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(), """
39
revision as _mod_revision,
45
from bzrlib.bundle import serializer
46
from bzrlib.revisiontree import RevisionTree
47
from bzrlib.store.versioned import VersionedFileStore
48
from bzrlib.testament import Testament
51
from bzrlib.decorators import needs_read_lock, needs_write_lock
52
from bzrlib.inter import InterObject
53
from bzrlib.inventory import (
59
from bzrlib import registry
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
text = file_obj.read()
498
ie.text_sha1, ie.text_size = self._add_text_to_weave(
499
ie.file_id, text, 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
517
self._add_text_to_weave(ie.file_id, '', heads, None)
518
elif kind == 'symlink':
519
current_link_target = content_summary[3]
521
# symlink target is not generic metadata, check if it has
523
if current_link_target != parent_entry.symlink_target:
526
# unchanged, carry over.
527
ie.revision = parent_entry.revision
528
ie.symlink_target = parent_entry.symlink_target
529
return self._get_delta(ie, basis_inv, path), False, None
530
ie.symlink_target = current_link_target
531
self._add_text_to_weave(ie.file_id, '', heads, None)
532
elif kind == 'tree-reference':
534
if content_summary[3] != parent_entry.reference_revision:
537
# unchanged, carry over.
538
ie.reference_revision = parent_entry.reference_revision
539
ie.revision = parent_entry.revision
540
return self._get_delta(ie, basis_inv, path), False, None
541
ie.reference_revision = content_summary[3]
542
self._add_text_to_weave(ie.file_id, '', heads, None)
544
raise NotImplementedError('unknown kind')
545
ie.revision = self._new_revision_id
546
self._any_changes = True
547
return self._get_delta(ie, basis_inv, path), True, fingerprint
549
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
550
_entry_factory=entry_factory):
551
"""Record a new tree via iter_changes.
553
:param tree: The tree to obtain text contents from for changed objects.
554
:param basis_revision_id: The revision id of the tree the iter_changes
555
has been generated against. Currently assumed to be the same
556
as self.parents[0] - if it is not, errors may occur.
557
:param iter_changes: An iter_changes iterator with the changes to apply
558
to basis_revision_id. The iterator must not include any items with
559
a current kind of None - missing items must be either filtered out
560
or errored-on beefore record_iter_changes sees the item.
561
:param _entry_factory: Private method to bind entry_factory locally for
563
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
566
# Create an inventory delta based on deltas between all the parents and
567
# deltas between all the parent inventories. We use inventory delta's
568
# between the inventory objects because iter_changes masks
569
# last-changed-field only changes.
571
# file_id -> change map, change is fileid, paths, changed, versioneds,
572
# parents, names, kinds, executables
574
# {file_id -> revision_id -> inventory entry, for entries in parent
575
# trees that are not parents[0]
579
revtrees = list(self.repository.revision_trees(self.parents))
580
except errors.NoSuchRevision:
581
# one or more ghosts, slow path.
583
for revision_id in self.parents:
585
revtrees.append(self.repository.revision_tree(revision_id))
586
except errors.NoSuchRevision:
588
basis_revision_id = _mod_revision.NULL_REVISION
590
revtrees.append(self.repository.revision_tree(
591
_mod_revision.NULL_REVISION))
592
# The basis inventory from a repository
594
basis_inv = revtrees[0].inventory
596
basis_inv = self.repository.revision_tree(
597
_mod_revision.NULL_REVISION).inventory
598
if len(self.parents) > 0:
599
if basis_revision_id != self.parents[0] and not ghost_basis:
601
"arbitrary basis parents not yet supported with merges")
602
for revtree in revtrees[1:]:
603
for change in revtree.inventory._make_delta(basis_inv):
604
if change[1] is None:
605
# Not present in this parent.
607
if change[2] not in merged_ids:
608
if change[0] is not None:
609
basis_entry = basis_inv[change[2]]
610
merged_ids[change[2]] = [
612
basis_entry.revision,
615
parent_entries[change[2]] = {
617
basis_entry.revision:basis_entry,
619
change[3].revision:change[3],
622
merged_ids[change[2]] = [change[3].revision]
623
parent_entries[change[2]] = {change[3].revision:change[3]}
625
merged_ids[change[2]].append(change[3].revision)
626
parent_entries[change[2]][change[3].revision] = change[3]
629
# Setup the changes from the tree:
630
# changes maps file_id -> (change, [parent revision_ids])
632
for change in iter_changes:
633
# This probably looks up in basis_inv way to much.
634
if change[1][0] is not None:
635
head_candidate = [basis_inv[change[0]].revision]
638
changes[change[0]] = change, merged_ids.get(change[0],
640
unchanged_merged = set(merged_ids) - set(changes)
641
# Extend the changes dict with synthetic changes to record merges of
643
for file_id in unchanged_merged:
644
# Record a merged version of these items that did not change vs the
645
# basis. This can be either identical parallel changes, or a revert
646
# of a specific file after a merge. The recorded content will be
647
# that of the current tree (which is the same as the basis), but
648
# the per-file graph will reflect a merge.
649
# NB:XXX: We are reconstructing path information we had, this
650
# should be preserved instead.
651
# inv delta change: (file_id, (path_in_source, path_in_target),
652
# changed_content, versioned, parent, name, kind,
655
basis_entry = basis_inv[file_id]
656
except errors.NoSuchId:
657
# a change from basis->some_parents but file_id isn't in basis
658
# so was new in the merge, which means it must have changed
659
# from basis -> current, and as it hasn't the add was reverted
660
# by the user. So we discard this change.
664
(basis_inv.id2path(file_id), tree.id2path(file_id)),
666
(basis_entry.parent_id, basis_entry.parent_id),
667
(basis_entry.name, basis_entry.name),
668
(basis_entry.kind, basis_entry.kind),
669
(basis_entry.executable, basis_entry.executable))
670
changes[file_id] = (change, merged_ids[file_id])
671
# changes contains tuples with the change and a set of inventory
672
# candidates for the file.
674
# old_path, new_path, file_id, new_inventory_entry
675
seen_root = False # Is the root in the basis delta?
676
inv_delta = self._basis_delta
677
modified_rev = self._new_revision_id
678
for change, head_candidates in changes.values():
679
if change[3][1]: # versioned in target.
680
# Several things may be happening here:
681
# We may have a fork in the per-file graph
682
# - record a change with the content from tree
683
# We may have a change against < all trees
684
# - carry over the tree that hasn't changed
685
# We may have a change against all trees
686
# - record the change with the content from tree
689
entry = _entry_factory[kind](file_id, change[5][1],
691
head_set = self._heads(change[0], set(head_candidates))
694
for head_candidate in head_candidates:
695
if head_candidate in head_set:
696
heads.append(head_candidate)
697
head_set.remove(head_candidate)
700
# Could be a carry-over situation:
701
parent_entry_revs = parent_entries.get(file_id, None)
702
if parent_entry_revs:
703
parent_entry = parent_entry_revs.get(heads[0], None)
706
if parent_entry is None:
707
# The parent iter_changes was called against is the one
708
# that is the per-file head, so any change is relevant
709
# iter_changes is valid.
710
carry_over_possible = False
712
# could be a carry over situation
713
# A change against the basis may just indicate a merge,
714
# we need to check the content against the source of the
715
# merge to determine if it was changed after the merge
717
if (parent_entry.kind != entry.kind or
718
parent_entry.parent_id != entry.parent_id or
719
parent_entry.name != entry.name):
720
# Metadata common to all entries has changed
721
# against per-file parent
722
carry_over_possible = False
724
carry_over_possible = True
725
# per-type checks for changes against the parent_entry
728
# Cannot be a carry-over situation
729
carry_over_possible = False
730
# Populate the entry in the delta
732
# XXX: There is still a small race here: If someone reverts the content of a file
733
# after iter_changes examines and decides it has changed,
734
# we will unconditionally record a new version even if some
735
# other process reverts it while commit is running (with
736
# the revert happening after iter_changes did it's
739
entry.executable = True
741
entry.executable = False
742
if (carry_over_possible and
743
parent_entry.executable == entry.executable):
744
# Check the file length, content hash after reading
746
nostore_sha = parent_entry.text_sha1
749
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
751
text = file_obj.read()
755
entry.text_sha1, entry.text_size = self._add_text_to_weave(
756
file_id, text, heads, nostore_sha)
757
yield file_id, change[1][1], (entry.text_sha1, stat_value)
758
except errors.ExistingContent:
759
# No content change against a carry_over parent
760
# Perhaps this should also yield a fs hash update?
762
entry.text_size = parent_entry.text_size
763
entry.text_sha1 = parent_entry.text_sha1
764
elif kind == 'symlink':
766
entry.symlink_target = tree.get_symlink_target(file_id)
767
if (carry_over_possible and
768
parent_entry.symlink_target == entry.symlink_target):
771
self._add_text_to_weave(change[0], '', heads, None)
772
elif kind == 'directory':
773
if carry_over_possible:
776
# Nothing to set on the entry.
777
# XXX: split into the Root and nonRoot versions.
778
if change[1][1] != '' or self.repository.supports_rich_root():
779
self._add_text_to_weave(change[0], '', heads, None)
780
elif kind == 'tree-reference':
781
if not self.repository._format.supports_tree_reference:
782
# This isn't quite sane as an error, but we shouldn't
783
# ever see this code path in practice: tree's don't
784
# permit references when the repo doesn't support tree
786
raise errors.UnsupportedOperation(tree.add_reference,
788
reference_revision = tree.get_reference_revision(change[0])
789
entry.reference_revision = reference_revision
790
if (carry_over_possible and
791
parent_entry.reference_revision == reference_revision):
794
self._add_text_to_weave(change[0], '', heads, None)
796
raise AssertionError('unknown kind %r' % kind)
798
entry.revision = modified_rev
800
entry.revision = parent_entry.revision
803
new_path = change[1][1]
804
inv_delta.append((change[1][0], new_path, change[0], entry))
807
self.new_inventory = None
809
self._any_changes = True
811
# housekeeping root entry changes do not affect no-change commits.
812
self._require_root_change(tree)
813
self.basis_delta_revision = basis_revision_id
815
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
816
parent_keys = tuple([(file_id, parent) for parent in parents])
817
return self.repository.texts._add_text(
818
(file_id, self._new_revision_id), parent_keys, new_text,
819
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
822
class RootCommitBuilder(CommitBuilder):
823
"""This commitbuilder actually records the root id"""
825
# the root entry gets versioned properly by this builder.
826
_versioned_root = True
828
def _check_root(self, ie, parent_invs, tree):
829
"""Helper for record_entry_contents.
831
:param ie: An entry being added.
832
:param parent_invs: The inventories of the parent revisions of the
834
:param tree: The tree that is being committed.
837
def _require_root_change(self, tree):
838
"""Enforce an appropriate root object change.
840
This is called once when record_iter_changes is called, if and only if
841
the root was not in the delta calculated by record_iter_changes.
843
:param tree: The tree which is being committed.
845
# versioned roots do not change unless the tree found a change.
848
######################################################################
852
class Repository(object):
853
"""Repository holding history for one or more branches.
855
The repository holds and retrieves historical information including
856
revisions and file history. It's normally accessed only by the Branch,
857
which views a particular line of development through that history.
859
The Repository builds on top of some byte storage facilies (the revisions,
860
signatures, inventories, texts and chk_bytes attributes) and a Transport,
861
which respectively provide byte storage and a means to access the (possibly
864
The byte storage facilities are addressed via tuples, which we refer to
865
as 'keys' throughout the code base. Revision_keys, inventory_keys and
866
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
867
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
868
byte string made up of a hash identifier and a hash value.
869
We use this interface because it allows low friction with the underlying
870
code that implements disk indices, network encoding and other parts of
873
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
874
the serialised revisions for the repository. This can be used to obtain
875
revision graph information or to access raw serialised revisions.
876
The result of trying to insert data into the repository via this store
877
is undefined: it should be considered read-only except for implementors
879
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
880
the serialised signatures for the repository. This can be used to
881
obtain access to raw serialised signatures. The result of trying to
882
insert data into the repository via this store is undefined: it should
883
be considered read-only except for implementors of repositories.
884
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
885
the serialised inventories for the repository. This can be used to
886
obtain unserialised inventories. The result of trying to insert data
887
into the repository via this store is undefined: it should be
888
considered read-only except for implementors of repositories.
889
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
890
texts of files and directories for the repository. This can be used to
891
obtain file texts or file graphs. Note that Repository.iter_file_bytes
892
is usually a better interface for accessing file texts.
893
The result of trying to insert data into the repository via this store
894
is undefined: it should be considered read-only except for implementors
896
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
897
any data the repository chooses to store or have indexed by its hash.
898
The result of trying to insert data into the repository via this store
899
is undefined: it should be considered read-only except for implementors
901
:ivar _transport: Transport for file access to repository, typically
902
pointing to .bzr/repository.
905
# What class to use for a CommitBuilder. Often its simpler to change this
906
# in a Repository class subclass rather than to override
907
# get_commit_builder.
908
_commit_builder_class = CommitBuilder
909
# The search regex used by xml based repositories to determine what things
910
# where changed in a single commit.
911
_file_ids_altered_regex = lazy_regex.lazy_compile(
912
r'file_id="(?P<file_id>[^"]+)"'
913
r'.* revision="(?P<revision_id>[^"]+)"'
916
def abort_write_group(self, suppress_errors=False):
917
"""Commit the contents accrued within the current write group.
919
:param suppress_errors: if true, abort_write_group will catch and log
920
unexpected errors that happen during the abort, rather than
921
allowing them to propagate. Defaults to False.
923
:seealso: start_write_group.
925
if self._write_group is not self.get_transaction():
926
# has an unlock or relock occured ?
927
raise errors.BzrError(
928
'mismatched lock context and write group. %r, %r' %
929
(self._write_group, self.get_transaction()))
931
self._abort_write_group()
932
except Exception, exc:
933
self._write_group = None
934
if not suppress_errors:
936
mutter('abort_write_group failed')
937
log_exception_quietly()
938
note('bzr: ERROR (ignored): %s', exc)
939
self._write_group = None
941
def _abort_write_group(self):
942
"""Template method for per-repository write group cleanup.
944
This is called during abort before the write group is considered to be
945
finished and should cleanup any internal state accrued during the write
946
group. There is no requirement that data handed to the repository be
947
*not* made available - this is not a rollback - but neither should any
948
attempt be made to ensure that data added is fully commited. Abort is
949
invoked when an error has occured so futher disk or network operations
950
may not be possible or may error and if possible should not be
954
def add_fallback_repository(self, repository):
955
"""Add a repository to use for looking up data not held locally.
957
:param repository: A repository.
959
if not self._format.supports_external_lookups:
960
raise errors.UnstackableRepositoryFormat(self._format, self.base)
962
# This repository will call fallback.unlock() when we transition to
963
# the unlocked state, so we make sure to increment the lock count
964
repository.lock_read()
965
self._check_fallback_repository(repository)
966
self._fallback_repositories.append(repository)
967
self.texts.add_fallback_versioned_files(repository.texts)
968
self.inventories.add_fallback_versioned_files(repository.inventories)
969
self.revisions.add_fallback_versioned_files(repository.revisions)
970
self.signatures.add_fallback_versioned_files(repository.signatures)
971
if self.chk_bytes is not None:
972
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
974
def _check_fallback_repository(self, repository):
975
"""Check that this repository can fallback to repository safely.
977
Raise an error if not.
979
:param repository: A repository to fallback to.
981
return InterRepository._assert_same_model(self, repository)
983
def add_inventory(self, revision_id, inv, parents):
984
"""Add the inventory inv to the repository as revision_id.
986
:param parents: The revision ids of the parents that revision_id
987
is known to have and are in the repository already.
989
:returns: The validator(which is a sha1 digest, though what is sha'd is
990
repository format specific) of the serialized inventory.
992
if not self.is_in_write_group():
993
raise AssertionError("%r not in write group" % (self,))
994
_mod_revision.check_not_reserved_id(revision_id)
995
if not (inv.revision_id is None or inv.revision_id == revision_id):
996
raise AssertionError(
997
"Mismatch between inventory revision"
998
" id and insertion revid (%r, %r)"
999
% (inv.revision_id, revision_id))
1000
if inv.root is None:
1001
raise AssertionError()
1002
return self._add_inventory_checked(revision_id, inv, parents)
1004
def _add_inventory_checked(self, revision_id, inv, parents):
1005
"""Add inv to the repository after checking the inputs.
1007
This function can be overridden to allow different inventory styles.
1009
:seealso: add_inventory, for the contract.
1011
inv_lines = self._serialise_inventory_to_lines(inv)
1012
return self._inventory_add_lines(revision_id, parents,
1013
inv_lines, check_content=False)
1015
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1016
parents, basis_inv=None, propagate_caches=False):
1017
"""Add a new inventory expressed as a delta against another revision.
