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
result = self._commit_builder_class(self, parents, config,
1699
timestamp, timezone, committer, revprops, revision_id)
1700
self.start_write_group()
1704
if (self.control_files._lock_count == 1 and
1705
self.control_files._lock_mode == 'w'):
1706
if self._write_group is not None:
1707
self.abort_write_group()
1708
self.control_files.unlock()
1709
raise errors.BzrError(
1710
'Must end write groups before releasing write locks.')
1711
self.control_files.unlock()
1712
if self.control_files._lock_count == 0:
1713
self._inventory_entry_cache.clear()
1714
for repo in self._fallback_repositories:
1718
def clone(self, a_bzrdir, revision_id=None):
1719
"""Clone this repository into a_bzrdir using the current format.
1721
Currently no check is made that the format of this repository and
1722
the bzrdir format are compatible. FIXME RBC 20060201.
1724
:return: The newly created destination repository.
1726
# TODO: deprecate after 0.16; cloning this with all its settings is
1727
# probably not very useful -- mbp 20070423
1728
dest_repo = self._create_sprouting_repo(a_bzrdir, shared=self.is_shared())
1729
self.copy_content_into(dest_repo, revision_id)
1732
def start_write_group(self):
1733
"""Start a write group in the repository.
1735
Write groups are used by repositories which do not have a 1:1 mapping
1736
between file ids and backend store to manage the insertion of data from
1737
both fetch and commit operations.
1739
A write lock is required around the start_write_group/commit_write_group
1740
for the support of lock-requiring repository formats.
1742
One can only insert data into a repository inside a write group.
1746
if not self.is_write_locked():
1747
raise errors.NotWriteLocked(self)
1748
if self._write_group:
1749
raise errors.BzrError('already in a write group')
1750
self._start_write_group()
1751
# so we can detect unlock/relock - the write group is now entered.
1752
self._write_group = self.get_transaction()
1754
def _start_write_group(self):
1755
"""Template method for per-repository write group startup.
1757
This is called before the write group is considered to be
1762
def sprout(self, to_bzrdir, revision_id=None):
1763
"""Create a descendent repository for new development.
1765
Unlike clone, this does not copy the settings of the repository.
1767
dest_repo = self._create_sprouting_repo(to_bzrdir, shared=False)
1768
dest_repo.fetch(self, revision_id=revision_id)
1771
def _create_sprouting_repo(self, a_bzrdir, shared):
1772
if not isinstance(a_bzrdir._format, self.bzrdir._format.__class__):
1773
# use target default format.
1774
dest_repo = a_bzrdir.create_repository()
1776
# Most control formats need the repository to be specifically
1777
# created, but on some old all-in-one formats it's not needed
1779
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
1780
except errors.UninitializableFormat:
1781
dest_repo = a_bzrdir.open_repository()
1784
def _get_sink(self):
1785
"""Return a sink for streaming into this repository."""
1786
return StreamSink(self)
1788
def _get_source(self, to_format):
1789
"""Return a source for streaming from this repository."""
1790
return StreamSource(self, to_format)
1793
def has_revision(self, revision_id):
1794
"""True if this repository has a copy of the revision."""
1795
return revision_id in self.has_revisions((revision_id,))
1798
def has_revisions(self, revision_ids):
1799
"""Probe to find out the presence of multiple revisions.
1801
:param revision_ids: An iterable of revision_ids.
1802
:return: A set of the revision_ids that were present.
1804
parent_map = self.revisions.get_parent_map(
1805
[(rev_id,) for rev_id in revision_ids])
1807
if _mod_revision.NULL_REVISION in revision_ids:
1808
result.add(_mod_revision.NULL_REVISION)
1809
result.update([key[0] for key in parent_map])
1813
def get_revision(self, revision_id):
1814
"""Return the Revision object for a named revision."""
1815
return self.get_revisions([revision_id])[0]
1818
def get_revision_reconcile(self, revision_id):
1819
"""'reconcile' helper routine that allows access to a revision always.
1821
This variant of get_revision does not cross check the weave graph
1822
against the revision one as get_revision does: but it should only
1823
be used by reconcile, or reconcile-alike commands that are correcting
1824
or testing the revision graph.
1826
return self._get_revisions([revision_id])[0]
1829
def get_revisions(self, revision_ids):
1830
"""Get many revisions at once.
1832
Repositories that need to check data on every revision read should
1833
subclass this method.
1835
return self._get_revisions(revision_ids)
1838
def _get_revisions(self, revision_ids):
1839
"""Core work logic to get many revisions without sanity checks."""
1841
for revid, rev in self._iter_revisions(revision_ids):
1843
raise errors.NoSuchRevision(self, revid)
1845
return [revs[revid] for revid in revision_ids]
1847
def _iter_revisions(self, revision_ids):
1848
"""Iterate over revision objects.
1850
:param revision_ids: An iterable of revisions to examine. None may be
1851
passed to request all revisions known to the repository. Note that
1852
not all repositories can find unreferenced revisions; for those
1853
repositories only referenced ones will be returned.
1854
:return: An iterator of (revid, revision) tuples. Absent revisions (
1855
those asked for but not available) are returned as (revid, None).
1857
if revision_ids is None:
1858
revision_ids = self.all_revision_ids()
1860
for rev_id in revision_ids:
1861
if not rev_id or not isinstance(rev_id, basestring):
1862
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1863
keys = [(key,) for key in revision_ids]
1864
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1865
for record in stream:
1866
revid = record.key[0]
1867
if record.storage_kind == 'absent':
1870
text = record.get_bytes_as('fulltext')
1871
rev = self._serializer.read_revision_from_string(text)
1875
def get_revision_xml(self, revision_id):
1876
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1877
# would have already do it.
1878
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1879
# TODO: this can't just be replaced by:
1880
# return self._serializer.write_revision_to_string(
1881
# self.get_revision(revision_id))
1882
# as cStringIO preservers the encoding unlike write_revision_to_string
1883
# or some other call down the path.
1884
rev = self.get_revision(revision_id)
1885
rev_tmp = cStringIO.StringIO()
1886
# the current serializer..
1887
self._serializer.write_revision(rev, rev_tmp)
1889
return rev_tmp.getvalue()
1891
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1892
"""Produce a generator of revision deltas.
1894
Note that the input is a sequence of REVISIONS, not revision_ids.
1895
Trees will be held in memory until the generator exits.
1896
Each delta is relative to the revision's lefthand predecessor.
1898
:param specific_fileids: if not None, the result is filtered
1899
so that only those file-ids, their parents and their
1900
children are included.
1902
# Get the revision-ids of interest
1903
required_trees = set()
1904
for revision in revisions:
1905
required_trees.add(revision.revision_id)
1906
required_trees.update(revision.parent_ids[:1])
1908
# Get the matching filtered trees. Note that it's more
1909
# efficient to pass filtered trees to changes_from() rather
1910
# than doing the filtering afterwards. changes_from() could
1911
# arguably do the filtering itself but it's path-based, not
1912
# file-id based, so filtering before or afterwards is
1914
if specific_fileids is None:
1915
trees = dict((t.get_revision_id(), t) for
1916
t in self.revision_trees(required_trees))
1918
trees = dict((t.get_revision_id(), t) for
1919
t in self._filtered_revision_trees(required_trees,
1922
# Calculate the deltas
1923
for revision in revisions:
1924
if not revision.parent_ids:
1925
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
1927
old_tree = trees[revision.parent_ids[0]]
1928
yield trees[revision.revision_id].changes_from(old_tree)
1931
def get_revision_delta(self, revision_id, specific_fileids=None):
1932
"""Return the delta for one revision.
1934
The delta is relative to the left-hand predecessor of the
1937
:param specific_fileids: if not None, the result is filtered
1938
so that only those file-ids, their parents and their
1939
children are included.
1941
r = self.get_revision(revision_id)
1942
return list(self.get_deltas_for_revisions([r],
1943
specific_fileids=specific_fileids))[0]
1946
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
1947
signature = gpg_strategy.sign(plaintext)
1948
self.add_signature_text(revision_id, signature)
1951
def add_signature_text(self, revision_id, signature):
1952
self.signatures.add_lines((revision_id,), (),
1953
osutils.split_lines(signature))
1955
def find_text_key_references(self):
1956
"""Find the text key references within the repository.
1958
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1959
to whether they were referred to by the inventory of the
1960
revision_id that they contain. The inventory texts from all present
1961
revision ids are assessed to generate this report.
1963
revision_keys = self.revisions.keys()
1964
w = self.inventories
1965
pb = ui.ui_factory.nested_progress_bar()
1967
return self._find_text_key_references_from_xml_inventory_lines(
1968
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1972
def _find_text_key_references_from_xml_inventory_lines(self,
1974
"""Core routine for extracting references to texts from inventories.
1976
This performs the translation of xml lines to revision ids.
1978
:param line_iterator: An iterator of lines, origin_version_id
1979
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1980
to whether they were referred to by the inventory of the
1981
revision_id that they contain. Note that if that revision_id was
1982
not part of the line_iterator's output then False will be given -
1983
even though it may actually refer to that key.
1985
if not self._serializer.support_altered_by_hack:
1986
raise AssertionError(
1987
"_find_text_key_references_from_xml_inventory_lines only "
1988
"supported for branches which store inventory as unnested xml"
1989
", not on %r" % self)
1992
# this code needs to read every new line in every inventory for the
1993
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
1994
# not present in one of those inventories is unnecessary but not
1995
# harmful because we are filtering by the revision id marker in the
1996
# inventory lines : we only select file ids altered in one of those
1997
# revisions. We don't need to see all lines in the inventory because
1998
# only those added in an inventory in rev X can contain a revision=X
2000
unescape_revid_cache = {}
2001
unescape_fileid_cache = {}
2003
# jam 20061218 In a big fetch, this handles hundreds of thousands
2004
# of lines, so it has had a lot of inlining and optimizing done.
2005
# Sorry that it is a little bit messy.
2006
# Move several functions to be local variables, since this is a long
2008
search = self._file_ids_altered_regex.search
2009
unescape = _unescape_xml
2010
setdefault = result.setdefault
2011
for line, line_key in line_iterator:
2012
match = search(line)
2015
# One call to match.group() returning multiple items is quite a
2016
# bit faster than 2 calls to match.group() each returning 1
2017
file_id, revision_id = match.group('file_id', 'revision_id')
2019
# Inlining the cache lookups helps a lot when you make 170,000
2020
# lines and 350k ids, versus 8.4 unique ids.
2021
# Using a cache helps in 2 ways:
2022
# 1) Avoids unnecessary decoding calls
2023
# 2) Re-uses cached strings, which helps in future set and
2025
# (2) is enough that removing encoding entirely along with
2026
# the cache (so we are using plain strings) results in no
2027
# performance improvement.
2029
revision_id = unescape_revid_cache[revision_id]
2031
unescaped = unescape(revision_id)
2032
unescape_revid_cache[revision_id] = unescaped
2033
revision_id = unescaped
2035
# Note that unconditionally unescaping means that we deserialise
2036
# every fileid, which for general 'pull' is not great, but we don't
2037
# really want to have some many fulltexts that this matters anyway.
2040
file_id = unescape_fileid_cache[file_id]
2042
unescaped = unescape(file_id)
2043
unescape_fileid_cache[file_id] = unescaped
2046
key = (file_id, revision_id)
2047
setdefault(key, False)
2048
if revision_id == line_key[-1]:
2052
def _inventory_xml_lines_for_keys(self, keys):
2053
"""Get a line iterator of the sort needed for findind references.
2055
Not relevant for non-xml inventory repositories.
2057
Ghosts in revision_keys are ignored.
2059
:param revision_keys: The revision keys for the inventories to inspect.
2060
:return: An iterator over (inventory line, revid) for the fulltexts of
2061
all of the xml inventories specified by revision_keys.
2063
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2064
for record in stream:
2065
if record.storage_kind != 'absent':
2066
chunks = record.get_bytes_as('chunked')
2067
revid = record.key[-1]
2068
lines = osutils.chunks_to_lines(chunks)
2072
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2074
"""Helper routine for fileids_altered_by_revision_ids.
