network_name() == self.to_format.network_name()):
4564
raise AssertionError(
4565
"this case should be handled by GroupCHKStreamSource")
4566
elif 'forceinvdeltas' in debug.debug_flags:
4567
return self._get_convertable_inventory_stream(revision_ids,
4568
delta_versus_null=missing)
4569
elif from_format.network_name() == self.to_format.network_name():
4571
return self._get_simple_inventory_stream(revision_ids,
4573
elif (not from_format.supports_chks and not self.to_format.supports_chks
4574
and from_format._serializer == self.to_format._serializer):
4575
# Essentially the same format.
4576
return self._get_simple_inventory_stream(revision_ids,
4579
# Any time we switch serializations, we want to use an
4580
# inventory-delta based approach.
4581
return self._get_convertable_inventory_stream(revision_ids,
4582
delta_versus_null=missing)
4584
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4585
# NB: This currently reopens the inventory weave in source;
4586
# using a single stream interface instead would avoid this.
4587
from_weave = self.from_repository.inventories
4589
delta_closure = True
4591
delta_closure = not self.delta_on_metadata()
4592
yield ('inventories', from_weave.get_record_stream(
4593
[(rev_id,) for rev_id in revision_ids],
4594
self.inventory_fetch_order(), delta_closure))
4596
def _get_convertable_inventory_stream(self, revision_ids,
4597
delta_versus_null=False):
4598
# The source is using CHKs, but the target either doesn't or it has a
4599
# different serializer. The StreamSink code expects to be able to
4600
# convert on the target, so we need to put bytes-on-the-wire that can
4601
# be converted. That means inventory deltas (if the remote is <1.19,
4602
# RemoteStreamSink will fallback to VFS to insert the deltas).
4603
yield ('inventory-deltas',
4604
self._stream_invs_as_deltas(revision_ids,
4605
delta_versus_null=delta_versus_null))
4607
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4608
"""Return a stream of inventory-deltas for the given rev ids.
4610
:param revision_ids: The list of inventories to transmit
4611
:param delta_versus_null: Don't try to find a minimal delta for this
4612
entry, instead compute the delta versus the NULL_REVISION. This
4613
effectively streams a complete inventory. Used for stuff like
4614
filling in missing parents, etc.
4616
from_repo = self.from_repository
4617
revision_keys = [(rev_id,) for rev_id in revision_ids]
4618
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4619
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4621
inventories = self.from_repository.iter_inventories(
4622
revision_ids, 'topological')
4623
format = from_repo._format
4624
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4625
inventory_cache = lru_cache.LRUCache(50)
4626
null_inventory = from_repo.revision_tree(
4627
_mod_revision.NULL_REVISION).inventory
4628
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4629
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4630
# repo back into a non-rich-root repo ought to be allowed)
4631
serializer = inventory_delta.InventoryDeltaSerializer(
4632
versioned_root=format.rich_root_data,
4633
tree_references=format.supports_tree_reference)
4634
for inv in inventories:
4635
key = (inv.revision_id,)
4636
parent_keys = parent_map.get(key, ())
4638
if not delta_versus_null and parent_keys:
4639
# The caller did not ask for complete inventories and we have
4640
# some parents that we can delta against. Make a delta against
4641
# each parent so that we can find the smallest.
4642
parent_ids = [parent_key[0] for parent_key in parent_keys]
4643
for parent_id in parent_ids:
4644
if parent_id not in invs_sent_so_far:
4645
# We don't know that the remote side has this basis, so
4648
if parent_id == _mod_revision.NULL_REVISION:
4649
parent_inv = null_inventory
4651
parent_inv = inventory_cache.get(parent_id, None)
4652
if parent_inv is None:
4653
parent_inv = from_repo.get_inventory(parent_id)
4654
candidate_delta = inv._make_delta(parent_inv)
4655
if (delta is None or
4656
len(delta) > len(candidate_delta)):
4657
delta = candidate_delta
4658
basis_id = parent_id
4660
# Either none of the parents ended up being suitable, or we
4661
# were asked to delta against NULL
4662
basis_id = _mod_revision.NULL_REVISION
4663
delta = inv._make_delta(null_inventory)
4664
invs_sent_so_far.add(inv.revision_id)
4665
inventory_cache[inv.revision_id] = inv
4666
delta_serialized = ''.join(
4667
serializer.delta_to_lines(basis_id, key[-1], delta))
4668
yield versionedfile.FulltextContentFactory(
4669
key, parent_keys, None, delta_serialized)
4672
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4673
stop_revision=None):
4674
"""Extend the partial history to include a given index
4676
If a stop_index is supplied, stop when that index has been reached.
4677
If a stop_revision is supplied, stop when that revision is
4678
encountered. Otherwise, stop when the beginning of history is
4681
:param stop_index: The index which should be present. When it is
4682
present, history extension will stop.
4683
:param stop_revision: The revision id which should be present. When
4684
it is encountered, history extension will stop.
4686
start_revision = partial_history_cache[-1]
4687
iterator = repo.iter_reverse_revision_history(start_revision)
4689
#skip the last revision in the list
4692
if (stop_index is not None and
4693
len(partial_history_cache) > stop_index):
4695
if partial_history_cache[-1] == stop_revision:
4697
revision_id = iterator.next()
4698
partial_history_cache.append(revision_id)
4699
except StopIteration:
added
removed
bzrlib/repository.py
