« back to all changes in this revision
Viewing changes to bzrlib/repository.py
- browse files at revision 5844
- compare with another revision
- compare with revision 5815.4.18
- download diff
- download tarball
- view history from revision 5844
- Committer: Canonical.com Patch Queue Manager
- Date: 2011-05-09 15:39:14 UTC
- mfrom: (5815.4.22 separate-vf)
- Revision ID: pqm@pqm.ubuntu.com-20110509153914-wbwvbi34ohzwpv9k
(jelmer) Split VersionedFiles-specific bits out of bzrlib.repository into
bzrlib.vf_repository. (Jelmer Vernooij)
bzrlib.vf_repository. (Jelmer Vernooij)
- files added:
- bzrlib/vf_repository.py
- files modified:
- bzrlib/bundle/apply_bundle.py
added
removed
bzrlib/repository.py
24
24
config,
25
25
controldir,
26
26
debug,
27
fetch as _mod_fetch,
28
fifo_cache,
29
27
generate_ids,
30
gpg,
31
28
graph,
32
inventory_delta,
33
29
lockable_files,
34
30
lockdir,
35
lru_cache,
36
31
osutils,
37
32
revision as _mod_revision,
38
static_tuple,
33
testament as _mod_testament,
39
34
tsort,
40
versionedfile,
41
35
)
42
36
from bzrlib.bundle import serializer
43
from bzrlib.recordcounter import RecordCounter
44
from bzrlib.revisiontree import InventoryRevisionTree
45
from bzrlib.store.versioned import VersionedFileStore
46
from bzrlib.testament import Testament
47
37
""")
48
38
49
39
from bzrlib import (
54
44
)
55
45
from bzrlib.decorators import needs_read_lock, needs_write_lock, only_raises
56
46
from bzrlib.inter import InterObject
57
from bzrlib.inventory import (
58
Inventory,
59
InventoryDirectory,
60
ROOT_ID,
61
entry_factory,
62
)
63
47
from bzrlib.lock import _RelockDebugMixin, LogicalLockResult
64
48
from bzrlib.trace import (
65
49
log_exception_quietly, note, mutter, mutter_callsite, warning)
86
70
87
71
# all clients should supply tree roots.
88
72
record_root_entry = True
89
# the default CommitBuilder does not manage trees whose root is versioned.
90
_versioned_root = False
91
# this commit builder supports the record_entry_contents interface
92
supports_record_entry_contents = True
73
# whether this commit builder supports the record_entry_contents interface
74
supports_record_entry_contents = False
93
75
94
76
def __init__(self, repository, parents, config, timestamp=None,
95
77
timezone=None, committer=None, revprops=None,
116
98
else:
117
99
self._committer = committer
118
100
119
self.new_inventory = Inventory(None)
120
101
self._new_revision_id = revision_id
121
102
self.parents = parents
122
103
self.repository = repository
137
118
self._timezone = int(timezone)
138
119
139
120
self._generate_revision_if_needed()
140
self.__heads = graph.HeadsCache(repository.get_graph()).heads
141
self._basis_delta = []
142
# API compatibility, older code that used CommitBuilder did not call
143
# .record_delete(), which means the delta that is computed would not be
144
# valid. Callers that will call record_delete() should call
145
# .will_record_deletes() to indicate that.
146
self._recording_deletes = False
147
# memo'd check for no-op commits.
148
self._any_changes = False
149
121
150
122
def any_changes(self):
151
123
"""Return True if any entries were changed.
152
124
153
125
This includes merge-only changes. It is the core for the --unchanged
154
126
detection in commit.
155
127
156
128
:return: True if any changes have occured.
157
129
"""
158
return self._any_changes
130
raise NotImplementedError(self.any_changes)
159
131
160
132
def _validate_unicode_text(self, text, context):
161
133
"""Verify things like commit messages don't have bogus characters."""
172
144
self._validate_unicode_text(value,
173
145
'revision property (%s)' % (key,))
174
146
175
def _ensure_fallback_inventories(self):
176
"""Ensure that appropriate inventories are available.
177
178
This only applies to repositories that are stacked, and is about
179
enusring the stacking invariants. Namely, that for any revision that is
180
present, we either have all of the file content, or we have the parent
181
inventory and the delta file content.
182
"""
183
if not self.repository._fallback_repositories:
184
return
185
if not self.repository._format.supports_chks:
186
raise errors.BzrError("Cannot commit directly to a stacked branch"
187
" in pre-2a formats. See "
188
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
189
# This is a stacked repo, we need to make sure we have the parent
190
# inventories for the parents.
191
parent_keys = [(p,) for p in self.parents]
192
parent_map = self.repository.inventories._index.get_parent_map(parent_keys)
193
missing_parent_keys = set([pk for pk in parent_keys
194
if pk not in parent_map])
195
fallback_repos = list(reversed(self.repository._fallback_repositories))
196
missing_keys = [('inventories', pk[0])
197
for pk in missing_parent_keys]
198
resume_tokens = []
199
while missing_keys and fallback_repos:
200
fallback_repo = fallback_repos.pop()
201
source = fallback_repo._get_source(self.repository._format)
202
sink = self.repository._get_sink()
203
stream = source.get_stream_for_missing_keys(missing_keys)
204
missing_keys = sink.insert_stream_without_locking(stream,
205
self.repository._format)
206
if missing_keys:
207
raise errors.BzrError('Unable to fill in parent inventories for a'
208
' stacked branch')
209
210
147
def commit(self, message):
211
148
"""Make the actual commit.
212
149
213
150
:return: The revision id of the recorded revision.
214
151
"""
215
self._validate_unicode_text(message, 'commit message')
216
rev = _mod_revision.Revision(
217
timestamp=self._timestamp,
218
timezone=self._timezone,
219
committer=self._committer,
220
message=message,
221
inventory_sha1=self.inv_sha1,
222
revision_id=self._new_revision_id,
223
properties=self._revprops)
224
rev.parent_ids = self.parents
225
self.repository.add_revision(self._new_revision_id, rev,
226
self.new_inventory, self._config)
227
self._ensure_fallback_inventories()
228
self.repository.commit_write_group()
229
return self._new_revision_id
152
raise NotImplementedError(self.commit)
230
153
231
154
def abort(self):
232
155
"""Abort the commit that is being built.
233
156
"""
234
self.repository.abort_write_group()
157
raise NotImplementedError(self.abort)
235
158
236
159
def revision_tree(self):
237
160
"""Return the tree that was just committed.
238
161
239
162
After calling commit() this can be called to get a
240
InventoryRevisionTree representing the newly committed tree. This is
163
RevisionTree representing the newly committed tree. This is
241
164
preferred to calling Repository.revision_tree() because that may
242
165
require deserializing the inventory, while we already have a copy in
243
166
memory.
244
167
"""
245
if self.new_inventory is None:
246
self.new_inventory = self.repository.get_inventory(
247
self._new_revision_id)
248
return InventoryRevisionTree(self.repository, self.new_inventory,
249
self._new_revision_id)
168
raise NotImplementedError(self.revision_tree)
250
169
251
170
def finish_inventory(self):
252
171
"""Tell the builder that the inventory is finished.
254
173
:return: The inventory id in the repository, which can be used with
255
174
repository.get_inventory.
256
175
"""
257
if self.new_inventory is None:
258
# an inventory delta was accumulated without creating a new
259
# inventory.
260
basis_id = self.basis_delta_revision
261
# We ignore the 'inventory' returned by add_inventory_by_delta
262
# because self.new_inventory is used to hint to the rest of the
263
# system what code path was taken
264
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
265
basis_id, self._basis_delta, self._new_revision_id,
266
self.parents)
267
else:
268
if self.new_inventory.root is None:
269
raise AssertionError('Root entry should be supplied to'
270
' record_entry_contents, as of bzr 0.10.')
271
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
272
self.new_inventory.revision_id = self._new_revision_id
273
self.inv_sha1 = self.repository.add_inventory(
274
self._new_revision_id,
275
self.new_inventory,
276
self.parents
277
)
278
return self._new_revision_id
176
raise NotImplementedError(self.finish_inventory)
279
177
280
178
def _gen_revision_id(self):
281
179
"""Return new revision-id."""
296
194
else:
297
195
self.random_revid = False
298
196
299
def _heads(self, file_id, revision_ids):
300
"""Calculate the graph heads for revision_ids in the graph of file_id.
301
302
This can use either a per-file graph or a global revision graph as we
303
have an identity relationship between the two graphs.
304
"""
305
return self.__heads(revision_ids)
306
307
def _check_root(self, ie, parent_invs, tree):
308
"""Helper for record_entry_contents.
309
310
:param ie: An entry being added.
311
:param parent_invs: The inventories of the parent revisions of the
312
commit.
313
:param tree: The tree that is being committed.
314
"""
315
# In this revision format, root entries have no knit or weave When
316
# serializing out to disk and back in root.revision is always
317
# _new_revision_id
318
ie.revision = self._new_revision_id
319
320
def _require_root_change(self, tree):
321
"""Enforce an appropriate root object change.
322
323
This is called once when record_iter_changes is called, if and only if
324
the root was not in the delta calculated by record_iter_changes.
325
326
:param tree: The tree which is being committed.
327
"""
328
if len(self.parents) == 0:
329
raise errors.RootMissing()
330
entry = entry_factory['directory'](tree.path2id(''), '',
331
None)
332
entry.revision = self._new_revision_id
333
self._basis_delta.append(('', '', entry.file_id, entry))
334
335
def _get_delta(self, ie, basis_inv, path):
336
"""Get a delta against the basis inventory for ie."""
337
if ie.file_id not in basis_inv:
338
# add
339
result = (None, path, ie.file_id, ie)
340
self._basis_delta.append(result)
341
return result
342
elif ie != basis_inv[ie.file_id]:
343
# common but altered
344
# TODO: avoid tis id2path call.
345
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
346
self._basis_delta.append(result)
347
return result
348
else:
349
# common, unaltered
350
return None
351
352
def get_basis_delta(self):
353
"""Return the complete inventory delta versus the basis inventory.
354
355
This has been built up with the calls to record_delete and
356
record_entry_contents. The client must have already called
357
will_record_deletes() to indicate that they will be generating a
358
complete delta.
359
360
:return: An inventory delta, suitable for use with apply_delta, or
361
Repository.add_inventory_by_delta, etc.
362
"""
363
if not self._recording_deletes:
364
raise AssertionError("recording deletes not activated.")
365
return self._basis_delta
366
367
def record_delete(self, path, file_id):
368
"""Record that a delete occured against a basis tree.
369
370
This is an optional API - when used it adds items to the basis_delta
371
being accumulated by the commit builder. It cannot be called unless the
372
method will_record_deletes() has been called to inform the builder that
373
a delta is being supplied.
374
375
:param path: The path of the thing deleted.
376
:param file_id: The file id that was deleted.
377
"""
378
if not self._recording_deletes:
379
raise AssertionError("recording deletes not activated.")
380
delta = (path, None, file_id, None)
381
self._basis_delta.append(delta)
382
self._any_changes = True
383
return delta
384
385
197
def will_record_deletes(self):
386
198
"""Tell the commit builder that deletes are being notified.
387
199
389
201
commit to be valid, deletes against the basis MUST be recorded via
390
202
builder.record_delete().
391
203
"""
392
self._recording_deletes = True
393
try:
394
basis_id = self.parents[0]
395
except IndexError:
396
basis_id = _mod_revision.NULL_REVISION
397
self.basis_delta_revision = basis_id
398
399
def record_entry_contents(self, ie, parent_invs, path, tree,
400
content_summary):
401
"""Record the content of ie from tree into the commit if needed.
402
403
Side effect: sets ie.revision when unchanged
404
405
:param ie: An inventory entry present in the commit.
406
:param parent_invs: The inventories of the parent revisions of the
407
commit.
408
:param path: The path the entry is at in the tree.
409
:param tree: The tree which contains this entry and should be used to
410
obtain content.
411
:param content_summary: Summary data from the tree about the paths
412
content - stat, length, exec, sha/link target. This is only
413
accessed when the entry has a revision of None - that is when it is
414
a candidate to commit.
415
:return: A tuple (change_delta, version_recorded, fs_hash).
416
change_delta is an inventory_delta change for this entry against
417
the basis tree of the commit, or None if no change occured against
418
the basis tree.
419
version_recorded is True if a new version of the entry has been
420
recorded. For instance, committing a merge where a file was only
421
changed on the other side will return (delta, False).
422
fs_hash is either None, or the hash details for the path (currently
423
a tuple of the contents sha1 and the statvalue returned by
424
tree.get_file_with_stat()).
425
"""
426
if self.new_inventory.root is None:
427
if ie.parent_id is not None:
428
raise errors.RootMissing()
429
self._check_root(ie, parent_invs, tree)
430
if ie.revision is None:
431
kind = content_summary[0]
432
else:
433
# ie is carried over from a prior commit
434
kind = ie.kind
435
# XXX: repository specific check for nested tree support goes here - if
436
# the repo doesn't want nested trees we skip it ?
437
if (kind == 'tree-reference' and
438
not self.repository._format.supports_tree_reference):
439
# mismatch between commit builder logic and repository:
440
# this needs the entry creation pushed down into the builder.
441
raise NotImplementedError('Missing repository subtree support.')