2076
This performs the translation of xml lines to revision ids.
2078
:param line_iterator: An iterator of lines, origin_version_id
2079
:param revision_keys: The revision ids to filter for. This should be a
2080
set or other type which supports efficient __contains__ lookups, as
2081
the revision key from each parsed line will be looked up in the
2082
revision_keys filter.
2083
:return: a dictionary mapping altered file-ids to an iterable of
2084
revision_ids. Each altered file-ids has the exact revision_ids that
2085
altered it listed explicitly.
2087
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2088
line_iterator).iterkeys())
2089
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2090
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2091
self._inventory_xml_lines_for_keys(parent_keys)))
2092
new_keys = seen - parent_seen
2094
setdefault = result.setdefault
2095
for key in new_keys:
2096
setdefault(key[0], set()).add(key[-1])
2099
def _find_parent_ids_of_revisions(self, revision_ids):
2100
"""Find all parent ids that are mentioned in the revision graph.
2102
:return: set of revisions that are parents of revision_ids which are
2103
not part of revision_ids themselves
2105
parent_map = self.get_parent_map(revision_ids)
2107
map(parent_ids.update, parent_map.itervalues())
2108
parent_ids.difference_update(revision_ids)
2109
parent_ids.discard(_mod_revision.NULL_REVISION)
2112
def _find_parent_keys_of_revisions(self, revision_keys):
2113
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2115
:param revision_keys: An iterable of revision_keys.
2116
:return: The parents of all revision_keys that are not already in
2119
parent_map = self.revisions.get_parent_map(revision_keys)
2121
map(parent_keys.update, parent_map.itervalues())
2122
parent_keys.difference_update(revision_keys)
2123
parent_keys.discard(_mod_revision.NULL_REVISION)
2126
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2127
"""Find the file ids and versions affected by revisions.
2129
:param revisions: an iterable containing revision ids.
2130
:param _inv_weave: The inventory weave from this repository or None.
2131
If None, the inventory weave will be opened automatically.
2132
:return: a dictionary mapping altered file-ids to an iterable of
2133
revision_ids. Each altered file-ids has the exact revision_ids that
2134
altered it listed explicitly.
2136
selected_keys = set((revid,) for revid in revision_ids)
2137
w = _inv_weave or self.inventories
2138
pb = ui.ui_factory.nested_progress_bar()
2140
return self._find_file_ids_from_xml_inventory_lines(
2141
w.iter_lines_added_or_present_in_keys(
2142
selected_keys, pb=pb),
2147
def iter_files_bytes(self, desired_files):
2148
"""Iterate through file versions.
2150
Files will not necessarily be returned in the order they occur in
2151
desired_files. No specific order is guaranteed.
2153
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2154
value supplied by the caller as part of desired_files. It should
2155
uniquely identify the file version in the caller's context. (Examples:
2156
an index number or a TreeTransform trans_id.)
2158
bytes_iterator is an iterable of bytestrings for the file. The
2159
kind of iterable and length of the bytestrings are unspecified, but for
2160
this implementation, it is a list of bytes produced by
2161
VersionedFile.get_record_stream().
2163
:param desired_files: a list of (file_id, revision_id, identifier)
2167
for file_id, revision_id, callable_data in desired_files:
2168
text_keys[(file_id, revision_id)] = callable_data
2169
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2170
if record.storage_kind == 'absent':
2171
raise errors.RevisionNotPresent(record.key, self)
2172
yield text_keys[record.key], record.get_bytes_as('chunked')
2174
def _generate_text_key_index(self, text_key_references=None,
2176
"""Generate a new text key index for the repository.
2178
This is an expensive function that will take considerable time to run.
2180
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2181
list of parents, also text keys. When a given key has no parents,
2182
the parents list will be [NULL_REVISION].
2184
# All revisions, to find inventory parents.
2185
if ancestors is None:
2186
graph = self.get_graph()
2187
ancestors = graph.get_parent_map(self.all_revision_ids())
2188
if text_key_references is None:
2189
text_key_references = self.find_text_key_references()
2190
pb = ui.ui_factory.nested_progress_bar()
2192
return self._do_generate_text_key_index(ancestors,
2193
text_key_references, pb)
2197
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2198
"""Helper for _generate_text_key_index to avoid deep nesting."""
2199
revision_order = tsort.topo_sort(ancestors)
2200
invalid_keys = set()
2202
for revision_id in revision_order:
2203
revision_keys[revision_id] = set()
2204
text_count = len(text_key_references)
2205
# a cache of the text keys to allow reuse; costs a dict of all the
2206
# keys, but saves a 2-tuple for every child of a given key.
2208
for text_key, valid in text_key_references.iteritems():
2210
invalid_keys.add(text_key)
2212
revision_keys[text_key[1]].add(text_key)
2213
text_key_cache[text_key] = text_key
2214
del text_key_references
2216
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2217
NULL_REVISION = _mod_revision.NULL_REVISION
2218
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2219
# too small for large or very branchy trees. However, for 55K path
2220
# trees, it would be easy to use too much memory trivially. Ideally we
2221
# could gauge this by looking at available real memory etc, but this is
2222
# always a tricky proposition.
2223
inventory_cache = lru_cache.LRUCache(10)
2224
batch_size = 10 # should be ~150MB on a 55K path tree
2225
batch_count = len(revision_order) / batch_size + 1
2227
pb.update("Calculating text parents", processed_texts, text_count)
2228
for offset in xrange(batch_count):
2229
to_query = revision_order[offset * batch_size:(offset + 1) *
2233
for revision_id in to_query:
2234
parent_ids = ancestors[revision_id]
2235
for text_key in revision_keys[revision_id]:
2236
pb.update("Calculating text parents", processed_texts)
2237
processed_texts += 1
2238
candidate_parents = []
2239
for parent_id in parent_ids:
2240
parent_text_key = (text_key[0], parent_id)
2242
check_parent = parent_text_key not in \
2243
revision_keys[parent_id]
2245
# the parent parent_id is a ghost:
2246
check_parent = False
2247
# truncate the derived graph against this ghost.
2248
parent_text_key = None
2250
# look at the parent commit details inventories to
2251
# determine possible candidates in the per file graph.
2254
inv = inventory_cache[parent_id]
2256
inv = self.revision_tree(parent_id).inventory
2257
inventory_cache[parent_id] = inv
2259
parent_entry = inv[text_key[0]]
2260
except (KeyError, errors.NoSuchId):
2262
if parent_entry is not None:
2264
text_key[0], parent_entry.revision)
2266
parent_text_key = None
2267
if parent_text_key is not None:
2268
candidate_parents.append(
2269
text_key_cache[parent_text_key])
2270
parent_heads = text_graph.heads(candidate_parents)
2271
new_parents = list(parent_heads)
2272
new_parents.sort(key=lambda x:candidate_parents.index(x))
2273
if new_parents == []:
2274
new_parents = [NULL_REVISION]
2275
text_index[text_key] = new_parents
2277
for text_key in invalid_keys:
2278
text_index[text_key] = [NULL_REVISION]
2281
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2282
"""Get an iterable listing the keys of all the data introduced by a set
2285
The keys will be ordered so that the corresponding items can be safely
2286
fetched and inserted in that order.
2288
:returns: An iterable producing tuples of (knit-kind, file-id,
2289
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2290
'revisions'. file-id is None unless knit-kind is 'file'.
2292
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2295
for result in self._find_non_file_keys_to_fetch(revision_ids):
2298
def _find_file_keys_to_fetch(self, revision_ids, pb):
2299
# XXX: it's a bit weird to control the inventory weave caching in this
2300
# generator. Ideally the caching would be done in fetch.py I think. Or
2301
# maybe this generator should explicitly have the contract that it
2302
# should not be iterated until the previously yielded item has been
2304
inv_w = self.inventories
2306
# file ids that changed
2307
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2309
num_file_ids = len(file_ids)
2310
for file_id, altered_versions in file_ids.iteritems():
2312
pb.update("fetch texts", count, num_file_ids)
2314
yield ("file", file_id, altered_versions)
2316
def _find_non_file_keys_to_fetch(self, revision_ids):
2318
yield ("inventory", None, revision_ids)
2321
# XXX: Note ATM no callers actually pay attention to this return
2322
# instead they just use the list of revision ids and ignore
2323
# missing sigs. Consider removing this work entirely
2324
revisions_with_signatures = set(self.signatures.get_parent_map(
2325
[(r,) for r in revision_ids]))
2326
revisions_with_signatures = set(
2327
[r for (r,) in revisions_with_signatures])
2328
revisions_with_signatures.intersection_update(revision_ids)
2329
yield ("signatures", None, revisions_with_signatures)
2332
yield ("revisions", None, revision_ids)
2335
def get_inventory(self, revision_id):
2336
"""Get Inventory object by revision id."""
2337
return self.iter_inventories([revision_id]).next()
2339
def iter_inventories(self, revision_ids):
2340
"""Get many inventories by revision_ids.
2342
This will buffer some or all of the texts used in constructing the
2343
inventories in memory, but will only parse a single inventory at a
2346
:param revision_ids: The expected revision ids of the inventories.
2347
:return: An iterator of inventories.
2349
if ((None in revision_ids)
2350
or (_mod_revision.NULL_REVISION in revision_ids)):
2351
raise ValueError('cannot get null revision inventory')
2352
return self._iter_inventories(revision_ids)
2354
def _iter_inventories(self, revision_ids):
2355
"""single-document based inventory iteration."""
2356
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2357
yield self.deserialise_inventory(revision_id, text)
2359
def _iter_inventory_xmls(self, revision_ids):
2360
keys = [(revision_id,) for revision_id in revision_ids]
2361
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2363
for record in stream:
2364
if record.storage_kind != 'absent':
2365
text_chunks[record.key] = record.get_bytes_as('chunked')
2367
raise errors.NoSuchRevision(self, record.key)
2369
chunks = text_chunks.pop(key)
2370
yield ''.join(chunks), key[-1]
2372
def deserialise_inventory(self, revision_id, xml):
2373
"""Transform the xml into an inventory object.
2375
:param revision_id: The expected revision id of the inventory.
2376
:param xml: A serialised inventory.
2378
result = self._serializer.read_inventory_from_string(xml, revision_id,
2379
entry_cache=self._inventory_entry_cache)
2380
if result.revision_id != revision_id:
2381
raise AssertionError('revision id mismatch %s != %s' % (
2382
result.revision_id, revision_id))
2385
def serialise_inventory(self, inv):
2386
return self._serializer.write_inventory_to_string(inv)
2388
def _serialise_inventory_to_lines(self, inv):
2389
return self._serializer.write_inventory_to_lines(inv)
2391
def get_serializer_format(self):
2392
return self._serializer.format_num
2395
def get_inventory_xml(self, revision_id):
2396
"""Get inventory XML as a file object."""
2397
texts = self._iter_inventory_xmls([revision_id])
2399
text, revision_id = texts.next()
2400
except StopIteration:
2401
raise errors.HistoryMissing(self, 'inventory', revision_id)
2405
def get_inventory_sha1(self, revision_id):
2406
"""Return the sha1 hash of the inventory entry
2408
return self.get_revision(revision_id).inventory_sha1
2410
def get_rev_id_for_revno(self, revno, known_pair):
2411
"""Return the revision id of a revno, given a later (revno, revid)
2412
pair in the same history.
2414
:return: if found (True, revid). If the available history ran out
2415
before reaching the revno, then this returns
2416
(False, (closest_revno, closest_revid)).
