« back to all changes in this revision
Viewing changes to bzrlib/repository.py
- browse files at revision 5718.9.2
- compare with another revision
- compare with revision 6170
- download diff
- download tarball
- view history from revision 5718.9.2
- Committer: Jelmer Vernooij
- Date: 2011-05-10 07:46:15 UTC
- mfrom: (5844 +trunk)
- mto: This revision was merged to the branch mainline in revision 5845.
- Revision ID: jelmer@samba.org-20110510074615-eptod049ndjxc4i7
Merge bzr.dev.
- files added:
- bzrlib/plugins/changelog_merge
- bzrlib/plugins/changelog_merge/tests
- files removed:
- bzrlib/tests/per_interbranch/test_update_revisions.py
- files renamed:
- bzrlib/tests/per_repository/helpers.py => bzrlib/tests/per_repository_vf/helpers.py
- bzrlib/tests/per_repository/test_add_inventory_by_delta.py => bzrlib/tests/per_repository_vf/test_add_inventory_by_delta.py
- files modified:
- Makefile
added
removed
bzrlib/repository.py
16
16
17
17
from bzrlib.lazy_import import lazy_import
18
18
lazy_import(globals(), """
19
import cStringIO
20
import re
21
19
import time
22
20
23
21
from bzrlib import (
24
22
bzrdir,
25
23
check,
26
chk_map,
27
24
config,
28
25
controldir,
29
26
debug,
30
fetch as _mod_fetch,
31
fifo_cache,
32
27
generate_ids,
33
gpg,
34
28
graph,
35
inventory,
36
inventory_delta,
37
29
lockable_files,
38
30
lockdir,
39
lru_cache,
40
31
osutils,
41
pyutils,
42
32
revision as _mod_revision,
43
static_tuple,
44
trace,
33
testament as _mod_testament,
45
34
tsort,
46
versionedfile,
47
35
)
48
36
from bzrlib.bundle import serializer
49
from bzrlib.revisiontree import RevisionTree
50
from bzrlib.store.versioned import VersionedFileStore
51
from bzrlib.testament import Testament
52
37
""")
53
38
54
39
from bzrlib import (
59
44
)
60
45
from bzrlib.decorators import needs_read_lock, needs_write_lock, only_raises
61
46
from bzrlib.inter import InterObject
62
from bzrlib.inventory import (
63
Inventory,
64
InventoryDirectory,
65
ROOT_ID,
66
entry_factory,
67
)
68
from bzrlib.recordcounter import RecordCounter
69
47
from bzrlib.lock import _RelockDebugMixin, LogicalLockResult
70
48
from bzrlib.trace import (
71
49
log_exception_quietly, note, mutter, mutter_callsite, warning)
92
70
93
71
# all clients should supply tree roots.
94
72
record_root_entry = True
95
# the default CommitBuilder does not manage trees whose root is versioned.
96
_versioned_root = False
73
# whether this commit builder supports the record_entry_contents interface
74
supports_record_entry_contents = False
97
75
98
76
def __init__(self, repository, parents, config, timestamp=None,
99
77
timezone=None, committer=None, revprops=None,
100
revision_id=None):
78
revision_id=None, lossy=False):
101
79
"""Initiate a CommitBuilder.
102
80
103
81
:param repository: Repository to commit to.
104
82
:param parents: Revision ids of the parents of the new revision.
105
:param config: Configuration to use.
106
83
:param timestamp: Optional timestamp recorded for commit.
107
84
:param timezone: Optional timezone for timestamp.
108
85
:param committer: Optional committer to set for commit.
109
86
:param revprops: Optional dictionary of revision properties.
110
87
:param revision_id: Optional revision id.
88
:param lossy: Whether to discard data that can not be natively
89
represented, when pushing to a foreign VCS
111
90
"""
112
91
self._config = config
92
self._lossy = lossy
113
93
114
94
if committer is None:
115
95
self._committer = self._config.username()
118
98
else:
119
99
self._committer = committer
120
100
121
self.new_inventory = Inventory(None)
122
101
self._new_revision_id = revision_id
123
102
self.parents = parents
124
103
self.repository = repository
139
118
self._timezone = int(timezone)
140
119
141
120
self._generate_revision_if_needed()
142
self.__heads = graph.HeadsCache(repository.get_graph()).heads
143
self._basis_delta = []
144
# API compatibility, older code that used CommitBuilder did not call
145
# .record_delete(), which means the delta that is computed would not be
146
# valid. Callers that will call record_delete() should call
147
# .will_record_deletes() to indicate that.
148
self._recording_deletes = False
149
# memo'd check for no-op commits.
150
self._any_changes = False
151
121
152
122
def any_changes(self):
153
123
"""Return True if any entries were changed.
154
124
155
125
This includes merge-only changes. It is the core for the --unchanged
156
126
detection in commit.
157
127
158
128
:return: True if any changes have occured.
159
129
"""
160
return self._any_changes
130
raise NotImplementedError(self.any_changes)
161
131
162
132
def _validate_unicode_text(self, text, context):
163
133
"""Verify things like commit messages don't have bogus characters."""
174
144
self._validate_unicode_text(value,
175
145
'revision property (%s)' % (key,))
176
146
177
def _ensure_fallback_inventories(self):
178
"""Ensure that appropriate inventories are available.
179
180
This only applies to repositories that are stacked, and is about
181
enusring the stacking invariants. Namely, that for any revision that is
182
present, we either have all of the file content, or we have the parent
183
inventory and the delta file content.
184
"""
185
if not self.repository._fallback_repositories:
186
return
187
if not self.repository._format.supports_chks:
188
raise errors.BzrError("Cannot commit directly to a stacked branch"
189
" in pre-2a formats. See "
190
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
191
# This is a stacked repo, we need to make sure we have the parent
192
# inventories for the parents.
193
parent_keys = [(p,) for p in self.parents]
194
parent_map = self.repository.inventories._index.get_parent_map(parent_keys)
195
missing_parent_keys = set([pk for pk in parent_keys
196
if pk not in parent_map])
197
fallback_repos = list(reversed(self.repository._fallback_repositories))
198
missing_keys = [('inventories', pk[0])
199
for pk in missing_parent_keys]
200
resume_tokens = []
201
while missing_keys and fallback_repos:
202
fallback_repo = fallback_repos.pop()
203
source = fallback_repo._get_source(self.repository._format)
204
sink = self.repository._get_sink()
205
stream = source.get_stream_for_missing_keys(missing_keys)
206
missing_keys = sink.insert_stream_without_locking(stream,
207
self.repository._format)
208
if missing_keys:
209
raise errors.BzrError('Unable to fill in parent inventories for a'
210
' stacked branch')
211
212
147
def commit(self, message):
213
148
"""Make the actual commit.
214
149
215
150
:return: The revision id of the recorded revision.
216
151
"""
217
self._validate_unicode_text(message, 'commit message')
218
rev = _mod_revision.Revision(
219
timestamp=self._timestamp,
220
timezone=self._timezone,
221
committer=self._committer,
222
message=message,
223
inventory_sha1=self.inv_sha1,
224
revision_id=self._new_revision_id,
225
properties=self._revprops)
226
rev.parent_ids = self.parents
227
self.repository.add_revision(self._new_revision_id, rev,
228
self.new_inventory, self._config)
229
self._ensure_fallback_inventories()
230
self.repository.commit_write_group()
231
return self._new_revision_id
152
raise NotImplementedError(self.commit)
232
153
233
154
def abort(self):
234
155
"""Abort the commit that is being built.
235
156
"""
236
self.repository.abort_write_group()
157
raise NotImplementedError(self.abort)
237
158
238
159
def revision_tree(self):
239
160
"""Return the tree that was just committed.
240
161
241
After calling commit() this can be called to get a RevisionTree
242
representing the newly committed tree. This is preferred to
243
calling Repository.revision_tree() because that may require
244
deserializing the inventory, while we already have a copy in
162
After calling commit() this can be called to get a
163
RevisionTree representing the newly committed tree. This is
164
preferred to calling Repository.revision_tree() because that may
165
require deserializing the inventory, while we already have a copy in
245
166
memory.
246
167
"""
247
if self.new_inventory is None:
248
self.new_inventory = self.repository.get_inventory(
249
self._new_revision_id)
250
return RevisionTree(self.repository, self.new_inventory,
251
self._new_revision_id)
168
raise NotImplementedError(self.revision_tree)
252
169
253
170
def finish_inventory(self):
254
171
"""Tell the builder that the inventory is finished.
256
173
:return: The inventory id in the repository, which can be used with
257
174
repository.get_inventory.
258
175
"""
259
if self.new_inventory is None:
260
# an inventory delta was accumulated without creating a new
261
# inventory.
262
basis_id = self.basis_delta_revision
263
# We ignore the 'inventory' returned by add_inventory_by_delta
264
# because self.new_inventory is used to hint to the rest of the
265
# system what code path was taken
266
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
267
basis_id, self._basis_delta, self._new_revision_id,
268
self.parents)
269
else:
270
if self.new_inventory.root is None:
271
raise AssertionError('Root entry should be supplied to'
272
' record_entry_contents, as of bzr 0.10.')
273
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
274
self.new_inventory.revision_id = self._new_revision_id
275
self.inv_sha1 = self.repository.add_inventory(
276
self._new_revision_id,
277
self.new_inventory,
278
self.parents
279
)
280
return self._new_revision_id
176
raise NotImplementedError(self.finish_inventory)
281
177
282
178
def _gen_revision_id(self):
283
179
"""Return new revision-id."""
298
194
else:
299
195
self.random_revid = False
300
196
301
def _heads(self, file_id, revision_ids):
302
"""Calculate the graph heads for revision_ids in the graph of file_id.
303
304
This can use either a per-file graph or a global revision graph as we
305
have an identity relationship between the two graphs.
306
"""
307
return self.__heads(revision_ids)
308
309
def _check_root(self, ie, parent_invs, tree):
310
"""Helper for record_entry_contents.
311
312
:param ie: An entry being added.
313
:param parent_invs: The inventories of the parent revisions of the
314
commit.
315
:param tree: The tree that is being committed.
316
"""
317
# In this revision format, root entries have no knit or weave When
318
# serializing out to disk and back in root.revision is always
319
# _new_revision_id
320
ie.revision = self._new_revision_id
321
322
def _require_root_change(self, tree):
323
"""Enforce an appropriate root object change.
324
325
This is called once when record_iter_changes is called, if and only if
326
the root was not in the delta calculated by record_iter_changes.
327
328
:param tree: The tree which is being committed.
329
"""
330
if len(self.parents) == 0:
331
raise errors.RootMissing()
332
entry = entry_factory['directory'](tree.path2id(''), '',
333
None)
334
entry.revision = self._new_revision_id
335
self._basis_delta.append(('', '', entry.file_id, entry))
336
337
def _get_delta(self, ie, basis_inv, path):
338
"""Get a delta against the basis inventory for ie."""
339
if ie.file_id not in basis_inv:
340
# add
341
result = (None, path, ie.file_id, ie)
342
self._basis_delta.append(result)
343
return result
344
elif ie != basis_inv[ie.file_id]:
345
# common but altered
346
# TODO: avoid tis id2path call.
347
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
348
self._basis_delta.append(result)
349
return result
350
else:
351
# common, unaltered
352
return None
353
354
def get_basis_delta(self):
355
"""Return the complete inventory delta versus the basis inventory.
356
357
This has been built up with the calls to record_delete and
358
record_entry_contents. The client must have already called
359
will_record_deletes() to indicate that they will be generating a
360
complete delta.
361
362
:return: An inventory delta, suitable for use with apply_delta, or
363
Repository.add_inventory_by_delta, etc.
364
"""
365
if not self._recording_deletes:
366
raise AssertionError("recording deletes not activated.")
367
return self._basis_delta
368
369
def record_delete(self, path, file_id):
370
"""Record that a delete occured against a basis tree.
371
372
This is an optional API - when used it adds items to the basis_delta
373
being accumulated by the commit builder. It cannot be called unless the
374
method will_record_deletes() has been called to inform the builder that
375
a delta is being supplied.
376
377
:param path: The path of the thing deleted.
378
:param file_id: The file id that was deleted.
379
"""
380
if not self._recording_deletes:
381
raise AssertionError("recording deletes not activated.")
382
delta = (path, None, file_id, None)
383
self._basis_delta.append(delta)
384
self._any_changes = True
385
return delta
386
387
197
def will_record_deletes(self):
388
198
"""Tell the commit builder that deletes are being notified.
389
199
391
201
commit to be valid, deletes against the basis MUST be recorded via
392
202
builder.record_delete().
393
203
"""
394
self._recording_deletes = True
395
try:
396
basis_id = self.parents[0]
397
except IndexError:
398
basis_id = _mod_revision.NULL_REVISION
399
self.basis_delta_revision = basis_id
400
401
def record_entry_contents(self, ie, parent_invs, path, tree,
402
content_summary):
403
"""Record the content of ie from tree into the commit if needed.
404
405
Side effect: sets ie.revision when unchanged
406
407
:param ie: An inventory entry present in the commit.
408
:param parent_invs: The inventories of the parent revisions of the
409
commit.
410
:param path: The path the entry is at in the tree.
411
:param tree: The tree which contains this entry and should be used to
412
obtain content.
413
:param content_summary: Summary data from the tree about the paths
414
content - stat, length, exec, sha/link target. This is only
415
accessed when the entry has a revision of None - that is when it is
416
a candidate to commit.
417
:return: A tuple (change_delta, version_recorded, fs_hash).
418
change_delta is an inventory_delta change for this entry against
419
the basis tree of the commit, or None if no change occured against
420
the basis tree.
421
version_recorded is True if a new version of the entry has been
422
recorded. For instance, committing a merge where a file was only
423
changed on the other side will return (delta, False).
424
fs_hash is either None, or the hash details for the path (currently
425
a tuple of the contents sha1 and the statvalue returned by
426
tree.get_file_with_stat()).
427
"""
428
if self.new_inventory.root is None:
429
if ie.parent_id is not None:
430
raise errors.RootMissing()
431
self._check_root(ie, parent_invs, tree)
432
if ie.revision is None:
433
kind = content_summary[0]
434
else:
435
# ie is carried over from a prior commit
436
kind = ie.kind
437
# XXX: repository specific check for nested tree support goes here - if
438
# the repo doesn't want nested trees we skip it ?
439
if (kind == 'tree-reference' and
440
not self.repository._format.supports_tree_reference):
441
# mismatch between commit builder logic and repository:
442
# this needs the entry creation pushed down into the builder.
443
raise NotImplementedError('Missing repository subtree support.')
