1
# Copyright (C) 2005, 2006, 2007, 2008, 2009 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
from bzrlib.lazy_import import lazy_import
18
lazy_import(globals(), """
37
revision as _mod_revision,
43
from bzrlib.bundle import serializer
44
from bzrlib.revisiontree import RevisionTree
45
from bzrlib.store.versioned import VersionedFileStore
46
from bzrlib.testament import Testament
49
from bzrlib.decorators import needs_read_lock, needs_write_lock
50
from bzrlib.inter import InterObject
51
from bzrlib.inventory import (
57
from bzrlib import registry
58
from bzrlib.symbol_versioning import (
61
from bzrlib.trace import (
62
log_exception_quietly, note, mutter, mutter_callsite, warning)
65
# Old formats display a warning, but only once
66
_deprecation_warning_done = False
69
class CommitBuilder(object):
70
"""Provides an interface to build up a commit.
72
This allows describing a tree to be committed without needing to
73
know the internals of the format of the repository.
76
# all clients should supply tree roots.
77
record_root_entry = True
78
# the default CommitBuilder does not manage trees whose root is versioned.
79
_versioned_root = False
81
def __init__(self, repository, parents, config, timestamp=None,
82
timezone=None, committer=None, revprops=None,
84
"""Initiate a CommitBuilder.
86
:param repository: Repository to commit to.
87
:param parents: Revision ids of the parents of the new revision.
88
:param config: Configuration to use.
89
:param timestamp: Optional timestamp recorded for commit.
90
:param timezone: Optional timezone for timestamp.
91
:param committer: Optional committer to set for commit.
92
:param revprops: Optional dictionary of revision properties.
93
:param revision_id: Optional revision id.
98
self._committer = self._config.username()
100
self._committer = committer
102
self.new_inventory = Inventory(None)
103
self._new_revision_id = revision_id
104
self.parents = parents
105
self.repository = repository
108
if revprops is not None:
109
self._validate_revprops(revprops)
110
self._revprops.update(revprops)
112
if timestamp is None:
113
timestamp = time.time()
114
# Restrict resolution to 1ms
115
self._timestamp = round(timestamp, 3)
118
self._timezone = osutils.local_time_offset()
120
self._timezone = int(timezone)
122
self._generate_revision_if_needed()
123
self.__heads = graph.HeadsCache(repository.get_graph()).heads
124
self._basis_delta = []
125
# API compatibility, older code that used CommitBuilder did not call
126
# .record_delete(), which means the delta that is computed would not be
127
# valid. Callers that will call record_delete() should call
128
# .will_record_deletes() to indicate that.
129
self._recording_deletes = False
130
# memo'd check for no-op commits.
131
self._any_changes = False
133
def any_changes(self):
134
"""Return True if any entries were changed.
136
This includes merge-only changes. It is the core for the --unchanged
139
:return: True if any changes have occured.
141
return self._any_changes
143
def _validate_unicode_text(self, text, context):
144
"""Verify things like commit messages don't have bogus characters."""
146
raise ValueError('Invalid value for %s: %r' % (context, text))
148
def _validate_revprops(self, revprops):
149
for key, value in revprops.iteritems():
150
# We know that the XML serializers do not round trip '\r'
151
# correctly, so refuse to accept them
152
if not isinstance(value, basestring):
153
raise ValueError('revision property (%s) is not a valid'
154
' (unicode) string: %r' % (key, value))
155
self._validate_unicode_text(value,
156
'revision property (%s)' % (key,))
158
def commit(self, message):
159
"""Make the actual commit.
161
:return: The revision id of the recorded revision.
163
self._validate_unicode_text(message, 'commit message')
164
rev = _mod_revision.Revision(
165
timestamp=self._timestamp,
166
timezone=self._timezone,
167
committer=self._committer,
169
inventory_sha1=self.inv_sha1,
170
revision_id=self._new_revision_id,
171
properties=self._revprops)
172
rev.parent_ids = self.parents
173
self.repository.add_revision(self._new_revision_id, rev,
174
self.new_inventory, self._config)
175
self.repository.commit_write_group()
176
return self._new_revision_id
179
"""Abort the commit that is being built.
181
self.repository.abort_write_group()
183
def revision_tree(self):
184
"""Return the tree that was just committed.
186
After calling commit() this can be called to get a RevisionTree
187
representing the newly committed tree. This is preferred to
188
calling Repository.revision_tree() because that may require
189
deserializing the inventory, while we already have a copy in
192
if self.new_inventory is None:
193
self.new_inventory = self.repository.get_inventory(
194
self._new_revision_id)
195
return RevisionTree(self.repository, self.new_inventory,
196
self._new_revision_id)
198
def finish_inventory(self):
199
"""Tell the builder that the inventory is finished.
201
:return: The inventory id in the repository, which can be used with
202
repository.get_inventory.
204
if self.new_inventory is None:
205
# an inventory delta was accumulated without creating a new
207
basis_id = self.basis_delta_revision
208
self.inv_sha1 = self.repository.add_inventory_by_delta(
209
basis_id, self._basis_delta, self._new_revision_id,
212
if self.new_inventory.root is None:
213
raise AssertionError('Root entry should be supplied to'
214
' record_entry_contents, as of bzr 0.10.')
215
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
216
self.new_inventory.revision_id = self._new_revision_id
217
self.inv_sha1 = self.repository.add_inventory(
218
self._new_revision_id,
222
return self._new_revision_id
224
def _gen_revision_id(self):
225
"""Return new revision-id."""
226
return generate_ids.gen_revision_id(self._config.username(),
229
def _generate_revision_if_needed(self):
230
"""Create a revision id if None was supplied.
232
If the repository can not support user-specified revision ids
233
they should override this function and raise CannotSetRevisionId
234
if _new_revision_id is not None.
236
:raises: CannotSetRevisionId
238
if self._new_revision_id is None:
239
self._new_revision_id = self._gen_revision_id()
240
self.random_revid = True
242
self.random_revid = False
244
def _heads(self, file_id, revision_ids):
245
"""Calculate the graph heads for revision_ids in the graph of file_id.
247
This can use either a per-file graph or a global revision graph as we
248
have an identity relationship between the two graphs.
250
return self.__heads(revision_ids)
252
def _check_root(self, ie, parent_invs, tree):
253
"""Helper for record_entry_contents.
255
:param ie: An entry being added.
256
:param parent_invs: The inventories of the parent revisions of the
258
:param tree: The tree that is being committed.
260
# In this revision format, root entries have no knit or weave When
261
# serializing out to disk and back in root.revision is always
263
ie.revision = self._new_revision_id
265
def _require_root_change(self, tree):
266
"""Enforce an appropriate root object change.
268
This is called once when record_iter_changes is called, if and only if
269
the root was not in the delta calculated by record_iter_changes.
271
:param tree: The tree which is being committed.
273
# NB: if there are no parents then this method is not called, so no
274
# need to guard on parents having length.
275
entry = entry_factory['directory'](tree.path2id(''), '',
277
entry.revision = self._new_revision_id
278
self._basis_delta.append(('', '', entry.file_id, entry))
280
def _get_delta(self, ie, basis_inv, path):
281
"""Get a delta against the basis inventory for ie."""
282
if ie.file_id not in basis_inv:
284
result = (None, path, ie.file_id, ie)
285
self._basis_delta.append(result)
287
elif ie != basis_inv[ie.file_id]:
289
# TODO: avoid tis id2path call.
290
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
291
self._basis_delta.append(result)
297
def get_basis_delta(self):
298
"""Return the complete inventory delta versus the basis inventory.
300
This has been built up with the calls to record_delete and
301
record_entry_contents. The client must have already called
302
will_record_deletes() to indicate that they will be generating a
305
:return: An inventory delta, suitable for use with apply_delta, or
306
Repository.add_inventory_by_delta, etc.
308
if not self._recording_deletes:
309
raise AssertionError("recording deletes not activated.")
310
return self._basis_delta
312
def record_delete(self, path, file_id):
313
"""Record that a delete occured against a basis tree.
315
This is an optional API - when used it adds items to the basis_delta
316
being accumulated by the commit builder. It cannot be called unless the
317
method will_record_deletes() has been called to inform the builder that
318
a delta is being supplied.
320
:param path: The path of the thing deleted.
321
:param file_id: The file id that was deleted.
323
if not self._recording_deletes:
324
raise AssertionError("recording deletes not activated.")
325
delta = (path, None, file_id, None)
326
self._basis_delta.append(delta)
329
def will_record_deletes(self):
330
"""Tell the commit builder that deletes are being notified.
332
This enables the accumulation of an inventory delta; for the resulting
333
commit to be valid, deletes against the basis MUST be recorded via
334
builder.record_delete().
336
self._recording_deletes = True
338
basis_id = self.parents[0]
340
basis_id = _mod_revision.NULL_REVISION
341
self.basis_delta_revision = basis_id
343
def record_entry_contents(self, ie, parent_invs, path, tree,
345
"""Record the content of ie from tree into the commit if needed.
347
Side effect: sets ie.revision when unchanged
349
:param ie: An inventory entry present in the commit.
350
:param parent_invs: The inventories of the parent revisions of the
352
:param path: The path the entry is at in the tree.
353
:param tree: The tree which contains this entry and should be used to
355
:param content_summary: Summary data from the tree about the paths
356
content - stat, length, exec, sha/link target. This is only
357
accessed when the entry has a revision of None - that is when it is
358
a candidate to commit.
359
:return: A tuple (change_delta, version_recorded, fs_hash).
360
change_delta is an inventory_delta change for this entry against
361
the basis tree of the commit, or None if no change occured against
363
version_recorded is True if a new version of the entry has been
364
recorded. For instance, committing a merge where a file was only
365
changed on the other side will return (delta, False).
366
fs_hash is either None, or the hash details for the path (currently
367
a tuple of the contents sha1 and the statvalue returned by
368
tree.get_file_with_stat()).
370
if self.new_inventory.root is None:
371
if ie.parent_id is not None:
372
raise errors.RootMissing()
373
self._check_root(ie, parent_invs, tree)
374
if ie.revision is None:
375
kind = content_summary[0]
377
# ie is carried over from a prior commit
379
# XXX: repository specific check for nested tree support goes here - if
380
# the repo doesn't want nested trees we skip it ?
381
if (kind == 'tree-reference' and
382
not self.repository._format.supports_tree_reference):
383
# mismatch between commit builder logic and repository:
384
# this needs the entry creation pushed down into the builder.
385
raise NotImplementedError('Missing repository subtree support.')
386
self.new_inventory.add(ie)
388
# TODO: slow, take it out of the inner loop.
390
basis_inv = parent_invs[0]
392
basis_inv = Inventory(root_id=None)
394
# ie.revision is always None if the InventoryEntry is considered
395
# for committing. We may record the previous parents revision if the
396
# content is actually unchanged against a sole head.
397
if ie.revision is not None:
398
if not self._versioned_root and path == '':
399
# repositories that do not version the root set the root's
400
# revision to the new commit even when no change occurs (more
401
# specifically, they do not record a revision on the root; and
402
# the rev id is assigned to the root during deserialisation -
403
# this masks when a change may have occurred against the basis.
404
# To match this we always issue a delta, because the revision
405
# of the root will always be changing.
406
if ie.file_id in basis_inv:
407
delta = (basis_inv.id2path(ie.file_id), path,
411
delta = (None, path, ie.file_id, ie)
412
self._basis_delta.append(delta)
413
return delta, False, None
415
# we don't need to commit this, because the caller already
416
# determined that an existing revision of this file is
417
# appropriate. If its not being considered for committing then
418
# it and all its parents to the root must be unaltered so
419
# no-change against the basis.
420
if ie.revision == self._new_revision_id:
421
raise AssertionError("Impossible situation, a skipped "
422
"inventory entry (%r) claims to be modified in this "
423
"commit (%r).", (ie, self._new_revision_id))
424
return None, False, None
425
# XXX: Friction: parent_candidates should return a list not a dict
426
# so that we don't have to walk the inventories again.
427
parent_candiate_entries = ie.parent_candidates(parent_invs)
428
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
430
for inv in parent_invs:
431
if ie.file_id in inv:
432
old_rev = inv[ie.file_id].revision
433
if old_rev in head_set:
434
heads.append(inv[ie.file_id].revision)
435
head_set.remove(inv[ie.file_id].revision)
438
# now we check to see if we need to write a new record to the
440
# We write a new entry unless there is one head to the ancestors, and
441
# the kind-derived content is unchanged.
443
# Cheapest check first: no ancestors, or more the one head in the
444
# ancestors, we write a new node.
448
# There is a single head, look it up for comparison
449
parent_entry = parent_candiate_entries[heads[0]]
450
# if the non-content specific data has changed, we'll be writing a
452
if (parent_entry.parent_id != ie.parent_id or
453
parent_entry.name != ie.name):
455
# now we need to do content specific checks:
457
# if the kind changed the content obviously has
458
if kind != parent_entry.kind:
460
# Stat cache fingerprint feedback for the caller - None as we usually
461
# don't generate one.
464
if content_summary[2] is None:
465
raise ValueError("Files must not have executable = None")
467
if (# if the file length changed we have to store:
468
parent_entry.text_size != content_summary[1] or
469
# if the exec bit has changed we have to store:
470
parent_entry.executable != content_summary[2]):
472
elif parent_entry.text_sha1 == content_summary[3]:
473
# all meta and content is unchanged (using a hash cache
474
# hit to check the sha)
475
ie.revision = parent_entry.revision
476
ie.text_size = parent_entry.text_size
477
ie.text_sha1 = parent_entry.text_sha1
478
ie.executable = parent_entry.executable
479
return self._get_delta(ie, basis_inv, path), False, None
481
# Either there is only a hash change(no hash cache entry,
482
# or same size content change), or there is no change on
484
# Provide the parent's hash to the store layer, so that the
485
# content is unchanged we will not store a new node.
486
nostore_sha = parent_entry.text_sha1
488
# We want to record a new node regardless of the presence or
489
# absence of a content change in the file.
491
ie.executable = content_summary[2]
492
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
494
lines = file_obj.readlines()
498
ie.text_sha1, ie.text_size = self._add_text_to_weave(
499
ie.file_id, lines, heads, nostore_sha)
500
# Let the caller know we generated a stat fingerprint.
501
fingerprint = (ie.text_sha1, stat_value)
502
except errors.ExistingContent:
503
# Turns out that the file content was unchanged, and we were
504
# only going to store a new node if it was changed. Carry over
506
ie.revision = parent_entry.revision
507
ie.text_size = parent_entry.text_size
508
ie.text_sha1 = parent_entry.text_sha1
509
ie.executable = parent_entry.executable
510
return self._get_delta(ie, basis_inv, path), False, None
511
elif kind == 'directory':
513
# all data is meta here, nothing specific to directory, so
515
ie.revision = parent_entry.revision
516
return self._get_delta(ie, basis_inv, path), False, None
518
self._add_text_to_weave(ie.file_id, lines, heads, None)
519
elif kind == 'symlink':
520
current_link_target = content_summary[3]
522
# symlink target is not generic metadata, check if it has
524
if current_link_target != parent_entry.symlink_target:
527
# unchanged, carry over.
528
ie.revision = parent_entry.revision
529
ie.symlink_target = parent_entry.symlink_target
530
return self._get_delta(ie, basis_inv, path), False, None
531
ie.symlink_target = current_link_target
533
self._add_text_to_weave(ie.file_id, lines, heads, None)
534
elif kind == 'tree-reference':
536
if content_summary[3] != parent_entry.reference_revision:
539
# unchanged, carry over.
540
ie.reference_revision = parent_entry.reference_revision
541
ie.revision = parent_entry.revision
542
return self._get_delta(ie, basis_inv, path), False, None
543
ie.reference_revision = content_summary[3]
545
self._add_text_to_weave(ie.file_id, lines, heads, None)
547
raise NotImplementedError('unknown kind')
548
ie.revision = self._new_revision_id
549
self._any_changes = True
550
return self._get_delta(ie, basis_inv, path), True, fingerprint
552
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
553
_entry_factory=entry_factory):
554
"""Record a new tree via iter_changes.
