« back to all changes in this revision
Viewing changes to bzrlib/repository.py
- browse files at revision 5820.1.27
- compare with another revision
- download diff
- download tarball
- view history from revision 5820.1.27
- Committer: INADA Naoki
- Date: 2011-05-14 14:32:02 UTC
- mto: This revision was merged to the branch mainline in revision 5874.
- Revision ID: songofacandy@gmail.com-20110514143202-47r07mp2ebipg2yq
Fix error when fix_sentence_endings=True.
- files added:
- bzrlib/deprecated_graph.py
- files removed:
- bzrlib/export_pot.py
- po
- files renamed:
- bzrlib/commit_signature_commands.py => bzrlib/sign_my_commits.py
- bzrlib/tests/per_repository_vf/test__generate_text_key_index.py => bzrlib/tests/per_repository/test__generate_text_key_index.py
- bzrlib/tests/per_repository_vf/test_add_inventory_by_delta.py => bzrlib/tests/per_repository/test_add_inventory_by_delta.py
- bzrlib/tests/per_repository_vf/test_find_text_key_references.py => bzrlib/tests/per_repository/test_find_text_key_references.py
- bzrlib/tests/per_repository/test_locking.py => bzrlib/tests/per_repository/test_is_write_locked.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 itertools
20
19
import time
21
20
22
21
from bzrlib import (
23
22
bzrdir,
23
check,
24
24
config,
25
25
controldir,
26
26
debug,
27
fetch as _mod_fetch,
28
fifo_cache,
27
29
generate_ids,
30
gpg,
28
31
graph,
32
inventory_delta,
29
33
lockable_files,
30
34
lockdir,
35
lru_cache,
31
36
osutils,
32
37
revision as _mod_revision,
33
testament as _mod_testament,
38
static_tuple,
34
39
tsort,
35
gpg,
40
versionedfile,
36
41
)
37
42
from bzrlib.bundle import serializer
43
from bzrlib.recordcounter import RecordCounter
44
from bzrlib.revisiontree import InventoryRevisionTree
45
from bzrlib.store.versioned import VersionedFileStore
46
from bzrlib.testament import Testament
38
47
""")
39
48
40
49
from bzrlib import (
45
54
)
46
55
from bzrlib.decorators import needs_read_lock, needs_write_lock, only_raises
47
56
from bzrlib.inter import InterObject
57
from bzrlib.inventory import (
58
Inventory,
59
InventoryDirectory,
60
ROOT_ID,
61
entry_factory,
62
)
48
63
from bzrlib.lock import _RelockDebugMixin, LogicalLockResult
49
64
from bzrlib.trace import (
50
65
log_exception_quietly, note, mutter, mutter_callsite, warning)
71
86
72
87
# all clients should supply tree roots.
73
88
record_root_entry = True
74
# whether this commit builder supports the record_entry_contents interface
75
supports_record_entry_contents = False
89
# the default CommitBuilder does not manage trees whose root is versioned.
90
_versioned_root = False
91
# this commit builder supports the record_entry_contents interface
92
supports_record_entry_contents = True
76
93
77
94
def __init__(self, repository, parents, config, timestamp=None,
78
95
timezone=None, committer=None, revprops=None,
99
116
else:
100
117
self._committer = committer
101
118
119
self.new_inventory = Inventory(None)
102
120
self._new_revision_id = revision_id
103
121
self.parents = parents
104
122
self.repository = repository
119
137
self._timezone = int(timezone)
120
138
121
139
self._generate_revision_if_needed()
140
self.__heads = graph.HeadsCache(repository.get_graph()).heads
141
self._basis_delta = []
142
# API compatibility, older code that used CommitBuilder did not call
143
# .record_delete(), which means the delta that is computed would not be
144
# valid. Callers that will call record_delete() should call
145
# .will_record_deletes() to indicate that.
146
self._recording_deletes = False
147
# memo'd check for no-op commits.
148
self._any_changes = False
122
149
123
150
def any_changes(self):
124
151
"""Return True if any entries were changed.
125
152
126
153
This includes merge-only changes. It is the core for the --unchanged
127
154
detection in commit.
128
155
129
156
:return: True if any changes have occured.
130
157
"""
131
raise NotImplementedError(self.any_changes)
158
return self._any_changes
132
159
133
160
def _validate_unicode_text(self, text, context):
134
161
"""Verify things like commit messages don't have bogus characters."""
145
172
self._validate_unicode_text(value,
146
173
'revision property (%s)' % (key,))
147
174
175
def _ensure_fallback_inventories(self):
176
"""Ensure that appropriate inventories are available.
177
178
This only applies to repositories that are stacked, and is about
179
enusring the stacking invariants. Namely, that for any revision that is
180
present, we either have all of the file content, or we have the parent
181
inventory and the delta file content.
182
"""
183
if not self.repository._fallback_repositories:
184
return
185
if not self.repository._format.supports_chks:
186
raise errors.BzrError("Cannot commit directly to a stacked branch"
187
" in pre-2a formats. See "
188
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
189
# This is a stacked repo, we need to make sure we have the parent
190
# inventories for the parents.
191
parent_keys = [(p,) for p in self.parents]
192
parent_map = self.repository.inventories._index.get_parent_map(parent_keys)
193
missing_parent_keys = set([pk for pk in parent_keys
194
if pk not in parent_map])
195
fallback_repos = list(reversed(self.repository._fallback_repositories))
196
missing_keys = [('inventories', pk[0])
197
for pk in missing_parent_keys]
198
resume_tokens = []
199
while missing_keys and fallback_repos:
200
fallback_repo = fallback_repos.pop()
201
source = fallback_repo._get_source(self.repository._format)
202
sink = self.repository._get_sink()
203
stream = source.get_stream_for_missing_keys(missing_keys)
204
missing_keys = sink.insert_stream_without_locking(stream,
205
self.repository._format)
206
if missing_keys:
207
raise errors.BzrError('Unable to fill in parent inventories for a'
208
' stacked branch')
209
148
210
def commit(self, message):
149
211
"""Make the actual commit.
150
212
151
213
:return: The revision id of the recorded revision.
152
214
"""
153
raise NotImplementedError(self.commit)
215
self._validate_unicode_text(message, 'commit message')
216
rev = _mod_revision.Revision(
217
timestamp=self._timestamp,
218
timezone=self._timezone,
219
committer=self._committer,
220
message=message,
221
inventory_sha1=self.inv_sha1,
222
revision_id=self._new_revision_id,
223
properties=self._revprops)
224
rev.parent_ids = self.parents
225
self.repository.add_revision(self._new_revision_id, rev,
226
self.new_inventory, self._config)
227
self._ensure_fallback_inventories()
228
self.repository.commit_write_group()
229
return self._new_revision_id
154
230
155
231
def abort(self):
156
232
"""Abort the commit that is being built.
157
233
"""
158
raise NotImplementedError(self.abort)
234
self.repository.abort_write_group()
159
235
160
236
def revision_tree(self):
161
237
"""Return the tree that was just committed.
162
238
163
239
After calling commit() this can be called to get a
164
RevisionTree representing the newly committed tree. This is
240
InventoryRevisionTree representing the newly committed tree. This is
165
241
preferred to calling Repository.revision_tree() because that may
166
242
require deserializing the inventory, while we already have a copy in
167
243
memory.
168
244
"""
169
raise NotImplementedError(self.revision_tree)
245
if self.new_inventory is None:
246
self.new_inventory = self.repository.get_inventory(
247
self._new_revision_id)
248
return InventoryRevisionTree(self.repository, self.new_inventory,
249
self._new_revision_id)
170
250
171
251
def finish_inventory(self):
172
252
"""Tell the builder that the inventory is finished.
174
254
:return: The inventory id in the repository, which can be used with
175
255
repository.get_inventory.
176
256
"""
177
raise NotImplementedError(self.finish_inventory)
257
if self.new_inventory is None:
258
# an inventory delta was accumulated without creating a new
259
# inventory.
260
basis_id = self.basis_delta_revision
261
# We ignore the 'inventory' returned by add_inventory_by_delta
262
# because self.new_inventory is used to hint to the rest of the
263
# system what code path was taken
264
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
265
basis_id, self._basis_delta, self._new_revision_id,
266
self.parents)
267
else:
268
if self.new_inventory.root is None:
269
raise AssertionError('Root entry should be supplied to'
270
' record_entry_contents, as of bzr 0.10.')
271
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
272
self.new_inventory.revision_id = self._new_revision_id
273
self.inv_sha1 = self.repository.add_inventory(
274
self._new_revision_id,
275
self.new_inventory,
276
self.parents
277
)
278
return self._new_revision_id
178
279
179
280
def _gen_revision_id(self):
180
281
"""Return new revision-id."""
195
296
else:
196
297
self.random_revid = False
197
298
299
def _heads(self, file_id, revision_ids):
300
"""Calculate the graph heads for revision_ids in the graph of file_id.
301
302
This can use either a per-file graph or a global revision graph as we
303
have an identity relationship between the two graphs.
304
"""
305
return self.__heads(revision_ids)
306
307
def _check_root(self, ie, parent_invs, tree):
308
"""Helper for record_entry_contents.
309
310
:param ie: An entry being added.
311
:param parent_invs: The inventories of the parent revisions of the
312
commit.
313
:param tree: The tree that is being committed.
314
"""
315
# In this revision format, root entries have no knit or weave When
316
# serializing out to disk and back in root.revision is always
317
# _new_revision_id
318
ie.revision = self._new_revision_id
319
320
def _require_root_change(self, tree):
321
"""Enforce an appropriate root object change.
322
323
This is called once when record_iter_changes is called, if and only if
324
the root was not in the delta calculated by record_iter_changes.
325
326
:param tree: The tree which is being committed.
327
"""
328
if len(self.parents) == 0:
329
raise errors.RootMissing()
330
entry = entry_factory['directory'](tree.path2id(''), '',
331
None)
332
entry.revision = self._new_revision_id
333
self._basis_delta.append(('', '', entry.file_id, entry))
334
335
def _get_delta(self, ie, basis_inv, path):
336
"""Get a delta against the basis inventory for ie."""