444
self.new_inventory.add(ie)
445
446
# TODO: slow, take it out of the inner loop.
447
try:
448
basis_inv = parent_invs[0]
449
except IndexError:
450
basis_inv = Inventory(root_id=None)
451
452
# ie.revision is always None if the InventoryEntry is considered
453
# for committing. We may record the previous parents revision if the
454
# content is actually unchanged against a sole head.
455
if ie.revision is not None:
456
if not self._versioned_root and path == '':
457
# repositories that do not version the root set the root's
458
# revision to the new commit even when no change occurs (more
459
# specifically, they do not record a revision on the root; and
460
# the rev id is assigned to the root during deserialisation -
461
# this masks when a change may have occurred against the basis.
462
# To match this we always issue a delta, because the revision
463
# of the root will always be changing.
464
if ie.file_id in basis_inv:
465
delta = (basis_inv.id2path(ie.file_id), path,
466
ie.file_id, ie)
467
else:
468
# add
469
delta = (None, path, ie.file_id, ie)
470
self._basis_delta.append(delta)
471
return delta, False, None
472
else:
473
# we don't need to commit this, because the caller already
474
# determined that an existing revision of this file is
475
# appropriate. If it's not being considered for committing then
476
# it and all its parents to the root must be unaltered so
477
# no-change against the basis.
478
if ie.revision == self._new_revision_id:
479
raise AssertionError("Impossible situation, a skipped "
480
"inventory entry (%r) claims to be modified in this "
481
"commit (%r).", (ie, self._new_revision_id))
482
return None, False, None
483
# XXX: Friction: parent_candidates should return a list not a dict
484
# so that we don't have to walk the inventories again.
485
parent_candiate_entries = ie.parent_candidates(parent_invs)
486
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
487
heads = []
488
for inv in parent_invs:
489
if ie.file_id in inv:
490
old_rev = inv[ie.file_id].revision
491
if old_rev in head_set:
492
heads.append(inv[ie.file_id].revision)
493
head_set.remove(inv[ie.file_id].revision)
494
495
store = False
496
# now we check to see if we need to write a new record to the
497
# file-graph.
498
# We write a new entry unless there is one head to the ancestors, and
499
# the kind-derived content is unchanged.
500
501
# Cheapest check first: no ancestors, or more the one head in the
502
# ancestors, we write a new node.
503
if len(heads) != 1:
504
store = True
505
if not store:
506
# There is a single head, look it up for comparison
507
parent_entry = parent_candiate_entries[heads[0]]
508
# if the non-content specific data has changed, we'll be writing a
509
# node:
510
if (parent_entry.parent_id != ie.parent_id or
511
parent_entry.name != ie.name):
512
store = True
513
# now we need to do content specific checks:
514
if not store:
515
# if the kind changed the content obviously has
516
if kind != parent_entry.kind:
517
store = True
518
# Stat cache fingerprint feedback for the caller - None as we usually
519
# don't generate one.
520
fingerprint = None
521
if kind == 'file':
522
if content_summary[2] is None:
523
raise ValueError("Files must not have executable = None")
524
if not store:
525
# We can't trust a check of the file length because of content
526
# filtering...
527
if (# if the exec bit has changed we have to store:
528
parent_entry.executable != content_summary[2]):
529
store = True
530
elif parent_entry.text_sha1 == content_summary[3]:
531
# all meta and content is unchanged (using a hash cache
532
# hit to check the sha)
533
ie.revision = parent_entry.revision
534
ie.text_size = parent_entry.text_size
535
ie.text_sha1 = parent_entry.text_sha1
536
ie.executable = parent_entry.executable
537
return self._get_delta(ie, basis_inv, path), False, None
538
else:
539
# Either there is only a hash change(no hash cache entry,
540
# or same size content change), or there is no change on
541
# this file at all.
542
# Provide the parent's hash to the store layer, so that the
543
# content is unchanged we will not store a new node.
544
nostore_sha = parent_entry.text_sha1
545
if store:
546
# We want to record a new node regardless of the presence or
547
# absence of a content change in the file.
548
nostore_sha = None
549
ie.executable = content_summary[2]
550
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
551
try:
552
text = file_obj.read()
553
finally:
554
file_obj.close()
555
try:
556
ie.text_sha1, ie.text_size = self._add_text_to_weave(
557
ie.file_id, text, heads, nostore_sha)
558
# Let the caller know we generated a stat fingerprint.
559
fingerprint = (ie.text_sha1, stat_value)
560
except errors.ExistingContent:
561
# Turns out that the file content was unchanged, and we were
562
# only going to store a new node if it was changed. Carry over
563
# the entry.
564
ie.revision = parent_entry.revision
565
ie.text_size = parent_entry.text_size
566
ie.text_sha1 = parent_entry.text_sha1
567
ie.executable = parent_entry.executable
568
return self._get_delta(ie, basis_inv, path), False, None
569
elif kind == 'directory':
570
if not store:
571
# all data is meta here, nothing specific to directory, so
572
# carry over:
573
ie.revision = parent_entry.revision
574
return self._get_delta(ie, basis_inv, path), False, None
575
self._add_text_to_weave(ie.file_id, '', heads, None)
576
elif kind == 'symlink':
577
current_link_target = content_summary[3]
578
if not store:
579
# symlink target is not generic metadata, check if it has
580
# changed.
581
if current_link_target != parent_entry.symlink_target:
582
store = True
583
if not store:
584
# unchanged, carry over.
585
ie.revision = parent_entry.revision
586
ie.symlink_target = parent_entry.symlink_target
587
return self._get_delta(ie, basis_inv, path), False, None
588
ie.symlink_target = current_link_target
589
self._add_text_to_weave(ie.file_id, '', heads, None)
590
elif kind == 'tree-reference':
591
if not store:
592
if content_summary[3] != parent_entry.reference_revision:
593
store = True
594
if not store:
595
# unchanged, carry over.
596
ie.reference_revision = parent_entry.reference_revision
597
ie.revision = parent_entry.revision
598
return self._get_delta(ie, basis_inv, path), False, None
599
ie.reference_revision = content_summary[3]
600
if ie.reference_revision is None:
601
raise AssertionError("invalid content_summary for nested tree: %r"
602
% (content_summary,))
603
self._add_text_to_weave(ie.file_id, '', heads, None)
604
else:
605
raise NotImplementedError('unknown kind')
606
ie.revision = self._new_revision_id
607
self._any_changes = True
608
return self._get_delta(ie, basis_inv, path), True, fingerprint
609
610
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
611
_entry_factory=entry_factory):
204
raise NotImplementedError(self.will_record_deletes)
205
206
def record_iter_changes(self, tree, basis_revision_id, iter_changes):
612
207
"""Record a new tree via iter_changes.
613
208
614
209
:param tree: The tree to obtain text contents from for changed objects.
619
214
to basis_revision_id. The iterator must not include any items with
620
215
a current kind of None - missing items must be either filtered out
621
216
or errored-on beefore record_iter_changes sees the item.
622
:param _entry_factory: Private method to bind entry_factory locally for
623
performance.
624
217
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
625
218
tree._observed_sha1.
626
219
"""
627
# Create an inventory delta based on deltas between all the parents and
628
# deltas between all the parent inventories. We use inventory delta's
629
# between the inventory objects because iter_changes masks
630
# last-changed-field only changes.
631
# Working data:
632
# file_id -> change map, change is fileid, paths, changed, versioneds,
633
# parents, names, kinds, executables
634
merged_ids = {}
635
# {file_id -> revision_id -> inventory entry, for entries in parent
636
# trees that are not parents[0]
637
parent_entries = {}
638
ghost_basis = False
639
try:
640
revtrees = list(self.repository.revision_trees(self.parents))
641
except errors.NoSuchRevision:
642
# one or more ghosts, slow path.
643
revtrees = []
644
for revision_id in self.parents:
645
try:
646
revtrees.append(self.repository.revision_tree(revision_id))
647
except errors.NoSuchRevision:
648
if not revtrees:
649
basis_revision_id = _mod_revision.NULL_REVISION
650
ghost_basis = True
651
revtrees.append(self.repository.revision_tree(
652
_mod_revision.NULL_REVISION))
653
# The basis inventory from a repository
654
if revtrees:
655
basis_inv = revtrees[0].inventory
656
else:
657
basis_inv = self.repository.revision_tree(
658
_mod_revision.NULL_REVISION).inventory
659
if len(self.parents) > 0:
660
if basis_revision_id != self.parents[0] and not ghost_basis:
661
raise Exception(
662
"arbitrary basis parents not yet supported with merges")
663
for revtree in revtrees[1:]:
664
for change in revtree.inventory._make_delta(basis_inv):
665
if change[1] is None:
666
# Not present in this parent.
667
continue
668
if change[2] not in merged_ids:
669
if change[0] is not None:
670
basis_entry = basis_inv[change[2]]
671
merged_ids[change[2]] = [
672
# basis revid
673
basis_entry.revision,
674
# new tree revid
675
change[3].revision]
676
parent_entries[change[2]] = {
677
# basis parent
678
basis_entry.revision:basis_entry,
679
# this parent
680
change[3].revision:change[3],
681
}
682
else:
683
merged_ids[change[2]] = [change[3].revision]
684
parent_entries[change[2]] = {change[3].revision:change[3]}
685
else:
686
merged_ids[change[2]].append(change[3].revision)
687
parent_entries[change[2]][change[3].revision] = change[3]
688
else:
689
merged_ids = {}
690
# Setup the changes from the tree:
691
# changes maps file_id -> (change, [parent revision_ids])
692
changes= {}
693
for change in iter_changes:
694
# This probably looks up in basis_inv way to much.
695
if change[1][0] is not None:
696
head_candidate = [basis_inv[change[0]].revision]
697
else:
698
head_candidate = []
699
changes[change[0]] = change, merged_ids.get(change[0],
700
head_candidate)
701
unchanged_merged = set(merged_ids) - set(changes)
702
# Extend the changes dict with synthetic changes to record merges of
703
# texts.
704
for file_id in unchanged_merged:
705
# Record a merged version of these items that did not change vs the
706
# basis. This can be either identical parallel changes, or a revert
707
# of a specific file after a merge. The recorded content will be
708
# that of the current tree (which is the same as the basis), but
709
# the per-file graph will reflect a merge.
710
# NB:XXX: We are reconstructing path information we had, this
711
# should be preserved instead.
712
# inv delta change: (file_id, (path_in_source, path_in_target),
713
# changed_content, versioned, parent, name, kind,
714
# executable)
715
try:
716
basis_entry = basis_inv[file_id]
717
except errors.NoSuchId:
718
# a change from basis->some_parents but file_id isn't in basis
719
# so was new in the merge, which means it must have changed
720
# from basis -> current, and as it hasn't the add was reverted
721
# by the user. So we discard this change.
722
pass
723
else:
724
change = (file_id,
725
(basis_inv.id2path(file_id), tree.id2path(file_id)),
726
False, (True, True),
727
(basis_entry.parent_id, basis_entry.parent_id),
728
(basis_entry.name, basis_entry.name),
729
(basis_entry.kind, basis_entry.kind),
730
(basis_entry.executable, basis_entry.executable))
731
changes[file_id] = (change, merged_ids[file_id])
732
# changes contains tuples with the change and a set of inventory
733
# candidates for the file.
734
# inv delta is:
735
# old_path, new_path, file_id, new_inventory_entry
736
seen_root = False # Is the root in the basis delta?
737
inv_delta = self._basis_delta
738
modified_rev = self._new_revision_id
739
for change, head_candidates in changes.values():
740
if change[3][1]: # versioned in target.
741
# Several things may be happening here:
742
# We may have a fork in the per-file graph
743
# - record a change with the content from tree
744
# We may have a change against < all trees
745
# - carry over the tree that hasn't changed
746
# We may have a change against all trees
747
# - record the change with the content from tree
748
kind = change[6][1]
749
file_id = change[0]
750
entry = _entry_factory[kind](file_id, change[5][1],
751
change[4][1])
752
head_set = self._heads(change[0], set(head_candidates))
753
heads = []
754
# Preserve ordering.
755
for head_candidate in head_candidates:
756
if head_candidate in head_set:
757
heads.append(head_candidate)
758
head_set.remove(head_candidate)
759
carried_over = False
760
if len(heads) == 1:
761
# Could be a carry-over situation:
762
parent_entry_revs = parent_entries.get(file_id, None)
763
if parent_entry_revs:
764
parent_entry = parent_entry_revs.get(heads[0], None)
765
else:
766
parent_entry = None
767
if parent_entry is None:
768
# The parent iter_changes was called against is the one
769
# that is the per-file head, so any change is relevant
770
# iter_changes is valid.
771
carry_over_possible = False
772
else:
773
# could be a carry over situation
774
# A change against the basis may just indicate a merge,
775
# we need to check the content against the source of the
776
# merge to determine if it was changed after the merge
777
# or carried over.
778
if (parent_entry.kind != entry.kind or
779
parent_entry.parent_id != entry.parent_id or
780
parent_entry.name != entry.name):
781
# Metadata common to all entries has changed
782
# against per-file parent
783
carry_over_possible = False
784
else:
785
carry_over_possible = True
786
# per-type checks for changes against the parent_entry
787
# are done below.
788
else:
789
# Cannot be a carry-over situation
790
carry_over_possible = False
791
# Populate the entry in the delta
792
if kind == 'file':
793
# XXX: There is still a small race here: If someone reverts the content of a file
794
# after iter_changes examines and decides it has changed,
795
# we will unconditionally record a new version even if some
796
# other process reverts it while commit is running (with
797
# the revert happening after iter_changes did its
798
# examination).
799
if change[7][1]:
800
entry.executable = True
801
else:
802
entry.executable = False
803
if (carry_over_possible and
804
parent_entry.executable == entry.executable):
805
# Check the file length, content hash after reading
806
# the file.
807
nostore_sha = parent_entry.text_sha1
808
else:
809
nostore_sha = None
810
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
811
try:
812
text = file_obj.read()
813
finally:
814
file_obj.close()
815
try:
816
entry.text_sha1, entry.text_size = self._add_text_to_weave(
817
file_id, text, heads, nostore_sha)
818
yield file_id, change[1][1], (entry.text_sha1, stat_value)
819
except errors.ExistingContent:
820
# No content change against a carry_over parent
821
# Perhaps this should also yield a fs hash update?
822
carried_over = True
823
entry.text_size = parent_entry.text_size
824
entry.text_sha1 = parent_entry.text_sha1
825
elif kind == 'symlink':
826
# Wants a path hint?
827
entry.symlink_target = tree.get_symlink_target(file_id)
828
if (carry_over_possible and
829
parent_entry.symlink_target == entry.symlink_target):
830
carried_over = True
831
else:
832
self._add_text_to_weave(change[0], '', heads, None)
833
elif kind == 'directory':
834
if carry_over_possible:
835
carried_over = True
836
else:
837
# Nothing to set on the entry.
838
# XXX: split into the Root and nonRoot versions.
839
if change[1][1] != '' or self.repository.supports_rich_root():
840
self._add_text_to_weave(change[0], '', heads, None)
841
elif kind == 'tree-reference':
842
if not self.repository._format.supports_tree_reference:
843
# This isn't quite sane as an error, but we shouldn't
844
# ever see this code path in practice: tree's don't
845
# permit references when the repo doesn't support tree
846
# references.
847
raise errors.UnsupportedOperation(tree.add_reference,
848
self.repository)
849
reference_revision = tree.get_reference_revision(change[0])
850
entry.reference_revision = reference_revision
851
if (carry_over_possible and
852
parent_entry.reference_revision == reference_revision):
853
carried_over = True
854
else:
855
self._add_text_to_weave(change[0], '', heads, None)
856
else:
857
raise AssertionError('unknown kind %r' % kind)
858
if not carried_over:
859
entry.revision = modified_rev
860
else:
861
entry.revision = parent_entry.revision
862
else:
863
entry = None
864
new_path = change[1][1]
865
inv_delta.append((change[1][0], new_path, change[0], entry))
866
if new_path == '':
867
seen_root = True
868
self.new_inventory = None
869
if len(inv_delta):
870
# This should perhaps be guarded by a check that the basis we
871
# commit against is the basis for the commit and if not do a delta
872
# against the basis.
873
self._any_changes = True
874
if not seen_root:
875
# housekeeping root entry changes do not affect no-change commits.
876
self._require_root_change(tree)
877
self.basis_delta_revision = basis_revision_id
878
879
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
880
parent_keys = tuple([(file_id, parent) for parent in parents])
881
return self.repository.texts._add_text(
882
(file_id, self._new_revision_id), parent_keys, new_text,
883
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
884
885
886
class RootCommitBuilder(CommitBuilder):
887
"""This commitbuilder actually records the root id"""
888
889
# the root entry gets versioned properly by this builder.
890
_versioned_root = True
891
892
def _check_root(self, ie, parent_invs, tree):
893
"""Helper for record_entry_contents.
894
895
:param ie: An entry being added.
896
:param parent_invs: The inventories of the parent revisions of the
897
commit.
898
:param tree: The tree that is being committed.
899
"""
900
901
def _require_root_change(self, tree):
902
"""Enforce an appropriate root object change.
903
904
This is called once when record_iter_changes is called, if and only if
905
the root was not in the delta calculated by record_iter_changes.
906
907
:param tree: The tree which is being committed.
908
"""
909
# versioned roots do not change unless the tree found a change.
220
raise NotImplementedError(self.record_iter_changes)
910
221
911
222
912
223
class RepositoryWriteLockResult(LogicalLockResult):
937
248
revisions and file history. It's normally accessed only by the Branch,
938
249
which views a particular line of development through that history.
939
250
940
The Repository builds on top of some byte storage facilies (the revisions,
941
signatures, inventories, texts and chk_bytes attributes) and a Transport,
942
which respectively provide byte storage and a means to access the (possibly
943
remote) disk.
944
945
The byte storage facilities are addressed via tuples, which we refer to
946
as 'keys' throughout the code base. Revision_keys, inventory_keys and
947
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
948
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
949
byte string made up of a hash identifier and a hash value.
950
We use this interface because it allows low friction with the underlying
951
code that implements disk indices, network encoding and other parts of
952
bzrlib.
953
954
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
955
the serialised revisions for the repository. This can be used to obtain
956
revision graph information or to access raw serialised revisions.
957
The result of trying to insert data into the repository via this store
958
is undefined: it should be considered read-only except for implementors
959
of repositories.
960
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
961
the serialised signatures for the repository. This can be used to
962
obtain access to raw serialised signatures. The result of trying to
963
insert data into the repository via this store is undefined: it should
964
be considered read-only except for implementors of repositories.
965
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
966
the serialised inventories for the repository. This can be used to
967
obtain unserialised inventories. The result of trying to insert data
968
into the repository via this store is undefined: it should be
969
considered read-only except for implementors of repositories.
970
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
971
texts of files and directories for the repository. This can be used to
972
obtain file texts or file graphs. Note that Repository.iter_file_bytes
973
is usually a better interface for accessing file texts.
974
The result of trying to insert data into the repository via this store
975
is undefined: it should be considered read-only except for implementors
976
of repositories.
977
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
978
any data the repository chooses to store or have indexed by its hash.
979
The result of trying to insert data into the repository via this store
980
is undefined: it should be considered read-only except for implementors
981
of repositories.
982
:ivar _transport: Transport for file access to repository, typically
983
pointing to .bzr/repository.
251
See VersionedFileRepository in bzrlib.vf_repository for the
252
base class for most Bazaar repositories.
984
253
"""
985
254
986
# What class to use for a CommitBuilder. Often it's simpler to change this
987
# in a Repository class subclass rather than to override
988
# get_commit_builder.
989
_commit_builder_class = CommitBuilder
990
991
255
def abort_write_group(self, suppress_errors=False):
992
256
"""Commit the contents accrued within the current write group.
993
257
1036
300
1037
301
:param repository: A repository.
1038
302
"""
1039
if not self._format.supports_external_lookups:
1040
raise errors.UnstackableRepositoryFormat(self._format, self.base)
1041
if self.is_locked():
1042
# This repository will call fallback.unlock() when we transition to
1043
# the unlocked state, so we make sure to increment the lock count
1044
repository.lock_read()
1045
self._check_fallback_repository(repository)
1046
self._fallback_repositories.append(repository)
1047
self.texts.add_fallback_versioned_files(repository.texts)
1048
self.inventories.add_fallback_versioned_files(repository.inventories)
1049
self.revisions.add_fallback_versioned_files(repository.revisions)
1050
self.signatures.add_fallback_versioned_files(repository.signatures)
1051
if self.chk_bytes is not None:
1052
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
303
raise NotImplementedError(self.add_fallback_repository)
1053
304
1054
305
def _check_fallback_repository(self, repository):
1055
306
"""Check that this repository can fallback to repository safely.
1060
311
"""
1061
312
return InterRepository._assert_same_model(self, repository)
1062
313
1063
def add_inventory(self, revision_id, inv, parents):
1064
"""Add the inventory inv to the repository as revision_id.
1065
1066
:param parents: The revision ids of the parents that revision_id
1067
is known to have and are in the repository already.
1068
1069
:returns: The validator(which is a sha1 digest, though what is sha'd is
1070
repository format specific) of the serialized inventory.
1071
"""
1072
if not self.is_in_write_group():
1073
raise AssertionError("%r not in write group" % (self,))
1074
_mod_revision.check_not_reserved_id(revision_id)
1075
if not (inv.revision_id is None or inv.revision_id == revision_id):
1076
raise AssertionError(
1077
"Mismatch between inventory revision"
1078
" id and insertion revid (%r, %r)"
1079
% (inv.revision_id, revision_id))
1080
if inv.root is None:
1081
raise errors.RootMissing()
1082
return self._add_inventory_checked(revision_id, inv, parents)
1083
1084
def _add_inventory_checked(self, revision_id, inv, parents):
1085
"""Add inv to the repository after checking the inputs.
1086
1087
This function can be overridden to allow different inventory styles.
1088
1089
:seealso: add_inventory, for the contract.
1090
"""
1091
inv_lines = self._serializer.write_inventory_to_lines(inv)
1092
return self._inventory_add_lines(revision_id, parents,
1093
inv_lines, check_content=False)
1094
1095
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1096
parents, basis_inv=None, propagate_caches=False):
1097
"""Add a new inventory expressed as a delta against another revision.
1098
1099
See the inventory developers documentation for the theory behind
1100
inventory deltas.
1101
1102
:param basis_revision_id: The inventory id the delta was created
1103
against. (This does not have to be a direct parent.)
1104
:param delta: The inventory delta (see Inventory.apply_delta for
1105
details).
1106
:param new_revision_id: The revision id that the inventory is being
1107
added for.
1108
:param parents: The revision ids of the parents that revision_id is
1109
known to have and are in the repository already. These are supplied
1110
for repositories that depend on the inventory graph for revision
1111
graph access, as well as for those that pun ancestry with delta
1112
compression.
1113
:param basis_inv: The basis inventory if it is already known,
1114
otherwise None.
1115
:param propagate_caches: If True, the caches for this inventory are
1116
copied to and updated for the result if possible.
1117
1118
:returns: (validator, new_inv)
1119
The validator(which is a sha1 digest, though what is sha'd is
1120
repository format specific) of the serialized inventory, and the
1121
resulting inventory.
1122
"""
1123
if not self.is_in_write_group():
1124
raise AssertionError("%r not in write group" % (self,))
1125
_mod_revision.check_not_reserved_id(new_revision_id)
1126
basis_tree = self.revision_tree(basis_revision_id)
1127
basis_tree.lock_read()
1128
try:
1129
# Note that this mutates the inventory of basis_tree, which not all
1130
# inventory implementations may support: A better idiom would be to
1131
# return a new inventory, but as there is no revision tree cache in
1132
# repository this is safe for now - RBC 20081013
1133
if basis_inv is None:
1134
basis_inv = basis_tree.inventory
1135
basis_inv.apply_delta(delta)
1136
basis_inv.revision_id = new_revision_id
1137
return (self.add_inventory(new_revision_id, basis_inv, parents),
1138
basis_inv)
1139
finally:
1140
basis_tree.unlock()
1141
1142
def _inventory_add_lines(self, revision_id, parents, lines,
1143
check_content=True):
1144
"""Store lines in inv_vf and return the sha1 of the inventory."""
1145
parents = [(parent,) for parent in parents]
1146
result = self.inventories.add_lines((revision_id,), parents, lines,
1147
check_content=check_content)[0]
1148
self.inventories._access.flush()
1149
return result
1150
1151
def add_revision(self, revision_id, rev, inv=None, config=None):
1152
"""Add rev to the revision store as revision_id.
1153
1154
:param revision_id: the revision id to use.
1155
:param rev: The revision object.
1156
:param inv: The inventory for the revision. if None, it will be looked
1157
up in the inventory storer
1158
:param config: If None no digital signature will be created.
1159
If supplied its signature_needed method will be used
1160
to determine if a signature should be made.
1161
"""
1162
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1163
# rev.parent_ids?
1164
_mod_revision.check_not_reserved_id(revision_id)
1165
if config is not None and config.signature_needed():
1166
if inv is None:
1167
inv = self.get_inventory(revision_id)
1168
plaintext = Testament(rev, inv).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
1787
726
1788
727
def get_commit_builder(self, branch, parents, config, timestamp=None,
1789
728
timezone=None, committer=None, revprops=None,
1790
revision_id=None):
729
revision_id=None, lossy=False):
1791
730
"""Obtain a CommitBuilder for this repository.
1792
731
1793
732
:param branch: Branch to commit to.
1798
737
:param committer: Optional committer to set for commit.
1799
738
:param revprops: Optional dictionary of revision properties.
1800
739
:param revision_id: Optional revision id.
740
:param lossy: Whether to discard data that can not be natively
741
represented, when pushing to a foreign VCS
1801
742
"""
1802
if self._fallback_repositories and not self._format.supports_chks:
1803
raise errors.BzrError("Cannot commit directly to a stacked branch"
1804
" in pre-2a formats. See "
1805
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1806
result = self._commit_builder_class(self, parents, config,
1807
timestamp, timezone, committer, revprops, revision_id)
1808
self.start_write_group()
1809
return result
743
raise NotImplementedError(self.get_commit_builder)
1810
744
1811
745
@only_raises(errors.LockNotHeld, errors.LockBroken)
1812
746
def unlock(self):
1819
753
'Must end write groups before releasing write locks.')
1820
754
self.control_files.unlock()
1821
755
if self.control_files._lock_count == 0:
1822
self._inventory_entry_cache.clear()
1823
756
for repo in self._fallback_repositories:
1824
757
repo.unlock()
1825
758
1890
823
dest_repo = a_bzrdir.open_repository()
1891
824
return dest_repo
1892
825
1893
def _get_sink(self):
1894
"""Return a sink for streaming into this repository."""
1895
return StreamSink(self)
1896
1897
def _get_source(self, to_format):
1898
"""Return a source for streaming from this repository."""
1899
return StreamSource(self, to_format)
1900
1901
826
@needs_read_lock
1902
827
def has_revision(self, revision_id):
1903
828
"""True if this repository has a copy of the revision."""
1910
835
:param revision_ids: An iterable of revision_ids.
1911
836
:return: A set of the revision_ids that were present.
1912
837
"""
1913
parent_map = self.revisions.get_parent_map(
1914
[(rev_id,) for rev_id in revision_ids])
1915
result = set()
1916
if _mod_revision.NULL_REVISION in revision_ids:
1917
result.add(_mod_revision.NULL_REVISION)
1918
result.update([key[0] for key in parent_map])
1919
return result
838
raise NotImplementedError(self.has_revisions)
1920
839
1921
840
@needs_read_lock
1922
841
def get_revision(self, revision_id):
1923
842
"""Return the Revision object for a named revision."""
1924
843
return self.get_revisions([revision_id])[0]
1925
844
1926
@needs_read_lock
1927
845
def get_revision_reconcile(self, revision_id):
1928
846
"""'reconcile' helper routine that allows access to a revision always.
1929
847
1932
850
be used by reconcile, or reconcile-alike commands that are correcting
1933
851
or testing the revision graph.
1934
852
"""
1935
return self._get_revisions([revision_id])[0]
853
raise NotImplementedError(self.get_revision_reconcile)
1936
854
1937
@needs_read_lock
1938
855
def get_revisions(self, revision_ids):
1939
856
"""Get many revisions at once.