442
self.new_inventory.add(ie)
443
444
# TODO: slow, take it out of the inner loop.
445
try:
446
basis_inv = parent_invs[0]
447
except IndexError:
448
basis_inv = Inventory(root_id=None)
449
450
# ie.revision is always None if the InventoryEntry is considered
451
# for committing. We may record the previous parents revision if the
452
# content is actually unchanged against a sole head.
453
if ie.revision is not None:
454
if not self._versioned_root and path == '':
455
# repositories that do not version the root set the root's
456
# revision to the new commit even when no change occurs (more
457
# specifically, they do not record a revision on the root; and
458
# the rev id is assigned to the root during deserialisation -
459
# this masks when a change may have occurred against the basis.
460
# To match this we always issue a delta, because the revision
461
# of the root will always be changing.
462
if ie.file_id in basis_inv:
463
delta = (basis_inv.id2path(ie.file_id), path,
464
ie.file_id, ie)
465
else:
466
# add
467
delta = (None, path, ie.file_id, ie)
468
self._basis_delta.append(delta)
469
return delta, False, None
470
else:
471
# we don't need to commit this, because the caller already
472
# determined that an existing revision of this file is
473
# appropriate. If it's not being considered for committing then
474
# it and all its parents to the root must be unaltered so
475
# no-change against the basis.
476
if ie.revision == self._new_revision_id:
477
raise AssertionError("Impossible situation, a skipped "
478
"inventory entry (%r) claims to be modified in this "
479
"commit (%r).", (ie, self._new_revision_id))
480
return None, False, None
481
# XXX: Friction: parent_candidates should return a list not a dict
482
# so that we don't have to walk the inventories again.
483
parent_candiate_entries = ie.parent_candidates(parent_invs)
484
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
485
heads = []
486
for inv in parent_invs:
487
if ie.file_id in inv:
488
old_rev = inv[ie.file_id].revision
489
if old_rev in head_set:
490
heads.append(inv[ie.file_id].revision)
491
head_set.remove(inv[ie.file_id].revision)
492
493
store = False
494
# now we check to see if we need to write a new record to the
495
# file-graph.
496
# We write a new entry unless there is one head to the ancestors, and
497
# the kind-derived content is unchanged.
498
499
# Cheapest check first: no ancestors, or more the one head in the
500
# ancestors, we write a new node.
501
if len(heads) != 1:
502
store = True
503
if not store:
504
# There is a single head, look it up for comparison
505
parent_entry = parent_candiate_entries[heads[0]]
506
# if the non-content specific data has changed, we'll be writing a
507
# node:
508
if (parent_entry.parent_id != ie.parent_id or
509
parent_entry.name != ie.name):
510
store = True
511
# now we need to do content specific checks:
512
if not store:
513
# if the kind changed the content obviously has
514
if kind != parent_entry.kind:
515
store = True
516
# Stat cache fingerprint feedback for the caller - None as we usually
517
# don't generate one.
518
fingerprint = None
519
if kind == 'file':
520
if content_summary[2] is None:
521
raise ValueError("Files must not have executable = None")
522
if not store:
523
# We can't trust a check of the file length because of content
524
# filtering...
525
if (# if the exec bit has changed we have to store:
526
parent_entry.executable != content_summary[2]):
527
store = True
528
elif parent_entry.text_sha1 == content_summary[3]:
529
# all meta and content is unchanged (using a hash cache
530
# hit to check the sha)
531
ie.revision = parent_entry.revision
532
ie.text_size = parent_entry.text_size
533
ie.text_sha1 = parent_entry.text_sha1
534
ie.executable = parent_entry.executable
535
return self._get_delta(ie, basis_inv, path), False, None
536
else:
537
# Either there is only a hash change(no hash cache entry,
538
# or same size content change), or there is no change on
539
# this file at all.
540
# Provide the parent's hash to the store layer, so that the
541
# content is unchanged we will not store a new node.
542
nostore_sha = parent_entry.text_sha1
543
if store:
544
# We want to record a new node regardless of the presence or
545
# absence of a content change in the file.
546
nostore_sha = None
547
ie.executable = content_summary[2]
548
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
549
try:
550
text = file_obj.read()
551
finally:
552
file_obj.close()
553
try:
554
ie.text_sha1, ie.text_size = self._add_text_to_weave(
555
ie.file_id, text, heads, nostore_sha)
556
# Let the caller know we generated a stat fingerprint.
557
fingerprint = (ie.text_sha1, stat_value)
558
except errors.ExistingContent:
559
# Turns out that the file content was unchanged, and we were
560
# only going to store a new node if it was changed. Carry over
561
# the entry.
562
ie.revision = parent_entry.revision
563
ie.text_size = parent_entry.text_size
564
ie.text_sha1 = parent_entry.text_sha1
565
ie.executable = parent_entry.executable
566
return self._get_delta(ie, basis_inv, path), False, None
567
elif kind == 'directory':
568
if not store:
569
# all data is meta here, nothing specific to directory, so
570
# carry over:
571
ie.revision = parent_entry.revision
572
return self._get_delta(ie, basis_inv, path), False, None
573
self._add_text_to_weave(ie.file_id, '', heads, None)
574
elif kind == 'symlink':
575
current_link_target = content_summary[3]
576
if not store:
577
# symlink target is not generic metadata, check if it has
578
# changed.
579
if current_link_target != parent_entry.symlink_target:
580
store = True
581
if not store:
582
# unchanged, carry over.
583
ie.revision = parent_entry.revision
584
ie.symlink_target = parent_entry.symlink_target
585
return self._get_delta(ie, basis_inv, path), False, None
586
ie.symlink_target = current_link_target
587
self._add_text_to_weave(ie.file_id, '', heads, None)
588
elif kind == 'tree-reference':
589
if not store:
590
if content_summary[3] != parent_entry.reference_revision:
591
store = True
592
if not store:
593
# unchanged, carry over.
594
ie.reference_revision = parent_entry.reference_revision
595
ie.revision = parent_entry.revision
596
return self._get_delta(ie, basis_inv, path), False, None
597
ie.reference_revision = content_summary[3]
598
if ie.reference_revision is None:
599
raise AssertionError("invalid content_summary for nested tree: %r"
600
% (content_summary,))
601
self._add_text_to_weave(ie.file_id, '', heads, None)
602
else:
603
raise NotImplementedError('unknown kind')
604
ie.revision = self._new_revision_id
605
self._any_changes = True
606
return self._get_delta(ie, basis_inv, path), True, fingerprint
607
608
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
609
_entry_factory=entry_factory):
204
raise NotImplementedError(self.will_record_deletes)
205
206
def record_iter_changes(self, tree, basis_revision_id, iter_changes):
610
207
"""Record a new tree via iter_changes.
611
208
612
209
:param tree: The tree to obtain text contents from for changed objects.
617
214
to basis_revision_id. The iterator must not include any items with
618
215
a current kind of None - missing items must be either filtered out
619
216
or errored-on beefore record_iter_changes sees the item.
620
:param _entry_factory: Private method to bind entry_factory locally for
621
performance.
622
217
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
623
218
tree._observed_sha1.
624
219
"""
625
# Create an inventory delta based on deltas between all the parents and
626
# deltas between all the parent inventories. We use inventory delta's
627
# between the inventory objects because iter_changes masks
628
# last-changed-field only changes.
629
# Working data:
630
# file_id -> change map, change is fileid, paths, changed, versioneds,
631
# parents, names, kinds, executables
632
merged_ids = {}
633
# {file_id -> revision_id -> inventory entry, for entries in parent
634
# trees that are not parents[0]
635
parent_entries = {}
636
ghost_basis = False
637
try:
638
revtrees = list(self.repository.revision_trees(self.parents))
639
except errors.NoSuchRevision:
640
# one or more ghosts, slow path.
641
revtrees = []
642
for revision_id in self.parents:
643
try:
644
revtrees.append(self.repository.revision_tree(revision_id))
645
except errors.NoSuchRevision:
646
if not revtrees:
647
basis_revision_id = _mod_revision.NULL_REVISION
648
ghost_basis = True
649
revtrees.append(self.repository.revision_tree(
650
_mod_revision.NULL_REVISION))
651
# The basis inventory from a repository
652
if revtrees:
653
basis_inv = revtrees[0].inventory
654
else:
655
basis_inv = self.repository.revision_tree(
656
_mod_revision.NULL_REVISION).inventory
657
if len(self.parents) > 0:
658
if basis_revision_id != self.parents[0] and not ghost_basis:
659
raise Exception(
660
"arbitrary basis parents not yet supported with merges")
661
for revtree in revtrees[1:]:
662
for change in revtree.inventory._make_delta(basis_inv):
663
if change[1] is None:
664
# Not present in this parent.
665
continue
666
if change[2] not in merged_ids:
667
if change[0] is not None:
668
basis_entry = basis_inv[change[2]]
669
merged_ids[change[2]] = [
670
# basis revid
671
basis_entry.revision,
672
# new tree revid
673
change[3].revision]
674
parent_entries[change[2]] = {
675
# basis parent
676
basis_entry.revision:basis_entry,
677
# this parent
678
change[3].revision:change[3],
679
}
680
else:
681
merged_ids[change[2]] = [change[3].revision]
682
parent_entries[change[2]] = {change[3].revision:change[3]}
683
else:
684
merged_ids[change[2]].append(change[3].revision)
685
parent_entries[change[2]][change[3].revision] = change[3]
686
else:
687
merged_ids = {}
688
# Setup the changes from the tree:
689
# changes maps file_id -> (change, [parent revision_ids])
690
changes= {}
691
for change in iter_changes:
692
# This probably looks up in basis_inv way to much.
693
if change[1][0] is not None:
694
head_candidate = [basis_inv[change[0]].revision]
695
else:
696
head_candidate = []
697
changes[change[0]] = change, merged_ids.get(change[0],
698
head_candidate)
699
unchanged_merged = set(merged_ids) - set(changes)
700
# Extend the changes dict with synthetic changes to record merges of
701
# texts.
702
for file_id in unchanged_merged:
703
# Record a merged version of these items that did not change vs the
704
# basis. This can be either identical parallel changes, or a revert
705
# of a specific file after a merge. The recorded content will be
706
# that of the current tree (which is the same as the basis), but
707
# the per-file graph will reflect a merge.
708
# NB:XXX: We are reconstructing path information we had, this
709
# should be preserved instead.
710
# inv delta change: (file_id, (path_in_source, path_in_target),
711
# changed_content, versioned, parent, name, kind,
712
# executable)
713
try:
714
basis_entry = basis_inv[file_id]
715
except errors.NoSuchId:
716
# a change from basis->some_parents but file_id isn't in basis
717
# so was new in the merge, which means it must have changed
718
# from basis -> current, and as it hasn't the add was reverted
719
# by the user. So we discard this change.
720
pass
721
else:
722
change = (file_id,
723
(basis_inv.id2path(file_id), tree.id2path(file_id)),
724
False, (True, True),
725
(basis_entry.parent_id, basis_entry.parent_id),
726
(basis_entry.name, basis_entry.name),
727
(basis_entry.kind, basis_entry.kind),
728
(basis_entry.executable, basis_entry.executable))
729
changes[file_id] = (change, merged_ids[file_id])
730
# changes contains tuples with the change and a set of inventory
731
# candidates for the file.
732
# inv delta is:
733
# old_path, new_path, file_id, new_inventory_entry
734
seen_root = False # Is the root in the basis delta?
735
inv_delta = self._basis_delta
736
modified_rev = self._new_revision_id
737
for change, head_candidates in changes.values():
738
if change[3][1]: # versioned in target.
739
# Several things may be happening here:
740
# We may have a fork in the per-file graph
741
# - record a change with the content from tree
742
# We may have a change against < all trees
743
# - carry over the tree that hasn't changed
744
# We may have a change against all trees
745
# - record the change with the content from tree
746
kind = change[6][1]
747
file_id = change[0]
748
entry = _entry_factory[kind](file_id, change[5][1],
749
change[4][1])
750
head_set = self._heads(change[0], set(head_candidates))
751
heads = []
752
# Preserve ordering.
753
for head_candidate in head_candidates:
754
if head_candidate in head_set:
755
heads.append(head_candidate)
756
head_set.remove(head_candidate)
757
carried_over = False
758
if len(heads) == 1:
759
# Could be a carry-over situation:
760
parent_entry_revs = parent_entries.get(file_id, None)
761
if parent_entry_revs:
762
parent_entry = parent_entry_revs.get(heads[0], None)
763
else:
764
parent_entry = None
765
if parent_entry is None:
766
# The parent iter_changes was called against is the one
767
# that is the per-file head, so any change is relevant
768
# iter_changes is valid.
769
carry_over_possible = False
770
else:
771
# could be a carry over situation
772
# A change against the basis may just indicate a merge,
773
# we need to check the content against the source of the
774
# merge to determine if it was changed after the merge
775
# or carried over.
776
if (parent_entry.kind != entry.kind or
777
parent_entry.parent_id != entry.parent_id or
778
parent_entry.name != entry.name):
779
# Metadata common to all entries has changed
780
# against per-file parent
781
carry_over_possible = False
782
else:
783
carry_over_possible = True
784
# per-type checks for changes against the parent_entry
785
# are done below.
786
else:
787
# Cannot be a carry-over situation
788
carry_over_possible = False
789
# Populate the entry in the delta
790
if kind == 'file':
791
# XXX: There is still a small race here: If someone reverts the content of a file
792
# after iter_changes examines and decides it has changed,
793
# we will unconditionally record a new version even if some
794
# other process reverts it while commit is running (with
795
# the revert happening after iter_changes did its
796
# examination).
797
if change[7][1]:
798
entry.executable = True
799
else:
800
entry.executable = False
801
if (carry_over_possible and
802
parent_entry.executable == entry.executable):
803
# Check the file length, content hash after reading
804
# the file.
805
nostore_sha = parent_entry.text_sha1
806
else:
807
nostore_sha = None
808
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
809
try:
810
text = file_obj.read()
811
finally:
812
file_obj.close()
813
try:
814
entry.text_sha1, entry.text_size = self._add_text_to_weave(
815
file_id, text, heads, nostore_sha)
816
yield file_id, change[1][1], (entry.text_sha1, stat_value)
817
except errors.ExistingContent:
818
# No content change against a carry_over parent
819
# Perhaps this should also yield a fs hash update?
820
carried_over = True
821
entry.text_size = parent_entry.text_size
822
entry.text_sha1 = parent_entry.text_sha1
823
elif kind == 'symlink':
824
# Wants a path hint?
825
entry.symlink_target = tree.get_symlink_target(file_id)
826
if (carry_over_possible and
827
parent_entry.symlink_target == entry.symlink_target):
828
carried_over = True
829
else:
830
self._add_text_to_weave(change[0], '', heads, None)
831
elif kind == 'directory':
832
if carry_over_possible:
833
carried_over = True
834
else:
835
# Nothing to set on the entry.
836
# XXX: split into the Root and nonRoot versions.
837
if change[1][1] != '' or self.repository.supports_rich_root():
838
self._add_text_to_weave(change[0], '', heads, None)
839
elif kind == 'tree-reference':
840
if not self.repository._format.supports_tree_reference:
841
# This isn't quite sane as an error, but we shouldn't
842
# ever see this code path in practice: tree's don't
843
# permit references when the repo doesn't support tree
844
# references.
845
raise errors.UnsupportedOperation(tree.add_reference,
846
self.repository)
847
reference_revision = tree.get_reference_revision(change[0])
848
entry.reference_revision = reference_revision
849
if (carry_over_possible and
850
parent_entry.reference_revision == reference_revision):
851
carried_over = True
852
else:
853
self._add_text_to_weave(change[0], '', heads, None)
854
else:
855
raise AssertionError('unknown kind %r' % kind)
856
if not carried_over:
857
entry.revision = modified_rev
858
else:
859
entry.revision = parent_entry.revision
860
else:
861
entry = None
862
new_path = change[1][1]
863
inv_delta.append((change[1][0], new_path, change[0], entry))
864
if new_path == '':
865
seen_root = True
866
self.new_inventory = None
867
if len(inv_delta):
868
# This should perhaps be guarded by a check that the basis we
869
# commit against is the basis for the commit and if not do a delta
870
# against the basis.
871
self._any_changes = True
872
if not seen_root:
873
# housekeeping root entry changes do not affect no-change commits.
874
self._require_root_change(tree)
875
self.basis_delta_revision = basis_revision_id
876
877
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
878
parent_keys = tuple([(file_id, parent) for parent in parents])
879
return self.repository.texts._add_text(
880
(file_id, self._new_revision_id), parent_keys, new_text,
881
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
882
883
884
class RootCommitBuilder(CommitBuilder):
885
"""This commitbuilder actually records the root id"""
886
887
# the root entry gets versioned properly by this builder.
888
_versioned_root = True
889
890
def _check_root(self, ie, parent_invs, tree):
891
"""Helper for record_entry_contents.