337
if ie.file_id not in basis_inv:
338
# add
339
result = (None, path, ie.file_id, ie)
340
self._basis_delta.append(result)
341
return result
342
elif ie != basis_inv[ie.file_id]:
343
# common but altered
344
# TODO: avoid tis id2path call.
345
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
346
self._basis_delta.append(result)
347
return result
348
else:
349
# common, unaltered
350
return None
351
352
def get_basis_delta(self):
353
"""Return the complete inventory delta versus the basis inventory.
354
355
This has been built up with the calls to record_delete and
356
record_entry_contents. The client must have already called
357
will_record_deletes() to indicate that they will be generating a
358
complete delta.
359
360
:return: An inventory delta, suitable for use with apply_delta, or
361
Repository.add_inventory_by_delta, etc.
362
"""
363
if not self._recording_deletes:
364
raise AssertionError("recording deletes not activated.")
365
return self._basis_delta
366
367
def record_delete(self, path, file_id):
368
"""Record that a delete occured against a basis tree.
369
370
This is an optional API - when used it adds items to the basis_delta
371
being accumulated by the commit builder. It cannot be called unless the
372
method will_record_deletes() has been called to inform the builder that
373
a delta is being supplied.
374
375
:param path: The path of the thing deleted.
376
:param file_id: The file id that was deleted.
377
"""
378
if not self._recording_deletes:
379
raise AssertionError("recording deletes not activated.")
380
delta = (path, None, file_id, None)
381
self._basis_delta.append(delta)
382
self._any_changes = True
383
return delta
384
198
385
def will_record_deletes(self):
199
386
"""Tell the commit builder that deletes are being notified.
200
387
202
389
commit to be valid, deletes against the basis MUST be recorded via
203
390
builder.record_delete().
204
391
"""
205
raise NotImplementedError(self.will_record_deletes)
206
207
def record_iter_changes(self, tree, basis_revision_id, iter_changes):
392
self._recording_deletes = True
393
try:
394
basis_id = self.parents[0]
395
except IndexError:
396
basis_id = _mod_revision.NULL_REVISION
397
self.basis_delta_revision = basis_id
398
399
def record_entry_contents(self, ie, parent_invs, path, tree,
400
content_summary):
401
"""Record the content of ie from tree into the commit if needed.
402
403
Side effect: sets ie.revision when unchanged
404
405
:param ie: An inventory entry present in the commit.
406
:param parent_invs: The inventories of the parent revisions of the
407
commit.
408
:param path: The path the entry is at in the tree.
409
:param tree: The tree which contains this entry and should be used to
410
obtain content.
411
:param content_summary: Summary data from the tree about the paths
412
content - stat, length, exec, sha/link target. This is only
413
accessed when the entry has a revision of None - that is when it is
414
a candidate to commit.
415
:return: A tuple (change_delta, version_recorded, fs_hash).
416
change_delta is an inventory_delta change for this entry against
417
the basis tree of the commit, or None if no change occured against
418
the basis tree.
419
version_recorded is True if a new version of the entry has been
420
recorded. For instance, committing a merge where a file was only
421
changed on the other side will return (delta, False).
422
fs_hash is either None, or the hash details for the path (currently
423
a tuple of the contents sha1 and the statvalue returned by
424
tree.get_file_with_stat()).
425
"""
426
if self.new_inventory.root is None:
427
if ie.parent_id is not None:
428
raise errors.RootMissing()
429
self._check_root(ie, parent_invs, tree)
430
if ie.revision is None:
431
kind = content_summary[0]
432
else:
433
# ie is carried over from a prior commit
434
kind = ie.kind
435
# XXX: repository specific check for nested tree support goes here - if
436
# the repo doesn't want nested trees we skip it ?
437
if (kind == 'tree-reference' and
438
not self.repository._format.supports_tree_reference):
439
# mismatch between commit builder logic and repository:
440
# this needs the entry creation pushed down into the builder.
441
raise NotImplementedError('Missing repository subtree support.')
442
self.new_inventory.add(ie)
443
444
# TODO: slow, take it out of the inner loop.
445
try:
446
basis_inv = parent_invs[0]
447
except IndexError:
448
basis_inv = Inventory(root_id=None)
449
450
# ie.revision is always None if the InventoryEntry is considered
451
# for committing. We may record the previous parents revision if the
452
# content is actually unchanged against a sole head.
453
if ie.revision is not None:
454
if not self._versioned_root and path == '':
455
# repositories that do not version the root set the root's
456
# revision to the new commit even when no change occurs (more
457
# specifically, they do not record a revision on the root; and
458
# the rev id is assigned to the root during deserialisation -
459
# this masks when a change may have occurred against the basis.
460
# To match this we always issue a delta, because the revision
461
# of the root will always be changing.
462
if ie.file_id in basis_inv:
463
delta = (basis_inv.id2path(ie.file_id), path,
464
ie.file_id, ie)
465
else:
466
# add
467
delta = (None, path, ie.file_id, ie)
468
self._basis_delta.append(delta)
469
return delta, False, None
470
else:
471
# we don't need to commit this, because the caller already
472
# determined that an existing revision of this file is
473
# appropriate. If it's not being considered for committing then
474
# it and all its parents to the root must be unaltered so
475
# no-change against the basis.
476
if ie.revision == self._new_revision_id:
477
raise AssertionError("Impossible situation, a skipped "
478
"inventory entry (%r) claims to be modified in this "
479
"commit (%r).", (ie, self._new_revision_id))
480
return None, False, None
481
# XXX: Friction: parent_candidates should return a list not a dict
482
# so that we don't have to walk the inventories again.
483
parent_candiate_entries = ie.parent_candidates(parent_invs)
484
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
485
heads = []
486
for inv in parent_invs:
487
if ie.file_id in inv:
488
old_rev = inv[ie.file_id].revision
489
if old_rev in head_set:
490
heads.append(inv[ie.file_id].revision)
491
head_set.remove(inv[ie.file_id].revision)
492
493
store = False
494
# now we check to see if we need to write a new record to the
495
# file-graph.
496
# We write a new entry unless there is one head to the ancestors, and
497
# the kind-derived content is unchanged.
498
499
# Cheapest check first: no ancestors, or more the one head in the
500
# ancestors, we write a new node.
501
if len(heads) != 1:
502
store = True
503
if not store:
504
# There is a single head, look it up for comparison
505
parent_entry = parent_candiate_entries[heads[0]]
506
# if the non-content specific data has changed, we'll be writing a
507
# node:
508
if (parent_entry.parent_id != ie.parent_id or
509
parent_entry.name != ie.name):
510
store = True
511
# now we need to do content specific checks:
512
if not store:
513
# if the kind changed the content obviously has
514
if kind != parent_entry.kind:
515
store = True
516
# Stat cache fingerprint feedback for the caller - None as we usually
517
# don't generate one.
518
fingerprint = None
519
if kind == 'file':
520
if content_summary[2] is None:
521
raise ValueError("Files must not have executable = None")
522
if not store:
523
# We can't trust a check of the file length because of content
524
# filtering...
525
if (# if the exec bit has changed we have to store:
526
parent_entry.executable != content_summary[2]):
527
store = True
528
elif parent_entry.text_sha1 == content_summary[3]:
529
# all meta and content is unchanged (using a hash cache
530
# hit to check the sha)
531
ie.revision = parent_entry.revision
532
ie.text_size = parent_entry.text_size
533
ie.text_sha1 = parent_entry.text_sha1
534
ie.executable = parent_entry.executable
535
return self._get_delta(ie, basis_inv, path), False, None
536
else:
537
# Either there is only a hash change(no hash cache entry,
538
# or same size content change), or there is no change on
539
# this file at all.
540
# Provide the parent's hash to the store layer, so that the
541
# content is unchanged we will not store a new node.
542
nostore_sha = parent_entry.text_sha1
543
if store:
544
# We want to record a new node regardless of the presence or
545
# absence of a content change in the file.
546
nostore_sha = None
547
ie.executable = content_summary[2]
548
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
549
try:
550
text = file_obj.read()
551
finally:
552
file_obj.close()
553
try:
554
ie.text_sha1, ie.text_size = self._add_text_to_weave(
555
ie.file_id, text, heads, nostore_sha)
556
# Let the caller know we generated a stat fingerprint.
557
fingerprint = (ie.text_sha1, stat_value)
558
except errors.ExistingContent:
559
# Turns out that the file content was unchanged, and we were
560
# only going to store a new node if it was changed. Carry over
561
# the entry.
562
ie.revision = parent_entry.revision
563
ie.text_size = parent_entry.text_size
564
ie.text_sha1 = parent_entry.text_sha1
565
ie.executable = parent_entry.executable
566
return self._get_delta(ie, basis_inv, path), False, None
567
elif kind == 'directory':
568
if not store:
569
# all data is meta here, nothing specific to directory, so
570
# carry over:
571
ie.revision = parent_entry.revision
572
return self._get_delta(ie, basis_inv, path), False, None
573
self._add_text_to_weave(ie.file_id, '', heads, None)
574
elif kind == 'symlink':
575
current_link_target = content_summary[3]
576
if not store:
577
# symlink target is not generic metadata, check if it has
578
# changed.
579
if current_link_target != parent_entry.symlink_target:
580
store = True
581
if not store:
582
# unchanged, carry over.
583
ie.revision = parent_entry.revision
584
ie.symlink_target = parent_entry.symlink_target
585
return self._get_delta(ie, basis_inv, path), False, None
586
ie.symlink_target = current_link_target
587
self._add_text_to_weave(ie.file_id, '', heads, None)
588
elif kind == 'tree-reference':
589
if not store:
590
if content_summary[3] != parent_entry.reference_revision:
591
store = True
592
if not store:
593
# unchanged, carry over.
594
ie.reference_revision = parent_entry.reference_revision
595
ie.revision = parent_entry.revision
596
return self._get_delta(ie, basis_inv, path), False, None
597
ie.reference_revision = content_summary[3]
598
if ie.reference_revision is None:
599
raise AssertionError("invalid content_summary for nested tree: %r"
600
% (content_summary,))
601
self._add_text_to_weave(ie.file_id, '', heads, None)
602
else:
603
raise NotImplementedError('unknown kind')
604
ie.revision = self._new_revision_id
605
self._any_changes = True
606
return self._get_delta(ie, basis_inv, path), True, fingerprint
607
608
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
609
_entry_factory=entry_factory):
208
610
"""Record a new tree via iter_changes.