1940
857
1941
858
Repositories that need to check data on every revision read should
1942
859
subclass this method.
1943
860
"""
1944
return self._get_revisions(revision_ids)
1945
1946
@needs_read_lock
1947
def _get_revisions(self, revision_ids):
1948
"""Core work logic to get many revisions without sanity checks."""
1949
revs = {}
1950
for revid, rev in self._iter_revisions(revision_ids):
1951
if rev is None:
1952
raise errors.NoSuchRevision(self, revid)
1953
revs[revid] = rev
1954
return [revs[revid] for revid in revision_ids]
1955
1956
def _iter_revisions(self, revision_ids):
1957
"""Iterate over revision objects.
1958
1959
:param revision_ids: An iterable of revisions to examine. None may be
1960
passed to request all revisions known to the repository. Note that
1961
not all repositories can find unreferenced revisions; for those
1962
repositories only referenced ones will be returned.
1963
:return: An iterator of (revid, revision) tuples. Absent revisions (
1964
those asked for but not available) are returned as (revid, None).
1965
"""
1966
if revision_ids is None:
1967
revision_ids = self.all_revision_ids()
1968
else:
1969
for rev_id in revision_ids:
1970
if not rev_id or not isinstance(rev_id, basestring):
1971
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1972
keys = [(key,) for key in revision_ids]
1973
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1974
for record in stream:
1975
revid = record.key[0]
1976
if record.storage_kind == 'absent':
1977
yield (revid, None)
1978
else:
1979
text = record.get_bytes_as('fulltext')
1980
rev = self._serializer.read_revision_from_string(text)
1981
yield (revid, rev)
861
raise NotImplementedError(self.get_revisions)
1982
862
1983
863
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
1984
864
"""Produce a generator of revision deltas.
2039
919
signature = gpg_strategy.sign(plaintext)
2040
920
self.add_signature_text(revision_id, signature)
2041
921
2042
@needs_write_lock
2043
922
def add_signature_text(self, revision_id, signature):
2044
self.signatures.add_lines((revision_id,), (),
2045
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)
2046
929
2047
930
def find_text_key_references(self):
2048
931
"""Find the text key references within the repository.
2049
932
2050
933
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2051
to whether they were referred to by the inventory of the
2052
revision_id that they contain. The inventory texts from all present
2053
revision ids are assessed to generate this report.
2054
"""
2055
revision_keys = self.revisions.keys()
2056
w = self.inventories
2057
pb = ui.ui_factory.nested_progress_bar()
2058
try:
2059
return self._serializer._find_text_key_references(
2060
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
2061
finally:
2062
pb.finished()
2063
2064
def _inventory_xml_lines_for_keys(self, keys):
2065
"""Get a line iterator of the sort needed for findind references.
2066
2067
Not relevant for non-xml inventory repositories.
2068
2069
Ghosts in revision_keys are ignored.
2070
2071
:param revision_keys: The revision keys for the inventories to inspect.
2072
:return: An iterator over (inventory line, revid) for the fulltexts of
2073
all of the xml inventories specified by revision_keys.
2074
"""
2075
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2076
for record in stream:
2077
if record.storage_kind != 'absent':
2078
chunks = record.get_bytes_as('chunked')
2079
revid = record.key[-1]
2080
lines = osutils.chunks_to_lines(chunks)
2081
for line in lines:
2082
yield line, revid
2083
2084
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2085
revision_keys):
2086
"""Helper routine for fileids_altered_by_revision_ids.
2087
2088
This performs the translation of xml lines to revision ids.
2089
2090
:param line_iterator: An iterator of lines, origin_version_id
2091
:param revision_keys: The revision ids to filter for. This should be a
2092
set or other type which supports efficient __contains__ lookups, as
2093
the revision key from each parsed line will be looked up in the
2094
revision_keys filter.
2095
:return: a dictionary mapping altered file-ids to an iterable of
2096
revision_ids. Each altered file-ids has the exact revision_ids that
2097
altered it listed explicitly.
2098
"""
2099
seen = set(self._serializer._find_text_key_references(
2100
line_iterator).iterkeys())
2101
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2102
parent_seen = set(self._serializer._find_text_key_references(
2103
self._inventory_xml_lines_for_keys(parent_keys)))
2104
new_keys = seen - parent_seen
2105
result = {}
2106
setdefault = result.setdefault
2107
for key in new_keys:
2108
setdefault(key[0], set()).add(key[-1])
2109
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)
2110
938
2111
939
def _find_parent_ids_of_revisions(self, revision_ids):
2112
940
"""Find all parent ids that are mentioned in the revision graph.
2121
949
parent_ids.discard(_mod_revision.NULL_REVISION)
2122
950
return parent_ids
2123
951
2124
def _find_parent_keys_of_revisions(self, revision_keys):
2125
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2126
2127
:param revision_keys: An iterable of revision_keys.
2128
:return: The parents of all revision_keys that are not already in
2129
revision_keys
2130
"""
2131
parent_map = self.revisions.get_parent_map(revision_keys)
2132
parent_keys = set()
2133
map(parent_keys.update, parent_map.itervalues())
2134
parent_keys.difference_update(revision_keys)
2135
parent_keys.discard(_mod_revision.NULL_REVISION)
2136
return parent_keys
2137
2138
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
952
def fileids_altered_by_revision_ids(self, revision_ids):
2139
953
"""Find the file ids and versions affected by revisions.
2140
954
2141
955
:param revisions: an iterable containing revision ids.
2142
:param _inv_weave: The inventory weave from this repository or None.
2143
If None, the inventory weave will be opened automatically.
2144
956
:return: a dictionary mapping altered file-ids to an iterable of
2145
revision_ids. Each altered file-ids has the exact revision_ids that
2146
altered it listed explicitly.
957
revision_ids. Each altered file-ids has the exact revision_ids
958
that altered it listed explicitly.
2147
959
"""
2148
selected_keys = set((revid,) for revid in revision_ids)
2149
w = _inv_weave or self.inventories
2150
return self._find_file_ids_from_xml_inventory_lines(
2151
w.iter_lines_added_or_present_in_keys(
2152
selected_keys, pb=None),
2153
selected_keys)
960
raise NotImplementedError(self.fileids_altered_by_revision_ids)
2154
961
2155
962
def iter_files_bytes(self, desired_files):
2156
963
"""Iterate through file versions.
2163
970
uniquely identify the file version in the caller's context. (Examples:
2164
971
an index number or a TreeTransform trans_id.)
2165
972
2166
bytes_iterator is an iterable of bytestrings for the file. The
2167
kind of iterable and length of the bytestrings are unspecified, but for
2168
this implementation, it is a list of bytes produced by
2169
VersionedFile.get_record_stream().
2170
2171
973
:param desired_files: a list of (file_id, revision_id, identifier)
2172
974
triples
2173
975
"""
2174
text_keys = {}
2175
for file_id, revision_id, callable_data in desired_files:
2176
text_keys[(file_id, revision_id)] = callable_data
2177
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2178
if record.storage_kind == 'absent':
2179
raise errors.RevisionNotPresent(record.key, self)
2180
yield text_keys[record.key], record.get_bytes_as('chunked')
2181
2182
def _generate_text_key_index(self, text_key_references=None,
2183
ancestors=None):
2184
"""Generate a new text key index for the repository.
2185
2186
This is an expensive function that will take considerable time to run.
2187
2188
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2189
list of parents, also text keys. When a given key has no parents,
2190
the parents list will be [NULL_REVISION].
2191
"""
2192
# All revisions, to find inventory parents.
2193
if ancestors is None:
2194
graph = self.get_graph()
2195
ancestors = graph.get_parent_map(self.all_revision_ids())
2196
if text_key_references is None:
2197
text_key_references = self.find_text_key_references()
2198
pb = ui.ui_factory.nested_progress_bar()
2199
try:
2200
return self._do_generate_text_key_index(ancestors,
2201
text_key_references, pb)
2202
finally:
2203
pb.finished()
2204
2205
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2206
"""Helper for _generate_text_key_index to avoid deep nesting."""
2207
revision_order = tsort.topo_sort(ancestors)
2208
invalid_keys = set()
2209
revision_keys = {}
2210
for revision_id in revision_order:
2211
revision_keys[revision_id] = set()
2212
text_count = len(text_key_references)
2213
# a cache of the text keys to allow reuse; costs a dict of all the
2214
# keys, but saves a 2-tuple for every child of a given key.
2215
text_key_cache = {}
2216
for text_key, valid in text_key_references.iteritems():
2217
if not valid:
2218
invalid_keys.add(text_key)
2219
else:
2220
revision_keys[text_key[1]].add(text_key)
2221
text_key_cache[text_key] = text_key
2222
del text_key_references
2223
text_index = {}
2224
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2225
NULL_REVISION = _mod_revision.NULL_REVISION
2226
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2227
# too small for large or very branchy trees. However, for 55K path
2228
# trees, it would be easy to use too much memory trivially. Ideally we
2229
# could gauge this by looking at available real memory etc, but this is
2230
# always a tricky proposition.
2231
inventory_cache = lru_cache.LRUCache(10)
2232
batch_size = 10 # should be ~150MB on a 55K path tree
2233
batch_count = len(revision_order) / batch_size + 1
2234
processed_texts = 0
2235
pb.update("Calculating text parents", processed_texts, text_count)
2236
for offset in xrange(batch_count):
2237
to_query = revision_order[offset * batch_size:(offset + 1) *
2238
batch_size]
2239
if not to_query:
2240
break
2241
for revision_id in to_query:
2242
parent_ids = ancestors[revision_id]
2243
for text_key in revision_keys[revision_id]:
2244
pb.update("Calculating text parents", processed_texts)
2245
processed_texts += 1
2246
candidate_parents = []
2247
for parent_id in parent_ids:
2248
parent_text_key = (text_key[0], parent_id)
2249
try:
2250
check_parent = parent_text_key not in \
2251
revision_keys[parent_id]
2252
except KeyError:
2253
# the parent parent_id is a ghost:
2254
check_parent = False
2255
# truncate the derived graph against this ghost.
2256
parent_text_key = None
2257
if check_parent:
2258
# look at the parent commit details inventories to
2259
# determine possible candidates in the per file graph.
2260
# TODO: cache here.
2261
try:
2262
inv = inventory_cache[parent_id]
2263
except KeyError:
2264
inv = self.revision_tree(parent_id).inventory
2265
inventory_cache[parent_id] = inv
2266
try:
2267
parent_entry = inv[text_key[0]]
2268
except (KeyError, errors.NoSuchId):
2269
parent_entry = None
2270
if parent_entry is not None:
2271
parent_text_key = (
2272
text_key[0], parent_entry.revision)
2273
else:
2274
parent_text_key = None
2275
if parent_text_key is not None:
2276
candidate_parents.append(
2277
text_key_cache[parent_text_key])
2278
parent_heads = text_graph.heads(candidate_parents)
2279
new_parents = list(parent_heads)
2280
new_parents.sort(key=lambda x:candidate_parents.index(x))
2281
if new_parents == []:
2282
new_parents = [NULL_REVISION]
2283
text_index[text_key] = new_parents
2284
2285
for text_key in invalid_keys:
2286
text_index[text_key] = [NULL_REVISION]
2287
return text_index
2288
2289
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2290
"""Get an iterable listing the keys of all the data introduced by a set
2291
of revision IDs.
2292
2293
The keys will be ordered so that the corresponding items can be safely
2294
fetched and inserted in that order.
2295
2296
:returns: An iterable producing tuples of (knit-kind, file-id,
2297
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2298
'revisions'. file-id is None unless knit-kind is 'file'.
2299
"""
2300
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2301
yield result
2302
del _files_pb
2303
for result in self._find_non_file_keys_to_fetch(revision_ids):
2304
yield result
2305
2306
def _find_file_keys_to_fetch(self, revision_ids, pb):
2307
# XXX: it's a bit weird to control the inventory weave caching in this
2308
# generator. Ideally the caching would be done in fetch.py I think. Or
2309
# maybe this generator should explicitly have the contract that it
2310
# should not be iterated until the previously yielded item has been
2311
# processed?
2312
inv_w = self.inventories
2313
2314
# file ids that changed
2315
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2316
count = 0
2317
num_file_ids = len(file_ids)
2318
for file_id, altered_versions in file_ids.iteritems():
2319
if pb is not None:
2320
pb.update("Fetch texts", count, num_file_ids)
2321
count += 1
2322
yield ("file", file_id, altered_versions)
2323
2324
def _find_non_file_keys_to_fetch(self, revision_ids):
2325
# inventory
2326
yield ("inventory", None, revision_ids)
2327
2328
# signatures
2329
# XXX: Note ATM no callers actually pay attention to this return
2330
# instead they just use the list of revision ids and ignore
2331
# missing sigs. Consider removing this work entirely
2332
revisions_with_signatures = set(self.signatures.get_parent_map(
2333
[(r,) for r in revision_ids]))
2334
revisions_with_signatures = set(
2335
[r for (r,) in revisions_with_signatures])
2336
revisions_with_signatures.intersection_update(revision_ids)
2337
yield ("signatures", None, revisions_with_signatures)
2338
2339
# revisions
2340
yield ("revisions", None, revision_ids)
2341
2342
@needs_read_lock
2343
def get_inventory(self, revision_id):
2344
"""Get Inventory object by revision id."""
2345
return self.iter_inventories([revision_id]).next()
2346
2347
def iter_inventories(self, revision_ids, ordering=None):
2348
"""Get many inventories by revision_ids.
2349
2350
This will buffer some or all of the texts used in constructing the
2351
inventories in memory, but will only parse a single inventory at a
2352
time.
2353
2354
:param revision_ids: The expected revision ids of the inventories.
2355
:param ordering: optional ordering, e.g. 'topological'. If not
2356
specified, the order of revision_ids will be preserved (by
2357
buffering if necessary).
2358
:return: An iterator of inventories.
2359
"""
2360
if ((None in revision_ids)
2361
or (_mod_revision.NULL_REVISION in revision_ids)):
2362
raise ValueError('cannot get null revision inventory')
2363
return self._iter_inventories(revision_ids, ordering)
2364
2365
def _iter_inventories(self, revision_ids, ordering):
2366
"""single-document based inventory iteration."""