1019
See the inventory developers documentation for the theory behind
1022
:param basis_revision_id: The inventory id the delta was created
1023
against. (This does not have to be a direct parent.)
1024
:param delta: The inventory delta (see Inventory.apply_delta for
1026
:param new_revision_id: The revision id that the inventory is being
1028
:param parents: The revision ids of the parents that revision_id is
1029
known to have and are in the repository already. These are supplied
1030
for repositories that depend on the inventory graph for revision
1031
graph access, as well as for those that pun ancestry with delta
1033
:param basis_inv: The basis inventory if it is already known,
1035
:param propagate_caches: If True, the caches for this inventory are
1036
copied to and updated for the result if possible.
1038
:returns: (validator, new_inv)
1039
The validator(which is a sha1 digest, though what is sha'd is
1040
repository format specific) of the serialized inventory, and the
1041
resulting inventory.
1043
if not self.is_in_write_group():
1044
raise AssertionError("%r not in write group" % (self,))
1045
_mod_revision.check_not_reserved_id(new_revision_id)
1046
basis_tree = self.revision_tree(basis_revision_id)
1047
basis_tree.lock_read()
1049
# Note that this mutates the inventory of basis_tree, which not all
1050
# inventory implementations may support: A better idiom would be to
1051
# return a new inventory, but as there is no revision tree cache in
1052
# repository this is safe for now - RBC 20081013
1053
if basis_inv is None:
1054
basis_inv = basis_tree.inventory
1055
basis_inv.apply_delta(delta)
1056
basis_inv.revision_id = new_revision_id
1057
return (self.add_inventory(new_revision_id, basis_inv, parents),
1062
def _inventory_add_lines(self, revision_id, parents, lines,
1063
check_content=True):
1064
"""Store lines in inv_vf and return the sha1 of the inventory."""
1065
parents = [(parent,) for parent in parents]
1066
return self.inventories.add_lines((revision_id,), parents, lines,
1067
check_content=check_content)[0]
1069
def add_revision(self, revision_id, rev, inv=None, config=None):
1070
"""Add rev to the revision store as revision_id.
1072
:param revision_id: the revision id to use.
1073
:param rev: The revision object.
1074
:param inv: The inventory for the revision. if None, it will be looked
1075
up in the inventory storer
1076
:param config: If None no digital signature will be created.
1077
If supplied its signature_needed method will be used
1078
to determine if a signature should be made.
1080
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1082
_mod_revision.check_not_reserved_id(revision_id)
1083
if config is not None and config.signature_needed():
1085
inv = self.get_inventory(revision_id)
1086
plaintext = Testament(rev, inv).as_short_text()
1087
self.store_revision_signature(
1088
gpg.GPGStrategy(config), plaintext, revision_id)
1089
# check inventory present
1090
if not self.inventories.get_parent_map([(revision_id,)]):
1092
raise errors.WeaveRevisionNotPresent(revision_id,
1095
# yes, this is not suitable for adding with ghosts.
1096
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1099
key = (revision_id,)
1100
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1101
self._add_revision(rev)
1103
def _add_revision(self, revision):
1104
text = self._serializer.write_revision_to_string(revision)
1105
key = (revision.revision_id,)
1106
parents = tuple((parent,) for parent in revision.parent_ids)
1107
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1109
def all_revision_ids(self):
1110
"""Returns a list of all the revision ids in the repository.
1112
This is conceptually deprecated because code should generally work on
1113
the graph reachable from a particular revision, and ignore any other
1114
revisions that might be present. There is no direct replacement
1117
if 'evil' in debug.debug_flags:
1118
mutter_callsite(2, "all_revision_ids is linear with history.")
1119
return self._all_revision_ids()
1121
def _all_revision_ids(self):
1122
"""Returns a list of all the revision ids in the repository.
1124
These are in as much topological order as the underlying store can
1127
raise NotImplementedError(self._all_revision_ids)
1129
def break_lock(self):
1130
"""Break a lock if one is present from another instance.
1132
Uses the ui factory to ask for confirmation if the lock may be from
1135
self.control_files.break_lock()
1138
def _eliminate_revisions_not_present(self, revision_ids):
1139
"""Check every revision id in revision_ids to see if we have it.
1141
Returns a set of the present revisions.
1144
graph = self.get_graph()
1145
parent_map = graph.get_parent_map(revision_ids)
1146
# The old API returned a list, should this actually be a set?
1147
return parent_map.keys()
1149
def _check_inventories(self, checker):
1150
"""Check the inventories found from the revision scan.
1152
This is responsible for verifying the sha1 of inventories and
1153
creating a pending_keys set that covers data referenced by inventories.
1155
bar = ui.ui_factory.nested_progress_bar()
1157
self._do_check_inventories(checker, bar)
1161
def _do_check_inventories(self, checker, bar):
1162
"""Helper for _check_inventories."""
1164
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1165
kinds = ['chk_bytes', 'texts']
1166
count = len(checker.pending_keys)
1167
bar.update("inventories", 0, 2)
1168
current_keys = checker.pending_keys
1169
checker.pending_keys = {}
1170
# Accumulate current checks.
1171
for key in current_keys:
1172
if key[0] != 'inventories' and key[0] not in kinds:
1173
checker._report_items.append('unknown key type %r' % (key,))
1174
keys[key[0]].add(key[1:])
1175
if keys['inventories']:
1176
# NB: output order *should* be roughly sorted - topo or
1177
# inverse topo depending on repository - either way decent
1178
# to just delta against. However, pre-CHK formats didn't
1179
# try to optimise inventory layout on disk. As such the
1180
# pre-CHK code path does not use inventory deltas.
1182
for record in self.inventories.check(keys=keys['inventories']):
1183
if record.storage_kind == 'absent':
1184
checker._report_items.append(
1185
'Missing inventory {%s}' % (record.key,))
1187
last_object = self._check_record('inventories', record,
1188
checker, last_object,
1189
current_keys[('inventories',) + record.key])
1190
del keys['inventories']
1193
bar.update("texts", 1)
1194
while (checker.pending_keys or keys['chk_bytes']
1196
# Something to check.
1197
current_keys = checker.pending_keys
1198
checker.pending_keys = {}
1199
# Accumulate current checks.
1200
for key in current_keys:
1201
if key[0] not in kinds:
1202
checker._report_items.append('unknown key type %r' % (key,))
1203
keys[key[0]].add(key[1:])
1204
# Check the outermost kind only - inventories || chk_bytes || texts
1208
for record in getattr(self, kind).check(keys=keys[kind]):
1209
if record.storage_kind == 'absent':
1210
checker._report_items.append(
1211
'Missing inventory {%s}' % (record.key,))
1213
last_object = self._check_record(kind, record,
1214
checker, last_object, current_keys[(kind,) + record.key])
1218
def _check_record(self, kind, record, checker, last_object, item_data):
1219
"""Check a single text from this repository."""
1220
if kind == 'inventories':
1221
rev_id = record.key[0]
1222
inv = self.deserialise_inventory(rev_id,
1223
record.get_bytes_as('fulltext'))
1224
if last_object is not None:
1225
delta = inv._make_delta(last_object)
1226
for old_path, path, file_id, ie in delta:
1229
ie.check(checker, rev_id, inv)
1231
for path, ie in inv.iter_entries():
1232
ie.check(checker, rev_id, inv)
1233
if self._format.fast_deltas:
1235
elif kind == 'chk_bytes':
1236
# No code written to check chk_bytes for this repo format.
1237
checker._report_items.append(
1238
'unsupported key type chk_bytes for %s' % (record.key,))
1239
elif kind == 'texts':
1240
self._check_text(record, checker, item_data)
1242
checker._report_items.append(
1243
'unknown key type %s for %s' % (kind, record.key))
1245
def _check_text(self, record, checker, item_data):
1246
"""Check a single text."""
1247
# Check it is extractable.
1248
# TODO: check length.
1249
if record.storage_kind == 'chunked':
1250
chunks = record.get_bytes_as(record.storage_kind)
1251
sha1 = osutils.sha_strings(chunks)
1252
length = sum(map(len, chunks))
1254
content = record.get_bytes_as('fulltext')
1255
sha1 = osutils.sha_string(content)
1256
length = len(content)
1257
if item_data and sha1 != item_data[1]:
1258
checker._report_items.append(
1259
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1260
(record.key, sha1, item_data[1], item_data[2]))
1263
def create(a_bzrdir):
1264
"""Construct the current default format repository in a_bzrdir."""
1265
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
1267
def __init__(self, _format, a_bzrdir, control_files):
1268
"""instantiate a Repository.
1270
:param _format: The format of the repository on disk.
1271
:param a_bzrdir: The BzrDir of the repository.
1273
In the future we will have a single api for all stores for
1274
getting file texts, inventories and revisions, then
1275
this construct will accept instances of those things.
1277
super(Repository, self).__init__()
1278
self._format = _format
1279
# the following are part of the public API for Repository:
1280
self.bzrdir = a_bzrdir
1281
self.control_files = control_files
1282
self._transport = control_files._transport
1283
self.base = self._transport.base
1285
self._reconcile_does_inventory_gc = True
1286
self._reconcile_fixes_text_parents = False
1287
self._reconcile_backsup_inventory = True
1288
# not right yet - should be more semantically clear ?
1290
# TODO: make sure to construct the right store classes, etc, depending
1291
# on whether escaping is required.
1292
self._warn_if_deprecated()
1293
self._write_group = None
1294
# Additional places to query for data.
1295
self._fallback_repositories = []
1296
# An InventoryEntry cache, used during deserialization
1297
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1300
if self._fallback_repositories:
1301
return '%s(%r, fallback_repositories=%r)' % (
1302
self.__class__.__name__,
1304
self._fallback_repositories)
1306
return '%s(%r)' % (self.__class__.__name__,
1309
def _has_same_fallbacks(self, other_repo):
1310
"""Returns true if the repositories have the same fallbacks."""
1311
my_fb = self._fallback_repositories
1312
other_fb = other_repo._fallback_repositories
1313
if len(my_fb) != len(other_fb):
1315
for f, g in zip(my_fb, other_fb):
1316
if not f.has_same_location(g):
1320
def has_same_location(self, other):
1321
"""Returns a boolean indicating if this repository is at the same
1322
location as another repository.
1324
This might return False even when two repository objects are accessing
1325
the same physical repository via different URLs.
1327
if self.__class__ is not other.__class__:
1329
return (self._transport.base == other._transport.base)
1331
def is_in_write_group(self):
1332
"""Return True if there is an open write group.
1334
:seealso: start_write_group.
1336
return self._write_group is not None
1338
def is_locked(self):
1339
return self.control_files.is_locked()
1341
def is_write_locked(self):
1342
"""Return True if this object is write locked."""
1343
return self.is_locked() and self.control_files._lock_mode == 'w'
1345
def lock_write(self, token=None):
1346
"""Lock this repository for writing.
1348
This causes caching within the repository obejct to start accumlating
1349
data during reads, and allows a 'write_group' to be obtained. Write
1350
groups must be used for actual data insertion.
1352
:param token: if this is already locked, then lock_write will fail
1353
unless the token matches the existing lock.
1354
:returns: a token if this instance supports tokens, otherwise None.
1355
:raises TokenLockingNotSupported: when a token is given but this
1356
instance doesn't support using token locks.
1357
:raises MismatchedToken: if the specified token doesn't match the token
1358
of the existing lock.
1359
:seealso: start_write_group.
1361
A token should be passed in if you know that you have locked the object
1362
some other way, and need to synchronise this object's state with that
1365
XXX: this docstring is duplicated in many places, e.g. lockable_files.py
1367
locked = self.is_locked()
1368
result = self.control_files.lock_write(token=token)
1370
for repo in self._fallback_repositories:
1371
# Writes don't affect fallback repos
1373
self._refresh_data()
1376
def lock_read(self):
1377
locked = self.is_locked()
1378
self.control_files.lock_read()
1380
for repo in self._fallback_repositories:
1382
self._refresh_data()
1384
def get_physical_lock_status(self):
1385
return self.control_files.get_physical_lock_status()
1387
def leave_lock_in_place(self):
1388
"""Tell this repository not to release the physical lock when this
1391
If lock_write doesn't return a token, then this method is not supported.
1393
self.control_files.leave_in_place()
1395
def dont_leave_lock_in_place(self):
1396
"""Tell this repository to release the physical lock when this
1397
object is unlocked, even if it didn't originally acquire it.
1399
If lock_write doesn't return a token, then this method is not supported.
1401
self.control_files.dont_leave_in_place()
1404
def gather_stats(self, revid=None, committers=None):
1405
"""Gather statistics from a revision id.
1407
:param revid: The revision id to gather statistics from, if None, then
1408
no revision specific statistics are gathered.
1409
:param committers: Optional parameter controlling whether to grab
1410
a count of committers from the revision specific statistics.
1411
:return: A dictionary of statistics. Currently this contains:
1412
committers: The number of committers if requested.
1413
firstrev: A tuple with timestamp, timezone for the penultimate left
1414
most ancestor of revid, if revid is not the NULL_REVISION.
1415
latestrev: A tuple with timestamp, timezone for revid, if revid is
1416
not the NULL_REVISION.
1417
revisions: The total revision count in the repository.
1418
size: An estimate disk size of the repository in bytes.
1421
if revid and committers:
1422
result['committers'] = 0
1423
if revid and revid != _mod_revision.NULL_REVISION:
1425
all_committers = set()
1426
revisions = self.get_ancestry(revid)
1427
# pop the leading None
1429
first_revision = None
1431
# ignore the revisions in the middle - just grab first and last
1432
revisions = revisions[0], revisions[-1]
1433
for revision in self.get_revisions(revisions):
1434
if not first_revision:
1435
first_revision = revision
1437
all_committers.add(revision.committer)
1438
last_revision = revision
1440
result['committers'] = len(all_committers)
1441
result['firstrev'] = (first_revision.timestamp,
1442
first_revision.timezone)
1443
result['latestrev'] = (last_revision.timestamp,
1444
last_revision.timezone)
1446
# now gather global repository information
1447
# XXX: This is available for many repos regardless of listability.
1448
if self.bzrdir.root_transport.listable():
1449
# XXX: do we want to __define len__() ?
1450
# Maybe the versionedfiles object should provide a different
1451
# method to get the number of keys.
1452
result['revisions'] = len(self.revisions.keys())
1453
# result['size'] = t
1456
def find_branches(self, using=False):
1457
"""Find branches underneath this repository.
1459
This will include branches inside other branches.
1461
:param using: If True, list only branches using this repository.
1463
if using and not self.is_shared():
1465
return [self.bzrdir.open_branch()]
1466
except errors.NotBranchError:
1468
class Evaluator(object):
1471
self.first_call = True
1473
def __call__(self, bzrdir):
1474
# On the first call, the parameter is always the bzrdir
1475
# containing the current repo.
1476
if not self.first_call:
1478
repository = bzrdir.open_repository()
1479
except errors.NoRepositoryPresent:
1482
return False, (None, repository)
1483
self.first_call = False
1485
value = (bzrdir.open_branch(), None)
1486
except errors.NotBranchError:
1487
value = (None, None)
1491
for branch, repository in bzrdir.BzrDir.find_bzrdirs(
1492
self.bzrdir.root_transport, evaluate=Evaluator()):
1493
if branch is not None:
1494
branches.append(branch)
1495
if not using and repository is not None:
1496
branches.extend(repository.find_branches())
1500
def search_missing_revision_ids(self, other, revision_id=None, find_ghosts=True):
1501
"""Return the revision ids that other has that this does not.
1503
These are returned in topological order.
1505
revision_id: only return revision ids included by revision_id.
1507
return InterRepository.get(other, self).search_missing_revision_ids(
1508
revision_id, find_ghosts)
1512
"""Open the repository rooted at base.
1514
For instance, if the repository is at URL/.bzr/repository,
1515
Repository.open(URL) -> a Repository instance.
1517
control = bzrdir.BzrDir.open(base)
1518
return control.open_repository()
1520
def copy_content_into(self, destination, revision_id=None):
1521
"""Make a complete copy of the content in self into destination.
1523
This is a destructive operation! Do not use it on existing
1526
return InterRepository.get(self, destination).copy_content(revision_id)
1528
def commit_write_group(self):
1529
"""Commit the contents accrued within the current write group.
1531
:seealso: start_write_group.
1533
if self._write_group is not self.get_transaction():
1534
# has an unlock or relock occured ?
1535
raise errors.BzrError('mismatched lock context %r and '
1537
(self.get_transaction(), self._write_group))
1538
result = self._commit_write_group()
1539
self._write_group = None
1542
def _commit_write_group(self):
1543
"""Template method for per-repository write group cleanup.
1545
This is called before the write group is considered to be
1546
finished and should ensure that all data handed to the repository
1547
for writing during the write group is safely committed (to the
1548
extent possible considering file system caching etc).