209
611
210
612
:param tree: The tree to obtain text contents from for changed objects.
215
617
to basis_revision_id. The iterator must not include any items with
216
618
a current kind of None - missing items must be either filtered out
217
619
or errored-on beefore record_iter_changes sees the item.
620
:param _entry_factory: Private method to bind entry_factory locally for
621
performance.
218
622
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
219
623
tree._observed_sha1.
220
624
"""
221
raise NotImplementedError(self.record_iter_changes)
625
# Create an inventory delta based on deltas between all the parents and
626
# deltas between all the parent inventories. We use inventory delta's
627
# between the inventory objects because iter_changes masks
628
# last-changed-field only changes.
629
# Working data:
630
# file_id -> change map, change is fileid, paths, changed, versioneds,
631
# parents, names, kinds, executables
632
merged_ids = {}
633
# {file_id -> revision_id -> inventory entry, for entries in parent
634
# trees that are not parents[0]
635
parent_entries = {}
636
ghost_basis = False
637
try:
638
revtrees = list(self.repository.revision_trees(self.parents))
639
except errors.NoSuchRevision:
640
# one or more ghosts, slow path.
641
revtrees = []
642
for revision_id in self.parents:
643
try:
644
revtrees.append(self.repository.revision_tree(revision_id))
645
except errors.NoSuchRevision:
646
if not revtrees:
647
basis_revision_id = _mod_revision.NULL_REVISION
648
ghost_basis = True
649
revtrees.append(self.repository.revision_tree(
650
_mod_revision.NULL_REVISION))
651
# The basis inventory from a repository
652
if revtrees:
653
basis_inv = revtrees[0].inventory
654
else:
655
basis_inv = self.repository.revision_tree(
656
_mod_revision.NULL_REVISION).inventory
657
if len(self.parents) > 0:
658
if basis_revision_id != self.parents[0] and not ghost_basis:
659
raise Exception(
660
"arbitrary basis parents not yet supported with merges")
661
for revtree in revtrees[1:]:
662
for change in revtree.inventory._make_delta(basis_inv):
663
if change[1] is None:
664
# Not present in this parent.
665
continue
666
if change[2] not in merged_ids:
667
if change[0] is not None:
668
basis_entry = basis_inv[change[2]]
669
merged_ids[change[2]] = [
670
# basis revid
671
basis_entry.revision,
672
# new tree revid
673
change[3].revision]
674
parent_entries[change[2]] = {
675
# basis parent
676
basis_entry.revision:basis_entry,
677
# this parent
678
change[3].revision:change[3],
679
}
680
else:
681
merged_ids[change[2]] = [change[3].revision]
682
parent_entries[change[2]] = {change[3].revision:change[3]}
683
else:
684
merged_ids[change[2]].append(change[3].revision)
685
parent_entries[change[2]][change[3].revision] = change[3]
686
else:
687
merged_ids = {}
688
# Setup the changes from the tree:
689
# changes maps file_id -> (change, [parent revision_ids])
690
changes= {}
691
for change in iter_changes:
692
# This probably looks up in basis_inv way to much.
693
if change[1][0] is not None:
694
head_candidate = [basis_inv[change[0]].revision]
695
else:
696
head_candidate = []
697
changes[change[0]] = change, merged_ids.get(change[0],
698
head_candidate)
699
unchanged_merged = set(merged_ids) - set(changes)
700
# Extend the changes dict with synthetic changes to record merges of
701
# texts.
702
for file_id in unchanged_merged:
703
# Record a merged version of these items that did not change vs the
704
# basis. This can be either identical parallel changes, or a revert
705
# of a specific file after a merge. The recorded content will be
706
# that of the current tree (which is the same as the basis), but
707
# the per-file graph will reflect a merge.
708
# NB:XXX: We are reconstructing path information we had, this
709
# should be preserved instead.
710
# inv delta change: (file_id, (path_in_source, path_in_target),
711
# changed_content, versioned, parent, name, kind,
712
# executable)
713
try:
714
basis_entry = basis_inv[file_id]
715
except errors.NoSuchId:
716
# a change from basis->some_parents but file_id isn't in basis
717
# so was new in the merge, which means it must have changed
718
# from basis -> current, and as it hasn't the add was reverted
719
# by the user. So we discard this change.
720
pass
721
else:
722
change = (file_id,
723
(basis_inv.id2path(file_id), tree.id2path(file_id)),
724
False, (True, True),
725
(basis_entry.parent_id, basis_entry.parent_id),
726
(basis_entry.name, basis_entry.name),
727
(basis_entry.kind, basis_entry.kind),
728
(basis_entry.executable, basis_entry.executable))
729
changes[file_id] = (change, merged_ids[file_id])
730
# changes contains tuples with the change and a set of inventory
731
# candidates for the file.
732
# inv delta is:
733
# old_path, new_path, file_id, new_inventory_entry
734
seen_root = False # Is the root in the basis delta?
735
inv_delta = self._basis_delta
736
modified_rev = self._new_revision_id
737
for change, head_candidates in changes.values():
738
if change[3][1]: # versioned in target.
739
# Several things may be happening here:
740
# We may have a fork in the per-file graph
741
# - record a change with the content from tree
742
# We may have a change against < all trees
743
# - carry over the tree that hasn't changed
744
# We may have a change against all trees
745
# - record the change with the content from tree
746
kind = change[6][1]
747
file_id = change[0]
748
entry = _entry_factory[kind](file_id, change[5][1],
749
change[4][1])
750
head_set = self._heads(change[0], set(head_candidates))
751
heads = []
752
# Preserve ordering.
753
for head_candidate in head_candidates:
754
if head_candidate in head_set:
755
heads.append(head_candidate)
756
head_set.remove(head_candidate)
757
carried_over = False
758
if len(heads) == 1:
759
# Could be a carry-over situation:
760
parent_entry_revs = parent_entries.get(file_id, None)
761
if parent_entry_revs:
762
parent_entry = parent_entry_revs.get(heads[0], None)
763
else:
764
parent_entry = None
765
if parent_entry is None:
766
# The parent iter_changes was called against is the one
767
# that is the per-file head, so any change is relevant
768
# iter_changes is valid.
769
carry_over_possible = False
770
else:
771
# could be a carry over situation
772
# A change against the basis may just indicate a merge,
773
# we need to check the content against the source of the
774
# merge to determine if it was changed after the merge
775
# or carried over.
776
if (parent_entry.kind != entry.kind or
777
parent_entry.parent_id != entry.parent_id or
778
parent_entry.name != entry.name):
779
# Metadata common to all entries has changed
780
# against per-file parent
781
carry_over_possible = False
782
else:
783
carry_over_possible = True
784
# per-type checks for changes against the parent_entry
785
# are done below.
786
else:
787
# Cannot be a carry-over situation
788
carry_over_possible = False
789
# Populate the entry in the delta
790
if kind == 'file':
791
# XXX: There is still a small race here: If someone reverts the content of a file
792
# after iter_changes examines and decides it has changed,
793
# we will unconditionally record a new version even if some
794
# other process reverts it while commit is running (with
795
# the revert happening after iter_changes did its
796
# examination).
797
if change[7][1]:
798
entry.executable = True
799
else:
800
entry.executable = False
801
if (carry_over_possible and
802
parent_entry.executable == entry.executable):
803
# Check the file length, content hash after reading
804
# the file.
805
nostore_sha = parent_entry.text_sha1
806
else:
807
nostore_sha = None
808
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
809
try:
810
text = file_obj.read()
811
finally:
812
file_obj.close()
813
try:
814
entry.text_sha1, entry.text_size = self._add_text_to_weave(
815
file_id, text, heads, nostore_sha)
816
yield file_id, change[1][1], (entry.text_sha1, stat_value)
817
except errors.ExistingContent:
818
# No content change against a carry_over parent
819
# Perhaps this should also yield a fs hash update?
820
carried_over = True
821
entry.text_size = parent_entry.text_size
822
entry.text_sha1 = parent_entry.text_sha1
823
elif kind == 'symlink':
824
# Wants a path hint?
825
entry.symlink_target = tree.get_symlink_target(file_id)
826
if (carry_over_possible and
827
parent_entry.symlink_target == entry.symlink_target):
828
carried_over = True
829
else:
830
self._add_text_to_weave(change[0], '', heads, None)
831
elif kind == 'directory':
832
if carry_over_possible:
833
carried_over = True
834
else:
835
# Nothing to set on the entry.
836
# XXX: split into the Root and nonRoot versions.
837
if change[1][1] != '' or self.repository.supports_rich_root():
838
self._add_text_to_weave(change[0], '', heads, None)
839
elif kind == 'tree-reference':
840
if not self.repository._format.supports_tree_reference:
841
# This isn't quite sane as an error, but we shouldn't
842
# ever see this code path in practice: tree's don't
843
# permit references when the repo doesn't support tree
844
# references.
845
raise errors.UnsupportedOperation(tree.add_reference,
846
self.repository)
847
reference_revision = tree.get_reference_revision(change[0])
848
entry.reference_revision = reference_revision
849
if (carry_over_possible and
850
parent_entry.reference_revision == reference_revision):
851
carried_over = True
852
else:
853
self._add_text_to_weave(change[0], '', heads, None)
854
else:
855
raise AssertionError('unknown kind %r' % kind)
856
if not carried_over:
857
entry.revision = modified_rev
858
else:
859
entry.revision = parent_entry.revision
860
else:
861
entry = None
862
new_path = change[1][1]
863
inv_delta.append((change[1][0], new_path, change[0], entry))
864
if new_path == '':
865
seen_root = True
866
self.new_inventory = None
867
if len(inv_delta):
868
# This should perhaps be guarded by a check that the basis we
869
# commit against is the basis for the commit and if not do a delta
870
# against the basis.
871
self._any_changes = True
872
if not seen_root:
873
# housekeeping root entry changes do not affect no-change commits.
874
self._require_root_change(tree)
875
self.basis_delta_revision = basis_revision_id
876
877
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
878
parent_keys = tuple([(file_id, parent) for parent in parents])
879
return self.repository.texts._add_text(
880
(file_id, self._new_revision_id), parent_keys, new_text,
881
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
882
883
884
class RootCommitBuilder(CommitBuilder):
885
"""This commitbuilder actually records the root id"""
886
887
# the root entry gets versioned properly by this builder.
888
_versioned_root = True
889
890
def _check_root(self, ie, parent_invs, tree):
891
"""Helper for record_entry_contents.