556
:param tree: The tree to obtain text contents from for changed objects.
557
:param basis_revision_id: The revision id of the tree the iter_changes
558
has been generated against. Currently assumed to be the same
559
as self.parents[0] - if it is not, errors may occur.
560
:param iter_changes: An iter_changes iterator with the changes to apply
561
to basis_revision_id.
562
:param _entry_factory: Private method to bind entry_factory locally for
566
# Create an inventory delta based on deltas between all the parents and
567
# deltas between all the parent inventories. We use inventory delta's
568
# between the inventory objects because iter_changes masks
569
# last-changed-field only changes.
571
# file_id -> change map, change is fileid, paths, changed, versioneds,
572
# parents, names, kinds, executables
574
# {file_id -> revision_id -> inventory entry, for entries in parent
575
# trees that are not parents[0]
577
revtrees = list(self.repository.revision_trees(self.parents))
578
# The basis inventory from a repository
580
basis_inv = revtrees[0].inventory
582
basis_inv = self.repository.revision_tree(
583
_mod_revision.NULL_REVISION).inventory
584
if len(self.parents) > 0:
585
if basis_revision_id != self.parents[0]:
587
"arbitrary basis parents not yet supported with merges")
588
for revtree in revtrees[1:]:
589
for change in revtree.inventory._make_delta(basis_inv):
590
if change[1] is None:
591
# Not present in this parent.
593
if change[2] not in merged_ids:
594
if change[0] is not None:
595
merged_ids[change[2]] = [
596
basis_inv[change[2]].revision,
599
merged_ids[change[2]] = [change[3].revision]
600
parent_entries[change[2]] = {change[3].revision:change[3]}
602
merged_ids[change[2]].append(change[3].revision)
603
parent_entries[change[2]][change[3].revision] = change[3]
606
# Setup the changes from the tree:
607
# changes maps file_id -> (change, [parent revision_ids])
609
for change in iter_changes:
610
# This probably looks up in basis_inv way to much.
611
if change[1][0] is not None:
612
head_candidate = [basis_inv[change[0]].revision]
615
changes[change[0]] = change, merged_ids.get(change[0],
617
unchanged_merged = set(merged_ids) - set(changes)
618
# Extend the changes dict with synthetic changes to record merges of
620
for file_id in unchanged_merged:
621
# Record a merged version of these items that did not change vs the
622
# basis. This can be either identical parallel changes, or a revert
623
# of a specific file after a merge. The recorded content will be
624
# that of the current tree (which is the same as the basis), but
625
# the per-file graph will reflect a merge.
626
# NB:XXX: We are reconstructing path information we had, this
627
# should be preserved instead.
628
# inv delta change: (file_id, (path_in_source, path_in_target),
629
# changed_content, versioned, parent, name, kind,
631
basis_entry = basis_inv[file_id]
633
(basis_inv.id2path(file_id), tree.id2path(file_id)),
635
(basis_entry.parent_id, basis_entry.parent_id),
636
(basis_entry.name, basis_entry.name),
637
(basis_entry.kind, basis_entry.kind),
638
(basis_entry.executable, basis_entry.executable))
639
changes[file_id] = (change, merged_ids[file_id])
640
# changes contains tuples with the change and a set of inventory
641
# candidates for the file.
643
# old_path, new_path, file_id, new_inventory_entry
644
seen_root = False # Is the root in the basis delta?
645
inv_delta = self._basis_delta
646
modified_rev = self._new_revision_id
647
for change, head_candidates in changes.values():
648
if change[3][1]: # versioned in target.
649
# Several things may be happening here:
650
# We may have a fork in the per-file graph
651
# - record a change with the content from tree
652
# We may have a change against < all trees
653
# - carry over the tree that hasn't changed
654
# We may have a change against all trees
655
# - record the change with the content from tree
658
entry = _entry_factory[kind](file_id, change[5][1],
660
head_set = self._heads(change[0], set(head_candidates))
663
for head_candidate in head_candidates:
664
if head_candidate in head_set:
665
heads.append(head_candidate)
666
head_set.remove(head_candidate)
669
# Could be a carry-over situation:
670
parent_entry_revs = parent_entries.get(file_id, None)
671
if parent_entry_revs:
672
parent_entry = parent_entry_revs.get(heads[0], None)
675
if parent_entry is None:
676
# The parent iter_changes was called against is the one
677
# that is the per-file head, so any change is relevant
678
# iter_changes is valid.
679
carry_over_possible = False
681
# could be a carry over situation
682
# A change against the basis may just indicate a merge,
683
# we need to check the content against the source of the
684
# merge to determine if it was changed after the merge
686
if (parent_entry.kind != entry.kind or
687
parent_entry.parent_id != entry.parent_id or
688
parent_entry.name != entry.name):
689
# Metadata common to all entries has changed
690
# against per-file parent
691
carry_over_possible = False
693
carry_over_possible = True
694
# per-type checks for changes against the parent_entry
697
# Cannot be a carry-over situation
698
carry_over_possible = False
699
# Populate the entry in the delta
701
# XXX: There is still a small race here: If someone reverts the content of a file
702
# after iter_changes examines and decides it has changed,
703
# we will unconditionally record a new version even if some
704
# other process reverts it while commit is running (with
705
# the revert happening after iter_changes did it's
708
entry.executable = True
710
entry.executable = False
711
if (carry_over_possible and
712
parent_entry.executable == entry.executable):
713
# Check the file length, content hash after reading
715
nostore_sha = parent_entry.text_sha1
718
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
720
lines = file_obj.readlines()
724
entry.text_sha1, entry.text_size = self._add_text_to_weave(
725
file_id, lines, heads, nostore_sha)
726
except errors.ExistingContent:
727
# No content change against a carry_over parent
729
entry.text_size = parent_entry.text_size
730
entry.text_sha1 = parent_entry.text_sha1
731
elif kind == 'symlink':
733
entry.symlink_target = tree.get_symlink_target(file_id)
734
if (carry_over_possible and
735
parent_entry.symlink_target == entry.symlink_target):
738
self._add_text_to_weave(change[0], [], heads, None)
739
elif kind == 'directory':
740
if carry_over_possible:
743
# Nothing to set on the entry.
744
# XXX: split into the Root and nonRoot versions.
745
if change[1][1] != '' or self.repository.supports_rich_root():
746
self._add_text_to_weave(change[0], [], heads, None)
747
elif kind == 'tree-reference':
748
raise AssertionError('unknown kind %r' % kind)
750
raise AssertionError('unknown kind %r' % kind)
752
entry.revision = modified_rev
754
entry.revision = parent_entry.revision
757
new_path = change[1][1]
758
inv_delta.append((change[1][0], new_path, change[0], entry))
761
self.new_inventory = None
763
self._any_changes = True
765
# housekeeping root entry changes do not affect no-change commits.
766
self._require_root_change(tree)
767
self.basis_delta_revision = basis_revision_id
769
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
770
# Note: as we read the content directly from the tree, we know its not
771
# been turned into unicode or badly split - but a broken tree
772
# implementation could give us bad output from readlines() so this is
773
# not a guarantee of safety. What would be better is always checking
774
# the content during test suite execution. RBC 20070912
775
parent_keys = tuple((file_id, parent) for parent in parents)
776
return self.repository.texts.add_lines(
777
(file_id, self._new_revision_id), parent_keys, new_lines,
778
nostore_sha=nostore_sha, random_id=self.random_revid,
779
check_content=False)[0:2]
782
class RootCommitBuilder(CommitBuilder):
783
"""This commitbuilder actually records the root id"""
785
# the root entry gets versioned properly by this builder.
786
_versioned_root = True
788
def _check_root(self, ie, parent_invs, tree):
789
"""Helper for record_entry_contents.
791
:param ie: An entry being added.
792
:param parent_invs: The inventories of the parent revisions of the
794
:param tree: The tree that is being committed.
797
def _require_root_change(self, tree):
798
"""Enforce an appropriate root object change.
800
This is called once when record_iter_changes is called, if and only if
801
the root was not in the delta calculated by record_iter_changes.
803
:param tree: The tree which is being committed.
805
# versioned roots do not change unless the tree found a change.
808
######################################################################
811
class Repository(object):
812
"""Repository holding history for one or more branches.
814
The repository holds and retrieves historical information including
815
revisions and file history. It's normally accessed only by the Branch,
816
which views a particular line of development through that history.
818
The Repository builds on top of some byte storage facilies (the revisions,
819
signatures, inventories and texts attributes) and a Transport, which
820
respectively provide byte storage and a means to access the (possibly
823
The byte storage facilities are addressed via tuples, which we refer to
824
as 'keys' throughout the code base. Revision_keys, inventory_keys and
825
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
826
(file_id, revision_id). We use this interface because it allows low
827
friction with the underlying code that implements disk indices, network
828
encoding and other parts of bzrlib.
830
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
831
the serialised revisions for the repository. This can be used to obtain
832
revision graph information or to access raw serialised revisions.
833
The result of trying to insert data into the repository via this store
834
is undefined: it should be considered read-only except for implementors
836
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
837
the serialised signatures for the repository. This can be used to
838
obtain access to raw serialised signatures. The result of trying to
839
insert data into the repository via this store is undefined: it should
840
be considered read-only except for implementors of repositories.
841
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
842
the serialised inventories for the repository. This can be used to
843
obtain unserialised inventories. The result of trying to insert data
844
into the repository via this store is undefined: it should be
845
considered read-only except for implementors of repositories.
846
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
847
texts of files and directories for the repository. This can be used to
848
obtain file texts or file graphs. Note that Repository.iter_file_bytes
849
is usually a better interface for accessing file texts.
850
The result of trying to insert data into the repository via this store
851
is undefined: it should be considered read-only except for implementors
853
:ivar _transport: Transport for file access to repository, typically
854
pointing to .bzr/repository.
857
# What class to use for a CommitBuilder. Often its simpler to change this
858
# in a Repository class subclass rather than to override
859
# get_commit_builder.
860
_commit_builder_class = CommitBuilder
861
# The search regex used by xml based repositories to determine what things
862
# where changed in a single commit.
863
_file_ids_altered_regex = lazy_regex.lazy_compile(
864
r'file_id="(?P<file_id>[^"]+)"'
865
r'.* revision="(?P<revision_id>[^"]+)"'
868
def abort_write_group(self, suppress_errors=False):
869
"""Commit the contents accrued within the current write group.
871
:param suppress_errors: if true, abort_write_group will catch and log
872
unexpected errors that happen during the abort, rather than
873
allowing them to propagate. Defaults to False.
875
:seealso: start_write_group.
877
if self._write_group is not self.get_transaction():
878
# has an unlock or relock occured ?
879
raise errors.BzrError('mismatched lock context and write group.')
881
self._abort_write_group()
882
except Exception, exc:
883
self._write_group = None
884
if not suppress_errors:
886
mutter('abort_write_group failed')
887
log_exception_quietly()
888
note('bzr: ERROR (ignored): %s', exc)
889
self._write_group = None
891
def _abort_write_group(self):
892
"""Template method for per-repository write group cleanup.
894
This is called during abort before the write group is considered to be
895
finished and should cleanup any internal state accrued during the write
896
group. There is no requirement that data handed to the repository be
897
*not* made available - this is not a rollback - but neither should any
898
attempt be made to ensure that data added is fully commited. Abort is
899
invoked when an error has occured so futher disk or network operations
900
may not be possible or may error and if possible should not be
904
def add_fallback_repository(self, repository):
905
"""Add a repository to use for looking up data not held locally.
907
:param repository: A repository.
909
if not self._format.supports_external_lookups:
910
raise errors.UnstackableRepositoryFormat(self._format, self.base)
911
self._check_fallback_repository(repository)
912
self._fallback_repositories.append(repository)
913
self.texts.add_fallback_versioned_files(repository.texts)
914
self.inventories.add_fallback_versioned_files(repository.inventories)
915
self.revisions.add_fallback_versioned_files(repository.revisions)
916
self.signatures.add_fallback_versioned_files(repository.signatures)
918
def _check_fallback_repository(self, repository):
919
"""Check that this repository can fallback to repository safely.
921
Raise an error if not.
923
:param repository: A repository to fallback to.
925
return InterRepository._assert_same_model(self, repository)
927
def add_inventory(self, revision_id, inv, parents):
928
"""Add the inventory inv to the repository as revision_id.
930
:param parents: The revision ids of the parents that revision_id
931
is known to have and are in the repository already.
933
:returns: The validator(which is a sha1 digest, though what is sha'd is
934
repository format specific) of the serialized inventory.
936
if not self.is_in_write_group():
937
raise AssertionError("%r not in write group" % (self,))
938
_mod_revision.check_not_reserved_id(revision_id)
939
if not (inv.revision_id is None or inv.revision_id == revision_id):
940
raise AssertionError(
941
"Mismatch between inventory revision"
942
" id and insertion revid (%r, %r)"
943
% (inv.revision_id, revision_id))
945
raise AssertionError()
946
inv_lines = self._serialise_inventory_to_lines(inv)
947
return self._inventory_add_lines(revision_id, parents,
948
inv_lines, check_content=False)
950
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
952
"""Add a new inventory expressed as a delta against another revision.
954
:param basis_revision_id: The inventory id the delta was created
955
against. (This does not have to be a direct parent.)
956
:param delta: The inventory delta (see Inventory.apply_delta for
958
:param new_revision_id: The revision id that the inventory is being
960
:param parents: The revision ids of the parents that revision_id is
961
known to have and are in the repository already. These are supplied
962
for repositories that depend on the inventory graph for revision
963
graph access, as well as for those that pun ancestry with delta
966
:returns: (validator, new_inv)
967
The validator(which is a sha1 digest, though what is sha'd is
968
repository format specific) of the serialized inventory, and the
971
if not self.is_in_write_group():
972
raise AssertionError("%r not in write group" % (self,))
973
_mod_revision.check_not_reserved_id(new_revision_id)
974
basis_tree = self.revision_tree(basis_revision_id)
975
basis_tree.lock_read()
977
# Note that this mutates the inventory of basis_tree, which not all
978
# inventory implementations may support: A better idiom would be to
979
# return a new inventory, but as there is no revision tree cache in
980
# repository this is safe for now - RBC 20081013
981
basis_inv = basis_tree.inventory
982
basis_inv.apply_delta(delta)
983
basis_inv.revision_id = new_revision_id
984
return (self.add_inventory(new_revision_id, basis_inv, parents),
989
def _inventory_add_lines(self, revision_id, parents, lines,
991
"""Store lines in inv_vf and return the sha1 of the inventory."""
992
parents = [(parent,) for parent in parents]
993
return self.inventories.add_lines((revision_id,), parents, lines,
994
check_content=check_content)[0]
996
def add_revision(self, revision_id, rev, inv=None, config=None):
997
"""Add rev to the revision store as revision_id.
999
:param revision_id: the revision id to use.
1000
:param rev: The revision object.
1001
:param inv: The inventory for the revision. if None, it will be looked
1002
up in the inventory storer
1003
:param config: If None no digital signature will be created.
1004
If supplied its signature_needed method will be used
1005
to determine if a signature should be made.
1007
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1009
_mod_revision.check_not_reserved_id(revision_id)
1010
if config is not None and config.signature_needed():
1012
inv = self.get_inventory(revision_id)
1013
plaintext = Testament(rev, inv).as_short_text()
1014
self.store_revision_signature(
1015
gpg.GPGStrategy(config), plaintext, revision_id)
1016
# check inventory present
1017
if not self.inventories.get_parent_map([(revision_id,)]):
1019
raise errors.WeaveRevisionNotPresent(revision_id,
1022
# yes, this is not suitable for adding with ghosts.
1023
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1026
key = (revision_id,)
1027
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1028
self._add_revision(rev)
1030
def _add_revision(self, revision):
1031
text = self._serializer.write_revision_to_string(revision)
1032
key = (revision.revision_id,)
1033
parents = tuple((parent,) for parent in revision.parent_ids)
1034
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1036
def all_revision_ids(self):
1037
"""Returns a list of all the revision ids in the repository.
