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(), """
36
revision as _mod_revision,
42
from bzrlib.bundle import serializer
43
from bzrlib.revisiontree import RevisionTree
44
from bzrlib.store.versioned import VersionedFileStore
45
from bzrlib.testament import Testament
48
from bzrlib.decorators import needs_read_lock, needs_write_lock
49
from bzrlib.inter import InterObject
50
from bzrlib.inventory import (
56
from bzrlib import registry
57
from bzrlib.symbol_versioning import (
60
from bzrlib.trace import (
61
log_exception_quietly, note, mutter, mutter_callsite, warning)
64
# Old formats display a warning, but only once
65
_deprecation_warning_done = False
68
class CommitBuilder(object):
69
"""Provides an interface to build up a commit.
71
This allows describing a tree to be committed without needing to
72
know the internals of the format of the repository.
75
# all clients should supply tree roots.
76
record_root_entry = True
77
# the default CommitBuilder does not manage trees whose root is versioned.
78
_versioned_root = False
80
def __init__(self, repository, parents, config, timestamp=None,
81
timezone=None, committer=None, revprops=None,
83
"""Initiate a CommitBuilder.
85
:param repository: Repository to commit to.
86
:param parents: Revision ids of the parents of the new revision.
87
:param config: Configuration to use.
88
:param timestamp: Optional timestamp recorded for commit.
89
:param timezone: Optional timezone for timestamp.
90
:param committer: Optional committer to set for commit.
91
:param revprops: Optional dictionary of revision properties.
92
:param revision_id: Optional revision id.
97
self._committer = self._config.username()
99
self._committer = committer
101
self.new_inventory = Inventory(None)
102
self._new_revision_id = revision_id
103
self.parents = parents
104
self.repository = repository
107
if revprops is not None:
108
self._validate_revprops(revprops)
109
self._revprops.update(revprops)
111
if timestamp is None:
112
timestamp = time.time()
113
# Restrict resolution to 1ms
114
self._timestamp = round(timestamp, 3)
117
self._timezone = osutils.local_time_offset()
119
self._timezone = int(timezone)
121
self._generate_revision_if_needed()
122
self.__heads = graph.HeadsCache(repository.get_graph()).heads
123
self._basis_delta = []
124
# API compatibility, older code that used CommitBuilder did not call
125
# .record_delete(), which means the delta that is computed would not be
126
# valid. Callers that will call record_delete() should call
127
# .will_record_deletes() to indicate that.
128
self._recording_deletes = False
129
# memo'd check for no-op commits.
130
self._any_changes = False
132
def any_changes(self):
133
"""Return True if any entries were changed.
135
This includes merge-only changes. It is the core for the --unchanged
138
:return: True if any changes have occured.
140
return self._any_changes
142
def _validate_unicode_text(self, text, context):
143
"""Verify things like commit messages don't have bogus characters."""
145
raise ValueError('Invalid value for %s: %r' % (context, text))
147
def _validate_revprops(self, revprops):
148
for key, value in revprops.iteritems():
149
# We know that the XML serializers do not round trip '\r'
150
# correctly, so refuse to accept them
151
if not isinstance(value, basestring):
152
raise ValueError('revision property (%s) is not a valid'
153
' (unicode) string: %r' % (key, value))
154
self._validate_unicode_text(value,
155
'revision property (%s)' % (key,))
157
def commit(self, message):
158
"""Make the actual commit.
160
:return: The revision id of the recorded revision.
162
self._validate_unicode_text(message, 'commit message')
163
rev = _mod_revision.Revision(
164
timestamp=self._timestamp,
165
timezone=self._timezone,
166
committer=self._committer,
168
inventory_sha1=self.inv_sha1,
169
revision_id=self._new_revision_id,
170
properties=self._revprops)
171
rev.parent_ids = self.parents
172
self.repository.add_revision(self._new_revision_id, rev,
173
self.new_inventory, self._config)
174
self.repository.commit_write_group()
175
return self._new_revision_id
178
"""Abort the commit that is being built.
180
self.repository.abort_write_group()
182
def revision_tree(self):
183
"""Return the tree that was just committed.
185
After calling commit() this can be called to get a RevisionTree
186
representing the newly committed tree. This is preferred to
187
calling Repository.revision_tree() because that may require
188
deserializing the inventory, while we already have a copy in
191
if self.new_inventory is None:
192
self.new_inventory = self.repository.get_inventory(
193
self._new_revision_id)
194
return RevisionTree(self.repository, self.new_inventory,
195
self._new_revision_id)
197
def finish_inventory(self):
198
"""Tell the builder that the inventory is finished.
200
:return: The inventory id in the repository, which can be used with
201
repository.get_inventory.
203
if self.new_inventory is None:
204
# an inventory delta was accumulated without creating a new
206
basis_id = self.basis_delta_revision
207
self.inv_sha1 = self.repository.add_inventory_by_delta(
208
basis_id, self._basis_delta, self._new_revision_id,
211
if self.new_inventory.root is None:
212
raise AssertionError('Root entry should be supplied to'
213
' record_entry_contents, as of bzr 0.10.')
214
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
215
self.new_inventory.revision_id = self._new_revision_id
216
self.inv_sha1 = self.repository.add_inventory(
217
self._new_revision_id,
221
return self._new_revision_id
223
def _gen_revision_id(self):
224
"""Return new revision-id."""
225
return generate_ids.gen_revision_id(self._config.username(),
228
def _generate_revision_if_needed(self):
229
"""Create a revision id if None was supplied.
231
If the repository can not support user-specified revision ids
232
they should override this function and raise CannotSetRevisionId
233
if _new_revision_id is not None.
235
:raises: CannotSetRevisionId
237
if self._new_revision_id is None:
238
self._new_revision_id = self._gen_revision_id()
239
self.random_revid = True
241
self.random_revid = False
243
def _heads(self, file_id, revision_ids):
244
"""Calculate the graph heads for revision_ids in the graph of file_id.
246
This can use either a per-file graph or a global revision graph as we
247
have an identity relationship between the two graphs.
249
return self.__heads(revision_ids)
251
def _check_root(self, ie, parent_invs, tree):
252
"""Helper for record_entry_contents.
254
:param ie: An entry being added.
255
:param parent_invs: The inventories of the parent revisions of the
257
:param tree: The tree that is being committed.
259
# In this revision format, root entries have no knit or weave When
260
# serializing out to disk and back in root.revision is always
262
ie.revision = self._new_revision_id
264
def _require_root_change(self, tree):
265
"""Enforce an appropriate root object change.
267
This is called once when record_iter_changes is called, if and only if
268
the root was not in the delta calculated by record_iter_changes.
270
:param tree: The tree which is being committed.
272
# NB: if there are no parents then this method is not called, so no
273
# need to guard on parents having length.
274
entry = entry_factory['directory'](tree.path2id(''), '',
276
entry.revision = self._new_revision_id
277
self._basis_delta.append(('', '', entry.file_id, entry))
279
def _get_delta(self, ie, basis_inv, path):
280
"""Get a delta against the basis inventory for ie."""
281
if ie.file_id not in basis_inv:
283
result = (None, path, ie.file_id, ie)
284
self._basis_delta.append(result)
286
elif ie != basis_inv[ie.file_id]:
288
# TODO: avoid tis id2path call.
289
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
290
self._basis_delta.append(result)
296
def get_basis_delta(self):
297
"""Return the complete inventory delta versus the basis inventory.
299
This has been built up with the calls to record_delete and
300
record_entry_contents. The client must have already called
301
will_record_deletes() to indicate that they will be generating a
304
:return: An inventory delta, suitable for use with apply_delta, or
305
Repository.add_inventory_by_delta, etc.
307
if not self._recording_deletes:
308
raise AssertionError("recording deletes not activated.")
309
return self._basis_delta
311
def record_delete(self, path, file_id):
312
"""Record that a delete occured against a basis tree.
314
This is an optional API - when used it adds items to the basis_delta
315
being accumulated by the commit builder. It cannot be called unless the
316
method will_record_deletes() has been called to inform the builder that
317
a delta is being supplied.
319
:param path: The path of the thing deleted.
320
:param file_id: The file id that was deleted.
322
if not self._recording_deletes:
323
raise AssertionError("recording deletes not activated.")
324
delta = (path, None, file_id, None)
325
self._basis_delta.append(delta)
326
self._any_changes = True
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. The iterator must not include any items with
562
a current kind of None - missing items must be either filtered out
563
or errored-on beefore record_iter_changes sees the item.
564
:param _entry_factory: Private method to bind entry_factory locally for
566
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
569
# Create an inventory delta based on deltas between all the parents and
570
# deltas between all the parent inventories. We use inventory delta's
571
# between the inventory objects because iter_changes masks
572
# last-changed-field only changes.
574
# file_id -> change map, change is fileid, paths, changed, versioneds,
575
# parents, names, kinds, executables
577
# {file_id -> revision_id -> inventory entry, for entries in parent
578
# trees that are not parents[0]
582
revtrees = list(self.repository.revision_trees(self.parents))
583
except errors.NoSuchRevision:
584
# one or more ghosts, slow path.
586
for revision_id in self.parents:
588
revtrees.append(self.repository.revision_tree(revision_id))
589
except errors.NoSuchRevision:
591
basis_revision_id = _mod_revision.NULL_REVISION
593
revtrees.append(self.repository.revision_tree(
594
_mod_revision.NULL_REVISION))
595
# The basis inventory from a repository
597
basis_inv = revtrees[0].inventory
599
basis_inv = self.repository.revision_tree(
600
_mod_revision.NULL_REVISION).inventory
601
if len(self.parents) > 0:
602
if basis_revision_id != self.parents[0] and not ghost_basis:
604
"arbitrary basis parents not yet supported with merges")
605
for revtree in revtrees[1:]:
606
for change in revtree.inventory._make_delta(basis_inv):
607
if change[1] is None:
608
# Not present in this parent.
610
if change[2] not in merged_ids:
611
if change[0] is not None:
612
basis_entry = basis_inv[change[2]]
613
merged_ids[change[2]] = [
615
basis_entry.revision,
618
parent_entries[change[2]] = {
620
basis_entry.revision:basis_entry,
622
change[3].revision:change[3],
625
merged_ids[change[2]] = [change[3].revision]
626
parent_entries[change[2]] = {change[3].revision:change[3]}
628
merged_ids[change[2]].append(change[3].revision)
629
parent_entries[change[2]][change[3].revision] = change[3]
632
# Setup the changes from the tree:
633
# changes maps file_id -> (change, [parent revision_ids])
635
for change in iter_changes:
636
# This probably looks up in basis_inv way to much.
637
if change[1][0] is not None:
638
head_candidate = [basis_inv[change[0]].revision]
641
changes[change[0]] = change, merged_ids.get(change[0],
643
unchanged_merged = set(merged_ids) - set(changes)
644
# Extend the changes dict with synthetic changes to record merges of
646
for file_id in unchanged_merged:
647
# Record a merged version of these items that did not change vs the
648
# basis. This can be either identical parallel changes, or a revert
649
# of a specific file after a merge. The recorded content will be
650
# that of the current tree (which is the same as the basis), but
651
# the per-file graph will reflect a merge.
652
# NB:XXX: We are reconstructing path information we had, this
653
# should be preserved instead.
654
# inv delta change: (file_id, (path_in_source, path_in_target),
655
# changed_content, versioned, parent, name, kind,
658
basis_entry = basis_inv[file_id]
659
except errors.NoSuchId:
660
# a change from basis->some_parents but file_id isn't in basis
661
# so was new in the merge, which means it must have changed
662
# from basis -> current, and as it hasn't the add was reverted
663
# by the user. So we discard this change.
667
(basis_inv.id2path(file_id), tree.id2path(file_id)),
669
(basis_entry.parent_id, basis_entry.parent_id),
670
(basis_entry.name, basis_entry.name),
671
(basis_entry.kind, basis_entry.kind),
672
(basis_entry.executable, basis_entry.executable))
673
changes[file_id] = (change, merged_ids[file_id])
674
# changes contains tuples with the change and a set of inventory
675
# candidates for the file.
677
# old_path, new_path, file_id, new_inventory_entry
678
seen_root = False # Is the root in the basis delta?
679
inv_delta = self._basis_delta
680
modified_rev = self._new_revision_id
681
for change, head_candidates in changes.values():
682
if change[3][1]: # versioned in target.
683
# Several things may be happening here:
684
# We may have a fork in the per-file graph
685
# - record a change with the content from tree
686
# We may have a change against < all trees
687
# - carry over the tree that hasn't changed
688
# We may have a change against all trees
689
# - record the change with the content from tree
692
entry = _entry_factory[kind](file_id, change[5][1],
694
head_set = self._heads(change[0], set(head_candidates))
697
for head_candidate in head_candidates:
698
if head_candidate in head_set:
699
heads.append(head_candidate)
700
head_set.remove(head_candidate)
703
# Could be a carry-over situation:
704
parent_entry_revs = parent_entries.get(file_id, None)
705
if parent_entry_revs:
706
parent_entry = parent_entry_revs.get(heads[0], None)
709
if parent_entry is None:
710
# The parent iter_changes was called against is the one
711
# that is the per-file head, so any change is relevant
712
# iter_changes is valid.
713
carry_over_possible = False
715
# could be a carry over situation
716
# A change against the basis may just indicate a merge,
717
# we need to check the content against the source of the
718
# merge to determine if it was changed after the merge
720
if (parent_entry.kind != entry.kind or
721
parent_entry.parent_id != entry.parent_id or
722
parent_entry.name != entry.name):
723
# Metadata common to all entries has changed
724
# against per-file parent
725
carry_over_possible = False
727
carry_over_possible = True
728
# per-type checks for changes against the parent_entry
731
# Cannot be a carry-over situation
732
carry_over_possible = False
733
# Populate the entry in the delta
735
# XXX: There is still a small race here: If someone reverts the content of a file
736
# after iter_changes examines and decides it has changed,
737
# we will unconditionally record a new version even if some
738
# other process reverts it while commit is running (with
739
# the revert happening after iter_changes did it's
742
entry.executable = True
744
entry.executable = False
745
if (carry_over_possible and
746
parent_entry.executable == entry.executable):
747
# Check the file length, content hash after reading
749
nostore_sha = parent_entry.text_sha1
752
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
754
lines = file_obj.readlines()
758
entry.text_sha1, entry.text_size = self._add_text_to_weave(
759
file_id, lines, heads, nostore_sha)
760
yield file_id, change[1][1], (entry.text_sha1, stat_value)
761
except errors.ExistingContent:
762
# No content change against a carry_over parent
763
# Perhaps this should also yield a fs hash update?
765
entry.text_size = parent_entry.text_size
766
entry.text_sha1 = parent_entry.text_sha1
767
elif kind == 'symlink':
769
entry.symlink_target = tree.get_symlink_target(file_id)
770
if (carry_over_possible and
771
parent_entry.symlink_target == entry.symlink_target):
774
self._add_text_to_weave(change[0], [], heads, None)
775
elif kind == 'directory':
776
if carry_over_possible:
779
# Nothing to set on the entry.
780
# XXX: split into the Root and nonRoot versions.
781
if change[1][1] != '' or self.repository.supports_rich_root():
782
self._add_text_to_weave(change[0], [], heads, None)
783
elif kind == 'tree-reference':
784
if not self.repository._format.supports_tree_reference:
785
# This isn't quite sane as an error, but we shouldn't
786
# ever see this code path in practice: tree's don't
787
# permit references when the repo doesn't support tree
789
raise errors.UnsupportedOperation(tree.add_reference,
791
entry.reference_revision = \
792
tree.get_reference_revision(change[0])
793
if (carry_over_possible and
794
parent_entry.reference_revision == reference_revision):
797
self._add_text_to_weave(change[0], [], heads, None)
799
raise AssertionError('unknown kind %r' % kind)
801
entry.revision = modified_rev
803
entry.revision = parent_entry.revision
806
new_path = change[1][1]
807
inv_delta.append((change[1][0], new_path, change[0], entry))
810
self.new_inventory = None
812
self._any_changes = True
814
# housekeeping root entry changes do not affect no-change commits.
815
self._require_root_change(tree)
816
self.basis_delta_revision = basis_revision_id
818
def _add_text_to_weave(self, file_id, new_lines, parents, nostore_sha):
819
# Note: as we read the content directly from the tree, we know its not
820
# been turned into unicode or badly split - but a broken tree
821
# implementation could give us bad output from readlines() so this is
822
# not a guarantee of safety. What would be better is always checking
823
# the content during test suite execution. RBC 20070912
824
parent_keys = tuple((file_id, parent) for parent in parents)
825
return self.repository.texts.add_lines(
826
(file_id, self._new_revision_id), parent_keys, new_lines,
827
nostore_sha=nostore_sha, random_id=self.random_revid,
828
check_content=False)[0:2]
831
class RootCommitBuilder(CommitBuilder):
832
"""This commitbuilder actually records the root id"""
834
# the root entry gets versioned properly by this builder.
835
_versioned_root = True
837
def _check_root(self, ie, parent_invs, tree):
838
"""Helper for record_entry_contents.