892
893
:param ie: An entry being added.
894
:param parent_invs: The inventories of the parent revisions of the
895
commit.
896
:param tree: The tree that is being committed.
897
"""
898
899
def _require_root_change(self, tree):
900
"""Enforce an appropriate root object change.
901
902
This is called once when record_iter_changes is called, if and only if
903
the root was not in the delta calculated by record_iter_changes.
904
905
:param tree: The tree which is being committed.
906
"""
907
# versioned roots do not change unless the tree found a change.
222
908
223
909
224
910
class RepositoryWriteLockResult(LogicalLockResult):
249
935
revisions and file history. It's normally accessed only by the Branch,
250
936
which views a particular line of development through that history.
251
937
252
See VersionedFileRepository in bzrlib.vf_repository for the
253
base class for most Bazaar repositories.
938
The Repository builds on top of some byte storage facilies (the revisions,
939
signatures, inventories, texts and chk_bytes attributes) and a Transport,
940
which respectively provide byte storage and a means to access the (possibly
941
remote) disk.
942
943
The byte storage facilities are addressed via tuples, which we refer to
944
as 'keys' throughout the code base. Revision_keys, inventory_keys and
945
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
946
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
947
byte string made up of a hash identifier and a hash value.
948
We use this interface because it allows low friction with the underlying
949
code that implements disk indices, network encoding and other parts of
950
bzrlib.
951
952
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
953
the serialised revisions for the repository. This can be used to obtain
954
revision graph information or to access raw serialised revisions.
955
The result of trying to insert data into the repository via this store
956
is undefined: it should be considered read-only except for implementors
957
of repositories.
958
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
959
the serialised signatures for the repository. This can be used to
960
obtain access to raw serialised signatures. The result of trying to
961
insert data into the repository via this store is undefined: it should
962
be considered read-only except for implementors of repositories.
963
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
964
the serialised inventories for the repository. This can be used to
965
obtain unserialised inventories. The result of trying to insert data
966
into the repository via this store is undefined: it should be
967
considered read-only except for implementors of repositories.
968
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
969
texts of files and directories for the repository. This can be used to
970
obtain file texts or file graphs. Note that Repository.iter_file_bytes
971
is usually a better interface for accessing file texts.
972
The result of trying to insert data into the repository via this store
973
is undefined: it should be considered read-only except for implementors
974
of repositories.
975
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
976
any data the repository chooses to store or have indexed by its hash.
977
The result of trying to insert data into the repository via this store
978
is undefined: it should be considered read-only except for implementors
979
of repositories.
980
:ivar _transport: Transport for file access to repository, typically
981
pointing to .bzr/repository.
254
982
"""
255
983
984
# What class to use for a CommitBuilder. Often it's simpler to change this
985
# in a Repository class subclass rather than to override
986
# get_commit_builder.
987
_commit_builder_class = CommitBuilder
988
256
989
def abort_write_group(self, suppress_errors=False):
257
990
"""Commit the contents accrued within the current write group.
258
991
301
1034
302
1035
:param repository: A repository.
303
1036
"""
304
raise NotImplementedError(self.add_fallback_repository)
1037
if not self._format.supports_external_lookups:
1038
raise errors.UnstackableRepositoryFormat(self._format, self.base)
1039
if self.is_locked():
1040
# This repository will call fallback.unlock() when we transition to
1041
# the unlocked state, so we make sure to increment the lock count
1042
repository.lock_read()
1043
self._check_fallback_repository(repository)
1044
self._fallback_repositories.append(repository)
1045
self.texts.add_fallback_versioned_files(repository.texts)
1046
self.inventories.add_fallback_versioned_files(repository.inventories)
1047
self.revisions.add_fallback_versioned_files(repository.revisions)
1048
self.signatures.add_fallback_versioned_files(repository.signatures)
1049
if self.chk_bytes is not None:
1050
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
305
1051
306
1052
def _check_fallback_repository(self, repository):
307
1053
"""Check that this repository can fallback to repository safely.
312
1058
"""
313
1059
return InterRepository._assert_same_model(self, repository)
314
1060
1061
def add_inventory(self, revision_id, inv, parents):
1062
"""Add the inventory inv to the repository as revision_id.
1063
1064
:param parents: The revision ids of the parents that revision_id
1065
is known to have and are in the repository already.
1066
1067
:returns: The validator(which is a sha1 digest, though what is sha'd is
1068
repository format specific) of the serialized inventory.
1069
"""
1070
if not self.is_in_write_group():
1071
raise AssertionError("%r not in write group" % (self,))
1072
_mod_revision.check_not_reserved_id(revision_id)
1073
if not (inv.revision_id is None or inv.revision_id == revision_id):
1074
raise AssertionError(
1075
"Mismatch between inventory revision"
1076
" id and insertion revid (%r, %r)"
1077
% (inv.revision_id, revision_id))
1078
if inv.root is None:
1079
raise errors.RootMissing()
1080
return self._add_inventory_checked(revision_id, inv, parents)
1081
1082
def _add_inventory_checked(self, revision_id, inv, parents):
1083
"""Add inv to the repository after checking the inputs.
1084
1085
This function can be overridden to allow different inventory styles.
1086
1087
:seealso: add_inventory, for the contract.
1088
"""
1089
inv_lines = self._serializer.write_inventory_to_lines(inv)
1090
return self._inventory_add_lines(revision_id, parents,
1091
inv_lines, check_content=False)
1092
1093
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1094
parents, basis_inv=None, propagate_caches=False):
1095
"""Add a new inventory expressed as a delta against another revision.
1096
1097
See the inventory developers documentation for the theory behind
1098
inventory deltas.
1099
1100
:param basis_revision_id: The inventory id the delta was created
1101
against. (This does not have to be a direct parent.)
1102
:param delta: The inventory delta (see Inventory.apply_delta for
1103
details).
1104
:param new_revision_id: The revision id that the inventory is being
1105
added for.
1106
:param parents: The revision ids of the parents that revision_id is
1107
known to have and are in the repository already. These are supplied
1108
for repositories that depend on the inventory graph for revision
1109
graph access, as well as for those that pun ancestry with delta
1110
compression.
1111
:param basis_inv: The basis inventory if it is already known,
1112
otherwise None.
1113
:param propagate_caches: If True, the caches for this inventory are
1114
copied to and updated for the result if possible.
1115
1116
:returns: (validator, new_inv)
1117
The validator(which is a sha1 digest, though what is sha'd is
1118
repository format specific) of the serialized inventory, and the
1119
resulting inventory.
1120
"""
1121
if not self.is_in_write_group():
1122
raise AssertionError("%r not in write group" % (self,))
1123
_mod_revision.check_not_reserved_id(new_revision_id)
1124
basis_tree = self.revision_tree(basis_revision_id)
1125
basis_tree.lock_read()
1126
try:
1127
# Note that this mutates the inventory of basis_tree, which not all
1128
# inventory implementations may support: A better idiom would be to
1129
# return a new inventory, but as there is no revision tree cache in
1130
# repository this is safe for now - RBC 20081013
1131
if basis_inv is None:
1132
basis_inv = basis_tree.inventory
1133
basis_inv.apply_delta(delta)
1134
basis_inv.revision_id = new_revision_id
1135
return (self.add_inventory(new_revision_id, basis_inv, parents),
1136
basis_inv)
1137
finally:
1138
basis_tree.unlock()
1139
1140
def _inventory_add_lines(self, revision_id, parents, lines,
1141
check_content=True):
1142
"""Store lines in inv_vf and return the sha1 of the inventory."""
1143
parents = [(parent,) for parent in parents]
1144
result = self.inventories.add_lines((revision_id,), parents, lines,
1145
check_content=check_content)[0]
1146
self.inventories._access.flush()
1147
return result
1148
1149
def add_revision(self, revision_id, rev, inv=None, config=None):
1150
"""Add rev to the revision store as revision_id.
1151
1152
:param revision_id: the revision id to use.
1153
:param rev: The revision object.
1154
:param inv: The inventory for the revision. if None, it will be looked
1155
up in the inventory storer
1156
:param config: If None no digital signature will be created.
1157
If supplied its signature_needed method will be used
1158
to determine if a signature should be made.