2418
known_revno, known_revid = known_pair
2419
partial_history = [known_revid]
2420
distance_from_known = known_revno - revno
2421
if distance_from_known < 0:
2423
'requested revno (%d) is later than given known revno (%d)'
2424
% (revno, known_revno))
2427
self, partial_history, stop_index=distance_from_known)
2428
except errors.RevisionNotPresent, err:
2429
if err.revision_id == known_revid:
2430
# The start revision (known_revid) wasn't found.
2432
# This is a stacked repository with no fallbacks, or a there's a
2433
# left-hand ghost. Either way, even though the revision named in
2434
# the error isn't in this repo, we know it's the next step in this
2435
# left-hand history.
2436
partial_history.append(err.revision_id)
2437
if len(partial_history) <= distance_from_known:
2438
# Didn't find enough history to get a revid for the revno.
2439
earliest_revno = known_revno - len(partial_history) + 1
2440
return (False, (earliest_revno, partial_history[-1]))
2441
if len(partial_history) - 1 > distance_from_known:
2442
raise AssertionError('_iter_for_revno returned too much history')
2443
return (True, partial_history[-1])
2445
def iter_reverse_revision_history(self, revision_id):
2446
"""Iterate backwards through revision ids in the lefthand history
2448
:param revision_id: The revision id to start with. All its lefthand
2449
ancestors will be traversed.
2451
graph = self.get_graph()
2452
next_id = revision_id
2454
if next_id in (None, _mod_revision.NULL_REVISION):
2457
parents = graph.get_parent_map([next_id])[next_id]
2459
raise errors.RevisionNotPresent(next_id, self)
2461
if len(parents) == 0:
2464
next_id = parents[0]
2467
def get_revision_inventory(self, revision_id):
2468
"""Return inventory of a past revision."""
2469
# TODO: Unify this with get_inventory()
2470
# bzr 0.0.6 and later imposes the constraint that the inventory_id
2471
# must be the same as its revision, so this is trivial.
2472
if revision_id is None:
2473
# This does not make sense: if there is no revision,
2474
# then it is the current tree inventory surely ?!
2475
# and thus get_root_id() is something that looks at the last
2476
# commit on the branch, and the get_root_id is an inventory check.
2477
raise NotImplementedError
2478
# return Inventory(self.get_root_id())
2480
return self.get_inventory(revision_id)
2482
def is_shared(self):
2483
"""Return True if this repository is flagged as a shared repository."""
2484
raise NotImplementedError(self.is_shared)
2487
def reconcile(self, other=None, thorough=False):
2488
"""Reconcile this repository."""
2489
from bzrlib.reconcile import RepoReconciler
2490
reconciler = RepoReconciler(self, thorough=thorough)
2491
reconciler.reconcile()
2494
def _refresh_data(self):
2495
"""Helper called from lock_* to ensure coherency with disk.
2497
The default implementation does nothing; it is however possible
2498
for repositories to maintain loaded indices across multiple locks
2499
by checking inside their implementation of this method to see
2500
whether their indices are still valid. This depends of course on
2501
the disk format being validatable in this manner. This method is
2502
also called by the refresh_data() public interface to cause a refresh
2503
to occur while in a write lock so that data inserted by a smart server
2504
push operation is visible on the client's instance of the physical
2509
def revision_tree(self, revision_id):
2510
"""Return Tree for a revision on this branch.
2512
`revision_id` may be NULL_REVISION for the empty tree revision.
2514
revision_id = _mod_revision.ensure_null(revision_id)
2515
# TODO: refactor this to use an existing revision object
2516
# so we don't need to read it in twice.
2517
if revision_id == _mod_revision.NULL_REVISION:
2518
return RevisionTree(self, Inventory(root_id=None),
2519
_mod_revision.NULL_REVISION)
2521
inv = self.get_revision_inventory(revision_id)
2522
return RevisionTree(self, inv, revision_id)
2524
def revision_trees(self, revision_ids):
2525
"""Return Trees for revisions in this repository.
2527
:param revision_ids: a sequence of revision-ids;
2528
a revision-id may not be None or 'null:'
2530
inventories = self.iter_inventories(revision_ids)
2531
for inv in inventories:
2532
yield RevisionTree(self, inv, inv.revision_id)
2534
def _filtered_revision_trees(self, revision_ids, file_ids):
2535
"""Return Tree for a revision on this branch with only some files.
2537
:param revision_ids: a sequence of revision-ids;
2538
a revision-id may not be None or 'null:'
2539
:param file_ids: if not None, the result is filtered
2540
so that only those file-ids, their parents and their
2541
children are included.
2543
inventories = self.iter_inventories(revision_ids)
2544
for inv in inventories:
2545
# Should we introduce a FilteredRevisionTree class rather
2546
# than pre-filter the inventory here?
2547
filtered_inv = inv.filter(file_ids)
2548
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
2551
def get_ancestry(self, revision_id, topo_sorted=True):
2552
"""Return a list of revision-ids integrated by a revision.
2554
The first element of the list is always None, indicating the origin
2555
revision. This might change when we have history horizons, or
2556
perhaps we should have a new API.
2558
This is topologically sorted.
2560
if _mod_revision.is_null(revision_id):
2562
if not self.has_revision(revision_id):
2563
raise errors.NoSuchRevision(self, revision_id)
2564
graph = self.get_graph()
2566
search = graph._make_breadth_first_searcher([revision_id])
2569
found, ghosts = search.next_with_ghosts()
2570
except StopIteration:
2573
if _mod_revision.NULL_REVISION in keys:
2574
keys.remove(_mod_revision.NULL_REVISION)
2576
parent_map = graph.get_parent_map(keys)
2577
keys = tsort.topo_sort(parent_map)
2578
return [None] + list(keys)
2580
def pack(self, hint=None):
2581
"""Compress the data within the repository.
2583
This operation only makes sense for some repository types. For other
2584
types it should be a no-op that just returns.
2586
This stub method does not require a lock, but subclasses should use
2587
@needs_write_lock as this is a long running call its reasonable to
2588
implicitly lock for the user.
2590
:param hint: If not supplied, the whole repository is packed.
2591
If supplied, the repository may use the hint parameter as a
2592
hint for the parts of the repository to pack. A hint can be
2593
obtained from the result of commit_write_group(). Out of
2594
date hints are simply ignored, because concurrent operations
2595
can obsolete them rapidly.
2598
def get_transaction(self):
2599
return self.control_files.get_transaction()
2601
def get_parent_map(self, revision_ids):
2602
"""See graph.StackedParentsProvider.get_parent_map"""
2603
# revisions index works in keys; this just works in revisions
2604
# therefore wrap and unwrap
2607
for revision_id in revision_ids:
2608
if revision_id == _mod_revision.NULL_REVISION:
2609
result[revision_id] = ()
2610
elif revision_id is None:
2611
raise ValueError('get_parent_map(None) is not valid')
2613
query_keys.append((revision_id ,))
2614
for ((revision_id,), parent_keys) in \
2615
self.revisions.get_parent_map(query_keys).iteritems():
2617
result[revision_id] = tuple(parent_revid
2618
for (parent_revid,) in parent_keys)
2620
result[revision_id] = (_mod_revision.NULL_REVISION,)
2623
def _make_parents_provider(self):
2626
def get_graph(self, other_repository=None):
2627
"""Return the graph walker for this repository format"""
2628
parents_provider = self._make_parents_provider()
2629
if (other_repository is not None and
2630
not self.has_same_location(other_repository)):
2631
parents_provider = graph.StackedParentsProvider(
2632
[parents_provider, other_repository._make_parents_provider()])
2633
return graph.Graph(parents_provider)
2635
def _get_versioned_file_checker(self, text_key_references=None,
2637
"""Return an object suitable for checking versioned files.
2639
:param text_key_references: if non-None, an already built
2640
dictionary mapping text keys ((fileid, revision_id) tuples)
2641
to whether they were referred to by the inventory of the
2642
revision_id that they contain. If None, this will be
2644
:param ancestors: Optional result from
2645
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2648
return _VersionedFileChecker(self,
2649
text_key_references=text_key_references, ancestors=ancestors)
2651
def revision_ids_to_search_result(self, result_set):
2652
"""Convert a set of revision ids to a graph SearchResult."""
2653
result_parents = set()
2654
for parents in self.get_graph().get_parent_map(
2655
result_set).itervalues():
2656
result_parents.update(parents)
2657
included_keys = result_set.intersection(result_parents)
2658
start_keys = result_set.difference(included_keys)
2659
exclude_keys = result_parents.difference(result_set)
2660
result = graph.SearchResult(start_keys, exclude_keys,
2661
len(result_set), result_set)
2665
def set_make_working_trees(self, new_value):
2666
"""Set the policy flag for making working trees when creating branches.
2668
This only applies to branches that use this repository.
2670
The default is 'True'.
2671
:param new_value: True to restore the default, False to disable making
2674
raise NotImplementedError(self.set_make_working_trees)
2676
def make_working_trees(self):
2677
"""Returns the policy for making working trees on new branches."""
2678
raise NotImplementedError(self.make_working_trees)
2681
def sign_revision(self, revision_id, gpg_strategy):
2682
plaintext = Testament.from_revision(self, revision_id).as_short_text()
2683
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2686
def has_signature_for_revision_id(self, revision_id):
2687
"""Query for a revision signature for revision_id in the repository."""
2688
if not self.has_revision(revision_id):
2689
raise errors.NoSuchRevision(self, revision_id)
2690
sig_present = (1 == len(
2691
self.signatures.get_parent_map([(revision_id,)])))
2695
def get_signature_text(self, revision_id):
2696
"""Return the text for a signature."""
2697
stream = self.signatures.get_record_stream([(revision_id,)],
2699
record = stream.next()
2700
if record.storage_kind == 'absent':
2701
raise errors.NoSuchRevision(self, revision_id)
2702
return record.get_bytes_as('fulltext')
2705
def check(self, revision_ids=None, callback_refs=None, check_repo=True):
2706
"""Check consistency of all history of given revision_ids.
2708
Different repository implementations should override _check().
2710
:param revision_ids: A non-empty list of revision_ids whose ancestry
2711
will be checked. Typically the last revision_id of a branch.
2712
:param callback_refs: A dict of check-refs to resolve and callback
2713
the check/_check method on the items listed as wanting the ref.
2715
:param check_repo: If False do not check the repository contents, just
2716
calculate the data callback_refs requires and call them back.
2718
return self._check(revision_ids, callback_refs=callback_refs,
2719
check_repo=check_repo)
2721
def _check(self, revision_ids, callback_refs, check_repo):
2722
result = check.Check(self, check_repo=check_repo)
2723
result.check(callback_refs)
2726
def _warn_if_deprecated(self):
2727
global _deprecation_warning_done
2728
if _deprecation_warning_done:
2730
_deprecation_warning_done = True
2731
warning("Format %s for %s is deprecated - please use 'bzr upgrade' to get better performance"
2732
% (self._format, self.bzrdir.transport.base))
2734
def supports_rich_root(self):
2735
return self._format.rich_root_data
2737
def _check_ascii_revisionid(self, revision_id, method):
2738
"""Private helper for ascii-only repositories."""
2739
# weave repositories refuse to store revisionids that are non-ascii.
2740
if revision_id is not None:
2741
# weaves require ascii revision ids.
2742
if isinstance(revision_id, unicode):
2744
revision_id.encode('ascii')
2745
except UnicodeEncodeError:
2746
raise errors.NonAsciiRevisionId(method, self)
2749
revision_id.decode('ascii')
2750
except UnicodeDecodeError:
2751
raise errors.NonAsciiRevisionId(method, self)
2753
def revision_graph_can_have_wrong_parents(self):
2754
"""Is it possible for this repository to have a revision graph with
2757
If True, then this repository must also implement
2758
_find_inconsistent_revision_parents so that check and reconcile can
2759
check for inconsistencies before proceeding with other checks that may
2760
depend on the revision index being consistent.
2762
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2765
# remove these delegates a while after bzr 0.15
2766
def __make_delegated(name, from_module):
2767
def _deprecated_repository_forwarder():
2768
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2769
% (name, from_module),
2772
m = __import__(from_module, globals(), locals(), [name])
2774
return getattr(m, name)
2775
except AttributeError:
2776
raise AttributeError('module %s has no name %s'
2778
globals()[name] = _deprecated_repository_forwarder
2781
'AllInOneRepository',
2782
'WeaveMetaDirRepository',
2783
'PreSplitOutRepositoryFormat',
2784
'RepositoryFormat4',
2785
'RepositoryFormat5',
2786
'RepositoryFormat6',
2787
'RepositoryFormat7',
2789
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2793
'RepositoryFormatKnit',
2794
'RepositoryFormatKnit1',
2796
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2799
def install_revision(repository, rev, revision_tree):
2800
"""Install all revision data into a repository."""