840
:param ie: An entry being added.
841
:param parent_invs: The inventories of the parent revisions of the
843
:param tree: The tree that is being committed.
846
def _require_root_change(self, tree):
847
"""Enforce an appropriate root object change.
849
This is called once when record_iter_changes is called, if and only if
850
the root was not in the delta calculated by record_iter_changes.
852
:param tree: The tree which is being committed.
854
# versioned roots do not change unless the tree found a change.
857
######################################################################
860
class Repository(object):
861
"""Repository holding history for one or more branches.
863
The repository holds and retrieves historical information including
864
revisions and file history. It's normally accessed only by the Branch,
865
which views a particular line of development through that history.
867
The Repository builds on top of some byte storage facilies (the revisions,
868
signatures, inventories and texts attributes) and a Transport, which
869
respectively provide byte storage and a means to access the (possibly
872
The byte storage facilities are addressed via tuples, which we refer to
873
as 'keys' throughout the code base. Revision_keys, inventory_keys and
874
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
875
(file_id, revision_id). We use this interface because it allows low
876
friction with the underlying code that implements disk indices, network
877
encoding and other parts of bzrlib.
879
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
880
the serialised revisions for the repository. This can be used to obtain
881
revision graph information or to access raw serialised revisions.
882
The result of trying to insert data into the repository via this store
883
is undefined: it should be considered read-only except for implementors
885
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
886
the serialised signatures for the repository. This can be used to
887
obtain access to raw serialised signatures. The result of trying to
888
insert data into the repository via this store is undefined: it should
889
be considered read-only except for implementors of repositories.
890
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
891
the serialised inventories for the repository. This can be used to
892
obtain unserialised inventories. The result of trying to insert data
893
into the repository via this store is undefined: it should be
894
considered read-only except for implementors of repositories.
895
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
896
texts of files and directories for the repository. This can be used to
897
obtain file texts or file graphs. Note that Repository.iter_file_bytes
898
is usually a better interface for accessing file texts.
899
The result of trying to insert data into the repository via this store
900
is undefined: it should be considered read-only except for implementors
902
:ivar _transport: Transport for file access to repository, typically
903
pointing to .bzr/repository.
906
# What class to use for a CommitBuilder. Often its simpler to change this
907
# in a Repository class subclass rather than to override
908
# get_commit_builder.
909
_commit_builder_class = CommitBuilder
910
# The search regex used by xml based repositories to determine what things
911
# where changed in a single commit.
912
_file_ids_altered_regex = lazy_regex.lazy_compile(
913
r'file_id="(?P<file_id>[^"]+)"'
914
r'.* revision="(?P<revision_id>[^"]+)"'
917
def abort_write_group(self, suppress_errors=False):
918
"""Commit the contents accrued within the current write group.
920
:param suppress_errors: if true, abort_write_group will catch and log
921
unexpected errors that happen during the abort, rather than
922
allowing them to propagate. Defaults to False.
924
:seealso: start_write_group.
926
if self._write_group is not self.get_transaction():
927
# has an unlock or relock occured ?
928
raise errors.BzrError('mismatched lock context and write group.')
930
self._abort_write_group()
931
except Exception, exc:
932
self._write_group = None
933
if not suppress_errors:
935
mutter('abort_write_group failed')
936
log_exception_quietly()
937
note('bzr: ERROR (ignored): %s', exc)
938
self._write_group = None
940
def _abort_write_group(self):
941
"""Template method for per-repository write group cleanup.
943
This is called during abort before the write group is considered to be
944
finished and should cleanup any internal state accrued during the write
945
group. There is no requirement that data handed to the repository be
946
*not* made available - this is not a rollback - but neither should any
947
attempt be made to ensure that data added is fully commited. Abort is
948
invoked when an error has occured so futher disk or network operations
949
may not be possible or may error and if possible should not be
953
def add_fallback_repository(self, repository):
954
"""Add a repository to use for looking up data not held locally.
956
:param repository: A repository.
958
if not self._format.supports_external_lookups:
959
raise errors.UnstackableRepositoryFormat(self._format, self.base)
960
self._check_fallback_repository(repository)
961
self._fallback_repositories.append(repository)
962
self.texts.add_fallback_versioned_files(repository.texts)
963
self.inventories.add_fallback_versioned_files(repository.inventories)
964
self.revisions.add_fallback_versioned_files(repository.revisions)
965
self.signatures.add_fallback_versioned_files(repository.signatures)
967
def _check_fallback_repository(self, repository):
968
"""Check that this repository can fallback to repository safely.
970
Raise an error if not.
972
:param repository: A repository to fallback to.
974
return InterRepository._assert_same_model(self, repository)
976
def add_inventory(self, revision_id, inv, parents):
977
"""Add the inventory inv to the repository as revision_id.
979
:param parents: The revision ids of the parents that revision_id
980
is known to have and are in the repository already.
982
:returns: The validator(which is a sha1 digest, though what is sha'd is
983
repository format specific) of the serialized inventory.
985
if not self.is_in_write_group():
986
raise AssertionError("%r not in write group" % (self,))
987
_mod_revision.check_not_reserved_id(revision_id)
988
if not (inv.revision_id is None or inv.revision_id == revision_id):
989
raise AssertionError(
990
"Mismatch between inventory revision"
991
" id and insertion revid (%r, %r)"
992
% (inv.revision_id, revision_id))
994
raise AssertionError()
995
inv_lines = self._serialise_inventory_to_lines(inv)
996
return self._inventory_add_lines(revision_id, parents,
997
inv_lines, check_content=False)
999
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1001
"""Add a new inventory expressed as a delta against another revision.
1003
:param basis_revision_id: The inventory id the delta was created
1004
against. (This does not have to be a direct parent.)
1005
:param delta: The inventory delta (see Inventory.apply_delta for
1007
:param new_revision_id: The revision id that the inventory is being
1009
:param parents: The revision ids of the parents that revision_id is
1010
known to have and are in the repository already. These are supplied
1011
for repositories that depend on the inventory graph for revision
1012
graph access, as well as for those that pun ancestry with delta
1015
:returns: (validator, new_inv)
1016
The validator(which is a sha1 digest, though what is sha'd is
1017
repository format specific) of the serialized inventory, and the
1018
resulting inventory.
1020
if not self.is_in_write_group():
1021
raise AssertionError("%r not in write group" % (self,))
1022
_mod_revision.check_not_reserved_id(new_revision_id)
1023
basis_tree = self.revision_tree(basis_revision_id)
1024
basis_tree.lock_read()
1026
# Note that this mutates the inventory of basis_tree, which not all
1027
# inventory implementations may support: A better idiom would be to
1028
# return a new inventory, but as there is no revision tree cache in
1029
# repository this is safe for now - RBC 20081013
1030
basis_inv = basis_tree.inventory
1031
basis_inv.apply_delta(delta)
1032
basis_inv.revision_id = new_revision_id
1033
return (self.add_inventory(new_revision_id, basis_inv, parents),
1038
def _inventory_add_lines(self, revision_id, parents, lines,
1039
check_content=True):
1040
"""Store lines in inv_vf and return the sha1 of the inventory."""
1041
parents = [(parent,) for parent in parents]
1042
return self.inventories.add_lines((revision_id,), parents, lines,
1043
check_content=check_content)[0]
1045
def add_revision(self, revision_id, rev, inv=None, config=None):
1046
"""Add rev to the revision store as revision_id.
1048
:param revision_id: the revision id to use.
1049
:param rev: The revision object.
1050
:param inv: The inventory for the revision. if None, it will be looked
1051
up in the inventory storer
1052
:param config: If None no digital signature will be created.
1053
If supplied its signature_needed method will be used
1054
to determine if a signature should be made.
1056
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1058
_mod_revision.check_not_reserved_id(revision_id)
1059
if config is not None and config.signature_needed():
1061
inv = self.get_inventory(revision_id)
1062
plaintext = Testament(rev, inv).as_short_text()
1063
self.store_revision_signature(
1064
gpg.GPGStrategy(config), plaintext, revision_id)
1065
# check inventory present
1066
if not self.inventories.get_parent_map([(revision_id,)]):
1068
raise errors.WeaveRevisionNotPresent(revision_id,
1071
# yes, this is not suitable for adding with ghosts.
1072
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1075
key = (revision_id,)
1076
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1077
self._add_revision(rev)
1079
def _add_revision(self, revision):
1080
text = self._serializer.write_revision_to_string(revision)
1081
key = (revision.revision_id,)
1082
parents = tuple((parent,) for parent in revision.parent_ids)
1083
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1085
def all_revision_ids(self):
1086
"""Returns a list of all the revision ids in the repository.
1088
This is conceptually deprecated because code should generally work on
1089
the graph reachable from a particular revision, and ignore any other
1090
revisions that might be present. There is no direct replacement
1093
if 'evil' in debug.debug_flags:
1094
mutter_callsite(2, "all_revision_ids is linear with history.")
1095
return self._all_revision_ids()
1097
def _all_revision_ids(self):
1098
"""Returns a list of all the revision ids in the repository.
1100
These are in as much topological order as the underlying store can
1103
raise NotImplementedError(self._all_revision_ids)
1105
def break_lock(self):
1106
"""Break a lock if one is present from another instance.
1108
Uses the ui factory to ask for confirmation if the lock may be from
1111
self.control_files.break_lock()
1114
def _eliminate_revisions_not_present(self, revision_ids):
1115
"""Check every revision id in revision_ids to see if we have it.
1117
Returns a set of the present revisions.
1120
graph = self.get_graph()
1121
parent_map = graph.get_parent_map(revision_ids)
1122
# The old API returned a list, should this actually be a set?
1123
return parent_map.keys()
1126
def create(a_bzrdir):
1127
"""Construct the current default format repository in a_bzrdir."""
1128
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1130
def __init__(self, _format, a_bzrdir, control_files):
1131
"""instantiate a Repository.
1133
:param _format: The format of the repository on disk.
1134
:param a_bzrdir: The BzrDir of the repository.
1136
In the future we will have a single api for all stores for
1137
getting file texts, inventories and revisions, then
1138
this construct will accept instances of those things.
1140
super(Repository, self).__init__()
1141
self._format = _format
1142
# the following are part of the public API for Repository:
1143
self.bzrdir = a_bzrdir
1144
self.control_files = control_files
1145
self._transport = control_files._transport
1146
self.base = self._transport.base
1148
self._reconcile_does_inventory_gc = True
1149
self._reconcile_fixes_text_parents = False
1150
self._reconcile_backsup_inventory = True
1151
# not right yet - should be more semantically clear ?
1153
# TODO: make sure to construct the right store classes, etc, depending
1154
# on whether escaping is required.
1155
self._warn_if_deprecated()
1156
self._write_group = None
1157
# Additional places to query for data.
1158
self._fallback_repositories = []
1159
# An InventoryEntry cache, used during deserialization
1160
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1163
return '%s(%r)' % (self.__class__.__name__,
1166
def has_same_location(self, other):
1167
"""Returns a boolean indicating if this repository is at the same
1168
location as another repository.
1170
This might return False even when two repository objects are accessing
1171
the same physical repository via different URLs.
1173
if self.__class__ is not other.__class__:
1175
return (self._transport.base == other._transport.base)
1177
def is_in_write_group(self):
1178
"""Return True if there is an open write group.
1180
:seealso: start_write_group.
1182
return self._write_group is not None
1184
def is_locked(self):
1185
return self.control_files.is_locked()
1187
def is_write_locked(self):
1188
"""Return True if this object is write locked."""
1189
return self.is_locked() and self.control_files._lock_mode == 'w'
1191
def lock_write(self, token=None):
1192
"""Lock this repository for writing.
1194
This causes caching within the repository obejct to start accumlating
1195
data during reads, and allows a 'write_group' to be obtained. Write
1196
groups must be used for actual data insertion.
1198
:param token: if this is already locked, then lock_write will fail
1199
unless the token matches the existing lock.
1200
:returns: a token if this instance supports tokens, otherwise None.
1201
:raises TokenLockingNotSupported: when a token is given but this
1202
instance doesn't support using token locks.
1203
:raises MismatchedToken: if the specified token doesn't match the token
1204
of the existing lock.
1205
:seealso: start_write_group.
1207
A token should be passed in if you know that you have locked the object
1208
some other way, and need to synchronise this object's state with that
1211
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1213
locked = self.is_locked()
1214
result = self.control_files.lock_write(token=token)
1215
for repo in self._fallback_repositories:
1216
# Writes don't affect fallback repos
1219
self._refresh_data()
1222
def lock_read(self):
1223
locked = self.is_locked()
1224
self.control_files.lock_read()
1225
for repo in self._fallback_repositories:
1228
self._refresh_data()
1230
def get_physical_lock_status(self):
1231
return self.control_files.get_physical_lock_status()
1233
def leave_lock_in_place(self):
1234
"""Tell this repository not to release the physical lock when this
1237
If lock_write doesn't return a token, then this method is not supported.
1239
self.control_files.leave_in_place()
1241
def dont_leave_lock_in_place(self):
1242
"""Tell this repository to release the physical lock when this
1243
object is unlocked, even if it didn't originally acquire it.
1245
If lock_write doesn't return a token, then this method is not supported.
1247
self.control_files.dont_leave_in_place()
1250
def gather_stats(self, revid=None, committers=None):
1251
"""Gather statistics from a revision id.