1039
This is conceptually deprecated because code should generally work on
1040
the graph reachable from a particular revision, and ignore any other
1041
revisions that might be present. There is no direct replacement
1044
if 'evil' in debug.debug_flags:
1045
mutter_callsite(2, "all_revision_ids is linear with history.")
1046
return self._all_revision_ids()
1048
def _all_revision_ids(self):
1049
"""Returns a list of all the revision ids in the repository.
1051
These are in as much topological order as the underlying store can
1054
raise NotImplementedError(self._all_revision_ids)
1056
def break_lock(self):
1057
"""Break a lock if one is present from another instance.
1059
Uses the ui factory to ask for confirmation if the lock may be from
1062
self.control_files.break_lock()
1065
def _eliminate_revisions_not_present(self, revision_ids):
1066
"""Check every revision id in revision_ids to see if we have it.
1068
Returns a set of the present revisions.
1071
graph = self.get_graph()
1072
parent_map = graph.get_parent_map(revision_ids)
1073
# The old API returned a list, should this actually be a set?
1074
return parent_map.keys()
1077
def create(a_bzrdir):
1078
"""Construct the current default format repository in a_bzrdir."""
1079
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1081
def __init__(self, _format, a_bzrdir, control_files):
1082
"""instantiate a Repository.
1084
:param _format: The format of the repository on disk.
1085
:param a_bzrdir: The BzrDir of the repository.
1087
In the future we will have a single api for all stores for
1088
getting file texts, inventories and revisions, then
1089
this construct will accept instances of those things.
1091
super(Repository, self).__init__()
1092
self._format = _format
1093
# the following are part of the public API for Repository:
1094
self.bzrdir = a_bzrdir
1095
self.control_files = control_files
1096
self._transport = control_files._transport
1097
self.base = self._transport.base
1099
self._reconcile_does_inventory_gc = True
1100
self._reconcile_fixes_text_parents = False
1101
self._reconcile_backsup_inventory = True
1102
# not right yet - should be more semantically clear ?
1104
# TODO: make sure to construct the right store classes, etc, depending
1105
# on whether escaping is required.
1106
self._warn_if_deprecated()
1107
self._write_group = None
1108
# Additional places to query for data.
1109
self._fallback_repositories = []
1110
# An InventoryEntry cache, used during deserialization
1111
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1114
return '%s(%r)' % (self.__class__.__name__,
1117
def has_same_location(self, other):
1118
"""Returns a boolean indicating if this repository is at the same
1119
location as another repository.
1121
This might return False even when two repository objects are accessing
1122
the same physical repository via different URLs.
1124
if self.__class__ is not other.__class__:
1126
return (self._transport.base == other._transport.base)
1128
def is_in_write_group(self):
1129
"""Return True if there is an open write group.
1131
:seealso: start_write_group.
1133
return self._write_group is not None
1135
def is_locked(self):
1136
return self.control_files.is_locked()
1138
def is_write_locked(self):
1139
"""Return True if this object is write locked."""
1140
return self.is_locked() and self.control_files._lock_mode == 'w'
1142
def lock_write(self, token=None):
1143
"""Lock this repository for writing.
1145
This causes caching within the repository obejct to start accumlating
1146
data during reads, and allows a 'write_group' to be obtained. Write
1147
groups must be used for actual data insertion.
1149
:param token: if this is already locked, then lock_write will fail
1150
unless the token matches the existing lock.
1151
:returns: a token if this instance supports tokens, otherwise None.
1152
:raises TokenLockingNotSupported: when a token is given but this
1153
instance doesn't support using token locks.
1154
:raises MismatchedToken: if the specified token doesn't match the token
1155
of the existing lock.
1156
:seealso: start_write_group.
1158
A token should be passed in if you know that you have locked the object
1159
some other way, and need to synchronise this object's state with that
1162
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1164
locked = self.is_locked()
1165
result = self.control_files.lock_write(token=token)
1166
for repo in self._fallback_repositories:
1167
# Writes don't affect fallback repos
1170
self._refresh_data()
1173
def lock_read(self):
1174
locked = self.is_locked()
1175
self.control_files.lock_read()
1176
for repo in self._fallback_repositories:
1179
self._refresh_data()
1181
def get_physical_lock_status(self):
1182
return self.control_files.get_physical_lock_status()
1184
def leave_lock_in_place(self):
1185
"""Tell this repository not to release the physical lock when this
1188
If lock_write doesn't return a token, then this method is not supported.
1190
self.control_files.leave_in_place()
1192
def dont_leave_lock_in_place(self):
1193
"""Tell this repository to release the physical lock when this
1194
object is unlocked, even if it didn't originally acquire it.
1196
If lock_write doesn't return a token, then this method is not supported.
1198
self.control_files.dont_leave_in_place()
1201
def gather_stats(self, revid=None, committers=None):
1202
"""Gather statistics from a revision id.
1204
:param revid: The revision id to gather statistics from, if None, then
1205
no revision specific statistics are gathered.
1206
:param committers: Optional parameter controlling whether to grab
1207
a count of committers from the revision specific statistics.
1208
:return: A dictionary of statistics. Currently this contains:
1209
committers: The number of committers if requested.
1210
firstrev: A tuple with timestamp, timezone for the penultimate left
1211
most ancestor of revid, if revid is not the NULL_REVISION.
1212
latestrev: A tuple with timestamp, timezone for revid, if revid is
1213
not the NULL_REVISION.
1214
revisions: The total revision count in the repository.
1215
size: An estimate disk size of the repository in bytes.
1218
if revid and committers:
1219
result['committers'] = 0
1220
if revid and revid != _mod_revision.NULL_REVISION:
1222
all_committers = set()
1223
revisions = self.get_ancestry(revid)
1224
# pop the leading None
1226
first_revision = None
1228
# ignore the revisions in the middle - just grab first and last
1229
revisions = revisions[0], revisions[-1]
1230
for revision in self.get_revisions(revisions):
1231
if not first_revision:
1232
first_revision = revision
1234
all_committers.add(revision.committer)
1235
last_revision = revision
1237
result['committers'] = len(all_committers)
1238
result['firstrev'] = (first_revision.timestamp,
1239
first_revision.timezone)
1240
result['latestrev'] = (last_revision.timestamp,
1241
last_revision.timezone)
1243
# now gather global repository information
1244
# XXX: This is available for many repos regardless of listability.
1245
if self.bzrdir.root_transport.listable():
1246
# XXX: do we want to __define len__() ?
1247
# Maybe the versionedfiles object should provide a different
1248
# method to get the number of keys.
1249
result['revisions'] = len(self.revisions.keys())
1250
# result['size'] = t
1253
def find_branches(self, using=False):
1254
"""Find branches underneath this repository.
1256
This will include branches inside other branches.
1258
:param using: If True, list only branches using this repository.
1260
if using and not self.is_shared():
1262
return [self.bzrdir.open_branch()]
1263
except errors.NotBranchError:
1265
class Evaluator(object):
1268
self.first_call = True
1270
def __call__(self, bzrdir):
1271
# On the first call, the parameter is always the bzrdir
1272
# containing the current repo.
1273
if not self.first_call:
1275
repository = bzrdir.open_repository()
1276
except errors.NoRepositoryPresent:
1279
return False, (None, repository)
1280
self.first_call = False
1282
value = (bzrdir.open_branch(), None)
1283
except errors.NotBranchError:
1284
value = (None, None)
1288
for branch, repository in bzrdir.BzrDir.find_bzrdirs(
1289
self.bzrdir.root_transport, evaluate=Evaluator()):
1290
if branch is not None:
1291
branches.append(branch)
1292
if not using and repository is not None:
1293
branches.extend(repository.find_branches())
1297
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
1298
"""Return the revision ids that other has that this does not.
1300
These are returned in topological order.
1302
revision_id: only return revision ids included by revision_id.
1304
return InterRepository.get(other, self).search_missing_revision_ids(
1305
revision_id, find_ghosts)
1309
"""Open the repository rooted at base.
1311
For instance, if the repository is at URL/.bzr/repository,
1312
Repository.open(URL) -> a Repository instance.
1314
control = bzrdir.BzrDir.open(base)
1315
return control.open_repository()
1317
def copy_content_into(self, destination, revision_id=None):
1318
"""Make a complete copy of the content in self into destination.
1320
This is a destructive operation! Do not use it on existing
1323
return InterRepository.get(self, destination).copy_content(revision_id)
1325
def commit_write_group(self):
1326
"""Commit the contents accrued within the current write group.
1328
:seealso: start_write_group.
1330
if self._write_group is not self.get_transaction():
1331
# has an unlock or relock occured ?
1332
raise errors.BzrError('mismatched lock context %r and '
1334
(self.get_transaction(), self._write_group))
1335
self._commit_write_group()
1336
self._write_group = None
1338
def _commit_write_group(self):
1339
"""Template method for per-repository write group cleanup.
1341
This is called before the write group is considered to be
1342
finished and should ensure that all data handed to the repository
1343
for writing during the write group is safely committed (to the
1344
extent possible considering file system caching etc).
1347
def suspend_write_group(self):
1348
raise errors.UnsuspendableWriteGroup(self)
1350
def refresh_data(self):
1351
"""Re-read any data needed to to synchronise with disk.
1353
This method is intended to be called after another repository instance
1354
(such as one used by a smart server) has inserted data into the
1355
repository. It may not be called during a write group, but may be
1356
called at any other time.
1358
if self.is_in_write_group():
1359
raise errors.InternalBzrError(
1360
"May not refresh_data while in a write group.")
1361
self._refresh_data()
1363
def resume_write_group(self, tokens):
1364
if not self.is_write_locked():
1365
raise errors.NotWriteLocked(self)
1366
if self._write_group:
1367
raise errors.BzrError('already in a write group')
1368
self._resume_write_group(tokens)
1369
# so we can detect unlock/relock - the write group is now entered.
1370
self._write_group = self.get_transaction()
1372
def _resume_write_group(self, tokens):
1373
raise errors.UnsuspendableWriteGroup(self)
1375
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1377
"""Fetch the content required to construct revision_id from source.
1379
If revision_id is None and fetch_spec is None, then all content is
1382
fetch() may not be used when the repository is in a write group -
1383
either finish the current write group before using fetch, or use
1384
fetch before starting the write group.
1386
:param find_ghosts: Find and copy revisions in the source that are
1387
ghosts in the target (and not reachable directly by walking out to
1388
the first-present revision in target from revision_id).
1389
:param revision_id: If specified, all the content needed for this
1390
revision ID will be copied to the target. Fetch will determine for
1391
itself which content needs to be copied.
1392
:param fetch_spec: If specified, a SearchResult or
1393
PendingAncestryResult that describes which revisions to copy. This
1394
allows copying multiple heads at once. Mutually exclusive with
1397
if fetch_spec is not None and revision_id is not None:
1398
raise AssertionError(
1399
"fetch_spec and revision_id are mutually exclusive.")
1400
if self.is_in_write_group():
1401
raise errors.InternalBzrError(
1402
"May not fetch while in a write group.")
1403
# fast path same-url fetch operations
1404
if self.has_same_location(source) and fetch_spec is None:
1405
# check that last_revision is in 'from' and then return a
1407
if (revision_id is not None and
1408
not _mod_revision.is_null(revision_id)):
1409
self.get_revision(revision_id)
1411
# if there is no specific appropriate InterRepository, this will get
1412
# the InterRepository base class, which raises an
1413
# IncompatibleRepositories when asked to fetch.
1414
inter = InterRepository.get(source, self)
1415
return inter.fetch(revision_id=revision_id, pb=pb,
1416
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1418
def create_bundle(self, target, base, fileobj, format=None):
1419
return serializer.write_bundle(self, target, base, fileobj, format)
1421
def get_commit_builder(self, branch, parents, config, timestamp=None,
1422
timezone=None, committer=None, revprops=None,
1424
"""Obtain a CommitBuilder for this repository.
1426
:param branch: Branch to commit to.
1427
:param parents: Revision ids of the parents of the new revision.
1428
:param config: Configuration to use.
1429
:param timestamp: Optional timestamp recorded for commit.
1430
:param timezone: Optional timezone for timestamp.
1431
:param committer: Optional committer to set for commit.
1432
:param revprops: Optional dictionary of revision properties.
1433
:param revision_id: Optional revision id.
1435
result = self._commit_builder_class(self, parents, config,
1436
timestamp, timezone, committer, revprops, revision_id)
1437
self.start_write_group()
1441
if (self.control_files._lock_count == 1 and
1442
self.control_files._lock_mode == 'w'):
1443
if self._write_group is not None:
1444
self.abort_write_group()
1445
self.control_files.unlock()
1446
raise errors.BzrError(
1447
'Must end write groups before releasing write locks.')
1448
self.control_files.unlock()
1449
if self.control_files._lock_count == 0:
1450
self._inventory_entry_cache.clear()
1451
for repo in self._fallback_repositories:
1455
def clone(self, a_bzrdir, revision_id=None):
1456
"""Clone this repository into a_bzrdir using the current format.
1458
Currently no check is made that the format of this repository and
1459
the bzrdir format are compatible. FIXME RBC 20060201.
1461
:return: The newly created destination repository.
1463
# TODO: deprecate after 0.16; cloning this with all its settings is
1464
# probably not very useful -- mbp 20070423
1465
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1466
self.copy_content_into(dest_repo, revision_id)
1469
def start_write_group(self):
1470
"""Start a write group in the repository.
1472
Write groups are used by repositories which do not have a 1:1 mapping
1473
between file ids and backend store to manage the insertion of data from
1474
both fetch and commit operations.
1476
A write lock is required around the start_write_group/commit_write_group
1477
for the support of lock-requiring repository formats.
1479
One can only insert data into a repository inside a write group.
1483
if not self.is_write_locked():
1484
raise errors.NotWriteLocked(self)
1485
if self._write_group:
1486
raise errors.BzrError('already in a write group')
1487
self._start_write_group()
1488
# so we can detect unlock/relock - the write group is now entered.
1489
self._write_group = self.get_transaction()
1491
def _start_write_group(self):
1492
"""Template method for per-repository write group startup.
1494
This is called before the write group is considered to be
1499
def sprout(self, to_bzrdir, revision_id=None):
1500
"""Create a descendent repository for new development.
1502
Unlike clone, this does not copy the settings of the repository.
1504
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1505
dest_repo.fetch(self, revision_id=revision_id)
1508
def _create_sprouting_repo(self, a_bzrdir, shared):
1509
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1510
# use target default format.
1511
dest_repo = a_bzrdir.create_repository()
1513
# Most control formats need the repository to be specifically
1514
# created, but on some old all-in-one formats it's not needed
1516
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1517
except errors.UninitializableFormat:
1518
dest_repo = a_bzrdir.open_repository()
1521
def _get_sink(self):
1522
"""Return a sink for streaming into this repository."""
1523
return StreamSink(self)
1525
def _get_source(self, to_format):
1526
"""Return a source for streaming from this repository."""
1527
return StreamSource(self, to_format)
1530
def has_revision(self, revision_id):
1531
"""True if this repository has a copy of the revision."""
1532
return revision_id in self.has_revisions((revision_id,))
1535
def has_revisions(self, revision_ids):
1536
"""Probe to find out the presence of multiple revisions.
1538
:param revision_ids: An iterable of revision_ids.
1539
:return: A set of the revision_ids that were present.
1541
parent_map = self.revisions.get_parent_map(
1542
[(rev_id,) for rev_id in revision_ids])
1544
if _mod_revision.NULL_REVISION in revision_ids:
1545
result.add(_mod_revision.NULL_REVISION)
1546
result.update([key[0] for key in parent_map])
1550
def get_revision(self, revision_id):
1551
"""Return the Revision object for a named revision."""
1552
return self.get_revisions([revision_id])[0]
1555
def get_revision_reconcile(self, revision_id):
1556
"""'reconcile' helper routine that allows access to a revision always.