2801
install_revisions(repository, [(rev, revision_tree, None)])
2804
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2805
"""Install all revision data into a repository.
2807
Accepts an iterable of revision, tree, signature tuples. The signature
2810
repository.start_write_group()
2812
inventory_cache = lru_cache.LRUCache(10)
2813
for n, (revision, revision_tree, signature) in enumerate(iterable):
2814
_install_revision(repository, revision, revision_tree, signature,
2817
pb.update('Transferring revisions', n + 1, num_revisions)
2819
repository.abort_write_group()
2822
repository.commit_write_group()
2825
def _install_revision(repository, rev, revision_tree, signature,
2827
"""Install all revision data into a repository."""
2828
present_parents = []
2830
for p_id in rev.parent_ids:
2831
if repository.has_revision(p_id):
2832
present_parents.append(p_id)
2833
parent_trees[p_id] = repository.revision_tree(p_id)
2835
parent_trees[p_id] = repository.revision_tree(
2836
_mod_revision.NULL_REVISION)
2838
inv = revision_tree.inventory
2839
entries = inv.iter_entries()
2840
# backwards compatibility hack: skip the root id.
2841
if not repository.supports_rich_root():
2842
path, root = entries.next()
2843
if root.revision != rev.revision_id:
2844
raise errors.IncompatibleRevision(repr(repository))
2846
for path, ie in entries:
2847
text_keys[(ie.file_id, ie.revision)] = ie
2848
text_parent_map = repository.texts.get_parent_map(text_keys)
2849
missing_texts = set(text_keys) - set(text_parent_map)
2850
# Add the texts that are not already present
2851
for text_key in missing_texts:
2852
ie = text_keys[text_key]
2854
# FIXME: TODO: The following loop overlaps/duplicates that done by
2855
# commit to determine parents. There is a latent/real bug here where
2856
# the parents inserted are not those commit would do - in particular
2857
# they are not filtered by heads(). RBC, AB
2858
for revision, tree in parent_trees.iteritems():
2859
if ie.file_id not in tree:
2861
parent_id = tree.inventory[ie.file_id].revision
2862
if parent_id in text_parents:
2864
text_parents.append((ie.file_id, parent_id))
2865
lines = revision_tree.get_file(ie.file_id).readlines()
2866
repository.texts.add_lines(text_key, text_parents, lines)
2868
# install the inventory
2869
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2870
# Cache this inventory
2871
inventory_cache[rev.revision_id] = inv
2873
basis_inv = inventory_cache[rev.parent_ids[0]]
2875
repository.add_inventory(rev.revision_id, inv, present_parents)
2877
delta = inv._make_delta(basis_inv)
2878
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2879
rev.revision_id, present_parents)
2881
repository.add_inventory(rev.revision_id, inv, present_parents)
2882
except errors.RevisionAlreadyPresent:
2884
if signature is not None:
2885
repository.add_signature_text(rev.revision_id, signature)
2886
repository.add_revision(rev.revision_id, rev, inv)
2889
class MetaDirRepository(Repository):
2890
"""Repositories in the new meta-dir layout.
2892
:ivar _transport: Transport for access to repository control files,
2893
typically pointing to .bzr/repository.
2896
def __init__(self, _format, a_bzrdir, control_files):
2897
super(MetaDirRepository, self).__init__(_format, a_bzrdir, control_files)
2898
self._transport = control_files._transport
2900
def is_shared(self):
2901
"""Return True if this repository is flagged as a shared repository."""
2902
return self._transport.has('shared-storage')
2905
def set_make_working_trees(self, new_value):
2906
"""Set the policy flag for making working trees when creating branches.
2908
This only applies to branches that use this repository.
2910
The default is 'True'.
2911
:param new_value: True to restore the default, False to disable making
2916
self._transport.delete('no-working-trees')
2917
except errors.NoSuchFile:
2920
self._transport.put_bytes('no-working-trees', '',
2921
mode=self.bzrdir._get_file_mode())
2923
def make_working_trees(self):
2924
"""Returns the policy for making working trees on new branches."""
2925
return not self._transport.has('no-working-trees')
2928
class MetaDirVersionedFileRepository(MetaDirRepository):
2929
"""Repositories in a meta-dir, that work via versioned file objects."""
2931
def __init__(self, _format, a_bzrdir, control_files):
2932
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2936
network_format_registry = registry.FormatRegistry()
2937
"""Registry of formats indexed by their network name.
2939
The network name for a repository format is an identifier that can be used when
2940
referring to formats with smart server operations. See
2941
RepositoryFormat.network_name() for more detail.
2945
format_registry = registry.FormatRegistry(network_format_registry)
2946
"""Registry of formats, indexed by their BzrDirMetaFormat format string.
2948
This can contain either format instances themselves, or classes/factories that
2949
can be called to obtain one.
2953
#####################################################################
2954
# Repository Formats
2956
class RepositoryFormat(object):
2957
"""A repository format.
2959
Formats provide four things:
2960
* An initialization routine to construct repository data on disk.
2961
* a optional format string which is used when the BzrDir supports
2963
* an open routine which returns a Repository instance.
2964
* A network name for referring to the format in smart server RPC
2967
There is one and only one Format subclass for each on-disk format. But
2968
there can be one Repository subclass that is used for several different
2969
formats. The _format attribute on a Repository instance can be used to
2970
determine the disk format.
2972
Formats are placed in a registry by their format string for reference
2973
during opening. These should be subclasses of RepositoryFormat for
2976
Once a format is deprecated, just deprecate the initialize and open
2977
methods on the format class. Do not deprecate the object, as the
2978
object may be created even when a repository instance hasn't been
2981
Common instance attributes:
2982
_matchingbzrdir - the bzrdir format that the repository format was
2983
originally written to work with. This can be used if manually
2984
constructing a bzrdir and repository, or more commonly for test suite
2988
# Set to True or False in derived classes. True indicates that the format
2989
# supports ghosts gracefully.
2990
supports_ghosts = None
2991
# Can this repository be given external locations to lookup additional
2992
# data. Set to True or False in derived classes.
2993
supports_external_lookups = None
2994
# Does this format support CHK bytestring lookups. Set to True or False in
2996
supports_chks = None
2997
# Should commit add an inventory, or an inventory delta to the repository.
2998
_commit_inv_deltas = True
2999
# What order should fetch operations request streams in?
3000
# The default is unordered as that is the cheapest for an origin to
3002
_fetch_order = 'unordered'
3003
# Does this repository format use deltas that can be fetched as-deltas ?
3004
# (E.g. knits, where the knit deltas can be transplanted intact.
3005
# We default to False, which will ensure that enough data to get
3006
# a full text out of any fetch stream will be grabbed.
3007
_fetch_uses_deltas = False
3008
# Should fetch trigger a reconcile after the fetch? Only needed for
3009
# some repository formats that can suffer internal inconsistencies.
3010
_fetch_reconcile = False
3011
# Does this format have < O(tree_size) delta generation. Used to hint what
3012
# code path for commit, amongst other things.
3014
# Does doing a pack operation compress data? Useful for the pack UI command
3015
# (so if there is one pack, the operation can still proceed because it may
3016
# help), and for fetching when data won't have come from the same
3018
pack_compresses = False
3021
return "<%s>" % self.__class__.__name__
3023
def __eq__(self, other):
3024
# format objects are generally stateless
3025
return isinstance(other, self.__class__)
3027
def __ne__(self, other):
3028
return not self == other
3031
def find_format(klass, a_bzrdir):
3032
"""Return the format for the repository object in a_bzrdir.
3034
This is used by bzr native formats that have a "format" file in
3035
the repository. Other methods may be used by different types of
3039
transport = a_bzrdir.get_repository_transport(None)
3040
format_string = transport.get("format").read()
3041
return format_registry.get(format_string)
3042
except errors.NoSuchFile:
3043
raise errors.NoRepositoryPresent(a_bzrdir)
3045
raise errors.UnknownFormatError(format=format_string,
3049
def register_format(klass, format):
3050
format_registry.register(format.get_format_string(), format)
3053
def unregister_format(klass, format):
3054
format_registry.remove(format.get_format_string())
3057
def get_default_format(klass):
3058
"""Return the current default format."""
3059
from bzrlib import bzrdir
3060
return bzrdir.format_registry.make_bzrdir('default').repository_format
3062
def get_format_string(self):
3063
"""Return the ASCII format string that identifies this format.
3065
Note that in pre format ?? repositories the format string is
3066
not permitted nor written to disk.
3068
raise NotImplementedError(self.get_format_string)
3070
def get_format_description(self):
3071
"""Return the short description for this format."""
3072
raise NotImplementedError(self.get_format_description)
3074
# TODO: this shouldn't be in the base class, it's specific to things that
3075
# use weaves or knits -- mbp 20070207
3076
def _get_versioned_file_store(self,
3081
versionedfile_class=None,
3082
versionedfile_kwargs={},
3084
if versionedfile_class is None:
3085
versionedfile_class = self._versionedfile_class
3086
weave_transport = control_files._transport.clone(name)
3087
dir_mode = control_files._dir_mode
3088
file_mode = control_files._file_mode
3089
return VersionedFileStore(weave_transport, prefixed=prefixed,
3091
file_mode=file_mode,
3092
versionedfile_class=versionedfile_class,
3093
versionedfile_kwargs=versionedfile_kwargs,
3096
def initialize(self, a_bzrdir, shared=False):
3097
"""Initialize a repository of this format in a_bzrdir.
3099
:param a_bzrdir: The bzrdir to put the new repository in it.
3100
:param shared: The repository should be initialized as a sharable one.
3101
:returns: The new repository object.
3103
This may raise UninitializableFormat if shared repository are not
3104
compatible the a_bzrdir.
3106
raise NotImplementedError(self.initialize)
3108
def is_supported(self):
3109
"""Is this format supported?
3111
Supported formats must be initializable and openable.
3112
Unsupported formats may not support initialization or committing or
3113
some other features depending on the reason for not being supported.
3117
def network_name(self):
3118
"""A simple byte string uniquely identifying this format for RPC calls.
3120
MetaDir repository formats use their disk format string to identify the
3121
repository over the wire. All in one formats such as bzr < 0.8, and
3122
foreign formats like svn/git and hg should use some marker which is
3123
unique and immutable.
3125
raise NotImplementedError(self.network_name)
3127
def check_conversion_target(self, target_format):
3128
raise NotImplementedError(self.check_conversion_target)
3130
def open(self, a_bzrdir, _found=False):
3131
"""Return an instance of this format for the bzrdir a_bzrdir.
3133
_found is a private parameter, do not use it.
3135
raise NotImplementedError(self.open)
3138
class MetaDirRepositoryFormat(RepositoryFormat):
3139
"""Common base class for the new repositories using the metadir layout."""
3141
rich_root_data = False
3142
supports_tree_reference = False
3143
supports_external_lookups = False
3146
def _matchingbzrdir(self):
3147
matching = bzrdir.BzrDirMetaFormat1()
3148
matching.repository_format = self
3152
super(MetaDirRepositoryFormat, self).__init__()
3154
def _create_control_files(self, a_bzrdir):
3155
"""Create the required files and the initial control_files object."""
3156
# FIXME: RBC 20060125 don't peek under the covers
3157
# NB: no need to escape relative paths that are url safe.
3158
repository_transport = a_bzrdir.get_repository_transport(self)
3159
control_files = lockable_files.LockableFiles(repository_transport,
3160
'lock', lockdir.LockDir)
3161
control_files.create_lock()
3162
return control_files
3164
def _upload_blank_content(self, a_bzrdir, dirs, files, utf8_files, shared):
3165
"""Upload the initial blank content."""