1253
:param revid: The revision id to gather statistics from, if None, then
1254
no revision specific statistics are gathered.
1255
:param committers: Optional parameter controlling whether to grab
1256
a count of committers from the revision specific statistics.
1257
:return: A dictionary of statistics. Currently this contains:
1258
committers: The number of committers if requested.
1259
firstrev: A tuple with timestamp, timezone for the penultimate left
1260
most ancestor of revid, if revid is not the NULL_REVISION.
1261
latestrev: A tuple with timestamp, timezone for revid, if revid is
1262
not the NULL_REVISION.
1263
revisions: The total revision count in the repository.
1264
size: An estimate disk size of the repository in bytes.
1267
if revid and committers:
1268
result['committers'] = 0
1269
if revid and revid != _mod_revision.NULL_REVISION:
1271
all_committers = set()
1272
revisions = self.get_ancestry(revid)
1273
# pop the leading None
1275
first_revision = None
1277
# ignore the revisions in the middle - just grab first and last
1278
revisions = revisions[0], revisions[-1]
1279
for revision in self.get_revisions(revisions):
1280
if not first_revision:
1281
first_revision = revision
1283
all_committers.add(revision.committer)
1284
last_revision = revision
1286
result['committers'] = len(all_committers)
1287
result['firstrev'] = (first_revision.timestamp,
1288
first_revision.timezone)
1289
result['latestrev'] = (last_revision.timestamp,
1290
last_revision.timezone)
1292
# now gather global repository information
1293
# XXX: This is available for many repos regardless of listability.
1294
if self.bzrdir.root_transport.listable():
1295
# XXX: do we want to __define len__() ?
1296
# Maybe the versionedfiles object should provide a different
1297
# method to get the number of keys.
1298
result['revisions'] = len(self.revisions.keys())
1299
# result['size'] = t
1302
def find_branches(self, using=False):
1303
"""Find branches underneath this repository.
1305
This will include branches inside other branches.
1307
:param using: If True, list only branches using this repository.
1309
if using and not self.is_shared():
1311
return [self.bzrdir.open_branch()]
1312
except errors.NotBranchError:
1314
class Evaluator(object):
1317
self.first_call = True
1319
def __call__(self, bzrdir):
1320
# On the first call, the parameter is always the bzrdir
1321
# containing the current repo.
1322
if not self.first_call:
1324
repository = bzrdir.open_repository()
1325
except errors.NoRepositoryPresent:
1328
return False, (None, repository)
1329
self.first_call = False
1331
value = (bzrdir.open_branch(), None)
1332
except errors.NotBranchError:
1333
value = (None, None)
1337
for branch, repository in bzrdir.BzrDir.find_bzrdirs(
1338
self.bzrdir.root_transport, evaluate=Evaluator()):
1339
if branch is not None:
1340
branches.append(branch)
1341
if not using and repository is not None:
1342
branches.extend(repository.find_branches())
1346
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
1347
"""Return the revision ids that other has that this does not.
1349
These are returned in topological order.
1351
revision_id: only return revision ids included by revision_id.
1353
return InterRepository.get(other, self).search_missing_revision_ids(
1354
revision_id, find_ghosts)
1358
"""Open the repository rooted at base.
1360
For instance, if the repository is at URL/.bzr/repository,
1361
Repository.open(URL) -> a Repository instance.
1363
control = bzrdir.BzrDir.open(base)
1364
return control.open_repository()
1366
def copy_content_into(self, destination, revision_id=None):
1367
"""Make a complete copy of the content in self into destination.
1369
This is a destructive operation! Do not use it on existing
1372
return InterRepository.get(self, destination).copy_content(revision_id)
1374
def commit_write_group(self):
1375
"""Commit the contents accrued within the current write group.
1377
:seealso: start_write_group.
1379
if self._write_group is not self.get_transaction():
1380
# has an unlock or relock occured ?
1381
raise errors.BzrError('mismatched lock context %r and '
1383
(self.get_transaction(), self._write_group))
1384
self._commit_write_group()
1385
self._write_group = None
1387
def _commit_write_group(self):
1388
"""Template method for per-repository write group cleanup.
1390
This is called before the write group is considered to be
1391
finished and should ensure that all data handed to the repository
1392
for writing during the write group is safely committed (to the
1393
extent possible considering file system caching etc).
1396
def suspend_write_group(self):
1397
raise errors.UnsuspendableWriteGroup(self)
1399
def refresh_data(self):
1400
"""Re-read any data needed to to synchronise with disk.
1402
This method is intended to be called after another repository instance
1403
(such as one used by a smart server) has inserted data into the
1404
repository. It may not be called during a write group, but may be
1405
called at any other time.
1407
if self.is_in_write_group():
1408
raise errors.InternalBzrError(
1409
"May not refresh_data while in a write group.")
1410
self._refresh_data()
1412
def resume_write_group(self, tokens):
1413
if not self.is_write_locked():
1414
raise errors.NotWriteLocked(self)
1415
if self._write_group:
1416
raise errors.BzrError('already in a write group')
1417
self._resume_write_group(tokens)
1418
# so we can detect unlock/relock - the write group is now entered.
1419
self._write_group = self.get_transaction()
1421
def _resume_write_group(self, tokens):
1422
raise errors.UnsuspendableWriteGroup(self)
1424
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1426
"""Fetch the content required to construct revision_id from source.
1428
If revision_id is None and fetch_spec is None, then all content is
1431
fetch() may not be used when the repository is in a write group -
1432
either finish the current write group before using fetch, or use
1433
fetch before starting the write group.
1435
:param find_ghosts: Find and copy revisions in the source that are
1436
ghosts in the target (and not reachable directly by walking out to
1437
the first-present revision in target from revision_id).
1438
:param revision_id: If specified, all the content needed for this
1439
revision ID will be copied to the target. Fetch will determine for
1440
itself which content needs to be copied.
1441
:param fetch_spec: If specified, a SearchResult or
1442
PendingAncestryResult that describes which revisions to copy. This
1443
allows copying multiple heads at once. Mutually exclusive with
1446
if fetch_spec is not None and revision_id is not None:
1447
raise AssertionError(
1448
"fetch_spec and revision_id are mutually exclusive.")
1449
if self.is_in_write_group():
1450
raise errors.InternalBzrError(
1451
"May not fetch while in a write group.")
1452
# fast path same-url fetch operations
1453
if self.has_same_location(source) and fetch_spec is None:
1454
# check that last_revision is in 'from' and then return a
1456
if (revision_id is not None and
1457
not _mod_revision.is_null(revision_id)):
1458
self.get_revision(revision_id)
1460
# if there is no specific appropriate InterRepository, this will get
1461
# the InterRepository base class, which raises an
1462
# IncompatibleRepositories when asked to fetch.
1463
inter = InterRepository.get(source, self)
1464
return inter.fetch(revision_id=revision_id, pb=pb,
1465
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1467
def create_bundle(self, target, base, fileobj, format=None):
1468
return serializer.write_bundle(self, target, base, fileobj, format)
1470
def get_commit_builder(self, branch, parents, config, timestamp=None,
1471
timezone=None, committer=None, revprops=None,
1473
"""Obtain a CommitBuilder for this repository.
1475
:param branch: Branch to commit to.
1476
:param parents: Revision ids of the parents of the new revision.
1477
:param config: Configuration to use.
1478
:param timestamp: Optional timestamp recorded for commit.
1479
:param timezone: Optional timezone for timestamp.
1480
:param committer: Optional committer to set for commit.
1481
:param revprops: Optional dictionary of revision properties.
1482
:param revision_id: Optional revision id.
1484
result = self._commit_builder_class(self, parents, config,
1485
timestamp, timezone, committer, revprops, revision_id)
1486
self.start_write_group()
1490
if (self.control_files._lock_count == 1 and
1491
self.control_files._lock_mode == 'w'):
1492
if self._write_group is not None:
1493
self.abort_write_group()
1494
self.control_files.unlock()
1495
raise errors.BzrError(
1496
'Must end write groups before releasing write locks.')
1497
self.control_files.unlock()
1498
if self.control_files._lock_count == 0:
1499
self._inventory_entry_cache.clear()
1500
for repo in self._fallback_repositories:
1504
def clone(self, a_bzrdir, revision_id=None):
1505
"""Clone this repository into a_bzrdir using the current format.
1507
Currently no check is made that the format of this repository and
1508
the bzrdir format are compatible. FIXME RBC 20060201.
1510
:return: The newly created destination repository.
1512
# TODO: deprecate after 0.16; cloning this with all its settings is
1513
# probably not very useful -- mbp 20070423
1514
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1515
self.copy_content_into(dest_repo, revision_id)
1518
def start_write_group(self):
1519
"""Start a write group in the repository.
1521
Write groups are used by repositories which do not have a 1:1 mapping
1522
between file ids and backend store to manage the insertion of data from
1523
both fetch and commit operations.
1525
A write lock is required around the start_write_group/commit_write_group
1526
for the support of lock-requiring repository formats.
1528
One can only insert data into a repository inside a write group.
1532
if not self.is_write_locked():
1533
raise errors.NotWriteLocked(self)
1534
if self._write_group:
1535
raise errors.BzrError('already in a write group')
1536
self._start_write_group()
1537
# so we can detect unlock/relock - the write group is now entered.
1538
self._write_group = self.get_transaction()
1540
def _start_write_group(self):
1541
"""Template method for per-repository write group startup.
1543
This is called before the write group is considered to be
1548
def sprout(self, to_bzrdir, revision_id=None):
1549
"""Create a descendent repository for new development.
1551
Unlike clone, this does not copy the settings of the repository.
1553
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1554
dest_repo.fetch(self, revision_id=revision_id)
1557
def _create_sprouting_repo(self, a_bzrdir, shared):
1558
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1559
# use target default format.
1560
dest_repo = a_bzrdir.create_repository()
1562
# Most control formats need the repository to be specifically
1563
# created, but on some old all-in-one formats it's not needed
1565
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1566
except errors.UninitializableFormat:
1567
dest_repo = a_bzrdir.open_repository()
1570
def _get_sink(self):
1571
"""Return a sink for streaming into this repository."""
1572
return StreamSink(self)
1574
def _get_source(self, to_format):
1575
"""Return a source for streaming from this repository."""
1576
return StreamSource(self, to_format)
1579
def has_revision(self, revision_id):
1580
"""True if this repository has a copy of the revision."""
1581
return revision_id in self.has_revisions((revision_id,))
1584
def has_revisions(self, revision_ids):
1585
"""Probe to find out the presence of multiple revisions.
1587
:param revision_ids: An iterable of revision_ids.
1588
:return: A set of the revision_ids that were present.
1590
parent_map = self.revisions.get_parent_map(
1591
[(rev_id,) for rev_id in revision_ids])
1593
if _mod_revision.NULL_REVISION in revision_ids:
1594
result.add(_mod_revision.NULL_REVISION)
1595
result.update([key[0] for key in parent_map])
1599
def get_revision(self, revision_id):
1600
"""Return the Revision object for a named revision."""
1601
return self.get_revisions([revision_id])[0]
1604
def get_revision_reconcile(self, revision_id):
1605
"""'reconcile' helper routine that allows access to a revision always.
1607
This variant of get_revision does not cross check the weave graph
1608
against the revision one as get_revision does: but it should only
1609
be used by reconcile, or reconcile-alike commands that are correcting
1610
or testing the revision graph.
1612
return self._get_revisions([revision_id])[0]
1615
def get_revisions(self, revision_ids):
1616
"""Get many revisions at once."""
1617
return self._get_revisions(revision_ids)
1620
def _get_revisions(self, revision_ids):
1621
"""Core work logic to get many revisions without sanity checks."""
1622
for rev_id in revision_ids:
1623
if not rev_id or not isinstance(rev_id, basestring):
1624
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1625
keys = [(key,) for key in revision_ids]
1626
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1628
for record in stream:
1629
if record.storage_kind == 'absent':
1630
raise errors.NoSuchRevision(self, record.key[0])
1631
text = record.get_bytes_as('fulltext')
1632
rev = self._serializer.read_revision_from_string(text)
1633
revs[record.key[0]] = rev
1634
return [revs[revid] for revid in revision_ids]
1637
def get_revision_xml(self, revision_id):
1638
"""Return the XML representation of a revision.
1640
:param revision_id: Revision for which to return the XML.
1643
return self._serializer.write_revision_to_string(
1644
self.get_revision(revision_id))
1646
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1647
"""Produce a generator of revision deltas.
1649
Note that the input is a sequence of REVISIONS, not revision_ids.
1650
Trees will be held in memory until the generator exits.
1651
Each delta is relative to the revision's lefthand predecessor.
1653
:param specific_fileids: if not None, the result is filtered
1654
so that only those file-ids, their parents and their
1655
children are included.
1657
# Get the revision-ids of interest
1658
required_trees = set()
1659
for revision in revisions:
1660
required_trees.add(revision.revision_id)
1661
required_trees.update(revision.parent_ids[:1])
1663
# Get the matching filtered trees. Note that it's more
1664
# efficient to pass filtered trees to changes_from() rather
1665
# than doing the filtering afterwards. changes_from() could
1666
# arguably do the filtering itself but it's path-based, not
1667
# file-id based, so filtering before or afterwards is
1669
if specific_fileids is None:
1670
trees = dict((t.get_revision_id(), t) for
1671
t in self.revision_trees(required_trees))
1673
trees = dict((t.get_revision_id(), t) for
1674
t in self._filtered_revision_trees(required_trees,
1677
# Calculate the deltas
1678
for revision in revisions:
1679
if not revision.parent_ids:
1680
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1682
old_tree = trees[revision.parent_ids[0]]
1683
yield trees[revision.revision_id].changes_from(old_tree)
1686
def get_revision_delta(self, revision_id, specific_fileids=None):
1687
"""Return the delta for one revision.
1689
The delta is relative to the left-hand predecessor of the
1692
:param specific_fileids: if not None, the result is filtered
1693
so that only those file-ids, their parents and their
1694
children are included.
1696
r = self.get_revision(revision_id)
1697
return list(self.get_deltas_for_revisions([r],
1698
specific_fileids=specific_fileids))[0]
1701
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1702
signature = gpg_strategy.sign(plaintext)
1703
self.add_signature_text(revision_id, signature)
1706
def add_signature_text(self, revision_id, signature):
1707
self.signatures.add_lines((revision_id,), (),
1708
osutils.split_lines(signature))
1710
def find_text_key_references(self):
1711
"""Find the text key references within the repository.
1713
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1714
to whether they were referred to by the inventory of the
1715
revision_id that they contain. The inventory texts from all present
1716
revision ids are assessed to generate this report.
1718
revision_keys = self.revisions.keys()
1719
w = self.inventories
1720
pb = ui.ui_factory.nested_progress_bar()
1722
return self._find_text_key_references_from_xml_inventory_lines(
1723
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1727
def _find_text_key_references_from_xml_inventory_lines(self,
1729
"""Core routine for extracting references to texts from inventories.
1731
This performs the translation of xml lines to revision ids.
1733
:param line_iterator: An iterator of lines, origin_version_id
1734
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1735
to whether they were referred to by the inventory of the
1736
revision_id that they contain. Note that if that revision_id was
1737
not part of the line_iterator's output then False will be given -
1738
even though it may actually refer to that key.