1558
This variant of get_revision does not cross check the weave graph
1559
against the revision one as get_revision does: but it should only
1560
be used by reconcile, or reconcile-alike commands that are correcting
1561
or testing the revision graph.
1563
return self._get_revisions([revision_id])[0]
1566
def get_revisions(self, revision_ids):
1567
"""Get many revisions at once."""
1568
return self._get_revisions(revision_ids)
1571
def _get_revisions(self, revision_ids):
1572
"""Core work logic to get many revisions without sanity checks."""
1573
for rev_id in revision_ids:
1574
if not rev_id or not isinstance(rev_id, basestring):
1575
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1576
keys = [(key,) for key in revision_ids]
1577
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1579
for record in stream:
1580
if record.storage_kind == 'absent':
1581
raise errors.NoSuchRevision(self, record.key[0])
1582
text = record.get_bytes_as('fulltext')
1583
rev = self._serializer.read_revision_from_string(text)
1584
revs[record.key[0]] = rev
1585
return [revs[revid] for revid in revision_ids]
1588
def get_revision_xml(self, revision_id):
1589
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1590
# would have already do it.
1591
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1592
rev = self.get_revision(revision_id)
1593
rev_tmp = cStringIO.StringIO()
1594
# the current serializer..
1595
self._serializer.write_revision(rev, rev_tmp)
1597
return rev_tmp.getvalue()
1599
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1600
"""Produce a generator of revision deltas.
1602
Note that the input is a sequence of REVISIONS, not revision_ids.
1603
Trees will be held in memory until the generator exits.
1604
Each delta is relative to the revision's lefthand predecessor.
1606
:param specific_fileids: if not None, the result is filtered
1607
so that only those file-ids, their parents and their
1608
children are included.
1610
# Get the revision-ids of interest
1611
required_trees = set()
1612
for revision in revisions:
1613
required_trees.add(revision.revision_id)
1614
required_trees.update(revision.parent_ids[:1])
1616
# Get the matching filtered trees. Note that it's more
1617
# efficient to pass filtered trees to changes_from() rather
1618
# than doing the filtering afterwards. changes_from() could
1619
# arguably do the filtering itself but it's path-based, not
1620
# file-id based, so filtering before or afterwards is
1622
if specific_fileids is None:
1623
trees = dict((t.get_revision_id(), t) for
1624
t in self.revision_trees(required_trees))
1626
trees = dict((t.get_revision_id(), t) for
1627
t in self._filtered_revision_trees(required_trees,
1630
# Calculate the deltas
1631
for revision in revisions:
1632
if not revision.parent_ids:
1633
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1635
old_tree = trees[revision.parent_ids[0]]
1636
yield trees[revision.revision_id].changes_from(old_tree)
1639
def get_revision_delta(self, revision_id, specific_fileids=None):
1640
"""Return the delta for one revision.
1642
The delta is relative to the left-hand predecessor of the
1645
:param specific_fileids: if not None, the result is filtered
1646
so that only those file-ids, their parents and their
1647
children are included.
1649
r = self.get_revision(revision_id)
1650
return list(self.get_deltas_for_revisions([r],
1651
specific_fileids=specific_fileids))[0]
1654
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1655
signature = gpg_strategy.sign(plaintext)
1656
self.add_signature_text(revision_id, signature)
1659
def add_signature_text(self, revision_id, signature):
1660
self.signatures.add_lines((revision_id,), (),
1661
osutils.split_lines(signature))
1663
def find_text_key_references(self):
1664
"""Find the text key references within the repository.
1666
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1667
to whether they were referred to by the inventory of the
1668
revision_id that they contain. The inventory texts from all present
1669
revision ids are assessed to generate this report.
1671
revision_keys = self.revisions.keys()
1672
w = self.inventories
1673
pb = ui.ui_factory.nested_progress_bar()
1675
return self._find_text_key_references_from_xml_inventory_lines(
1676
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1680
def _find_text_key_references_from_xml_inventory_lines(self,
1682
"""Core routine for extracting references to texts from inventories.
1684
This performs the translation of xml lines to revision ids.
1686
:param line_iterator: An iterator of lines, origin_version_id
1687
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1688
to whether they were referred to by the inventory of the
1689
revision_id that they contain. Note that if that revision_id was
1690
not part of the line_iterator's output then False will be given -
1691
even though it may actually refer to that key.
1693
if not self._serializer.support_altered_by_hack:
1694
raise AssertionError(
1695
"_find_text_key_references_from_xml_inventory_lines only "
1696
"supported for branches which store inventory as unnested xml"
1697
", not on %r" % self)
1700
# this code needs to read every new line in every inventory for the
1701
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
1702
# not present in one of those inventories is unnecessary but not
1703
# harmful because we are filtering by the revision id marker in the
1704
# inventory lines : we only select file ids altered in one of those
1705
# revisions. We don't need to see all lines in the inventory because
1706
# only those added in an inventory in rev X can contain a revision=X
1708
unescape_revid_cache = {}
1709
unescape_fileid_cache = {}
1711
# jam 20061218 In a big fetch, this handles hundreds of thousands
1712
# of lines, so it has had a lot of inlining and optimizing done.
1713
# Sorry that it is a little bit messy.
1714
# Move several functions to be local variables, since this is a long
1716
search = self._file_ids_altered_regex.search
1717
unescape = _unescape_xml
1718
setdefault = result.setdefault
1719
for line, line_key in line_iterator:
1720
match = search(line)
1723
# One call to match.group() returning multiple items is quite a
1724
# bit faster than 2 calls to match.group() each returning 1
1725
file_id, revision_id = match.group('file_id', 'revision_id')
1727
# Inlining the cache lookups helps a lot when you make 170,000
1728
# lines and 350k ids, versus 8.4 unique ids.
1729
# Using a cache helps in 2 ways:
1730
# 1) Avoids unnecessary decoding calls
1731
# 2) Re-uses cached strings, which helps in future set and
1733
# (2) is enough that removing encoding entirely along with
1734
# the cache (so we are using plain strings) results in no
1735
# performance improvement.
1737
revision_id = unescape_revid_cache[revision_id]
1739
unescaped = unescape(revision_id)
1740
unescape_revid_cache[revision_id] = unescaped
1741
revision_id = unescaped
1743
# Note that unconditionally unescaping means that we deserialise
1744
# every fileid, which for general 'pull' is not great, but we don't
1745
# really want to have some many fulltexts that this matters anyway.
1748
file_id = unescape_fileid_cache[file_id]
1750
unescaped = unescape(file_id)
1751
unescape_fileid_cache[file_id] = unescaped
1754
key = (file_id, revision_id)
1755
setdefault(key, False)
1756
if revision_id == line_key[-1]:
1760
def _inventory_xml_lines_for_keys(self, keys):
1761
"""Get a line iterator of the sort needed for findind references.
1763
Not relevant for non-xml inventory repositories.
1765
Ghosts in revision_keys are ignored.
1767
:param revision_keys: The revision keys for the inventories to inspect.
1768
:return: An iterator over (inventory line, revid) for the fulltexts of
1769
all of the xml inventories specified by revision_keys.
1771
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1772
for record in stream:
1773
if record.storage_kind != 'absent':
1774
chunks = record.get_bytes_as('chunked')
1775
revid = record.key[-1]
1776
lines = osutils.chunks_to_lines(chunks)
1780
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1782
"""Helper routine for fileids_altered_by_revision_ids.
1784
This performs the translation of xml lines to revision ids.
1786
:param line_iterator: An iterator of lines, origin_version_id
1787
:param revision_ids: The revision ids to filter for. This should be a
1788
set or other type which supports efficient __contains__ lookups, as
1789
the revision id from each parsed line will be looked up in the
1790
revision_ids filter.
1791
:return: a dictionary mapping altered file-ids to an iterable of
1792
revision_ids. Each altered file-ids has the exact revision_ids that
1793
altered it listed explicitly.
1795
seen = set(self._find_text_key_references_from_xml_inventory_lines(
1796
line_iterator).iterkeys())
1797
# Note that revision_ids are revision keys.
1798
parent_maps = self.revisions.get_parent_map(revision_ids)
1800
map(parents.update, parent_maps.itervalues())
1801
parents.difference_update(revision_ids)
1802
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
1803
self._inventory_xml_lines_for_keys(parents)))
1804
new_keys = seen - parent_seen
1806
setdefault = result.setdefault
1807
for key in new_keys:
1808
setdefault(key[0], set()).add(key[-1])
1811
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
1812
"""Find the file ids and versions affected by revisions.
1814
:param revisions: an iterable containing revision ids.
1815
:param _inv_weave: The inventory weave from this repository or None.
1816
If None, the inventory weave will be opened automatically.
1817
:return: a dictionary mapping altered file-ids to an iterable of
1818
revision_ids. Each altered file-ids has the exact revision_ids that
1819
altered it listed explicitly.
1821
selected_keys = set((revid,) for revid in revision_ids)
1822
w = _inv_weave or self.inventories
1823
pb = ui.ui_factory.nested_progress_bar()
1825
return self._find_file_ids_from_xml_inventory_lines(
1826
w.iter_lines_added_or_present_in_keys(
1827
selected_keys, pb=pb),
1832
def iter_files_bytes(self, desired_files):
1833
"""Iterate through file versions.
1835
Files will not necessarily be returned in the order they occur in
1836
desired_files. No specific order is guaranteed.
1838
Yields pairs of identifier, bytes_iterator. identifier is an opaque
1839
value supplied by the caller as part of desired_files. It should
1840
uniquely identify the file version in the caller's context. (Examples:
1841
an index number or a TreeTransform trans_id.)
1843
bytes_iterator is an iterable of bytestrings for the file. The
1844
kind of iterable and length of the bytestrings are unspecified, but for
1845
this implementation, it is a list of bytes produced by
1846
VersionedFile.get_record_stream().
1848
:param desired_files: a list of (file_id, revision_id, identifier)
1852
for file_id, revision_id, callable_data in desired_files:
1853
text_keys[(file_id, revision_id)] = callable_data
1854
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
1855
if record.storage_kind == 'absent':
1856
raise errors.RevisionNotPresent(record.key, self)
1857
yield text_keys[record.key], record.get_bytes_as('chunked')
1859
def _generate_text_key_index(self, text_key_references=None,
1861
"""Generate a new text key index for the repository.
1863
This is an expensive function that will take considerable time to run.
1865
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
1866
list of parents, also text keys. When a given key has no parents,
1867
the parents list will be [NULL_REVISION].
1869
# All revisions, to find inventory parents.
1870
if ancestors is None:
1871
graph = self.get_graph()
1872
ancestors = graph.get_parent_map(self.all_revision_ids())
1873
if text_key_references is None:
1874
text_key_references = self.find_text_key_references()
1875
pb = ui.ui_factory.nested_progress_bar()
1877
return self._do_generate_text_key_index(ancestors,
1878
text_key_references, pb)
1882
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
1883
"""Helper for _generate_text_key_index to avoid deep nesting."""
1884
revision_order = tsort.topo_sort(ancestors)
1885
invalid_keys = set()
1887
for revision_id in revision_order:
1888
revision_keys[revision_id] = set()
1889
text_count = len(text_key_references)
1890
# a cache of the text keys to allow reuse; costs a dict of all the
1891
# keys, but saves a 2-tuple for every child of a given key.
1893
for text_key, valid in text_key_references.iteritems():
1895
invalid_keys.add(text_key)
1897
revision_keys[text_key[1]].add(text_key)
1898
text_key_cache[text_key] = text_key
1899
del text_key_references
1901
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
1902
NULL_REVISION = _mod_revision.NULL_REVISION
1903
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
1904
# too small for large or very branchy trees. However, for 55K path
1905
# trees, it would be easy to use too much memory trivially. Ideally we
1906
# could gauge this by looking at available real memory etc, but this is
1907
# always a tricky proposition.
1908
inventory_cache = lru_cache.LRUCache(10)
1909
batch_size = 10 # should be ~150MB on a 55K path tree
1910
batch_count = len(revision_order) / batch_size + 1
1912
pb.update("Calculating text parents", processed_texts, text_count)
1913
for offset in xrange(batch_count):
1914
to_query = revision_order[offset * batch_size:(offset + 1) *
1918
for rev_tree in self.revision_trees(to_query):
1919
revision_id = rev_tree.get_revision_id()
1920
parent_ids = ancestors[revision_id]
1921
for text_key in revision_keys[revision_id]:
1922
pb.update("Calculating text parents", processed_texts)
1923
processed_texts += 1
1924
candidate_parents = []
1925
for parent_id in parent_ids:
1926
parent_text_key = (text_key[0], parent_id)
1928
check_parent = parent_text_key not in \
1929
revision_keys[parent_id]
1931
# the parent parent_id is a ghost:
1932
check_parent = False
1933
# truncate the derived graph against this ghost.
1934
parent_text_key = None
1936
# look at the parent commit details inventories to
1937
# determine possible candidates in the per file graph.
1940
inv = inventory_cache[parent_id]
1942
inv = self.revision_tree(parent_id).inventory
1943
inventory_cache[parent_id] = inv
1944
parent_entry = inv._byid.get(text_key[0], None)
1945
if parent_entry is not None:
1947
text_key[0], parent_entry.revision)
1949
parent_text_key = None
1950
if parent_text_key is not None:
1951
candidate_parents.append(
1952
text_key_cache[parent_text_key])
1953
parent_heads = text_graph.heads(candidate_parents)
1954
new_parents = list(parent_heads)
1955
new_parents.sort(key=lambda x:candidate_parents.index(x))
1956
if new_parents == []:
1957
new_parents = [NULL_REVISION]
1958
text_index[text_key] = new_parents
1960
for text_key in invalid_keys:
1961
text_index[text_key] = [NULL_REVISION]
1964
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
1965
"""Get an iterable listing the keys of all the data introduced by a set
1968
The keys will be ordered so that the corresponding items can be safely
1969
fetched and inserted in that order.
1971
:returns: An iterable producing tuples of (knit-kind, file-id,
1972
versions). knit-kind is one of 'file', 'inventory', 'signatures',
1973
'revisions'. file-id is None unless knit-kind is 'file'.
1975
# XXX: it's a bit weird to control the inventory weave caching in this
1976
# generator. Ideally the caching would be done in fetch.py I think. Or
1977
# maybe this generator should explicitly have the contract that it
1978
# should not be iterated until the previously yielded item has been
1980
inv_w = self.inventories
1982
# file ids that changed
1983
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
1985
num_file_ids = len(file_ids)
1986
for file_id, altered_versions in file_ids.iteritems():
1987
if _files_pb is not None:
1988
_files_pb.update("fetch texts", count, num_file_ids)
1990
yield ("file", file_id, altered_versions)
1991
# We're done with the files_pb. Note that it finished by the caller,
1992
# just as it was created by the caller.
1996
yield ("inventory", None, revision_ids)
1999
# XXX: Note ATM no callers actually pay attention to this return
2000
# instead they just use the list of revision ids and ignore
2001
# missing sigs. Consider removing this work entirely
2002
revisions_with_signatures = set(self.signatures.get_parent_map(
2003
[(r,) for r in revision_ids]))
2004
revisions_with_signatures = set(
2005
[r for (r,) in revisions_with_signatures])
2006
revisions_with_signatures.intersection_update(revision_ids)
2007
yield ("signatures", None, revisions_with_signatures)
2010
yield ("revisions", None, revision_ids)
2013
def get_inventory(self, revision_id):
2014
"""Get Inventory object by revision id."""
2015
return self.iter_inventories([revision_id]).next()
2017
def iter_inventories(self, revision_ids):
2018
"""Get many inventories by revision_ids.
2020
This will buffer some or all of the texts used in constructing the
2021
inventories in memory, but will only parse a single inventory at a
2024
:return: An iterator of inventories.
2026
if ((None in revision_ids)
2027
or (_mod_revision.NULL_REVISION in revision_ids)):
2028
raise ValueError('cannot get null revision inventory')
2029
return self._iter_inventories(revision_ids)
2031
def _iter_inventories(self, revision_ids):
2032
"""single-document based inventory iteration."""