892
893
:param ie: An entry being added.
894
:param parent_invs: The inventories of the parent revisions of the
895
commit.
896
:param tree: The tree that is being committed.
897
"""
898
899
def _require_root_change(self, tree):
900
"""Enforce an appropriate root object change.
901
902
This is called once when record_iter_changes is called, if and only if
903
the root was not in the delta calculated by record_iter_changes.
904
905
:param tree: The tree which is being committed.
906
"""
907
# versioned roots do not change unless the tree found a change.
220
raise NotImplementedError(self.record_iter_changes)
908
221
909
222
910
223
class RepositoryWriteLockResult(LogicalLockResult):
935
248
revisions and file history. It's normally accessed only by the Branch,
936
249
which views a particular line of development through that history.
937
250
938
The Repository builds on top of some byte storage facilies (the revisions,
939
signatures, inventories, texts and chk_bytes attributes) and a Transport,
940
which respectively provide byte storage and a means to access the (possibly
941
remote) disk.
942
943
The byte storage facilities are addressed via tuples, which we refer to
944
as 'keys' throughout the code base. Revision_keys, inventory_keys and
945
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
946
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
947
byte string made up of a hash identifier and a hash value.
948
We use this interface because it allows low friction with the underlying
949
code that implements disk indices, network encoding and other parts of
950
bzrlib.
951
952
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
953
the serialised revisions for the repository. This can be used to obtain
954
revision graph information or to access raw serialised revisions.
955
The result of trying to insert data into the repository via this store
956
is undefined: it should be considered read-only except for implementors
957
of repositories.
958
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
959
the serialised signatures for the repository. This can be used to
960
obtain access to raw serialised signatures. The result of trying to
961
insert data into the repository via this store is undefined: it should
962
be considered read-only except for implementors of repositories.
963
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
964
the serialised inventories for the repository. This can be used to
965
obtain unserialised inventories. The result of trying to insert data
966
into the repository via this store is undefined: it should be
967
considered read-only except for implementors of repositories.
968
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
969
texts of files and directories for the repository. This can be used to
970
obtain file texts or file graphs. Note that Repository.iter_file_bytes
971
is usually a better interface for accessing file texts.
972
The result of trying to insert data into the repository via this store
973
is undefined: it should be considered read-only except for implementors
974
of repositories.
975
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
976
any data the repository chooses to store or have indexed by its hash.
977
The result of trying to insert data into the repository via this store
978
is undefined: it should be considered read-only except for implementors
979
of repositories.
980
:ivar _transport: Transport for file access to repository, typically
981
pointing to .bzr/repository.
251
See VersionedFileRepository in bzrlib.vf_repository for the
252
base class for most Bazaar repositories.
982
253
"""
983
254
984
# What class to use for a CommitBuilder. Often it's simpler to change this
985
# in a Repository class subclass rather than to override
986
# get_commit_builder.
987
_commit_builder_class = CommitBuilder
988
989
255
def abort_write_group(self, suppress_errors=False):
990
256
"""Commit the contents accrued within the current write group.
991
257
1034
300
1035
301
:param repository: A repository.
1036
302
"""
1037
if not self._format.supports_external_lookups:
1038
raise errors.UnstackableRepositoryFormat(self._format, self.base)
1039
if self.is_locked():
1040
# This repository will call fallback.unlock() when we transition to
1041
# the unlocked state, so we make sure to increment the lock count
1042
repository.lock_read()
1043
self._check_fallback_repository(repository)
1044
self._fallback_repositories.append(repository)
1045
self.texts.add_fallback_versioned_files(repository.texts)
1046
self.inventories.add_fallback_versioned_files(repository.inventories)
1047
self.revisions.add_fallback_versioned_files(repository.revisions)
1048
self.signatures.add_fallback_versioned_files(repository.signatures)
1049
if self.chk_bytes is not None:
1050
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
303
raise NotImplementedError(self.add_fallback_repository)
1051
304
1052
305
def _check_fallback_repository(self, repository):
1053
306
"""Check that this repository can fallback to repository safely.
1058
311
"""
1059
312
return InterRepository._assert_same_model(self, repository)
1060
313
1061
def add_inventory(self, revision_id, inv, parents):
1062
"""Add the inventory inv to the repository as revision_id.
1063
1064
:param parents: The revision ids of the parents that revision_id
1065
is known to have and are in the repository already.
1066
1067
:returns: The validator(which is a sha1 digest, though what is sha'd is
1068
repository format specific) of the serialized inventory.
1069
"""
1070
if not self.is_in_write_group():
1071
raise AssertionError("%r not in write group" % (self,))
1072
_mod_revision.check_not_reserved_id(revision_id)
1073
if not (inv.revision_id is None or inv.revision_id == revision_id):
1074
raise AssertionError(
1075
"Mismatch between inventory revision"
1076
" id and insertion revid (%r, %r)"
1077
% (inv.revision_id, revision_id))
1078
if inv.root is None:
1079
raise errors.RootMissing()
1080
return self._add_inventory_checked(revision_id, inv, parents)
1081
1082
def _add_inventory_checked(self, revision_id, inv, parents):
1083
"""Add inv to the repository after checking the inputs.
1084
1085
This function can be overridden to allow different inventory styles.
1086
1087
:seealso: add_inventory, for the contract.
1088
"""
1089
inv_lines = self._serializer.write_inventory_to_lines(inv)
1090
return self._inventory_add_lines(revision_id, parents,
1091
inv_lines, check_content=False)
1092
1093
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1094
parents, basis_inv=None, propagate_caches=False):
1095
"""Add a new inventory expressed as a delta against another revision.
1096
1097
See the inventory developers documentation for the theory behind
1098
inventory deltas.
1099
1100
:param basis_revision_id: The inventory id the delta was created
1101
against. (This does not have to be a direct parent.)
1102
:param delta: The inventory delta (see Inventory.apply_delta for
1103
details).
1104
:param new_revision_id: The revision id that the inventory is being
1105
added for.
1106
:param parents: The revision ids of the parents that revision_id is
1107
known to have and are in the repository already. These are supplied
1108
for repositories that depend on the inventory graph for revision
1109
graph access, as well as for those that pun ancestry with delta
1110
compression.
1111
:param basis_inv: The basis inventory if it is already known,
1112
otherwise None.
1113
:param propagate_caches: If True, the caches for this inventory are
1114
copied to and updated for the result if possible.
1115
1116
:returns: (validator, new_inv)
1117
The validator(which is a sha1 digest, though what is sha'd is
1118
repository format specific) of the serialized inventory, and the
1119
resulting inventory.
1120
"""
1121
if not self.is_in_write_group():
1122
raise AssertionError("%r not in write group" % (self,))
1123
_mod_revision.check_not_reserved_id(new_revision_id)
1124
basis_tree = self.revision_tree(basis_revision_id)
1125
basis_tree.lock_read()
1126
try:
1127
# Note that this mutates the inventory of basis_tree, which not all
1128
# inventory implementations may support: A better idiom would be to
1129
# return a new inventory, but as there is no revision tree cache in
1130
# repository this is safe for now - RBC 20081013
1131
if basis_inv is None:
1132
basis_inv = basis_tree.inventory
1133
basis_inv.apply_delta(delta)
1134
basis_inv.revision_id = new_revision_id
1135
return (self.add_inventory(new_revision_id, basis_inv, parents),
1136
basis_inv)
1137
finally:
1138
basis_tree.unlock()
1139
1140
def _inventory_add_lines(self, revision_id, parents, lines,
1141
check_content=True):
1142
"""Store lines in inv_vf and return the sha1 of the inventory."""
1143
parents = [(parent,) for parent in parents]
1144
result = self.inventories.add_lines((revision_id,), parents, lines,
1145
check_content=check_content)[0]
1146
self.inventories._access.flush()
1147
return result
1148
1149
def add_revision(self, revision_id, rev, inv=None, config=None):
1150
"""Add rev to the revision store as revision_id.
1151
1152
:param revision_id: the revision id to use.
1153
:param rev: The revision object.
1154
:param inv: The inventory for the revision. if None, it will be looked
1155
up in the inventory storer
1156
:param config: If None no digital signature will be created.
1157
If supplied its signature_needed method will be used
1158
to determine if a signature should be made.
1159
"""
1160
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1161
# rev.parent_ids?
1162
_mod_revision.check_not_reserved_id(revision_id)
1163
if config is not None and config.signature_needed():
1164
if inv is None:
1165
inv = self.get_inventory(revision_id)
1166
tree = InventoryRevisionTree(self, inv, revision_id)
1167
testament = Testament(rev, tree)
1168
plaintext = testament.as_short_text()
1169
self.store_revision_signature(
1170
gpg.GPGStrategy(config), plaintext, revision_id)
1171
# check inventory present
1172
if not self.inventories.get_parent_map([(revision_id,)]):
1173
if inv is None:
1174
raise errors.WeaveRevisionNotPresent(revision_id,
1175
self.inventories)
1176
else:
1177
# yes, this is not suitable for adding with ghosts.
1178
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1179
rev.parent_ids)
1180
else:
1181
key = (revision_id,)
1182
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1183
self._add_revision(rev)
1184
1185
def _add_revision(self, revision):
1186
text = self._serializer.write_revision_to_string(revision)
1187
key = (revision.revision_id,)
1188
parents = tuple((parent,) for parent in revision.parent_ids)
1189
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1190
1191
314
def all_revision_ids(self):
1192
315
"""Returns a list of all the revision ids in the repository.
1193
316
1228
351
# The old API returned a list, should this actually be a set?
1229
352
return parent_map.keys()
1230
353
1231
def _check_inventories(self, checker):
1232
"""Check the inventories found from the revision scan.
1233
1234
This is responsible for verifying the sha1 of inventories and
1235
creating a pending_keys set that covers data referenced by inventories.
1236
"""
1237
bar = ui.ui_factory.nested_progress_bar()
1238
try:
1239
self._do_check_inventories(checker, bar)
1240
finally:
1241
bar.finished()
1242
1243
def _do_check_inventories(self, checker, bar):
1244
"""Helper for _check_inventories."""
1245
revno = 0
1246
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1247
kinds = ['chk_bytes', 'texts']
1248
count = len(checker.pending_keys)
1249
bar.update("inventories", 0, 2)
1250
current_keys = checker.pending_keys
1251
checker.pending_keys = {}
1252
# Accumulate current checks.
1253
for key in current_keys:
1254
if key[0] != 'inventories' and key[0] not in kinds:
1255
checker._report_items.append('unknown key type %r' % (key,))
1256
keys[key[0]].add(key[1:])
1257
if keys['inventories']:
1258
# NB: output order *should* be roughly sorted - topo or
1259
# inverse topo depending on repository - either way decent
1260
# to just delta against. However, pre-CHK formats didn't
1261
# try to optimise inventory layout on disk. As such the
1262
# pre-CHK code path does not use inventory deltas.
1263
last_object = None
1264
for record in self.inventories.check(keys=keys['inventories']):
1265
if record.storage_kind == 'absent':
1266
checker._report_items.append(
1267
'Missing inventory {%s}' % (record.key,))
1268
else:
1269
last_object = self._check_record('inventories', record,
1270
checker, last_object,
1271
current_keys[('inventories',) + record.key])
1272
del keys['inventories']
1273
else:
1274
return
1275
bar.update("texts", 1)
1276
while (checker.pending_keys or keys['chk_bytes']
1277
or keys['texts']):
1278
# Something to check.
1279
current_keys = checker.pending_keys
1280
checker.pending_keys = {}
1281
# Accumulate current checks.
1282
for key in current_keys:
1283
if key[0] not in kinds:
1284
checker._report_items.append('unknown key type %r' % (key,))
1285
keys[key[0]].add(key[1:])
1286
# Check the outermost kind only - inventories || chk_bytes || texts
1287
for kind in kinds:
1288
if keys[kind]:
1289
last_object = None
1290
for record in getattr(self, kind).check(keys=keys[kind]):
1291
if record.storage_kind == 'absent':
1292
checker._report_items.append(
1293
'Missing %s {%s}' % (kind, record.key,))
1294
else:
1295
last_object = self._check_record(kind, record,
1296
checker, last_object, current_keys[(kind,) + record.key])
1297
keys[kind] = set()
1298
break
1299
1300
def _check_record(self, kind, record, checker, last_object, item_data):
1301
"""Check a single text from this repository."""
1302
if kind == 'inventories':
1303
rev_id = record.key[0]
1304
inv = self._deserialise_inventory(rev_id,
1305
record.get_bytes_as('fulltext'))
1306
if last_object is not None:
1307
delta = inv._make_delta(last_object)
1308
for old_path, path, file_id, ie in delta:
1309
if ie is None:
1310
continue
1311
ie.check(checker, rev_id, inv)
1312
else:
1313
for path, ie in inv.iter_entries():
1314
ie.check(checker, rev_id, inv)
1315
if self._format.fast_deltas:
1316
return inv
1317
elif kind == 'chk_bytes':
1318
# No code written to check chk_bytes for this repo format.
1319
checker._report_items.append(
1320
'unsupported key type chk_bytes for %s' % (record.key,))
1321
elif kind == 'texts':
1322
self._check_text(record, checker, item_data)
1323
else:
1324
checker._report_items.append(
1325
'unknown key type %s for %s' % (kind, record.key))
1326
1327
def _check_text(self, record, checker, item_data):
1328
"""Check a single text."""
1329
# Check it is extractable.
1330
# TODO: check length.
1331
if record.storage_kind == 'chunked':
1332
chunks = record.get_bytes_as(record.storage_kind)
1333
sha1 = osutils.sha_strings(chunks)
1334
length = sum(map(len, chunks))
1335
else:
1336
content = record.get_bytes_as('fulltext')
1337
sha1 = osutils.sha_string(content)
1338
length = len(content)
1339
if item_data and sha1 != item_data[1]:
1340
checker._report_items.append(
1341
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1342
(record.key, sha1, item_data[1], item_data[2]))
1343
1344
354
@staticmethod
1345
355
def create(a_bzrdir):
1346
356
"""Construct the current default format repository in a_bzrdir."""
1351
361
1352
362
:param _format: The format of the repository on disk.