3166
control_files = self._create_control_files(a_bzrdir)
3167
control_files.lock_write()
3168
transport = control_files._transport
3170
utf8_files += [('shared-storage', '')]
3172
transport.mkdir_multi(dirs, mode=a_bzrdir._get_dir_mode())
3173
for (filename, content_stream) in files:
3174
transport.put_file(filename, content_stream,
3175
mode=a_bzrdir._get_file_mode())
3176
for (filename, content_bytes) in utf8_files:
3177
transport.put_bytes_non_atomic(filename, content_bytes,
3178
mode=a_bzrdir._get_file_mode())
3180
control_files.unlock()
3182
def network_name(self):
3183
"""Metadir formats have matching disk and network format strings."""
3184
return self.get_format_string()
3187
# Pre-0.8 formats that don't have a disk format string (because they are
3188
# versioned by the matching control directory). We use the control directories
3189
# disk format string as a key for the network_name because they meet the
3190
# constraints (simple string, unique, immutable).
3191
network_format_registry.register_lazy(
3192
"Bazaar-NG branch, format 5\n",
3193
'bzrlib.repofmt.weaverepo',
3194
'RepositoryFormat5',
3196
network_format_registry.register_lazy(
3197
"Bazaar-NG branch, format 6\n",
3198
'bzrlib.repofmt.weaverepo',
3199
'RepositoryFormat6',
3202
# formats which have no format string are not discoverable or independently
3203
# creatable on disk, so are not registered in format_registry. They're
3204
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
3205
# needed, it's constructed directly by the BzrDir. Non-native formats where
3206
# the repository is not separately opened are similar.
3208
format_registry.register_lazy(
3209
'Bazaar-NG Repository format 7',
3210
'bzrlib.repofmt.weaverepo',
3214
format_registry.register_lazy(
3215
'Bazaar-NG Knit Repository Format 1',
3216
'bzrlib.repofmt.knitrepo',
3217
'RepositoryFormatKnit1',
3220
format_registry.register_lazy(
3221
'Bazaar Knit Repository Format 3 (bzr 0.15)\n',
3222
'bzrlib.repofmt.knitrepo',
3223
'RepositoryFormatKnit3',
3226
format_registry.register_lazy(
3227
'Bazaar Knit Repository Format 4 (bzr 1.0)\n',
3228
'bzrlib.repofmt.knitrepo',
3229
'RepositoryFormatKnit4',
3232
# Pack-based formats. There is one format for pre-subtrees, and one for
3233
# post-subtrees to allow ease of testing.
3234
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
3235
format_registry.register_lazy(
3236
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
3237
'bzrlib.repofmt.pack_repo',
3238
'RepositoryFormatKnitPack1',
3240
format_registry.register_lazy(
3241
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
3242
'bzrlib.repofmt.pack_repo',
3243
'RepositoryFormatKnitPack3',
3245
format_registry.register_lazy(
3246
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
3247
'bzrlib.repofmt.pack_repo',
3248
'RepositoryFormatKnitPack4',
3250
format_registry.register_lazy(
3251
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
3252
'bzrlib.repofmt.pack_repo',
3253
'RepositoryFormatKnitPack5',
3255
format_registry.register_lazy(
3256
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
3257
'bzrlib.repofmt.pack_repo',
3258
'RepositoryFormatKnitPack5RichRoot',
3260
format_registry.register_lazy(
3261
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3262
'bzrlib.repofmt.pack_repo',
3263
'RepositoryFormatKnitPack5RichRootBroken',
3265
format_registry.register_lazy(
3266
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3267
'bzrlib.repofmt.pack_repo',
3268
'RepositoryFormatKnitPack6',
3270
format_registry.register_lazy(
3271
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3272
'bzrlib.repofmt.pack_repo',
3273
'RepositoryFormatKnitPack6RichRoot',
3276
# Development formats.
3277
# Obsolete but kept pending a CHK based subtree format.
3278
format_registry.register_lazy(
3279
("Bazaar development format 2 with subtree support "
3280
"(needs bzr.dev from before 1.8)\n"),
3281
'bzrlib.repofmt.pack_repo',
3282
'RepositoryFormatPackDevelopment2Subtree',
3285
# 1.14->1.16 go below here
3286
format_registry.register_lazy(
3287
'Bazaar development format - group compression and chk inventory'
3288
' (needs bzr.dev from 1.14)\n',
3289
'bzrlib.repofmt.groupcompress_repo',
3290
'RepositoryFormatCHK1',
3293
format_registry.register_lazy(
3294
'Bazaar development format - chk repository with bencode revision '
3295
'serialization (needs bzr.dev from 1.16)\n',
3296
'bzrlib.repofmt.groupcompress_repo',
3297
'RepositoryFormatCHK2',
3299
format_registry.register_lazy(
3300
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3301
'bzrlib.repofmt.groupcompress_repo',
3302
'RepositoryFormat2a',
3306
class InterRepository(InterObject):
3307
"""This class represents operations taking place between two repositories.
3309
Its instances have methods like copy_content and fetch, and contain
3310
references to the source and target repositories these operations can be
3313
Often we will provide convenience methods on 'repository' which carry out
3314
operations with another repository - they will always forward to
3315
InterRepository.get(other).method_name(parameters).
3318
_walk_to_common_revisions_batch_size = 50
3320
"""The available optimised InterRepository types."""
3323
def copy_content(self, revision_id=None):
3324
"""Make a complete copy of the content in self into destination.
3326
This is a destructive operation! Do not use it on existing
3329
:param revision_id: Only copy the content needed to construct
3330
revision_id and its parents.
3333
self.target.set_make_working_trees(self.source.make_working_trees())
3334
except NotImplementedError:
3336
self.target.fetch(self.source, revision_id=revision_id)
3339
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3341
"""Fetch the content required to construct revision_id.
3343
The content is copied from self.source to self.target.
3345
:param revision_id: if None all content is copied, if NULL_REVISION no
3347
:param pb: optional progress bar to use for progress reports. If not
3348
provided a default one will be created.
3351
from bzrlib.fetch import RepoFetcher
3352
f = RepoFetcher(to_repository=self.target,
3353
from_repository=self.source,
3354
last_revision=revision_id,
3355
fetch_spec=fetch_spec,
3356
pb=pb, find_ghosts=find_ghosts)
3358
def _walk_to_common_revisions(self, revision_ids):
3359
"""Walk out from revision_ids in source to revisions target has.
3361
:param revision_ids: The start point for the search.
3362
:return: A set of revision ids.
3364
target_graph = self.target.get_graph()
3365
revision_ids = frozenset(revision_ids)
3366
missing_revs = set()
3367
source_graph = self.source.get_graph()
3368
# ensure we don't pay silly lookup costs.
3369
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3370
null_set = frozenset([_mod_revision.NULL_REVISION])
3371
searcher_exhausted = False
3375
# Iterate the searcher until we have enough next_revs
3376
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3378
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3379
next_revs.update(next_revs_part)
3380
ghosts.update(ghosts_part)
3381
except StopIteration:
3382
searcher_exhausted = True
3384
# If there are ghosts in the source graph, and the caller asked for
3385
# them, make sure that they are present in the target.
3386
# We don't care about other ghosts as we can't fetch them and
3387
# haven't been asked to.
3388
ghosts_to_check = set(revision_ids.intersection(ghosts))
3389
revs_to_get = set(next_revs).union(ghosts_to_check)
3391
have_revs = set(target_graph.get_parent_map(revs_to_get))
3392
# we always have NULL_REVISION present.
3393
have_revs = have_revs.union(null_set)
3394
# Check if the target is missing any ghosts we need.
3395
ghosts_to_check.difference_update(have_revs)
3397
# One of the caller's revision_ids is a ghost in both the
3398
# source and the target.
3399
raise errors.NoSuchRevision(
3400
self.source, ghosts_to_check.pop())
3401
missing_revs.update(next_revs - have_revs)
3402
# Because we may have walked past the original stop point, make
3403
# sure everything is stopped
3404
stop_revs = searcher.find_seen_ancestors(have_revs)
3405
searcher.stop_searching_any(stop_revs)
3406
if searcher_exhausted:
3408
return searcher.get_result()
3411
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3412
"""Return the revision ids that source has that target does not.
3414
:param revision_id: only return revision ids included by this
3416
:param find_ghosts: If True find missing revisions in deep history
3417
rather than just finding the surface difference.
3418
:return: A bzrlib.graph.SearchResult.
3420
# stop searching at found target revisions.
3421
if not find_ghosts and revision_id is not None:
3422
return self._walk_to_common_revisions([revision_id])
3423
# generic, possibly worst case, slow code path.
3424
target_ids = set(self.target.all_revision_ids())
3425
if revision_id is not None:
3426
source_ids = self.source.get_ancestry(revision_id)
3427
if source_ids[0] is not None:
3428
raise AssertionError()
3431
source_ids = self.source.all_revision_ids()
3432
result_set = set(source_ids).difference(target_ids)
3433
return self.source.revision_ids_to_search_result(result_set)
3436
def _same_model(source, target):
3437
"""True if source and target have the same data representation.
3439
Note: this is always called on the base class; overriding it in a
3440
subclass will have no effect.
3443
InterRepository._assert_same_model(source, target)
3445
except errors.IncompatibleRepositories, e:
3449
def _assert_same_model(source, target):
3450
"""Raise an exception if two repositories do not use the same model.
3452
if source.supports_rich_root() != target.supports_rich_root():
3453
raise errors.IncompatibleRepositories(source, target,
3454
"different rich-root support")
3455
if source._serializer != target._serializer:
3456
raise errors.IncompatibleRepositories(source, target,
3457
"different serializers")
3460
class InterSameDataRepository(InterRepository):
3461
"""Code for converting between repositories that represent the same data.
3463
Data format and model must match for this to work.
3467
def _get_repo_format_to_test(self):
3468
"""Repository format for testing with.
3470
InterSameData can pull from subtree to subtree and from non-subtree to
3471
non-subtree, so we test this with the richest repository format.
3473
from bzrlib.repofmt import knitrepo
3474
return knitrepo.RepositoryFormatKnit3()
3477
def is_compatible(source, target):
3478
return InterRepository._same_model(source, target)
3481
class InterWeaveRepo(InterSameDataRepository):
3482
"""Optimised code paths between Weave based repositories.
3484
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3485
implemented lazy inter-object optimisation.
3489
def _get_repo_format_to_test(self):
3490
from bzrlib.repofmt import weaverepo
3491
return weaverepo.RepositoryFormat7()
3494
def is_compatible(source, target):
3495
"""Be compatible with known Weave formats.
3497
We don't test for the stores being of specific types because that
3498
could lead to confusing results, and there is no need to be
3501
from bzrlib.repofmt.weaverepo import (
3507
return (isinstance(source._format, (RepositoryFormat5,
3509
RepositoryFormat7)) and
3510
isinstance(target._format, (RepositoryFormat5,
3512
RepositoryFormat7)))
3513
except AttributeError:
3517
def copy_content(self, revision_id=None):
3518
"""See InterRepository.copy_content()."""
3519
# weave specific optimised path:
3521
self.target.set_make_working_trees(self.source.make_working_trees())
3522
except (errors.RepositoryUpgradeRequired, NotImplemented):
3524
# FIXME do not peek!
3525
if self.source._transport.listable():
3526
pb = ui.ui_factory.nested_progress_bar()
3528
self.target.texts.insert_record_stream(
3529
self.source.texts.get_record_stream(
3530
self.source.texts.keys(), 'topological', False))
3531
pb.update('copying inventory', 0, 1)
3532
self.target.inventories.insert_record_stream(
3533
self.source.inventories.get_record_stream(
3534
self.source.inventories.keys(), 'topological', False))
3535
self.target.signatures.insert_record_stream(
3536
self.source.signatures.get_record_stream(
3537
self.source.signatures.keys(),
3539
self.target.revisions.insert_record_stream(
3540
self.source.revisions.get_record_stream(
3541
self.source.revisions.keys(),
3542
'topological', True))
3546
self.target.fetch(self.source, revision_id=revision_id)
3549
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3550
"""See InterRepository.missing_revision_ids()."""