1551
def suspend_write_group(self):
1552
raise errors.UnsuspendableWriteGroup(self)
1554
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1555
"""Return the keys of missing inventory parents for revisions added in
1558
A revision is not complete if the inventory delta for that revision
1559
cannot be calculated. Therefore if the parent inventories of a
1560
revision are not present, the revision is incomplete, and e.g. cannot
1561
be streamed by a smart server. This method finds missing inventory
1562
parents for revisions added in this write group.
1564
if not self._format.supports_external_lookups:
1565
# This is only an issue for stacked repositories
1567
if not self.is_in_write_group():
1568
raise AssertionError('not in a write group')
1570
# XXX: We assume that every added revision already has its
1571
# corresponding inventory, so we only check for parent inventories that
1572
# might be missing, rather than all inventories.
1573
parents = set(self.revisions._index.get_missing_parents())
1574
parents.discard(_mod_revision.NULL_REVISION)
1575
unstacked_inventories = self.inventories._index
1576
present_inventories = unstacked_inventories.get_parent_map(
1577
key[-1:] for key in parents)
1578
parents.difference_update(present_inventories)
1579
if len(parents) == 0:
1580
# No missing parent inventories.
1582
if not check_for_missing_texts:
1583
return set(('inventories', rev_id) for (rev_id,) in parents)
1584
# Ok, now we have a list of missing inventories. But these only matter
1585
# if the inventories that reference them are missing some texts they
1586
# appear to introduce.
1587
# XXX: Texts referenced by all added inventories need to be present,
1588
# but at the moment we're only checking for texts referenced by
1589
# inventories at the graph's edge.
1590
key_deps = self.revisions._index._key_dependencies
1591
key_deps.add_keys(present_inventories)
1592
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1593
file_ids = self.fileids_altered_by_revision_ids(referrers)
1594
missing_texts = set()
1595
for file_id, version_ids in file_ids.iteritems():
1596
missing_texts.update(
1597
(file_id, version_id) for version_id in version_ids)
1598
present_texts = self.texts.get_parent_map(missing_texts)
1599
missing_texts.difference_update(present_texts)
1600
if not missing_texts:
1601
# No texts are missing, so all revisions and their deltas are
1604
# Alternatively the text versions could be returned as the missing
1605
# keys, but this is likely to be less data.
1606
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1609
def refresh_data(self):
1610
"""Re-read any data needed to to synchronise with disk.
1612
This method is intended to be called after another repository instance
1613
(such as one used by a smart server) has inserted data into the
1614
repository. It may not be called during a write group, but may be
1615
called at any other time.
1617
if self.is_in_write_group():
1618
raise errors.InternalBzrError(
1619
"May not refresh_data while in a write group.")
1620
self._refresh_data()
1622
def resume_write_group(self, tokens):
1623
if not self.is_write_locked():
1624
raise errors.NotWriteLocked(self)
1625
if self._write_group:
1626
raise errors.BzrError('already in a write group')
1627
self._resume_write_group(tokens)
1628
# so we can detect unlock/relock - the write group is now entered.
1629
self._write_group = self.get_transaction()
1631
def _resume_write_group(self, tokens):
1632
raise errors.UnsuspendableWriteGroup(self)
1634
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
1636
"""Fetch the content required to construct revision_id from source.
1638
If revision_id is None and fetch_spec is None, then all content is
1641
fetch() may not be used when the repository is in a write group -
1642
either finish the current write group before using fetch, or use
1643
fetch before starting the write group.
1645
:param find_ghosts: Find and copy revisions in the source that are
1646
ghosts in the target (and not reachable directly by walking out to
1647
the first-present revision in target from revision_id).
1648
:param revision_id: If specified, all the content needed for this
1649
revision ID will be copied to the target. Fetch will determine for
1650
itself which content needs to be copied.
1651
:param fetch_spec: If specified, a SearchResult or
1652
PendingAncestryResult that describes which revisions to copy. This
1653
allows copying multiple heads at once. Mutually exclusive with
1656
if fetch_spec is not None and revision_id is not None:
1657
raise AssertionError(
1658
"fetch_spec and revision_id are mutually exclusive.")
1659
if self.is_in_write_group():
1660
raise errors.InternalBzrError(
1661
"May not fetch while in a write group.")
1662
# fast path same-url fetch operations
1663
# TODO: lift out to somewhere common with RemoteRepository
1664
# <https://bugs.edge.launchpad.net/bzr/+bug/401646>
1665
if (self.has_same_location(source)
1666
and fetch_spec is None
1667
and self._has_same_fallbacks(source)):
1668
# check that last_revision is in 'from' and then return a
1670
if (revision_id is not None and
1671
not _mod_revision.is_null(revision_id)):
1672
self.get_revision(revision_id)
1674
# if there is no specific appropriate InterRepository, this will get
1675
# the InterRepository base class, which raises an
1676
# IncompatibleRepositories when asked to fetch.
1677
inter = InterRepository.get(source, self)
1678
return inter.fetch(revision_id=revision_id, pb=pb,
1679
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1681
def create_bundle(self, target, base, fileobj, format=None):
1682
return serializer.write_bundle(self, target, base, fileobj, format)
1684
def get_commit_builder(self, branch, parents, config, timestamp=None,
1685
timezone=None, committer=None, revprops=None,
1687
"""Obtain a CommitBuilder for this repository.
1689
:param branch: Branch to commit to.
1690
:param parents: Revision ids of the parents of the new revision.
1691
:param config: Configuration to use.
1692
:param timestamp: Optional timestamp recorded for commit.
1693
:param timezone: Optional timezone for timestamp.
1694
:param committer: Optional committer to set for commit.
1695
:param revprops: Optional dictionary of revision properties.
1696
:param revision_id: Optional revision id.
1698
if self._fallback_repositories:
1699
raise errors.BzrError("Cannot commit from a lightweight checkout "
1700
"to a stacked branch. See "
1701
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1702
result = self._commit_builder_class(self, parents, config,
1703
timestamp, timezone, committer, revprops, revision_id)
1704
self.start_write_group()
1708
if (self.control_files._lock_count == 1 and
1709
self.control_files._lock_mode == 'w'):
1710
if self._write_group is not None:
1711
self.abort_write_group()
1712
self.control_files.unlock()
1713
raise errors.BzrError(
1714
'Must end write groups before releasing write locks.')
1715
self.control_files.unlock()
1716
if self.control_files._lock_count == 0:
1717
self._inventory_entry_cache.clear()
1718
for repo in self._fallback_repositories:
1722
def clone(self, a_bzrdir, revision_id=None):
1723
"""Clone this repository into a_bzrdir using the current format.
1725
Currently no check is made that the format of this repository and
1726
the bzrdir format are compatible. FIXME RBC 20060201.
1728
:return: The newly created destination repository.
1730
# TODO: deprecate after 0.16; cloning this with all its settings is
1731
# probably not very useful -- mbp 20070423
1732
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1733
self.copy_content_into(dest_repo, revision_id)
1736
def start_write_group(self):
1737
"""Start a write group in the repository.
1739
Write groups are used by repositories which do not have a 1:1 mapping
1740
between file ids and backend store to manage the insertion of data from
1741
both fetch and commit operations.
1743
A write lock is required around the start_write_group/commit_write_group
1744
for the support of lock-requiring repository formats.
1746
One can only insert data into a repository inside a write group.
1750
if not self.is_write_locked():
1751
raise errors.NotWriteLocked(self)
1752
if self._write_group:
1753
raise errors.BzrError('already in a write group')
1754
self._start_write_group()
1755
# so we can detect unlock/relock - the write group is now entered.
1756
self._write_group = self.get_transaction()
1758
def _start_write_group(self):
1759
"""Template method for per-repository write group startup.
1761
This is called before the write group is considered to be
1766
def sprout(self, to_bzrdir, revision_id=None):
1767
"""Create a descendent repository for new development.
1769
Unlike clone, this does not copy the settings of the repository.
1771
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1772
dest_repo.fetch(self, revision_id=revision_id)
1775
def _create_sprouting_repo(self, a_bzrdir, shared):
1776
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1777
# use target default format.
1778
dest_repo = a_bzrdir.create_repository()
1780
# Most control formats need the repository to be specifically
1781
# created, but on some old all-in-one formats it's not needed
1783
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1784
except errors.UninitializableFormat:
1785
dest_repo = a_bzrdir.open_repository()
1788
def _get_sink(self):
1789
"""Return a sink for streaming into this repository."""
1790
return StreamSink(self)
1792
def _get_source(self, to_format):
1793
"""Return a source for streaming from this repository."""
1794
return StreamSource(self, to_format)
1797
def has_revision(self, revision_id):
1798
"""True if this repository has a copy of the revision."""
1799
return revision_id in self.has_revisions((revision_id,))
1802
def has_revisions(self, revision_ids):
1803
"""Probe to find out the presence of multiple revisions.
1805
:param revision_ids: An iterable of revision_ids.
1806
:return: A set of the revision_ids that were present.
1808
parent_map = self.revisions.get_parent_map(
1809
[(rev_id,) for rev_id in revision_ids])
1811
if _mod_revision.NULL_REVISION in revision_ids:
1812
result.add(_mod_revision.NULL_REVISION)
1813
result.update([key[0] for key in parent_map])
1817
def get_revision(self, revision_id):
1818
"""Return the Revision object for a named revision."""
1819
return self.get_revisions([revision_id])[0]
1822
def get_revision_reconcile(self, revision_id):
1823
"""'reconcile' helper routine that allows access to a revision always.
1825
This variant of get_revision does not cross check the weave graph
1826
against the revision one as get_revision does: but it should only
1827
be used by reconcile, or reconcile-alike commands that are correcting
1828
or testing the revision graph.
1830
return self._get_revisions([revision_id])[0]
1833
def get_revisions(self, revision_ids):
1834
"""Get many revisions at once.
1836
Repositories that need to check data on every revision read should
1837
subclass this method.
1839
return self._get_revisions(revision_ids)
1842
def _get_revisions(self, revision_ids):
1843
"""Core work logic to get many revisions without sanity checks."""
1845
for revid, rev in self._iter_revisions(revision_ids):
1847
raise errors.NoSuchRevision(self, revid)
1849
return [revs[revid] for revid in revision_ids]
1851
def _iter_revisions(self, revision_ids):
1852
"""Iterate over revision objects.
1854
:param revision_ids: An iterable of revisions to examine. None may be
1855
passed to request all revisions known to the repository. Note that
1856
not all repositories can find unreferenced revisions; for those
1857
repositories only referenced ones will be returned.
1858
:return: An iterator of (revid, revision) tuples. Absent revisions (
1859
those asked for but not available) are returned as (revid, None).
1861
if revision_ids is None:
1862
revision_ids = self.all_revision_ids()
1864
for rev_id in revision_ids:
1865
if not rev_id or not isinstance(rev_id, basestring):
1866
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1867
keys = [(key,) for key in revision_ids]
1868
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1869
for record in stream:
1870
revid = record.key[0]
1871
if record.storage_kind == 'absent':
1874
text = record.get_bytes_as('fulltext')
1875
rev = self._serializer.read_revision_from_string(text)
1879
def get_revision_xml(self, revision_id):
1880
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1881
# would have already do it.
1882
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1883
# TODO: this can't just be replaced by:
1884
# return self._serializer.write_revision_to_string(
1885
# self.get_revision(revision_id))
1886
# as cStringIO preservers the encoding unlike write_revision_to_string
1887
# or some other call down the path.
1888
rev = self.get_revision(revision_id)
1889
rev_tmp = cStringIO.StringIO()
1890
# the current serializer..
1891
self._serializer.write_revision(rev, rev_tmp)
1893
return rev_tmp.getvalue()
1895
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1896
"""Produce a generator of revision deltas.
1898
Note that the input is a sequence of REVISIONS, not revision_ids.
1899
Trees will be held in memory until the generator exits.
1900
Each delta is relative to the revision's lefthand predecessor.
1902
:param specific_fileids: if not None, the result is filtered
1903
so that only those file-ids, their parents and their
1904
children are included.
1906
# Get the revision-ids of interest
1907
required_trees = set()
1908
for revision in revisions:
1909
required_trees.add(revision.revision_id)
1910
required_trees.update(revision.parent_ids[:1])
1912
# Get the matching filtered trees. Note that it's more
1913
# efficient to pass filtered trees to changes_from() rather
1914
# than doing the filtering afterwards. changes_from() could
1915
# arguably do the filtering itself but it's path-based, not
1916
# file-id based, so filtering before or afterwards is
1918
if specific_fileids is None:
1919
trees = dict((t.get_revision_id(), t) for
1920
t in self.revision_trees(required_trees))
1922
trees = dict((t.get_revision_id(), t) for
1923
t in self._filtered_revision_trees(required_trees,
1926
# Calculate the deltas
1927
for revision in revisions:
1928
if not revision.parent_ids:
1929
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1931
old_tree = trees[revision.parent_ids[0]]
1932
yield trees[revision.revision_id].changes_from(old_tree)
1935
def get_revision_delta(self, revision_id, specific_fileids=None):
1936
"""Return the delta for one revision.
1938
The delta is relative to the left-hand predecessor of the
1941
:param specific_fileids: if not None, the result is filtered
1942
so that only those file-ids, their parents and their
1943
children are included.
1945
r = self.get_revision(revision_id)
1946
return list(self.get_deltas_for_revisions([r],
1947
specific_fileids=specific_fileids))[0]
1950
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1951
signature = gpg_strategy.sign(plaintext)
1952
self.add_signature_text(revision_id, signature)
1955
def add_signature_text(self, revision_id, signature):
1956
self.signatures.add_lines((revision_id,), (),
1957
osutils.split_lines(signature))
1959
def find_text_key_references(self):
1960
"""Find the text key references within the repository.
1962
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1963
to whether they were referred to by the inventory of the
1964
revision_id that they contain. The inventory texts from all present
1965
revision ids are assessed to generate this report.
1967
revision_keys = self.revisions.keys()
1968
w = self.inventories
1969
pb = ui.ui_factory.nested_progress_bar()
1971
return self._find_text_key_references_from_xml_inventory_lines(
1972
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1976
def _find_text_key_references_from_xml_inventory_lines(self,
1978
"""Core routine for extracting references to texts from inventories.
1980
This performs the translation of xml lines to revision ids.
1982
:param line_iterator: An iterator of lines, origin_version_id
1983
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1984
to whether they were referred to by the inventory of the
1985
revision_id that they contain. Note that if that revision_id was
1986
not part of the line_iterator's output then False will be given -
1987
even though it may actually refer to that key.
1989
if not self._serializer.support_altered_by_hack:
1990
raise AssertionError(
1991
"_find_text_key_references_from_xml_inventory_lines only "
1992
"supported for branches which store inventory as unnested xml"
1993
", not on %r" % self)
1996
# this code needs to read every new line in every inventory for the
1997
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
1998
# not present in one of those inventories is unnecessary but not
1999
# harmful because we are filtering by the revision id marker in the
2000
# inventory lines : we only select file ids altered in one of those
2001
# revisions. We don't need to see all lines in the inventory because
2002
# only those added in an inventory in rev X can contain a revision=X
2004
unescape_revid_cache = {}
2005
unescape_fileid_cache = {}
2007
# jam 20061218 In a big fetch, this handles hundreds of thousands
2008
# of lines, so it has had a lot of inlining and optimizing done.
2009
# Sorry that it is a little bit messy.
2010
# Move several functions to be local variables, since this is a long
2012
search = self._file_ids_altered_regex.search
2013
unescape = _unescape_xml
2014
setdefault = result.setdefault
2015
for line, line_key in line_iterator:
2016
match = search(line)
2019
# One call to match.group() returning multiple items is quite a
2020
# bit faster than 2 calls to match.group() each returning 1
2021
file_id, revision_id = match.group('file_id', 'revision_id')
2023
# Inlining the cache lookups helps a lot when you make 170,000
2024
# lines and 350k ids, versus 8.4 unique ids.
2025
# Using a cache helps in 2 ways:
2026
# 1) Avoids unnecessary decoding calls
2027
# 2) Re-uses cached strings, which helps in future set and
2029
# (2) is enough that removing encoding entirely along with
2030
# the cache (so we are using plain strings) results in no
2031
# performance improvement.
2033
revision_id = unescape_revid_cache[revision_id]
2035
unescaped = unescape(revision_id)
2036
unescape_revid_cache[revision_id] = unescaped
2037
revision_id = unescaped
2039
# Note that unconditionally unescaping means that we deserialise
2040
# every fileid, which for general 'pull' is not great, but we don't
2041
# really want to have some many fulltexts that this matters anyway.
2044
file_id = unescape_fileid_cache[file_id]
2046
unescaped = unescape(file_id)
2047
unescape_fileid_cache[file_id] = unescaped
2050
key = (file_id, revision_id)
2051
setdefault(key, False)
2052
if revision_id == line_key[-1]:
2056
def _inventory_xml_lines_for_keys(self, keys):
2057
"""Get a line iterator of the sort needed for findind references.