1740
if not self._serializer.support_altered_by_hack:
1741
raise AssertionError(
1742
"_find_text_key_references_from_xml_inventory_lines only "
1743
"supported for branches which store inventory as unnested xml"
1744
", not on %r" % self)
1747
# this code needs to read every new line in every inventory for the
1748
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
1749
# not present in one of those inventories is unnecessary but not
1750
# harmful because we are filtering by the revision id marker in the
1751
# inventory lines : we only select file ids altered in one of those
1752
# revisions. We don't need to see all lines in the inventory because
1753
# only those added in an inventory in rev X can contain a revision=X
1755
unescape_revid_cache = {}
1756
unescape_fileid_cache = {}
1758
# jam 20061218 In a big fetch, this handles hundreds of thousands
1759
# of lines, so it has had a lot of inlining and optimizing done.
1760
# Sorry that it is a little bit messy.
1761
# Move several functions to be local variables, since this is a long
1763
search = self._file_ids_altered_regex.search
1764
unescape = _unescape_xml
1765
setdefault = result.setdefault
1766
for line, line_key in line_iterator:
1767
match = search(line)
1770
# One call to match.group() returning multiple items is quite a
1771
# bit faster than 2 calls to match.group() each returning 1
1772
file_id, revision_id = match.group('file_id', 'revision_id')
1774
# Inlining the cache lookups helps a lot when you make 170,000
1775
# lines and 350k ids, versus 8.4 unique ids.
1776
# Using a cache helps in 2 ways:
1777
# 1) Avoids unnecessary decoding calls
1778
# 2) Re-uses cached strings, which helps in future set and
1780
# (2) is enough that removing encoding entirely along with
1781
# the cache (so we are using plain strings) results in no
1782
# performance improvement.
1784
revision_id = unescape_revid_cache[revision_id]
1786
unescaped = unescape(revision_id)
1787
unescape_revid_cache[revision_id] = unescaped
1788
revision_id = unescaped
1790
# Note that unconditionally unescaping means that we deserialise
1791
# every fileid, which for general 'pull' is not great, but we don't
1792
# really want to have some many fulltexts that this matters anyway.
1795
file_id = unescape_fileid_cache[file_id]
1797
unescaped = unescape(file_id)
1798
unescape_fileid_cache[file_id] = unescaped
1801
key = (file_id, revision_id)
1802
setdefault(key, False)
1803
if revision_id == line_key[-1]:
1807
def _inventory_xml_lines_for_keys(self, keys):
1808
"""Get a line iterator of the sort needed for findind references.
1810
Not relevant for non-xml inventory repositories.
1812
Ghosts in revision_keys are ignored.
1814
:param revision_keys: The revision keys for the inventories to inspect.
1815
:return: An iterator over (inventory line, revid) for the fulltexts of
1816
all of the xml inventories specified by revision_keys.
1818
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1819
for record in stream:
1820
if record.storage_kind != 'absent':
1821
chunks = record.get_bytes_as('chunked')
1822
revid = record.key[-1]
1823
lines = osutils.chunks_to_lines(chunks)
1827
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1829
"""Helper routine for fileids_altered_by_revision_ids.
1831
This performs the translation of xml lines to revision ids.
1833
:param line_iterator: An iterator of lines, origin_version_id
1834
:param revision_ids: The revision ids to filter for. This should be a
1835
set or other type which supports efficient __contains__ lookups, as
1836
the revision id from each parsed line will be looked up in the
1837
revision_ids filter.
1838
:return: a dictionary mapping altered file-ids to an iterable of
1839
revision_ids. Each altered file-ids has the exact revision_ids that
1840
altered it listed explicitly.
1842
seen = set(self._find_text_key_references_from_xml_inventory_lines(
1843
line_iterator).iterkeys())
1844
# Note that revision_ids are revision keys.
1845
parent_maps = self.revisions.get_parent_map(revision_ids)
1847
map(parents.update, parent_maps.itervalues())
1848
parents.difference_update(revision_ids)
1849
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
1850
self._inventory_xml_lines_for_keys(parents)))
1851
new_keys = seen - parent_seen
1853
setdefault = result.setdefault
1854
for key in new_keys:
1855
setdefault(key[0], set()).add(key[-1])
1858
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
1859
"""Find the file ids and versions affected by revisions.
1861
:param revisions: an iterable containing revision ids.
1862
:param _inv_weave: The inventory weave from this repository or None.
1863
If None, the inventory weave will be opened automatically.
1864
:return: a dictionary mapping altered file-ids to an iterable of
1865
revision_ids. Each altered file-ids has the exact revision_ids that
1866
altered it listed explicitly.
1868
selected_keys = set((revid,) for revid in revision_ids)
1869
w = _inv_weave or self.inventories
1870
pb = ui.ui_factory.nested_progress_bar()
1872
return self._find_file_ids_from_xml_inventory_lines(
1873
w.iter_lines_added_or_present_in_keys(
1874
selected_keys, pb=pb),
1879
def iter_files_bytes(self, desired_files):
1880
"""Iterate through file versions.
1882
Files will not necessarily be returned in the order they occur in
1883
desired_files. No specific order is guaranteed.
1885
Yields pairs of identifier, bytes_iterator. identifier is an opaque
1886
value supplied by the caller as part of desired_files. It should
1887
uniquely identify the file version in the caller's context. (Examples:
1888
an index number or a TreeTransform trans_id.)
1890
bytes_iterator is an iterable of bytestrings for the file. The
1891
kind of iterable and length of the bytestrings are unspecified, but for
1892
this implementation, it is a list of bytes produced by
1893
VersionedFile.get_record_stream().
1895
:param desired_files: a list of (file_id, revision_id, identifier)
1899
for file_id, revision_id, callable_data in desired_files:
1900
text_keys[(file_id, revision_id)] = callable_data
1901
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
1902
if record.storage_kind == 'absent':
1903
raise errors.RevisionNotPresent(record.key, self)
1904
yield text_keys[record.key], record.get_bytes_as('chunked')
1906
def _generate_text_key_index(self, text_key_references=None,
1908
"""Generate a new text key index for the repository.
1910
This is an expensive function that will take considerable time to run.
1912
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
1913
list of parents, also text keys. When a given key has no parents,
1914
the parents list will be [NULL_REVISION].
1916
# All revisions, to find inventory parents.
1917
if ancestors is None:
1918
graph = self.get_graph()
1919
ancestors = graph.get_parent_map(self.all_revision_ids())
1920
if text_key_references is None:
1921
text_key_references = self.find_text_key_references()
1922
pb = ui.ui_factory.nested_progress_bar()
1924
return self._do_generate_text_key_index(ancestors,
1925
text_key_references, pb)
1929
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
1930
"""Helper for _generate_text_key_index to avoid deep nesting."""
1931
revision_order = tsort.topo_sort(ancestors)
1932
invalid_keys = set()
1934
for revision_id in revision_order:
1935
revision_keys[revision_id] = set()
1936
text_count = len(text_key_references)
1937
# a cache of the text keys to allow reuse; costs a dict of all the
1938
# keys, but saves a 2-tuple for every child of a given key.
1940
for text_key, valid in text_key_references.iteritems():
1942
invalid_keys.add(text_key)
1944
revision_keys[text_key[1]].add(text_key)
1945
text_key_cache[text_key] = text_key
1946
del text_key_references
1948
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
1949
NULL_REVISION = _mod_revision.NULL_REVISION
1950
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
1951
# too small for large or very branchy trees. However, for 55K path
1952
# trees, it would be easy to use too much memory trivially. Ideally we
1953
# could gauge this by looking at available real memory etc, but this is
1954
# always a tricky proposition.
1955
inventory_cache = lru_cache.LRUCache(10)
1956
batch_size = 10 # should be ~150MB on a 55K path tree
1957
batch_count = len(revision_order) / batch_size + 1
1959
pb.update("Calculating text parents", processed_texts, text_count)
1960
for offset in xrange(batch_count):
1961
to_query = revision_order[offset * batch_size:(offset + 1) *
1965
for rev_tree in self.revision_trees(to_query):
1966
revision_id = rev_tree.get_revision_id()
1967
parent_ids = ancestors[revision_id]
1968
for text_key in revision_keys[revision_id]:
1969
pb.update("Calculating text parents", processed_texts)
1970
processed_texts += 1
1971
candidate_parents = []
1972
for parent_id in parent_ids:
1973
parent_text_key = (text_key[0], parent_id)
1975
check_parent = parent_text_key not in \
1976
revision_keys[parent_id]
1978
# the parent parent_id is a ghost:
1979
check_parent = False
1980
# truncate the derived graph against this ghost.
1981
parent_text_key = None
1983
# look at the parent commit details inventories to
1984
# determine possible candidates in the per file graph.
1987
inv = inventory_cache[parent_id]
1989
inv = self.revision_tree(parent_id).inventory
1990
inventory_cache[parent_id] = inv
1991
parent_entry = inv._byid.get(text_key[0], None)
1992
if parent_entry is not None:
1994
text_key[0], parent_entry.revision)
1996
parent_text_key = None
1997
if parent_text_key is not None:
1998
candidate_parents.append(
1999
text_key_cache[parent_text_key])
2000
parent_heads = text_graph.heads(candidate_parents)
2001
new_parents = list(parent_heads)
2002
new_parents.sort(key=lambda x:candidate_parents.index(x))
2003
if new_parents == []:
2004
new_parents = [NULL_REVISION]
2005
text_index[text_key] = new_parents
2007
for text_key in invalid_keys:
2008
text_index[text_key] = [NULL_REVISION]
2011
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2012
"""Get an iterable listing the keys of all the data introduced by a set
2015
The keys will be ordered so that the corresponding items can be safely
2016
fetched and inserted in that order.
2018
:returns: An iterable producing tuples of (knit-kind, file-id,
2019
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2020
'revisions'. file-id is None unless knit-kind is 'file'.
2022
# XXX: it's a bit weird to control the inventory weave caching in this
2023
# generator. Ideally the caching would be done in fetch.py I think. Or
2024
# maybe this generator should explicitly have the contract that it
2025
# should not be iterated until the previously yielded item has been
2027
inv_w = self.inventories
2029
# file ids that changed
2030
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2032
num_file_ids = len(file_ids)
2033
for file_id, altered_versions in file_ids.iteritems():
2034
if _files_pb is not None:
2035
_files_pb.update("fetch texts", count, num_file_ids)
2037
yield ("file", file_id, altered_versions)
2038
# We're done with the files_pb. Note that it finished by the caller,
2039
# just as it was created by the caller.
2043
yield ("inventory", None, revision_ids)
2046
# XXX: Note ATM no callers actually pay attention to this return
2047
# instead they just use the list of revision ids and ignore
2048
# missing sigs. Consider removing this work entirely
2049
revisions_with_signatures = set(self.signatures.get_parent_map(
2050
[(r,) for r in revision_ids]))
2051
revisions_with_signatures = set(
2052
[r for (r,) in revisions_with_signatures])
2053
revisions_with_signatures.intersection_update(revision_ids)
2054
yield ("signatures", None, revisions_with_signatures)
2057
yield ("revisions", None, revision_ids)
2060
def get_inventory(self, revision_id):
2061
"""Get Inventory object by revision id."""
2062
return self.iter_inventories([revision_id]).next()
2064
def iter_inventories(self, revision_ids):
2065
"""Get many inventories by revision_ids.
2067
This will buffer some or all of the texts used in constructing the
2068
inventories in memory, but will only parse a single inventory at a
2071
:param revision_ids: The expected revision ids of the inventories.
2072
:return: An iterator of inventories.
2074
if ((None in revision_ids)
2075
or (_mod_revision.NULL_REVISION in revision_ids)):
2076
raise ValueError('cannot get null revision inventory')
2077
return self._iter_inventories(revision_ids)
2079
def _iter_inventories(self, revision_ids):
2080
"""single-document based inventory iteration."""
2081
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2082
yield self.deserialise_inventory(revision_id, text)
2084
def _iter_inventory_xmls(self, revision_ids):
2085
keys = [(revision_id,) for revision_id in revision_ids]
2086
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2088
for record in stream:
2089
if record.storage_kind != 'absent':
2090
text_chunks[record.key] = record.get_bytes_as('chunked')
2092
raise errors.NoSuchRevision(self, record.key)
2094
chunks = text_chunks.pop(key)
2095
yield ''.join(chunks), key[-1]
2097
def deserialise_inventory(self, revision_id, xml):
2098
"""Transform the xml into an inventory object.
2100
:param revision_id: The expected revision id of the inventory.
2101
:param xml: A serialised inventory.
2103
result = self._serializer.read_inventory_from_string(xml, revision_id,
2104
entry_cache=self._inventory_entry_cache)
2105
if result.revision_id != revision_id:
2106
raise AssertionError('revision id mismatch %s != %s' % (
2107
result.revision_id, revision_id))
2110
def serialise_inventory(self, inv):
2111
return self._serializer.write_inventory_to_string(inv)
2113
def _serialise_inventory_to_lines(self, inv):
2114
return self._serializer.write_inventory_to_lines(inv)
2116
def get_serializer_format(self):
2117
return self._serializer.format_num
2120
def get_inventory_xml(self, revision_id):
2121
"""Get inventory XML as a file object."""
2122
texts = self._iter_inventory_xmls([revision_id])
2124
text, revision_id = texts.next()
2125
except StopIteration:
2126
raise errors.HistoryMissing(self, 'inventory', revision_id)
2130
def get_inventory_sha1(self, revision_id):
2131
"""Return the sha1 hash of the inventory entry
2133
return self.get_revision(revision_id).inventory_sha1
2135
def iter_reverse_revision_history(self, revision_id):
2136
"""Iterate backwards through revision ids in the lefthand history
2138
:param revision_id: The revision id to start with. All its lefthand
2139
ancestors will be traversed.
2141
graph = self.get_graph()
2142
next_id = revision_id
2144
if next_id in (None, _mod_revision.NULL_REVISION):
2147
# Note: The following line may raise KeyError in the event of
2148
# truncated history. We decided not to have a try:except:raise
2149
# RevisionNotPresent here until we see a use for it, because of the
2150
# cost in an inner loop that is by its very nature O(history).
2151
# Robert Collins 20080326
2152
parents = graph.get_parent_map([next_id])[next_id]
2153
if len(parents) == 0:
2156
next_id = parents[0]
2159
def get_revision_inventory(self, revision_id):
2160
"""Return inventory of a past revision."""
2161
# TODO: Unify this with get_inventory()
2162
# bzr 0.0.6 and later imposes the constraint that the inventory_id
2163
# must be the same as its revision, so this is trivial.
2164
if revision_id is None:
2165
# This does not make sense: if there is no revision,
2166
# then it is the current tree inventory surely ?!
2167
# and thus get_root_id() is something that looks at the last
2168
# commit on the branch, and the get_root_id is an inventory check.
2169
raise NotImplementedError
2170
# return Inventory(self.get_root_id())
2172
return self.get_inventory(revision_id)
2174
def is_shared(self):
2175
"""Return True if this repository is flagged as a shared repository."""