3551
# we want all revisions to satisfy revision_id in source.
3552
# but we don't want to stat every file here and there.
3553
# we want then, all revisions other needs to satisfy revision_id
3554
# checked, but not those that we have locally.
3555
# so the first thing is to get a subset of the revisions to
3556
# satisfy revision_id in source, and then eliminate those that
3557
# we do already have.
3558
# this is slow on high latency connection to self, but as this
3559
# disk format scales terribly for push anyway due to rewriting
3560
# inventory.weave, this is considered acceptable.
3562
if revision_id is not None:
3563
source_ids = self.source.get_ancestry(revision_id)
3564
if source_ids[0] is not None:
3565
raise AssertionError()
3568
source_ids = self.source._all_possible_ids()
3569
source_ids_set = set(source_ids)
3570
# source_ids is the worst possible case we may need to pull.
3571
# now we want to filter source_ids against what we actually
3572
# have in target, but don't try to check for existence where we know
3573
# we do not have a revision as that would be pointless.
3574
target_ids = set(self.target._all_possible_ids())
3575
possibly_present_revisions = target_ids.intersection(source_ids_set)
3576
actually_present_revisions = set(
3577
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3578
required_revisions = source_ids_set.difference(actually_present_revisions)
3579
if revision_id is not None:
3580
# we used get_ancestry to determine source_ids then we are assured all
3581
# revisions referenced are present as they are installed in topological order.
3582
# and the tip revision was validated by get_ancestry.
3583
result_set = required_revisions
3585
# if we just grabbed the possibly available ids, then
3586
# we only have an estimate of whats available and need to validate
3587
# that against the revision records.
3589
self.source._eliminate_revisions_not_present(required_revisions))
3590
return self.source.revision_ids_to_search_result(result_set)
3593
class InterKnitRepo(InterSameDataRepository):
3594
"""Optimised code paths between Knit based repositories."""
3597
def _get_repo_format_to_test(self):
3598
from bzrlib.repofmt import knitrepo
3599
return knitrepo.RepositoryFormatKnit1()
3602
def is_compatible(source, target):
3603
"""Be compatible with known Knit formats.
3605
We don't test for the stores being of specific types because that
3606
could lead to confusing results, and there is no need to be
3609
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3611
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3612
isinstance(target._format, RepositoryFormatKnit))
3613
except AttributeError:
3615
return are_knits and InterRepository._same_model(source, target)
3618
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3619
"""See InterRepository.missing_revision_ids()."""
3620
if revision_id is not None:
3621
source_ids = self.source.get_ancestry(revision_id)
3622
if source_ids[0] is not None:
3623
raise AssertionError()
3626
source_ids = self.source.all_revision_ids()
3627
source_ids_set = set(source_ids)
3628
# source_ids is the worst possible case we may need to pull.
3629
# now we want to filter source_ids against what we actually
3630
# have in target, but don't try to check for existence where we know
3631
# we do not have a revision as that would be pointless.
3632
target_ids = set(self.target.all_revision_ids())
3633
possibly_present_revisions = target_ids.intersection(source_ids_set)
3634
actually_present_revisions = set(
3635
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3636
required_revisions = source_ids_set.difference(actually_present_revisions)
3637
if revision_id is not None:
3638
# we used get_ancestry to determine source_ids then we are assured all
3639
# revisions referenced are present as they are installed in topological order.
3640
# and the tip revision was validated by get_ancestry.
3641
result_set = required_revisions
3643
# if we just grabbed the possibly available ids, then
3644
# we only have an estimate of whats available and need to validate
3645
# that against the revision records.
3647
self.source._eliminate_revisions_not_present(required_revisions))
3648
return self.source.revision_ids_to_search_result(result_set)
3651
class InterDifferingSerializer(InterRepository):
3654
def _get_repo_format_to_test(self):
3658
def is_compatible(source, target):
3659
"""Be compatible with Knit2 source and Knit3 target"""
3660
# This is redundant with format.check_conversion_target(), however that
3661
# raises an exception, and we just want to say "False" as in we won't
3662
# support converting between these formats.
3663
if source.supports_rich_root() and not target.supports_rich_root():
3665
if (source._format.supports_tree_reference
3666
and not target._format.supports_tree_reference):
3670
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3671
"""Get the best delta and base for this revision.
3673
:return: (basis_id, delta)
3675
possible_trees = [(parent_id, cache[parent_id])
3676
for parent_id in parent_ids
3677
if parent_id in cache]
3678
if len(possible_trees) == 0:
3679
# There either aren't any parents, or the parents aren't in the
3680
# cache, so just use the last converted tree
3681
possible_trees.append((basis_id, cache[basis_id]))
3683
for basis_id, basis_tree in possible_trees:
3684
delta = tree.inventory._make_delta(basis_tree.inventory)
3685
deltas.append((len(delta), basis_id, delta))
3687
return deltas[0][1:]
3689
def _get_parent_keys(self, root_key, parent_map):
3690
"""Get the parent keys for a given root id."""
3691
root_id, rev_id = root_key
3692
# Include direct parents of the revision, but only if they used
3693
# the same root_id and are heads.
3695
for parent_id in parent_map[rev_id]:
3696
if parent_id == _mod_revision.NULL_REVISION:
3698
if parent_id not in self._revision_id_to_root_id:
3699
# We probably didn't read this revision, go spend the
3700
# extra effort to actually check
3702
tree = self.source.revision_tree(parent_id)
3703
except errors.NoSuchRevision:
3704
# Ghost, fill out _revision_id_to_root_id in case we
3705
# encounter this again.
3706
# But set parent_root_id to None since we don't really know
3707
parent_root_id = None
3709
parent_root_id = tree.get_root_id()
3710
self._revision_id_to_root_id[parent_id] = None
3712
parent_root_id = self._revision_id_to_root_id[parent_id]
3713
if root_id == parent_root_id:
3714
# With stacking we _might_ want to refer to a non-local
3715
# revision, but this code path only applies when we have the
3716
# full content available, so ghosts really are ghosts, not just
3717
# the edge of local data.
3718
parent_keys.append((parent_id,))
3720
# root_id may be in the parent anyway.
3722
tree = self.source.revision_tree(parent_id)
3723
except errors.NoSuchRevision:
3724
# ghost, can't refer to it.
3728
parent_keys.append((tree.inventory[root_id].revision,))
3729
except errors.NoSuchId:
3732
g = graph.Graph(self.source.revisions)
3733
heads = g.heads(parent_keys)
3735
for key in parent_keys:
3736
if key in heads and key not in selected_keys:
3737
selected_keys.append(key)
3738
return tuple([(root_id,)+ key for key in selected_keys])
3740
def _new_root_data_stream(self, root_keys_to_create, parent_map):
3741
for root_key in root_keys_to_create:
3742
parent_keys = self._get_parent_keys(root_key, parent_map)
3743
yield versionedfile.FulltextContentFactory(root_key,
3744
parent_keys, None, '')
3746
def _fetch_batch(self, revision_ids, basis_id, cache):
3747
"""Fetch across a few revisions.
3749
:param revision_ids: The revisions to copy
3750
:param basis_id: The revision_id of a tree that must be in cache, used
3751
as a basis for delta when no other base is available
3752
:param cache: A cache of RevisionTrees that we can use.
3753
:return: The revision_id of the last converted tree. The RevisionTree
3754
for it will be in cache
3756
# Walk though all revisions; get inventory deltas, copy referenced
3757
# texts that delta references, insert the delta, revision and
3759
root_keys_to_create = set()
3762
pending_revisions = []
3763
parent_map = self.source.get_parent_map(revision_ids)
3764
for tree in self.source.revision_trees(revision_ids):
3765
current_revision_id = tree.get_revision_id()
3766
parent_ids = parent_map.get(current_revision_id, ())
3767
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3769
if self._converting_to_rich_root:
3770
self._revision_id_to_root_id[current_revision_id] = \
3772
# Find text entries that need to be copied
3773
for old_path, new_path, file_id, entry in delta:
3774
if new_path is not None:
3777
if not self.target.supports_rich_root():
3778
# The target doesn't support rich root, so we don't
3781
if self._converting_to_rich_root:
3782
# This can't be copied normally, we have to insert
3784
root_keys_to_create.add((file_id, entry.revision))
3786
text_keys.add((file_id, entry.revision))
3787
revision = self.source.get_revision(current_revision_id)
3788
pending_deltas.append((basis_id, delta,
3789
current_revision_id, revision.parent_ids))
3790
pending_revisions.append(revision)
3791
cache[current_revision_id] = tree
3792
basis_id = current_revision_id
3794
from_texts = self.source.texts
3795
to_texts = self.target.texts
3796
if root_keys_to_create:
3797
root_stream = self._new_root_data_stream(root_keys_to_create,
3799
to_texts.insert_record_stream(root_stream)
3800
to_texts.insert_record_stream(from_texts.get_record_stream(
3801
text_keys, self.target._format._fetch_order,
3802
not self.target._format._fetch_uses_deltas))
3803
# insert inventory deltas
3804
for delta in pending_deltas:
3805
self.target.add_inventory_by_delta(*delta)
3806
if self.target._fallback_repositories:
3807
# Make sure this stacked repository has all the parent inventories
3808
# for the new revisions that we are about to insert. We do this
3809
# before adding the revisions so that no revision is added until
3810
# all the inventories it may depend on are added.
3812
revision_ids = set()
3813
for revision in pending_revisions:
3814
revision_ids.add(revision.revision_id)
3815
parent_ids.update(revision.parent_ids)
3816
parent_ids.difference_update(revision_ids)
3817
parent_ids.discard(_mod_revision.NULL_REVISION)
3818
parent_map = self.source.get_parent_map(parent_ids)
3819
for parent_tree in self.source.revision_trees(parent_ids):
3820
basis_id, delta = self._get_delta_for_revision(tree, parent_ids, basis_id, cache)
3821
current_revision_id = parent_tree.get_revision_id()
3822
parents_parents = parent_map[current_revision_id]
3823
self.target.add_inventory_by_delta(
3824
basis_id, delta, current_revision_id, parents_parents)
3825
# insert signatures and revisions
3826
for revision in pending_revisions:
3828
signature = self.source.get_signature_text(
3829
revision.revision_id)
3830
self.target.add_signature_text(revision.revision_id,
3832
except errors.NoSuchRevision:
3834
self.target.add_revision(revision.revision_id, revision)
3837
def _fetch_all_revisions(self, revision_ids, pb):
3838
"""Fetch everything for the list of revisions.
3840
:param revision_ids: The list of revisions to fetch. Must be in
3842
:param pb: A ProgressTask
3845
basis_id, basis_tree = self._get_basis(revision_ids[0])
3847
cache = lru_cache.LRUCache(100)
3848
cache[basis_id] = basis_tree
3849
del basis_tree # We don't want to hang on to it here
3851
for offset in range(0, len(revision_ids), batch_size):
3852
self.target.start_write_group()
3854
pb.update('Transferring revisions', offset,
3856
batch = revision_ids[offset:offset+batch_size]
3857
basis_id = self._fetch_batch(batch, basis_id, cache)
3859
self.target.abort_write_group()
3862
hint = self.target.commit_write_group()
3865
if hints and self.target._format.pack_compresses:
3866
self.target.pack(hint=hints)
3867
pb.update('Transferring revisions', len(revision_ids),
3871
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3873
"""See InterRepository.fetch()."""
3874
if fetch_spec is not None:
3875
raise AssertionError("Not implemented yet...")
3876
if (not self.source.supports_rich_root()
3877
and self.target.supports_rich_root()):
3878
self._converting_to_rich_root = True
3879
self._revision_id_to_root_id = {}
3881
self._converting_to_rich_root = False
3882
revision_ids = self.target.search_missing_revision_ids(self.source,
3883
revision_id, find_ghosts=find_ghosts).get_keys()
3884
if not revision_ids:
3886
revision_ids = tsort.topo_sort(
3887
self.source.get_graph().get_parent_map(revision_ids))
3888
if not revision_ids:
3890
# Walk though all revisions; get inventory deltas, copy referenced
3891
# texts that delta references, insert the delta, revision and
3893
first_rev = self.source.get_revision(revision_ids[0])
3895
my_pb = ui.ui_factory.nested_progress_bar()
3898
symbol_versioning.warn(
3899
symbol_versioning.deprecated_in((1, 14, 0))
3900
% "pb parameter to fetch()")
3903
self._fetch_all_revisions(revision_ids, pb)
3905
if my_pb is not None:
3907
return len(revision_ids), 0
3909
def _get_basis(self, first_revision_id):
3910
"""Get a revision and tree which exists in the target.