2059
Not relevant for non-xml inventory repositories.
2061
Ghosts in revision_keys are ignored.
2063
:param revision_keys: The revision keys for the inventories to inspect.
2064
:return: An iterator over (inventory line, revid) for the fulltexts of
2065
all of the xml inventories specified by revision_keys.
2067
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2068
for record in stream:
2069
if record.storage_kind != 'absent':
2070
chunks = record.get_bytes_as('chunked')
2071
revid = record.key[-1]
2072
lines = osutils.chunks_to_lines(chunks)
2076
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2078
"""Helper routine for fileids_altered_by_revision_ids.
2080
This performs the translation of xml lines to revision ids.
2082
:param line_iterator: An iterator of lines, origin_version_id
2083
:param revision_keys: The revision ids to filter for. This should be a
2084
set or other type which supports efficient __contains__ lookups, as
2085
the revision key from each parsed line will be looked up in the
2086
revision_keys filter.
2087
:return: a dictionary mapping altered file-ids to an iterable of
2088
revision_ids. Each altered file-ids has the exact revision_ids that
2089
altered it listed explicitly.
2091
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2092
line_iterator).iterkeys())
2093
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2094
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2095
self._inventory_xml_lines_for_keys(parent_keys)))
2096
new_keys = seen - parent_seen
2098
setdefault = result.setdefault
2099
for key in new_keys:
2100
setdefault(key[0], set()).add(key[-1])
2103
def _find_parent_ids_of_revisions(self, revision_ids):
2104
"""Find all parent ids that are mentioned in the revision graph.
2106
:return: set of revisions that are parents of revision_ids which are
2107
not part of revision_ids themselves
2109
parent_map = self.get_parent_map(revision_ids)
2111
map(parent_ids.update, parent_map.itervalues())
2112
parent_ids.difference_update(revision_ids)
2113
parent_ids.discard(_mod_revision.NULL_REVISION)
2116
def _find_parent_keys_of_revisions(self, revision_keys):
2117
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2119
:param revision_keys: An iterable of revision_keys.
2120
:return: The parents of all revision_keys that are not already in
2123
parent_map = self.revisions.get_parent_map(revision_keys)
2125
map(parent_keys.update, parent_map.itervalues())
2126
parent_keys.difference_update(revision_keys)
2127
parent_keys.discard(_mod_revision.NULL_REVISION)
2130
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2131
"""Find the file ids and versions affected by revisions.
2133
:param revisions: an iterable containing revision ids.
2134
:param _inv_weave: The inventory weave from this repository or None.
2135
If None, the inventory weave will be opened automatically.
2136
:return: a dictionary mapping altered file-ids to an iterable of
2137
revision_ids. Each altered file-ids has the exact revision_ids that
2138
altered it listed explicitly.
2140
selected_keys = set((revid,) for revid in revision_ids)
2141
w = _inv_weave or self.inventories
2142
pb = ui.ui_factory.nested_progress_bar()
2144
return self._find_file_ids_from_xml_inventory_lines(
2145
w.iter_lines_added_or_present_in_keys(
2146
selected_keys, pb=pb),
2151
def iter_files_bytes(self, desired_files):
2152
"""Iterate through file versions.
2154
Files will not necessarily be returned in the order they occur in
2155
desired_files. No specific order is guaranteed.
2157
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2158
value supplied by the caller as part of desired_files. It should
2159
uniquely identify the file version in the caller's context. (Examples:
2160
an index number or a TreeTransform trans_id.)
2162
bytes_iterator is an iterable of bytestrings for the file. The
2163
kind of iterable and length of the bytestrings are unspecified, but for
2164
this implementation, it is a list of bytes produced by
2165
VersionedFile.get_record_stream().
2167
:param desired_files: a list of (file_id, revision_id, identifier)
2171
for file_id, revision_id, callable_data in desired_files:
2172
text_keys[(file_id, revision_id)] = callable_data
2173
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2174
if record.storage_kind == 'absent':
2175
raise errors.RevisionNotPresent(record.key, self)
2176
yield text_keys[record.key], record.get_bytes_as('chunked')
2178
def _generate_text_key_index(self, text_key_references=None,
2180
"""Generate a new text key index for the repository.
2182
This is an expensive function that will take considerable time to run.
2184
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2185
list of parents, also text keys. When a given key has no parents,
2186
the parents list will be [NULL_REVISION].
2188
# All revisions, to find inventory parents.
2189
if ancestors is None:
2190
graph = self.get_graph()
2191
ancestors = graph.get_parent_map(self.all_revision_ids())
2192
if text_key_references is None:
2193
text_key_references = self.find_text_key_references()
2194
pb = ui.ui_factory.nested_progress_bar()
2196
return self._do_generate_text_key_index(ancestors,
2197
text_key_references, pb)
2201
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2202
"""Helper for _generate_text_key_index to avoid deep nesting."""
2203
revision_order = tsort.topo_sort(ancestors)
2204
invalid_keys = set()
2206
for revision_id in revision_order:
2207
revision_keys[revision_id] = set()
2208
text_count = len(text_key_references)
2209
# a cache of the text keys to allow reuse; costs a dict of all the
2210
# keys, but saves a 2-tuple for every child of a given key.
2212
for text_key, valid in text_key_references.iteritems():
2214
invalid_keys.add(text_key)
2216
revision_keys[text_key[1]].add(text_key)
2217
text_key_cache[text_key] = text_key
2218
del text_key_references
2220
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2221
NULL_REVISION = _mod_revision.NULL_REVISION
2222
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2223
# too small for large or very branchy trees. However, for 55K path
2224
# trees, it would be easy to use too much memory trivially. Ideally we
2225
# could gauge this by looking at available real memory etc, but this is
2226
# always a tricky proposition.
2227
inventory_cache = lru_cache.LRUCache(10)
2228
batch_size = 10 # should be ~150MB on a 55K path tree
2229
batch_count = len(revision_order) / batch_size + 1
2231
pb.update("Calculating text parents", processed_texts, text_count)
2232
for offset in xrange(batch_count):
2233
to_query = revision_order[offset * batch_size:(offset + 1) *
2237
for revision_id in to_query:
2238
parent_ids = ancestors[revision_id]
2239
for text_key in revision_keys[revision_id]:
2240
pb.update("Calculating text parents", processed_texts)
2241
processed_texts += 1
2242
candidate_parents = []
2243
for parent_id in parent_ids:
2244
parent_text_key = (text_key[0], parent_id)
2246
check_parent = parent_text_key not in \
2247
revision_keys[parent_id]
2249
# the parent parent_id is a ghost:
2250
check_parent = False
2251
# truncate the derived graph against this ghost.
2252
parent_text_key = None
2254
# look at the parent commit details inventories to
2255
# determine possible candidates in the per file graph.
2258
inv = inventory_cache[parent_id]
2260
inv = self.revision_tree(parent_id).inventory
2261
inventory_cache[parent_id] = inv
2263
parent_entry = inv[text_key[0]]
2264
except (KeyError, errors.NoSuchId):
2266
if parent_entry is not None:
2268
text_key[0], parent_entry.revision)
2270
parent_text_key = None
2271
if parent_text_key is not None:
2272
candidate_parents.append(
2273
text_key_cache[parent_text_key])
2274
parent_heads = text_graph.heads(candidate_parents)
2275
new_parents = list(parent_heads)
2276
new_parents.sort(key=lambda x:candidate_parents.index(x))
2277
if new_parents == []:
2278
new_parents = [NULL_REVISION]
2279
text_index[text_key] = new_parents
2281
for text_key in invalid_keys:
2282
text_index[text_key] = [NULL_REVISION]
2285
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2286
"""Get an iterable listing the keys of all the data introduced by a set
2289
The keys will be ordered so that the corresponding items can be safely
2290
fetched and inserted in that order.
2292
:returns: An iterable producing tuples of (knit-kind, file-id,
2293
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2294
'revisions'. file-id is None unless knit-kind is 'file'.
2296
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2299
for result in self._find_non_file_keys_to_fetch(revision_ids):
2302
def _find_file_keys_to_fetch(self, revision_ids, pb):
2303
# XXX: it's a bit weird to control the inventory weave caching in this
2304
# generator. Ideally the caching would be done in fetch.py I think. Or
2305
# maybe this generator should explicitly have the contract that it
2306
# should not be iterated until the previously yielded item has been
2308
inv_w = self.inventories
2310
# file ids that changed
2311
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2313
num_file_ids = len(file_ids)
2314
for file_id, altered_versions in file_ids.iteritems():
2316
pb.update("fetch texts", count, num_file_ids)
2318
yield ("file", file_id, altered_versions)
2320
def _find_non_file_keys_to_fetch(self, revision_ids):
2322
yield ("inventory", None, revision_ids)
2325
# XXX: Note ATM no callers actually pay attention to this return
2326
# instead they just use the list of revision ids and ignore
2327
# missing sigs. Consider removing this work entirely
2328
revisions_with_signatures = set(self.signatures.get_parent_map(
2329
[(r,) for r in revision_ids]))
2330
revisions_with_signatures = set(
2331
[r for (r,) in revisions_with_signatures])
2332
revisions_with_signatures.intersection_update(revision_ids)
2333
yield ("signatures", None, revisions_with_signatures)
2336
yield ("revisions", None, revision_ids)
2339
def get_inventory(self, revision_id):
2340
"""Get Inventory object by revision id."""
2341
return self.iter_inventories([revision_id]).next()
2343
def iter_inventories(self, revision_ids):
2344
"""Get many inventories by revision_ids.
2346
This will buffer some or all of the texts used in constructing the
2347
inventories in memory, but will only parse a single inventory at a
2350
:param revision_ids: The expected revision ids of the inventories.
2351
:return: An iterator of inventories.
2353
if ((None in revision_ids)
2354
or (_mod_revision.NULL_REVISION in revision_ids)):
2355
raise ValueError('cannot get null revision inventory')
2356
return self._iter_inventories(revision_ids)
2358
def _iter_inventories(self, revision_ids):
2359
"""single-document based inventory iteration."""
2360
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2361
yield self.deserialise_inventory(revision_id, text)
2363
def _iter_inventory_xmls(self, revision_ids):
2364
keys = [(revision_id,) for revision_id in revision_ids]
2365
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2367
for record in stream:
2368
if record.storage_kind != 'absent':
2369
text_chunks[record.key] = record.get_bytes_as('chunked')
2371
raise errors.NoSuchRevision(self, record.key)
2373
chunks = text_chunks.pop(key)
2374
yield ''.join(chunks), key[-1]
2376
def deserialise_inventory(self, revision_id, xml):
2377
"""Transform the xml into an inventory object.
2379
:param revision_id: The expected revision id of the inventory.
2380
:param xml: A serialised inventory.
2382
result = self._serializer.read_inventory_from_string(xml, revision_id,
2383
entry_cache=self._inventory_entry_cache)
2384
if result.revision_id != revision_id:
2385
raise AssertionError('revision id mismatch %s != %s' % (
2386
result.revision_id, revision_id))
2389
def serialise_inventory(self, inv):
2390
return self._serializer.write_inventory_to_string(inv)
2392
def _serialise_inventory_to_lines(self, inv):
2393
return self._serializer.write_inventory_to_lines(inv)
2395
def get_serializer_format(self):
2396
return self._serializer.format_num
2399
def get_inventory_xml(self, revision_id):
2400
"""Get inventory XML as a file object."""
2401
texts = self._iter_inventory_xmls([revision_id])
2403
text, revision_id = texts.next()
2404
except StopIteration:
2405
raise errors.HistoryMissing(self, 'inventory', revision_id)
2409
def get_inventory_sha1(self, revision_id):
2410
"""Return the sha1 hash of the inventory entry
2412
return self.get_revision(revision_id).inventory_sha1
2414
def get_rev_id_for_revno(self, revno, known_pair):
2415
"""Return the revision id of a revno, given a later (revno, revid)
2416
pair in the same history.
2418
:return: if found (True, revid). If the available history ran out
2419
before reaching the revno, then this returns
2420
(False, (closest_revno, closest_revid)).
2422
known_revno, known_revid = known_pair
2423
partial_history = [known_revid]
2424
distance_from_known = known_revno - revno
2425
if distance_from_known < 0:
2427
'requested revno (%d) is later than given known revno (%d)'
2428
% (revno, known_revno))
2431
self, partial_history, stop_index=distance_from_known)
2432
except errors.RevisionNotPresent, err:
2433
if err.revision_id == known_revid:
2434
# The start revision (known_revid) wasn't found.
2436
# This is a stacked repository with no fallbacks, or a there's a
2437
# left-hand ghost. Either way, even though the revision named in
2438
# the error isn't in this repo, we know it's the next step in this
2439
# left-hand history.
2440
partial_history.append(err.revision_id)
2441
if len(partial_history) <= distance_from_known:
2442
# Didn't find enough history to get a revid for the revno.
2443
earliest_revno = known_revno - len(partial_history) + 1
2444
return (False, (earliest_revno, partial_history[-1]))
2445
if len(partial_history) - 1 > distance_from_known:
2446
raise AssertionError('_iter_for_revno returned too much history')
2447
return (True, partial_history[-1])
2449
def iter_reverse_revision_history(self, revision_id):
2450
"""Iterate backwards through revision ids in the lefthand history
2452
:param revision_id: The revision id to start with. All its lefthand
2453
ancestors will be traversed.
2455
graph = self.get_graph()
2456
next_id = revision_id
2458
if next_id in (None, _mod_revision.NULL_REVISION):
2461
parents = graph.get_parent_map([next_id])[next_id]
2463
raise errors.RevisionNotPresent(next_id, self)
2465
if len(parents) == 0:
2468
next_id = parents[0]
2471
def get_revision_inventory(self, revision_id):
2472
"""Return inventory of a past revision."""
2473
# TODO: Unify this with get_inventory()
2474
# bzr 0.0.6 and later imposes the constraint that the inventory_id
2475
# must be the same as its revision, so this is trivial.
2476
if revision_id is None:
2477
# This does not make sense: if there is no revision,
2478
# then it is the current tree inventory surely ?!
2479
# and thus get_root_id() is something that looks at the last
2480
# commit on the branch, and the get_root_id is an inventory check.
2481
raise NotImplementedError
2482
# return Inventory(self.get_root_id())
2484
return self.get_inventory(revision_id)
2486
def is_shared(self):
2487
"""Return True if this repository is flagged as a shared repository."""
2488
raise NotImplementedError(self.is_shared)
2491
def reconcile(self, other=None, thorough=False):
2492
"""Reconcile this repository."""
2493
from bzrlib.reconcile import RepoReconciler
2494
reconciler = RepoReconciler(self, thorough=thorough)
2495
reconciler.reconcile()
2498
def _refresh_data(self):
2499
"""Helper called from lock_* to ensure coherency with disk.
2501
The default implementation does nothing; it is however possible
2502
for repositories to maintain loaded indices across multiple locks
2503
by checking inside their implementation of this method to see
2504
whether their indices are still valid. This depends of course on
2505
the disk format being validatable in this manner. This method is
2506
also called by the refresh_data() public interface to cause a refresh
2507
to occur while in a write lock so that data inserted by a smart server
2508
push operation is visible on the client's instance of the physical
2513
def revision_tree(self, revision_id):
2514
"""Return Tree for a revision on this branch.
2516
`revision_id` may be NULL_REVISION for the empty tree revision.
2518
revision_id = _mod_revision.ensure_null(revision_id)
2519
# TODO: refactor this to use an existing revision object
2520
# so we don't need to read it in twice.
2521
if revision_id == _mod_revision.NULL_REVISION:
2522
return RevisionTree(self, Inventory(root_id=None),
2523
_mod_revision.NULL_REVISION)
2525
inv = self.get_revision_inventory(revision_id)
2526
return RevisionTree(self, inv, revision_id)
2528
def revision_trees(self, revision_ids):
2529
"""Return Trees for revisions in this repository.
2531
:param revision_ids: a sequence of revision-ids;
2532
a revision-id may not be None or 'null:'
2534
inventories = self.iter_inventories(revision_ids)
2535
for inv in inventories:
2536
yield RevisionTree(self, inv, inv.revision_id)
2538
def _filtered_revision_trees(self, revision_ids, file_ids):
2539
"""Return Tree for a revision on this branch with only some files.
2541
:param revision_ids: a sequence of revision-ids;
2542
a revision-id may not be None or 'null:'
2543
:param file_ids: if not None, the result is filtered
2544
so that only those file-ids, their parents and their
2545
children are included.
2547
inventories = self.iter_inventories(revision_ids)
2548
for inv in inventories:
2549
# Should we introduce a FilteredRevisionTree class rather
2550
# than pre-filter the inventory here?
2551
filtered_inv = inv.filter(file_ids)
2552
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2555
def get_ancestry(self, revision_id, topo_sorted=True):
2556
"""Return a list of revision-ids integrated by a revision.