1159
"""
1160
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1161
# rev.parent_ids?
1162
_mod_revision.check_not_reserved_id(revision_id)
1163
if config is not None and config.signature_needed():
1164
if inv is None:
1165
inv = self.get_inventory(revision_id)
1166
tree = InventoryRevisionTree(self, inv, revision_id)
1167
testament = Testament(rev, tree)
1168
plaintext = testament.as_short_text()
1169
self.store_revision_signature(
1170
gpg.GPGStrategy(config), plaintext, revision_id)
1171
# check inventory present
1172
if not self.inventories.get_parent_map([(revision_id,)]):
1173
if inv is None:
1174
raise errors.WeaveRevisionNotPresent(revision_id,
1175
self.inventories)
1176
else:
1177
# yes, this is not suitable for adding with ghosts.
1178
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1179
rev.parent_ids)
1180
else:
1181
key = (revision_id,)
1182
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1183
self._add_revision(rev)
1184
1185
def _add_revision(self, revision):
1186
text = self._serializer.write_revision_to_string(revision)
1187
key = (revision.revision_id,)
1188
parents = tuple((parent,) for parent in revision.parent_ids)
1189
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1190
315
1191
def all_revision_ids(self):
316
1192
"""Returns a list of all the revision ids in the repository.
317
1193
340
1216
"""
341
1217
self.control_files.break_lock()
342
1218
1219
@needs_read_lock
1220
def _eliminate_revisions_not_present(self, revision_ids):
1221
"""Check every revision id in revision_ids to see if we have it.
1222
1223
Returns a set of the present revisions.
1224
"""
1225
result = []
1226
graph = self.get_graph()
1227
parent_map = graph.get_parent_map(revision_ids)
1228
# The old API returned a list, should this actually be a set?
1229
return parent_map.keys()
1230
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
343
1344
@staticmethod
344
1345
def create(a_bzrdir):
345
1346
"""Construct the current default format repository in a_bzrdir."""
350
1351
351
1352
:param _format: The format of the repository on disk.
352
1353
:param a_bzrdir: The BzrDir of the repository.
353
:param control_files: Control files to use for locking, etc.
354
1354
"""
355
1355
# In the future we will have a single api for all stores for
356
1356
# getting file texts, inventories and revisions, then
363
1363
self._transport = control_files._transport
364
1364
self.base = self._transport.base
365
1365
# for tests
1366
self._reconcile_does_inventory_gc = True
1367
self._reconcile_fixes_text_parents = False
1368
self._reconcile_backsup_inventory = True
366
1369
self._write_group = None
367
1370
# Additional places to query for data.
368
1371
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
369
1377
370
1378
@property
371
1379
def user_transport(self):
511
1519
if revid and committers:
512
1520
result['committers'] = 0
513
1521
if revid and revid != _mod_revision.NULL_REVISION:
514
graph = self.get_graph()
515
1522
if committers:
516
1523
all_committers = set()
517
revisions = [r for (r, p) in graph.iter_ancestry([revid])
518
if r != _mod_revision.NULL_REVISION]
519
last_revision = None
1524
revisions = self.get_ancestry(revid)
1525
# pop the leading None
1526
revisions.pop(0)
1527
first_revision = None
520
1528
if not committers:
521
1529
# ignore the revisions in the middle - just grab first and last
522
1530
revisions = revisions[0], revisions[-1]
523
1531
for revision in self.get_revisions(revisions):
524
if not last_revision:
525
last_revision = revision
1532
if not first_revision:
1533
first_revision = revision
526
1534
if committers:
527
1535
all_committers.add(revision.committer)
528
first_revision = revision
1536
last_revision = revision
529
1537
if committers:
530
1538
result['committers'] = len(all_committers)
531
1539
result['firstrev'] = (first_revision.timestamp,
532
1540
first_revision.timezone)
533
1541
result['latestrev'] = (last_revision.timestamp,
534
1542
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
535
1552
return result
536
1553
537
1554
def find_branches(self, using=False):
574
1591
@needs_read_lock
575
1592
def search_missing_revision_ids(self, other,
576
1593
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
577
find_ghosts=True, revision_ids=None, if_present_ids=None,
578
limit=None):
1594
find_ghosts=True, revision_ids=None, if_present_ids=None):
579
1595
"""Return the revision ids that other has that this does not.
580
1596
581
1597
These are returned in topological order.
594
1610
revision_ids = [revision_id]
595
1611
return InterRepository.get(other, self).search_missing_revision_ids(
596
1612
find_ghosts=find_ghosts, revision_ids=revision_ids,
597
if_present_ids=if_present_ids, limit=limit)
1613
if_present_ids=if_present_ids)
598
1614
599
1615
@staticmethod
600
1616
def open(base):
642
1658
def suspend_write_group(self):
643
1659
raise errors.UnsuspendableWriteGroup(self)
644
1660
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
645
1716
def refresh_data(self):
646
1717
"""Re-read any data needed to synchronise with disk.
647
1718
730
1801
:param lossy: Whether to discard data that can not be natively
731
1802
represented, when pushing to a foreign VCS
732
1803
"""
733
raise NotImplementedError(self.get_commit_builder)
1804
if self._fallback_repositories and not self._format.supports_chks:
1805
raise errors.BzrError("Cannot commit directly to a stacked branch"
1806
" in pre-2a formats. See "
1807
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1808
result = self._commit_builder_class(self, parents, config,
1809
timestamp, timezone, committer, revprops, revision_id,
1810
lossy)
1811
self.start_write_group()
1812
return result
734
1813
735
1814
@only_raises(errors.LockNotHeld, errors.LockBroken)
736
1815
def unlock(self):
743
1822
'Must end write groups before releasing write locks.')
744
1823
self.control_files.unlock()
745
1824
if self.control_files._lock_count == 0:
1825
self._inventory_entry_cache.clear()
746
1826
for repo in self._fallback_repositories:
747
1827
repo.unlock()
748
1828
813
1893
dest_repo = a_bzrdir.open_repository()
814
1894
return dest_repo
815
1895
1896
def _get_sink(self):
1897
"""Return a sink for streaming into this repository."""
1898
return StreamSink(self)
1899
1900
def _get_source(self, to_format):
1901
"""Return a source for streaming from this repository."""
1902
return StreamSource(self, to_format)
1903
816
1904
@needs_read_lock
817
1905
def has_revision(self, revision_id):
818
1906
"""True if this repository has a copy of the revision."""
825
1913
:param revision_ids: An iterable of revision_ids.
826
1914
:return: A set of the revision_ids that were present.
827
1915
"""
828
raise NotImplementedError(self.has_revisions)
1916
parent_map = self.revisions.get_parent_map(
1917
[(rev_id,) for rev_id in revision_ids])
1918
result = set()
1919
if _mod_revision.NULL_REVISION in revision_ids:
1920
result.add(_mod_revision.NULL_REVISION)
1921
result.update([key[0] for key in parent_map])
1922
return result
829
1923
830
1924
@needs_read_lock
831
1925
def get_revision(self, revision_id):
832
1926
"""Return the Revision object for a named revision."""
833
1927
return self.get_revisions([revision_id])[0]
834
1928
1929
@needs_read_lock
835
1930
def get_revision_reconcile(self, revision_id):
836
1931
"""'reconcile' helper routine that allows access to a revision always.
837
1932
840
1935
be used by reconcile, or reconcile-alike commands that are correcting
841
1936
or testing the revision graph.
842
1937
"""
843
raise NotImplementedError(self.get_revision_reconcile)
1938
return self._get_revisions([revision_id])[0]
844
1939
1940
@needs_read_lock
845
1941
def get_revisions(self, revision_ids):
846
1942
"""Get many revisions at once.
847
1943
848
1944
Repositories that need to check data on every revision read should
849
1945
subclass this method.
850
1946
"""
851
raise NotImplementedError(self.get_revisions)
1947
return self._get_revisions(revision_ids)
1948
1949
@needs_read_lock
1950
def _get_revisions(self, revision_ids):
1951
"""Core work logic to get many revisions without sanity checks."""
1952
revs = {}
1953
for revid, rev in self._iter_revisions(revision_ids):
1954
if rev is None:
1955
raise errors.NoSuchRevision(self, revid)
1956
revs[revid] = rev
1957
return [revs[revid] for revid in revision_ids]
1958
1959
def _iter_revisions(self, revision_ids):
1960
"""Iterate over revision objects.
1961
1962
:param revision_ids: An iterable of revisions to examine. None may be
1963
passed to request all revisions known to the repository. Note that
1964
not all repositories can find unreferenced revisions; for those
1965
repositories only referenced ones will be returned.
1966
:return: An iterator of (revid, revision) tuples. Absent revisions (
1967
those asked for but not available) are returned as (revid, None).
1968
"""
1969
if revision_ids is None:
1970
revision_ids = self.all_revision_ids()
1971
else:
1972
for rev_id in revision_ids:
1973
if not rev_id or not isinstance(rev_id, basestring):
1974
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1975
keys = [(key,) for key in revision_ids]
1976
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1977
for record in stream:
1978
revid = record.key[0]
1979
if record.storage_kind == 'absent':
1980
yield (revid, None)
1981
else:
1982
text = record.get_bytes_as('fulltext')
1983
rev = self._serializer.read_revision_from_string(text)
1984
yield (revid, rev)
852
1985
853
1986
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
854
1987
"""Produce a generator of revision deltas.
909
2042
signature = gpg_strategy.sign(plaintext)
910
2043
self.add_signature_text(revision_id, signature)
911
2044
2045
@needs_write_lock
912
2046
def add_signature_text(self, revision_id, signature):
913
"""Store a signature text for a revision.
914
915
:param revision_id: Revision id of the revision
916
:param signature: Signature text.
917
"""
918
raise NotImplementedError(self.add_signature_text)
2047
self.signatures.add_lines((revision_id,), (),
2048
osutils.split_lines(signature))
2049
2050
def find_text_key_references(self):
2051
"""Find the text key references within the repository.
2052
2053
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2054
to whether they were referred to by the inventory of the
2055
revision_id that they contain. The inventory texts from all present
2056
revision ids are assessed to generate this report.
2057
"""
2058
revision_keys = self.revisions.keys()
2059
w = self.inventories
2060
pb = ui.ui_factory.nested_progress_bar()
2061
try:
2062
return self._serializer._find_text_key_references(
2063
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
2064
finally:
2065
pb.finished()
2066
2067
def _inventory_xml_lines_for_keys(self, keys):
2068
"""Get a line iterator of the sort needed for findind references.