2367
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2368
for text, revision_id in inv_xmls:
2369
yield self._deserialise_inventory(revision_id, text)
2370
2371
def _iter_inventory_xmls(self, revision_ids, ordering):
2372
if ordering is None:
2373
order_as_requested = True
2374
ordering = 'unordered'
2375
else:
2376
order_as_requested = False
2377
keys = [(revision_id,) for revision_id in revision_ids]
2378
if not keys:
2379
return
2380
if order_as_requested:
2381
key_iter = iter(keys)
2382
next_key = key_iter.next()
2383
stream = self.inventories.get_record_stream(keys, ordering, True)
2384
text_chunks = {}
2385
for record in stream:
2386
if record.storage_kind != 'absent':
2387
chunks = record.get_bytes_as('chunked')
2388
if order_as_requested:
2389
text_chunks[record.key] = chunks
2390
else:
2391
yield ''.join(chunks), record.key[-1]
2392
else:
2393
raise errors.NoSuchRevision(self, record.key)
2394
if order_as_requested:
2395
# Yield as many results as we can while preserving order.
2396
while next_key in text_chunks:
2397
chunks = text_chunks.pop(next_key)
2398
yield ''.join(chunks), next_key[-1]
2399
try:
2400
next_key = key_iter.next()
2401
except StopIteration:
2402
# We still want to fully consume the get_record_stream,
2403
# just in case it is not actually finished at this point
2404
next_key = None
2405
break
2406
2407
def _deserialise_inventory(self, revision_id, xml):
2408
"""Transform the xml into an inventory object.
2409
2410
:param revision_id: The expected revision id of the inventory.
2411
:param xml: A serialised inventory.
2412
"""
2413
result = self._serializer.read_inventory_from_string(xml, revision_id,
2414
entry_cache=self._inventory_entry_cache,
2415
return_from_cache=self._safe_to_return_from_cache)
2416
if result.revision_id != revision_id:
2417
raise AssertionError('revision id mismatch %s != %s' % (
2418
result.revision_id, revision_id))
2419
return result
2420
2421
def get_serializer_format(self):
2422
return self._serializer.format_num
2423
2424
@needs_read_lock
2425
def _get_inventory_xml(self, revision_id):
2426
"""Get serialized inventory as a string."""
2427
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2428
try:
2429
text, revision_id = texts.next()
2430
except StopIteration:
2431
raise errors.HistoryMissing(self, 'inventory', revision_id)
2432
return text
976
raise NotImplementedError(self.iter_files_bytes)
2433
977
2434
978
def get_rev_id_for_revno(self, revno, known_pair):
2435
979
"""Return the revision id of a revno, given a later (revno, revid)
2508
1052
2509
1053
`revision_id` may be NULL_REVISION for the empty tree revision.
2510
1054
"""
2511
revision_id = _mod_revision.ensure_null(revision_id)
2512
# TODO: refactor this to use an existing revision object
2513
# so we don't need to read it in twice.
2514
if revision_id == _mod_revision.NULL_REVISION:
2515
return RevisionTree(self, Inventory(root_id=None),
2516
_mod_revision.NULL_REVISION)
2517
else:
2518
inv = self.get_inventory(revision_id)
2519
return RevisionTree(self, inv, revision_id)
1055
raise NotImplementedError(self.revision_tree)
2520
1056
2521
1057
def revision_trees(self, revision_ids):
2522
1058
"""Return Trees for revisions in this repository.
2524
1060
:param revision_ids: a sequence of revision-ids;
2525
1061
a revision-id may not be None or 'null:'
2526
1062
"""
2527
inventories = self.iter_inventories(revision_ids)
2528
for inv in inventories:
2529
yield RevisionTree(self, inv, inv.revision_id)
2530
2531
def _filtered_revision_trees(self, revision_ids, file_ids):
2532
"""Return Tree for a revision on this branch with only some files.
2533
2534
:param revision_ids: a sequence of revision-ids;
2535
a revision-id may not be None or 'null:'
2536
:param file_ids: if not None, the result is filtered
2537
so that only those file-ids, their parents and their
2538
children are included.
2539
"""
2540
inventories = self.iter_inventories(revision_ids)
2541
for inv in inventories:
2542
# Should we introduce a FilteredRevisionTree class rather
2543
# than pre-filter the inventory here?
2544
filtered_inv = inv.filter(file_ids)
2545
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
1063
raise NotImplementedError(self.revision_trees)
2546
1064
2547
1065
@needs_read_lock
2548
1066
def get_ancestry(self, revision_id, topo_sorted=True):
2600
1118
2601
1119
def get_parent_map(self, revision_ids):
2602
1120
"""See graph.StackedParentsProvider.get_parent_map"""
1121
raise NotImplementedError(self.get_parent_map)
1122
1123
def _get_parent_map_no_fallbacks(self, revision_ids):
1124
"""Same as Repository.get_parent_map except doesn't query fallbacks."""
2603
1125
# revisions index works in keys; this just works in revisions
2604
1126
# therefore wrap and unwrap
2605
1127
query_keys = []
2611
1133
raise ValueError('get_parent_map(None) is not valid')
2612
1134
else:
2613
1135
query_keys.append((revision_id ,))
1136
vf = self.revisions.without_fallbacks()
2614
1137
for ((revision_id,), parent_keys) in \
2615
self.revisions.get_parent_map(query_keys).iteritems():
1138
vf.get_parent_map(query_keys).iteritems():
2616
1139
if parent_keys:
2617
1140
result[revision_id] = tuple([parent_revid
2618
1141
for (parent_revid,) in parent_keys])
2621
1144
return result
2622
1145
2623
1146
def _make_parents_provider(self):
2624
return self
1147
if not self._format.supports_external_lookups:
1148
return self
1149
return graph.StackedParentsProvider(_LazyListJoin(
1150
[self._make_parents_provider_unstacked()],
1151
self._fallback_repositories))
1152
1153
def _make_parents_provider_unstacked(self):
1154
return graph.CallableToParentsProviderAdapter(
1155
self._get_parent_map_no_fallbacks)
2625
1156
2626
1157
@needs_read_lock
2627
1158
def get_known_graph_ancestry(self, revision_ids):
2628
1159
"""Return the known graph for a set of revision ids and their ancestors.
2629
1160
"""
2630
st = static_tuple.StaticTuple
2631
revision_keys = [st(r_id).intern() for r_id in revision_ids]
2632
known_graph = self.revisions.get_known_graph_ancestry(revision_keys)
2633
return graph.GraphThunkIdsToKeys(known_graph)
1161
raise NotImplementedError(self.get_known_graph_ancestry)
2634
1162
2635
1163
def get_graph(self, other_repository=None):
2636
1164
"""Return the graph walker for this repository format"""
2641
1169
[parents_provider, other_repository._make_parents_provider()])
2642
1170
return graph.Graph(parents_provider)
2643
1171
2644
def _get_versioned_file_checker(self, text_key_references=None,
2645
ancestors=None):
2646
"""Return an object suitable for checking versioned files.
2647
2648
:param text_key_references: if non-None, an already built
2649
dictionary mapping text keys ((fileid, revision_id) tuples)
2650
to whether they were referred to by the inventory of the
2651
revision_id that they contain. If None, this will be
2652
calculated.
2653
:param ancestors: Optional result from
2654
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2655
available.
2656
"""
2657
return _VersionedFileChecker(self,
2658
text_key_references=text_key_references, ancestors=ancestors)
2659
2660
1172
def revision_ids_to_search_result(self, result_set):
2661
1173
"""Convert a set of revision ids to a graph SearchResult."""
2662
1174
result_parents = set()
2688
1200
2689
1201
@needs_write_lock
2690
1202
def sign_revision(self, revision_id, gpg_strategy):
2691
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()
2692
1205
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
2693
1206
2694
@needs_read_lock
2695
1207
def has_signature_for_revision_id(self, revision_id):
2696
1208
"""Query for a revision signature for revision_id in the repository."""
2697
if not self.has_revision(revision_id):
2698
raise errors.NoSuchRevision(self, revision_id)
2699
sig_present = (1 == len(
2700
self.signatures.get_parent_map([(revision_id,)])))
2701
return sig_present
1209
raise NotImplementedError(self.has_signature_for_revision_id)
2702
1210
2703
@needs_read_lock
2704
1211
def get_signature_text(self, revision_id):
2705
1212
"""Return the text for a signature."""
2706
stream = self.signatures.get_record_stream([(revision_id,)],
2707
'unordered', True)
2708
record = stream.next()
2709
if record.storage_kind == 'absent':
2710
raise errors.NoSuchRevision(self, revision_id)
2711
return record.get_bytes_as('fulltext')
1213
raise NotImplementedError(self.get_signature_text)
2712
1214
2713
1215
@needs_read_lock
2714
1216
def check(self, revision_ids=None, callback_refs=None, check_repo=True):
2770
1272
except UnicodeDecodeError:
2771
1273
raise errors.NonAsciiRevisionId(method, self)
2772
1274
2773
def revision_graph_can_have_wrong_parents(self):
2774
"""Is it possible for this repository to have a revision graph with
2775
incorrect parents?
2776
2777
If True, then this repository must also implement
2778
_find_inconsistent_revision_parents so that check and reconcile can
2779
check for inconsistencies before proceeding with other checks that may
2780
depend on the revision index being consistent.
2781
"""
2782
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2783
2784
2785
def install_revision(repository, rev, revision_tree):
2786
"""Install all revision data into a repository."""
2787
install_revisions(repository, [(rev, revision_tree, None)])
2788
2789
2790
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2791
"""Install all revision data into a repository.
2792
2793
Accepts an iterable of revision, tree, signature tuples. The signature
2794
may be None.
2795
"""
2796
repository.start_write_group()
2797
try:
2798
inventory_cache = lru_cache.LRUCache(10)
2799
for n, (revision, revision_tree, signature) in enumerate(iterable):
2800
_install_revision(repository, revision, revision_tree, signature,
2801
inventory_cache)
2802
if pb is not None:
2803
pb.update('Transferring revisions', n + 1, num_revisions)
2804
except:
2805
repository.abort_write_group()
2806
raise
2807
else:
2808
repository.commit_write_group()
2809
2810
2811
def _install_revision(repository, rev, revision_tree, signature,
2812
inventory_cache):
2813
"""Install all revision data into a repository."""
2814
present_parents = []
2815
parent_trees = {}
2816
for p_id in rev.parent_ids:
2817
if repository.has_revision(p_id):
2818
present_parents.append(p_id)
2819
parent_trees[p_id] = repository.revision_tree(p_id)
2820
else:
2821
parent_trees[p_id] = repository.revision_tree(
2822
_mod_revision.NULL_REVISION)
2823
2824
inv = revision_tree.inventory
2825
entries = inv.iter_entries()
2826
# backwards compatibility hack: skip the root id.
2827
if not repository.supports_rich_root():
2828
path, root = entries.next()
2829
if root.revision != rev.revision_id:
2830
raise errors.IncompatibleRevision(repr(repository))
2831
text_keys = {}
2832
for path, ie in entries:
2833
text_keys[(ie.file_id, ie.revision)] = ie
2834
text_parent_map = repository.texts.get_parent_map(text_keys)
2835
missing_texts = set(text_keys) - set(text_parent_map)
2836
# Add the texts that are not already present
2837
for text_key in missing_texts:
2838
ie = text_keys[text_key]
2839
text_parents = []
2840
# FIXME: TODO: The following loop overlaps/duplicates that done by
2841
# commit to determine parents. There is a latent/real bug here where
2842
# the parents inserted are not those commit would do - in particular
2843
# they are not filtered by heads(). RBC, AB
2844
for revision, tree in parent_trees.iteritems():
2845
if ie.file_id not in tree:
2846
continue
2847
parent_id = tree.inventory[ie.file_id].revision
2848
if parent_id in text_parents:
2849
continue
2850
text_parents.append((ie.file_id, parent_id))
2851
lines = revision_tree.get_file(ie.file_id).readlines()
2852
repository.texts.add_lines(text_key, text_parents, lines)
2853
try:
2854
# install the inventory
2855
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2856
# Cache this inventory
2857
inventory_cache[rev.revision_id] = inv
2858
try:
2859
basis_inv = inventory_cache[rev.parent_ids[0]]
2860
except KeyError:
2861
repository.add_inventory(rev.revision_id, inv, present_parents)
2862
else:
2863
delta = inv._make_delta(basis_inv)
2864
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2865
rev.revision_id, present_parents)
2866
else:
2867
repository.add_inventory(rev.revision_id, inv, present_parents)
2868
except errors.RevisionAlreadyPresent:
2869
pass
2870
if signature is not None:
2871
repository.add_signature_text(rev.revision_id, signature)
2872
repository.add_revision(rev.revision_id, rev, inv)
2873
2874
1275
2875
1276
class MetaDirRepository(Repository):
2876
1277
"""Repositories in the new meta-dir layout.
2911
1312
return not self._transport.has('no-working-trees')
2912
1313
2913
1314
2914
class MetaDirVersionedFileRepository(MetaDirRepository):
2915
"""Repositories in a meta-dir, that work via versioned file objects."""
2916
2917
def __init__(self, _format, a_bzrdir, control_files):
2918
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2919
control_files)
2920
2921
2922
1315
class RepositoryFormatRegistry(controldir.ControlComponentFormatRegistry):
2923
1316
"""Repository format registry."""
2924
1317
2989
1382
# Does this format support CHK bytestring lookups. Set to True or False in
2990
1383
# derived classes.
2991
1384
supports_chks = None
2992
# Should commit add an inventory, or an inventory delta to the repository.
2993
_commit_inv_deltas = True
2994
# What order should fetch operations request streams in?
2995
# The default is unordered as that is the cheapest for an origin to
2996
# provide.
2997
_fetch_order = 'unordered'
2998
# Does this repository format use deltas that can be fetched as-deltas ?
2999
# (E.g. knits, where the knit deltas can be transplanted intact.
3000
# We default to False, which will ensure that enough data to get
3001
# a full text out of any fetch stream will be grabbed.
3002
_fetch_uses_deltas = False
3003
1385
# Should fetch trigger a reconcile after the fetch? Only needed for
3004
1386
# some repository formats that can suffer internal inconsistencies.
3005
1387
_fetch_reconcile = False
3011
1393
# help), and for fetching when data won't have come from the same
3012
1394
# compressor.
3013
1395
pack_compresses = False
3014
# Does the repository inventory storage understand references to trees?
1396
# Does the repository storage understand references to trees?
3015
1397
supports_tree_reference = None
3016
1398
# Is the format experimental ?
3017
1399
experimental = False
3018
# Does this repository format escape funky characters, or does it create files with
3019
# 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
# ?
3020
1403
supports_funky_characters = None
3021
1404
# Does this repository format support leaving locks?
3022
1405
supports_leaving_lock = None
3023
1406
# Does this format support the full VersionedFiles interface?
3024
1407
supports_full_versioned_files = None
1408
# Does this format support signing revision signatures?
1409
supports_revision_signatures = True
1410
# Can the revision graph have incorrect parents?
1411
revision_graph_can_have_wrong_parents = None
1412
# Does this format support rich root data?
1413
rich_root_data = None
3025
1414
3026
1415
def __repr__(self):
3027
1416
return "%s()" % self.__class__.__name__
3079
1468
"""Return the short description for this format."""