2558
The first element of the list is always None, indicating the origin
2559
revision. This might change when we have history horizons, or
2560
perhaps we should have a new API.
2562
This is topologically sorted.
2564
if _mod_revision.is_null(revision_id):
2566
if not self.has_revision(revision_id):
2567
raise errors.NoSuchRevision(self, revision_id)
2568
graph = self.get_graph()
2570
search = graph._make_breadth_first_searcher([revision_id])
2573
found, ghosts = search.next_with_ghosts()
2574
except StopIteration:
2577
if _mod_revision.NULL_REVISION in keys:
2578
keys.remove(_mod_revision.NULL_REVISION)
2580
parent_map = graph.get_parent_map(keys)
2581
keys = tsort.topo_sort(parent_map)
2582
return [None] + list(keys)
2584
def pack(self, hint=None):
2585
"""Compress the data within the repository.
2587
This operation only makes sense for some repository types. For other
2588
types it should be a no-op that just returns.
2590
This stub method does not require a lock, but subclasses should use
2591
@needs_write_lock as this is a long running call its reasonable to
2592
implicitly lock for the user.
2594
:param hint: If not supplied, the whole repository is packed.
2595
If supplied, the repository may use the hint parameter as a
2596
hint for the parts of the repository to pack. A hint can be
2597
obtained from the result of commit_write_group(). Out of
2598
date hints are simply ignored, because concurrent operations
2599
can obsolete them rapidly.
2602
def get_transaction(self):
2603
return self.control_files.get_transaction()
2605
def get_parent_map(self, revision_ids):
2606
"""See graph.StackedParentsProvider.get_parent_map"""
2607
# revisions index works in keys; this just works in revisions
2608
# therefore wrap and unwrap
2611
for revision_id in revision_ids:
2612
if revision_id == _mod_revision.NULL_REVISION:
2613
result[revision_id] = ()
2614
elif revision_id is None:
2615
raise ValueError('get_parent_map(None) is not valid')
2617
query_keys.append((revision_id ,))
2618
for ((revision_id,), parent_keys) in \
2619
self.revisions.get_parent_map(query_keys).iteritems():
2621
result[revision_id] = tuple(parent_revid
2622
for (parent_revid,) in parent_keys)
2624
result[revision_id] = (_mod_revision.NULL_REVISION,)
2627
def _make_parents_provider(self):
2630
def get_graph(self, other_repository=None):
2631
"""Return the graph walker for this repository format"""
2632
parents_provider = self._make_parents_provider()
2633
if (other_repository is not None and
2634
not self.has_same_location(other_repository)):
2635
parents_provider = graph.StackedParentsProvider(
2636
[parents_provider, other_repository._make_parents_provider()])
2637
return graph.Graph(parents_provider)
2639
def _get_versioned_file_checker(self, text_key_references=None,
2641
"""Return an object suitable for checking versioned files.
2643
:param text_key_references: if non-None, an already built
2644
dictionary mapping text keys ((fileid, revision_id) tuples)
2645
to whether they were referred to by the inventory of the
2646
revision_id that they contain. If None, this will be
2648
:param ancestors: Optional result from
2649
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2652
return _VersionedFileChecker(self,
2653
text_key_references=text_key_references, ancestors=ancestors)
2655
def revision_ids_to_search_result(self, result_set):
2656
"""Convert a set of revision ids to a graph SearchResult."""
2657
result_parents = set()
2658
for parents in self.get_graph().get_parent_map(
2659
result_set).itervalues():
2660
result_parents.update(parents)
2661
included_keys = result_set.intersection(result_parents)
2662
start_keys = result_set.difference(included_keys)
2663
exclude_keys = result_parents.difference(result_set)
2664
result = graph.SearchResult(start_keys, exclude_keys,
2665
len(result_set), result_set)
2669
def set_make_working_trees(self, new_value):
2670
"""Set the policy flag for making working trees when creating branches.
2672
This only applies to branches that use this repository.
2674
The default is 'True'.
2675
:param new_value: True to restore the default, False to disable making
2678
raise NotImplementedError(self.set_make_working_trees)
2680
def make_working_trees(self):
2681
"""Returns the policy for making working trees on new branches."""
2682
raise NotImplementedError(self.make_working_trees)
2685
def sign_revision(self, revision_id, gpg_strategy):
2686
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2687
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2690
def has_signature_for_revision_id(self, revision_id):
2691
"""Query for a revision signature for revision_id in the repository."""
2692
if not self.has_revision(revision_id):
2693
raise errors.NoSuchRevision(self, revision_id)
2694
sig_present = (1 == len(
2695
self.signatures.get_parent_map([(revision_id,)])))
2699
def get_signature_text(self, revision_id):
2700
"""Return the text for a signature."""
2701
stream = self.signatures.get_record_stream([(revision_id,)],
2703
record = stream.next()
2704
if record.storage_kind == 'absent':
2705
raise errors.NoSuchRevision(self, revision_id)
2706
return record.get_bytes_as('fulltext')
2709
def check(self, revision_ids=None, callback_refs=None, check_repo=True):
2710
"""Check consistency of all history of given revision_ids.
2712
Different repository implementations should override _check().
2714
:param revision_ids: A non-empty list of revision_ids whose ancestry
2715
will be checked. Typically the last revision_id of a branch.
2716
:param callback_refs: A dict of check-refs to resolve and callback
2717
the check/_check method on the items listed as wanting the ref.
2719
:param check_repo: If False do not check the repository contents, just
2720
calculate the data callback_refs requires and call them back.
2722
return self._check(revision_ids, callback_refs=callback_refs,
2723
check_repo=check_repo)
2725
def _check(self, revision_ids, callback_refs, check_repo):
2726
result = check.Check(self, check_repo=check_repo)
2727
result.check(callback_refs)
2730
def _warn_if_deprecated(self):
2731
global _deprecation_warning_done
2732
if _deprecation_warning_done:
2734
_deprecation_warning_done = True
2735
warning("Format %s for %s is deprecated - please use 'bzr upgrade' to get better performance"
2736
% (self._format, self.bzrdir.transport.base))
2738
def supports_rich_root(self):
2739
return self._format.rich_root_data
2741
def _check_ascii_revisionid(self, revision_id, method):
2742
"""Private helper for ascii-only repositories."""
2743
# weave repositories refuse to store revisionids that are non-ascii.
2744
if revision_id is not None:
2745
# weaves require ascii revision ids.
2746
if isinstance(revision_id, unicode):
2748
revision_id.encode('ascii')
2749
except UnicodeEncodeError:
2750
raise errors.NonAsciiRevisionId(method, self)
2753
revision_id.decode('ascii')
2754
except UnicodeDecodeError:
2755
raise errors.NonAsciiRevisionId(method, self)
2757
def revision_graph_can_have_wrong_parents(self):
2758
"""Is it possible for this repository to have a revision graph with
2761
If True, then this repository must also implement
2762
_find_inconsistent_revision_parents so that check and reconcile can
2763
check for inconsistencies before proceeding with other checks that may
2764
depend on the revision index being consistent.
2766
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2769
# remove these delegates a while after bzr 0.15
2770
def __make_delegated(name, from_module):
2771
def _deprecated_repository_forwarder():
2772
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2773
% (name, from_module),
2776
m = __import__(from_module, globals(), locals(), [name])
2778
return getattr(m, name)
2779
except AttributeError:
2780
raise AttributeError('module %s has no name %s'
2782
globals()[name] = _deprecated_repository_forwarder
2785
'AllInOneRepository',
2786
'WeaveMetaDirRepository',
2787
'PreSplitOutRepositoryFormat',
2788
'RepositoryFormat4',
2789
'RepositoryFormat5',
2790
'RepositoryFormat6',
2791
'RepositoryFormat7',
2793
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2797
'RepositoryFormatKnit',
2798
'RepositoryFormatKnit1',
2800
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2803
def install_revision(repository, rev, revision_tree):
2804
"""Install all revision data into a repository."""
2805
install_revisions(repository, [(rev, revision_tree, None)])
2808
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2809
"""Install all revision data into a repository.
2811
Accepts an iterable of revision, tree, signature tuples. The signature
2814
repository.start_write_group()
2816
inventory_cache = lru_cache.LRUCache(10)
2817
for n, (revision, revision_tree, signature) in enumerate(iterable):
2818
_install_revision(repository, revision, revision_tree, signature,
2821
pb.update('Transferring revisions', n + 1, num_revisions)
2823
repository.abort_write_group()
2826
repository.commit_write_group()
2829
def _install_revision(repository, rev, revision_tree, signature,
2831
"""Install all revision data into a repository."""
2832
present_parents = []
2834
for p_id in rev.parent_ids:
2835
if repository.has_revision(p_id):
2836
present_parents.append(p_id)
2837
parent_trees[p_id] = repository.revision_tree(p_id)
2839
parent_trees[p_id] = repository.revision_tree(
2840
_mod_revision.NULL_REVISION)
2842
inv = revision_tree.inventory
2843
entries = inv.iter_entries()
2844
# backwards compatibility hack: skip the root id.
2845
if not repository.supports_rich_root():
2846
path, root = entries.next()
2847
if root.revision != rev.revision_id:
2848
raise errors.IncompatibleRevision(repr(repository))
2850
for path, ie in entries:
2851
text_keys[(ie.file_id, ie.revision)] = ie
2852
text_parent_map = repository.texts.get_parent_map(text_keys)
2853
missing_texts = set(text_keys) - set(text_parent_map)
2854
# Add the texts that are not already present
2855
for text_key in missing_texts:
2856
ie = text_keys[text_key]
2858
# FIXME: TODO: The following loop overlaps/duplicates that done by
2859
# commit to determine parents. There is a latent/real bug here where
2860
# the parents inserted are not those commit would do - in particular
2861
# they are not filtered by heads(). RBC, AB
2862
for revision, tree in parent_trees.iteritems():
2863
if ie.file_id not in tree:
2865
parent_id = tree.inventory[ie.file_id].revision
2866
if parent_id in text_parents:
2868
text_parents.append((ie.file_id, parent_id))
2869
lines = revision_tree.get_file(ie.file_id).readlines()
2870
repository.texts.add_lines(text_key, text_parents, lines)
2872
# install the inventory
2873
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2874
# Cache this inventory
2875
inventory_cache[rev.revision_id] = inv
2877
basis_inv = inventory_cache[rev.parent_ids[0]]
2879
repository.add_inventory(rev.revision_id, inv, present_parents)
2881
delta = inv._make_delta(basis_inv)
2882
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2883
rev.revision_id, present_parents)
2885
repository.add_inventory(rev.revision_id, inv, present_parents)
2886
except errors.RevisionAlreadyPresent:
2888
if signature is not None:
2889
repository.add_signature_text(rev.revision_id, signature)
2890
repository.add_revision(rev.revision_id, rev, inv)
2893
class MetaDirRepository(Repository):
2894
"""Repositories in the new meta-dir layout.
2896
:ivar _transport: Transport for access to repository control files,
2897
typically pointing to .bzr/repository.
2900
def __init__(self, _format, a_bzrdir, control_files):
2901
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2902
self._transport = control_files._transport
2904
def is_shared(self):
2905
"""Return True if this repository is flagged as a shared repository."""
2906
return self._transport.has('shared-storage')
2909
def set_make_working_trees(self, new_value):
2910
"""Set the policy flag for making working trees when creating branches.
2912
This only applies to branches that use this repository.
2914
The default is 'True'.
2915
:param new_value: True to restore the default, False to disable making
2920
self._transport.delete('no-working-trees')
2921
except errors.NoSuchFile:
2924
self._transport.put_bytes('no-working-trees', '',
2925
mode=self.bzrdir._get_file_mode())
2927
def make_working_trees(self):
2928
"""Returns the policy for making working trees on new branches."""
2929
return not self._transport.has('no-working-trees')
2932
class MetaDirVersionedFileRepository(MetaDirRepository):
2933
"""Repositories in a meta-dir, that work via versioned file objects."""
2935
def __init__(self, _format, a_bzrdir, control_files):
2936
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2940
network_format_registry = registry.FormatRegistry()
2941
"""Registry of formats indexed by their network name.
2943
The network name for a repository format is an identifier that can be used when
2944
referring to formats with smart server operations. See
2945
RepositoryFormat.network_name() for more detail.
2949
format_registry = registry.FormatRegistry(network_format_registry)
2950
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
2952
This can contain either format instances themselves, or classes/factories that
2953
can be called to obtain one.
2957
#####################################################################
2958
# Repository Formats
2960
class RepositoryFormat(object):
2961
"""A repository format.
2963
Formats provide four things:
2964
* An initialization routine to construct repository data on disk.
2965
* a optional format string which is used when the BzrDir supports
2967
* an open routine which returns a Repository instance.
2968
* A network name for referring to the format in smart server RPC
2971
There is one and only one Format subclass for each on-disk format. But
2972
there can be one Repository subclass that is used for several different
2973
formats. The _format attribute on a Repository instance can be used to
2974
determine the disk format.
2976
Formats are placed in a registry by their format string for reference
2977
during opening. These should be subclasses of RepositoryFormat for
2980
Once a format is deprecated, just deprecate the initialize and open
2981
methods on the format class. Do not deprecate the object, as the
2982
object may be created even when a repository instance hasn't been
2985
Common instance attributes:
2986
_matchingbzrdir - the bzrdir format that the repository format was
2987
originally written to work with. This can be used if manually
2988
constructing a bzrdir and repository, or more commonly for test suite
2992
# Set to True or False in derived classes. True indicates that the format
2993
# supports ghosts gracefully.
2994
supports_ghosts = None
2995
# Can this repository be given external locations to lookup additional
2996
# data. Set to True or False in derived classes.
2997
supports_external_lookups = None
2998
# Does this format support CHK bytestring lookups. Set to True or False in
3000
supports_chks = None
3001
# Should commit add an inventory, or an inventory delta to the repository.
3002
_commit_inv_deltas = True
3003
# What order should fetch operations request streams in?
3004
# The default is unordered as that is the cheapest for an origin to
3006
_fetch_order = 'unordered'
3007
# Does this repository format use deltas that can be fetched as-deltas ?
3008
# (E.g. knits, where the knit deltas can be transplanted intact.
3009
# We default to False, which will ensure that enough data to get
3010
# a full text out of any fetch stream will be grabbed.
3011
_fetch_uses_deltas = False
3012
# Should fetch trigger a reconcile after the fetch? Only needed for
3013
# some repository formats that can suffer internal inconsistencies.
3014
_fetch_reconcile = False
3015
# Does this format have < O(tree_size) delta generation. Used to hint what
3016
# code path for commit, amongst other things.
3018
# Does doing a pack operation compress data? Useful for the pack UI command
3019
# (so if there is one pack, the operation can still proceed because it may
3020
# help), and for fetching when data won't have come from the same
3022
pack_compresses = False
3025
return "<%s>" % self.__class__.__name__
3027
def __eq__(self, other):
3028
# format objects are generally stateless
3029
return isinstance(other, self.__class__)
3031
def __ne__(self, other):
3032
return not self == other
3035
def find_format(klass, a_bzrdir):
3036
"""Return the format for the repository object in a_bzrdir.
3038
This is used by bzr native formats that have a "format" file in
3039
the repository. Other methods may be used by different types of
3043
transport = a_bzrdir.get_repository_transport(None)
3044
format_string = transport.get("format").read()
3045
return format_registry.get(format_string)
3046
except errors.NoSuchFile:
3047
raise errors.NoRepositoryPresent(a_bzrdir)
3049
raise errors.UnknownFormatError(format=format_string,
3053
def register_format(klass, format):
3054
format_registry.register(format.get_format_string(), format)
3057
def unregister_format(klass, format):
3058
format_registry.remove(format.get_format_string())
3061
def get_default_format(klass):
3062
"""Return the current default format."""
3063
from bzrlib import bzrdir
3064
return bzrdir.format_registry.make_bzrdir('default').repository_format
3066
def get_format_string(self):
3067
"""Return the ASCII format string that identifies this format.
3069
Note that in pre format ?? repositories the format string is
3070
not permitted nor written to disk.
3072
raise NotImplementedError(self.get_format_string)
3074
def get_format_description(self):
3075
"""Return the short description for this format."""
3076
raise NotImplementedError(self.get_format_description)
3078
# TODO: this shouldn't be in the base class, it's specific to things that
3079
# use weaves or knits -- mbp 20070207
3080
def _get_versioned_file_store(self,
3085
versionedfile_class=None,
3086
versionedfile_kwargs={},
3088
if versionedfile_class is None:
3089
versionedfile_class = self._versionedfile_class
3090
weave_transport = control_files._transport.clone(name)
3091
dir_mode = control_files._dir_mode
3092
file_mode = control_files._file_mode
3093
return VersionedFileStore(weave_transport, prefixed=prefixed,
3095
file_mode=file_mode,
3096
versionedfile_class=versionedfile_class,
3097
versionedfile_kwargs=versionedfile_kwargs,
3100
def initialize(self, a_bzrdir, shared=False):
3101
"""Initialize a repository of this format in a_bzrdir.