2033
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2034
yield self.deserialise_inventory(revision_id, text)
2036
def _iter_inventory_xmls(self, revision_ids):
2037
keys = [(revision_id,) for revision_id in revision_ids]
2038
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2040
for record in stream:
2041
if record.storage_kind != 'absent':
2042
text_chunks[record.key] = record.get_bytes_as('chunked')
2044
raise errors.NoSuchRevision(self, record.key)
2046
chunks = text_chunks.pop(key)
2047
yield ''.join(chunks), key[-1]
2049
def deserialise_inventory(self, revision_id, xml):
2050
"""Transform the xml into an inventory object.
2052
:param revision_id: The expected revision id of the inventory.
2053
:param xml: A serialised inventory.
2055
result = self._serializer.read_inventory_from_string(xml, revision_id,
2056
entry_cache=self._inventory_entry_cache)
2057
if result.revision_id != revision_id:
2058
raise AssertionError('revision id mismatch %s != %s' % (
2059
result.revision_id, revision_id))
2062
def serialise_inventory(self, inv):
2063
return self._serializer.write_inventory_to_string(inv)
2065
def _serialise_inventory_to_lines(self, inv):
2066
return self._serializer.write_inventory_to_lines(inv)
2068
def get_serializer_format(self):
2069
return self._serializer.format_num
2072
def get_inventory_xml(self, revision_id):
2073
"""Get inventory XML as a file object."""
2074
texts = self._iter_inventory_xmls([revision_id])
2076
text, revision_id = texts.next()
2077
except StopIteration:
2078
raise errors.HistoryMissing(self, 'inventory', revision_id)
2082
def get_inventory_sha1(self, revision_id):
2083
"""Return the sha1 hash of the inventory entry
2085
return self.get_revision(revision_id).inventory_sha1
2087
def iter_reverse_revision_history(self, revision_id):
2088
"""Iterate backwards through revision ids in the lefthand history
2090
:param revision_id: The revision id to start with. All its lefthand
2091
ancestors will be traversed.
2093
graph = self.get_graph()
2094
next_id = revision_id
2096
if next_id in (None, _mod_revision.NULL_REVISION):
2099
# Note: The following line may raise KeyError in the event of
2100
# truncated history. We decided not to have a try:except:raise
2101
# RevisionNotPresent here until we see a use for it, because of the
2102
# cost in an inner loop that is by its very nature O(history).
2103
# Robert Collins 20080326
2104
parents = graph.get_parent_map([next_id])[next_id]
2105
if len(parents) == 0:
2108
next_id = parents[0]
2111
def get_revision_inventory(self, revision_id):
2112
"""Return inventory of a past revision."""
2113
# TODO: Unify this with get_inventory()
2114
# bzr 0.0.6 and later imposes the constraint that the inventory_id
2115
# must be the same as its revision, so this is trivial.
2116
if revision_id is None:
2117
# This does not make sense: if there is no revision,
2118
# then it is the current tree inventory surely ?!
2119
# and thus get_root_id() is something that looks at the last
2120
# commit on the branch, and the get_root_id is an inventory check.
2121
raise NotImplementedError
2122
# return Inventory(self.get_root_id())
2124
return self.get_inventory(revision_id)
2126
def is_shared(self):
2127
"""Return True if this repository is flagged as a shared repository."""
2128
raise NotImplementedError(self.is_shared)
2131
def reconcile(self, other=None, thorough=False):
2132
"""Reconcile this repository."""
2133
from bzrlib.reconcile import RepoReconciler
2134
reconciler = RepoReconciler(self, thorough=thorough)
2135
reconciler.reconcile()
2138
def _refresh_data(self):
2139
"""Helper called from lock_* to ensure coherency with disk.
2141
The default implementation does nothing; it is however possible
2142
for repositories to maintain loaded indices across multiple locks
2143
by checking inside their implementation of this method to see
2144
whether their indices are still valid. This depends of course on
2145
the disk format being validatable in this manner. This method is
2146
also called by the refresh_data() public interface to cause a refresh
2147
to occur while in a write lock so that data inserted by a smart server
2148
push operation is visible on the client's instance of the physical
2153
def revision_tree(self, revision_id):
2154
"""Return Tree for a revision on this branch.
2156
`revision_id` may be NULL_REVISION for the empty tree revision.
2158
revision_id = _mod_revision.ensure_null(revision_id)
2159
# TODO: refactor this to use an existing revision object
2160
# so we don't need to read it in twice.
2161
if revision_id == _mod_revision.NULL_REVISION:
2162
return RevisionTree(self, Inventory(root_id=None),
2163
_mod_revision.NULL_REVISION)
2165
inv = self.get_revision_inventory(revision_id)
2166
return RevisionTree(self, inv, revision_id)
2168
def revision_trees(self, revision_ids):
2169
"""Return Trees for revisions in this repository.
2171
:param revision_ids: a sequence of revision-ids;
2172
a revision-id may not be None or 'null:'
2174
inventories = self.iter_inventories(revision_ids)
2175
for inv in inventories:
2176
yield RevisionTree(self, inv, inv.revision_id)
2178
def _filtered_revision_trees(self, revision_ids, file_ids):
2179
"""Return Tree for a revision on this branch with only some files.
2181
:param revision_ids: a sequence of revision-ids;
2182
a revision-id may not be None or 'null:'
2183
:param file_ids: if not None, the result is filtered
2184
so that only those file-ids, their parents and their
2185
children are included.
2187
inventories = self.iter_inventories(revision_ids)
2188
for inv in inventories:
2189
# Should we introduce a FilteredRevisionTree class rather
2190
# than pre-filter the inventory here?
2191
filtered_inv = inv.filter(file_ids)
2192
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2195
def get_ancestry(self, revision_id, topo_sorted=True):
2196
"""Return a list of revision-ids integrated by a revision.
2198
The first element of the list is always None, indicating the origin
2199
revision. This might change when we have history horizons, or
2200
perhaps we should have a new API.
2202
This is topologically sorted.
2204
if _mod_revision.is_null(revision_id):
2206
if not self.has_revision(revision_id):
2207
raise errors.NoSuchRevision(self, revision_id)
2208
graph = self.get_graph()
2210
search = graph._make_breadth_first_searcher([revision_id])
2213
found, ghosts = search.next_with_ghosts()
2214
except StopIteration:
2217
if _mod_revision.NULL_REVISION in keys:
2218
keys.remove(_mod_revision.NULL_REVISION)
2220
parent_map = graph.get_parent_map(keys)
2221
keys = tsort.topo_sort(parent_map)
2222
return [None] + list(keys)
2225
"""Compress the data within the repository.
2227
This operation only makes sense for some repository types. For other
2228
types it should be a no-op that just returns.
2230
This stub method does not require a lock, but subclasses should use
2231
@needs_write_lock as this is a long running call its reasonable to
2232
implicitly lock for the user.
2235
def get_transaction(self):
2236
return self.control_files.get_transaction()
2238
def get_parent_map(self, revision_ids):
2239
"""See graph._StackedParentsProvider.get_parent_map"""
2240
# revisions index works in keys; this just works in revisions
2241
# therefore wrap and unwrap
2244
for revision_id in revision_ids:
2245
if revision_id == _mod_revision.NULL_REVISION:
2246
result[revision_id] = ()
2247
elif revision_id is None:
2248
raise ValueError('get_parent_map(None) is not valid')
2250
query_keys.append((revision_id ,))
2251
for ((revision_id,), parent_keys) in \
2252
self.revisions.get_parent_map(query_keys).iteritems():
2254
result[revision_id] = tuple(parent_revid
2255
for (parent_revid,) in parent_keys)
2257
result[revision_id] = (_mod_revision.NULL_REVISION,)
2260
def _make_parents_provider(self):
2263
def get_graph(self, other_repository=None):
2264
"""Return the graph walker for this repository format"""
2265
parents_provider = self._make_parents_provider()
2266
if (other_repository is not None and
2267
not self.has_same_location(other_repository)):
2268
parents_provider = graph._StackedParentsProvider(
2269
[parents_provider, other_repository._make_parents_provider()])
2270
return graph.Graph(parents_provider)
2272
def _get_versioned_file_checker(self, text_key_references=None):
2273
"""Return an object suitable for checking versioned files.
2275
:param text_key_references: if non-None, an already built
2276
dictionary mapping text keys ((fileid, revision_id) tuples)
2277
to whether they were referred to by the inventory of the
2278
revision_id that they contain. If None, this will be
2281
return _VersionedFileChecker(self,
2282
text_key_references=text_key_references)
2284
def revision_ids_to_search_result(self, result_set):
2285
"""Convert a set of revision ids to a graph SearchResult."""
2286
result_parents = set()
2287
for parents in self.get_graph().get_parent_map(
2288
result_set).itervalues():
2289
result_parents.update(parents)
2290
included_keys = result_set.intersection(result_parents)
2291
start_keys = result_set.difference(included_keys)
2292
exclude_keys = result_parents.difference(result_set)
2293
result = graph.SearchResult(start_keys, exclude_keys,
2294
len(result_set), result_set)
2298
def set_make_working_trees(self, new_value):
2299
"""Set the policy flag for making working trees when creating branches.
2301
This only applies to branches that use this repository.
2303
The default is 'True'.
2304
:param new_value: True to restore the default, False to disable making
2307
raise NotImplementedError(self.set_make_working_trees)
2309
def make_working_trees(self):
2310
"""Returns the policy for making working trees on new branches."""
2311
raise NotImplementedError(self.make_working_trees)
2314
def sign_revision(self, revision_id, gpg_strategy):
2315
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2316
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2319
def has_signature_for_revision_id(self, revision_id):
2320
"""Query for a revision signature for revision_id in the repository."""
2321
if not self.has_revision(revision_id):
2322
raise errors.NoSuchRevision(self, revision_id)
2323
sig_present = (1 == len(
2324
self.signatures.get_parent_map([(revision_id,)])))
2328
def get_signature_text(self, revision_id):
2329
"""Return the text for a signature."""
2330
stream = self.signatures.get_record_stream([(revision_id,)],
2332
record = stream.next()
2333
if record.storage_kind == 'absent':
2334
raise errors.NoSuchRevision(self, revision_id)
2335
return record.get_bytes_as('fulltext')
2338
def check(self, revision_ids=None):
2339
"""Check consistency of all history of given revision_ids.
2341
Different repository implementations should override _check().
2343
:param revision_ids: A non-empty list of revision_ids whose ancestry
2344
will be checked. Typically the last revision_id of a branch.
2346
return self._check(revision_ids)
2348
def _check(self, revision_ids):
2349
result = check.Check(self)
2353
def _warn_if_deprecated(self):
2354
global _deprecation_warning_done
2355
if _deprecation_warning_done:
2357
_deprecation_warning_done = True
2358
warning("Format %s for %s is deprecated - please use 'bzr upgrade' to get better performance"
2359
% (self._format, self.bzrdir.transport.base))
2361
def supports_rich_root(self):
2362
return self._format.rich_root_data
2364
def _check_ascii_revisionid(self, revision_id, method):
2365
"""Private helper for ascii-only repositories."""
2366
# weave repositories refuse to store revisionids that are non-ascii.
2367
if revision_id is not None:
2368
# weaves require ascii revision ids.
2369
if isinstance(revision_id, unicode):
2371
revision_id.encode('ascii')
2372
except UnicodeEncodeError:
2373
raise errors.NonAsciiRevisionId(method, self)
2376
revision_id.decode('ascii')
2377
except UnicodeDecodeError:
2378
raise errors.NonAsciiRevisionId(method, self)
2380
def revision_graph_can_have_wrong_parents(self):
2381
"""Is it possible for this repository to have a revision graph with
2384
If True, then this repository must also implement
2385
_find_inconsistent_revision_parents so that check and reconcile can
2386
check for inconsistencies before proceeding with other checks that may
2387
depend on the revision index being consistent.
2389
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2392
# remove these delegates a while after bzr 0.15
2393
def __make_delegated(name, from_module):
2394
def _deprecated_repository_forwarder():
2395
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2396
% (name, from_module),
2399
m = __import__(from_module, globals(), locals(), [name])
2401
return getattr(m, name)
2402
except AttributeError:
2403
raise AttributeError('module %s has no name %s'
2405
globals()[name] = _deprecated_repository_forwarder
2408
'AllInOneRepository',
2409
'WeaveMetaDirRepository',
2410
'PreSplitOutRepositoryFormat',
2411
'RepositoryFormat4',
2412
'RepositoryFormat5',
2413
'RepositoryFormat6',
2414
'RepositoryFormat7',
2416
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2420
'RepositoryFormatKnit',
2421
'RepositoryFormatKnit1',
2423
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2426
def install_revision(repository, rev, revision_tree):
2427
"""Install all revision data into a repository."""
2428
install_revisions(repository, [(rev, revision_tree, None)])
2431
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2432
"""Install all revision data into a repository.
2434
Accepts an iterable of revision, tree, signature tuples. The signature
2437
repository.start_write_group()
2439
for n, (revision, revision_tree, signature) in enumerate(iterable):
2440
_install_revision(repository, revision, revision_tree, signature)
2442
pb.update('Transferring revisions', n + 1, num_revisions)
2444
repository.abort_write_group()
2447
repository.commit_write_group()
2450
def _install_revision(repository, rev, revision_tree, signature):
2451
"""Install all revision data into a repository."""
2452
present_parents = []
2454
for p_id in rev.parent_ids:
2455
if repository.has_revision(p_id):
2456
present_parents.append(p_id)
2457
parent_trees[p_id] = repository.revision_tree(p_id)
2459
parent_trees[p_id] = repository.revision_tree(
2460
_mod_revision.NULL_REVISION)
2462
inv = revision_tree.inventory
2463
entries = inv.iter_entries()
2464
# backwards compatibility hack: skip the root id.
2465
if not repository.supports_rich_root():
2466
path, root = entries.next()
2467
if root.revision != rev.revision_id:
2468
raise errors.IncompatibleRevision(repr(repository))
2470
for path, ie in entries:
2471
text_keys[(ie.file_id, ie.revision)] = ie
2472
text_parent_map = repository.texts.get_parent_map(text_keys)
2473
missing_texts = set(text_keys) - set(text_parent_map)
2474
# Add the texts that are not already present
2475
for text_key in missing_texts:
2476
ie = text_keys[text_key]
2478
# FIXME: TODO: The following loop overlaps/duplicates that done by
2479
# commit to determine parents. There is a latent/real bug here where
2480
# the parents inserted are not those commit would do - in particular
2481
# they are not filtered by heads(). RBC, AB
2482
for revision, tree in parent_trees.iteritems():
2483
if ie.file_id not in tree:
2485
parent_id = tree.inventory[ie.file_id].revision
2486
if parent_id in text_parents:
2488
text_parents.append((ie.file_id, parent_id))
2489
lines = revision_tree.get_file(ie.file_id).readlines()
2490
repository.texts.add_lines(text_key, text_parents, lines)
2492
# install the inventory
2493
repository.add_inventory(rev.revision_id, inv, present_parents)
2494
except errors.RevisionAlreadyPresent:
2496
if signature is not None:
2497
repository.add_signature_text(rev.revision_id, signature)
2498
repository.add_revision(rev.revision_id, rev, inv)
2501
class MetaDirRepository(Repository):
2502
"""Repositories in the new meta-dir layout.
2504
:ivar _transport: Transport for access to repository control files,
2505
typically pointing to .bzr/repository.
2508
def __init__(self, _format, a_bzrdir, control_files):
2509
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2510
self._transport = control_files._transport
2512
def is_shared(self):
2513
"""Return True if this repository is flagged as a shared repository."""
2514
return self._transport.has('shared-storage')
2517
def set_make_working_trees(self, new_value):
2518
"""Set the policy flag for making working trees when creating branches.
2520
This only applies to branches that use this repository.
2522
The default is 'True'.
2523
:param new_value: True to restore the default, False to disable making
2528
self._transport.delete('no-working-trees')
2529
except errors.NoSuchFile:
2532
self._transport.put_bytes('no-working-trees', '',
2533
mode=self.bzrdir._get_file_mode())
2535
def make_working_trees(self):
2536
"""Returns the policy for making working trees on new branches."""