1353
363
:param a_bzrdir: The BzrDir of the repository.
364
:param control_files: Control files to use for locking, etc.
1354
365
"""
1355
366
# In the future we will have a single api for all stores for
1356
367
# getting file texts, inventories and revisions, then
1363
374
self._transport = control_files._transport
1364
375
self.base = self._transport.base
1365
376
# for tests
1366
self._reconcile_does_inventory_gc = True
1367
self._reconcile_fixes_text_parents = False
1368
self._reconcile_backsup_inventory = True
1369
377
self._write_group = None
1370
378
# Additional places to query for data.
1371
379
self._fallback_repositories = []
1372
# An InventoryEntry cache, used during deserialization
1373
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1374
# Is it safe to return inventory entries directly from the entry cache,
1375
# rather copying them?
1376
self._safe_to_return_from_cache = False
1377
380
1378
381
@property
1379
382
def user_transport(self):
1540
543
first_revision.timezone)
1541
544
result['latestrev'] = (last_revision.timestamp,
1542
545
last_revision.timezone)
1543
1544
# now gather global repository information
1545
# XXX: This is available for many repos regardless of listability.
1546
if self.user_transport.listable():
1547
# XXX: do we want to __define len__() ?
1548
# Maybe the versionedfiles object should provide a different
1549
# method to get the number of keys.
1550
result['revisions'] = len(self.revisions.keys())
1551
# result['size'] = t
1552
546
return result
1553
547
1554
548
def find_branches(self, using=False):
1658
652
def suspend_write_group(self):
1659
653
raise errors.UnsuspendableWriteGroup(self)
1660
654
1661
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1662
"""Return the keys of missing inventory parents for revisions added in
1663
this write group.
1664
1665
A revision is not complete if the inventory delta for that revision
1666
cannot be calculated. Therefore if the parent inventories of a
1667
revision are not present, the revision is incomplete, and e.g. cannot
1668
be streamed by a smart server. This method finds missing inventory
1669
parents for revisions added in this write group.
1670
"""
1671
if not self._format.supports_external_lookups:
1672
# This is only an issue for stacked repositories
1673
return set()
1674
if not self.is_in_write_group():
1675
raise AssertionError('not in a write group')
1676
1677
# XXX: We assume that every added revision already has its
1678
# corresponding inventory, so we only check for parent inventories that
1679
# might be missing, rather than all inventories.
1680
parents = set(self.revisions._index.get_missing_parents())
1681
parents.discard(_mod_revision.NULL_REVISION)
1682
unstacked_inventories = self.inventories._index
1683
present_inventories = unstacked_inventories.get_parent_map(
1684
key[-1:] for key in parents)
1685
parents.difference_update(present_inventories)
1686
if len(parents) == 0:
1687
# No missing parent inventories.
1688
return set()
1689
if not check_for_missing_texts:
1690
return set(('inventories', rev_id) for (rev_id,) in parents)
1691
# Ok, now we have a list of missing inventories. But these only matter
1692
# if the inventories that reference them are missing some texts they
1693
# appear to introduce.
1694
# XXX: Texts referenced by all added inventories need to be present,
1695
# but at the moment we're only checking for texts referenced by
1696
# inventories at the graph's edge.
1697
key_deps = self.revisions._index._key_dependencies
1698
key_deps.satisfy_refs_for_keys(present_inventories)
1699
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1700
file_ids = self.fileids_altered_by_revision_ids(referrers)
1701
missing_texts = set()
1702
for file_id, version_ids in file_ids.iteritems():
1703
missing_texts.update(
1704
(file_id, version_id) for version_id in version_ids)
1705
present_texts = self.texts.get_parent_map(missing_texts)
1706
missing_texts.difference_update(present_texts)
1707
if not missing_texts:
1708
# No texts are missing, so all revisions and their deltas are
1709
# reconstructable.
1710
return set()
1711
# Alternatively the text versions could be returned as the missing
1712
# keys, but this is likely to be less data.
1713
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1714
return missing_keys
1715
1716
655
def refresh_data(self):
1717
656
"""Re-read any data needed to synchronise with disk.
1718
657
1801
740
:param lossy: Whether to discard data that can not be natively
1802
741
represented, when pushing to a foreign VCS
1803
742
"""
1804
if self._fallback_repositories and not self._format.supports_chks:
1805
raise errors.BzrError("Cannot commit directly to a stacked branch"
1806
" in pre-2a formats. See "
1807
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1808
result = self._commit_builder_class(self, parents, config,
1809
timestamp, timezone, committer, revprops, revision_id,
1810
lossy)
1811
self.start_write_group()
1812
return result
743
raise NotImplementedError(self.get_commit_builder)
1813
744
1814
745
@only_raises(errors.LockNotHeld, errors.LockBroken)
1815
746
def unlock(self):
1822
753
'Must end write groups before releasing write locks.')
1823
754
self.control_files.unlock()
1824
755
if self.control_files._lock_count == 0:
1825
self._inventory_entry_cache.clear()
1826
756
for repo in self._fallback_repositories:
1827
757
repo.unlock()
1828
758
1893
823
dest_repo = a_bzrdir.open_repository()
1894
824
return dest_repo
1895
825
1896
def _get_sink(self):
1897
"""Return a sink for streaming into this repository."""
1898
return StreamSink(self)
1899
1900
def _get_source(self, to_format):
1901
"""Return a source for streaming from this repository."""
1902
return StreamSource(self, to_format)
1903
1904
826
@needs_read_lock
1905
827
def has_revision(self, revision_id):
1906
828
"""True if this repository has a copy of the revision."""
1913
835
:param revision_ids: An iterable of revision_ids.
1914
836
:return: A set of the revision_ids that were present.
1915
837
"""
1916
parent_map = self.revisions.get_parent_map(
1917
[(rev_id,) for rev_id in revision_ids])
1918
result = set()
1919
if _mod_revision.NULL_REVISION in revision_ids:
1920
result.add(_mod_revision.NULL_REVISION)
1921
result.update([key[0] for key in parent_map])
1922
return result
838
raise NotImplementedError(self.has_revisions)
1923
839
1924
840
@needs_read_lock
1925
841
def get_revision(self, revision_id):
1926
842
"""Return the Revision object for a named revision."""
1927
843
return self.get_revisions([revision_id])[0]
1928
844
1929
@needs_read_lock
1930
845
def get_revision_reconcile(self, revision_id):
1931
846
"""'reconcile' helper routine that allows access to a revision always.
1932
847
1935
850
be used by reconcile, or reconcile-alike commands that are correcting
1936
851
or testing the revision graph.
1937
852
"""
1938
return self._get_revisions([revision_id])[0]
853
raise NotImplementedError(self.get_revision_reconcile)
1939
854
1940
@needs_read_lock
1941
855
def get_revisions(self, revision_ids):
1942
856
"""Get many revisions at once.
1943
857
1944
858
Repositories that need to check data on every revision read should
1945
859
subclass this method.
1946
860
"""
1947
return self._get_revisions(revision_ids)
1948
1949
@needs_read_lock
1950
def _get_revisions(self, revision_ids):
1951
"""Core work logic to get many revisions without sanity checks."""
1952
revs = {}
1953
for revid, rev in self._iter_revisions(revision_ids):
1954
if rev is None:
1955
raise errors.NoSuchRevision(self, revid)
1956
revs[revid] = rev
1957
return [revs[revid] for revid in revision_ids]
1958
1959
def _iter_revisions(self, revision_ids):
1960
"""Iterate over revision objects.
1961
1962
:param revision_ids: An iterable of revisions to examine. None may be
1963
passed to request all revisions known to the repository. Note that
1964
not all repositories can find unreferenced revisions; for those
1965
repositories only referenced ones will be returned.
1966
:return: An iterator of (revid, revision) tuples. Absent revisions (
1967
those asked for but not available) are returned as (revid, None).
1968
"""
1969
if revision_ids is None:
1970
revision_ids = self.all_revision_ids()
1971
else:
1972
for rev_id in revision_ids:
1973
if not rev_id or not isinstance(rev_id, basestring):
1974
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1975
keys = [(key,) for key in revision_ids]
1976
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1977
for record in stream:
1978
revid = record.key[0]
1979
if record.storage_kind == 'absent':
1980
yield (revid, None)
1981
else:
1982
text = record.get_bytes_as('fulltext')
1983
rev = self._serializer.read_revision_from_string(text)
1984
yield (revid, rev)
861
raise NotImplementedError(self.get_revisions)
1985
862
1986
863
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1987
864
"""Produce a generator of revision deltas.
2042
919
signature = gpg_strategy.sign(plaintext)
2043
920
self.add_signature_text(revision_id, signature)
2044
921
2045
@needs_write_lock
2046
922
def add_signature_text(self, revision_id, signature):
2047
self.signatures.add_lines((revision_id,), (),
2048
osutils.split_lines(signature))
923
"""Store a signature text for a revision.
924
925
:param revision_id: Revision id of the revision
926
:param signature: Signature text.
927
"""
928
raise NotImplementedError(self.add_signature_text)
2049
929
2050
930
def find_text_key_references(self):
2051
931
"""Find the text key references within the repository.
2052
932
2053
933
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2054
to whether they were referred to by the inventory of the
2055
revision_id that they contain. The inventory texts from all present
2056
revision ids are assessed to generate this report.
2057
"""
2058
revision_keys = self.revisions.keys()
2059
w = self.inventories
2060
pb = ui.ui_factory.nested_progress_bar()
2061
try:
2062
return self._serializer._find_text_key_references(
2063
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
2064
finally:
2065
pb.finished()
2066
2067
def _inventory_xml_lines_for_keys(self, keys):
2068
"""Get a line iterator of the sort needed for findind references.
2069
2070
Not relevant for non-xml inventory repositories.
2071
2072
Ghosts in revision_keys are ignored.
2073
2074
:param revision_keys: The revision keys for the inventories to inspect.
2075
:return: An iterator over (inventory line, revid) for the fulltexts of
2076
all of the xml inventories specified by revision_keys.
2077
"""
2078
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2079
for record in stream:
2080
if record.storage_kind != 'absent':
2081
chunks = record.get_bytes_as('chunked')
2082
revid = record.key[-1]
2083
lines = osutils.chunks_to_lines(chunks)
2084
for line in lines:
2085
yield line, revid
2086
2087
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2088
revision_keys):
2089
"""Helper routine for fileids_altered_by_revision_ids.
2090
2091
This performs the translation of xml lines to revision ids.
2092
2093
:param line_iterator: An iterator of lines, origin_version_id
2094
:param revision_keys: The revision ids to filter for. This should be a
2095
set or other type which supports efficient __contains__ lookups, as
2096
the revision key from each parsed line will be looked up in the
2097
revision_keys filter.
2098
:return: a dictionary mapping altered file-ids to an iterable of
2099
revision_ids. Each altered file-ids has the exact revision_ids that
2100
altered it listed explicitly.
2101
"""
2102
seen = set(self._serializer._find_text_key_references(
2103
line_iterator).iterkeys())
2104
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2105
parent_seen = set(self._serializer._find_text_key_references(
2106
self._inventory_xml_lines_for_keys(parent_keys)))
2107
new_keys = seen - parent_seen
2108
result = {}
2109
setdefault = result.setdefault
2110
for key in new_keys:
2111
setdefault(key[0], set()).add(key[-1])
2112
return result
934
to whether they were referred to by the tree of the
935
revision_id that they contain.
936
"""
937
raise NotImplementedError(self.find_text_key_references)
2113
938
2114
939
def _find_parent_ids_of_revisions(self, revision_ids):
2115
940
"""Find all parent ids that are mentioned in the revision graph.
2124
949
parent_ids.discard(_mod_revision.NULL_REVISION)
2125
950
return parent_ids
2126
951
2127
def _find_parent_keys_of_revisions(self, revision_keys):
2128
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2129
2130
:param revision_keys: An iterable of revision_keys.
2131
:return: The parents of all revision_keys that are not already in
2132
revision_keys
2133
"""
2134
parent_map = self.revisions.get_parent_map(revision_keys)
2135
parent_keys = set()
2136
map(parent_keys.update, parent_map.itervalues())
2137
parent_keys.difference_update(revision_keys)
2138
parent_keys.discard(_mod_revision.NULL_REVISION)
2139
return parent_keys
2140
2141
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
952
def fileids_altered_by_revision_ids(self, revision_ids):
2142
953
"""Find the file ids and versions affected by revisions.
2143
954
2144
955
:param revisions: an iterable containing revision ids.
2145
:param _inv_weave: The inventory weave from this repository or None.
2146
If None, the inventory weave will be opened automatically.
2147
956
:return: a dictionary mapping altered file-ids to an iterable of
2148
revision_ids. Each altered file-ids has the exact revision_ids that
2149
altered it listed explicitly.
957
revision_ids. Each altered file-ids has the exact revision_ids
958
that altered it listed explicitly.
2150
959
"""
2151
selected_keys = set((revid,) for revid in revision_ids)
2152
w = _inv_weave or self.inventories
2153
return self._find_file_ids_from_xml_inventory_lines(
2154
w.iter_lines_added_or_present_in_keys(
2155
selected_keys, pb=None),
2156
selected_keys)
960
raise NotImplementedError(self.fileids_altered_by_revision_ids)
2157
961
2158
962
def iter_files_bytes(self, desired_files):
2159
963
"""Iterate through file versions.
2166
970
uniquely identify the file version in the caller's context. (Examples:
2167
971
an index number or a TreeTransform trans_id.)
2168
972
2169
bytes_iterator is an iterable of bytestrings for the file. The
2170
kind of iterable and length of the bytestrings are unspecified, but for
2171
this implementation, it is a list of bytes produced by
2172
VersionedFile.get_record_stream().
2173
2174
973
:param desired_files: a list of (file_id, revision_id, identifier)
2175
974
triples
2176
975
"""
2177
text_keys = {}
2178
for file_id, revision_id, callable_data in desired_files:
2179
text_keys[(file_id, revision_id)] = callable_data
2180
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2181
if record.storage_kind == 'absent':
2182
raise errors.RevisionNotPresent(record.key, self)
2183
yield text_keys[record.key], record.get_bytes_as('chunked')
2184
2185
def _generate_text_key_index(self, text_key_references=None,
2186
ancestors=None):
2187
"""Generate a new text key index for the repository.
2188
2189
This is an expensive function that will take considerable time to run.
2190
2191
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2192
list of parents, also text keys. When a given key has no parents,
2193
the parents list will be [NULL_REVISION].
2194
"""
2195
# All revisions, to find inventory parents.
2196
if ancestors is None:
2197
graph = self.get_graph()
2198
ancestors = graph.get_parent_map(self.all_revision_ids())
2199
if text_key_references is None:
2200
text_key_references = self.find_text_key_references()
2201
pb = ui.ui_factory.nested_progress_bar()
2202
try:
2203
return self._do_generate_text_key_index(ancestors,
2204
text_key_references, pb)
2205
finally:
2206
pb.finished()
2207
2208
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2209
"""Helper for _generate_text_key_index to avoid deep nesting."""