3103
:param a_bzrdir: The bzrdir to put the new repository in it.
3104
:param shared: The repository should be initialized as a sharable one.
3105
:returns: The new repository object.
3107
This may raise UninitializableFormat if shared repository are not
3108
compatible the a_bzrdir.
3110
raise NotImplementedError(self.initialize)
3112
def is_supported(self):
3113
"""Is this format supported?
3115
Supported formats must be initializable and openable.
3116
Unsupported formats may not support initialization or committing or
3117
some other features depending on the reason for not being supported.
3121
def network_name(self):
3122
"""A simple byte string uniquely identifying this format for RPC calls.
3124
MetaDir repository formats use their disk format string to identify the
3125
repository over the wire. All in one formats such as bzr < 0.8, and
3126
foreign formats like svn/git and hg should use some marker which is
3127
unique and immutable.
3129
raise NotImplementedError(self.network_name)
3131
def check_conversion_target(self, target_format):
3132
raise NotImplementedError(self.check_conversion_target)
3134
def open(self, a_bzrdir, _found=False):
3135
"""Return an instance of this format for the bzrdir a_bzrdir.
3137
_found is a private parameter, do not use it.
3139
raise NotImplementedError(self.open)
3142
class MetaDirRepositoryFormat(RepositoryFormat):
3143
"""Common base class for the new repositories using the metadir layout."""
3145
rich_root_data = False
3146
supports_tree_reference = False
3147
supports_external_lookups = False
3150
def _matchingbzrdir(self):
3151
matching = bzrdir.BzrDirMetaFormat1()
3152
matching.repository_format = self
3156
super(MetaDirRepositoryFormat, self).__init__()
3158
def _create_control_files(self, a_bzrdir):
3159
"""Create the required files and the initial control_files object."""
3160
# FIXME: RBC 20060125 don't peek under the covers
3161
# NB: no need to escape relative paths that are url safe.
3162
repository_transport = a_bzrdir.get_repository_transport(self)
3163
control_files = lockable_files.LockableFiles(repository_transport,
3164
'lock', lockdir.LockDir)
3165
control_files.create_lock()
3166
return control_files
3168
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
3169
"""Upload the initial blank content."""
3170
control_files = self._create_control_files(a_bzrdir)
3171
control_files.lock_write()
3172
transport = control_files._transport
3174
utf8_files += [('shared-storage', '')]
3176
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
3177
for (filename, content_stream) in files:
3178
transport.put_file(filename, content_stream,
3179
mode=a_bzrdir._get_file_mode())
3180
for (filename, content_bytes) in utf8_files:
3181
transport.put_bytes_non_atomic(filename, content_bytes,
3182
mode=a_bzrdir._get_file_mode())
3184
control_files.unlock()
3186
def network_name(self):
3187
"""Metadir formats have matching disk and network format strings."""
3188
return self.get_format_string()
3191
# Pre-0.8 formats that don't have a disk format string (because they are
3192
# versioned by the matching control directory). We use the control directories
3193
# disk format string as a key for the network_name because they meet the
3194
# constraints (simple string, unique, immutable).
3195
network_format_registry.register_lazy(
3196
"Bazaar-NG branch, format 5\n",
3197
'bzrlib.repofmt.weaverepo',
3198
'RepositoryFormat5',
3200
network_format_registry.register_lazy(
3201
"Bazaar-NG branch, format 6\n",
3202
'bzrlib.repofmt.weaverepo',
3203
'RepositoryFormat6',
3206
# formats which have no format string are not discoverable or independently
3207
# creatable on disk, so are not registered in format_registry. They're
3208
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
3209
# needed, it's constructed directly by the BzrDir. Non-native formats where
3210
# the repository is not separately opened are similar.
3212
format_registry.register_lazy(
3213
'Bazaar-NG Repository format 7',
3214
'bzrlib.repofmt.weaverepo',
3218
format_registry.register_lazy(
3219
'Bazaar-NG Knit Repository Format 1',
3220
'bzrlib.repofmt.knitrepo',
3221
'RepositoryFormatKnit1',
3224
format_registry.register_lazy(
3225
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
3226
'bzrlib.repofmt.knitrepo',
3227
'RepositoryFormatKnit3',
3230
format_registry.register_lazy(
3231
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
3232
'bzrlib.repofmt.knitrepo',
3233
'RepositoryFormatKnit4',
3236
# Pack-based formats. There is one format for pre-subtrees, and one for
3237
# post-subtrees to allow ease of testing.
3238
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
3239
format_registry.register_lazy(
3240
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
3241
'bzrlib.repofmt.pack_repo',
3242
'RepositoryFormatKnitPack1',
3244
format_registry.register_lazy(
3245
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
3246
'bzrlib.repofmt.pack_repo',
3247
'RepositoryFormatKnitPack3',
3249
format_registry.register_lazy(
3250
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
3251
'bzrlib.repofmt.pack_repo',
3252
'RepositoryFormatKnitPack4',
3254
format_registry.register_lazy(
3255
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
3256
'bzrlib.repofmt.pack_repo',
3257
'RepositoryFormatKnitPack5',
3259
format_registry.register_lazy(
3260
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
3261
'bzrlib.repofmt.pack_repo',
3262
'RepositoryFormatKnitPack5RichRoot',
3264
format_registry.register_lazy(
3265
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3266
'bzrlib.repofmt.pack_repo',
3267
'RepositoryFormatKnitPack5RichRootBroken',
3269
format_registry.register_lazy(
3270
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3271
'bzrlib.repofmt.pack_repo',
3272
'RepositoryFormatKnitPack6',
3274
format_registry.register_lazy(
3275
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3276
'bzrlib.repofmt.pack_repo',
3277
'RepositoryFormatKnitPack6RichRoot',
3280
# Development formats.
3281
# Obsolete but kept pending a CHK based subtree format.
3282
format_registry.register_lazy(
3283
("Bazaar development format 2 with subtree support "
3284
"(needs bzr.dev from before 1.8)\n"),
3285
'bzrlib.repofmt.pack_repo',
3286
'RepositoryFormatPackDevelopment2Subtree',
3289
# 1.14->1.16 go below here
3290
format_registry.register_lazy(
3291
'Bazaar development format - group compression and chk inventory'
3292
' (needs bzr.dev from 1.14)\n',
3293
'bzrlib.repofmt.groupcompress_repo',
3294
'RepositoryFormatCHK1',
3297
format_registry.register_lazy(
3298
'Bazaar development format - chk repository with bencode revision '
3299
'serialization (needs bzr.dev from 1.16)\n',
3300
'bzrlib.repofmt.groupcompress_repo',
3301
'RepositoryFormatCHK2',
3303
format_registry.register_lazy(
3304
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3305
'bzrlib.repofmt.groupcompress_repo',
3306
'RepositoryFormat2a',
3310
class InterRepository(InterObject):
3311
"""This class represents operations taking place between two repositories.
3313
Its instances have methods like copy_content and fetch, and contain
3314
references to the source and target repositories these operations can be
3317
Often we will provide convenience methods on 'repository' which carry out
3318
operations with another repository - they will always forward to
3319
InterRepository.get(other).method_name(parameters).
3322
_walk_to_common_revisions_batch_size = 50
3324
"""The available optimised InterRepository types."""
3327
def copy_content(self, revision_id=None):
3328
"""Make a complete copy of the content in self into destination.
3330
This is a destructive operation! Do not use it on existing
3333
:param revision_id: Only copy the content needed to construct
3334
revision_id and its parents.
3337
self.target.set_make_working_trees(self.source.make_working_trees())
3338
except NotImplementedError:
3340
self.target.fetch(self.source, revision_id=revision_id)
3343
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3345
"""Fetch the content required to construct revision_id.
3347
The content is copied from self.source to self.target.
3349
:param revision_id: if None all content is copied, if NULL_REVISION no
3351
:param pb: optional progress bar to use for progress reports. If not
3352
provided a default one will be created.
3355
from bzrlib.fetch import RepoFetcher
3356
f = RepoFetcher(to_repository=self.target,
3357
from_repository=self.source,
3358
last_revision=revision_id,
3359
fetch_spec=fetch_spec,
3360
pb=pb, find_ghosts=find_ghosts)
3362
def _walk_to_common_revisions(self, revision_ids):
3363
"""Walk out from revision_ids in source to revisions target has.
3365
:param revision_ids: The start point for the search.
3366
:return: A set of revision ids.
3368
target_graph = self.target.get_graph()
3369
revision_ids = frozenset(revision_ids)
3370
missing_revs = set()
3371
source_graph = self.source.get_graph()
3372
# ensure we don't pay silly lookup costs.
3373
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3374
null_set = frozenset([_mod_revision.NULL_REVISION])
3375
searcher_exhausted = False
3379
# Iterate the searcher until we have enough next_revs
3380
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3382
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3383
next_revs.update(next_revs_part)
3384
ghosts.update(ghosts_part)
3385
except StopIteration:
3386
searcher_exhausted = True
3388
# If there are ghosts in the source graph, and the caller asked for
3389
# them, make sure that they are present in the target.
3390
# We don't care about other ghosts as we can't fetch them and
3391
# haven't been asked to.
3392
ghosts_to_check = set(revision_ids.intersection(ghosts))
3393
revs_to_get = set(next_revs).union(ghosts_to_check)
3395
have_revs = set(target_graph.get_parent_map(revs_to_get))
3396
# we always have NULL_REVISION present.
3397
have_revs = have_revs.union(null_set)
3398
# Check if the target is missing any ghosts we need.
3399
ghosts_to_check.difference_update(have_revs)
3401
# One of the caller's revision_ids is a ghost in both the
3402
# source and the target.
3403
raise errors.NoSuchRevision(
3404
self.source, ghosts_to_check.pop())
3405
missing_revs.update(next_revs - have_revs)
3406
# Because we may have walked past the original stop point, make
3407
# sure everything is stopped
3408
stop_revs = searcher.find_seen_ancestors(have_revs)
3409
searcher.stop_searching_any(stop_revs)
3410
if searcher_exhausted:
3412
return searcher.get_result()
3415
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3416
"""Return the revision ids that source has that target does not.
3418
:param revision_id: only return revision ids included by this
3420
:param find_ghosts: If True find missing revisions in deep history
3421
rather than just finding the surface difference.
3422
:return: A bzrlib.graph.SearchResult.
3424
# stop searching at found target revisions.
3425
if not find_ghosts and revision_id is not None:
3426
return self._walk_to_common_revisions([revision_id])
3427
# generic, possibly worst case, slow code path.
3428
target_ids = set(self.target.all_revision_ids())
3429
if revision_id is not None:
3430
source_ids = self.source.get_ancestry(revision_id)
3431
if source_ids[0] is not None:
3432
raise AssertionError()
3435
source_ids = self.source.all_revision_ids()
3436
result_set = set(source_ids).difference(target_ids)
3437
return self.source.revision_ids_to_search_result(result_set)
3440
def _same_model(source, target):
3441
"""True if source and target have the same data representation.
3443
Note: this is always called on the base class; overriding it in a
3444
subclass will have no effect.
3447
InterRepository._assert_same_model(source, target)
3449
except errors.IncompatibleRepositories, e:
3453
def _assert_same_model(source, target):
3454
"""Raise an exception if two repositories do not use the same model.
3456
if source.supports_rich_root() != target.supports_rich_root():
3457
raise errors.IncompatibleRepositories(source, target,
3458
"different rich-root support")
3459
if source._serializer != target._serializer:
3460
raise errors.IncompatibleRepositories(source, target,
3461
"different serializers")
3464
class InterSameDataRepository(InterRepository):
3465
"""Code for converting between repositories that represent the same data.
3467
Data format and model must match for this to work.
3471
def _get_repo_format_to_test(self):
3472
"""Repository format for testing with.
3474
InterSameData can pull from subtree to subtree and from non-subtree to
3475
non-subtree, so we test this with the richest repository format.
3477
from bzrlib.repofmt import knitrepo
3478
return knitrepo.RepositoryFormatKnit3()
3481
def is_compatible(source, target):
3482
return InterRepository._same_model(source, target)
3485
class InterWeaveRepo(InterSameDataRepository):
3486
"""Optimised code paths between Weave based repositories.
3488
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3489
implemented lazy inter-object optimisation.
3493
def _get_repo_format_to_test(self):
3494
from bzrlib.repofmt import weaverepo
3495
return weaverepo.RepositoryFormat7()
3498
def is_compatible(source, target):
3499
"""Be compatible with known Weave formats.
3501
We don't test for the stores being of specific types because that
3502
could lead to confusing results, and there is no need to be
3505
from bzrlib.repofmt.weaverepo import (
3511
return (isinstance(source._format, (RepositoryFormat5,
3513
RepositoryFormat7)) and
3514
isinstance(target._format, (RepositoryFormat5,
3516
RepositoryFormat7)))
3517
except AttributeError:
3521
def copy_content(self, revision_id=None):
3522
"""See InterRepository.copy_content()."""
3523
# weave specific optimised path:
3525
self.target.set_make_working_trees(self.source.make_working_trees())
3526
except (errors.RepositoryUpgradeRequired, NotImplemented):
3528
# FIXME do not peek!
3529
if self.source._transport.listable():
3530
pb = ui.ui_factory.nested_progress_bar()
3532
self.target.texts.insert_record_stream(
3533
self.source.texts.get_record_stream(
3534
self.source.texts.keys(), 'topological', False))
3535
pb.update('copying inventory', 0, 1)
3536
self.target.inventories.insert_record_stream(
3537
self.source.inventories.get_record_stream(
3538
self.source.inventories.keys(), 'topological', False))
3539
self.target.signatures.insert_record_stream(
3540
self.source.signatures.get_record_stream(
3541
self.source.signatures.keys(),
3543
self.target.revisions.insert_record_stream(
3544
self.source.revisions.get_record_stream(
3545
self.source.revisions.keys(),
3546
'topological', True))
3550
self.target.fetch(self.source, revision_id=revision_id)
3553
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3554
"""See InterRepository.missing_revision_ids()."""
3555
# we want all revisions to satisfy revision_id in source.
3556
# but we don't want to stat every file here and there.
3557
# we want then, all revisions other needs to satisfy revision_id
3558
# checked, but not those that we have locally.
3559
# so the first thing is to get a subset of the revisions to
3560
# satisfy revision_id in source, and then eliminate those that
3561
# we do already have.
3562
# this is slow on high latency connection to self, but as this
3563
# disk format scales terribly for push anyway due to rewriting
3564
# inventory.weave, this is considered acceptable.
3566
if revision_id is not None:
3567
source_ids = self.source.get_ancestry(revision_id)
3568
if source_ids[0] is not None:
3569
raise AssertionError()
3572
source_ids = self.source._all_possible_ids()
3573
source_ids_set = set(source_ids)
3574
# source_ids is the worst possible case we may need to pull.
3575
# now we want to filter source_ids against what we actually
3576
# have in target, but don't try to check for existence where we know
3577
# we do not have a revision as that would be pointless.
3578
target_ids = set(self.target._all_possible_ids())
3579
possibly_present_revisions = target_ids.intersection(source_ids_set)
3580
actually_present_revisions = set(
3581
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3582
required_revisions = source_ids_set.difference(actually_present_revisions)
3583
if revision_id is not None:
3584
# we used get_ancestry to determine source_ids then we are assured all
3585
# revisions referenced are present as they are installed in topological order.
3586
# and the tip revision was validated by get_ancestry.
3587
result_set = required_revisions
3589
# if we just grabbed the possibly available ids, then
3590
# we only have an estimate of whats available and need to validate
3591
# that against the revision records.
3593
self.source._eliminate_revisions_not_present(required_revisions))
3594
return self.source.revision_ids_to_search_result(result_set)
3597
class InterKnitRepo(InterSameDataRepository):
3598
"""Optimised code paths between Knit based repositories."""
3601
def _get_repo_format_to_test(self):
3602
from bzrlib.repofmt import knitrepo
3603
return knitrepo.RepositoryFormatKnit1()
3606
def is_compatible(source, target):
3607
"""Be compatible with known Knit formats.
3609
We don't test for the stores being of specific types because that
3610
could lead to confusing results, and there is no need to be
3613
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3615
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3616
isinstance(target._format, RepositoryFormatKnit))
3617
except AttributeError:
3619
return are_knits and InterRepository._same_model(source, target)
3622
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3623
"""See InterRepository.missing_revision_ids()."""