2537
return not self._transport.has('no-working-trees')
2540
class MetaDirVersionedFileRepository(MetaDirRepository):
2541
"""Repositories in a meta-dir, that work via versioned file objects."""
2543
def __init__(self, _format, a_bzrdir, control_files):
2544
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2548
network_format_registry = registry.FormatRegistry()
2549
"""Registry of formats indexed by their network name.
2551
The network name for a repository format is an identifier that can be used when
2552
referring to formats with smart server operations. See
2553
RepositoryFormat.network_name() for more detail.
2557
format_registry = registry.FormatRegistry(network_format_registry)
2558
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
2560
This can contain either format instances themselves, or classes/factories that
2561
can be called to obtain one.
2565
#####################################################################
2566
# Repository Formats
2568
class RepositoryFormat(object):
2569
"""A repository format.
2571
Formats provide four things:
2572
* An initialization routine to construct repository data on disk.
2573
* a optional format string which is used when the BzrDir supports
2575
* an open routine which returns a Repository instance.
2576
* A network name for referring to the format in smart server RPC
2579
There is one and only one Format subclass for each on-disk format. But
2580
there can be one Repository subclass that is used for several different
2581
formats. The _format attribute on a Repository instance can be used to
2582
determine the disk format.
2584
Formats are placed in a registry by their format string for reference
2585
during opening. These should be subclasses of RepositoryFormat for
2588
Once a format is deprecated, just deprecate the initialize and open
2589
methods on the format class. Do not deprecate the object, as the
2590
object may be created even when a repository instnace hasn't been
2593
Common instance attributes:
2594
_matchingbzrdir - the bzrdir format that the repository format was
2595
originally written to work with. This can be used if manually
2596
constructing a bzrdir and repository, or more commonly for test suite
2600
# Set to True or False in derived classes. True indicates that the format
2601
# supports ghosts gracefully.
2602
supports_ghosts = None
2603
# Can this repository be given external locations to lookup additional
2604
# data. Set to True or False in derived classes.
2605
supports_external_lookups = None
2606
# What order should fetch operations request streams in?
2607
# The default is unordered as that is the cheapest for an origin to
2609
_fetch_order = 'unordered'
2610
# Does this repository format use deltas that can be fetched as-deltas ?
2611
# (E.g. knits, where the knit deltas can be transplanted intact.
2612
# We default to False, which will ensure that enough data to get
2613
# a full text out of any fetch stream will be grabbed.
2614
_fetch_uses_deltas = False
2615
# Should fetch trigger a reconcile after the fetch? Only needed for
2616
# some repository formats that can suffer internal inconsistencies.
2617
_fetch_reconcile = False
2618
# Does this format have < O(tree_size) delta generation. Used to hint what
2619
# code path for commit, amongst other things.
2623
return "<%s>" % self.__class__.__name__
2625
def __eq__(self, other):
2626
# format objects are generally stateless
2627
return isinstance(other, self.__class__)
2629
def __ne__(self, other):
2630
return not self == other
2633
def find_format(klass, a_bzrdir):
2634
"""Return the format for the repository object in a_bzrdir.
2636
This is used by bzr native formats that have a "format" file in
2637
the repository. Other methods may be used by different types of
2641
transport = a_bzrdir.get_repository_transport(None)
2642
format_string = transport.get("format").read()
2643
return format_registry.get(format_string)
2644
except errors.NoSuchFile:
2645
raise errors.NoRepositoryPresent(a_bzrdir)
2647
raise errors.UnknownFormatError(format=format_string,
2651
def register_format(klass, format):
2652
format_registry.register(format.get_format_string(), format)
2655
def unregister_format(klass, format):
2656
format_registry.remove(format.get_format_string())
2659
def get_default_format(klass):
2660
"""Return the current default format."""
2661
from bzrlib import bzrdir
2662
return bzrdir.format_registry.make_bzrdir('default').repository_format
2664
def get_format_string(self):
2665
"""Return the ASCII format string that identifies this format.
2667
Note that in pre format ?? repositories the format string is
2668
not permitted nor written to disk.
2670
raise NotImplementedError(self.get_format_string)
2672
def get_format_description(self):
2673
"""Return the short description for this format."""
2674
raise NotImplementedError(self.get_format_description)
2676
# TODO: this shouldn't be in the base class, it's specific to things that
2677
# use weaves or knits -- mbp 20070207
2678
def _get_versioned_file_store(self,
2683
versionedfile_class=None,
2684
versionedfile_kwargs={},
2686
if versionedfile_class is None:
2687
versionedfile_class = self._versionedfile_class
2688
weave_transport = control_files._transport.clone(name)
2689
dir_mode = control_files._dir_mode
2690
file_mode = control_files._file_mode
2691
return VersionedFileStore(weave_transport, prefixed=prefixed,
2693
file_mode=file_mode,
2694
versionedfile_class=versionedfile_class,
2695
versionedfile_kwargs=versionedfile_kwargs,
2698
def initialize(self, a_bzrdir, shared=False):
2699
"""Initialize a repository of this format in a_bzrdir.
2701
:param a_bzrdir: The bzrdir to put the new repository in it.
2702
:param shared: The repository should be initialized as a sharable one.
2703
:returns: The new repository object.
2705
This may raise UninitializableFormat if shared repository are not
2706
compatible the a_bzrdir.
2708
raise NotImplementedError(self.initialize)
2710
def is_supported(self):
2711
"""Is this format supported?
2713
Supported formats must be initializable and openable.
2714
Unsupported formats may not support initialization or committing or
2715
some other features depending on the reason for not being supported.
2719
def network_name(self):
2720
"""A simple byte string uniquely identifying this format for RPC calls.
2722
MetaDir repository formats use their disk format string to identify the
2723
repository over the wire. All in one formats such as bzr < 0.8, and
2724
foreign formats like svn/git and hg should use some marker which is
2725
unique and immutable.
2727
raise NotImplementedError(self.network_name)
2729
def check_conversion_target(self, target_format):
2730
raise NotImplementedError(self.check_conversion_target)
2732
def open(self, a_bzrdir, _found=False):
2733
"""Return an instance of this format for the bzrdir a_bzrdir.
2735
_found is a private parameter, do not use it.
2737
raise NotImplementedError(self.open)
2740
class MetaDirRepositoryFormat(RepositoryFormat):
2741
"""Common base class for the new repositories using the metadir layout."""
2743
rich_root_data = False
2744
supports_tree_reference = False
2745
supports_external_lookups = False
2748
def _matchingbzrdir(self):
2749
matching = bzrdir.BzrDirMetaFormat1()
2750
matching.repository_format = self
2754
super(MetaDirRepositoryFormat, self).__init__()
2756
def _create_control_files(self, a_bzrdir):
2757
"""Create the required files and the initial control_files object."""
2758
# FIXME: RBC 20060125 don't peek under the covers
2759
# NB: no need to escape relative paths that are url safe.
2760
repository_transport = a_bzrdir.get_repository_transport(self)
2761
control_files = lockable_files.LockableFiles(repository_transport,
2762
'lock', lockdir.LockDir)
2763
control_files.create_lock()
2764
return control_files
2766
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
2767
"""Upload the initial blank content."""
2768
control_files = self._create_control_files(a_bzrdir)
2769
control_files.lock_write()
2770
transport = control_files._transport
2772
utf8_files += [('shared-storage', '')]
2774
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
2775
for (filename, content_stream) in files:
2776
transport.put_file(filename, content_stream,
2777
mode=a_bzrdir._get_file_mode())
2778
for (filename, content_bytes) in utf8_files:
2779
transport.put_bytes_non_atomic(filename, content_bytes,
2780
mode=a_bzrdir._get_file_mode())
2782
control_files.unlock()
2784
def network_name(self):
2785
"""Metadir formats have matching disk and network format strings."""
2786
return self.get_format_string()
2789
# Pre-0.8 formats that don't have a disk format string (because they are
2790
# versioned by the matching control directory). We use the control directories
2791
# disk format string as a key for the network_name because they meet the
2792
# constraints (simple string, unique, immmutable).
2793
network_format_registry.register_lazy(
2794
"Bazaar-NG branch, format 5\n",
2795
'bzrlib.repofmt.weaverepo',
2796
'RepositoryFormat5',
2798
network_format_registry.register_lazy(
2799
"Bazaar-NG branch, format 6\n",
2800
'bzrlib.repofmt.weaverepo',
2801
'RepositoryFormat6',
2804
# formats which have no format string are not discoverable or independently
2805
# creatable on disk, so are not registered in format_registry. They're
2806
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
2807
# needed, it's constructed directly by the BzrDir. Non-native formats where
2808
# the repository is not separately opened are similar.
2810
format_registry.register_lazy(
2811
'Bazaar-NG Repository format 7',
2812
'bzrlib.repofmt.weaverepo',
2816
format_registry.register_lazy(
2817
'Bazaar-NG Knit Repository Format 1',
2818
'bzrlib.repofmt.knitrepo',
2819
'RepositoryFormatKnit1',
2822
format_registry.register_lazy(
2823
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
2824
'bzrlib.repofmt.knitrepo',
2825
'RepositoryFormatKnit3',
2828
format_registry.register_lazy(
2829
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
2830
'bzrlib.repofmt.knitrepo',
2831
'RepositoryFormatKnit4',
2834
# Pack-based formats. There is one format for pre-subtrees, and one for
2835
# post-subtrees to allow ease of testing.
2836
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
2837
format_registry.register_lazy(
2838
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
2839
'bzrlib.repofmt.pack_repo',
2840
'RepositoryFormatKnitPack1',
2842
format_registry.register_lazy(
2843
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
2844
'bzrlib.repofmt.pack_repo',
2845
'RepositoryFormatKnitPack3',
2847
format_registry.register_lazy(
2848
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
2849
'bzrlib.repofmt.pack_repo',
2850
'RepositoryFormatKnitPack4',
2852
format_registry.register_lazy(
2853
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
2854
'bzrlib.repofmt.pack_repo',
2855
'RepositoryFormatKnitPack5',
2857
format_registry.register_lazy(
2858
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
2859
'bzrlib.repofmt.pack_repo',
2860
'RepositoryFormatKnitPack5RichRoot',
2862
format_registry.register_lazy(
2863
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
2864
'bzrlib.repofmt.pack_repo',
2865
'RepositoryFormatKnitPack5RichRootBroken',
2867
format_registry.register_lazy(
2868
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
2869
'bzrlib.repofmt.pack_repo',
2870
'RepositoryFormatKnitPack6',
2872
format_registry.register_lazy(
2873
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
2874
'bzrlib.repofmt.pack_repo',
2875
'RepositoryFormatKnitPack6RichRoot',
2878
# Development formats.
2879
# 1.7->1.8 go below here
2880
format_registry.register_lazy(
2881
"Bazaar development format 2 (needs bzr.dev from before 1.8)\n",
2882
'bzrlib.repofmt.pack_repo',
2883
'RepositoryFormatPackDevelopment2',
2885
format_registry.register_lazy(
2886
("Bazaar development format 2 with subtree support "
2887
"(needs bzr.dev from before 1.8)\n"),
2888
'bzrlib.repofmt.pack_repo',
2889
'RepositoryFormatPackDevelopment2Subtree',
2893
class InterRepository(InterObject):
2894
"""This class represents operations taking place between two repositories.
2896
Its instances have methods like copy_content and fetch, and contain
2897
references to the source and target repositories these operations can be
2900
Often we will provide convenience methods on 'repository' which carry out
2901
operations with another repository - they will always forward to
2902
InterRepository.get(other).method_name(parameters).
2905
_walk_to_common_revisions_batch_size = 50
2907
"""The available optimised InterRepository types."""
2910
def copy_content(self, revision_id=None):
2911
"""Make a complete copy of the content in self into destination.
2913
This is a destructive operation! Do not use it on existing
2916
:param revision_id: Only copy the content needed to construct
2917
revision_id and its parents.
2920
self.target.set_make_working_trees(self.source.make_working_trees())
2921
except NotImplementedError:
2923
self.target.fetch(self.source, revision_id=revision_id)
2926
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
2928
"""Fetch the content required to construct revision_id.
2930
The content is copied from self.source to self.target.
2932
:param revision_id: if None all content is copied, if NULL_REVISION no
2934
:param pb: optional progress bar to use for progress reports. If not
2935
provided a default one will be created.
2938
from bzrlib.fetch import RepoFetcher
2939
f = RepoFetcher(to_repository=self.target,
2940
from_repository=self.source,
2941
last_revision=revision_id,
2942
fetch_spec=fetch_spec,
2943
pb=pb, find_ghosts=find_ghosts)
2945
def _walk_to_common_revisions(self, revision_ids):
2946
"""Walk out from revision_ids in source to revisions target has.
2948
:param revision_ids: The start point for the search.
2949
:return: A set of revision ids.
2951
target_graph = self.target.get_graph()
2952
revision_ids = frozenset(revision_ids)
2953
# Fast path for the case where all the revisions are already in the
2955
# (Although this does incur an extra round trip for the
2956
# fairly common case where the target doesn't already have the revision
2958
if set(target_graph.get_parent_map(revision_ids)) == revision_ids:
2959
return graph.SearchResult(revision_ids, set(), 0, set())
2960
missing_revs = set()
2961
source_graph = self.source.get_graph()
2962
# ensure we don't pay silly lookup costs.
2963
searcher = source_graph._make_breadth_first_searcher(revision_ids)
2964
null_set = frozenset([_mod_revision.NULL_REVISION])
2965
searcher_exhausted = False
2969
# Iterate the searcher until we have enough next_revs
2970
while len(next_revs) < self._walk_to_common_revisions_batch_size:
2972
next_revs_part, ghosts_part = searcher.next_with_ghosts()
2973
next_revs.update(next_revs_part)
2974
ghosts.update(ghosts_part)
2975
except StopIteration:
2976
searcher_exhausted = True
2978
# If there are ghosts in the source graph, and the caller asked for
2979
# them, make sure that they are present in the target.
2980
# We don't care about other ghosts as we can't fetch them and
2981
# haven't been asked to.
2982
ghosts_to_check = set(revision_ids.intersection(ghosts))
2983
revs_to_get = set(next_revs).union(ghosts_to_check)
2985
have_revs = set(target_graph.get_parent_map(revs_to_get))
2986
# we always have NULL_REVISION present.
2987
have_revs = have_revs.union(null_set)
2988
# Check if the target is missing any ghosts we need.
2989
ghosts_to_check.difference_update(have_revs)
2991
# One of the caller's revision_ids is a ghost in both the
2992
# source and the target.
2993
raise errors.NoSuchRevision(
2994
self.source, ghosts_to_check.pop())
2995
missing_revs.update(next_revs - have_revs)
2996
# Because we may have walked past the original stop point, make
2997
# sure everything is stopped
2998
stop_revs = searcher.find_seen_ancestors(have_revs)
2999
searcher.stop_searching_any(stop_revs)
3000
if searcher_exhausted:
3002
return searcher.get_result()
3005
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3006
"""Return the revision ids that source has that target does not.
3008
:param revision_id: only return revision ids included by this
3010
:param find_ghosts: If True find missing revisions in deep history
3011
rather than just finding the surface difference.
3012
:return: A bzrlib.graph.SearchResult.
3014
# stop searching at found target revisions.
3015
if not find_ghosts and revision_id is not None:
3016
return self._walk_to_common_revisions([revision_id])
3017
# generic, possibly worst case, slow code path.
3018
target_ids = set(self.target.all_revision_ids())
3019
if revision_id is not None:
3020
source_ids = self.source.get_ancestry(revision_id)
3021
if source_ids[0] is not None:
3022
raise AssertionError()
3025
source_ids = self.source.all_revision_ids()
3026
result_set = set(source_ids).difference(target_ids)
3027
return self.source.revision_ids_to_search_result(result_set)
3030
def _same_model(source, target):
3031
"""True if source and target have the same data representation.