2210
revision_order = tsort.topo_sort(ancestors)
2211
invalid_keys = set()
2212
revision_keys = {}
2213
for revision_id in revision_order:
2214
revision_keys[revision_id] = set()
2215
text_count = len(text_key_references)
2216
# a cache of the text keys to allow reuse; costs a dict of all the
2217
# keys, but saves a 2-tuple for every child of a given key.
2218
text_key_cache = {}
2219
for text_key, valid in text_key_references.iteritems():
2220
if not valid:
2221
invalid_keys.add(text_key)
2222
else:
2223
revision_keys[text_key[1]].add(text_key)
2224
text_key_cache[text_key] = text_key
2225
del text_key_references
2226
text_index = {}
2227
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2228
NULL_REVISION = _mod_revision.NULL_REVISION
2229
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2230
# too small for large or very branchy trees. However, for 55K path
2231
# trees, it would be easy to use too much memory trivially. Ideally we
2232
# could gauge this by looking at available real memory etc, but this is
2233
# always a tricky proposition.
2234
inventory_cache = lru_cache.LRUCache(10)
2235
batch_size = 10 # should be ~150MB on a 55K path tree
2236
batch_count = len(revision_order) / batch_size + 1
2237
processed_texts = 0
2238
pb.update("Calculating text parents", processed_texts, text_count)
2239
for offset in xrange(batch_count):
2240
to_query = revision_order[offset * batch_size:(offset + 1) *
2241
batch_size]
2242
if not to_query:
2243
break
2244
for revision_id in to_query:
2245
parent_ids = ancestors[revision_id]
2246
for text_key in revision_keys[revision_id]:
2247
pb.update("Calculating text parents", processed_texts)
2248
processed_texts += 1
2249
candidate_parents = []
2250
for parent_id in parent_ids:
2251
parent_text_key = (text_key[0], parent_id)
2252
try:
2253
check_parent = parent_text_key not in \
2254
revision_keys[parent_id]
2255
except KeyError:
2256
# the parent parent_id is a ghost:
2257
check_parent = False
2258
# truncate the derived graph against this ghost.
2259
parent_text_key = None
2260
if check_parent:
2261
# look at the parent commit details inventories to
2262
# determine possible candidates in the per file graph.
2263
# TODO: cache here.
2264
try:
2265
inv = inventory_cache[parent_id]
2266
except KeyError:
2267
inv = self.revision_tree(parent_id).inventory
2268
inventory_cache[parent_id] = inv
2269
try:
2270
parent_entry = inv[text_key[0]]
2271
except (KeyError, errors.NoSuchId):
2272
parent_entry = None
2273
if parent_entry is not None:
2274
parent_text_key = (
2275
text_key[0], parent_entry.revision)
2276
else:
2277
parent_text_key = None
2278
if parent_text_key is not None:
2279
candidate_parents.append(
2280
text_key_cache[parent_text_key])
2281
parent_heads = text_graph.heads(candidate_parents)
2282
new_parents = list(parent_heads)
2283
new_parents.sort(key=lambda x:candidate_parents.index(x))
2284
if new_parents == []:
2285
new_parents = [NULL_REVISION]
2286
text_index[text_key] = new_parents
2287
2288
for text_key in invalid_keys:
2289
text_index[text_key] = [NULL_REVISION]
2290
return text_index
2291
2292
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2293
"""Get an iterable listing the keys of all the data introduced by a set
2294
of revision IDs.
2295
2296
The keys will be ordered so that the corresponding items can be safely
2297
fetched and inserted in that order.
2298
2299
:returns: An iterable producing tuples of (knit-kind, file-id,
2300
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2301
'revisions'. file-id is None unless knit-kind is 'file'.
2302
"""
2303
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2304
yield result
2305
del _files_pb
2306
for result in self._find_non_file_keys_to_fetch(revision_ids):
2307
yield result
2308
2309
def _find_file_keys_to_fetch(self, revision_ids, pb):
2310
# XXX: it's a bit weird to control the inventory weave caching in this
2311
# generator. Ideally the caching would be done in fetch.py I think. Or
2312
# maybe this generator should explicitly have the contract that it
2313
# should not be iterated until the previously yielded item has been
2314
# processed?
2315
inv_w = self.inventories
2316
2317
# file ids that changed
2318
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2319
count = 0
2320
num_file_ids = len(file_ids)
2321
for file_id, altered_versions in file_ids.iteritems():
2322
if pb is not None:
2323
pb.update("Fetch texts", count, num_file_ids)
2324
count += 1
2325
yield ("file", file_id, altered_versions)
2326
2327
def _find_non_file_keys_to_fetch(self, revision_ids):
2328
# inventory
2329
yield ("inventory", None, revision_ids)
2330
2331
# signatures
2332
# XXX: Note ATM no callers actually pay attention to this return
2333
# instead they just use the list of revision ids and ignore
2334
# missing sigs. Consider removing this work entirely
2335
revisions_with_signatures = set(self.signatures.get_parent_map(
2336
[(r,) for r in revision_ids]))
2337
revisions_with_signatures = set(
2338
[r for (r,) in revisions_with_signatures])
2339
revisions_with_signatures.intersection_update(revision_ids)
2340
yield ("signatures", None, revisions_with_signatures)
2341
2342
# revisions
2343
yield ("revisions", None, revision_ids)
2344
2345
@needs_read_lock
2346
def get_inventory(self, revision_id):
2347
"""Get Inventory object by revision id."""
2348
return self.iter_inventories([revision_id]).next()
2349
2350
def iter_inventories(self, revision_ids, ordering=None):
2351
"""Get many inventories by revision_ids.
2352
2353
This will buffer some or all of the texts used in constructing the
2354
inventories in memory, but will only parse a single inventory at a
2355
time.
2356
2357
:param revision_ids: The expected revision ids of the inventories.
2358
:param ordering: optional ordering, e.g. 'topological'. If not
2359
specified, the order of revision_ids will be preserved (by
2360
buffering if necessary).
2361
:return: An iterator of inventories.
2362
"""
2363
if ((None in revision_ids)
2364
or (_mod_revision.NULL_REVISION in revision_ids)):
2365
raise ValueError('cannot get null revision inventory')
2366
return self._iter_inventories(revision_ids, ordering)
2367
2368
def _iter_inventories(self, revision_ids, ordering):
2369
"""single-document based inventory iteration."""
2370
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2371
for text, revision_id in inv_xmls:
2372
yield self._deserialise_inventory(revision_id, text)
2373
2374
def _iter_inventory_xmls(self, revision_ids, ordering):
2375
if ordering is None:
2376
order_as_requested = True
2377
ordering = 'unordered'
2378
else:
2379
order_as_requested = False
2380
keys = [(revision_id,) for revision_id in revision_ids]
2381
if not keys:
2382
return
2383
if order_as_requested:
2384
key_iter = iter(keys)
2385
next_key = key_iter.next()
2386
stream = self.inventories.get_record_stream(keys, ordering, True)
2387
text_chunks = {}
2388
for record in stream:
2389
if record.storage_kind != 'absent':
2390
chunks = record.get_bytes_as('chunked')
2391
if order_as_requested:
2392
text_chunks[record.key] = chunks
2393
else:
2394
yield ''.join(chunks), record.key[-1]
2395
else:
2396
raise errors.NoSuchRevision(self, record.key)
2397
if order_as_requested:
2398
# Yield as many results as we can while preserving order.
2399
while next_key in text_chunks:
2400
chunks = text_chunks.pop(next_key)
2401
yield ''.join(chunks), next_key[-1]
2402
try:
2403
next_key = key_iter.next()
2404
except StopIteration:
2405
# We still want to fully consume the get_record_stream,
2406
# just in case it is not actually finished at this point
2407
next_key = None
2408
break
2409
2410
def _deserialise_inventory(self, revision_id, xml):
2411
"""Transform the xml into an inventory object.
2412
2413
:param revision_id: The expected revision id of the inventory.
2414
:param xml: A serialised inventory.
2415
"""
2416
result = self._serializer.read_inventory_from_string(xml, revision_id,
2417
entry_cache=self._inventory_entry_cache,
2418
return_from_cache=self._safe_to_return_from_cache)
2419
if result.revision_id != revision_id:
2420
raise AssertionError('revision id mismatch %s != %s' % (
2421
result.revision_id, revision_id))
2422
return result
2423
2424
def get_serializer_format(self):
2425
return self._serializer.format_num
2426
2427
@needs_read_lock
2428
def _get_inventory_xml(self, revision_id):
2429
"""Get serialized inventory as a string."""
2430
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2431
try:
2432
text, revision_id = texts.next()
2433
except StopIteration:
2434
raise errors.HistoryMissing(self, 'inventory', revision_id)
2435
return text
976
raise NotImplementedError(self.iter_files_bytes)
2436
977
2437
978
def get_rev_id_for_revno(self, revno, known_pair):
2438
979
"""Return the revision id of a revno, given a later (revno, revid)
2511
1052
2512
1053
`revision_id` may be NULL_REVISION for the empty tree revision.
2513
1054
"""
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 InventoryRevisionTree(self,
2519
Inventory(root_id=None), _mod_revision.NULL_REVISION)
2520
else:
2521
inv = self.get_inventory(revision_id)
2522
return InventoryRevisionTree(self, inv, revision_id)
1055
raise NotImplementedError(self.revision_tree)
2523
1056
2524
1057
def revision_trees(self, revision_ids):
2525
1058
"""Return Trees for revisions in this repository.
2527
1060
:param revision_ids: a sequence of revision-ids;
2528
1061
a revision-id may not be None or 'null:'
2529
1062
"""
2530
inventories = self.iter_inventories(revision_ids)
2531
for inv in inventories:
2532
yield InventoryRevisionTree(self, inv, inv.revision_id)
2533
2534
def _filtered_revision_trees(self, revision_ids, file_ids):
2535
"""Return Tree for a revision on this branch with only some files.
2536
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.
2542
"""
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 InventoryRevisionTree(self, filtered_inv, filtered_inv.revision_id)
1063
raise NotImplementedError(self.revision_trees)
2549
1064
2550
1065
@needs_read_lock
2551
1066
def get_ancestry(self, revision_id, topo_sorted=True):
2603
1118
2604
1119
def get_parent_map(self, revision_ids):
2605
1120
"""See graph.StackedParentsProvider.get_parent_map"""
2606
# revisions index works in keys; this just works in revisions
2607
# therefore wrap and unwrap
2608
query_keys = []
2609
result = {}
2610
for revision_id in revision_ids:
2611
if revision_id == _mod_revision.NULL_REVISION:
2612
result[revision_id] = ()
2613
elif revision_id is None:
2614
raise ValueError('get_parent_map(None) is not valid')
2615
else:
2616
query_keys.append((revision_id ,))
2617
for ((revision_id,), parent_keys) in \
2618
self.revisions.get_parent_map(query_keys).iteritems():
2619
if parent_keys:
2620
result[revision_id] = tuple([parent_revid
2621
for (parent_revid,) in parent_keys])
2622
else:
2623
result[revision_id] = (_mod_revision.NULL_REVISION,)
2624
return result
1121
raise NotImplementedError(self.get_parent_map)
2625
1122
2626
1123
def _get_parent_map_no_fallbacks(self, revision_ids):
2627
1124
"""Same as Repository.get_parent_map except doesn't query fallbacks."""
2661
1158
def get_known_graph_ancestry(self, revision_ids):
2662
1159
"""Return the known graph for a set of revision ids and their ancestors.
2663
1160
"""
2664
st = static_tuple.StaticTuple
2665
revision_keys = [st(r_id).intern() for r_id in revision_ids]
2666
known_graph = self.revisions.get_known_graph_ancestry(revision_keys)
2667
return graph.GraphThunkIdsToKeys(known_graph)
1161
raise NotImplementedError(self.get_known_graph_ancestry)
2668
1162
2669
1163
def get_graph(self, other_repository=None):
2670
1164
"""Return the graph walker for this repository format"""
2675
1169
[parents_provider, other_repository._make_parents_provider()])
2676
1170
return graph.Graph(parents_provider)
2677
1171
2678
def _get_versioned_file_checker(self, text_key_references=None,
2679
ancestors=None):
2680
"""Return an object suitable for checking versioned files.
2681
2682
:param text_key_references: if non-None, an already built
2683
dictionary mapping text keys ((fileid, revision_id) tuples)
2684
to whether they were referred to by the inventory of the
2685
revision_id that they contain. If None, this will be
2686
calculated.
2687
:param ancestors: Optional result from
2688
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2689
available.
2690
"""
2691
return _VersionedFileChecker(self,
2692
text_key_references=text_key_references, ancestors=ancestors)
2693
2694
1172
def revision_ids_to_search_result(self, result_set):
2695
1173
"""Convert a set of revision ids to a graph SearchResult."""
2696
1174
result_parents = set()
2722
1200
2723
1201
@needs_write_lock
2724
1202
def sign_revision(self, revision_id, gpg_strategy):
2725
plaintext = Testament.from_revision(self, revision_id).as_short_text()
1203
testament = _mod_testament.Testament.from_revision(self, revision_id)
1204
plaintext = testament.as_short_text()
2726
1205
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2727
1206
2728
@needs_read_lock
2729
1207
def has_signature_for_revision_id(self, revision_id):
2730
1208
"""Query for a revision signature for revision_id in the repository."""
2731
if not self.has_revision(revision_id):
2732
raise errors.NoSuchRevision(self, revision_id)
2733
sig_present = (1 == len(
2734
self.signatures.get_parent_map([(revision_id,)])))
2735
return sig_present
1209
raise NotImplementedError(self.has_signature_for_revision_id)
2736
1210
2737
@needs_read_lock
2738
1211
def get_signature_text(self, revision_id):
2739
1212
"""Return the text for a signature."""
2740
stream = self.signatures.get_record_stream([(revision_id,)],
2741
'unordered', True)
2742
record = stream.next()
2743
if record.storage_kind == 'absent':
2744
raise errors.NoSuchRevision(self, revision_id)
2745
return record.get_bytes_as('fulltext')
1213
raise NotImplementedError(self.get_signature_text)
2746
1214
2747
1215
@needs_read_lock
2748
1216
def check(self, revision_ids=None, callback_refs=None, check_repo=True):
2804
1272
except UnicodeDecodeError:
2805
1273
raise errors.NonAsciiRevisionId(method, self)
2806
1274
2807
def _find_inconsistent_revision_parents(self, revisions_iterator=None):
2808
"""Find revisions with different parent lists in the revision object
2809
and in the index graph.