2176
raise NotImplementedError(self.is_shared)
2179
def reconcile(self, other=None, thorough=False):
2180
"""Reconcile this repository."""
2181
from bzrlib.reconcile import RepoReconciler
2182
reconciler = RepoReconciler(self, thorough=thorough)
2183
reconciler.reconcile()
2186
def _refresh_data(self):
2187
"""Helper called from lock_* to ensure coherency with disk.
2189
The default implementation does nothing; it is however possible
2190
for repositories to maintain loaded indices across multiple locks
2191
by checking inside their implementation of this method to see
2192
whether their indices are still valid. This depends of course on
2193
the disk format being validatable in this manner. This method is
2194
also called by the refresh_data() public interface to cause a refresh
2195
to occur while in a write lock so that data inserted by a smart server
2196
push operation is visible on the client's instance of the physical
2201
def revision_tree(self, revision_id):
2202
"""Return Tree for a revision on this branch.
2204
`revision_id` may be NULL_REVISION for the empty tree revision.
2206
revision_id = _mod_revision.ensure_null(revision_id)
2207
# TODO: refactor this to use an existing revision object
2208
# so we don't need to read it in twice.
2209
if revision_id == _mod_revision.NULL_REVISION:
2210
return RevisionTree(self, Inventory(root_id=None),
2211
_mod_revision.NULL_REVISION)
2213
inv = self.get_revision_inventory(revision_id)
2214
return RevisionTree(self, inv, revision_id)
2216
def revision_trees(self, revision_ids):
2217
"""Return Trees for revisions in this repository.
2219
:param revision_ids: a sequence of revision-ids;
2220
a revision-id may not be None or 'null:'
2222
inventories = self.iter_inventories(revision_ids)
2223
for inv in inventories:
2224
yield RevisionTree(self, inv, inv.revision_id)
2226
def _filtered_revision_trees(self, revision_ids, file_ids):
2227
"""Return Tree for a revision on this branch with only some files.
2229
:param revision_ids: a sequence of revision-ids;
2230
a revision-id may not be None or 'null:'
2231
:param file_ids: if not None, the result is filtered
2232
so that only those file-ids, their parents and their
2233
children are included.
2235
inventories = self.iter_inventories(revision_ids)
2236
for inv in inventories:
2237
# Should we introduce a FilteredRevisionTree class rather
2238
# than pre-filter the inventory here?
2239
filtered_inv = inv.filter(file_ids)
2240
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2243
def get_ancestry(self, revision_id, topo_sorted=True):
2244
"""Return a list of revision-ids integrated by a revision.
2246
The first element of the list is always None, indicating the origin
2247
revision. This might change when we have history horizons, or
2248
perhaps we should have a new API.
2250
This is topologically sorted.
2252
if _mod_revision.is_null(revision_id):
2254
if not self.has_revision(revision_id):
2255
raise errors.NoSuchRevision(self, revision_id)
2256
graph = self.get_graph()
2258
search = graph._make_breadth_first_searcher([revision_id])
2261
found, ghosts = search.next_with_ghosts()
2262
except StopIteration:
2265
if _mod_revision.NULL_REVISION in keys:
2266
keys.remove(_mod_revision.NULL_REVISION)
2268
parent_map = graph.get_parent_map(keys)
2269
keys = tsort.topo_sort(parent_map)
2270
return [None] + list(keys)
2273
"""Compress the data within the repository.
2275
This operation only makes sense for some repository types. For other
2276
types it should be a no-op that just returns.
2278
This stub method does not require a lock, but subclasses should use
2279
@needs_write_lock as this is a long running call its reasonable to
2280
implicitly lock for the user.
2283
def get_transaction(self):
2284
return self.control_files.get_transaction()
2286
def get_parent_map(self, revision_ids):
2287
"""See graph._StackedParentsProvider.get_parent_map"""
2288
# revisions index works in keys; this just works in revisions
2289
# therefore wrap and unwrap
2292
for revision_id in revision_ids:
2293
if revision_id == _mod_revision.NULL_REVISION:
2294
result[revision_id] = ()
2295
elif revision_id is None:
2296
raise ValueError('get_parent_map(None) is not valid')
2298
query_keys.append((revision_id ,))
2299
for ((revision_id,), parent_keys) in \
2300
self.revisions.get_parent_map(query_keys).iteritems():
2302
result[revision_id] = tuple(parent_revid
2303
for (parent_revid,) in parent_keys)
2305
result[revision_id] = (_mod_revision.NULL_REVISION,)
2308
def _make_parents_provider(self):
2311
def get_graph(self, other_repository=None):
2312
"""Return the graph walker for this repository format"""
2313
parents_provider = self._make_parents_provider()
2314
if (other_repository is not None and
2315
not self.has_same_location(other_repository)):
2316
parents_provider = graph._StackedParentsProvider(
2317
[parents_provider, other_repository._make_parents_provider()])
2318
return graph.Graph(parents_provider)
2320
def _get_versioned_file_checker(self, text_key_references=None):
2321
"""Return an object suitable for checking versioned files.
2323
:param text_key_references: if non-None, an already built
2324
dictionary mapping text keys ((fileid, revision_id) tuples)
2325
to whether they were referred to by the inventory of the
2326
revision_id that they contain. If None, this will be
2329
return _VersionedFileChecker(self,
2330
text_key_references=text_key_references)
2332
def revision_ids_to_search_result(self, result_set):
2333
"""Convert a set of revision ids to a graph SearchResult."""
2334
result_parents = set()
2335
for parents in self.get_graph().get_parent_map(
2336
result_set).itervalues():
2337
result_parents.update(parents)
2338
included_keys = result_set.intersection(result_parents)
2339
start_keys = result_set.difference(included_keys)
2340
exclude_keys = result_parents.difference(result_set)
2341
result = graph.SearchResult(start_keys, exclude_keys,
2342
len(result_set), result_set)
2346
def set_make_working_trees(self, new_value):
2347
"""Set the policy flag for making working trees when creating branches.
2349
This only applies to branches that use this repository.
2351
The default is 'True'.
2352
:param new_value: True to restore the default, False to disable making
2355
raise NotImplementedError(self.set_make_working_trees)
2357
def make_working_trees(self):
2358
"""Returns the policy for making working trees on new branches."""
2359
raise NotImplementedError(self.make_working_trees)
2362
def sign_revision(self, revision_id, gpg_strategy):
2363
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2364
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2367
def has_signature_for_revision_id(self, revision_id):
2368
"""Query for a revision signature for revision_id in the repository."""
2369
if not self.has_revision(revision_id):
2370
raise errors.NoSuchRevision(self, revision_id)
2371
sig_present = (1 == len(
2372
self.signatures.get_parent_map([(revision_id,)])))
2376
def get_signature_text(self, revision_id):
2377
"""Return the text for a signature."""
2378
stream = self.signatures.get_record_stream([(revision_id,)],
2380
record = stream.next()
2381
if record.storage_kind == 'absent':
2382
raise errors.NoSuchRevision(self, revision_id)
2383
return record.get_bytes_as('fulltext')
2386
def check(self, revision_ids=None):
2387
"""Check consistency of all history of given revision_ids.
2389
Different repository implementations should override _check().
2391
:param revision_ids: A non-empty list of revision_ids whose ancestry
2392
will be checked. Typically the last revision_id of a branch.
2394
return self._check(revision_ids)
2396
def _check(self, revision_ids):
2397
result = check.Check(self)
2401
def _warn_if_deprecated(self):
2402
global _deprecation_warning_done
2403
if _deprecation_warning_done:
2405
_deprecation_warning_done = True
2406
warning("Format %s for %s is deprecated - please use 'bzr upgrade' to get better performance"
2407
% (self._format, self.bzrdir.transport.base))
2409
def supports_rich_root(self):
2410
return self._format.rich_root_data
2412
def _check_ascii_revisionid(self, revision_id, method):
2413
"""Private helper for ascii-only repositories."""
2414
# weave repositories refuse to store revisionids that are non-ascii.
2415
if revision_id is not None:
2416
# weaves require ascii revision ids.
2417
if isinstance(revision_id, unicode):
2419
revision_id.encode('ascii')
2420
except UnicodeEncodeError:
2421
raise errors.NonAsciiRevisionId(method, self)
2424
revision_id.decode('ascii')
2425
except UnicodeDecodeError:
2426
raise errors.NonAsciiRevisionId(method, self)
2428
def revision_graph_can_have_wrong_parents(self):
2429
"""Is it possible for this repository to have a revision graph with
2432
If True, then this repository must also implement
2433
_find_inconsistent_revision_parents so that check and reconcile can
2434
check for inconsistencies before proceeding with other checks that may
2435
depend on the revision index being consistent.
2437
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2440
# remove these delegates a while after bzr 0.15
2441
def __make_delegated(name, from_module):
2442
def _deprecated_repository_forwarder():
2443
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2444
% (name, from_module),
2447
m = __import__(from_module, globals(), locals(), [name])
2449
return getattr(m, name)
2450
except AttributeError:
2451
raise AttributeError('module %s has no name %s'
2453
globals()[name] = _deprecated_repository_forwarder
2456
'AllInOneRepository',
2457
'WeaveMetaDirRepository',
2458
'PreSplitOutRepositoryFormat',
2459
'RepositoryFormat4',
2460
'RepositoryFormat5',
2461
'RepositoryFormat6',
2462
'RepositoryFormat7',
2464
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2468
'RepositoryFormatKnit',
2469
'RepositoryFormatKnit1',
2471
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2474
def install_revision(repository, rev, revision_tree):
2475
"""Install all revision data into a repository."""
2476
install_revisions(repository, [(rev, revision_tree, None)])
2479
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2480
"""Install all revision data into a repository.
2482
Accepts an iterable of revision, tree, signature tuples. The signature
2485
repository.start_write_group()
2487
for n, (revision, revision_tree, signature) in enumerate(iterable):
2488
_install_revision(repository, revision, revision_tree, signature)
2490
pb.update('Transferring revisions', n + 1, num_revisions)
2492
repository.abort_write_group()
2495
repository.commit_write_group()
2498
def _install_revision(repository, rev, revision_tree, signature):
2499
"""Install all revision data into a repository."""
2500
present_parents = []
2502
for p_id in rev.parent_ids:
2503
if repository.has_revision(p_id):
2504
present_parents.append(p_id)
2505
parent_trees[p_id] = repository.revision_tree(p_id)
2507
parent_trees[p_id] = repository.revision_tree(
2508
_mod_revision.NULL_REVISION)
2510
inv = revision_tree.inventory
2511
entries = inv.iter_entries()
2512
# backwards compatibility hack: skip the root id.
2513
if not repository.supports_rich_root():
2514
path, root = entries.next()
2515
if root.revision != rev.revision_id:
2516
raise errors.IncompatibleRevision(repr(repository))
2518
for path, ie in entries:
2519
text_keys[(ie.file_id, ie.revision)] = ie
2520
text_parent_map = repository.texts.get_parent_map(text_keys)
2521
missing_texts = set(text_keys) - set(text_parent_map)
2522
# Add the texts that are not already present
2523
for text_key in missing_texts:
2524
ie = text_keys[text_key]
2526
# FIXME: TODO: The following loop overlaps/duplicates that done by
2527
# commit to determine parents. There is a latent/real bug here where
2528
# the parents inserted are not those commit would do - in particular
2529
# they are not filtered by heads(). RBC, AB
2530
for revision, tree in parent_trees.iteritems():
2531
if ie.file_id not in tree:
2533
parent_id = tree.inventory[ie.file_id].revision
2534
if parent_id in text_parents:
2536
text_parents.append((ie.file_id, parent_id))
2537
lines = revision_tree.get_file(ie.file_id).readlines()
2538
repository.texts.add_lines(text_key, text_parents, lines)
2540
# install the inventory
2541
repository.add_inventory(rev.revision_id, inv, present_parents)
2542
except errors.RevisionAlreadyPresent:
2544
if signature is not None:
2545
repository.add_signature_text(rev.revision_id, signature)
2546
repository.add_revision(rev.revision_id, rev, inv)
2549
class MetaDirRepository(Repository):
2550
"""Repositories in the new meta-dir layout.
2552
:ivar _transport: Transport for access to repository control files,
2553
typically pointing to .bzr/repository.
2556
def __init__(self, _format, a_bzrdir, control_files):
2557
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2558
self._transport = control_files._transport
2560
def is_shared(self):
2561
"""Return True if this repository is flagged as a shared repository."""
2562
return self._transport.has('shared-storage')
2565
def set_make_working_trees(self, new_value):
2566
"""Set the policy flag for making working trees when creating branches.
2568
This only applies to branches that use this repository.
2570
The default is 'True'.
2571
:param new_value: True to restore the default, False to disable making
2576
self._transport.delete('no-working-trees')
2577
except errors.NoSuchFile:
2580
self._transport.put_bytes('no-working-trees', '',
2581
mode=self.bzrdir._get_file_mode())
2583
def make_working_trees(self):
2584
"""Returns the policy for making working trees on new branches."""
2585
return not self._transport.has('no-working-trees')
2588
class MetaDirVersionedFileRepository(MetaDirRepository):
2589
"""Repositories in a meta-dir, that work via versioned file objects."""
2591
def __init__(self, _format, a_bzrdir, control_files):
2592
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2596
network_format_registry = registry.FormatRegistry()
2597
"""Registry of formats indexed by their network name.
2599
The network name for a repository format is an identifier that can be used when
2600
referring to formats with smart server operations. See
2601
RepositoryFormat.network_name() for more detail.
2605
format_registry = registry.FormatRegistry(network_format_registry)
2606
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
2608
This can contain either format instances themselves, or classes/factories that
2609
can be called to obtain one.
2613
#####################################################################
2614
# Repository Formats
2616
class RepositoryFormat(object):
2617
"""A repository format.
2619
Formats provide four things:
2620
* An initialization routine to construct repository data on disk.
2621
* a optional format string which is used when the BzrDir supports
2623
* an open routine which returns a Repository instance.
2624
* A network name for referring to the format in smart server RPC
2627
There is one and only one Format subclass for each on-disk format. But
2628
there can be one Repository subclass that is used for several different
2629
formats. The _format attribute on a Repository instance can be used to
2630
determine the disk format.
2632
Formats are placed in a registry by their format string for reference
2633
during opening. These should be subclasses of RepositoryFormat for
2636
Once a format is deprecated, just deprecate the initialize and open
2637
methods on the format class. Do not deprecate the object, as the
2638
object may be created even when a repository instnace hasn't been
2641
Common instance attributes:
2642
_matchingbzrdir - the bzrdir format that the repository format was
2643
originally written to work with. This can be used if manually
2644
constructing a bzrdir and repository, or more commonly for test suite
2648
# Set to True or False in derived classes. True indicates that the format
2649
# supports ghosts gracefully.
2650
supports_ghosts = None
2651
# Can this repository be given external locations to lookup additional
2652
# data. Set to True or False in derived classes.
2653
supports_external_lookups = None
2654
# What order should fetch operations request streams in?
2655
# The default is unordered as that is the cheapest for an origin to
2657
_fetch_order = 'unordered'
2658
# Does this repository format use deltas that can be fetched as-deltas ?
2659
# (E.g. knits, where the knit deltas can be transplanted intact.
2660
# We default to False, which will ensure that enough data to get
2661
# a full text out of any fetch stream will be grabbed.
2662
_fetch_uses_deltas = False
2663
# Should fetch trigger a reconcile after the fetch? Only needed for
2664
# some repository formats that can suffer internal inconsistencies.
2665
_fetch_reconcile = False
2666
# Does this format have < O(tree_size) delta generation. Used to hint what
2667
# code path for commit, amongst other things.
2671
return "<%s>" % self.__class__.__name__
2673
def __eq__(self, other):
2674
# format objects are generally stateless
2675
return isinstance(other, self.__class__)
2677
def __ne__(self, other):
2678
return not self == other
2681
def find_format(klass, a_bzrdir):
2682
"""Return the format for the repository object in a_bzrdir.