3033
Note: this is always called on the base class; overriding it in a
3034
subclass will have no effect.
3037
InterRepository._assert_same_model(source, target)
3039
except errors.IncompatibleRepositories, e:
3043
def _assert_same_model(source, target):
3044
"""Raise an exception if two repositories do not use the same model.
3046
if source.supports_rich_root() != target.supports_rich_root():
3047
raise errors.IncompatibleRepositories(source, target,
3048
"different rich-root support")
3049
if source._serializer != target._serializer:
3050
raise errors.IncompatibleRepositories(source, target,
3051
"different serializers")
3054
class InterSameDataRepository(InterRepository):
3055
"""Code for converting between repositories that represent the same data.
3057
Data format and model must match for this to work.
3061
def _get_repo_format_to_test(self):
3062
"""Repository format for testing with.
3064
InterSameData can pull from subtree to subtree and from non-subtree to
3065
non-subtree, so we test this with the richest repository format.
3067
from bzrlib.repofmt import knitrepo
3068
return knitrepo.RepositoryFormatKnit3()
3071
def is_compatible(source, target):
3072
return InterRepository._same_model(source, target)
3075
class InterWeaveRepo(InterSameDataRepository):
3076
"""Optimised code paths between Weave based repositories.
3078
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3079
implemented lazy inter-object optimisation.
3083
def _get_repo_format_to_test(self):
3084
from bzrlib.repofmt import weaverepo
3085
return weaverepo.RepositoryFormat7()
3088
def is_compatible(source, target):
3089
"""Be compatible with known Weave formats.
3091
We don't test for the stores being of specific types because that
3092
could lead to confusing results, and there is no need to be
3095
from bzrlib.repofmt.weaverepo import (
3101
return (isinstance(source._format, (RepositoryFormat5,
3103
RepositoryFormat7)) and
3104
isinstance(target._format, (RepositoryFormat5,
3106
RepositoryFormat7)))
3107
except AttributeError:
3111
def copy_content(self, revision_id=None):
3112
"""See InterRepository.copy_content()."""
3113
# weave specific optimised path:
3115
self.target.set_make_working_trees(self.source.make_working_trees())
3116
except (errors.RepositoryUpgradeRequired, NotImplemented):
3118
# FIXME do not peek!
3119
if self.source._transport.listable():
3120
pb = ui.ui_factory.nested_progress_bar()
3122
self.target.texts.insert_record_stream(
3123
self.source.texts.get_record_stream(
3124
self.source.texts.keys(), 'topological', False))
3125
pb.update('copying inventory', 0, 1)
3126
self.target.inventories.insert_record_stream(
3127
self.source.inventories.get_record_stream(
3128
self.source.inventories.keys(), 'topological', False))
3129
self.target.signatures.insert_record_stream(
3130
self.source.signatures.get_record_stream(
3131
self.source.signatures.keys(),
3133
self.target.revisions.insert_record_stream(
3134
self.source.revisions.get_record_stream(
3135
self.source.revisions.keys(),
3136
'topological', True))
3140
self.target.fetch(self.source, revision_id=revision_id)
3143
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3144
"""See InterRepository.missing_revision_ids()."""
3145
# we want all revisions to satisfy revision_id in source.
3146
# but we don't want to stat every file here and there.
3147
# we want then, all revisions other needs to satisfy revision_id
3148
# checked, but not those that we have locally.
3149
# so the first thing is to get a subset of the revisions to
3150
# satisfy revision_id in source, and then eliminate those that
3151
# we do already have.
3152
# this is slow on high latency connection to self, but as as this
3153
# disk format scales terribly for push anyway due to rewriting
3154
# inventory.weave, this is considered acceptable.
3156
if revision_id is not None:
3157
source_ids = self.source.get_ancestry(revision_id)
3158
if source_ids[0] is not None:
3159
raise AssertionError()
3162
source_ids = self.source._all_possible_ids()
3163
source_ids_set = set(source_ids)
3164
# source_ids is the worst possible case we may need to pull.
3165
# now we want to filter source_ids against what we actually
3166
# have in target, but don't try to check for existence where we know
3167
# we do not have a revision as that would be pointless.
3168
target_ids = set(self.target._all_possible_ids())
3169
possibly_present_revisions = target_ids.intersection(source_ids_set)
3170
actually_present_revisions = set(
3171
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3172
required_revisions = source_ids_set.difference(actually_present_revisions)
3173
if revision_id is not None:
3174
# we used get_ancestry to determine source_ids then we are assured all
3175
# revisions referenced are present as they are installed in topological order.
3176
# and the tip revision was validated by get_ancestry.
3177
result_set = required_revisions
3179
# if we just grabbed the possibly available ids, then
3180
# we only have an estimate of whats available and need to validate
3181
# that against the revision records.
3183
self.source._eliminate_revisions_not_present(required_revisions))
3184
return self.source.revision_ids_to_search_result(result_set)
3187
class InterKnitRepo(InterSameDataRepository):
3188
"""Optimised code paths between Knit based repositories."""
3191
def _get_repo_format_to_test(self):
3192
from bzrlib.repofmt import knitrepo
3193
return knitrepo.RepositoryFormatKnit1()
3196
def is_compatible(source, target):
3197
"""Be compatible with known Knit formats.
3199
We don't test for the stores being of specific types because that
3200
could lead to confusing results, and there is no need to be
3203
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3205
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3206
isinstance(target._format, RepositoryFormatKnit))
3207
except AttributeError:
3209
return are_knits and InterRepository._same_model(source, target)
3212
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3213
"""See InterRepository.missing_revision_ids()."""
3214
if revision_id is not None:
3215
source_ids = self.source.get_ancestry(revision_id)
3216
if source_ids[0] is not None:
3217
raise AssertionError()
3220
source_ids = self.source.all_revision_ids()
3221
source_ids_set = set(source_ids)
3222
# source_ids is the worst possible case we may need to pull.
3223
# now we want to filter source_ids against what we actually
3224
# have in target, but don't try to check for existence where we know
3225
# we do not have a revision as that would be pointless.
3226
target_ids = set(self.target.all_revision_ids())
3227
possibly_present_revisions = target_ids.intersection(source_ids_set)
3228
actually_present_revisions = set(
3229
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3230
required_revisions = source_ids_set.difference(actually_present_revisions)
3231
if revision_id is not None:
3232
# we used get_ancestry to determine source_ids then we are assured all
3233
# revisions referenced are present as they are installed in topological order.
3234
# and the tip revision was validated by get_ancestry.
3235
result_set = required_revisions
3237
# if we just grabbed the possibly available ids, then
3238
# we only have an estimate of whats available and need to validate
3239
# that against the revision records.
3241
self.source._eliminate_revisions_not_present(required_revisions))
3242
return self.source.revision_ids_to_search_result(result_set)
3245
class InterPackRepo(InterSameDataRepository):
3246
"""Optimised code paths between Pack based repositories."""
3249
def _get_repo_format_to_test(self):
3250
from bzrlib.repofmt import pack_repo
3251
return pack_repo.RepositoryFormatKnitPack1()
3254
def is_compatible(source, target):
3255
"""Be compatible with known Pack formats.
3257
We don't test for the stores being of specific types because that
3258
could lead to confusing results, and there is no need to be
3261
from bzrlib.repofmt.pack_repo import RepositoryFormatPack
3263
are_packs = (isinstance(source._format, RepositoryFormatPack) and
3264
isinstance(target._format, RepositoryFormatPack))
3265
except AttributeError:
3267
return are_packs and InterRepository._same_model(source, target)
3270
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3272
"""See InterRepository.fetch()."""
3273
if (len(self.source._fallback_repositories) > 0 or
3274
len(self.target._fallback_repositories) > 0):
3275
# The pack layer is not aware of fallback repositories, so when
3276
# fetching from a stacked repository or into a stacked repository
3277
# we use the generic fetch logic which uses the VersionedFiles
3278
# attributes on repository.
3279
from bzrlib.fetch import RepoFetcher
3280
fetcher = RepoFetcher(self.target, self.source, revision_id,
3281
pb, find_ghosts, fetch_spec=fetch_spec)
3282
if fetch_spec is not None:
3283
if len(list(fetch_spec.heads)) != 1:
3284
raise AssertionError(
3285
"InterPackRepo.fetch doesn't support "
3286
"fetching multiple heads yet.")
3287
revision_id = list(fetch_spec.heads)[0]
3289
if revision_id is None:
3291
# everything to do - use pack logic
3292
# to fetch from all packs to one without
3293
# inventory parsing etc, IFF nothing to be copied is in the target.
3295
source_revision_ids = frozenset(self.source.all_revision_ids())
3296
revision_ids = source_revision_ids - \
3297
frozenset(self.target.get_parent_map(source_revision_ids))
3298
revision_keys = [(revid,) for revid in revision_ids]
3299
index = self.target._pack_collection.revision_index.combined_index
3300
present_revision_ids = set(item[1][0] for item in
3301
index.iter_entries(revision_keys))
3302
revision_ids = set(revision_ids) - present_revision_ids
3303
# implementing the TODO will involve:
3304
# - detecting when all of a pack is selected
3305
# - avoiding as much as possible pre-selection, so the
3306
# more-core routines such as create_pack_from_packs can filter in
3307
# a just-in-time fashion. (though having a HEADS list on a
3308
# repository might make this a lot easier, because we could
3309
# sensibly detect 'new revisions' without doing a full index scan.
3310
elif _mod_revision.is_null(revision_id):
3315
revision_ids = self.search_missing_revision_ids(revision_id,
3316
find_ghosts=find_ghosts).get_keys()
3317
except errors.NoSuchRevision:
3318
raise errors.InstallFailed([revision_id])
3319
if len(revision_ids) == 0:
3321
return self._pack(self.source, self.target, revision_ids)
3323
def _pack(self, source, target, revision_ids):
3324
from bzrlib.repofmt.pack_repo import Packer
3325
packs = source._pack_collection.all_packs()
3326
pack = Packer(self.target._pack_collection, packs, '.fetch',
3327
revision_ids).pack()
3328
if pack is not None:
3329
self.target._pack_collection._save_pack_names()
3330
copied_revs = pack.get_revision_count()
3331
# Trigger an autopack. This may duplicate effort as we've just done
3332
# a pack creation, but for now it is simpler to think about as
3333
# 'upload data, then repack if needed'.
3334
self.target._pack_collection.autopack()
3335
return (copied_revs, [])
3340
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3341
"""See InterRepository.missing_revision_ids().
3343
:param find_ghosts: Find ghosts throughout the ancestry of
3346
if not find_ghosts and revision_id is not None:
3347
return self._walk_to_common_revisions([revision_id])
3348
elif revision_id is not None:
3349
# Find ghosts: search for revisions pointing from one repository to
3350
# the other, and vice versa, anywhere in the history of revision_id.
3351
graph = self.target.get_graph(other_repository=self.source)
3352
searcher = graph._make_breadth_first_searcher([revision_id])
3356
next_revs, ghosts = searcher.next_with_ghosts()
3357
except StopIteration:
3359
if revision_id in ghosts:
3360
raise errors.NoSuchRevision(self.source, revision_id)
3361
found_ids.update(next_revs)
3362
found_ids.update(ghosts)
3363
found_ids = frozenset(found_ids)
3364
# Double query here: should be able to avoid this by changing the
3365
# graph api further.
3366
result_set = found_ids - frozenset(
3367
self.target.get_parent_map(found_ids))
3369
source_ids = self.source.all_revision_ids()
3370
# source_ids is the worst possible case we may need to pull.
3371
# now we want to filter source_ids against what we actually
3372
# have in target, but don't try to check for existence where we know
3373
# we do not have a revision as that would be pointless.
3374
target_ids = set(self.target.all_revision_ids())
3375
result_set = set(source_ids).difference(target_ids)
3376
return self.source.revision_ids_to_search_result(result_set)
3379
class InterDifferingSerializer(InterKnitRepo):
3382
def _get_repo_format_to_test(self):
3386
def is_compatible(source, target):
3387
"""Be compatible with Knit2 source and Knit3 target"""
3388
if source.supports_rich_root() != target.supports_rich_root():
3390
# Ideally, we'd support fetching if the source had no tree references
3391
# even if it supported them...
3392
if (getattr(source, '_format.supports_tree_reference', False) and
3393
not getattr(target, '_format.supports_tree_reference', False)):
3397
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3398
"""Get the best delta and base for this revision.
3400
:return: (basis_id, delta)
3402
possible_trees = [(parent_id, cache[parent_id])
3403
for parent_id in parent_ids
3404
if parent_id in cache]
3405
if len(possible_trees) == 0:
3406
# There either aren't any parents, or the parents aren't in the
3407
# cache, so just use the last converted tree
3408
possible_trees.append((basis_id, cache[basis_id]))
3410
for basis_id, basis_tree in possible_trees:
3411
delta = tree.inventory._make_delta(basis_tree.inventory)
3412
deltas.append((len(delta), basis_id, delta))
3414
return deltas[0][1:]
3416
def _fetch_batch(self, revision_ids, basis_id, cache):
3417
"""Fetch across a few revisions.
3419
:param revision_ids: The revisions to copy
3420
:param basis_id: The revision_id of a tree that must be in cache, used
3421
as a basis for delta when no other base is available
3422
:param cache: A cache of RevisionTrees that we can use.
3423
:return: The revision_id of the last converted tree. The RevisionTree
3424
for it will be in cache
3426
# Walk though all revisions; get inventory deltas, copy referenced
3427
# texts that delta references, insert the delta, revision and
3431
pending_revisions = []
3432
parent_map = self.source.get_parent_map(revision_ids)
3433
for tree in self.source.revision_trees(revision_ids):
3434
current_revision_id = tree.get_revision_id()
3435
parent_ids = parent_map.get(current_revision_id, ())
3436
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3438
# Find text entries that need to be copied
3439
for old_path, new_path, file_id, entry in delta:
3440
if new_path is not None:
3441
if not (new_path or self.target.supports_rich_root()):
3442
# We don't copy the text for the root node unless the
3443
# target supports_rich_root.
3445
text_keys.add((file_id, entry.revision))
3446
revision = self.source.get_revision(current_revision_id)
3447
pending_deltas.append((basis_id, delta,
3448
current_revision_id, revision.parent_ids))
3449
pending_revisions.append(revision)
3450
cache[current_revision_id] = tree
3451
basis_id = current_revision_id
3453
from_texts = self.source.texts
3454
to_texts = self.target.texts
3455
to_texts.insert_record_stream(from_texts.get_record_stream(
3456
text_keys, self.target._format._fetch_order,
3457
not self.target._format._fetch_uses_deltas))
3459
for delta in pending_deltas:
3460
self.target.add_inventory_by_delta(*delta)
3461
# insert signatures and revisions
3462
for revision in pending_revisions:
3464
signature = self.source.get_signature_text(
3465
revision.revision_id)
3466
self.target.add_signature_text(revision.revision_id,
3468
except errors.NoSuchRevision:
3470
self.target.add_revision(revision.revision_id, revision)
3473
def _fetch_all_revisions(self, revision_ids, pb):
3474
"""Fetch everything for the list of revisions.
3476
:param revision_ids: The list of revisions to fetch. Must be in
3478
:param pb: A ProgressBar
3481
basis_id, basis_tree = self._get_basis(revision_ids[0])
3483
cache = lru_cache.LRUCache(100)
3484
cache[basis_id] = basis_tree
3485
del basis_tree # We don't want to hang on to it here
3486
for offset in range(0, len(revision_ids), batch_size):
3487
self.target.start_write_group()
3489
pb.update('Transferring revisions', offset,
3491
batch = revision_ids[offset:offset+batch_size]
3492
basis_id = self._fetch_batch(batch, basis_id, cache)
3494
self.target.abort_write_group()
3497
self.target.commit_write_group()
3498
pb.update('Transferring revisions', len(revision_ids),
3502
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3504
"""See InterRepository.fetch()."""
3505
if fetch_spec is not None:
3506
raise AssertionError("Not implemented yet...")