3624
if revision_id is not None:
3625
source_ids = self.source.get_ancestry(revision_id)
3626
if source_ids[0] is not None:
3627
raise AssertionError()
3630
source_ids = self.source.all_revision_ids()
3631
source_ids_set = set(source_ids)
3632
# source_ids is the worst possible case we may need to pull.
3633
# now we want to filter source_ids against what we actually
3634
# have in target, but don't try to check for existence where we know
3635
# we do not have a revision as that would be pointless.
3636
target_ids = set(self.target.all_revision_ids())
3637
possibly_present_revisions = target_ids.intersection(source_ids_set)
3638
actually_present_revisions = set(
3639
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3640
required_revisions = source_ids_set.difference(actually_present_revisions)
3641
if revision_id is not None:
3642
# we used get_ancestry to determine source_ids then we are assured all
3643
# revisions referenced are present as they are installed in topological order.
3644
# and the tip revision was validated by get_ancestry.
3645
result_set = required_revisions
3647
# if we just grabbed the possibly available ids, then
3648
# we only have an estimate of whats available and need to validate
3649
# that against the revision records.
3651
self.source._eliminate_revisions_not_present(required_revisions))
3652
return self.source.revision_ids_to_search_result(result_set)
3655
class InterDifferingSerializer(InterRepository):
3658
def _get_repo_format_to_test(self):
3662
def is_compatible(source, target):
3663
"""Be compatible with Knit2 source and Knit3 target"""
3664
# This is redundant with format.check_conversion_target(), however that
3665
# raises an exception, and we just want to say "False" as in we won't
3666
# support converting between these formats.
3667
if source.supports_rich_root() and not target.supports_rich_root():
3669
if (source._format.supports_tree_reference
3670
and not target._format.supports_tree_reference):
3674
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3675
"""Get the best delta and base for this revision.
3677
:return: (basis_id, delta)
3679
possible_trees = [(parent_id, cache[parent_id])
3680
for parent_id in parent_ids
3681
if parent_id in cache]
3682
if len(possible_trees) == 0:
3683
# There either aren't any parents, or the parents aren't in the
3684
# cache, so just use the last converted tree
3685
possible_trees.append((basis_id, cache[basis_id]))
3687
for basis_id, basis_tree in possible_trees:
3688
delta = tree.inventory._make_delta(basis_tree.inventory)
3689
deltas.append((len(delta), basis_id, delta))
3691
return deltas[0][1:]
3693
def _get_parent_keys(self, root_key, parent_map):
3694
"""Get the parent keys for a given root id."""
3695
root_id, rev_id = root_key
3696
# Include direct parents of the revision, but only if they used
3697
# the same root_id and are heads.
3699
for parent_id in parent_map[rev_id]:
3700
if parent_id == _mod_revision.NULL_REVISION:
3702
if parent_id not in self._revision_id_to_root_id:
3703
# We probably didn't read this revision, go spend the
3704
# extra effort to actually check
3706
tree = self.source.revision_tree(parent_id)
3707
except errors.NoSuchRevision:
3708
# Ghost, fill out _revision_id_to_root_id in case we
3709
# encounter this again.
3710
# But set parent_root_id to None since we don't really know
3711
parent_root_id = None
3713
parent_root_id = tree.get_root_id()
3714
self._revision_id_to_root_id[parent_id] = None
3716
parent_root_id = self._revision_id_to_root_id[parent_id]
3717
if root_id == parent_root_id:
3718
# With stacking we _might_ want to refer to a non-local
3719
# revision, but this code path only applies when we have the
3720
# full content available, so ghosts really are ghosts, not just
3721
# the edge of local data.
3722
parent_keys.append((parent_id,))
3724
# root_id may be in the parent anyway.
3726
tree = self.source.revision_tree(parent_id)
3727
except errors.NoSuchRevision:
3728
# ghost, can't refer to it.
3732
parent_keys.append((tree.inventory[root_id].revision,))
3733
except errors.NoSuchId:
3736
g = graph.Graph(self.source.revisions)
3737
heads = g.heads(parent_keys)
3739
for key in parent_keys:
3740
if key in heads and key not in selected_keys:
3741
selected_keys.append(key)
3742
return tuple([(root_id,)+ key for key in selected_keys])
3744
def _new_root_data_stream(self, root_keys_to_create, parent_map):
3745
for root_key in root_keys_to_create:
3746
parent_keys = self._get_parent_keys(root_key, parent_map)
3747
yield versionedfile.FulltextContentFactory(root_key,
3748
parent_keys, None, '')
3750
def _fetch_batch(self, revision_ids, basis_id, cache):
3751
"""Fetch across a few revisions.
3753
:param revision_ids: The revisions to copy
3754
:param basis_id: The revision_id of a tree that must be in cache, used
3755
as a basis for delta when no other base is available
3756
:param cache: A cache of RevisionTrees that we can use.
3757
:return: The revision_id of the last converted tree. The RevisionTree
3758
for it will be in cache
3760
# Walk though all revisions; get inventory deltas, copy referenced
3761
# texts that delta references, insert the delta, revision and
3763
root_keys_to_create = set()
3766
pending_revisions = []
3767
parent_map = self.source.get_parent_map(revision_ids)
3768
for tree in self.source.revision_trees(revision_ids):
3769
current_revision_id = tree.get_revision_id()
3770
parent_ids = parent_map.get(current_revision_id, ())
3771
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3773
if self._converting_to_rich_root:
3774
self._revision_id_to_root_id[current_revision_id] = \
3776
# Find text entries that need to be copied
3777
for old_path, new_path, file_id, entry in delta:
3778
if new_path is not None:
3781
if not self.target.supports_rich_root():
3782
# The target doesn't support rich root, so we don't
3785
if self._converting_to_rich_root:
3786
# This can't be copied normally, we have to insert
3788
root_keys_to_create.add((file_id, entry.revision))
3790
text_keys.add((file_id, entry.revision))
3791
revision = self.source.get_revision(current_revision_id)
3792
pending_deltas.append((basis_id, delta,
3793
current_revision_id, revision.parent_ids))
3794
pending_revisions.append(revision)
3795
cache[current_revision_id] = tree
3796
basis_id = current_revision_id
3798
from_texts = self.source.texts
3799
to_texts = self.target.texts
3800
if root_keys_to_create:
3801
root_stream = self._new_root_data_stream(root_keys_to_create,
3803
to_texts.insert_record_stream(root_stream)
3804
to_texts.insert_record_stream(from_texts.get_record_stream(
3805
text_keys, self.target._format._fetch_order,
3806
not self.target._format._fetch_uses_deltas))
3807
# insert inventory deltas
3808
for delta in pending_deltas:
3809
self.target.add_inventory_by_delta(*delta)
3810
if self.target._fallback_repositories:
3811
# Make sure this stacked repository has all the parent inventories
3812
# for the new revisions that we are about to insert. We do this
3813
# before adding the revisions so that no revision is added until
3814
# all the inventories it may depend on are added.
3815
# Note that this is overzealous, as we may have fetched these in an
3818
revision_ids = set()
3819
for revision in pending_revisions:
3820
revision_ids.add(revision.revision_id)
3821
parent_ids.update(revision.parent_ids)
3822
parent_ids.difference_update(revision_ids)
3823
parent_ids.discard(_mod_revision.NULL_REVISION)
3824
parent_map = self.source.get_parent_map(parent_ids)
3825
# we iterate over parent_map and not parent_ids because we don't
3826
# want to try copying any revision which is a ghost
3827
for parent_tree in self.source.revision_trees(parent_map):
3828
current_revision_id = parent_tree.get_revision_id()
3829
parents_parents = parent_map[current_revision_id]
3830
basis_id, delta = self._get_delta_for_revision(parent_tree,
3831
parents_parents, basis_id, cache)
3832
self.target.add_inventory_by_delta(
3833
basis_id, delta, current_revision_id, parents_parents)
3834
# insert signatures and revisions
3835
for revision in pending_revisions:
3837
signature = self.source.get_signature_text(
3838
revision.revision_id)
3839
self.target.add_signature_text(revision.revision_id,
3841
except errors.NoSuchRevision:
3843
self.target.add_revision(revision.revision_id, revision)
3846
def _fetch_all_revisions(self, revision_ids, pb):
3847
"""Fetch everything for the list of revisions.
3849
:param revision_ids: The list of revisions to fetch. Must be in
3851
:param pb: A ProgressTask
3854
basis_id, basis_tree = self._get_basis(revision_ids[0])
3856
cache = lru_cache.LRUCache(100)
3857
cache[basis_id] = basis_tree
3858
del basis_tree # We don't want to hang on to it here
3860
for offset in range(0, len(revision_ids), batch_size):
3861
self.target.start_write_group()
3863
pb.update('Transferring revisions', offset,
3865
batch = revision_ids[offset:offset+batch_size]
3866
basis_id = self._fetch_batch(batch, basis_id, cache)
3868
self.target.abort_write_group()
3871
hint = self.target.commit_write_group()
3874
if hints and self.target._format.pack_compresses:
3875
self.target.pack(hint=hints)
3876
pb.update('Transferring revisions', len(revision_ids),
3880
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3882
"""See InterRepository.fetch()."""
3883
if fetch_spec is not None:
3884
raise AssertionError("Not implemented yet...")
3885
if (not self.source.supports_rich_root()
3886
and self.target.supports_rich_root()):
3887
self._converting_to_rich_root = True
3888
self._revision_id_to_root_id = {}
3890
self._converting_to_rich_root = False
3891
revision_ids = self.target.search_missing_revision_ids(self.source,
3892
revision_id, find_ghosts=find_ghosts).get_keys()
3893
if not revision_ids:
3895
revision_ids = tsort.topo_sort(
3896
self.source.get_graph().get_parent_map(revision_ids))
3897
if not revision_ids:
3899
# Walk though all revisions; get inventory deltas, copy referenced
3900
# texts that delta references, insert the delta, revision and
3902
first_rev = self.source.get_revision(revision_ids[0])
3904
my_pb = ui.ui_factory.nested_progress_bar()
3907
symbol_versioning.warn(
3908
symbol_versioning.deprecated_in((1, 14, 0))
3909
% "pb parameter to fetch()")
3912
self._fetch_all_revisions(revision_ids, pb)
3914
if my_pb is not None:
3916
return len(revision_ids), 0
3918
def _get_basis(self, first_revision_id):
3919
"""Get a revision and tree which exists in the target.
3921
This assumes that first_revision_id is selected for transmission
3922
because all other ancestors are already present. If we can't find an
3923
ancestor we fall back to NULL_REVISION since we know that is safe.
3925
:return: (basis_id, basis_tree)
3927
first_rev = self.source.get_revision(first_revision_id)
3929
basis_id = first_rev.parent_ids[0]
3930
# only valid as a basis if the target has it
3931
self.target.get_revision(basis_id)
3932
# Try to get a basis tree - if its a ghost it will hit the
3933
# NoSuchRevision case.
3934
basis_tree = self.source.revision_tree(basis_id)
3935
except (IndexError, errors.NoSuchRevision):
3936
basis_id = _mod_revision.NULL_REVISION
3937
basis_tree = self.source.revision_tree(basis_id)
3938
return basis_id, basis_tree
3941
InterRepository.register_optimiser(InterDifferingSerializer)
3942
InterRepository.register_optimiser(InterSameDataRepository)
3943
InterRepository.register_optimiser(InterWeaveRepo)
3944
InterRepository.register_optimiser(InterKnitRepo)
3947
class CopyConverter(object):
3948
"""A repository conversion tool which just performs a copy of the content.
3950
This is slow but quite reliable.
3953
def __init__(self, target_format):
3954
"""Create a CopyConverter.
3956
:param target_format: The format the resulting repository should be.
3958
self.target_format = target_format
3960
def convert(self, repo, pb):
3961
"""Perform the conversion of to_convert, giving feedback via pb.
3963
:param to_convert: The disk object to convert.
3964
:param pb: a progress bar to use for progress information.
3969
# this is only useful with metadir layouts - separated repo content.
3970
# trigger an assertion if not such
3971
repo._format.get_format_string()
3972
self.repo_dir = repo.bzrdir
3973
self.step('Moving repository to repository.backup')
3974
self.repo_dir.transport.move('repository', 'repository.backup')
3975
backup_transport = self.repo_dir.transport.clone('repository.backup')
3976
repo._format.check_conversion_target(self.target_format)
3977
self.source_repo = repo._format.open(self.repo_dir,
3979
_override_transport=backup_transport)
3980
self.step('Creating new repository')
3981
converted = self.target_format.initialize(self.repo_dir,
3982
self.source_repo.is_shared())
3983
converted.lock_write()
3985
self.step('Copying content into repository.')
3986
self.source_repo.copy_content_into(converted)
3989
self.step('Deleting old repository content.')
3990
self.repo_dir.transport.delete_tree('repository.backup')
3991
self.pb.note('repository converted')
3993
def step(self, message):
3994
"""Update the pb by a step."""
3996
self.pb.update(message, self.count, self.total)
4008
def _unescaper(match, _map=_unescape_map):
4009
code = match.group(1)
4013
if not code.startswith('#'):
4015
return unichr(int(code[1:])).encode('utf8')
4021
def _unescape_xml(data):
4022
"""Unescape predefined XML entities in a string of data."""
4024
if _unescape_re is None:
4025
_unescape_re = re.compile('\&([^;]*);')
4026
return _unescape_re.sub(_unescaper, data)
4029
class _VersionedFileChecker(object):
4031
def __init__(self, repository, text_key_references=None, ancestors=None):
4032
self.repository = repository
4033
self.text_index = self.repository._generate_text_key_index(
4034
text_key_references=text_key_references, ancestors=ancestors)
4036
def calculate_file_version_parents(self, text_key):
4037
"""Calculate the correct parents for a file version according to
4040
parent_keys = self.text_index[text_key]
4041
if parent_keys == [_mod_revision.NULL_REVISION]:
4043
return tuple(parent_keys)
4045
def check_file_version_parents(self, texts, progress_bar=None):
4046
"""Check the parents stored in a versioned file are correct.
4048
It also detects file versions that are not referenced by their
4049
corresponding revision's inventory.
4051
:returns: A tuple of (wrong_parents, dangling_file_versions).
4052
wrong_parents is a dict mapping {revision_id: (stored_parents,
4053
correct_parents)} for each revision_id where the stored parents
4054
are not correct. dangling_file_versions is a set of (file_id,
4055
revision_id) tuples for versions that are present in this versioned
4056
file, but not used by the corresponding inventory.
4058
local_progress = None
4059
if progress_bar is None:
4060
local_progress = ui.ui_factory.nested_progress_bar()
4061
progress_bar = local_progress
4063
return self._check_file_version_parents(texts, progress_bar)
4066
local_progress.finished()
4068
def _check_file_version_parents(self, texts, progress_bar):
4069
"""See check_file_version_parents."""
4071
self.file_ids = set([file_id for file_id, _ in
4072
self.text_index.iterkeys()])
4073
# text keys is now grouped by file_id
4074
n_weaves = len(self.file_ids)
4075
files_in_revisions = {}
4076
revisions_of_files = {}
4077
n_versions = len(self.text_index)
4078
progress_bar.update('loading text store', 0, n_versions)
4079
parent_map = self.repository.texts.get_parent_map(self.text_index)
4080
# On unlistable transports this could well be empty/error...
4081
text_keys = self.repository.texts.keys()
4082
unused_keys = frozenset(text_keys) - set(self.text_index)
4083
for num, key in enumerate(self.text_index.iterkeys()):
4084
progress_bar.update('checking text graph', num, n_versions)
4085
correct_parents = self.calculate_file_version_parents(key)
4087
knit_parents = parent_map[key]
4088
except errors.RevisionNotPresent:
4091
if correct_parents != knit_parents:
4092
wrong_parents[key] = (knit_parents, correct_parents)
4093
return wrong_parents, unused_keys
4096
def _old_get_graph(repository, revision_id):
4097
"""DO NOT USE. That is all. I'm serious."""
4098
graph = repository.get_graph()
4099
revision_graph = dict(((key, value) for key, value in
4100
graph.iter_ancestry([revision_id]) if value is not None))
4101
return _strip_NULL_ghosts(revision_graph)
4104
def _strip_NULL_ghosts(revision_graph):
4105
"""Also don't use this. more compatibility code for unmigrated clients."""
4106
# Filter ghosts, and null:
4107
if _mod_revision.NULL_REVISION in revision_graph:
4108
del revision_graph[_mod_revision.NULL_REVISION]
4109
for key, parents in revision_graph.items():
4110
revision_graph[key] = tuple(parent for parent in parents if parent
4112
return revision_graph
4115
class StreamSink(object):
4116
"""An object that can insert a stream into a repository.
4118
This interface handles the complexity of reserialising inventories and
4119
revisions from different formats, and allows unidirectional insertion into
4120
stacked repositories without looking for the missing basis parents
4124
def __init__(self, target_repo):
4125
self.target_repo = target_repo
4127
def insert_stream(self, stream, src_format, resume_tokens):
4128
"""Insert a stream's content into the target repository.
4130
:param src_format: a bzr repository format.
4132
:return: a list of resume tokens and an iterable of keys additional
4133
items required before the insertion can be completed.