2684
This is used by bzr native formats that have a "format" file in
2685
the repository. Other methods may be used by different types of
2689
transport = a_bzrdir.get_repository_transport(None)
2690
format_string = transport.get("format").read()
2691
return format_registry.get(format_string)
2692
except errors.NoSuchFile:
2693
raise errors.NoRepositoryPresent(a_bzrdir)
2695
raise errors.UnknownFormatError(format=format_string,
2699
def register_format(klass, format):
2700
format_registry.register(format.get_format_string(), format)
2703
def unregister_format(klass, format):
2704
format_registry.remove(format.get_format_string())
2707
def get_default_format(klass):
2708
"""Return the current default format."""
2709
from bzrlib import bzrdir
2710
return bzrdir.format_registry.make_bzrdir('default').repository_format
2712
def get_format_string(self):
2713
"""Return the ASCII format string that identifies this format.
2715
Note that in pre format ?? repositories the format string is
2716
not permitted nor written to disk.
2718
raise NotImplementedError(self.get_format_string)
2720
def get_format_description(self):
2721
"""Return the short description for this format."""
2722
raise NotImplementedError(self.get_format_description)
2724
# TODO: this shouldn't be in the base class, it's specific to things that
2725
# use weaves or knits -- mbp 20070207
2726
def _get_versioned_file_store(self,
2731
versionedfile_class=None,
2732
versionedfile_kwargs={},
2734
if versionedfile_class is None:
2735
versionedfile_class = self._versionedfile_class
2736
weave_transport = control_files._transport.clone(name)
2737
dir_mode = control_files._dir_mode
2738
file_mode = control_files._file_mode
2739
return VersionedFileStore(weave_transport, prefixed=prefixed,
2741
file_mode=file_mode,
2742
versionedfile_class=versionedfile_class,
2743
versionedfile_kwargs=versionedfile_kwargs,
2746
def initialize(self, a_bzrdir, shared=False):
2747
"""Initialize a repository of this format in a_bzrdir.
2749
:param a_bzrdir: The bzrdir to put the new repository in it.
2750
:param shared: The repository should be initialized as a sharable one.
2751
:returns: The new repository object.
2753
This may raise UninitializableFormat if shared repository are not
2754
compatible the a_bzrdir.
2756
raise NotImplementedError(self.initialize)
2758
def is_supported(self):
2759
"""Is this format supported?
2761
Supported formats must be initializable and openable.
2762
Unsupported formats may not support initialization or committing or
2763
some other features depending on the reason for not being supported.
2767
def network_name(self):
2768
"""A simple byte string uniquely identifying this format for RPC calls.
2770
MetaDir repository formats use their disk format string to identify the
2771
repository over the wire. All in one formats such as bzr < 0.8, and
2772
foreign formats like svn/git and hg should use some marker which is
2773
unique and immutable.
2775
raise NotImplementedError(self.network_name)
2777
def check_conversion_target(self, target_format):
2778
raise NotImplementedError(self.check_conversion_target)
2780
def open(self, a_bzrdir, _found=False):
2781
"""Return an instance of this format for the bzrdir a_bzrdir.
2783
_found is a private parameter, do not use it.
2785
raise NotImplementedError(self.open)
2788
class MetaDirRepositoryFormat(RepositoryFormat):
2789
"""Common base class for the new repositories using the metadir layout."""
2791
rich_root_data = False
2792
supports_tree_reference = False
2793
supports_external_lookups = False
2796
def _matchingbzrdir(self):
2797
matching = bzrdir.BzrDirMetaFormat1()
2798
matching.repository_format = self
2802
super(MetaDirRepositoryFormat, self).__init__()
2804
def _create_control_files(self, a_bzrdir):
2805
"""Create the required files and the initial control_files object."""
2806
# FIXME: RBC 20060125 don't peek under the covers
2807
# NB: no need to escape relative paths that are url safe.
2808
repository_transport = a_bzrdir.get_repository_transport(self)
2809
control_files = lockable_files.LockableFiles(repository_transport,
2810
'lock', lockdir.LockDir)
2811
control_files.create_lock()
2812
return control_files
2814
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
2815
"""Upload the initial blank content."""
2816
control_files = self._create_control_files(a_bzrdir)
2817
control_files.lock_write()
2818
transport = control_files._transport
2820
utf8_files += [('shared-storage', '')]
2822
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
2823
for (filename, content_stream) in files:
2824
transport.put_file(filename, content_stream,
2825
mode=a_bzrdir._get_file_mode())
2826
for (filename, content_bytes) in utf8_files:
2827
transport.put_bytes_non_atomic(filename, content_bytes,
2828
mode=a_bzrdir._get_file_mode())
2830
control_files.unlock()
2832
def network_name(self):
2833
"""Metadir formats have matching disk and network format strings."""
2834
return self.get_format_string()
2837
# Pre-0.8 formats that don't have a disk format string (because they are
2838
# versioned by the matching control directory). We use the control directories
2839
# disk format string as a key for the network_name because they meet the
2840
# constraints (simple string, unique, immmutable).
2841
network_format_registry.register_lazy(
2842
"Bazaar-NG branch, format 5\n",
2843
'bzrlib.repofmt.weaverepo',
2844
'RepositoryFormat5',
2846
network_format_registry.register_lazy(
2847
"Bazaar-NG branch, format 6\n",
2848
'bzrlib.repofmt.weaverepo',
2849
'RepositoryFormat6',
2852
# formats which have no format string are not discoverable or independently
2853
# creatable on disk, so are not registered in format_registry. They're
2854
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
2855
# needed, it's constructed directly by the BzrDir. Non-native formats where
2856
# the repository is not separately opened are similar.
2858
format_registry.register_lazy(
2859
'Bazaar-NG Repository format 7',
2860
'bzrlib.repofmt.weaverepo',
2864
format_registry.register_lazy(
2865
'Bazaar-NG Knit Repository Format 1',
2866
'bzrlib.repofmt.knitrepo',
2867
'RepositoryFormatKnit1',
2870
format_registry.register_lazy(
2871
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
2872
'bzrlib.repofmt.knitrepo',
2873
'RepositoryFormatKnit3',
2876
format_registry.register_lazy(
2877
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
2878
'bzrlib.repofmt.knitrepo',
2879
'RepositoryFormatKnit4',
2882
# Pack-based formats. There is one format for pre-subtrees, and one for
2883
# post-subtrees to allow ease of testing.
2884
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
2885
format_registry.register_lazy(
2886
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
2887
'bzrlib.repofmt.pack_repo',
2888
'RepositoryFormatKnitPack1',
2890
format_registry.register_lazy(
2891
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
2892
'bzrlib.repofmt.pack_repo',
2893
'RepositoryFormatKnitPack3',
2895
format_registry.register_lazy(
2896
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
2897
'bzrlib.repofmt.pack_repo',
2898
'RepositoryFormatKnitPack4',
2900
format_registry.register_lazy(
2901
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
2902
'bzrlib.repofmt.pack_repo',
2903
'RepositoryFormatKnitPack5',
2905
format_registry.register_lazy(
2906
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
2907
'bzrlib.repofmt.pack_repo',
2908
'RepositoryFormatKnitPack5RichRoot',
2910
format_registry.register_lazy(
2911
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
2912
'bzrlib.repofmt.pack_repo',
2913
'RepositoryFormatKnitPack5RichRootBroken',
2915
format_registry.register_lazy(
2916
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
2917
'bzrlib.repofmt.pack_repo',
2918
'RepositoryFormatKnitPack6',
2920
format_registry.register_lazy(
2921
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
2922
'bzrlib.repofmt.pack_repo',
2923
'RepositoryFormatKnitPack6RichRoot',
2926
# Development formats.
2927
# 1.7->1.8 go below here
2928
format_registry.register_lazy(
2929
"Bazaar development format 2 (needs bzr.dev from before 1.8)\n",
2930
'bzrlib.repofmt.pack_repo',
2931
'RepositoryFormatPackDevelopment2',
2933
format_registry.register_lazy(
2934
("Bazaar development format 2 with subtree support "
2935
"(needs bzr.dev from before 1.8)\n"),
2936
'bzrlib.repofmt.pack_repo',
2937
'RepositoryFormatPackDevelopment2Subtree',
2941
class InterRepository(InterObject):
2942
"""This class represents operations taking place between two repositories.
2944
Its instances have methods like copy_content and fetch, and contain
2945
references to the source and target repositories these operations can be
2948
Often we will provide convenience methods on 'repository' which carry out
2949
operations with another repository - they will always forward to
2950
InterRepository.get(other).method_name(parameters).
2953
_walk_to_common_revisions_batch_size = 50
2955
"""The available optimised InterRepository types."""
2958
def copy_content(self, revision_id=None):
2959
"""Make a complete copy of the content in self into destination.
2961
This is a destructive operation! Do not use it on existing
2964
:param revision_id: Only copy the content needed to construct
2965
revision_id and its parents.
2968
self.target.set_make_working_trees(self.source.make_working_trees())
2969
except NotImplementedError:
2971
self.target.fetch(self.source, revision_id=revision_id)
2974
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
2976
"""Fetch the content required to construct revision_id.
2978
The content is copied from self.source to self.target.
2980
:param revision_id: if None all content is copied, if NULL_REVISION no
2982
:param pb: optional progress bar to use for progress reports. If not
2983
provided a default one will be created.
2986
from bzrlib.fetch import RepoFetcher
2987
f = RepoFetcher(to_repository=self.target,
2988
from_repository=self.source,
2989
last_revision=revision_id,
2990
fetch_spec=fetch_spec,
2991
pb=pb, find_ghosts=find_ghosts)
2993
def _walk_to_common_revisions(self, revision_ids):
2994
"""Walk out from revision_ids in source to revisions target has.
2996
:param revision_ids: The start point for the search.
2997
:return: A set of revision ids.
2999
target_graph = self.target.get_graph()
3000
revision_ids = frozenset(revision_ids)
3001
# Fast path for the case where all the revisions are already in the
3003
# (Although this does incur an extra round trip for the
3004
# fairly common case where the target doesn't already have the revision
3006
if set(target_graph.get_parent_map(revision_ids)) == revision_ids:
3007
return graph.SearchResult(revision_ids, set(), 0, set())
3008
missing_revs = set()
3009
source_graph = self.source.get_graph()
3010
# ensure we don't pay silly lookup costs.
3011
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3012
null_set = frozenset([_mod_revision.NULL_REVISION])
3013
searcher_exhausted = False
3017
# Iterate the searcher until we have enough next_revs
3018
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3020
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3021
next_revs.update(next_revs_part)
3022
ghosts.update(ghosts_part)
3023
except StopIteration:
3024
searcher_exhausted = True
3026
# If there are ghosts in the source graph, and the caller asked for
3027
# them, make sure that they are present in the target.
3028
# We don't care about other ghosts as we can't fetch them and
3029
# haven't been asked to.
3030
ghosts_to_check = set(revision_ids.intersection(ghosts))
3031
revs_to_get = set(next_revs).union(ghosts_to_check)
3033
have_revs = set(target_graph.get_parent_map(revs_to_get))
3034
# we always have NULL_REVISION present.
3035
have_revs = have_revs.union(null_set)
3036
# Check if the target is missing any ghosts we need.
3037
ghosts_to_check.difference_update(have_revs)
3039
# One of the caller's revision_ids is a ghost in both the
3040
# source and the target.
3041
raise errors.NoSuchRevision(
3042
self.source, ghosts_to_check.pop())
3043
missing_revs.update(next_revs - have_revs)
3044
# Because we may have walked past the original stop point, make
3045
# sure everything is stopped
3046
stop_revs = searcher.find_seen_ancestors(have_revs)
3047
searcher.stop_searching_any(stop_revs)
3048
if searcher_exhausted:
3050
return searcher.get_result()
3053
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3054
"""Return the revision ids that source has that target does not.
3056
:param revision_id: only return revision ids included by this
3058
:param find_ghosts: If True find missing revisions in deep history
3059
rather than just finding the surface difference.
3060
:return: A bzrlib.graph.SearchResult.
3062
# stop searching at found target revisions.
3063
if not find_ghosts and revision_id is not None:
3064
return self._walk_to_common_revisions([revision_id])
3065
# generic, possibly worst case, slow code path.
3066
target_ids = set(self.target.all_revision_ids())
3067
if revision_id is not None:
3068
source_ids = self.source.get_ancestry(revision_id)
3069
if source_ids[0] is not None:
3070
raise AssertionError()
3073
source_ids = self.source.all_revision_ids()
3074
result_set = set(source_ids).difference(target_ids)
3075
return self.source.revision_ids_to_search_result(result_set)
3078
def _same_model(source, target):
3079
"""True if source and target have the same data representation.
3081
Note: this is always called on the base class; overriding it in a
3082
subclass will have no effect.
3085
InterRepository._assert_same_model(source, target)
3087
except errors.IncompatibleRepositories, e:
3091
def _assert_same_model(source, target):
3092
"""Raise an exception if two repositories do not use the same model.
3094
if source.supports_rich_root() != target.supports_rich_root():
3095
raise errors.IncompatibleRepositories(source, target,
3096
"different rich-root support")
3097
if source._serializer != target._serializer:
3098
raise errors.IncompatibleRepositories(source, target,
3099
"different serializers")
3102
class InterSameDataRepository(InterRepository):
3103
"""Code for converting between repositories that represent the same data.
3105
Data format and model must match for this to work.
3109
def _get_repo_format_to_test(self):
3110
"""Repository format for testing with.
3112
InterSameData can pull from subtree to subtree and from non-subtree to
3113
non-subtree, so we test this with the richest repository format.
3115
from bzrlib.repofmt import knitrepo
3116
return knitrepo.RepositoryFormatKnit3()
3119
def is_compatible(source, target):
3120
return InterRepository._same_model(source, target)
3123
class InterWeaveRepo(InterSameDataRepository):
3124
"""Optimised code paths between Weave based repositories.
3126
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3127
implemented lazy inter-object optimisation.
3131
def _get_repo_format_to_test(self):
3132
from bzrlib.repofmt import weaverepo
3133
return weaverepo.RepositoryFormat7()
3136
def is_compatible(source, target):
3137
"""Be compatible with known Weave formats.