3912
This assumes that first_revision_id is selected for transmission
3913
because all other ancestors are already present. If we can't find an
3914
ancestor we fall back to NULL_REVISION since we know that is safe.
3916
:return: (basis_id, basis_tree)
3918
first_rev = self.source.get_revision(first_revision_id)
3920
basis_id = first_rev.parent_ids[0]
3921
# only valid as a basis if the target has it
3922
self.target.get_revision(basis_id)
3923
# Try to get a basis tree - if its a ghost it will hit the
3924
# NoSuchRevision case.
3925
basis_tree = self.source.revision_tree(basis_id)
3926
except (IndexError, errors.NoSuchRevision):
3927
basis_id = _mod_revision.NULL_REVISION
3928
basis_tree = self.source.revision_tree(basis_id)
3929
return basis_id, basis_tree
3932
InterRepository.register_optimiser(InterDifferingSerializer)
3933
InterRepository.register_optimiser(InterSameDataRepository)
3934
InterRepository.register_optimiser(InterWeaveRepo)
3935
InterRepository.register_optimiser(InterKnitRepo)
3938
class CopyConverter(object):
3939
"""A repository conversion tool which just performs a copy of the content.
3941
This is slow but quite reliable.
3944
def __init__(self, target_format):
3945
"""Create a CopyConverter.
3947
:param target_format: The format the resulting repository should be.
3949
self.target_format = target_format
3951
def convert(self, repo, pb):
3952
"""Perform the conversion of to_convert, giving feedback via pb.
3954
:param to_convert: The disk object to convert.
3955
:param pb: a progress bar to use for progress information.
3960
# this is only useful with metadir layouts - separated repo content.
3961
# trigger an assertion if not such
3962
repo._format.get_format_string()
3963
self.repo_dir = repo.bzrdir
3964
self.step('Moving repository to repository.backup')
3965
self.repo_dir.transport.move('repository', 'repository.backup')
3966
backup_transport = self.repo_dir.transport.clone('repository.backup')
3967
repo._format.check_conversion_target(self.target_format)
3968
self.source_repo = repo._format.open(self.repo_dir,
3970
_override_transport=backup_transport)
3971
self.step('Creating new repository')
3972
converted = self.target_format.initialize(self.repo_dir,
3973
self.source_repo.is_shared())
3974
converted.lock_write()
3976
self.step('Copying content into repository.')
3977
self.source_repo.copy_content_into(converted)
3980
self.step('Deleting old repository content.')
3981
self.repo_dir.transport.delete_tree('repository.backup')
3982
self.pb.note('repository converted')
3984
def step(self, message):
3985
"""Update the pb by a step."""
3987
self.pb.update(message, self.count, self.total)
3999
def _unescaper(match, _map=_unescape_map):
4000
code = match.group(1)
4004
if not code.startswith('#'):
4006
return unichr(int(code[1:])).encode('utf8')
4012
def _unescape_xml(data):
4013
"""Unescape predefined XML entities in a string of data."""
4015
if _unescape_re is None:
4016
_unescape_re = re.compile('\&([^;]*);')
4017
return _unescape_re.sub(_unescaper, data)
4020
class _VersionedFileChecker(object):
4022
def __init__(self, repository, text_key_references=None, ancestors=None):
4023
self.repository = repository
4024
self.text_index = self.repository._generate_text_key_index(
4025
text_key_references=text_key_references, ancestors=ancestors)
4027
def calculate_file_version_parents(self, text_key):
4028
"""Calculate the correct parents for a file version according to
4031
parent_keys = self.text_index[text_key]
4032
if parent_keys == [_mod_revision.NULL_REVISION]:
4034
return tuple(parent_keys)
4036
def check_file_version_parents(self, texts, progress_bar=None):
4037
"""Check the parents stored in a versioned file are correct.
4039
It also detects file versions that are not referenced by their
4040
corresponding revision's inventory.
4042
:returns: A tuple of (wrong_parents, dangling_file_versions).
4043
wrong_parents is a dict mapping {revision_id: (stored_parents,
4044
correct_parents)} for each revision_id where the stored parents
4045
are not correct. dangling_file_versions is a set of (file_id,
4046
revision_id) tuples for versions that are present in this versioned
4047
file, but not used by the corresponding inventory.
4049
local_progress = None
4050
if progress_bar is None:
4051
local_progress = ui.ui_factory.nested_progress_bar()
4052
progress_bar = local_progress
4054
return self._check_file_version_parents(texts, progress_bar)
4057
local_progress.finished()
4059
def _check_file_version_parents(self, texts, progress_bar):
4060
"""See check_file_version_parents."""
4062
self.file_ids = set([file_id for file_id, _ in
4063
self.text_index.iterkeys()])
4064
# text keys is now grouped by file_id
4065
n_weaves = len(self.file_ids)
4066
files_in_revisions = {}
4067
revisions_of_files = {}
4068
n_versions = len(self.text_index)
4069
progress_bar.update('loading text store', 0, n_versions)
4070
parent_map = self.repository.texts.get_parent_map(self.text_index)
4071
# On unlistable transports this could well be empty/error...
4072
text_keys = self.repository.texts.keys()
4073
unused_keys = frozenset(text_keys) - set(self.text_index)
4074
for num, key in enumerate(self.text_index.iterkeys()):
4075
progress_bar.update('checking text graph', num, n_versions)
4076
correct_parents = self.calculate_file_version_parents(key)
4078
knit_parents = parent_map[key]
4079
except errors.RevisionNotPresent:
4082
if correct_parents != knit_parents:
4083
wrong_parents[key] = (knit_parents, correct_parents)
4084
return wrong_parents, unused_keys
4087
def _old_get_graph(repository, revision_id):
4088
"""DO NOT USE. That is all. I'm serious."""
4089
graph = repository.get_graph()
4090
revision_graph = dict(((key, value) for key, value in
4091
graph.iter_ancestry([revision_id]) if value is not None))
4092
return _strip_NULL_ghosts(revision_graph)
4095
def _strip_NULL_ghosts(revision_graph):
4096
"""Also don't use this. more compatibility code for unmigrated clients."""
4097
# Filter ghosts, and null:
4098
if _mod_revision.NULL_REVISION in revision_graph:
4099
del revision_graph[_mod_revision.NULL_REVISION]
4100
for key, parents in revision_graph.items():
4101
revision_graph[key] = tuple(parent for parent in parents if parent
4103
return revision_graph
4106
class StreamSink(object):
4107
"""An object that can insert a stream into a repository.
4109
This interface handles the complexity of reserialising inventories and
4110
revisions from different formats, and allows unidirectional insertion into
4111
stacked repositories without looking for the missing basis parents
4115
def __init__(self, target_repo):
4116
self.target_repo = target_repo
4118
def insert_stream(self, stream, src_format, resume_tokens):
4119
"""Insert a stream's content into the target repository.
4121
:param src_format: a bzr repository format.
4123
:return: a list of resume tokens and an iterable of keys additional
4124
items required before the insertion can be completed.
4126
self.target_repo.lock_write()
4129
self.target_repo.resume_write_group(resume_tokens)
4132
self.target_repo.start_write_group()
4135
# locked_insert_stream performs a commit|suspend.
4136
return self._locked_insert_stream(stream, src_format, is_resume)
4138
self.target_repo.abort_write_group(suppress_errors=True)
4141
self.target_repo.unlock()
4143
def _locked_insert_stream(self, stream, src_format, is_resume):
4144
to_serializer = self.target_repo._format._serializer
4145
src_serializer = src_format._serializer
4147
if to_serializer == src_serializer:
4148
# If serializers match and the target is a pack repository, set the
4149
# write cache size on the new pack. This avoids poor performance
4150
# on transports where append is unbuffered (such as
4151
# RemoteTransport). This is safe to do because nothing should read
4152
# back from the target repository while a stream with matching
4153
# serialization is being inserted.
4154
# The exception is that a delta record from the source that should
4155
# be a fulltext may need to be expanded by the target (see
4156
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4157
# explicitly flush any buffered writes first in that rare case.
4159
new_pack = self.target_repo._pack_collection._new_pack
4160
except AttributeError:
4161
# Not a pack repository
4164
new_pack.set_write_cache_size(1024*1024)
4165
for substream_type, substream in stream:
4166
if substream_type == 'texts':
4167
self.target_repo.texts.insert_record_stream(substream)
4168
elif substream_type == 'inventories':
4169
if src_serializer == to_serializer:
4170
self.target_repo.inventories.insert_record_stream(
4173
self._extract_and_insert_inventories(
4174
substream, src_serializer)
4175
elif substream_type == 'chk_bytes':
4176
# XXX: This doesn't support conversions, as it assumes the
4177
# conversion was done in the fetch code.
4178
self.target_repo.chk_bytes.insert_record_stream(substream)
4179
elif substream_type == 'revisions':
4180
# This may fallback to extract-and-insert more often than
4181
# required if the serializers are different only in terms of
4183
if src_serializer == to_serializer:
4184
self.target_repo.revisions.insert_record_stream(
4187
self._extract_and_insert_revisions(substream,
4189
elif substream_type == 'signatures':
4190
self.target_repo.signatures.insert_record_stream(substream)
4192
raise AssertionError('kaboom! %s' % (substream_type,))
4193
# Done inserting data, and the missing_keys calculations will try to
4194
# read back from the inserted data, so flush the writes to the new pack
4195
# (if this is pack format).
4196
if new_pack is not None:
4197
new_pack._write_data('', flush=True)
4198
# Find all the new revisions (including ones from resume_tokens)
4199
missing_keys = self.target_repo.get_missing_parent_inventories(
4200
check_for_missing_texts=is_resume)
4202
for prefix, versioned_file in (
4203
('texts', self.target_repo.texts),
4204
('inventories', self.target_repo.inventories),
4205
('revisions', self.target_repo.revisions),
4206
('signatures', self.target_repo.signatures),
4207
('chk_bytes', self.target_repo.chk_bytes),
4209
if versioned_file is None:
4211
missing_keys.update((prefix,) + key for key in
4212
versioned_file.get_missing_compression_parent_keys())
4213
except NotImplementedError:
4214
# cannot even attempt suspending, and missing would have failed
4215
# during stream insertion.
4216
missing_keys = set()
4219
# suspend the write group and tell the caller what we is
4220
# missing. We know we can suspend or else we would not have
4221
# entered this code path. (All repositories that can handle
4222
# missing keys can handle suspending a write group).
4223
write_group_tokens = self.target_repo.suspend_write_group()
4224
return write_group_tokens, missing_keys
4225
hint = self.target_repo.commit_write_group()
4226
if (to_serializer != src_serializer and
4227
self.target_repo._format.pack_compresses):
4228
self.target_repo.pack(hint=hint)
4231
def _extract_and_insert_inventories(self, substream, serializer):
4232
"""Generate a new inventory versionedfile in target, converting data.
4234
The inventory is retrieved from the source, (deserializing it), and
4235
stored in the target (reserializing it in a different format).
4237
for record in substream:
4238
bytes = record.get_bytes_as('fulltext')
4239
revision_id = record.key[0]
4240
inv = serializer.read_inventory_from_string(bytes, revision_id)
4241
parents = [key[0] for key in record.parents]
4242
self.target_repo.add_inventory(revision_id, inv, parents)
4244
def _extract_and_insert_revisions(self, substream, serializer):
4245
for record in substream:
4246
bytes = record.get_bytes_as('fulltext')
4247
revision_id = record.key[0]
4248
rev = serializer.read_revision_from_string(bytes)
4249
if rev.revision_id != revision_id:
4250
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4251
self.target_repo.add_revision(revision_id, rev)
4254
if self.target_repo._format._fetch_reconcile:
4255
self.target_repo.reconcile()
4258
class StreamSource(object):
4259
"""A source of a stream for fetching between repositories."""