2810
2811
:param revisions_iterator: None, or an iterator of (revid,
2812
Revision-or-None). This iterator controls the revisions checked.
2813
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
2814
parents-in-revision).
2815
"""
2816
if not self.is_locked():
2817
raise AssertionError()
2818
vf = self.revisions
2819
if revisions_iterator is None:
2820
revisions_iterator = self._iter_revisions(None)
2821
for revid, revision in revisions_iterator:
2822
if revision is None:
2823
pass
2824
parent_map = vf.get_parent_map([(revid,)])
2825
parents_according_to_index = tuple(parent[-1] for parent in
2826
parent_map[(revid,)])
2827
parents_according_to_revision = tuple(revision.parent_ids)
2828
if parents_according_to_index != parents_according_to_revision:
2829
yield (revid, parents_according_to_index,
2830
parents_according_to_revision)
2831
2832
def _check_for_inconsistent_revision_parents(self):
2833
inconsistencies = list(self._find_inconsistent_revision_parents())
2834
if inconsistencies:
2835
raise errors.BzrCheckError(
2836
"Revision knit has inconsistent parents.")
2837
2838
2839
def install_revision(repository, rev, revision_tree):
2840
"""Install all revision data into a repository."""
2841
install_revisions(repository, [(rev, revision_tree, None)])
2842
2843
2844
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2845
"""Install all revision data into a repository.
2846
2847
Accepts an iterable of revision, tree, signature tuples. The signature
2848
may be None.
2849
"""
2850
repository.start_write_group()
2851
try:
2852
inventory_cache = lru_cache.LRUCache(10)
2853
for n, (revision, revision_tree, signature) in enumerate(iterable):
2854
_install_revision(repository, revision, revision_tree, signature,
2855
inventory_cache)
2856
if pb is not None:
2857
pb.update('Transferring revisions', n + 1, num_revisions)
2858
except:
2859
repository.abort_write_group()
2860
raise
2861
else:
2862
repository.commit_write_group()
2863
2864
2865
def _install_revision(repository, rev, revision_tree, signature,
2866
inventory_cache):
2867
"""Install all revision data into a repository."""
2868
present_parents = []
2869
parent_trees = {}
2870
for p_id in rev.parent_ids:
2871
if repository.has_revision(p_id):
2872
present_parents.append(p_id)
2873
parent_trees[p_id] = repository.revision_tree(p_id)
2874
else:
2875
parent_trees[p_id] = repository.revision_tree(
2876
_mod_revision.NULL_REVISION)
2877
2878
inv = revision_tree.inventory
2879
entries = inv.iter_entries()
2880
# backwards compatibility hack: skip the root id.
2881
if not repository.supports_rich_root():
2882
path, root = entries.next()
2883
if root.revision != rev.revision_id:
2884
raise errors.IncompatibleRevision(repr(repository))
2885
text_keys = {}
2886
for path, ie in entries:
2887
text_keys[(ie.file_id, ie.revision)] = ie
2888
text_parent_map = repository.texts.get_parent_map(text_keys)
2889
missing_texts = set(text_keys) - set(text_parent_map)
2890
# Add the texts that are not already present
2891
for text_key in missing_texts:
2892
ie = text_keys[text_key]
2893
text_parents = []
2894
# FIXME: TODO: The following loop overlaps/duplicates that done by
2895
# commit to determine parents. There is a latent/real bug here where
2896
# the parents inserted are not those commit would do - in particular
2897
# they are not filtered by heads(). RBC, AB
2898
for revision, tree in parent_trees.iteritems():
2899
if ie.file_id not in tree:
2900
continue
2901
parent_id = tree.get_file_revision(ie.file_id)
2902
if parent_id in text_parents:
2903
continue
2904
text_parents.append((ie.file_id, parent_id))
2905
lines = revision_tree.get_file(ie.file_id).readlines()
2906
repository.texts.add_lines(text_key, text_parents, lines)
2907
try:
2908
# install the inventory
2909
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2910
# Cache this inventory
2911
inventory_cache[rev.revision_id] = inv
2912
try:
2913
basis_inv = inventory_cache[rev.parent_ids[0]]
2914
except KeyError:
2915
repository.add_inventory(rev.revision_id, inv, present_parents)
2916
else:
2917
delta = inv._make_delta(basis_inv)
2918
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2919
rev.revision_id, present_parents)
2920
else:
2921
repository.add_inventory(rev.revision_id, inv, present_parents)
2922
except errors.RevisionAlreadyPresent:
2923
pass
2924
if signature is not None:
2925
repository.add_signature_text(rev.revision_id, signature)
2926
repository.add_revision(rev.revision_id, rev, inv)
2927
2928
1275
2929
1276
class MetaDirRepository(Repository):
2930
1277
"""Repositories in the new meta-dir layout.
2965
1312
return not self._transport.has('no-working-trees')
2966
1313
2967
1314
2968
class MetaDirVersionedFileRepository(MetaDirRepository):
2969
"""Repositories in a meta-dir, that work via versioned file objects."""
2970
2971
def __init__(self, _format, a_bzrdir, control_files):
2972
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2973
control_files)
2974
2975
2976
1315
class RepositoryFormatRegistry(controldir.ControlComponentFormatRegistry):
2977
1316
"""Repository format registry."""
2978
1317
3043
1382
# Does this format support CHK bytestring lookups. Set to True or False in
3044
1383
# derived classes.
3045
1384
supports_chks = None
3046
# Should commit add an inventory, or an inventory delta to the repository.
3047
_commit_inv_deltas = True
3048
# What order should fetch operations request streams in?
3049
# The default is unordered as that is the cheapest for an origin to
3050
# provide.
3051
_fetch_order = 'unordered'
3052
# Does this repository format use deltas that can be fetched as-deltas ?
3053
# (E.g. knits, where the knit deltas can be transplanted intact.
3054
# We default to False, which will ensure that enough data to get
3055
# a full text out of any fetch stream will be grabbed.
3056
_fetch_uses_deltas = False
3057
1385
# Should fetch trigger a reconcile after the fetch? Only needed for
3058
1386
# some repository formats that can suffer internal inconsistencies.
3059
1387
_fetch_reconcile = False
3065
1393
# help), and for fetching when data won't have come from the same
3066
1394
# compressor.
3067
1395
pack_compresses = False
3068
# Does the repository inventory storage understand references to trees?
1396
# Does the repository storage understand references to trees?
3069
1397
supports_tree_reference = None
3070
1398
# Is the format experimental ?
3071
1399
experimental = False
3072
# Does this repository format escape funky characters, or does it create files with
3073
# similar names as the versioned files in its contents on disk ?
1400
# Does this repository format escape funky characters, or does it create
1401
# files with similar names as the versioned files in its contents on disk
1402
# ?
3074
1403
supports_funky_characters = None
3075
1404
# Does this repository format support leaving locks?
3076
1405
supports_leaving_lock = None
3080
1409
supports_revision_signatures = True
3081
1410
# Can the revision graph have incorrect parents?
3082
1411
revision_graph_can_have_wrong_parents = None
1412
# Does this format support rich root data?
1413
rich_root_data = None
3083
1414
3084
1415
def __repr__(self):
3085
1416
return "%s()" % self.__class__.__name__
3562
1893
return True
3563
1894
3564
1895
3565
class InterSameDataRepository(InterRepository):
3566
"""Code for converting between repositories that represent the same data.
3567
3568
Data format and model must match for this to work.
3569
"""
3570
3571
@classmethod
3572
def _get_repo_format_to_test(self):
3573
"""Repository format for testing with.
3574
3575
InterSameData can pull from subtree to subtree and from non-subtree to
3576
non-subtree, so we test this with the richest repository format.
3577
"""
3578
from bzrlib.repofmt import knitrepo
3579
return knitrepo.RepositoryFormatKnit3()
3580
3581
@staticmethod
3582
def is_compatible(source, target):
3583
return InterRepository._same_model(source, target)
3584
3585
3586
class InterDifferingSerializer(InterRepository):
3587
3588
@classmethod
3589
def _get_repo_format_to_test(self):
3590
return None
3591
3592
@staticmethod
3593
def is_compatible(source, target):
3594
"""Be compatible with Knit2 source and Knit3 target"""
3595
# This is redundant with format.check_conversion_target(), however that
3596
# raises an exception, and we just want to say "False" as in we won't
3597
# support converting between these formats.
3598
if 'IDS_never' in debug.debug_flags:
3599
return False
3600
if source.supports_rich_root() and not target.supports_rich_root():
3601
return False
3602
if (source._format.supports_tree_reference
3603
and not target._format.supports_tree_reference):
3604
return False
3605
if target._fallback_repositories and target._format.supports_chks:
3606
# IDS doesn't know how to copy CHKs for the parent inventories it
3607
# adds to stacked repos.
3608
return False
3609
if 'IDS_always' in debug.debug_flags:
3610
return True
3611
# Only use this code path for local source and target. IDS does far
3612
# too much IO (both bandwidth and roundtrips) over a network.
3613
if not source.bzrdir.transport.base.startswith('file:///'):
3614
return False
3615
if not target.bzrdir.transport.base.startswith('file:///'):
3616
return False
3617
return True
3618
3619
def _get_trees(self, revision_ids, cache):
3620
possible_trees = []
3621
for rev_id in revision_ids:
3622
if rev_id in cache:
3623
possible_trees.append((rev_id, cache[rev_id]))
3624
else:
3625
# Not cached, but inventory might be present anyway.
3626
try:
3627
tree = self.source.revision_tree(rev_id)
3628
except errors.NoSuchRevision:
3629
# Nope, parent is ghost.
3630
pass
3631
else:
3632
cache[rev_id] = tree
3633
possible_trees.append((rev_id, tree))
3634
return possible_trees
3635
3636
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3637
"""Get the best delta and base for this revision.
3638
3639
:return: (basis_id, delta)
3640
"""
3641
deltas = []
3642
# Generate deltas against each tree, to find the shortest.
3643
texts_possibly_new_in_tree = set()
3644
for basis_id, basis_tree in possible_trees:
3645
delta = tree.inventory._make_delta(basis_tree.inventory)
3646
for old_path, new_path, file_id, new_entry in delta:
3647
if new_path is None:
3648
# This file_id isn't present in the new rev, so we don't
3649
# care about it.
3650
continue
3651
if not new_path:
3652
# Rich roots are handled elsewhere...
3653
continue
3654
kind = new_entry.kind
3655
if kind != 'directory' and kind != 'file':
3656
# No text record associated with this inventory entry.
3657
continue
3658
# This is a directory or file that has changed somehow.
3659
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3660
deltas.append((len(delta), basis_id, delta))
3661
deltas.sort()
3662
return deltas[0][1:]
3663
3664
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3665
"""Find all parent revisions that are absent, but for which the
3666
inventory is present, and copy those inventories.
3667
3668
This is necessary to preserve correctness when the source is stacked
3669
without fallbacks configured. (Note that in cases like upgrade the
3670
source may be not have _fallback_repositories even though it is
3671
stacked.)
3672
"""
3673
parent_revs = set()
3674
for parents in parent_map.values():
3675
parent_revs.update(parents)
3676
present_parents = self.source.get_parent_map(parent_revs)
3677
absent_parents = set(parent_revs).difference(present_parents)
3678
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3679
(rev_id,) for rev_id in absent_parents)
3680
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3681
for parent_tree in self.source.revision_trees(parent_inv_ids):
3682
current_revision_id = parent_tree.get_revision_id()
3683
parents_parents_keys = parent_invs_keys_for_stacking[
3684
(current_revision_id,)]
3685
parents_parents = [key[-1] for key in parents_parents_keys]
3686
basis_id = _mod_revision.NULL_REVISION
3687
basis_tree = self.source.revision_tree(basis_id)
3688
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3689
self.target.add_inventory_by_delta(
3690
basis_id, delta, current_revision_id, parents_parents)
3691
cache[current_revision_id] = parent_tree
3692
3693
def _fetch_batch(self, revision_ids, basis_id, cache):
3694
"""Fetch across a few revisions.
3695
3696
:param revision_ids: The revisions to copy
3697
:param basis_id: The revision_id of a tree that must be in cache, used
3698
as a basis for delta when no other base is available
3699
:param cache: A cache of RevisionTrees that we can use.
3700
:return: The revision_id of the last converted tree. The RevisionTree
3701
for it will be in cache
3702
"""
3703
# Walk though all revisions; get inventory deltas, copy referenced
3704
# texts that delta references, insert the delta, revision and
3705
# signature.
3706
root_keys_to_create = set()
3707
text_keys = set()
3708
pending_deltas = []
3709
pending_revisions = []
3710
parent_map = self.source.get_parent_map(revision_ids)
3711
self._fetch_parent_invs_for_stacking(parent_map, cache)
3712
self.source._safe_to_return_from_cache = True
3713
for tree in self.source.revision_trees(revision_ids):
3714
# Find a inventory delta for this revision.
3715
# Find text entries that need to be copied, too.
3716
current_revision_id = tree.get_revision_id()
3717
parent_ids = parent_map.get(current_revision_id, ())
3718
parent_trees = self._get_trees(parent_ids, cache)
3719
possible_trees = list(parent_trees)
3720
if len(possible_trees) == 0:
3721
# There either aren't any parents, or the parents are ghosts,
3722
# so just use the last converted tree.
3723
possible_trees.append((basis_id, cache[basis_id]))
3724
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3725
possible_trees)
3726
revision = self.source.get_revision(current_revision_id)
3727
pending_deltas.append((basis_id, delta,
3728
current_revision_id, revision.parent_ids))
3729
if self._converting_to_rich_root:
3730
self._revision_id_to_root_id[current_revision_id] = \
3731
tree.get_root_id()
3732
# Determine which texts are in present in this revision but not in
3733
# any of the available parents.
3734
texts_possibly_new_in_tree = set()
3735
for old_path, new_path, file_id, entry in delta:
3736
if new_path is None:
3737
# This file_id isn't present in the new rev
3738
continue
3739
if not new_path:
3740
# This is the root
3741
if not self.target.supports_rich_root():
3742
# The target doesn't support rich root, so we don't
3743
# copy
3744
continue
3745
if self._converting_to_rich_root:
3746
# This can't be copied normally, we have to insert
3747
# it specially
3748
root_keys_to_create.add((file_id, entry.revision))
3749
continue
3750
kind = entry.kind
3751
texts_possibly_new_in_tree.add((file_id, entry.revision))
3752
for basis_id, basis_tree in possible_trees:
3753
basis_inv = basis_tree.inventory
3754
for file_key in list(texts_possibly_new_in_tree):
3755
file_id, file_revision = file_key
3756
try:
3757
entry = basis_inv[file_id]
3758
except errors.NoSuchId:
3759
continue
3760
if entry.revision == file_revision:
3761
texts_possibly_new_in_tree.remove(file_key)
3762
text_keys.update(texts_possibly_new_in_tree)
3763
pending_revisions.append(revision)
3764
cache[current_revision_id] = tree
3765
basis_id = current_revision_id
3766
self.source._safe_to_return_from_cache = False
3767
# Copy file texts
3768
from_texts = self.source.texts
3769
to_texts = self.target.texts
3770
if root_keys_to_create:
3771
root_stream = _mod_fetch._new_root_data_stream(
3772
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3773
self.source)
3774
to_texts.insert_record_stream(root_stream)
3775
to_texts.insert_record_stream(from_texts.get_record_stream(
3776
text_keys, self.target._format._fetch_order,
3777
not self.target._format._fetch_uses_deltas))
3778
# insert inventory deltas
3779
for delta in pending_deltas:
3780
self.target.add_inventory_by_delta(*delta)
3781
if self.target._fallback_repositories:
3782
# Make sure this stacked repository has all the parent inventories
3783
# for the new revisions that we are about to insert. We do this
3784
# before adding the revisions so that no revision is added until
3785
# all the inventories it may depend on are added.
3786
# Note that this is overzealous, as we may have fetched these in an
3787
# earlier batch.
3788
parent_ids = set()
3789
revision_ids = set()
3790
for revision in pending_revisions:
3791
revision_ids.add(revision.revision_id)
3792
parent_ids.update(revision.parent_ids)
3793
parent_ids.difference_update(revision_ids)
3794
parent_ids.discard(_mod_revision.NULL_REVISION)
3795
parent_map = self.source.get_parent_map(parent_ids)
3796
# we iterate over parent_map and not parent_ids because we don't
3797
# want to try copying any revision which is a ghost
3798
for parent_tree in self.source.revision_trees(parent_map):
3799
current_revision_id = parent_tree.get_revision_id()
3800
parents_parents = parent_map[current_revision_id]
3801
possible_trees = self._get_trees(parents_parents, cache)
3802
if len(possible_trees) == 0:
3803
# There either aren't any parents, or the parents are
3804
# ghosts, so just use the last converted tree.
3805
possible_trees.append((basis_id, cache[basis_id]))
3806
basis_id, delta = self._get_delta_for_revision(parent_tree,
3807
parents_parents, possible_trees)
3808
self.target.add_inventory_by_delta(
3809
basis_id, delta, current_revision_id, parents_parents)
3810
# insert signatures and revisions
3811
for revision in pending_revisions:
3812
try:
3813
signature = self.source.get_signature_text(
3814
revision.revision_id)
3815
self.target.add_signature_text(revision.revision_id,
3816
signature)
3817
except errors.NoSuchRevision:
3818
pass
3819
self.target.add_revision(revision.revision_id, revision)
3820
return basis_id
3821
3822
def _fetch_all_revisions(self, revision_ids, pb):
3823
"""Fetch everything for the list of revisions.
3824
3825
:param revision_ids: The list of revisions to fetch. Must be in
3826
topological order.
3827
:param pb: A ProgressTask
3828
:return: None
3829
"""
3830
basis_id, basis_tree = self._get_basis(revision_ids[0])
3831
batch_size = 100
3832
cache = lru_cache.LRUCache(100)
3833
cache[basis_id] = basis_tree
3834
del basis_tree # We don't want to hang on to it here
3835
hints = []
3836
a_graph = None
3837
3838
for offset in range(0, len(revision_ids), batch_size):
3839
self.target.start_write_group()
3840
try:
3841
pb.update('Transferring revisions', offset,
3842
len(revision_ids))
3843
batch = revision_ids[offset:offset+batch_size]
3844
basis_id = self._fetch_batch(batch, basis_id, cache)
3845
except:
3846
self.source._safe_to_return_from_cache = False
3847
self.target.abort_write_group()
3848
raise
3849
else:
3850
hint = self.target.commit_write_group()
3851
if hint:
3852
hints.extend(hint)
3853
if hints and self.target._format.pack_compresses:
3854
self.target.pack(hint=hints)
3855
pb.update('Transferring revisions', len(revision_ids),
3856
len(revision_ids))
3857
3858
@needs_write_lock
3859
def fetch(self, revision_id=None, find_ghosts=False,
3860
fetch_spec=None):
3861
"""See InterRepository.fetch()."""
3862
if fetch_spec is not None:
3863
revision_ids = fetch_spec.get_keys()
3864
else:
3865
revision_ids = None
3866
ui.ui_factory.warn_experimental_format_fetch(self)
3867
if (not self.source.supports_rich_root()
3868
and self.target.supports_rich_root()):
3869
self._converting_to_rich_root = True
3870
self._revision_id_to_root_id = {}
3871
else:
3872
self._converting_to_rich_root = False
3873
# See <https://launchpad.net/bugs/456077> asking for a warning here
3874
if self.source._format.network_name() != self.target._format.network_name():
3875
ui.ui_factory.show_user_warning('cross_format_fetch',
3876
from_format=self.source._format,
3877
to_format=self.target._format)
3878
if revision_ids is None:
3879
if revision_id:
3880
search_revision_ids = [revision_id]
3881
else:
3882
search_revision_ids = None
3883
revision_ids = self.target.search_missing_revision_ids(self.source,
3884
revision_ids=search_revision_ids,
3885
find_ghosts=find_ghosts).get_keys()
3886
if not revision_ids:
3887
return 0, 0
3888
revision_ids = tsort.topo_sort(
3889
self.source.get_graph().get_parent_map(revision_ids))
3890
if not revision_ids:
3891
return 0, 0
3892
# Walk though all revisions; get inventory deltas, copy referenced
3893
# texts that delta references, insert the delta, revision and
3894
# signature.
3895
pb = ui.ui_factory.nested_progress_bar()
3896
try:
3897
self._fetch_all_revisions(revision_ids, pb)
3898
finally:
3899
pb.finished()
3900
return len(revision_ids), 0
3901
3902
def _get_basis(self, first_revision_id):
3903
"""Get a revision and tree which exists in the target.
3904
3905
This assumes that first_revision_id is selected for transmission
3906
because all other ancestors are already present. If we can't find an
3907
ancestor we fall back to NULL_REVISION since we know that is safe.
3908
3909
:return: (basis_id, basis_tree)
3910
"""
3911
first_rev = self.source.get_revision(first_revision_id)
3912
try:
3913
basis_id = first_rev.parent_ids[0]
3914
# only valid as a basis if the target has it
3915
self.target.get_revision(basis_id)
3916
# Try to get a basis tree - if it's a ghost it will hit the
3917
# NoSuchRevision case.
3918
basis_tree = self.source.revision_tree(basis_id)
3919
except (IndexError, errors.NoSuchRevision):
3920
basis_id = _mod_revision.NULL_REVISION
3921
basis_tree = self.source.revision_tree(basis_id)
3922
return basis_id, basis_tree
3923
3924
3925
1896
InterRepository.register_optimiser(InterRepository)
3926
InterRepository.register_optimiser(InterDifferingSerializer)
3927
InterRepository.register_optimiser(InterSameDataRepository)
3928
1897
3929
1898
3930
1899
class CopyConverter(object):
3975
1944
pb.finished()
3976
1945
3977
1946
3978
class _VersionedFileChecker(object):
3979
3980
def __init__(self, repository, text_key_references=None, ancestors=None):
3981
self.repository = repository
3982
self.text_index = self.repository._generate_text_key_index(
3983
text_key_references=text_key_references, ancestors=ancestors)
3984
3985
def calculate_file_version_parents(self, text_key):
3986
"""Calculate the correct parents for a file version according to
3987
the inventories.
3988
"""
3989
parent_keys = self.text_index[text_key]
3990
if parent_keys == [_mod_revision.NULL_REVISION]:
3991
return ()
3992
return tuple(parent_keys)
3993
3994
def check_file_version_parents(self, texts, progress_bar=None):
3995
"""Check the parents stored in a versioned file are correct.
3996
3997
It also detects file versions that are not referenced by their
3998
corresponding revision's inventory.
3999
4000
:returns: A tuple of (wrong_parents, dangling_file_versions).
4001
wrong_parents is a dict mapping {revision_id: (stored_parents,
4002
correct_parents)} for each revision_id where the stored parents
4003
are not correct. dangling_file_versions is a set of (file_id,
4004
revision_id) tuples for versions that are present in this versioned
4005
file, but not used by the corresponding inventory.
4006
"""
4007
local_progress = None
4008
if progress_bar is None:
4009
local_progress = ui.ui_factory.nested_progress_bar()
4010
progress_bar = local_progress
4011
try:
4012
return self._check_file_version_parents(texts, progress_bar)
4013
finally:
4014
if local_progress:
4015
local_progress.finished()
4016
4017
def _check_file_version_parents(self, texts, progress_bar):
4018
"""See check_file_version_parents."""
4019
wrong_parents = {}
4020
self.file_ids = set([file_id for file_id, _ in
4021
self.text_index.iterkeys()])
4022
# text keys is now grouped by file_id
4023
n_versions = len(self.text_index)
4024
progress_bar.update('loading text store', 0, n_versions)
4025
parent_map = self.repository.texts.get_parent_map(self.text_index)
4026
# On unlistable transports this could well be empty/error...
4027
text_keys = self.repository.texts.keys()
4028
unused_keys = frozenset(text_keys) - set(self.text_index)
4029
for num, key in enumerate(self.text_index.iterkeys()):
4030
progress_bar.update('checking text graph', num, n_versions)
4031
correct_parents = self.calculate_file_version_parents(key)
4032
try:
4033
knit_parents = parent_map[key]
4034
except errors.RevisionNotPresent:
4035
# Missing text!
4036
knit_parents = None
4037
if correct_parents != knit_parents:
4038
wrong_parents[key] = (knit_parents, correct_parents)
4039
return wrong_parents, unused_keys
4040
4041
4042
1947
def _strip_NULL_ghosts(revision_graph):
4043
1948
"""Also don't use this. more compatibility code for unmigrated clients."""
4044
1949
# Filter ghosts, and null:
4050
1955
return revision_graph
4051
1956
4052
1957
4053
class StreamSink(object):
4054
"""An object that can insert a stream into a repository.
4055
4056
This interface handles the complexity of reserialising inventories and
4057
revisions from different formats, and allows unidirectional insertion into
4058
stacked repositories without looking for the missing basis parents
4059
beforehand.
4060
"""
4061
4062
def __init__(self, target_repo):
4063
self.target_repo = target_repo
4064
4065
def insert_stream(self, stream, src_format, resume_tokens):
4066
"""Insert a stream's content into the target repository.
4067
4068
:param src_format: a bzr repository format.
4069
4070
:return: a list of resume tokens and an iterable of keys additional
4071
items required before the insertion can be completed.
4072
"""
4073
self.target_repo.lock_write()
4074
try:
4075
if resume_tokens:
4076
self.target_repo.resume_write_group(resume_tokens)
4077
is_resume = True
4078
else:
4079
self.target_repo.start_write_group()
4080
is_resume = False
4081
try:
4082
# locked_insert_stream performs a commit|suspend.
4083
missing_keys = self.insert_stream_without_locking(stream,
4084
src_format, is_resume)
4085
if missing_keys:
4086
# suspend the write group and tell the caller what we is
4087
# missing. We know we can suspend or else we would not have
4088
# entered this code path. (All repositories that can handle
4089
# missing keys can handle suspending a write group).
4090
write_group_tokens = self.target_repo.suspend_write_group()
4091
return write_group_tokens, missing_keys
4092
hint = self.target_repo.commit_write_group()
4093
to_serializer = self.target_repo._format._serializer
4094
src_serializer = src_format._serializer
4095
if (to_serializer != src_serializer and
4096
self.target_repo._format.pack_compresses):
4097
self.target_repo.pack(hint=hint)
4098
return [], set()
4099
except:
4100
self.target_repo.abort_write_group(suppress_errors=True)
4101
raise
4102
finally:
4103
self.target_repo.unlock()
4104
4105
def insert_stream_without_locking(self, stream, src_format,
4106
is_resume=False):
4107
"""Insert a stream's content into the target repository.
4108
4109
This assumes that you already have a locked repository and an active
4110
write group.
4111
4112
:param src_format: a bzr repository format.
4113
:param is_resume: Passed down to get_missing_parent_inventories to
4114
indicate if we should be checking for missing texts at the same
4115
time.
4116
4117
:return: A set of keys that are missing.
4118
"""
4119
if not self.target_repo.is_write_locked():
4120
raise errors.ObjectNotLocked(self)
4121
if not self.target_repo.is_in_write_group():
4122
raise errors.BzrError('you must already be in a write group')
4123
to_serializer = self.target_repo._format._serializer
4124
src_serializer = src_format._serializer
4125
new_pack = None
4126
if to_serializer == src_serializer:
4127
# If serializers match and the target is a pack repository, set the
4128
# write cache size on the new pack. This avoids poor performance
4129
# on transports where append is unbuffered (such as
4130
# RemoteTransport). This is safe to do because nothing should read
4131
# back from the target repository while a stream with matching
4132
# serialization is being inserted.
4133
# The exception is that a delta record from the source that should
4134
# be a fulltext may need to be expanded by the target (see
4135
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4136
# explicitly flush any buffered writes first in that rare case.
4137
try:
4138
new_pack = self.target_repo._pack_collection._new_pack
4139
except AttributeError:
4140
# Not a pack repository
4141
pass
4142
else:
4143
new_pack.set_write_cache_size(1024*1024)
4144
for substream_type, substream in stream:
4145
if 'stream' in debug.debug_flags:
4146
mutter('inserting substream: %s', substream_type)
4147
if substream_type == 'texts':
4148
self.target_repo.texts.insert_record_stream(substream)
4149
elif substream_type == 'inventories':
4150
if src_serializer == to_serializer:
4151
self.target_repo.inventories.insert_record_stream(
4152
substream)
4153
else:
4154
self._extract_and_insert_inventories(
4155
substream, src_serializer)
4156
elif substream_type == 'inventory-deltas':
4157
self._extract_and_insert_inventory_deltas(
4158
substream, src_serializer)
4159
elif substream_type == 'chk_bytes':
4160
# XXX: This doesn't support conversions, as it assumes the
4161
# conversion was done in the fetch code.
4162
self.target_repo.chk_bytes.insert_record_stream(substream)
4163
elif substream_type == 'revisions':
4164
# This may fallback to extract-and-insert more often than
4165
# required if the serializers are different only in terms of
4166
# the inventory.
4167
if src_serializer == to_serializer:
4168
self.target_repo.revisions.insert_record_stream(substream)
4169
else:
4170
self._extract_and_insert_revisions(substream,
4171
src_serializer)
4172
elif substream_type == 'signatures':
4173
self.target_repo.signatures.insert_record_stream(substream)
4174
else:
4175
raise AssertionError('kaboom! %s' % (substream_type,))