2069
2070
Not relevant for non-xml inventory repositories.
2071
2072
Ghosts in revision_keys are ignored.
2073
2074
:param revision_keys: The revision keys for the inventories to inspect.
2075
:return: An iterator over (inventory line, revid) for the fulltexts of
2076
all of the xml inventories specified by revision_keys.
2077
"""
2078
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2079
for record in stream:
2080
if record.storage_kind != 'absent':
2081
chunks = record.get_bytes_as('chunked')
2082
revid = record.key[-1]
2083
lines = osutils.chunks_to_lines(chunks)
2084
for line in lines:
2085
yield line, revid
2086
2087
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2088
revision_keys):
2089
"""Helper routine for fileids_altered_by_revision_ids.
2090
2091
This performs the translation of xml lines to revision ids.
2092
2093
:param line_iterator: An iterator of lines, origin_version_id
2094
:param revision_keys: The revision ids to filter for. This should be a
2095
set or other type which supports efficient __contains__ lookups, as
2096
the revision key from each parsed line will be looked up in the
2097
revision_keys filter.
2098
:return: a dictionary mapping altered file-ids to an iterable of
2099
revision_ids. Each altered file-ids has the exact revision_ids that
2100
altered it listed explicitly.
2101
"""
2102
seen = set(self._serializer._find_text_key_references(
2103
line_iterator).iterkeys())
2104
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2105
parent_seen = set(self._serializer._find_text_key_references(
2106
self._inventory_xml_lines_for_keys(parent_keys)))
2107
new_keys = seen - parent_seen
2108
result = {}
2109
setdefault = result.setdefault
2110
for key in new_keys:
2111
setdefault(key[0], set()).add(key[-1])
2112
return result
919
2113
920
2114
def _find_parent_ids_of_revisions(self, revision_ids):
921
2115
"""Find all parent ids that are mentioned in the revision graph.
930
2124
parent_ids.discard(_mod_revision.NULL_REVISION)
931
2125
return parent_ids
932
2126
933
def fileids_altered_by_revision_ids(self, revision_ids):
2127
def _find_parent_keys_of_revisions(self, revision_keys):
2128
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2129
2130
:param revision_keys: An iterable of revision_keys.
2131
:return: The parents of all revision_keys that are not already in
2132
revision_keys
2133
"""
2134
parent_map = self.revisions.get_parent_map(revision_keys)
2135
parent_keys = set()
2136
map(parent_keys.update, parent_map.itervalues())
2137
parent_keys.difference_update(revision_keys)
2138
parent_keys.discard(_mod_revision.NULL_REVISION)
2139
return parent_keys
2140
2141
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
934
2142
"""Find the file ids and versions affected by revisions.
935
2143
936
2144
:param revisions: an iterable containing revision ids.
2145
:param _inv_weave: The inventory weave from this repository or None.
2146
If None, the inventory weave will be opened automatically.
937
2147
:return: a dictionary mapping altered file-ids to an iterable of
938
revision_ids. Each altered file-ids has the exact revision_ids
939
that altered it listed explicitly.
2148
revision_ids. Each altered file-ids has the exact revision_ids that
2149
altered it listed explicitly.
940
2150
"""
941
raise NotImplementedError(self.fileids_altered_by_revision_ids)
2151
selected_keys = set((revid,) for revid in revision_ids)
2152
w = _inv_weave or self.inventories
2153
return self._find_file_ids_from_xml_inventory_lines(
2154
w.iter_lines_added_or_present_in_keys(
2155
selected_keys, pb=None),
2156
selected_keys)
942
2157
943
2158
def iter_files_bytes(self, desired_files):
944
2159
"""Iterate through file versions.
951
2166
uniquely identify the file version in the caller's context. (Examples:
952
2167
an index number or a TreeTransform trans_id.)
953
2168
2169
bytes_iterator is an iterable of bytestrings for the file. The
2170
kind of iterable and length of the bytestrings are unspecified, but for
2171
this implementation, it is a list of bytes produced by
2172
VersionedFile.get_record_stream().
2173
954
2174
:param desired_files: a list of (file_id, revision_id, identifier)
955
2175
triples
956
2176
"""
957
raise NotImplementedError(self.iter_files_bytes)
2177
text_keys = {}
2178
for file_id, revision_id, callable_data in desired_files:
2179
text_keys[(file_id, revision_id)] = callable_data
2180
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2181
if record.storage_kind == 'absent':
2182
raise errors.RevisionNotPresent(record.key, self)
2183
yield text_keys[record.key], record.get_bytes_as('chunked')
2184
2185
def _generate_text_key_index(self, text_key_references=None,
2186
ancestors=None):
2187
"""Generate a new text key index for the repository.
2188
2189
This is an expensive function that will take considerable time to run.
2190
2191
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2192
list of parents, also text keys. When a given key has no parents,
2193
the parents list will be [NULL_REVISION].
2194
"""
2195
# All revisions, to find inventory parents.
2196
if ancestors is None:
2197
graph = self.get_graph()
2198
ancestors = graph.get_parent_map(self.all_revision_ids())
2199
if text_key_references is None:
2200
text_key_references = self.find_text_key_references()
2201
pb = ui.ui_factory.nested_progress_bar()
2202
try:
2203
return self._do_generate_text_key_index(ancestors,
2204
text_key_references, pb)
2205
finally:
2206
pb.finished()
2207
2208
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2209
"""Helper for _generate_text_key_index to avoid deep nesting."""
2210
revision_order = tsort.topo_sort(ancestors)
2211
invalid_keys = set()
2212
revision_keys = {}
2213
for revision_id in revision_order:
2214
revision_keys[revision_id] = set()
2215
text_count = len(text_key_references)
2216
# a cache of the text keys to allow reuse; costs a dict of all the
2217
# keys, but saves a 2-tuple for every child of a given key.
2218
text_key_cache = {}
2219
for text_key, valid in text_key_references.iteritems():
2220
if not valid:
2221
invalid_keys.add(text_key)
2222
else:
2223
revision_keys[text_key[1]].add(text_key)
2224
text_key_cache[text_key] = text_key
2225
del text_key_references
2226
text_index = {}
2227
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2228
NULL_REVISION = _mod_revision.NULL_REVISION
2229
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2230
# too small for large or very branchy trees. However, for 55K path
2231
# trees, it would be easy to use too much memory trivially. Ideally we
2232
# could gauge this by looking at available real memory etc, but this is
2233
# always a tricky proposition.
2234
inventory_cache = lru_cache.LRUCache(10)
2235
batch_size = 10 # should be ~150MB on a 55K path tree
2236
batch_count = len(revision_order) / batch_size + 1
2237
processed_texts = 0
2238
pb.update("Calculating text parents", processed_texts, text_count)
2239
for offset in xrange(batch_count):
2240
to_query = revision_order[offset * batch_size:(offset + 1) *
2241
batch_size]
2242
if not to_query:
2243
break
2244
for revision_id in to_query:
2245
parent_ids = ancestors[revision_id]
2246
for text_key in revision_keys[revision_id]:
2247
pb.update("Calculating text parents", processed_texts)
2248
processed_texts += 1
2249
candidate_parents = []
2250
for parent_id in parent_ids:
2251
parent_text_key = (text_key[0], parent_id)
2252
try:
2253
check_parent = parent_text_key not in \
2254
revision_keys[parent_id]
2255
except KeyError:
2256
# the parent parent_id is a ghost:
2257
check_parent = False
2258
# truncate the derived graph against this ghost.
2259
parent_text_key = None
2260
if check_parent:
2261
# look at the parent commit details inventories to
2262
# determine possible candidates in the per file graph.
2263
# TODO: cache here.
2264
try:
2265
inv = inventory_cache[parent_id]
2266
except KeyError:
2267
inv = self.revision_tree(parent_id).inventory
2268
inventory_cache[parent_id] = inv
2269
try:
2270
parent_entry = inv[text_key[0]]
2271
except (KeyError, errors.NoSuchId):
2272
parent_entry = None
2273
if parent_entry is not None:
2274
parent_text_key = (
2275
text_key[0], parent_entry.revision)
2276
else:
2277
parent_text_key = None
2278
if parent_text_key is not None:
2279
candidate_parents.append(
2280
text_key_cache[parent_text_key])
2281
parent_heads = text_graph.heads(candidate_parents)
2282
new_parents = list(parent_heads)
2283
new_parents.sort(key=lambda x:candidate_parents.index(x))
2284
if new_parents == []:
2285
new_parents = [NULL_REVISION]
2286
text_index[text_key] = new_parents
2287
2288
for text_key in invalid_keys:
2289
text_index[text_key] = [NULL_REVISION]
2290
return text_index
2291
2292
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2293
"""Get an iterable listing the keys of all the data introduced by a set
2294
of revision IDs.
2295
2296
The keys will be ordered so that the corresponding items can be safely
2297
fetched and inserted in that order.
2298
2299
:returns: An iterable producing tuples of (knit-kind, file-id,
2300
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2301
'revisions'. file-id is None unless knit-kind is 'file'.