3139
We don't test for the stores being of specific types because that
3140
could lead to confusing results, and there is no need to be
3143
from bzrlib.repofmt.weaverepo import (
3149
return (isinstance(source._format, (RepositoryFormat5,
3151
RepositoryFormat7)) and
3152
isinstance(target._format, (RepositoryFormat5,
3154
RepositoryFormat7)))
3155
except AttributeError:
3159
def copy_content(self, revision_id=None):
3160
"""See InterRepository.copy_content()."""
3161
# weave specific optimised path:
3163
self.target.set_make_working_trees(self.source.make_working_trees())
3164
except (errors.RepositoryUpgradeRequired, NotImplemented):
3166
# FIXME do not peek!
3167
if self.source._transport.listable():
3168
pb = ui.ui_factory.nested_progress_bar()
3170
self.target.texts.insert_record_stream(
3171
self.source.texts.get_record_stream(
3172
self.source.texts.keys(), 'topological', False))
3173
pb.update('copying inventory', 0, 1)
3174
self.target.inventories.insert_record_stream(
3175
self.source.inventories.get_record_stream(
3176
self.source.inventories.keys(), 'topological', False))
3177
self.target.signatures.insert_record_stream(
3178
self.source.signatures.get_record_stream(
3179
self.source.signatures.keys(),
3181
self.target.revisions.insert_record_stream(
3182
self.source.revisions.get_record_stream(
3183
self.source.revisions.keys(),
3184
'topological', True))
3188
self.target.fetch(self.source, revision_id=revision_id)
3191
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3192
"""See InterRepository.missing_revision_ids()."""
3193
# we want all revisions to satisfy revision_id in source.
3194
# but we don't want to stat every file here and there.
3195
# we want then, all revisions other needs to satisfy revision_id
3196
# checked, but not those that we have locally.
3197
# so the first thing is to get a subset of the revisions to
3198
# satisfy revision_id in source, and then eliminate those that
3199
# we do already have.
3200
# this is slow on high latency connection to self, but as as this
3201
# disk format scales terribly for push anyway due to rewriting
3202
# inventory.weave, this is considered acceptable.
3204
if revision_id is not None:
3205
source_ids = self.source.get_ancestry(revision_id)
3206
if source_ids[0] is not None:
3207
raise AssertionError()
3210
source_ids = self.source._all_possible_ids()
3211
source_ids_set = set(source_ids)
3212
# source_ids is the worst possible case we may need to pull.
3213
# now we want to filter source_ids against what we actually
3214
# have in target, but don't try to check for existence where we know
3215
# we do not have a revision as that would be pointless.
3216
target_ids = set(self.target._all_possible_ids())
3217
possibly_present_revisions = target_ids.intersection(source_ids_set)
3218
actually_present_revisions = set(
3219
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3220
required_revisions = source_ids_set.difference(actually_present_revisions)
3221
if revision_id is not None:
3222
# we used get_ancestry to determine source_ids then we are assured all
3223
# revisions referenced are present as they are installed in topological order.
3224
# and the tip revision was validated by get_ancestry.
3225
result_set = required_revisions
3227
# if we just grabbed the possibly available ids, then
3228
# we only have an estimate of whats available and need to validate
3229
# that against the revision records.
3231
self.source._eliminate_revisions_not_present(required_revisions))
3232
return self.source.revision_ids_to_search_result(result_set)
3235
class InterKnitRepo(InterSameDataRepository):
3236
"""Optimised code paths between Knit based repositories."""
3239
def _get_repo_format_to_test(self):
3240
from bzrlib.repofmt import knitrepo
3241
return knitrepo.RepositoryFormatKnit1()
3244
def is_compatible(source, target):
3245
"""Be compatible with known Knit formats.
3247
We don't test for the stores being of specific types because that
3248
could lead to confusing results, and there is no need to be
3251
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3253
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3254
isinstance(target._format, RepositoryFormatKnit))
3255
except AttributeError:
3257
return are_knits and InterRepository._same_model(source, target)
3260
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3261
"""See InterRepository.missing_revision_ids()."""
3262
if revision_id is not None:
3263
source_ids = self.source.get_ancestry(revision_id)
3264
if source_ids[0] is not None:
3265
raise AssertionError()
3268
source_ids = self.source.all_revision_ids()
3269
source_ids_set = set(source_ids)
3270
# source_ids is the worst possible case we may need to pull.
3271
# now we want to filter source_ids against what we actually
3272
# have in target, but don't try to check for existence where we know
3273
# we do not have a revision as that would be pointless.
3274
target_ids = set(self.target.all_revision_ids())
3275
possibly_present_revisions = target_ids.intersection(source_ids_set)
3276
actually_present_revisions = set(
3277
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3278
required_revisions = source_ids_set.difference(actually_present_revisions)
3279
if revision_id is not None:
3280
# we used get_ancestry to determine source_ids then we are assured all
3281
# revisions referenced are present as they are installed in topological order.
3282
# and the tip revision was validated by get_ancestry.
3283
result_set = required_revisions
3285
# if we just grabbed the possibly available ids, then
3286
# we only have an estimate of whats available and need to validate
3287
# that against the revision records.
3289
self.source._eliminate_revisions_not_present(required_revisions))
3290
return self.source.revision_ids_to_search_result(result_set)
3293
class InterPackRepo(InterSameDataRepository):
3294
"""Optimised code paths between Pack based repositories."""
3297
def _get_repo_format_to_test(self):
3298
from bzrlib.repofmt import pack_repo
3299
return pack_repo.RepositoryFormatKnitPack1()
3302
def is_compatible(source, target):
3303
"""Be compatible with known Pack formats.
3305
We don't test for the stores being of specific types because that
3306
could lead to confusing results, and there is no need to be
3309
from bzrlib.repofmt.pack_repo import RepositoryFormatPack
3311
are_packs = (isinstance(source._format, RepositoryFormatPack) and
3312
isinstance(target._format, RepositoryFormatPack))
3313
except AttributeError:
3315
return are_packs and InterRepository._same_model(source, target)
3318
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3320
"""See InterRepository.fetch()."""
3321
if (len(self.source._fallback_repositories) > 0 or
3322
len(self.target._fallback_repositories) > 0):
3323
# The pack layer is not aware of fallback repositories, so when
3324
# fetching from a stacked repository or into a stacked repository
3325
# we use the generic fetch logic which uses the VersionedFiles
3326
# attributes on repository.
3327
from bzrlib.fetch import RepoFetcher
3328
fetcher = RepoFetcher(self.target, self.source, revision_id,
3329
pb, find_ghosts, fetch_spec=fetch_spec)
3330
if fetch_spec is not None:
3331
if len(list(fetch_spec.heads)) != 1:
3332
raise AssertionError(
3333
"InterPackRepo.fetch doesn't support "
3334
"fetching multiple heads yet.")
3335
revision_id = list(fetch_spec.heads)[0]
3337
if revision_id is None:
3339
# everything to do - use pack logic
3340
# to fetch from all packs to one without
3341
# inventory parsing etc, IFF nothing to be copied is in the target.
3343
source_revision_ids = frozenset(self.source.all_revision_ids())
3344
revision_ids = source_revision_ids - \
3345
frozenset(self.target.get_parent_map(source_revision_ids))
3346
revision_keys = [(revid,) for revid in revision_ids]
3347
index = self.target._pack_collection.revision_index.combined_index
3348
present_revision_ids = set(item[1][0] for item in
3349
index.iter_entries(revision_keys))
3350
revision_ids = set(revision_ids) - present_revision_ids
3351
# implementing the TODO will involve:
3352
# - detecting when all of a pack is selected
3353
# - avoiding as much as possible pre-selection, so the
3354
# more-core routines such as create_pack_from_packs can filter in
3355
# a just-in-time fashion. (though having a HEADS list on a
3356
# repository might make this a lot easier, because we could
3357
# sensibly detect 'new revisions' without doing a full index scan.
3358
elif _mod_revision.is_null(revision_id):
3363
revision_ids = self.search_missing_revision_ids(revision_id,
3364
find_ghosts=find_ghosts).get_keys()
3365
except errors.NoSuchRevision:
3366
raise errors.InstallFailed([revision_id])
3367
if len(revision_ids) == 0:
3369
return self._pack(self.source, self.target, revision_ids)
3371
def _pack(self, source, target, revision_ids):
3372
from bzrlib.repofmt.pack_repo import Packer
3373
packs = source._pack_collection.all_packs()
3374
pack = Packer(self.target._pack_collection, packs, '.fetch',
3375
revision_ids).pack()
3376
if pack is not None:
3377
self.target._pack_collection._save_pack_names()
3378
copied_revs = pack.get_revision_count()
3379
# Trigger an autopack. This may duplicate effort as we've just done
3380
# a pack creation, but for now it is simpler to think about as
3381
# 'upload data, then repack if needed'.
3382
self.target._pack_collection.autopack()
3383
return (copied_revs, [])
3388
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3389
"""See InterRepository.missing_revision_ids().
3391
:param find_ghosts: Find ghosts throughout the ancestry of
3394
if not find_ghosts and revision_id is not None:
3395
return self._walk_to_common_revisions([revision_id])
3396
elif revision_id is not None:
3397
# Find ghosts: search for revisions pointing from one repository to
3398
# the other, and vice versa, anywhere in the history of revision_id.
3399
graph = self.target.get_graph(other_repository=self.source)
3400
searcher = graph._make_breadth_first_searcher([revision_id])
3404
next_revs, ghosts = searcher.next_with_ghosts()
3405
except StopIteration:
3407
if revision_id in ghosts:
3408
raise errors.NoSuchRevision(self.source, revision_id)
3409
found_ids.update(next_revs)
3410
found_ids.update(ghosts)
3411
found_ids = frozenset(found_ids)
3412
# Double query here: should be able to avoid this by changing the
3413
# graph api further.
3414
result_set = found_ids - frozenset(
3415
self.target.get_parent_map(found_ids))
3417
source_ids = self.source.all_revision_ids()
3418
# source_ids is the worst possible case we may need to pull.
3419
# now we want to filter source_ids against what we actually
3420
# have in target, but don't try to check for existence where we know
3421
# we do not have a revision as that would be pointless.
3422
target_ids = set(self.target.all_revision_ids())
3423
result_set = set(source_ids).difference(target_ids)
3424
return self.source.revision_ids_to_search_result(result_set)
3427
class InterDifferingSerializer(InterKnitRepo):
3430
def _get_repo_format_to_test(self):
3434
def is_compatible(source, target):
3435
"""Be compatible with Knit2 source and Knit3 target"""
3436
if source.supports_rich_root() != target.supports_rich_root():
3438
# Ideally, we'd support fetching if the source had no tree references
3439
# even if it supported them...
3440
if (getattr(source, '_format.supports_tree_reference', False) and
3441
not getattr(target, '_format.supports_tree_reference', False)):
3445
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3446
"""Get the best delta and base for this revision.
3448
:return: (basis_id, delta)
3450
possible_trees = [(parent_id, cache[parent_id])
3451
for parent_id in parent_ids
3452
if parent_id in cache]
3453
if len(possible_trees) == 0:
3454
# There either aren't any parents, or the parents aren't in the
3455
# cache, so just use the last converted tree
3456
possible_trees.append((basis_id, cache[basis_id]))
3458
for basis_id, basis_tree in possible_trees:
3459
delta = tree.inventory._make_delta(basis_tree.inventory)
3460
deltas.append((len(delta), basis_id, delta))
3462
return deltas[0][1:]
3464
def _fetch_batch(self, revision_ids, basis_id, cache):
3465
"""Fetch across a few revisions.
3467
:param revision_ids: The revisions to copy
3468
:param basis_id: The revision_id of a tree that must be in cache, used
3469
as a basis for delta when no other base is available
3470
:param cache: A cache of RevisionTrees that we can use.
3471
:return: The revision_id of the last converted tree. The RevisionTree
3472
for it will be in cache
3474
# Walk though all revisions; get inventory deltas, copy referenced
3475
# texts that delta references, insert the delta, revision and
3479
pending_revisions = []
3480
parent_map = self.source.get_parent_map(revision_ids)
3481
for tree in self.source.revision_trees(revision_ids):
3482
current_revision_id = tree.get_revision_id()
3483
parent_ids = parent_map.get(current_revision_id, ())
3484
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3486
# Find text entries that need to be copied
3487
for old_path, new_path, file_id, entry in delta:
3488
if new_path is not None:
3489
if not (new_path or self.target.supports_rich_root()):
3490
# We don't copy the text for the root node unless the
3491
# target supports_rich_root.
3493
text_keys.add((file_id, entry.revision))
3494
revision = self.source.get_revision(current_revision_id)
3495
pending_deltas.append((basis_id, delta,
3496
current_revision_id, revision.parent_ids))
3497
pending_revisions.append(revision)
3498
cache[current_revision_id] = tree
3499
basis_id = current_revision_id
3501
from_texts = self.source.texts
3502
to_texts = self.target.texts
3503
to_texts.insert_record_stream(from_texts.get_record_stream(
3504
text_keys, self.target._format._fetch_order,
3505
not self.target._format._fetch_uses_deltas))
3507
for delta in pending_deltas:
3508
self.target.add_inventory_by_delta(*delta)
3509
# insert signatures and revisions
3510
for revision in pending_revisions:
3512
signature = self.source.get_signature_text(
3513
revision.revision_id)
3514
self.target.add_signature_text(revision.revision_id,
3516
except errors.NoSuchRevision:
3518
self.target.add_revision(revision.revision_id, revision)
3521
def _fetch_all_revisions(self, revision_ids, pb):
3522
"""Fetch everything for the list of revisions.
3524
:param revision_ids: The list of revisions to fetch. Must be in
3526
:param pb: A ProgressBar
3529
basis_id, basis_tree = self._get_basis(revision_ids[0])
3531
cache = lru_cache.LRUCache(100)
3532
cache[basis_id] = basis_tree
3533
del basis_tree # We don't want to hang on to it here
3534
for offset in range(0, len(revision_ids), batch_size):
3535
self.target.start_write_group()
3537
pb.update('Transferring revisions', offset,
3539
batch = revision_ids[offset:offset+batch_size]
3540
basis_id = self._fetch_batch(batch, basis_id, cache)
3542
self.target.abort_write_group()
3545
self.target.commit_write_group()
3546
pb.update('Transferring revisions', len(revision_ids),
3550
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3552
"""See InterRepository.fetch()."""
3553
if fetch_spec is not None:
3554
raise AssertionError("Not implemented yet...")
3555
revision_ids = self.target.search_missing_revision_ids(self.source,
3556
revision_id, find_ghosts=find_ghosts).get_keys()
3557
if not revision_ids:
3559
revision_ids = tsort.topo_sort(
3560
self.source.get_graph().get_parent_map(revision_ids))
3562
my_pb = ui.ui_factory.nested_progress_bar()
3565
symbol_versioning.warn(
3566
symbol_versioning.deprecated_in((1, 14, 0))
3567
% "pb parameter to fetch()")
3570
self._fetch_all_revisions(revision_ids, pb)
3572
if my_pb is not None:
3574
return len(revision_ids), 0
3576
def _get_basis(self, first_revision_id):
3577
"""Get a revision and tree which exists in the target.
3579
This assumes that first_revision_id is selected for transmission
3580
because all other ancestors are already present. If we can't find an
3581
ancestor we fall back to NULL_REVISION since we know that is safe.