3080
1469
raise NotImplementedError(self.get_format_description)
3081
1470
3082
# TODO: this shouldn't be in the base class, it's specific to things that
3083
# use weaves or knits -- mbp 20070207
3084
def _get_versioned_file_store(self,
3085
name,
3086
transport,
3087
control_files,
3088
prefixed=True,
3089
versionedfile_class=None,
3090
versionedfile_kwargs={},
3091
escaped=False):
3092
if versionedfile_class is None:
3093
versionedfile_class = self._versionedfile_class
3094
weave_transport = control_files._transport.clone(name)
3095
dir_mode = control_files._dir_mode
3096
file_mode = control_files._file_mode
3097
return VersionedFileStore(weave_transport, prefixed=prefixed,
3098
dir_mode=dir_mode,
3099
file_mode=file_mode,
3100
versionedfile_class=versionedfile_class,
3101
versionedfile_kwargs=versionedfile_kwargs,
3102
escaped=escaped)
3103
3104
1471
def initialize(self, a_bzrdir, shared=False):
3105
1472
"""Initialize a repository of this format in a_bzrdir.
3106
1473
3247
1614
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
3248
1615
format_registry.register_lazy(
3249
1616
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
3250
'bzrlib.repofmt.pack_repo',
1617
'bzrlib.repofmt.knitpack_repo',
3251
1618
'RepositoryFormatKnitPack1',
3252
1619
)
3253
1620
format_registry.register_lazy(
3254
1621
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
3255
'bzrlib.repofmt.pack_repo',
1622
'bzrlib.repofmt.knitpack_repo',
3256
1623
'RepositoryFormatKnitPack3',
3257
1624
)
3258
1625
format_registry.register_lazy(
3259
1626
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
3260
'bzrlib.repofmt.pack_repo',
1627
'bzrlib.repofmt.knitpack_repo',
3261
1628
'RepositoryFormatKnitPack4',
3262
1629
)
3263
1630
format_registry.register_lazy(
3264
1631
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
3265
'bzrlib.repofmt.pack_repo',
1632
'bzrlib.repofmt.knitpack_repo',
3266
1633
'RepositoryFormatKnitPack5',
3267
1634
)
3268
1635
format_registry.register_lazy(
3269
1636
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
3270
'bzrlib.repofmt.pack_repo',
1637
'bzrlib.repofmt.knitpack_repo',
3271
1638
'RepositoryFormatKnitPack5RichRoot',
3272
1639
)
3273
1640
format_registry.register_lazy(
3274
1641
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3275
'bzrlib.repofmt.pack_repo',
1642
'bzrlib.repofmt.knitpack_repo',
3276
1643
'RepositoryFormatKnitPack5RichRootBroken',
3277
1644
)
3278
1645
format_registry.register_lazy(
3279
1646
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3280
'bzrlib.repofmt.pack_repo',
1647
'bzrlib.repofmt.knitpack_repo',
3281
1648
'RepositoryFormatKnitPack6',
3282
1649
)
3283
1650
format_registry.register_lazy(
3284
1651
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3285
'bzrlib.repofmt.pack_repo',
1652
'bzrlib.repofmt.knitpack_repo',
3286
1653
'RepositoryFormatKnitPack6RichRoot',
3287
1654
)
3288
1655
format_registry.register_lazy(
3296
1663
format_registry.register_lazy(
3297
1664
("Bazaar development format 2 with subtree support "
3298
1665
"(needs bzr.dev from before 1.8)\n"),
3299
'bzrlib.repofmt.pack_repo',
1666
'bzrlib.repofmt.knitpack_repo',
3300
1667
'RepositoryFormatPackDevelopment2Subtree',
3301
1668
)
3302
1669
format_registry.register_lazy(
3516
1883
raise errors.IncompatibleRepositories(source, target,
3517
1884
"different serializers")
3518
1885
3519
3520
class InterSameDataRepository(InterRepository):
3521
"""Code for converting between repositories that represent the same data.
3522
3523
Data format and model must match for this to work.
3524
"""
3525
3526
@classmethod
3527
def _get_repo_format_to_test(self):
3528
"""Repository format for testing with.
3529
3530
InterSameData can pull from subtree to subtree and from non-subtree to
3531
non-subtree, so we test this with the richest repository format.
3532
"""
3533
from bzrlib.repofmt import knitrepo
3534
return knitrepo.RepositoryFormatKnit3()
3535
3536
@staticmethod
3537
def is_compatible(source, target):
3538
return InterRepository._same_model(source, target)
3539
3540
3541
class InterDifferingSerializer(InterRepository):
3542
3543
1886
@classmethod
3544
1887
def _get_repo_format_to_test(self):
3545
1888
return None
3546
1889
3547
@staticmethod
3548
def is_compatible(source, target):
3549
"""Be compatible with Knit2 source and Knit3 target"""
3550
# This is redundant with format.check_conversion_target(), however that
3551
# raises an exception, and we just want to say "False" as in we won't
3552
# support converting between these formats.
3553
if 'IDS_never' in debug.debug_flags:
3554
return False
3555
if source.supports_rich_root() and not target.supports_rich_root():
3556
return False
3557
if (source._format.supports_tree_reference
3558
and not target._format.supports_tree_reference):
3559
return False
3560
if target._fallback_repositories and target._format.supports_chks:
3561
# IDS doesn't know how to copy CHKs for the parent inventories it
3562
# adds to stacked repos.
3563
return False
3564
if 'IDS_always' in debug.debug_flags:
3565
return True
3566
# Only use this code path for local source and target. IDS does far
3567
# too much IO (both bandwidth and roundtrips) over a network.
3568
if not source.bzrdir.transport.base.startswith('file:///'):
3569
return False
3570
if not target.bzrdir.transport.base.startswith('file:///'):
3571
return False
1890
@classmethod
1891
def is_compatible(cls, source, target):
1892
# The default implementation is compatible with everything
3572
1893
return True
3573
1894
3574
def _get_trees(self, revision_ids, cache):
3575
possible_trees = []
3576
for rev_id in revision_ids:
3577
if rev_id in cache:
3578
possible_trees.append((rev_id, cache[rev_id]))
3579
else:
3580
# Not cached, but inventory might be present anyway.
3581
try:
3582
tree = self.source.revision_tree(rev_id)
3583
except errors.NoSuchRevision:
3584
# Nope, parent is ghost.
3585
pass
3586
else:
3587
cache[rev_id] = tree
3588
possible_trees.append((rev_id, tree))
3589
return possible_trees
3590
3591
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3592
"""Get the best delta and base for this revision.
3593
3594
:return: (basis_id, delta)
3595
"""
3596
deltas = []
3597
# Generate deltas against each tree, to find the shortest.
3598
texts_possibly_new_in_tree = set()
3599
for basis_id, basis_tree in possible_trees:
3600
delta = tree.inventory._make_delta(basis_tree.inventory)
3601
for old_path, new_path, file_id, new_entry in delta:
3602
if new_path is None:
3603
# This file_id isn't present in the new rev, so we don't
3604
# care about it.
3605
continue
3606
if not new_path:
3607
# Rich roots are handled elsewhere...
3608
continue
3609
kind = new_entry.kind
3610
if kind != 'directory' and kind != 'file':
3611
# No text record associated with this inventory entry.
3612
continue
3613
# This is a directory or file that has changed somehow.
3614
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3615
deltas.append((len(delta), basis_id, delta))
3616
deltas.sort()
3617
return deltas[0][1:]
3618
3619
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3620
"""Find all parent revisions that are absent, but for which the
3621
inventory is present, and copy those inventories.
3622
3623
This is necessary to preserve correctness when the source is stacked
3624
without fallbacks configured. (Note that in cases like upgrade the
3625
source may be not have _fallback_repositories even though it is
3626
stacked.)
3627
"""
3628
parent_revs = set()
3629
for parents in parent_map.values():
3630
parent_revs.update(parents)
3631
present_parents = self.source.get_parent_map(parent_revs)
3632
absent_parents = set(parent_revs).difference(present_parents)
3633
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3634
(rev_id,) for rev_id in absent_parents)
3635
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3636
for parent_tree in self.source.revision_trees(parent_inv_ids):
3637
current_revision_id = parent_tree.get_revision_id()
3638
parents_parents_keys = parent_invs_keys_for_stacking[
3639
(current_revision_id,)]
3640
parents_parents = [key[-1] for key in parents_parents_keys]
3641
basis_id = _mod_revision.NULL_REVISION
3642
basis_tree = self.source.revision_tree(basis_id)
3643
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3644
self.target.add_inventory_by_delta(
3645
basis_id, delta, current_revision_id, parents_parents)
3646
cache[current_revision_id] = parent_tree
3647
3648
def _fetch_batch(self, revision_ids, basis_id, cache):
3649
"""Fetch across a few revisions.
3650
3651
:param revision_ids: The revisions to copy
3652
:param basis_id: The revision_id of a tree that must be in cache, used
3653
as a basis for delta when no other base is available
3654
:param cache: A cache of RevisionTrees that we can use.
3655
:return: The revision_id of the last converted tree. The RevisionTree
3656
for it will be in cache
3657
"""
3658
# Walk though all revisions; get inventory deltas, copy referenced
3659
# texts that delta references, insert the delta, revision and
3660
# signature.
3661
root_keys_to_create = set()
3662
text_keys = set()
3663
pending_deltas = []
3664
pending_revisions = []
3665
parent_map = self.source.get_parent_map(revision_ids)
3666
self._fetch_parent_invs_for_stacking(parent_map, cache)
3667
self.source._safe_to_return_from_cache = True
3668
for tree in self.source.revision_trees(revision_ids):
3669
# Find a inventory delta for this revision.
3670
# Find text entries that need to be copied, too.
3671
current_revision_id = tree.get_revision_id()
3672
parent_ids = parent_map.get(current_revision_id, ())
3673
parent_trees = self._get_trees(parent_ids, cache)
3674
possible_trees = list(parent_trees)
3675
if len(possible_trees) == 0:
3676
# There either aren't any parents, or the parents are ghosts,
3677
# so just use the last converted tree.
3678
possible_trees.append((basis_id, cache[basis_id]))
3679
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3680
possible_trees)
3681
revision = self.source.get_revision(current_revision_id)
3682
pending_deltas.append((basis_id, delta,
3683
current_revision_id, revision.parent_ids))
3684
if self._converting_to_rich_root:
3685
self._revision_id_to_root_id[current_revision_id] = \
3686
tree.get_root_id()
3687
# Determine which texts are in present in this revision but not in
3688
# any of the available parents.
3689
texts_possibly_new_in_tree = set()
3690
for old_path, new_path, file_id, entry in delta:
3691
if new_path is None:
3692
# This file_id isn't present in the new rev
3693
continue
3694
if not new_path:
3695
# This is the root
3696
if not self.target.supports_rich_root():
3697
# The target doesn't support rich root, so we don't
3698
# copy
3699
continue
3700
if self._converting_to_rich_root:
3701
# This can't be copied normally, we have to insert
3702
# it specially
3703
root_keys_to_create.add((file_id, entry.revision))
3704
continue
3705
kind = entry.kind
3706
texts_possibly_new_in_tree.add((file_id, entry.revision))
3707
for basis_id, basis_tree in possible_trees:
3708
basis_inv = basis_tree.inventory
3709
for file_key in list(texts_possibly_new_in_tree):
3710
file_id, file_revision = file_key
3711
try:
3712
entry = basis_inv[file_id]
3713
except errors.NoSuchId:
3714
continue
3715
if entry.revision == file_revision:
3716
texts_possibly_new_in_tree.remove(file_key)
3717
text_keys.update(texts_possibly_new_in_tree)
3718
pending_revisions.append(revision)
3719
cache[current_revision_id] = tree
3720
basis_id = current_revision_id
3721
self.source._safe_to_return_from_cache = False
3722
# Copy file texts
3723
from_texts = self.source.texts
3724
to_texts = self.target.texts
3725
if root_keys_to_create:
3726
root_stream = _mod_fetch._new_root_data_stream(
3727
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3728
self.source)
3729
to_texts.insert_record_stream(root_stream)
3730
to_texts.insert_record_stream(from_texts.get_record_stream(
3731
text_keys, self.target._format._fetch_order,
3732
not self.target._format._fetch_uses_deltas))
3733
# insert inventory deltas
3734
for delta in pending_deltas:
3735
self.target.add_inventory_by_delta(*delta)
3736
if self.target._fallback_repositories:
3737
# Make sure this stacked repository has all the parent inventories
3738
# for the new revisions that we are about to insert. We do this
3739
# before adding the revisions so that no revision is added until
3740
# all the inventories it may depend on are added.
3741
# Note that this is overzealous, as we may have fetched these in an
3742
# earlier batch.
3743
parent_ids = set()
3744
revision_ids = set()
3745
for revision in pending_revisions:
3746
revision_ids.add(revision.revision_id)
3747
parent_ids.update(revision.parent_ids)
3748
parent_ids.difference_update(revision_ids)
3749
parent_ids.discard(_mod_revision.NULL_REVISION)
3750
parent_map = self.source.get_parent_map(parent_ids)
3751
# we iterate over parent_map and not parent_ids because we don't
3752
# want to try copying any revision which is a ghost
3753
for parent_tree in self.source.revision_trees(parent_map):
3754
current_revision_id = parent_tree.get_revision_id()
3755
parents_parents = parent_map[current_revision_id]
3756
possible_trees = self._get_trees(parents_parents, cache)
3757
if len(possible_trees) == 0:
3758
# There either aren't any parents, or the parents are
3759
# ghosts, so just use the last converted tree.
3760
possible_trees.append((basis_id, cache[basis_id]))
3761
basis_id, delta = self._get_delta_for_revision(parent_tree,
3762
parents_parents, possible_trees)
3763
self.target.add_inventory_by_delta(
3764
basis_id, delta, current_revision_id, parents_parents)
3765
# insert signatures and revisions
3766
for revision in pending_revisions:
3767
try:
3768
signature = self.source.get_signature_text(
3769
revision.revision_id)
3770
self.target.add_signature_text(revision.revision_id,
3771
signature)
3772
except errors.NoSuchRevision:
3773
pass
3774
self.target.add_revision(revision.revision_id, revision)
3775
return basis_id
3776
3777
def _fetch_all_revisions(self, revision_ids, pb):
3778
"""Fetch everything for the list of revisions.