4135
self.target_repo.lock_write()
4138
self.target_repo.resume_write_group(resume_tokens)
4141
self.target_repo.start_write_group()
4144
# locked_insert_stream performs a commit|suspend.
4145
return self._locked_insert_stream(stream, src_format, is_resume)
4147
self.target_repo.abort_write_group(suppress_errors=True)
4150
self.target_repo.unlock()
4152
def _locked_insert_stream(self, stream, src_format, is_resume):
4153
to_serializer = self.target_repo._format._serializer
4154
src_serializer = src_format._serializer
4156
if to_serializer == src_serializer:
4157
# If serializers match and the target is a pack repository, set the
4158
# write cache size on the new pack. This avoids poor performance
4159
# on transports where append is unbuffered (such as
4160
# RemoteTransport). This is safe to do because nothing should read
4161
# back from the target repository while a stream with matching
4162
# serialization is being inserted.
4163
# The exception is that a delta record from the source that should
4164
# be a fulltext may need to be expanded by the target (see
4165
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4166
# explicitly flush any buffered writes first in that rare case.
4168
new_pack = self.target_repo._pack_collection._new_pack
4169
except AttributeError:
4170
# Not a pack repository
4173
new_pack.set_write_cache_size(1024*1024)
4174
for substream_type, substream in stream:
4175
if substream_type == 'texts':
4176
self.target_repo.texts.insert_record_stream(substream)
4177
elif substream_type == 'inventories':
4178
if src_serializer == to_serializer:
4179
self.target_repo.inventories.insert_record_stream(
4182
self._extract_and_insert_inventories(
4183
substream, src_serializer)
4184
elif substream_type == 'chk_bytes':
4185
# XXX: This doesn't support conversions, as it assumes the
4186
# conversion was done in the fetch code.
4187
self.target_repo.chk_bytes.insert_record_stream(substream)
4188
elif substream_type == 'revisions':
4189
# This may fallback to extract-and-insert more often than
4190
# required if the serializers are different only in terms of
4192
if src_serializer == to_serializer:
4193
self.target_repo.revisions.insert_record_stream(
4196
self._extract_and_insert_revisions(substream,
4198
elif substream_type == 'signatures':
4199
self.target_repo.signatures.insert_record_stream(substream)
4201
raise AssertionError('kaboom! %s' % (substream_type,))
4202
# Done inserting data, and the missing_keys calculations will try to
4203
# read back from the inserted data, so flush the writes to the new pack
4204
# (if this is pack format).
4205
if new_pack is not None:
4206
new_pack._write_data('', flush=True)
4207
# Find all the new revisions (including ones from resume_tokens)
4208
missing_keys = self.target_repo.get_missing_parent_inventories(
4209
check_for_missing_texts=is_resume)
4211
for prefix, versioned_file in (
4212
('texts', self.target_repo.texts),
4213
('inventories', self.target_repo.inventories),
4214
('revisions', self.target_repo.revisions),
4215
('signatures', self.target_repo.signatures),
4216
('chk_bytes', self.target_repo.chk_bytes),
4218
if versioned_file is None:
4220
missing_keys.update((prefix,) + key for key in
4221
versioned_file.get_missing_compression_parent_keys())
4222
except NotImplementedError:
4223
# cannot even attempt suspending, and missing would have failed
4224
# during stream insertion.
4225
missing_keys = set()
4228
# suspend the write group and tell the caller what we is
4229
# missing. We know we can suspend or else we would not have
4230
# entered this code path. (All repositories that can handle
4231
# missing keys can handle suspending a write group).
4232
write_group_tokens = self.target_repo.suspend_write_group()
4233
return write_group_tokens, missing_keys
4234
hint = self.target_repo.commit_write_group()
4235
if (to_serializer != src_serializer and
4236
self.target_repo._format.pack_compresses):
4237
self.target_repo.pack(hint=hint)
4240
def _extract_and_insert_inventories(self, substream, serializer):
4241
"""Generate a new inventory versionedfile in target, converting data.
4243
The inventory is retrieved from the source, (deserializing it), and
4244
stored in the target (reserializing it in a different format).
4246
for record in substream:
4247
bytes = record.get_bytes_as('fulltext')
4248
revision_id = record.key[0]
4249
inv = serializer.read_inventory_from_string(bytes, revision_id)
4250
parents = [key[0] for key in record.parents]
4251
self.target_repo.add_inventory(revision_id, inv, parents)
4253
def _extract_and_insert_revisions(self, substream, serializer):
4254
for record in substream:
4255
bytes = record.get_bytes_as('fulltext')
4256
revision_id = record.key[0]
4257
rev = serializer.read_revision_from_string(bytes)
4258
if rev.revision_id != revision_id:
4259
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4260
self.target_repo.add_revision(revision_id, rev)
4263
if self.target_repo._format._fetch_reconcile:
4264
self.target_repo.reconcile()
4267
class StreamSource(object):
4268
"""A source of a stream for fetching between repositories."""
4270
def __init__(self, from_repository, to_format):
4271
"""Create a StreamSource streaming from from_repository."""
4272
self.from_repository = from_repository
4273
self.to_format = to_format
4275
def delta_on_metadata(self):
4276
"""Return True if delta's are permitted on metadata streams.
4278
That is on revisions and signatures.
4280
src_serializer = self.from_repository._format._serializer
4281
target_serializer = self.to_format._serializer
4282
return (self.to_format._fetch_uses_deltas and
4283
src_serializer == target_serializer)
4285
def _fetch_revision_texts(self, revs):
4286
# fetch signatures first and then the revision texts
4287
# may need to be a InterRevisionStore call here.
4288
from_sf = self.from_repository.signatures
4289
# A missing signature is just skipped.
4290
keys = [(rev_id,) for rev_id in revs]
4291
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4293
self.to_format._fetch_order,
4294
not self.to_format._fetch_uses_deltas))
4295
# If a revision has a delta, this is actually expanded inside the
4296
# insert_record_stream code now, which is an alternate fix for
4298
from_rf = self.from_repository.revisions
4299
revisions = from_rf.get_record_stream(
4301
self.to_format._fetch_order,
4302
not self.delta_on_metadata())
4303
return [('signatures', signatures), ('revisions', revisions)]
4305
def _generate_root_texts(self, revs):
4306
"""This will be called by __fetch between fetching weave texts and
4307
fetching the inventory weave.
4309
Subclasses should override this if they need to generate root texts
4310
after fetching weave texts.
4312
if self._rich_root_upgrade():
4314
return bzrlib.fetch.Inter1and2Helper(
4315
self.from_repository).generate_root_texts(revs)
4319
def get_stream(self, search):
4321
revs = search.get_keys()
4322
graph = self.from_repository.get_graph()
4323
revs = list(graph.iter_topo_order(revs))
4324
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4326
for knit_kind, file_id, revisions in data_to_fetch:
4327
if knit_kind != phase:
4329
# Make a new progress bar for this phase
4330
if knit_kind == "file":
4331
# Accumulate file texts
4332
text_keys.extend([(file_id, revision) for revision in
4334
elif knit_kind == "inventory":
4335
# Now copy the file texts.
4336
from_texts = self.from_repository.texts
4337
yield ('texts', from_texts.get_record_stream(
4338
text_keys, self.to_format._fetch_order,
4339
not self.to_format._fetch_uses_deltas))
4340
# Cause an error if a text occurs after we have done the
4343
# Before we process the inventory we generate the root
4344
# texts (if necessary) so that the inventories references
4346
for _ in self._generate_root_texts(revs):
4348
# NB: This currently reopens the inventory weave in source;
4349
# using a single stream interface instead would avoid this.
4350
from_weave = self.from_repository.inventories
4351
# we fetch only the referenced inventories because we do not
4352
# know for unselected inventories whether all their required
4353
# texts are present in the other repository - it could be
4355
for info in self._get_inventory_stream(revs):
4357
elif knit_kind == "signatures":
4358
# Nothing to do here; this will be taken care of when
4359
# _fetch_revision_texts happens.
4361
elif knit_kind == "revisions":
4362
for record in self._fetch_revision_texts(revs):
4365
raise AssertionError("Unknown knit kind %r" % knit_kind)
4367
def get_stream_for_missing_keys(self, missing_keys):
4368
# missing keys can only occur when we are byte copying and not
4369
# translating (because translation means we don't send
4370
# unreconstructable deltas ever).
4372
keys['texts'] = set()
4373
keys['revisions'] = set()
4374
keys['inventories'] = set()
4375
keys['chk_bytes'] = set()
4376
keys['signatures'] = set()
4377
for key in missing_keys:
4378
keys[key[0]].add(key[1:])
4379
if len(keys['revisions']):
4380
# If we allowed copying revisions at this point, we could end up
4381
# copying a revision without copying its required texts: a
4382
# violation of the requirements for repository integrity.
4383
raise AssertionError(
4384
'cannot copy revisions to fill in missing deltas %s' % (
4385
keys['revisions'],))
4386
for substream_kind, keys in keys.iteritems():
4387
vf = getattr(self.from_repository, substream_kind)
4388
if vf is None and keys:
4389
raise AssertionError(
4390
"cannot fill in keys for a versioned file we don't"
4391
" have: %s needs %s" % (substream_kind, keys))
4393
# No need to stream something we don't have
4395
# Ask for full texts always so that we don't need more round trips
4396
# after this stream.
4397
# Some of the missing keys are genuinely ghosts, so filter absent
4398
# records. The Sink is responsible for doing another check to
4399
# ensure that ghosts don't introduce missing data for future
4401
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4402
self.to_format._fetch_order, True))
4403
yield substream_kind, stream
4405
def inventory_fetch_order(self):
4406
if self._rich_root_upgrade():
4407
return 'topological'
4409
return self.to_format._fetch_order
4411
def _rich_root_upgrade(self):
4412
return (not self.from_repository._format.rich_root_data and
4413
self.to_format.rich_root_data)
4415
def _get_inventory_stream(self, revision_ids):
4416
from_format = self.from_repository._format
4417
if (from_format.supports_chks and self.to_format.supports_chks
4418
and (from_format._serializer == self.to_format._serializer)):
4419
# Both sides support chks, and they use the same serializer, so it
4420
# is safe to transmit the chk pages and inventory pages across
4422
return self._get_chk_inventory_stream(revision_ids)
4423
elif (not from_format.supports_chks):
4424
# Source repository doesn't support chks. So we can transmit the
4425
# inventories 'as-is' and either they are just accepted on the
4426
# target, or the Sink will properly convert it.
4427
return self._get_simple_inventory_stream(revision_ids)
4429
# XXX: Hack to make not-chk->chk fetch: copy the inventories as
4430
# inventories. Note that this should probably be done somehow
4431
# as part of bzrlib.repository.StreamSink. Except JAM couldn't
4432
# figure out how a non-chk repository could possibly handle
4433
# deserializing an inventory stream from a chk repo, as it
4434
# doesn't have a way to understand individual pages.
4435
return self._get_convertable_inventory_stream(revision_ids)
4437
def _get_simple_inventory_stream(self, revision_ids):
4438
from_weave = self.from_repository.inventories
4439
yield ('inventories', from_weave.get_record_stream(
4440
[(rev_id,) for rev_id in revision_ids],
4441
self.inventory_fetch_order(),
4442
not self.delta_on_metadata()))
4444
def _get_chk_inventory_stream(self, revision_ids):
4445
"""Fetch the inventory texts, along with the associated chk maps."""
4446
# We want an inventory outside of the search set, so that we can filter
4447
# out uninteresting chk pages. For now we use
4448
# _find_revision_outside_set, but if we had a Search with cut_revs, we
4449
# could use that instead.
4450
start_rev_id = self.from_repository._find_revision_outside_set(
4452
start_rev_key = (start_rev_id,)
4453
inv_keys_to_fetch = [(rev_id,) for rev_id in revision_ids]
4454
if start_rev_id != _mod_revision.NULL_REVISION:
4455
inv_keys_to_fetch.append((start_rev_id,))
4456
# Any repo that supports chk_bytes must also support out-of-order
4457
# insertion. At least, that is how we expect it to work
4458
# We use get_record_stream instead of iter_inventories because we want
4459
# to be able to insert the stream as well. We could instead fetch
4460
# allowing deltas, and then iter_inventories, but we don't know whether
4461
# source or target is more 'local' anway.
4462
inv_stream = self.from_repository.inventories.get_record_stream(
4463
inv_keys_to_fetch, 'unordered',
4464
True) # We need them as full-texts so we can find their references
4465
uninteresting_chk_roots = set()
4466
interesting_chk_roots = set()
4467
def filter_inv_stream(inv_stream):
4468
for idx, record in enumerate(inv_stream):
4469
### child_pb.update('fetch inv', idx, len(inv_keys_to_fetch))
4470
bytes = record.get_bytes_as('fulltext')
4471
chk_inv = inventory.CHKInventory.deserialise(
4472
self.from_repository.chk_bytes, bytes, record.key)
4473
if record.key == start_rev_key:
4474
uninteresting_chk_roots.add(chk_inv.id_to_entry.key())
4475
p_id_map = chk_inv.parent_id_basename_to_file_id
4476
if p_id_map is not None:
4477
uninteresting_chk_roots.add(p_id_map.key())
4480
interesting_chk_roots.add(chk_inv.id_to_entry.key())
4481
p_id_map = chk_inv.parent_id_basename_to_file_id
4482
if p_id_map is not None:
4483
interesting_chk_roots.add(p_id_map.key())
4484
### pb.update('fetch inventory', 0, 2)
4485
yield ('inventories', filter_inv_stream(inv_stream))
4486
# Now that we have worked out all of the interesting root nodes, grab
4487
# all of the interesting pages and insert them
4488
### pb.update('fetch inventory', 1, 2)
4489
interesting = chk_map.iter_interesting_nodes(
4490
self.from_repository.chk_bytes, interesting_chk_roots,
4491
uninteresting_chk_roots)
4492
def to_stream_adapter():
4493
"""Adapt the iter_interesting_nodes result to a single stream.
4495
iter_interesting_nodes returns records as it processes them, along
4496
with keys. However, we only want to return the records themselves.
4498
for record, items in interesting:
4499
if record is not None:
4501
# XXX: We could instead call get_record_stream(records.keys())
4502
# ATM, this will always insert the records as fulltexts, and
4503
# requires that you can hang on to records once you have gone
4504
# on to the next one. Further, it causes the target to
4505
# recompress the data. Testing shows it to be faster than
4506
# requesting the records again, though.
4507
yield ('chk_bytes', to_stream_adapter())
4508
### pb.update('fetch inventory', 2, 2)
4510
def _get_convertable_inventory_stream(self, revision_ids):
4511
# XXX: One of source or target is using chks, and they don't have
4512
# compatible serializations. The StreamSink code expects to be
4513
# able to convert on the target, so we need to put
4514
# bytes-on-the-wire that can be converted
4515
yield ('inventories', self._stream_invs_as_fulltexts(revision_ids))
4517
def _stream_invs_as_fulltexts(self, revision_ids):
4518
from_repo = self.from_repository
4519
from_serializer = from_repo._format._serializer
4520
revision_keys = [(rev_id,) for rev_id in revision_ids]
4521
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4522
for inv in self.from_repository.iter_inventories(revision_ids):
4523
# XXX: This is a bit hackish, but it works. Basically,
4524
# CHKSerializer 'accidentally' supports
4525
# read/write_inventory_to_string, even though that is never
4526
# the format that is stored on disk. It *does* give us a
4527
# single string representation for an inventory, so live with
4529
# This would be far better if we had a 'serialized inventory
4530
# delta' form. Then we could use 'inventory._make_delta', and
4531
# transmit that. This would both be faster to generate, and
4532
# result in fewer bytes-on-the-wire.
4533
as_bytes = from_serializer.write_inventory_to_string(inv)
4534
key = (inv.revision_id,)
4535
parent_keys = parent_map.get(key, ())
4536
yield versionedfile.FulltextContentFactory(
4537
key, parent_keys, None, as_bytes)
4540
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4541
stop_revision=None):
4542
"""Extend the partial history to include a given index
4544
If a stop_index is supplied, stop when that index has been reached.
4545
If a stop_revision is supplied, stop when that revision is
4546
encountered. Otherwise, stop when the beginning of history is
4549
:param stop_index: The index which should be present. When it is
4550
present, history extension will stop.
4551
:param stop_revision: The revision id which should be present. When
4552
it is encountered, history extension will stop.
4554
start_revision = partial_history_cache[-1]
4555
iterator = repo.iter_reverse_revision_history(start_revision)
4557
#skip the last revision in the list
4560
if (stop_index is not None and
4561
len(partial_history_cache) > stop_index):
4563
if partial_history_cache[-1] == stop_revision:
4565
revision_id = iterator.next()
4566
partial_history_cache.append(revision_id)
4567
except StopIteration:
added
removed
bzrlib/repository.py