4261
def __init__(self, from_repository, to_format):
4262
"""Create a StreamSource streaming from from_repository."""
4263
self.from_repository = from_repository
4264
self.to_format = to_format
4266
def delta_on_metadata(self):
4267
"""Return True if delta's are permitted on metadata streams.
4269
That is on revisions and signatures.
4271
src_serializer = self.from_repository._format._serializer
4272
target_serializer = self.to_format._serializer
4273
return (self.to_format._fetch_uses_deltas and
4274
src_serializer == target_serializer)
4276
def _fetch_revision_texts(self, revs):
4277
# fetch signatures first and then the revision texts
4278
# may need to be a InterRevisionStore call here.
4279
from_sf = self.from_repository.signatures
4280
# A missing signature is just skipped.
4281
keys = [(rev_id,) for rev_id in revs]
4282
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4284
self.to_format._fetch_order,
4285
not self.to_format._fetch_uses_deltas))
4286
# If a revision has a delta, this is actually expanded inside the
4287
# insert_record_stream code now, which is an alternate fix for
4289
from_rf = self.from_repository.revisions
4290
revisions = from_rf.get_record_stream(
4292
self.to_format._fetch_order,
4293
not self.delta_on_metadata())
4294
return [('signatures', signatures), ('revisions', revisions)]
4296
def _generate_root_texts(self, revs):
4297
"""This will be called by __fetch between fetching weave texts and
4298
fetching the inventory weave.
4300
Subclasses should override this if they need to generate root texts
4301
after fetching weave texts.
4303
if self._rich_root_upgrade():
4305
return bzrlib.fetch.Inter1and2Helper(
4306
self.from_repository).generate_root_texts(revs)
4310
def get_stream(self, search):
4312
revs = search.get_keys()
4313
graph = self.from_repository.get_graph()
4314
revs = list(graph.iter_topo_order(revs))
4315
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4317
for knit_kind, file_id, revisions in data_to_fetch:
4318
if knit_kind != phase:
4320
# Make a new progress bar for this phase
4321
if knit_kind == "file":
4322
# Accumulate file texts
4323
text_keys.extend([(file_id, revision) for revision in
4325
elif knit_kind == "inventory":
4326
# Now copy the file texts.
4327
from_texts = self.from_repository.texts
4328
yield ('texts', from_texts.get_record_stream(
4329
text_keys, self.to_format._fetch_order,
4330
not self.to_format._fetch_uses_deltas))
4331
# Cause an error if a text occurs after we have done the
4334
# Before we process the inventory we generate the root
4335
# texts (if necessary) so that the inventories references
4337
for _ in self._generate_root_texts(revs):
4339
# NB: This currently reopens the inventory weave in source;
4340
# using a single stream interface instead would avoid this.
4341
from_weave = self.from_repository.inventories
4342
# we fetch only the referenced inventories because we do not
4343
# know for unselected inventories whether all their required
4344
# texts are present in the other repository - it could be
4346
for info in self._get_inventory_stream(revs):
4348
elif knit_kind == "signatures":
4349
# Nothing to do here; this will be taken care of when
4350
# _fetch_revision_texts happens.
4352
elif knit_kind == "revisions":
4353
for record in self._fetch_revision_texts(revs):
4356
raise AssertionError("Unknown knit kind %r" % knit_kind)
4358
def get_stream_for_missing_keys(self, missing_keys):
4359
# missing keys can only occur when we are byte copying and not
4360
# translating (because translation means we don't send
4361
# unreconstructable deltas ever).
4363
keys['texts'] = set()
4364
keys['revisions'] = set()
4365
keys['inventories'] = set()
4366
keys['chk_bytes'] = set()
4367
keys['signatures'] = set()
4368
for key in missing_keys:
4369
keys[key[0]].add(key[1:])
4370
if len(keys['revisions']):
4371
# If we allowed copying revisions at this point, we could end up
4372
# copying a revision without copying its required texts: a
4373
# violation of the requirements for repository integrity.
4374
raise AssertionError(
4375
'cannot copy revisions to fill in missing deltas %s' % (
4376
keys['revisions'],))
4377
for substream_kind, keys in keys.iteritems():
4378
vf = getattr(self.from_repository, substream_kind)
4379
if vf is None and keys:
4380
raise AssertionError(
4381
"cannot fill in keys for a versioned file we don't"
4382
" have: %s needs %s" % (substream_kind, keys))
4384
# No need to stream something we don't have
4386
# Ask for full texts always so that we don't need more round trips
4387
# after this stream.
4388
# Some of the missing keys are genuinely ghosts, so filter absent
4389
# records. The Sink is responsible for doing another check to
4390
# ensure that ghosts don't introduce missing data for future
4392
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4393
self.to_format._fetch_order, True))
4394
yield substream_kind, stream
4396
def inventory_fetch_order(self):
4397
if self._rich_root_upgrade():
4398
return 'topological'
4400
return self.to_format._fetch_order
4402
def _rich_root_upgrade(self):
4403
return (not self.from_repository._format.rich_root_data and
4404
self.to_format.rich_root_data)
4406
def _get_inventory_stream(self, revision_ids):
4407
from_format = self.from_repository._format
4408
if (from_format.supports_chks and self.to_format.supports_chks
4409
and (from_format._serializer == self.to_format._serializer)):
4410
# Both sides support chks, and they use the same serializer, so it
4411
# is safe to transmit the chk pages and inventory pages across
4413
return self._get_chk_inventory_stream(revision_ids)
4414
elif (not from_format.supports_chks):
4415
# Source repository doesn't support chks. So we can transmit the
4416
# inventories 'as-is' and either they are just accepted on the
4417
# target, or the Sink will properly convert it.
4418
return self._get_simple_inventory_stream(revision_ids)
4420
# XXX: Hack to make not-chk->chk fetch: copy the inventories as
4421
# inventories. Note that this should probably be done somehow
4422
# as part of bzrlib.repository.StreamSink. Except JAM couldn't
4423
# figure out how a non-chk repository could possibly handle
4424
# deserializing an inventory stream from a chk repo, as it
4425
# doesn't have a way to understand individual pages.
4426
return self._get_convertable_inventory_stream(revision_ids)
4428
def _get_simple_inventory_stream(self, revision_ids):
4429
from_weave = self.from_repository.inventories
4430
yield ('inventories', from_weave.get_record_stream(
4431
[(rev_id,) for rev_id in revision_ids],
4432
self.inventory_fetch_order(),
4433
not self.delta_on_metadata()))
4435
def _get_chk_inventory_stream(self, revision_ids):
4436
"""Fetch the inventory texts, along with the associated chk maps."""
4437
# We want an inventory outside of the search set, so that we can filter
4438
# out uninteresting chk pages. For now we use
4439
# _find_revision_outside_set, but if we had a Search with cut_revs, we
4440
# could use that instead.
4441
start_rev_id = self.from_repository._find_revision_outside_set(
4443
start_rev_key = (start_rev_id,)
4444
inv_keys_to_fetch = [(rev_id,) for rev_id in revision_ids]
4445
if start_rev_id != _mod_revision.NULL_REVISION:
4446
inv_keys_to_fetch.append((start_rev_id,))
4447
# Any repo that supports chk_bytes must also support out-of-order
4448
# insertion. At least, that is how we expect it to work
4449
# We use get_record_stream instead of iter_inventories because we want
4450
# to be able to insert the stream as well. We could instead fetch
4451
# allowing deltas, and then iter_inventories, but we don't know whether
4452
# source or target is more 'local' anway.
4453
inv_stream = self.from_repository.inventories.get_record_stream(
4454
inv_keys_to_fetch, 'unordered',
4455
True) # We need them as full-texts so we can find their references
4456
uninteresting_chk_roots = set()
4457
interesting_chk_roots = set()
4458
def filter_inv_stream(inv_stream):
4459
for idx, record in enumerate(inv_stream):
4460
### child_pb.update('fetch inv', idx, len(inv_keys_to_fetch))
4461
bytes = record.get_bytes_as('fulltext')
4462
chk_inv = inventory.CHKInventory.deserialise(
4463
self.from_repository.chk_bytes, bytes, record.key)
4464
if record.key == start_rev_key:
4465
uninteresting_chk_roots.add(chk_inv.id_to_entry.key())
4466
p_id_map = chk_inv.parent_id_basename_to_file_id
4467
if p_id_map is not None:
4468
uninteresting_chk_roots.add(p_id_map.key())
4471
interesting_chk_roots.add(chk_inv.id_to_entry.key())
4472
p_id_map = chk_inv.parent_id_basename_to_file_id
4473
if p_id_map is not None:
4474
interesting_chk_roots.add(p_id_map.key())
4475
### pb.update('fetch inventory', 0, 2)
4476
yield ('inventories', filter_inv_stream(inv_stream))
4477
# Now that we have worked out all of the interesting root nodes, grab
4478
# all of the interesting pages and insert them
4479
### pb.update('fetch inventory', 1, 2)
4480
interesting = chk_map.iter_interesting_nodes(
4481
self.from_repository.chk_bytes, interesting_chk_roots,
4482
uninteresting_chk_roots)
4483
def to_stream_adapter():
4484
"""Adapt the iter_interesting_nodes result to a single stream.
4486
iter_interesting_nodes returns records as it processes them, along
4487
with keys. However, we only want to return the records themselves.
4489
for record, items in interesting:
4490
if record is not None:
4492
# XXX: We could instead call get_record_stream(records.keys())
4493
# ATM, this will always insert the records as fulltexts, and
4494
# requires that you can hang on to records once you have gone
4495
# on to the next one. Further, it causes the target to
4496
# recompress the data. Testing shows it to be faster than
4497
# requesting the records again, though.
4498
yield ('chk_bytes', to_stream_adapter())
4499
### pb.update('fetch inventory', 2, 2)
4501
def _get_convertable_inventory_stream(self, revision_ids):
4502
# XXX: One of source or target is using chks, and they don't have
4503
# compatible serializations. The StreamSink code expects to be
4504
# able to convert on the target, so we need to put
4505
# bytes-on-the-wire that can be converted
4506
yield ('inventories', self._stream_invs_as_fulltexts(revision_ids))
4508
def _stream_invs_as_fulltexts(self, revision_ids):
4509
from_repo = self.from_repository
4510
from_serializer = from_repo._format._serializer
4511
revision_keys = [(rev_id,) for rev_id in revision_ids]
4512
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4513
for inv in self.from_repository.iter_inventories(revision_ids):
4514
# XXX: This is a bit hackish, but it works. Basically,
4515
# CHKSerializer 'accidentally' supports
4516
# read/write_inventory_to_string, even though that is never
4517
# the format that is stored on disk. It *does* give us a
4518
# single string representation for an inventory, so live with
4520
# This would be far better if we had a 'serialized inventory
4521
# delta' form. Then we could use 'inventory._make_delta', and
4522
# transmit that. This would both be faster to generate, and
4523
# result in fewer bytes-on-the-wire.
4524
as_bytes = from_serializer.write_inventory_to_string(inv)
4525
key = (inv.revision_id,)
4526
parent_keys = parent_map.get(key, ())
4527
yield versionedfile.FulltextContentFactory(
4528
key, parent_keys, None, as_bytes)
4531
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4532
stop_revision=None):
4533
"""Extend the partial history to include a given index
4535
If a stop_index is supplied, stop when that index has been reached.
4536
If a stop_revision is supplied, stop when that revision is
4537
encountered. Otherwise, stop when the beginning of history is
4540
:param stop_index: The index which should be present. When it is
4541
present, history extension will stop.
4542
:param stop_revision: The revision id which should be present. When
4543
it is encountered, history extension will stop.
4545
start_revision = partial_history_cache[-1]
4546
iterator = repo.iter_reverse_revision_history(start_revision)
4548
#skip the last revision in the list
4551
if (stop_index is not None and
4552
len(partial_history_cache) > stop_index):
4554
if partial_history_cache[-1] == stop_revision:
4556
revision_id = iterator.next()
4557
partial_history_cache.append(revision_id)
4558
except StopIteration:
added
removed
bzrlib/repository.py