2302
"""
2303
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2304
yield result
2305
del _files_pb
2306
for result in self._find_non_file_keys_to_fetch(revision_ids):
2307
yield result
2308
2309
def _find_file_keys_to_fetch(self, revision_ids, pb):
2310
# XXX: it's a bit weird to control the inventory weave caching in this
2311
# generator. Ideally the caching would be done in fetch.py I think. Or
2312
# maybe this generator should explicitly have the contract that it
2313
# should not be iterated until the previously yielded item has been
2314
# processed?
2315
inv_w = self.inventories
2316
2317
# file ids that changed
2318
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2319
count = 0
2320
num_file_ids = len(file_ids)
2321
for file_id, altered_versions in file_ids.iteritems():
2322
if pb is not None:
2323
pb.update("Fetch texts", count, num_file_ids)
2324
count += 1
2325
yield ("file", file_id, altered_versions)
2326
2327
def _find_non_file_keys_to_fetch(self, revision_ids):
2328
# inventory
2329
yield ("inventory", None, revision_ids)
2330
2331
# signatures
2332
# XXX: Note ATM no callers actually pay attention to this return
2333
# instead they just use the list of revision ids and ignore
2334
# missing sigs. Consider removing this work entirely
2335
revisions_with_signatures = set(self.signatures.get_parent_map(
2336
[(r,) for r in revision_ids]))
2337
revisions_with_signatures = set(
2338
[r for (r,) in revisions_with_signatures])
2339
revisions_with_signatures.intersection_update(revision_ids)
2340
yield ("signatures", None, revisions_with_signatures)
2341
2342
# revisions
2343
yield ("revisions", None, revision_ids)
2344
2345
@needs_read_lock
2346
def get_inventory(self, revision_id):
2347
"""Get Inventory object by revision id."""
2348
return self.iter_inventories([revision_id]).next()
2349
2350
def iter_inventories(self, revision_ids, ordering=None):
2351
"""Get many inventories by revision_ids.
2352
2353
This will buffer some or all of the texts used in constructing the
2354
inventories in memory, but will only parse a single inventory at a
2355
time.
2356
2357
:param revision_ids: The expected revision ids of the inventories.
2358
:param ordering: optional ordering, e.g. 'topological'. If not
2359
specified, the order of revision_ids will be preserved (by
2360
buffering if necessary).
2361
:return: An iterator of inventories.
2362
"""
2363
if ((None in revision_ids)
2364
or (_mod_revision.NULL_REVISION in revision_ids)):
2365
raise ValueError('cannot get null revision inventory')
2366
return self._iter_inventories(revision_ids, ordering)
2367
2368
def _iter_inventories(self, revision_ids, ordering):
2369
"""single-document based inventory iteration."""
2370
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2371
for text, revision_id in inv_xmls:
2372
yield self._deserialise_inventory(revision_id, text)
2373
2374
def _iter_inventory_xmls(self, revision_ids, ordering):
2375
if ordering is None:
2376
order_as_requested = True
2377
ordering = 'unordered'
2378
else:
2379
order_as_requested = False
2380
keys = [(revision_id,) for revision_id in revision_ids]
2381
if not keys:
2382
return
2383
if order_as_requested:
2384
key_iter = iter(keys)
2385
next_key = key_iter.next()
2386
stream = self.inventories.get_record_stream(keys, ordering, True)
2387
text_chunks = {}
2388
for record in stream:
2389
if record.storage_kind != 'absent':
2390
chunks = record.get_bytes_as('chunked')
2391
if order_as_requested:
2392
text_chunks[record.key] = chunks
2393
else:
2394
yield ''.join(chunks), record.key[-1]
2395
else:
2396
raise errors.NoSuchRevision(self, record.key)
2397
if order_as_requested:
2398
# Yield as many results as we can while preserving order.
2399
while next_key in text_chunks:
2400
chunks = text_chunks.pop(next_key)
2401
yield ''.join(chunks), next_key[-1]
2402
try:
2403
next_key = key_iter.next()
2404
except StopIteration:
2405
# We still want to fully consume the get_record_stream,
2406
# just in case it is not actually finished at this point
2407
next_key = None
2408
break
2409
2410
def _deserialise_inventory(self, revision_id, xml):
2411
"""Transform the xml into an inventory object.
2412
2413
:param revision_id: The expected revision id of the inventory.
2414
:param xml: A serialised inventory.
2415
"""
2416
result = self._serializer.read_inventory_from_string(xml, revision_id,
2417
entry_cache=self._inventory_entry_cache,
2418
return_from_cache=self._safe_to_return_from_cache)
2419
if result.revision_id != revision_id:
2420
raise AssertionError('revision id mismatch %s != %s' % (
2421
result.revision_id, revision_id))
2422
return result
2423
2424
def get_serializer_format(self):
2425
return self._serializer.format_num
2426
2427
@needs_read_lock
2428
def _get_inventory_xml(self, revision_id):
2429
"""Get serialized inventory as a string."""
2430
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2431
try:
2432
text, revision_id = texts.next()
2433
except StopIteration:
2434
raise errors.HistoryMissing(self, 'inventory', revision_id)
2435
return text
958
2436
959
2437
def get_rev_id_for_revno(self, revno, known_pair):
960
2438
"""Return the revision id of a revno, given a later (revno, revid)
991
2469
raise AssertionError('_iter_for_revno returned too much history')
992
2470
return (True, partial_history[-1])
993
2471
994
@symbol_versioning.deprecated_method(symbol_versioning.deprecated_in((2, 4, 0)))
995
2472
def iter_reverse_revision_history(self, revision_id):
996
2473
"""Iterate backwards through revision ids in the lefthand history
997
2474
1034
2511
1035
2512
`revision_id` may be NULL_REVISION for the empty tree revision.
1036
2513
"""
1037
raise NotImplementedError(self.revision_tree)
2514
revision_id = _mod_revision.ensure_null(revision_id)
2515
# TODO: refactor this to use an existing revision object
2516
# so we don't need to read it in twice.
2517
if revision_id == _mod_revision.NULL_REVISION:
2518
return InventoryRevisionTree(self,
2519
Inventory(root_id=None), _mod_revision.NULL_REVISION)
2520
else:
2521
inv = self.get_inventory(revision_id)
2522
return InventoryRevisionTree(self, inv, revision_id)
1038
2523
1039
2524
def revision_trees(self, revision_ids):
1040
2525
"""Return Trees for revisions in this repository.
1042
2527
:param revision_ids: a sequence of revision-ids;
1043
2528
a revision-id may not be None or 'null:'
1044
2529
"""
1045
raise NotImplementedError(self.revision_trees)
2530
inventories = self.iter_inventories(revision_ids)
2531
for inv in inventories:
2532
yield InventoryRevisionTree(self, inv, inv.revision_id)
2533
2534
def _filtered_revision_trees(self, revision_ids, file_ids):
2535
"""Return Tree for a revision on this branch with only some files.
2536
2537
:param revision_ids: a sequence of revision-ids;
2538
a revision-id may not be None or 'null:'
2539
:param file_ids: if not None, the result is filtered
2540
so that only those file-ids, their parents and their
2541
children are included.
2542
"""
2543
inventories = self.iter_inventories(revision_ids)
2544
for inv in inventories:
2545
# Should we introduce a FilteredRevisionTree class rather
2546
# than pre-filter the inventory here?
2547
filtered_inv = inv.filter(file_ids)
2548
yield InventoryRevisionTree(self, filtered_inv, filtered_inv.revision_id)
1046
2549
1047
2550
@needs_read_lock
1048
@symbol_versioning.deprecated_method(
1049
symbol_versioning.deprecated_in((2, 4, 0)))
1050
2551
def get_ancestry(self, revision_id, topo_sorted=True):
1051
2552
"""Return a list of revision-ids integrated by a revision.
1052
2553
1056
2557
1057
2558
This is topologically sorted.
1058
2559
"""
1059
if 'evil' in debug.debug_flags:
1060
mutter_callsite(2, "get_ancestry is linear with history.")
1061
2560
if _mod_revision.is_null(revision_id):
1062
2561
return [None]
1063
2562
if not self.has_revision(revision_id):
1104
2603
1105
2604
def get_parent_map(self, revision_ids):
1106
2605
"""See graph.StackedParentsProvider.get_parent_map"""
1107
raise NotImplementedError(self.get_parent_map)
1108
1109
def _get_parent_map_no_fallbacks(self, revision_ids):
1110
"""Same as Repository.get_parent_map except doesn't query fallbacks."""
1111
2606
# revisions index works in keys; this just works in revisions
1112
2607
# therefore wrap and unwrap
1113
2608
query_keys = []
1119
2614
raise ValueError('get_parent_map(None) is not valid')
1120
2615
else:
1121
2616
query_keys.append((revision_id ,))
1122
vf = self.revisions.without_fallbacks()
1123
2617
for ((revision_id,), parent_keys) in \
1124
vf.get_parent_map(query_keys).iteritems():
2618
self.revisions.get_parent_map(query_keys).iteritems():
1125
2619
if parent_keys:
1126
2620
result[revision_id] = tuple([parent_revid
1127
2621
for (parent_revid,) in parent_keys])
1130
2624
return result
1131
2625
1132
2626
def _make_parents_provider(self):
1133
if not self._format.supports_external_lookups:
1134
return self
1135
return graph.StackedParentsProvider(_LazyListJoin(
1136
[self._make_parents_provider_unstacked()],
1137
self._fallback_repositories))
1138
1139
def _make_parents_provider_unstacked(self):
1140
return graph.CallableToParentsProviderAdapter(
1141
self._get_parent_map_no_fallbacks)
2627
return self
1142
2628
1143
2629
@needs_read_lock
1144
2630
def get_known_graph_ancestry(self, revision_ids):
1145
2631
"""Return the known graph for a set of revision ids and their ancestors.
1146
2632
"""
1147
raise NotImplementedError(self.get_known_graph_ancestry)
1148
1149
def get_file_graph(self):
1150
"""Return the graph walker for files."""
1151
raise NotImplementedError(self.get_file_graph)
2633
st = static_tuple.StaticTuple
2634
revision_keys = [st(r_id).intern() for r_id in revision_ids]
2635
known_graph = self.revisions.get_known_graph_ancestry(revision_keys)
2636
return graph.GraphThunkIdsToKeys(known_graph)
1152
2637
1153
2638
def get_graph(self, other_repository=None):
1154
2639
"""Return the graph walker for this repository format"""
1159
2644
[parents_provider, other_repository._make_parents_provider()])
1160
2645
return graph.Graph(parents_provider)
1161
2646
2647
def _get_versioned_file_checker(self, text_key_references=None,
2648
ancestors=None):
2649
"""Return an object suitable for checking versioned files.