3507
revision_ids = self.target.search_missing_revision_ids(self.source,
3508
revision_id, find_ghosts=find_ghosts).get_keys()
3509
if not revision_ids:
3511
revision_ids = tsort.topo_sort(
3512
self.source.get_graph().get_parent_map(revision_ids))
3514
my_pb = ui.ui_factory.nested_progress_bar()
3517
symbol_versioning.warn(
3518
symbol_versioning.deprecated_in((1, 14, 0))
3519
% "pb parameter to fetch()")
3522
self._fetch_all_revisions(revision_ids, pb)
3524
if my_pb is not None:
3526
return len(revision_ids), 0
3528
def _get_basis(self, first_revision_id):
3529
"""Get a revision and tree which exists in the target.
3531
This assumes that first_revision_id is selected for transmission
3532
because all other ancestors are already present. If we can't find an
3533
ancestor we fall back to NULL_REVISION since we know that is safe.
3535
:return: (basis_id, basis_tree)
3537
first_rev = self.source.get_revision(first_revision_id)
3539
basis_id = first_rev.parent_ids[0]
3540
# only valid as a basis if the target has it
3541
self.target.get_revision(basis_id)
3542
# Try to get a basis tree - if its a ghost it will hit the
3543
# NoSuchRevision case.
3544
basis_tree = self.source.revision_tree(basis_id)
3545
except (IndexError, errors.NoSuchRevision):
3546
basis_id = _mod_revision.NULL_REVISION
3547
basis_tree = self.source.revision_tree(basis_id)
3548
return basis_id, basis_tree
3551
InterRepository.register_optimiser(InterDifferingSerializer)
3552
InterRepository.register_optimiser(InterSameDataRepository)
3553
InterRepository.register_optimiser(InterWeaveRepo)
3554
InterRepository.register_optimiser(InterKnitRepo)
3555
InterRepository.register_optimiser(InterPackRepo)
3558
class CopyConverter(object):
3559
"""A repository conversion tool which just performs a copy of the content.
3561
This is slow but quite reliable.
3564
def __init__(self, target_format):
3565
"""Create a CopyConverter.
3567
:param target_format: The format the resulting repository should be.
3569
self.target_format = target_format
3571
def convert(self, repo, pb):
3572
"""Perform the conversion of to_convert, giving feedback via pb.
3574
:param to_convert: The disk object to convert.
3575
:param pb: a progress bar to use for progress information.
3580
# this is only useful with metadir layouts - separated repo content.
3581
# trigger an assertion if not such
3582
repo._format.get_format_string()
3583
self.repo_dir = repo.bzrdir
3584
self.step('Moving repository to repository.backup')
3585
self.repo_dir.transport.move('repository', 'repository.backup')
3586
backup_transport = self.repo_dir.transport.clone('repository.backup')
3587
repo._format.check_conversion_target(self.target_format)
3588
self.source_repo = repo._format.open(self.repo_dir,
3590
_override_transport=backup_transport)
3591
self.step('Creating new repository')
3592
converted = self.target_format.initialize(self.repo_dir,
3593
self.source_repo.is_shared())
3594
converted.lock_write()
3596
self.step('Copying content into repository.')
3597
self.source_repo.copy_content_into(converted)
3600
self.step('Deleting old repository content.')
3601
self.repo_dir.transport.delete_tree('repository.backup')
3602
self.pb.note('repository converted')
3604
def step(self, message):
3605
"""Update the pb by a step."""
3607
self.pb.update(message, self.count, self.total)
3619
def _unescaper(match, _map=_unescape_map):
3620
code = match.group(1)
3624
if not code.startswith('#'):
3626
return unichr(int(code[1:])).encode('utf8')
3632
def _unescape_xml(data):
3633
"""Unescape predefined XML entities in a string of data."""
3635
if _unescape_re is None:
3636
_unescape_re = re.compile('\&([^;]*);')
3637
return _unescape_re.sub(_unescaper, data)
3640
class _VersionedFileChecker(object):
3642
def __init__(self, repository, text_key_references=None):
3643
self.repository = repository
3644
self.text_index = self.repository._generate_text_key_index(
3645
text_key_references=text_key_references)
3647
def calculate_file_version_parents(self, text_key):
3648
"""Calculate the correct parents for a file version according to
3651
parent_keys = self.text_index[text_key]
3652
if parent_keys == [_mod_revision.NULL_REVISION]:
3654
return tuple(parent_keys)
3656
def check_file_version_parents(self, texts, progress_bar=None):
3657
"""Check the parents stored in a versioned file are correct.
3659
It also detects file versions that are not referenced by their
3660
corresponding revision's inventory.
3662
:returns: A tuple of (wrong_parents, dangling_file_versions).
3663
wrong_parents is a dict mapping {revision_id: (stored_parents,
3664
correct_parents)} for each revision_id where the stored parents
3665
are not correct. dangling_file_versions is a set of (file_id,
3666
revision_id) tuples for versions that are present in this versioned
3667
file, but not used by the corresponding inventory.
3670
self.file_ids = set([file_id for file_id, _ in
3671
self.text_index.iterkeys()])
3672
# text keys is now grouped by file_id
3673
n_weaves = len(self.file_ids)
3674
files_in_revisions = {}
3675
revisions_of_files = {}
3676
n_versions = len(self.text_index)
3677
progress_bar.update('loading text store', 0, n_versions)
3678
parent_map = self.repository.texts.get_parent_map(self.text_index)
3679
# On unlistable transports this could well be empty/error...
3680
text_keys = self.repository.texts.keys()
3681
unused_keys = frozenset(text_keys) - set(self.text_index)
3682
for num, key in enumerate(self.text_index.iterkeys()):
3683
if progress_bar is not None:
3684
progress_bar.update('checking text graph', num, n_versions)
3685
correct_parents = self.calculate_file_version_parents(key)
3687
knit_parents = parent_map[key]
3688
except errors.RevisionNotPresent:
3691
if correct_parents != knit_parents:
3692
wrong_parents[key] = (knit_parents, correct_parents)
3693
return wrong_parents, unused_keys
3696
def _old_get_graph(repository, revision_id):
3697
"""DO NOT USE. That is all. I'm serious."""
3698
graph = repository.get_graph()
3699
revision_graph = dict(((key, value) for key, value in
3700
graph.iter_ancestry([revision_id]) if value is not None))
3701
return _strip_NULL_ghosts(revision_graph)
3704
def _strip_NULL_ghosts(revision_graph):
3705
"""Also don't use this. more compatibility code for unmigrated clients."""
3706
# Filter ghosts, and null:
3707
if _mod_revision.NULL_REVISION in revision_graph:
3708
del revision_graph[_mod_revision.NULL_REVISION]
3709
for key, parents in revision_graph.items():
3710
revision_graph[key] = tuple(parent for parent in parents if parent
3712
return revision_graph
3715
class StreamSink(object):
3716
"""An object that can insert a stream into a repository.
3718
This interface handles the complexity of reserialising inventories and
3719
revisions from different formats, and allows unidirectional insertion into
3720
stacked repositories without looking for the missing basis parents
3724
def __init__(self, target_repo):
3725
self.target_repo = target_repo
3727
def insert_stream(self, stream, src_format, resume_tokens):
3728
"""Insert a stream's content into the target repository.
3730
:param src_format: a bzr repository format.
3732
:return: a list of resume tokens and an iterable of keys additional
3733
items required before the insertion can be completed.
3735
self.target_repo.lock_write()
3738
self.target_repo.resume_write_group(resume_tokens)
3740
self.target_repo.start_write_group()
3742
# locked_insert_stream performs a commit|suspend.
3743
return self._locked_insert_stream(stream, src_format)
3745
self.target_repo.abort_write_group(suppress_errors=True)
3748
self.target_repo.unlock()
3750
def _locked_insert_stream(self, stream, src_format):
3751
to_serializer = self.target_repo._format._serializer
3752
src_serializer = src_format._serializer
3753
if to_serializer == src_serializer:
3754
# If serializers match and the target is a pack repository, set the
3755
# write cache size on the new pack. This avoids poor performance
3756
# on transports where append is unbuffered (such as
3757
# RemoteTransport). This is safe to do because nothing should read
3758
# back from the target repository while a stream with matching
3759
# serialization is being inserted.
3760
# The exception is that a delta record from the source that should
3761
# be a fulltext may need to be expanded by the target (see
3762
# test_fetch_revisions_with_deltas_into_pack); but we take care to
3763
# explicitly flush any buffered writes first in that rare case.
3765
new_pack = self.target_repo._pack_collection._new_pack
3766
except AttributeError:
3767
# Not a pack repository
3770
new_pack.set_write_cache_size(1024*1024)
3771
for substream_type, substream in stream:
3772
if substream_type == 'texts':
3773
self.target_repo.texts.insert_record_stream(substream)
3774
elif substream_type == 'inventories':
3775
if src_serializer == to_serializer:
3776
self.target_repo.inventories.insert_record_stream(
3779
self._extract_and_insert_inventories(
3780
substream, src_serializer)
3781
elif substream_type == 'revisions':
3782
# This may fallback to extract-and-insert more often than
3783
# required if the serializers are different only in terms of
3785
if src_serializer == to_serializer:
3786
self.target_repo.revisions.insert_record_stream(
3789
self._extract_and_insert_revisions(substream,
3791
elif substream_type == 'signatures':
3792
self.target_repo.signatures.insert_record_stream(substream)
3794
raise AssertionError('kaboom! %s' % (substream_type,))
3796
missing_keys = set()
3797
for prefix, versioned_file in (
3798
('texts', self.target_repo.texts),
3799
('inventories', self.target_repo.inventories),
3800
('revisions', self.target_repo.revisions),
3801
('signatures', self.target_repo.signatures),
3803
missing_keys.update((prefix,) + key for key in
3804
versioned_file.get_missing_compression_parent_keys())
3805
except NotImplementedError:
3806
# cannot even attempt suspending, and missing would have failed
3807
# during stream insertion.
3808
missing_keys = set()
3811
# suspend the write group and tell the caller what we is
3812
# missing. We know we can suspend or else we would not have
3813
# entered this code path. (All repositories that can handle
3814
# missing keys can handle suspending a write group).
3815
write_group_tokens = self.target_repo.suspend_write_group()
3816
return write_group_tokens, missing_keys
3817
self.target_repo.commit_write_group()
3820
def _extract_and_insert_inventories(self, substream, serializer):
3821
"""Generate a new inventory versionedfile in target, converting data.
3823
The inventory is retrieved from the source, (deserializing it), and
3824
stored in the target (reserializing it in a different format).
3826
for record in substream:
3827
bytes = record.get_bytes_as('fulltext')
3828
revision_id = record.key[0]
3829
inv = serializer.read_inventory_from_string(bytes, revision_id)
3830
parents = [key[0] for key in record.parents]
3831
self.target_repo.add_inventory(revision_id, inv, parents)
3833
def _extract_and_insert_revisions(self, substream, serializer):
3834
for record in substream:
3835
bytes = record.get_bytes_as('fulltext')
3836
revision_id = record.key[0]
3837
rev = serializer.read_revision_from_string(bytes)
3838
if rev.revision_id != revision_id:
3839
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
3840
self.target_repo.add_revision(revision_id, rev)
3843
if self.target_repo._format._fetch_reconcile:
3844
self.target_repo.reconcile()
3847
class StreamSource(object):
3848
"""A source of a stream for fetching between repositories."""
3850
def __init__(self, from_repository, to_format):
3851
"""Create a StreamSource streaming from from_repository."""
3852
self.from_repository = from_repository
3853
self.to_format = to_format
3855
def delta_on_metadata(self):
3856
"""Return True if delta's are permitted on metadata streams.
3858
That is on revisions and signatures.
3860
src_serializer = self.from_repository._format._serializer
3861
target_serializer = self.to_format._serializer
3862
return (self.to_format._fetch_uses_deltas and
3863
src_serializer == target_serializer)
3865
def _fetch_revision_texts(self, revs):
3866
# fetch signatures first and then the revision texts
3867
# may need to be a InterRevisionStore call here.
3868
from_sf = self.from_repository.signatures
3869
# A missing signature is just skipped.
3870
keys = [(rev_id,) for rev_id in revs]
3871
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
3873
self.to_format._fetch_order,
3874
not self.to_format._fetch_uses_deltas))
3875
# If a revision has a delta, this is actually expanded inside the
3876
# insert_record_stream code now, which is an alternate fix for
3878
from_rf = self.from_repository.revisions
3879
revisions = from_rf.get_record_stream(
3881
self.to_format._fetch_order,
3882
not self.delta_on_metadata())
3883
return [('signatures', signatures), ('revisions', revisions)]
3885
def _generate_root_texts(self, revs):
3886
"""This will be called by __fetch between fetching weave texts and
3887
fetching the inventory weave.
3889
Subclasses should override this if they need to generate root texts
3890
after fetching weave texts.
3892
if self._rich_root_upgrade():
3894
return bzrlib.fetch.Inter1and2Helper(
3895
self.from_repository).generate_root_texts(revs)
3899
def get_stream(self, search):
3901
revs = search.get_keys()
3902
graph = self.from_repository.get_graph()
3903
revs = list(graph.iter_topo_order(revs))
3904
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
3906
for knit_kind, file_id, revisions in data_to_fetch:
3907
if knit_kind != phase:
3909
# Make a new progress bar for this phase
3910
if knit_kind == "file":
3911
# Accumulate file texts
3912
text_keys.extend([(file_id, revision) for revision in
3914
elif knit_kind == "inventory":
3915
# Now copy the file texts.
3916
from_texts = self.from_repository.texts
3917
yield ('texts', from_texts.get_record_stream(
3918
text_keys, self.to_format._fetch_order,
3919
not self.to_format._fetch_uses_deltas))
3920
# Cause an error if a text occurs after we have done the
3923
# Before we process the inventory we generate the root
3924
# texts (if necessary) so that the inventories references
3926
for _ in self._generate_root_texts(revs):
3928
# NB: This currently reopens the inventory weave in source;
3929
# using a single stream interface instead would avoid this.
3930
from_weave = self.from_repository.inventories
3931
# we fetch only the referenced inventories because we do not
3932
# know for unselected inventories whether all their required
3933
# texts are present in the other repository - it could be
3935
yield ('inventories', from_weave.get_record_stream(
3936
[(rev_id,) for rev_id in revs],
3937
self.inventory_fetch_order(),
3938
not self.delta_on_metadata()))
3939
elif knit_kind == "signatures":
3940
# Nothing to do here; this will be taken care of when
3941
# _fetch_revision_texts happens.
3943
elif knit_kind == "revisions":
3944
for record in self._fetch_revision_texts(revs):
3947
raise AssertionError("Unknown knit kind %r" % knit_kind)
3949
def get_stream_for_missing_keys(self, missing_keys):
3950
# missing keys can only occur when we are byte copying and not
3951
# translating (because translation means we don't send
3952
# unreconstructable deltas ever).
3954
keys['texts'] = set()
3955
keys['revisions'] = set()
3956
keys['inventories'] = set()
3957
keys['signatures'] = set()
3958
for key in missing_keys:
3959
keys[key[0]].add(key[1:])
3960
if len(keys['revisions']):
3961
# If we allowed copying revisions at this point, we could end up
3962
# copying a revision without copying its required texts: a
3963
# violation of the requirements for repository integrity.
3964
raise AssertionError(
3965
'cannot copy revisions to fill in missing deltas %s' % (
3966
keys['revisions'],))
3967
for substream_kind, keys in keys.iteritems():
3968
vf = getattr(self.from_repository, substream_kind)
3969
# Ask for full texts always so that we don't need more round trips
3970
# after this stream.
3971
stream = vf.get_record_stream(keys,
3972
self.to_format._fetch_order, True)
3973
yield substream_kind, stream
3975
def inventory_fetch_order(self):
3976
if self._rich_root_upgrade():
3977
return 'topological'
3979
return self.to_format._fetch_order
3981
def _rich_root_upgrade(self):
3982
return (not self.from_repository._format.rich_root_data and
3983
self.to_format.rich_root_data)
added
removed
bzrlib/repository.py