3779
3780
:param revision_ids: The list of revisions to fetch. Must be in
3781
topological order.
3782
:param pb: A ProgressTask
3783
:return: None
3784
"""
3785
basis_id, basis_tree = self._get_basis(revision_ids[0])
3786
batch_size = 100
3787
cache = lru_cache.LRUCache(100)
3788
cache[basis_id] = basis_tree
3789
del basis_tree # We don't want to hang on to it here
3790
hints = []
3791
a_graph = None
3792
3793
for offset in range(0, len(revision_ids), batch_size):
3794
self.target.start_write_group()
3795
try:
3796
pb.update('Transferring revisions', offset,
3797
len(revision_ids))
3798
batch = revision_ids[offset:offset+batch_size]
3799
basis_id = self._fetch_batch(batch, basis_id, cache)
3800
except:
3801
self.source._safe_to_return_from_cache = False
3802
self.target.abort_write_group()
3803
raise
3804
else:
3805
hint = self.target.commit_write_group()
3806
if hint:
3807
hints.extend(hint)
3808
if hints and self.target._format.pack_compresses:
3809
self.target.pack(hint=hints)
3810
pb.update('Transferring revisions', len(revision_ids),
3811
len(revision_ids))
3812
3813
@needs_write_lock
3814
def fetch(self, revision_id=None, find_ghosts=False,
3815
fetch_spec=None):
3816
"""See InterRepository.fetch()."""
3817
if fetch_spec is not None:
3818
revision_ids = fetch_spec.get_keys()
3819
else:
3820
revision_ids = None
3821
ui.ui_factory.warn_experimental_format_fetch(self)
3822
if (not self.source.supports_rich_root()
3823
and self.target.supports_rich_root()):
3824
self._converting_to_rich_root = True
3825
self._revision_id_to_root_id = {}
3826
else:
3827
self._converting_to_rich_root = False
3828
# See <https://launchpad.net/bugs/456077> asking for a warning here
3829
if self.source._format.network_name() != self.target._format.network_name():
3830
ui.ui_factory.show_user_warning('cross_format_fetch',
3831
from_format=self.source._format,
3832
to_format=self.target._format)
3833
if revision_ids is None:
3834
if revision_id:
3835
search_revision_ids = [revision_id]
3836
else:
3837
search_revision_ids = None
3838
revision_ids = self.target.search_missing_revision_ids(self.source,
3839
revision_ids=search_revision_ids,
3840
find_ghosts=find_ghosts).get_keys()
3841
if not revision_ids:
3842
return 0, 0
3843
revision_ids = tsort.topo_sort(
3844
self.source.get_graph().get_parent_map(revision_ids))
3845
if not revision_ids:
3846
return 0, 0
3847
# Walk though all revisions; get inventory deltas, copy referenced
3848
# texts that delta references, insert the delta, revision and
3849
# signature.
3850
pb = ui.ui_factory.nested_progress_bar()
3851
try:
3852
self._fetch_all_revisions(revision_ids, pb)
3853
finally:
3854
pb.finished()
3855
return len(revision_ids), 0
3856
3857
def _get_basis(self, first_revision_id):
3858
"""Get a revision and tree which exists in the target.
3859
3860
This assumes that first_revision_id is selected for transmission
3861
because all other ancestors are already present. If we can't find an
3862
ancestor we fall back to NULL_REVISION since we know that is safe.
3863
3864
:return: (basis_id, basis_tree)
3865
"""
3866
first_rev = self.source.get_revision(first_revision_id)
3867
try:
3868
basis_id = first_rev.parent_ids[0]
3869
# only valid as a basis if the target has it
3870
self.target.get_revision(basis_id)
3871
# Try to get a basis tree - if it's a ghost it will hit the
3872
# NoSuchRevision case.
3873
basis_tree = self.source.revision_tree(basis_id)
3874
except (IndexError, errors.NoSuchRevision):
3875
basis_id = _mod_revision.NULL_REVISION
3876
basis_tree = self.source.revision_tree(basis_id)
3877
return basis_id, basis_tree
3878
3879
3880
InterRepository.register_optimiser(InterDifferingSerializer)
3881
InterRepository.register_optimiser(InterSameDataRepository)
1895
1896
InterRepository.register_optimiser(InterRepository)
3882
1897
3883
1898
3884
1899
class CopyConverter(object):
3929
1944
pb.finished()
3930
1945
3931
1946
3932
class _VersionedFileChecker(object):
3933
3934
def __init__(self, repository, text_key_references=None, ancestors=None):
3935
self.repository = repository
3936
self.text_index = self.repository._generate_text_key_index(
3937
text_key_references=text_key_references, ancestors=ancestors)
3938
3939
def calculate_file_version_parents(self, text_key):
3940
"""Calculate the correct parents for a file version according to
3941
the inventories.
3942
"""
3943
parent_keys = self.text_index[text_key]
3944
if parent_keys == [_mod_revision.NULL_REVISION]:
3945
return ()
3946
return tuple(parent_keys)
3947
3948
def check_file_version_parents(self, texts, progress_bar=None):
3949
"""Check the parents stored in a versioned file are correct.
3950
3951
It also detects file versions that are not referenced by their
3952
corresponding revision's inventory.
3953
3954
:returns: A tuple of (wrong_parents, dangling_file_versions).
3955
wrong_parents is a dict mapping {revision_id: (stored_parents,
3956
correct_parents)} for each revision_id where the stored parents
3957
are not correct. dangling_file_versions is a set of (file_id,
3958
revision_id) tuples for versions that are present in this versioned
3959
file, but not used by the corresponding inventory.
3960
"""
3961
local_progress = None
3962
if progress_bar is None:
3963
local_progress = ui.ui_factory.nested_progress_bar()
3964
progress_bar = local_progress
3965
try:
3966
return self._check_file_version_parents(texts, progress_bar)
3967
finally:
3968
if local_progress:
3969
local_progress.finished()
3970
3971
def _check_file_version_parents(self, texts, progress_bar):
3972
"""See check_file_version_parents."""
3973
wrong_parents = {}
3974
self.file_ids = set([file_id for file_id, _ in
3975
self.text_index.iterkeys()])
3976
# text keys is now grouped by file_id
3977
n_versions = len(self.text_index)
3978
progress_bar.update('loading text store', 0, n_versions)
3979
parent_map = self.repository.texts.get_parent_map(self.text_index)
3980
# On unlistable transports this could well be empty/error...
3981
text_keys = self.repository.texts.keys()
3982
unused_keys = frozenset(text_keys) - set(self.text_index)
3983
for num, key in enumerate(self.text_index.iterkeys()):
3984
progress_bar.update('checking text graph', num, n_versions)
3985
correct_parents = self.calculate_file_version_parents(key)
3986
try:
3987
knit_parents = parent_map[key]
3988
except errors.RevisionNotPresent:
3989
# Missing text!
3990
knit_parents = None
3991
if correct_parents != knit_parents:
3992
wrong_parents[key] = (knit_parents, correct_parents)
3993
return wrong_parents, unused_keys
3994
3995
3996
def _old_get_graph(repository, revision_id):
3997
"""DO NOT USE. That is all. I'm serious."""
3998
graph = repository.get_graph()
3999
revision_graph = dict(((key, value) for key, value in
4000
graph.iter_ancestry([revision_id]) if value is not None))
4001
return _strip_NULL_ghosts(revision_graph)
4002
4003
4004
1947
def _strip_NULL_ghosts(revision_graph):
4005
1948
"""Also don't use this. more compatibility code for unmigrated clients."""
4006
1949
# Filter ghosts, and null:
4012
1955
return revision_graph
4013
1956
4014
1957
4015
class StreamSink(object):
4016
"""An object that can insert a stream into a repository.
4017
4018
This interface handles the complexity of reserialising inventories and
4019
revisions from different formats, and allows unidirectional insertion into
4020
stacked repositories without looking for the missing basis parents
4021
beforehand.
4022
"""
4023
4024
def __init__(self, target_repo):
4025
self.target_repo = target_repo
4026
4027
def insert_stream(self, stream, src_format, resume_tokens):
4028
"""Insert a stream's content into the target repository.
4029
4030
:param src_format: a bzr repository format.
4031
4032
:return: a list of resume tokens and an iterable of keys additional
4033
items required before the insertion can be completed.
4034
"""
4035
self.target_repo.lock_write()
4036
try:
4037
if resume_tokens:
4038
self.target_repo.resume_write_group(resume_tokens)
4039
is_resume = True
4040
else:
4041
self.target_repo.start_write_group()
4042
is_resume = False
4043
try:
4044
# locked_insert_stream performs a commit|suspend.
4045
missing_keys = self.insert_stream_without_locking(stream,
4046
src_format, is_resume)
4047
if missing_keys:
4048
# suspend the write group and tell the caller what we is
4049
# missing. We know we can suspend or else we would not have
4050
# entered this code path. (All repositories that can handle
4051
# missing keys can handle suspending a write group).
4052
write_group_tokens = self.target_repo.suspend_write_group()
4053
return write_group_tokens, missing_keys
4054
hint = self.target_repo.commit_write_group()
4055
to_serializer = self.target_repo._format._serializer
4056
src_serializer = src_format._serializer
4057
if (to_serializer != src_serializer and
4058
self.target_repo._format.pack_compresses):
4059
self.target_repo.pack(hint=hint)
4060
return [], set()
4061
except:
4062
self.target_repo.abort_write_group(suppress_errors=True)
4063
raise
4064
finally:
4065
self.target_repo.unlock()
4066
4067
def insert_stream_without_locking(self, stream, src_format,
4068
is_resume=False):
4069
"""Insert a stream's content into the target repository.
4070
4071
This assumes that you already have a locked repository and an active
4072
write group.
4073
4074
:param src_format: a bzr repository format.
4075
:param is_resume: Passed down to get_missing_parent_inventories to
4076
indicate if we should be checking for missing texts at the same
4077
time.
4078
4079
:return: A set of keys that are missing.
4080
"""
4081
if not self.target_repo.is_write_locked():
4082
raise errors.ObjectNotLocked(self)
4083
if not self.target_repo.is_in_write_group():
4084
raise errors.BzrError('you must already be in a write group')
4085
to_serializer = self.target_repo._format._serializer
4086
src_serializer = src_format._serializer
4087
new_pack = None
4088
if to_serializer == src_serializer:
4089
# If serializers match and the target is a pack repository, set the
4090
# write cache size on the new pack. This avoids poor performance
4091
# on transports where append is unbuffered (such as
4092
# RemoteTransport). This is safe to do because nothing should read
4093
# back from the target repository while a stream with matching
4094
# serialization is being inserted.
4095
# The exception is that a delta record from the source that should
4096
# be a fulltext may need to be expanded by the target (see
4097
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4098
# explicitly flush any buffered writes first in that rare case.
4099
try:
4100
new_pack = self.target_repo._pack_collection._new_pack
4101
except AttributeError:
4102
# Not a pack repository
4103
pass
4104
else:
4105
new_pack.set_write_cache_size(1024*1024)
4106
for substream_type, substream in stream:
4107
if 'stream' in debug.debug_flags:
4108
mutter('inserting substream: %s', substream_type)
4109
if substream_type == 'texts':
4110
self.target_repo.texts.insert_record_stream(substream)
4111
elif substream_type == 'inventories':
4112
if src_serializer == to_serializer:
4113
self.target_repo.inventories.insert_record_stream(
4114
substream)
4115
else:
4116
self._extract_and_insert_inventories(
4117
substream, src_serializer)
4118
elif substream_type == 'inventory-deltas':
4119
self._extract_and_insert_inventory_deltas(
4120
substream, src_serializer)
4121
elif substream_type == 'chk_bytes':
4122
# XXX: This doesn't support conversions, as it assumes the
4123
# conversion was done in the fetch code.
4124
self.target_repo.chk_bytes.insert_record_stream(substream)
4125
elif substream_type == 'revisions':
4126
# This may fallback to extract-and-insert more often than
4127
# required if the serializers are different only in terms of
4128
# the inventory.
4129
if src_serializer == to_serializer:
4130
self.target_repo.revisions.insert_record_stream(substream)
4131
else:
4132
self._extract_and_insert_revisions(substream,
4133
src_serializer)
4134
elif substream_type == 'signatures':
4135
self.target_repo.signatures.insert_record_stream(substream)
4136
else:
4137
raise AssertionError('kaboom! %s' % (substream_type,))
4138
# Done inserting data, and the missing_keys calculations will try to
4139
# read back from the inserted data, so flush the writes to the new pack
4140
# (if this is pack format).
4141
if new_pack is not None:
4142
new_pack._write_data('', flush=True)
4143
# Find all the new revisions (including ones from resume_tokens)
4144
missing_keys = self.target_repo.get_missing_parent_inventories(
4145
check_for_missing_texts=is_resume)
4146
try:
4147
for prefix, versioned_file in (
4148
('texts', self.target_repo.texts),
4149
('inventories', self.target_repo.inventories),
4150
('revisions', self.target_repo.revisions),
4151
('signatures', self.target_repo.signatures),
4152
('chk_bytes', self.target_repo.chk_bytes),
4153
):
4154
if versioned_file is None:
4155
continue
4156
# TODO: key is often going to be a StaticTuple object
4157
# I don't believe we can define a method by which
4158
# (prefix,) + StaticTuple will work, though we could
4159
# define a StaticTuple.sq_concat that would allow you to
4160
# pass in either a tuple or a StaticTuple as the second
4161
# object, so instead we could have:
4162
# StaticTuple(prefix) + key here...
4163
missing_keys.update((prefix,) + key for key in
4164
versioned_file.get_missing_compression_parent_keys())
4165
except NotImplementedError:
4166
# cannot even attempt suspending, and missing would have failed
4167
# during stream insertion.
4168
missing_keys = set()
4169
return missing_keys
4170
4171
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4172
target_rich_root = self.target_repo._format.rich_root_data
4173
target_tree_refs = self.target_repo._format.supports_tree_reference
4174
for record in substream:
4175
# Insert the delta directly
4176
inventory_delta_bytes = record.get_bytes_as('fulltext')
4177
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4178
try:
4179
parse_result = deserialiser.parse_text_bytes(
4180
inventory_delta_bytes)
4181
except inventory_delta.IncompatibleInventoryDelta, err:
4182
trace.mutter("Incompatible delta: %s", err.msg)
4183
raise errors.IncompatibleRevision(self.target_repo._format)
4184
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4185
revision_id = new_id
4186
parents = [key[0] for key in record.parents]
4187
self.target_repo.add_inventory_by_delta(
4188
basis_id, inv_delta, revision_id, parents)
4189
4190
def _extract_and_insert_inventories(self, substream, serializer,
4191
parse_delta=None):
4192
"""Generate a new inventory versionedfile in target, converting data.