4176
# Done inserting data, and the missing_keys calculations will try to
4177
# read back from the inserted data, so flush the writes to the new pack
4178
# (if this is pack format).
4179
if new_pack is not None:
4180
new_pack._write_data('', flush=True)
4181
# Find all the new revisions (including ones from resume_tokens)
4182
missing_keys = self.target_repo.get_missing_parent_inventories(
4183
check_for_missing_texts=is_resume)
4184
try:
4185
for prefix, versioned_file in (
4186
('texts', self.target_repo.texts),
4187
('inventories', self.target_repo.inventories),
4188
('revisions', self.target_repo.revisions),
4189
('signatures', self.target_repo.signatures),
4190
('chk_bytes', self.target_repo.chk_bytes),
4191
):
4192
if versioned_file is None:
4193
continue
4194
# TODO: key is often going to be a StaticTuple object
4195
# I don't believe we can define a method by which
4196
# (prefix,) + StaticTuple will work, though we could
4197
# define a StaticTuple.sq_concat that would allow you to
4198
# pass in either a tuple or a StaticTuple as the second
4199
# object, so instead we could have:
4200
# StaticTuple(prefix) + key here...
4201
missing_keys.update((prefix,) + key for key in
4202
versioned_file.get_missing_compression_parent_keys())
4203
except NotImplementedError:
4204
# cannot even attempt suspending, and missing would have failed
4205
# during stream insertion.
4206
missing_keys = set()
4207
return missing_keys
4208
4209
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4210
target_rich_root = self.target_repo._format.rich_root_data
4211
target_tree_refs = self.target_repo._format.supports_tree_reference
4212
for record in substream:
4213
# Insert the delta directly
4214
inventory_delta_bytes = record.get_bytes_as('fulltext')
4215
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4216
try:
4217
parse_result = deserialiser.parse_text_bytes(
4218
inventory_delta_bytes)
4219
except inventory_delta.IncompatibleInventoryDelta, err:
4220
mutter("Incompatible delta: %s", err.msg)
4221
raise errors.IncompatibleRevision(self.target_repo._format)
4222
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4223
revision_id = new_id
4224
parents = [key[0] for key in record.parents]
4225
self.target_repo.add_inventory_by_delta(
4226
basis_id, inv_delta, revision_id, parents)
4227
4228
def _extract_and_insert_inventories(self, substream, serializer,
4229
parse_delta=None):
4230
"""Generate a new inventory versionedfile in target, converting data.
4231
4232
The inventory is retrieved from the source, (deserializing it), and
4233
stored in the target (reserializing it in a different format).
4234
"""
4235
target_rich_root = self.target_repo._format.rich_root_data
4236
target_tree_refs = self.target_repo._format.supports_tree_reference
4237
for record in substream:
4238
# It's not a delta, so it must be a fulltext in the source
4239
# serializer's format.
4240
bytes = record.get_bytes_as('fulltext')
4241
revision_id = record.key[0]
4242
inv = serializer.read_inventory_from_string(bytes, revision_id)
4243
parents = [key[0] for key in record.parents]
4244
self.target_repo.add_inventory(revision_id, inv, parents)
4245
# No need to keep holding this full inv in memory when the rest of
4246
# the substream is likely to be all deltas.
4247
del inv
4248
4249
def _extract_and_insert_revisions(self, substream, serializer):
4250
for record in substream:
4251
bytes = record.get_bytes_as('fulltext')
4252
revision_id = record.key[0]
4253
rev = serializer.read_revision_from_string(bytes)
4254
if rev.revision_id != revision_id:
4255
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4256
self.target_repo.add_revision(revision_id, rev)
4257
4258
def finished(self):
4259
if self.target_repo._format._fetch_reconcile:
4260
self.target_repo.reconcile()
4261
4262
4263
class StreamSource(object):
4264
"""A source of a stream for fetching between repositories."""
4265
4266
def __init__(self, from_repository, to_format):
4267
"""Create a StreamSource streaming from from_repository."""
4268
self.from_repository = from_repository
4269
self.to_format = to_format
4270
self._record_counter = RecordCounter()
4271
4272
def delta_on_metadata(self):
4273
"""Return True if delta's are permitted on metadata streams.
4274
4275
That is on revisions and signatures.
4276
"""
4277
src_serializer = self.from_repository._format._serializer
4278
target_serializer = self.to_format._serializer
4279
return (self.to_format._fetch_uses_deltas and
4280
src_serializer == target_serializer)
4281
4282
def _fetch_revision_texts(self, revs):
4283
# fetch signatures first and then the revision texts
4284
# may need to be a InterRevisionStore call here.
4285
from_sf = self.from_repository.signatures
4286
# A missing signature is just skipped.
4287
keys = [(rev_id,) for rev_id in revs]
4288
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4289
keys,
4290
self.to_format._fetch_order,
4291
not self.to_format._fetch_uses_deltas))
4292
# If a revision has a delta, this is actually expanded inside the
4293
# insert_record_stream code now, which is an alternate fix for
4294
# bug #261339
4295
from_rf = self.from_repository.revisions
4296
revisions = from_rf.get_record_stream(
4297
keys,
4298
self.to_format._fetch_order,
4299
not self.delta_on_metadata())
4300
return [('signatures', signatures), ('revisions', revisions)]
4301
4302
def _generate_root_texts(self, revs):
4303
"""This will be called by get_stream between fetching weave texts and
4304
fetching the inventory weave.
4305
"""
4306
if self._rich_root_upgrade():
4307
return _mod_fetch.Inter1and2Helper(
4308
self.from_repository).generate_root_texts(revs)
4309
else:
4310
return []
4311
4312
def get_stream(self, search):
4313
phase = 'file'
4314
revs = search.get_keys()
4315
graph = self.from_repository.get_graph()
4316
revs = tsort.topo_sort(graph.get_parent_map(revs))
4317
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4318
text_keys = []
4319
for knit_kind, file_id, revisions in data_to_fetch:
4320
if knit_kind != phase:
4321
phase = knit_kind
4322
# Make a new progress bar for this phase
4323
if knit_kind == "file":
4324
# Accumulate file texts
4325
text_keys.extend([(file_id, revision) for revision in
4326
revisions])
4327
elif knit_kind == "inventory":
4328
# Now copy the file texts.
4329
from_texts = self.from_repository.texts
4330
yield ('texts', from_texts.get_record_stream(
4331
text_keys, self.to_format._fetch_order,
4332
not self.to_format._fetch_uses_deltas))
4333
# Cause an error if a text occurs after we have done the
4334
# copy.
4335
text_keys = None
4336
# Before we process the inventory we generate the root
4337
# texts (if necessary) so that the inventories references
4338
# will be valid.
4339
for _ in self._generate_root_texts(revs):
4340
yield _
4341
# we fetch only the referenced inventories because we do not
4342
# know for unselected inventories whether all their required
4343
# texts are present in the other repository - it could be
4344
# corrupt.
4345
for info in self._get_inventory_stream(revs):
4346
yield info
4347
elif knit_kind == "signatures":
4348
# Nothing to do here; this will be taken care of when
4349
# _fetch_revision_texts happens.
4350
pass
4351
elif knit_kind == "revisions":
4352
for record in self._fetch_revision_texts(revs):
4353
yield record
4354
else:
4355
raise AssertionError("Unknown knit kind %r" % knit_kind)
4356
4357
def get_stream_for_missing_keys(self, missing_keys):
4358
# missing keys can only occur when we are byte copying and not
4359
# translating (because translation means we don't send
4360
# unreconstructable deltas ever).
4361
keys = {}
4362
keys['texts'] = set()
4363
keys['revisions'] = set()
4364
keys['inventories'] = set()
4365
keys['chk_bytes'] = set()
4366
keys['signatures'] = set()
4367
for key in missing_keys:
4368
keys[key[0]].add(key[1:])
4369
if len(keys['revisions']):
4370
# If we allowed copying revisions at this point, we could end up
4371
# copying a revision without copying its required texts: a
4372
# violation of the requirements for repository integrity.
4373
raise AssertionError(
4374
'cannot copy revisions to fill in missing deltas %s' % (
4375
keys['revisions'],))
4376
for substream_kind, keys in keys.iteritems():
4377
vf = getattr(self.from_repository, substream_kind)
4378
if vf is None and keys:
4379
raise AssertionError(
4380
"cannot fill in keys for a versioned file we don't"
4381
" have: %s needs %s" % (substream_kind, keys))
4382
if not keys:
4383
# No need to stream something we don't have
4384
continue
4385
if substream_kind == 'inventories':
4386
# Some missing keys are genuinely ghosts, filter those out.
4387
present = self.from_repository.inventories.get_parent_map(keys)
4388
revs = [key[0] for key in present]
4389
# Get the inventory stream more-or-less as we do for the
4390
# original stream; there's no reason to assume that records
4391
# direct from the source will be suitable for the sink. (Think
4392
# e.g. 2a -> 1.9-rich-root).
4393
for info in self._get_inventory_stream(revs, missing=True):
4394
yield info
4395
continue
4396
4397
# Ask for full texts always so that we don't need more round trips
4398
# after this stream.
4399
# Some of the missing keys are genuinely ghosts, so filter absent
4400
# records. The Sink is responsible for doing another check to
4401
# ensure that ghosts don't introduce missing data for future
4402
# fetches.
4403
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4404
self.to_format._fetch_order, True))
4405
yield substream_kind, stream
4406
4407
def inventory_fetch_order(self):
4408
if self._rich_root_upgrade():
4409
return 'topological'
4410
else:
4411
return self.to_format._fetch_order
4412
4413
def _rich_root_upgrade(self):
4414
return (not self.from_repository._format.rich_root_data and
4415
self.to_format.rich_root_data)
4416
4417
def _get_inventory_stream(self, revision_ids, missing=False):
4418
from_format = self.from_repository._format
4419
if (from_format.supports_chks and self.to_format.supports_chks and
4420
from_format.network_name() == self.to_format.network_name()):
4421
raise AssertionError(
4422
"this case should be handled by GroupCHKStreamSource")
4423
elif 'forceinvdeltas' in debug.debug_flags:
4424
return self._get_convertable_inventory_stream(revision_ids,
4425
delta_versus_null=missing)
4426
elif from_format.network_name() == self.to_format.network_name():
4427
# Same format.
4428
return self._get_simple_inventory_stream(revision_ids,
4429
missing=missing)
4430
elif (not from_format.supports_chks and not self.to_format.supports_chks
4431
and from_format._serializer == self.to_format._serializer):
4432
# Essentially the same format.
4433
return self._get_simple_inventory_stream(revision_ids,
4434
missing=missing)
4435
else:
4436
# Any time we switch serializations, we want to use an
4437
# inventory-delta based approach.
4438
return self._get_convertable_inventory_stream(revision_ids,
4439
delta_versus_null=missing)
4440
4441
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4442
# NB: This currently reopens the inventory weave in source;
4443
# using a single stream interface instead would avoid this.
4444
from_weave = self.from_repository.inventories
4445
if missing:
4446
delta_closure = True
4447
else:
4448
delta_closure = not self.delta_on_metadata()
4449
yield ('inventories', from_weave.get_record_stream(
4450
[(rev_id,) for rev_id in revision_ids],
4451
self.inventory_fetch_order(), delta_closure))
4452
4453
def _get_convertable_inventory_stream(self, revision_ids,
4454
delta_versus_null=False):
4455
# The two formats are sufficiently different that there is no fast
4456
# path, so we need to send just inventorydeltas, which any
4457
# sufficiently modern client can insert into any repository.
4458
# The StreamSink code expects to be able to
4459
# convert on the target, so we need to put bytes-on-the-wire that can
4460
# be converted. That means inventory deltas (if the remote is <1.19,
4461
# RemoteStreamSink will fallback to VFS to insert the deltas).
4462
yield ('inventory-deltas',
4463
self._stream_invs_as_deltas(revision_ids,
4464
delta_versus_null=delta_versus_null))
4465
4466
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4467
"""Return a stream of inventory-deltas for the given rev ids.
4468
4469
:param revision_ids: The list of inventories to transmit
4470
:param delta_versus_null: Don't try to find a minimal delta for this
4471
entry, instead compute the delta versus the NULL_REVISION. This
4472
effectively streams a complete inventory. Used for stuff like
4473
filling in missing parents, etc.
4474
"""
4475
from_repo = self.from_repository
4476
revision_keys = [(rev_id,) for rev_id in revision_ids]
4477
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4478
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4479
# method...
4480
inventories = self.from_repository.iter_inventories(
4481
revision_ids, 'topological')
4482
format = from_repo._format
4483
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4484
inventory_cache = lru_cache.LRUCache(50)
4485
null_inventory = from_repo.revision_tree(
4486
_mod_revision.NULL_REVISION).inventory
4487
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4488
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4489
# repo back into a non-rich-root repo ought to be allowed)
4490
serializer = inventory_delta.InventoryDeltaSerializer(
4491
versioned_root=format.rich_root_data,
4492
tree_references=format.supports_tree_reference)
4493
for inv in inventories:
4494
key = (inv.revision_id,)
4495
parent_keys = parent_map.get(key, ())
4496
delta = None
4497
if not delta_versus_null and parent_keys:
4498
# The caller did not ask for complete inventories and we have
4499
# some parents that we can delta against. Make a delta against
4500
# each parent so that we can find the smallest.
4501
parent_ids = [parent_key[0] for parent_key in parent_keys]
4502
for parent_id in parent_ids:
4503
if parent_id not in invs_sent_so_far:
4504
# We don't know that the remote side has this basis, so
4505
# we can't use it.
4506
continue
4507
if parent_id == _mod_revision.NULL_REVISION:
4508
parent_inv = null_inventory
4509
else:
4510
parent_inv = inventory_cache.get(parent_id, None)
4511
if parent_inv is None:
4512
parent_inv = from_repo.get_inventory(parent_id)
4513
candidate_delta = inv._make_delta(parent_inv)
4514
if (delta is None or
4515
len(delta) > len(candidate_delta)):
4516
delta = candidate_delta
4517
basis_id = parent_id
4518
if delta is None:
4519
# Either none of the parents ended up being suitable, or we
4520
# were asked to delta against NULL
4521
basis_id = _mod_revision.NULL_REVISION
4522
delta = inv._make_delta(null_inventory)
4523
invs_sent_so_far.add(inv.revision_id)
4524
inventory_cache[inv.revision_id] = inv
4525
delta_serialized = ''.join(
4526
serializer.delta_to_lines(basis_id, key[-1], delta))
4527
yield versionedfile.FulltextContentFactory(
4528
key, parent_keys, None, delta_serialized)
4529
4530
4531
1958
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4532
1959
stop_revision=None):
4533
1960
"""Extend the partial history to include a given index
Loggerhead is a web-based interface for Breezy
Version: 2.0.1