3583
:return: (basis_id, basis_tree)
3585
first_rev = self.source.get_revision(first_revision_id)
3587
basis_id = first_rev.parent_ids[0]
3588
# only valid as a basis if the target has it
3589
self.target.get_revision(basis_id)
3590
# Try to get a basis tree - if its a ghost it will hit the
3591
# NoSuchRevision case.
3592
basis_tree = self.source.revision_tree(basis_id)
3593
except (IndexError, errors.NoSuchRevision):
3594
basis_id = _mod_revision.NULL_REVISION
3595
basis_tree = self.source.revision_tree(basis_id)
3596
return basis_id, basis_tree
3599
InterRepository.register_optimiser(InterDifferingSerializer)
3600
InterRepository.register_optimiser(InterSameDataRepository)
3601
InterRepository.register_optimiser(InterWeaveRepo)
3602
InterRepository.register_optimiser(InterKnitRepo)
3603
InterRepository.register_optimiser(InterPackRepo)
3606
class CopyConverter(object):
3607
"""A repository conversion tool which just performs a copy of the content.
3609
This is slow but quite reliable.
3612
def __init__(self, target_format):
3613
"""Create a CopyConverter.
3615
:param target_format: The format the resulting repository should be.
3617
self.target_format = target_format
3619
def convert(self, repo, pb):
3620
"""Perform the conversion of to_convert, giving feedback via pb.
3622
:param to_convert: The disk object to convert.
3623
:param pb: a progress bar to use for progress information.
3628
# this is only useful with metadir layouts - separated repo content.
3629
# trigger an assertion if not such
3630
repo._format.get_format_string()
3631
self.repo_dir = repo.bzrdir
3632
self.step('Moving repository to repository.backup')
3633
self.repo_dir.transport.move('repository', 'repository.backup')
3634
backup_transport = self.repo_dir.transport.clone('repository.backup')
3635
repo._format.check_conversion_target(self.target_format)
3636
self.source_repo = repo._format.open(self.repo_dir,
3638
_override_transport=backup_transport)
3639
self.step('Creating new repository')
3640
converted = self.target_format.initialize(self.repo_dir,
3641
self.source_repo.is_shared())
3642
converted.lock_write()
3644
self.step('Copying content into repository.')
3645
self.source_repo.copy_content_into(converted)
3648
self.step('Deleting old repository content.')
3649
self.repo_dir.transport.delete_tree('repository.backup')
3650
self.pb.note('repository converted')
3652
def step(self, message):
3653
"""Update the pb by a step."""
3655
self.pb.update(message, self.count, self.total)
3667
def _unescaper(match, _map=_unescape_map):
3668
code = match.group(1)
3672
if not code.startswith('#'):
3674
return unichr(int(code[1:])).encode('utf8')
3680
def _unescape_xml(data):
3681
"""Unescape predefined XML entities in a string of data."""
3683
if _unescape_re is None:
3684
_unescape_re = re.compile('\&([^;]*);')
3685
return _unescape_re.sub(_unescaper, data)
3688
class _VersionedFileChecker(object):
3690
def __init__(self, repository, text_key_references=None):
3691
self.repository = repository
3692
self.text_index = self.repository._generate_text_key_index(
3693
text_key_references=text_key_references)
3695
def calculate_file_version_parents(self, text_key):
3696
"""Calculate the correct parents for a file version according to
3699
parent_keys = self.text_index[text_key]
3700
if parent_keys == [_mod_revision.NULL_REVISION]:
3702
return tuple(parent_keys)
3704
def check_file_version_parents(self, texts, progress_bar=None):
3705
"""Check the parents stored in a versioned file are correct.
3707
It also detects file versions that are not referenced by their
3708
corresponding revision's inventory.
3710
:returns: A tuple of (wrong_parents, dangling_file_versions).
3711
wrong_parents is a dict mapping {revision_id: (stored_parents,
3712
correct_parents)} for each revision_id where the stored parents
3713
are not correct. dangling_file_versions is a set of (file_id,
3714
revision_id) tuples for versions that are present in this versioned
3715
file, but not used by the corresponding inventory.
3718
self.file_ids = set([file_id for file_id, _ in
3719
self.text_index.iterkeys()])
3720
# text keys is now grouped by file_id
3721
n_weaves = len(self.file_ids)
3722
files_in_revisions = {}
3723
revisions_of_files = {}
3724
n_versions = len(self.text_index)
3725
progress_bar.update('loading text store', 0, n_versions)
3726
parent_map = self.repository.texts.get_parent_map(self.text_index)
3727
# On unlistable transports this could well be empty/error...
3728
text_keys = self.repository.texts.keys()
3729
unused_keys = frozenset(text_keys) - set(self.text_index)
3730
for num, key in enumerate(self.text_index.iterkeys()):
3731
if progress_bar is not None:
3732
progress_bar.update('checking text graph', num, n_versions)
3733
correct_parents = self.calculate_file_version_parents(key)
3735
knit_parents = parent_map[key]
3736
except errors.RevisionNotPresent:
3739
if correct_parents != knit_parents:
3740
wrong_parents[key] = (knit_parents, correct_parents)
3741
return wrong_parents, unused_keys
3744
def _old_get_graph(repository, revision_id):
3745
"""DO NOT USE. That is all. I'm serious."""
3746
graph = repository.get_graph()
3747
revision_graph = dict(((key, value) for key, value in
3748
graph.iter_ancestry([revision_id]) if value is not None))
3749
return _strip_NULL_ghosts(revision_graph)
3752
def _strip_NULL_ghosts(revision_graph):
3753
"""Also don't use this. more compatibility code for unmigrated clients."""
3754
# Filter ghosts, and null:
3755
if _mod_revision.NULL_REVISION in revision_graph:
3756
del revision_graph[_mod_revision.NULL_REVISION]
3757
for key, parents in revision_graph.items():
3758
revision_graph[key] = tuple(parent for parent in parents if parent
3760
return revision_graph
3763
class StreamSink(object):
3764
"""An object that can insert a stream into a repository.
3766
This interface handles the complexity of reserialising inventories and
3767
revisions from different formats, and allows unidirectional insertion into
3768
stacked repositories without looking for the missing basis parents
3772
def __init__(self, target_repo):
3773
self.target_repo = target_repo
3775
def insert_stream(self, stream, src_format, resume_tokens):
3776
"""Insert a stream's content into the target repository.
3778
:param src_format: a bzr repository format.
3780
:return: a list of resume tokens and an iterable of keys additional
3781
items required before the insertion can be completed.
3783
self.target_repo.lock_write()
3786
self.target_repo.resume_write_group(resume_tokens)
3788
self.target_repo.start_write_group()
3790
# locked_insert_stream performs a commit|suspend.
3791
return self._locked_insert_stream(stream, src_format)
3793
self.target_repo.abort_write_group(suppress_errors=True)
3796
self.target_repo.unlock()
3798
def _locked_insert_stream(self, stream, src_format):
3799
to_serializer = self.target_repo._format._serializer
3800
src_serializer = src_format._serializer
3801
if to_serializer == src_serializer:
3802
# If serializers match and the target is a pack repository, set the
3803
# write cache size on the new pack. This avoids poor performance
3804
# on transports where append is unbuffered (such as
3805
# RemoteTransport). This is safe to do because nothing should read
3806
# back from the target repository while a stream with matching
3807
# serialization is being inserted.
3808
# The exception is that a delta record from the source that should
3809
# be a fulltext may need to be expanded by the target (see
3810
# test_fetch_revisions_with_deltas_into_pack); but we take care to
3811
# explicitly flush any buffered writes first in that rare case.
3813
new_pack = self.target_repo._pack_collection._new_pack
3814
except AttributeError:
3815
# Not a pack repository
3818
new_pack.set_write_cache_size(1024*1024)
3819
for substream_type, substream in stream:
3820
if substream_type == 'texts':
3821
self.target_repo.texts.insert_record_stream(substream)
3822
elif substream_type == 'inventories':
3823
if src_serializer == to_serializer:
3824
self.target_repo.inventories.insert_record_stream(
3827
self._extract_and_insert_inventories(
3828
substream, src_serializer)
3829
elif substream_type == 'revisions':
3830
# This may fallback to extract-and-insert more often than
3831
# required if the serializers are different only in terms of
3833
if src_serializer == to_serializer:
3834
self.target_repo.revisions.insert_record_stream(
3837
self._extract_and_insert_revisions(substream,
3839
elif substream_type == 'signatures':
3840
self.target_repo.signatures.insert_record_stream(substream)
3842
raise AssertionError('kaboom! %s' % (substream_type,))
3844
missing_keys = set()
3845
for prefix, versioned_file in (
3846
('texts', self.target_repo.texts),
3847
('inventories', self.target_repo.inventories),
3848
('revisions', self.target_repo.revisions),
3849
('signatures', self.target_repo.signatures),
3851
missing_keys.update((prefix,) + key for key in
3852
versioned_file.get_missing_compression_parent_keys())
3853
except NotImplementedError:
3854
# cannot even attempt suspending, and missing would have failed
3855
# during stream insertion.
3856
missing_keys = set()
3859
# suspend the write group and tell the caller what we is
3860
# missing. We know we can suspend or else we would not have
3861
# entered this code path. (All repositories that can handle
3862
# missing keys can handle suspending a write group).
3863
write_group_tokens = self.target_repo.suspend_write_group()
3864
return write_group_tokens, missing_keys
3865
self.target_repo.commit_write_group()
3868
def _extract_and_insert_inventories(self, substream, serializer):
3869
"""Generate a new inventory versionedfile in target, converting data.
3871
The inventory is retrieved from the source, (deserializing it), and
3872
stored in the target (reserializing it in a different format).
3874
for record in substream:
3875
bytes = record.get_bytes_as('fulltext')
3876
revision_id = record.key[0]
3877
inv = serializer.read_inventory_from_string(bytes, revision_id)
3878
parents = [key[0] for key in record.parents]
3879
self.target_repo.add_inventory(revision_id, inv, parents)
3881
def _extract_and_insert_revisions(self, substream, serializer):
3882
for record in substream:
3883
bytes = record.get_bytes_as('fulltext')
3884
revision_id = record.key[0]
3885
rev = serializer.read_revision_from_string(bytes)
3886
if rev.revision_id != revision_id:
3887
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
3888
self.target_repo.add_revision(revision_id, rev)
3891
if self.target_repo._format._fetch_reconcile:
3892
self.target_repo.reconcile()
3895
class StreamSource(object):
3896
"""A source of a stream for fetching between repositories."""
3898
def __init__(self, from_repository, to_format):
3899
"""Create a StreamSource streaming from from_repository."""
3900
self.from_repository = from_repository
3901
self.to_format = to_format
3903
def delta_on_metadata(self):
3904
"""Return True if delta's are permitted on metadata streams.
3906
That is on revisions and signatures.
3908
src_serializer = self.from_repository._format._serializer
3909
target_serializer = self.to_format._serializer
3910
return (self.to_format._fetch_uses_deltas and
3911
src_serializer == target_serializer)
3913
def _fetch_revision_texts(self, revs):
3914
# fetch signatures first and then the revision texts
3915
# may need to be a InterRevisionStore call here.
3916
from_sf = self.from_repository.signatures
3917
# A missing signature is just skipped.
3918
keys = [(rev_id,) for rev_id in revs]
3919
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
3921
self.to_format._fetch_order,
3922
not self.to_format._fetch_uses_deltas))
3923
# If a revision has a delta, this is actually expanded inside the
3924
# insert_record_stream code now, which is an alternate fix for
3926
from_rf = self.from_repository.revisions
3927
revisions = from_rf.get_record_stream(
3929
self.to_format._fetch_order,
3930
not self.delta_on_metadata())
3931
return [('signatures', signatures), ('revisions', revisions)]
3933
def _generate_root_texts(self, revs):
3934
"""This will be called by __fetch between fetching weave texts and
3935
fetching the inventory weave.
3937
Subclasses should override this if they need to generate root texts
3938
after fetching weave texts.
3940
if self._rich_root_upgrade():
3942
return bzrlib.fetch.Inter1and2Helper(
3943
self.from_repository).generate_root_texts(revs)
3947
def get_stream(self, search):
3949
revs = search.get_keys()
3950
graph = self.from_repository.get_graph()
3951
revs = list(graph.iter_topo_order(revs))
3952
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
3954
for knit_kind, file_id, revisions in data_to_fetch:
3955
if knit_kind != phase:
3957
# Make a new progress bar for this phase
3958
if knit_kind == "file":
3959
# Accumulate file texts
3960
text_keys.extend([(file_id, revision) for revision in
3962
elif knit_kind == "inventory":
3963
# Now copy the file texts.
3964
from_texts = self.from_repository.texts
3965
yield ('texts', from_texts.get_record_stream(
3966
text_keys, self.to_format._fetch_order,
3967
not self.to_format._fetch_uses_deltas))
3968
# Cause an error if a text occurs after we have done the
3971
# Before we process the inventory we generate the root
3972
# texts (if necessary) so that the inventories references
3974
for _ in self._generate_root_texts(revs):
3976
# NB: This currently reopens the inventory weave in source;
3977
# using a single stream interface instead would avoid this.
3978
from_weave = self.from_repository.inventories
3979
# we fetch only the referenced inventories because we do not
3980
# know for unselected inventories whether all their required
3981
# texts are present in the other repository - it could be
3983
yield ('inventories', from_weave.get_record_stream(
3984
[(rev_id,) for rev_id in revs],
3985
self.inventory_fetch_order(),
3986
not self.delta_on_metadata()))
3987
elif knit_kind == "signatures":
3988
# Nothing to do here; this will be taken care of when
3989
# _fetch_revision_texts happens.
3991
elif knit_kind == "revisions":
3992
for record in self._fetch_revision_texts(revs):
3995
raise AssertionError("Unknown knit kind %r" % knit_kind)
3997
def get_stream_for_missing_keys(self, missing_keys):
3998
# missing keys can only occur when we are byte copying and not
3999
# translating (because translation means we don't send
4000
# unreconstructable deltas ever).
4002
keys['texts'] = set()
4003
keys['revisions'] = set()
4004
keys['inventories'] = set()
4005
keys['signatures'] = set()
4006
for key in missing_keys:
4007
keys[key[0]].add(key[1:])
4008
if len(keys['revisions']):
4009
# If we allowed copying revisions at this point, we could end up
4010
# copying a revision without copying its required texts: a
4011
# violation of the requirements for repository integrity.
4012
raise AssertionError(
4013
'cannot copy revisions to fill in missing deltas %s' % (
4014
keys['revisions'],))
4015
for substream_kind, keys in keys.iteritems():
4016
vf = getattr(self.from_repository, substream_kind)
4017
# Ask for full texts always so that we don't need more round trips
4018
# after this stream.
4019
stream = vf.get_record_stream(keys,
4020
self.to_format._fetch_order, True)
4021
yield substream_kind, stream
4023
def inventory_fetch_order(self):
4024
if self._rich_root_upgrade():
4025
return 'topological'
4027
return self.to_format._fetch_order
4029
def _rich_root_upgrade(self):
4030
return (not self.from_repository._format.rich_root_data and
4031
self.to_format.rich_root_data)
added
removed
bzrlib/repository.py