4193
4194
The inventory is retrieved from the source, (deserializing it), and
4195
stored in the target (reserializing it in a different format).
4196
"""
4197
target_rich_root = self.target_repo._format.rich_root_data
4198
target_tree_refs = self.target_repo._format.supports_tree_reference
4199
for record in substream:
4200
# It's not a delta, so it must be a fulltext in the source
4201
# serializer's format.
4202
bytes = record.get_bytes_as('fulltext')
4203
revision_id = record.key[0]
4204
inv = serializer.read_inventory_from_string(bytes, revision_id)
4205
parents = [key[0] for key in record.parents]
4206
self.target_repo.add_inventory(revision_id, inv, parents)
4207
# No need to keep holding this full inv in memory when the rest of
4208
# the substream is likely to be all deltas.
4209
del inv
4210
4211
def _extract_and_insert_revisions(self, substream, serializer):
4212
for record in substream:
4213
bytes = record.get_bytes_as('fulltext')
4214
revision_id = record.key[0]
4215
rev = serializer.read_revision_from_string(bytes)
4216
if rev.revision_id != revision_id:
4217
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4218
self.target_repo.add_revision(revision_id, rev)
4219
4220
def finished(self):
4221
if self.target_repo._format._fetch_reconcile:
4222
self.target_repo.reconcile()
4223
4224
4225
class StreamSource(object):
4226
"""A source of a stream for fetching between repositories."""
4227
4228
def __init__(self, from_repository, to_format):
4229
"""Create a StreamSource streaming from from_repository."""
4230
self.from_repository = from_repository
4231
self.to_format = to_format
4232
self._record_counter = RecordCounter()
4233
4234
def delta_on_metadata(self):
4235
"""Return True if delta's are permitted on metadata streams.
4236
4237
That is on revisions and signatures.
4238
"""
4239
src_serializer = self.from_repository._format._serializer
4240
target_serializer = self.to_format._serializer
4241
return (self.to_format._fetch_uses_deltas and
4242
src_serializer == target_serializer)
4243
4244
def _fetch_revision_texts(self, revs):
4245
# fetch signatures first and then the revision texts
4246
# may need to be a InterRevisionStore call here.
4247
from_sf = self.from_repository.signatures
4248
# A missing signature is just skipped.
4249
keys = [(rev_id,) for rev_id in revs]
4250
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4251
keys,
4252
self.to_format._fetch_order,
4253
not self.to_format._fetch_uses_deltas))
4254
# If a revision has a delta, this is actually expanded inside the
4255
# insert_record_stream code now, which is an alternate fix for
4256
# bug #261339
4257
from_rf = self.from_repository.revisions
4258
revisions = from_rf.get_record_stream(
4259
keys,
4260
self.to_format._fetch_order,
4261
not self.delta_on_metadata())
4262
return [('signatures', signatures), ('revisions', revisions)]
4263
4264
def _generate_root_texts(self, revs):
4265
"""This will be called by get_stream between fetching weave texts and
4266
fetching the inventory weave.
4267
"""
4268
if self._rich_root_upgrade():
4269
return _mod_fetch.Inter1and2Helper(
4270
self.from_repository).generate_root_texts(revs)
4271
else:
4272
return []
4273
4274
def get_stream(self, search):
4275
phase = 'file'
4276
revs = search.get_keys()
4277
graph = self.from_repository.get_graph()
4278
revs = tsort.topo_sort(graph.get_parent_map(revs))
4279
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4280
text_keys = []
4281
for knit_kind, file_id, revisions in data_to_fetch:
4282
if knit_kind != phase:
4283
phase = knit_kind
4284
# Make a new progress bar for this phase
4285
if knit_kind == "file":
4286
# Accumulate file texts
4287
text_keys.extend([(file_id, revision) for revision in
4288
revisions])
4289
elif knit_kind == "inventory":
4290
# Now copy the file texts.
4291
from_texts = self.from_repository.texts
4292
yield ('texts', from_texts.get_record_stream(
4293
text_keys, self.to_format._fetch_order,
4294
not self.to_format._fetch_uses_deltas))
4295
# Cause an error if a text occurs after we have done the
4296
# copy.
4297
text_keys = None
4298
# Before we process the inventory we generate the root
4299
# texts (if necessary) so that the inventories references
4300
# will be valid.
4301
for _ in self._generate_root_texts(revs):
4302
yield _
4303
# we fetch only the referenced inventories because we do not
4304
# know for unselected inventories whether all their required
4305
# texts are present in the other repository - it could be
4306
# corrupt.
4307
for info in self._get_inventory_stream(revs):
4308
yield info
4309
elif knit_kind == "signatures":
4310
# Nothing to do here; this will be taken care of when
4311
# _fetch_revision_texts happens.
4312
pass
4313
elif knit_kind == "revisions":
4314
for record in self._fetch_revision_texts(revs):
4315
yield record
4316
else:
4317
raise AssertionError("Unknown knit kind %r" % knit_kind)
4318
4319
def get_stream_for_missing_keys(self, missing_keys):
4320
# missing keys can only occur when we are byte copying and not
4321
# translating (because translation means we don't send
4322
# unreconstructable deltas ever).
4323
keys = {}
4324
keys['texts'] = set()
4325
keys['revisions'] = set()
4326
keys['inventories'] = set()
4327
keys['chk_bytes'] = set()
4328
keys['signatures'] = set()
4329
for key in missing_keys:
4330
keys[key[0]].add(key[1:])
4331
if len(keys['revisions']):
4332
# If we allowed copying revisions at this point, we could end up
4333
# copying a revision without copying its required texts: a
4334
# violation of the requirements for repository integrity.
4335
raise AssertionError(
4336
'cannot copy revisions to fill in missing deltas %s' % (
4337
keys['revisions'],))
4338
for substream_kind, keys in keys.iteritems():
4339
vf = getattr(self.from_repository, substream_kind)
4340
if vf is None and keys:
4341
raise AssertionError(
4342
"cannot fill in keys for a versioned file we don't"
4343
" have: %s needs %s" % (substream_kind, keys))
4344
if not keys:
4345
# No need to stream something we don't have
4346
continue
4347
if substream_kind == 'inventories':
4348
# Some missing keys are genuinely ghosts, filter those out.
4349
present = self.from_repository.inventories.get_parent_map(keys)
4350
revs = [key[0] for key in present]
4351
# Get the inventory stream more-or-less as we do for the
4352
# original stream; there's no reason to assume that records
4353
# direct from the source will be suitable for the sink. (Think
4354
# e.g. 2a -> 1.9-rich-root).
4355
for info in self._get_inventory_stream(revs, missing=True):
4356
yield info
4357
continue
4358
4359
# Ask for full texts always so that we don't need more round trips
4360
# after this stream.
4361
# Some of the missing keys are genuinely ghosts, so filter absent
4362
# records. The Sink is responsible for doing another check to
4363
# ensure that ghosts don't introduce missing data for future
4364
# fetches.
4365
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4366
self.to_format._fetch_order, True))
4367
yield substream_kind, stream
4368
4369
def inventory_fetch_order(self):
4370
if self._rich_root_upgrade():
4371
return 'topological'
4372
else:
4373
return self.to_format._fetch_order
4374
4375
def _rich_root_upgrade(self):
4376
return (not self.from_repository._format.rich_root_data and
4377
self.to_format.rich_root_data)
4378
4379
def _get_inventory_stream(self, revision_ids, missing=False):
4380
from_format = self.from_repository._format
4381
if (from_format.supports_chks and self.to_format.supports_chks and
4382
from_format.network_name() == self.to_format.network_name()):
4383
raise AssertionError(
4384
"this case should be handled by GroupCHKStreamSource")
4385
elif 'forceinvdeltas' in debug.debug_flags:
4386
return self._get_convertable_inventory_stream(revision_ids,
4387
delta_versus_null=missing)
4388
elif from_format.network_name() == self.to_format.network_name():
4389
# Same format.
4390
return self._get_simple_inventory_stream(revision_ids,
4391
missing=missing)
4392
elif (not from_format.supports_chks and not self.to_format.supports_chks
4393
and from_format._serializer == self.to_format._serializer):
4394
# Essentially the same format.
4395
return self._get_simple_inventory_stream(revision_ids,
4396
missing=missing)
4397
else:
4398
# Any time we switch serializations, we want to use an
4399
# inventory-delta based approach.
4400
return self._get_convertable_inventory_stream(revision_ids,
4401
delta_versus_null=missing)
4402
4403
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4404
# NB: This currently reopens the inventory weave in source;
4405
# using a single stream interface instead would avoid this.
4406
from_weave = self.from_repository.inventories
4407
if missing:
4408
delta_closure = True
4409
else:
4410
delta_closure = not self.delta_on_metadata()
4411
yield ('inventories', from_weave.get_record_stream(
4412
[(rev_id,) for rev_id in revision_ids],
4413
self.inventory_fetch_order(), delta_closure))
4414
4415
def _get_convertable_inventory_stream(self, revision_ids,
4416
delta_versus_null=False):
4417
# The two formats are sufficiently different that there is no fast
4418
# path, so we need to send just inventorydeltas, which any
4419
# sufficiently modern client can insert into any repository.
4420
# The StreamSink code expects to be able to
4421
# convert on the target, so we need to put bytes-on-the-wire that can
4422
# be converted. That means inventory deltas (if the remote is <1.19,
4423
# RemoteStreamSink will fallback to VFS to insert the deltas).
4424
yield ('inventory-deltas',
4425
self._stream_invs_as_deltas(revision_ids,
4426
delta_versus_null=delta_versus_null))
4427
4428
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4429
"""Return a stream of inventory-deltas for the given rev ids.
4430
4431
:param revision_ids: The list of inventories to transmit
4432
:param delta_versus_null: Don't try to find a minimal delta for this
4433
entry, instead compute the delta versus the NULL_REVISION. This
4434
effectively streams a complete inventory. Used for stuff like
4435
filling in missing parents, etc.
4436
"""
4437
from_repo = self.from_repository
4438
revision_keys = [(rev_id,) for rev_id in revision_ids]
4439
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4440
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4441
# method...
4442
inventories = self.from_repository.iter_inventories(
4443
revision_ids, 'topological')
4444
format = from_repo._format
4445
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4446
inventory_cache = lru_cache.LRUCache(50)
4447
null_inventory = from_repo.revision_tree(
4448
_mod_revision.NULL_REVISION).inventory
4449
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4450
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4451
# repo back into a non-rich-root repo ought to be allowed)
4452
serializer = inventory_delta.InventoryDeltaSerializer(
4453
versioned_root=format.rich_root_data,
4454
tree_references=format.supports_tree_reference)
4455
for inv in inventories:
4456
key = (inv.revision_id,)
4457
parent_keys = parent_map.get(key, ())
4458
delta = None
4459
if not delta_versus_null and parent_keys:
4460
# The caller did not ask for complete inventories and we have
4461
# some parents that we can delta against. Make a delta against
4462
# each parent so that we can find the smallest.
4463
parent_ids = [parent_key[0] for parent_key in parent_keys]
4464
for parent_id in parent_ids:
4465
if parent_id not in invs_sent_so_far:
4466
# We don't know that the remote side has this basis, so
4467
# we can't use it.
4468
continue
4469
if parent_id == _mod_revision.NULL_REVISION:
4470
parent_inv = null_inventory
4471
else:
4472
parent_inv = inventory_cache.get(parent_id, None)
4473
if parent_inv is None:
4474
parent_inv = from_repo.get_inventory(parent_id)
4475
candidate_delta = inv._make_delta(parent_inv)
4476
if (delta is None or
4477
len(delta) > len(candidate_delta)):
4478
delta = candidate_delta
4479
basis_id = parent_id
4480
if delta is None:
4481
# Either none of the parents ended up being suitable, or we
4482
# were asked to delta against NULL
4483
basis_id = _mod_revision.NULL_REVISION
4484
delta = inv._make_delta(null_inventory)
4485
invs_sent_so_far.add(inv.revision_id)
4486
inventory_cache[inv.revision_id] = inv
4487
delta_serialized = ''.join(
4488
serializer.delta_to_lines(basis_id, key[-1], delta))
4489
yield versionedfile.FulltextContentFactory(
4490
key, parent_keys, None, delta_serialized)
4491
4492
4493
1958
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4494
1959
stop_revision=None):
4495
1960
"""Extend the partial history to include a given index
4522
1987
return
4523
1988
4524
1989
4525
1990
class _LazyListJoin(object):
1991
"""An iterable yielding the contents of many lists as one list.
1992
1993
Each iterator made from this will reflect the current contents of the lists
1994
at the time the iterator is made.
1995
1996
This is used by Repository's _make_parents_provider implementation so that
1997
it is safe to do::
1998
1999
pp = repo._make_parents_provider() # uses a list of fallback repos
2000
pp.add_fallback_repository(other_repo) # appends to that list
2001
result = pp.get_parent_map(...)
2002
# The result will include revs from other_repo
2003
"""
2004
2005
def __init__(self, *list_parts):
2006
self.list_parts = list_parts
2007
2008
def __iter__(self):
2009
full_list = []
2010
for list_part in self.list_parts:
2011
full_list.extend(list_part)
2012
return iter(full_list)
Loggerhead is a web-based interface for Breezy
Version: 2.0.1