2650
2651
:param text_key_references: if non-None, an already built
2652
dictionary mapping text keys ((fileid, revision_id) tuples)
2653
to whether they were referred to by the inventory of the
2654
revision_id that they contain. If None, this will be
2655
calculated.
2656
:param ancestors: Optional result from
2657
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2658
available.
2659
"""
2660
return _VersionedFileChecker(self,
2661
text_key_references=text_key_references, ancestors=ancestors)
2662
1162
2663
def revision_ids_to_search_result(self, result_set):
1163
2664
"""Convert a set of revision ids to a graph SearchResult."""
1164
2665
result_parents = set()
1190
2691
1191
2692
@needs_write_lock
1192
2693
def sign_revision(self, revision_id, gpg_strategy):
1193
testament = _mod_testament.Testament.from_revision(self, revision_id)
1194
plaintext = testament.as_short_text()
2694
plaintext = Testament.from_revision(self, revision_id).as_short_text()
1195
2695
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
1196
2696
1197
2697
@needs_read_lock
1198
def verify_revision(self, revision_id, gpg_strategy):
1199
"""Verify the signature on a revision.
1200
1201
:param revision_id: the revision to verify
1202
:gpg_strategy: the GPGStrategy object to used
1203
1204
:return: gpg.SIGNATURE_VALID or a failed SIGNATURE_ value
1205
"""
1206
if not self.has_signature_for_revision_id(revision_id):
1207
return gpg.SIGNATURE_NOT_SIGNED, None
1208
signature = self.get_signature_text(revision_id)
1209
1210
testament = _mod_testament.Testament.from_revision(self, revision_id)
1211
plaintext = testament.as_short_text()
1212
1213
return gpg_strategy.verify(signature, plaintext)
1214
1215
2698
def has_signature_for_revision_id(self, revision_id):
1216
2699
"""Query for a revision signature for revision_id in the repository."""
1217
raise NotImplementedError(self.has_signature_for_revision_id)
2700
if not self.has_revision(revision_id):
2701
raise errors.NoSuchRevision(self, revision_id)
2702
sig_present = (1 == len(
2703
self.signatures.get_parent_map([(revision_id,)])))
2704
return sig_present
1218
2705
2706
@needs_read_lock
1219
2707
def get_signature_text(self, revision_id):
1220
2708
"""Return the text for a signature."""
1221
raise NotImplementedError(self.get_signature_text)
2709
stream = self.signatures.get_record_stream([(revision_id,)],
2710
'unordered', True)
2711
record = stream.next()
2712
if record.storage_kind == 'absent':
2713
raise errors.NoSuchRevision(self, revision_id)
2714
return record.get_bytes_as('fulltext')
1222
2715
2716
@needs_read_lock
1223
2717
def check(self, revision_ids=None, callback_refs=None, check_repo=True):
1224
2718
"""Check consistency of all history of given revision_ids.
1225
2719
1233
2727
:param check_repo: If False do not check the repository contents, just
1234
2728
calculate the data callback_refs requires and call them back.
1235
2729
"""
1236
return self._check(revision_ids=revision_ids, callback_refs=callback_refs,
2730
return self._check(revision_ids, callback_refs=callback_refs,
1237
2731
check_repo=check_repo)
1238
2732
1239
def _check(self, revision_ids=None, callback_refs=None, check_repo=True):
1240
raise NotImplementedError(self.check)
2733
def _check(self, revision_ids, callback_refs, check_repo):
2734
result = check.Check(self, check_repo=check_repo)
2735
result.check(callback_refs)
2736
return result
1241
2737
1242
2738
def _warn_if_deprecated(self, branch=None):
1243
2739
if not self._format.is_deprecated():
1277
2773
except UnicodeDecodeError:
1278
2774
raise errors.NonAsciiRevisionId(method, self)
1279
2775
2776
def _find_inconsistent_revision_parents(self, revisions_iterator=None):
2777
"""Find revisions with different parent lists in the revision object
2778
and in the index graph.
2779
2780
:param revisions_iterator: None, or an iterator of (revid,
2781
Revision-or-None). This iterator controls the revisions checked.
2782
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
2783
parents-in-revision).
2784
"""
2785
if not self.is_locked():
2786
raise AssertionError()
2787
vf = self.revisions
2788
if revisions_iterator is None:
2789
revisions_iterator = self._iter_revisions(None)
2790
for revid, revision in revisions_iterator:
2791
if revision is None:
2792
pass
2793
parent_map = vf.get_parent_map([(revid,)])
2794
parents_according_to_index = tuple(parent[-1] for parent in
2795
parent_map[(revid,)])
2796
parents_according_to_revision = tuple(revision.parent_ids)
2797
if parents_according_to_index != parents_according_to_revision:
2798
yield (revid, parents_according_to_index,
2799
parents_according_to_revision)
2800
2801
def _check_for_inconsistent_revision_parents(self):
2802
inconsistencies = list(self._find_inconsistent_revision_parents())
2803
if inconsistencies:
2804
raise errors.BzrCheckError(
2805
"Revision knit has inconsistent parents.")
2806
2807
2808
def install_revision(repository, rev, revision_tree):
2809
"""Install all revision data into a repository."""
2810
install_revisions(repository, [(rev, revision_tree, None)])
2811
2812
2813
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2814
"""Install all revision data into a repository.
2815
2816
Accepts an iterable of revision, tree, signature tuples. The signature
2817
may be None.
2818
"""
2819
repository.start_write_group()
2820
try:
2821
inventory_cache = lru_cache.LRUCache(10)
2822
for n, (revision, revision_tree, signature) in enumerate(iterable):
2823
_install_revision(repository, revision, revision_tree, signature,
2824
inventory_cache)
2825
if pb is not None:
2826
pb.update('Transferring revisions', n + 1, num_revisions)
2827
except:
2828
repository.abort_write_group()
2829
raise
2830
else:
2831
repository.commit_write_group()
2832
2833
2834
def _install_revision(repository, rev, revision_tree, signature,
2835
inventory_cache):
2836
"""Install all revision data into a repository."""
2837
present_parents = []
2838
parent_trees = {}
2839
for p_id in rev.parent_ids:
2840
if repository.has_revision(p_id):
2841
present_parents.append(p_id)
2842
parent_trees[p_id] = repository.revision_tree(p_id)
2843
else:
2844
parent_trees[p_id] = repository.revision_tree(
2845
_mod_revision.NULL_REVISION)
2846
2847
inv = revision_tree.inventory
2848
entries = inv.iter_entries()
2849
# backwards compatibility hack: skip the root id.
2850
if not repository.supports_rich_root():
2851
path, root = entries.next()
2852
if root.revision != rev.revision_id:
2853
raise errors.IncompatibleRevision(repr(repository))
2854
text_keys = {}
2855
for path, ie in entries:
2856
text_keys[(ie.file_id, ie.revision)] = ie
2857
text_parent_map = repository.texts.get_parent_map(text_keys)
2858
missing_texts = set(text_keys) - set(text_parent_map)
2859
# Add the texts that are not already present
2860
for text_key in missing_texts:
2861
ie = text_keys[text_key]
2862
text_parents = []
2863
# FIXME: TODO: The following loop overlaps/duplicates that done by
2864
# commit to determine parents. There is a latent/real bug here where
2865
# the parents inserted are not those commit would do - in particular
2866
# they are not filtered by heads(). RBC, AB
2867
for revision, tree in parent_trees.iteritems():
2868
if ie.file_id not in tree:
2869
continue
2870
parent_id = tree.get_file_revision(ie.file_id)
2871
if parent_id in text_parents:
2872
continue
2873
text_parents.append((ie.file_id, parent_id))
2874
lines = revision_tree.get_file(ie.file_id).readlines()
2875
repository.texts.add_lines(text_key, text_parents, lines)
2876
try:
2877
# install the inventory
2878
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2879
# Cache this inventory
2880
inventory_cache[rev.revision_id] = inv
2881
try:
2882
basis_inv = inventory_cache[rev.parent_ids[0]]
2883
except KeyError:
2884
repository.add_inventory(rev.revision_id, inv, present_parents)
2885
else:
2886
delta = inv._make_delta(basis_inv)
2887
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2888
rev.revision_id, present_parents)
2889
else:
2890
repository.add_inventory(rev.revision_id, inv, present_parents)
2891
except errors.RevisionAlreadyPresent:
2892
pass
2893
if signature is not None:
2894
repository.add_signature_text(rev.revision_id, signature)
2895
repository.add_revision(rev.revision_id, rev, inv)
2896
1280
2897
1281
2898
class MetaDirRepository(Repository):
1282
2899
"""Repositories in the new meta-dir layout.
1317
2934
return not self._transport.has('no-working-trees')
1318
2935
1319
2936
2937
class MetaDirVersionedFileRepository(MetaDirRepository):
2938
"""Repositories in a meta-dir, that work via versioned file objects."""
2939
2940
def __init__(self, _format, a_bzrdir, control_files):
2941
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2942
control_files)
2943
2944
1320
2945
class RepositoryFormatRegistry(controldir.ControlComponentFormatRegistry):
1321
2946
"""Repository format registry."""
1322
2947
1387
3012
# Does this format support CHK bytestring lookups. Set to True or False in
1388
3013
# derived classes.
1389
3014
supports_chks = None
3015
# Should commit add an inventory, or an inventory delta to the repository.
3016
_commit_inv_deltas = True
3017
# What order should fetch operations request streams in?
3018
# The default is unordered as that is the cheapest for an origin to
3019
# provide.
3020
_fetch_order = 'unordered'
3021
# Does this repository format use deltas that can be fetched as-deltas ?
3022
# (E.g. knits, where the knit deltas can be transplanted intact.
3023
# We default to False, which will ensure that enough data to get
3024
# a full text out of any fetch stream will be grabbed.
3025
_fetch_uses_deltas = False
1390
3026
# Should fetch trigger a reconcile after the fetch? Only needed for
1391
3027
# some repository formats that can suffer internal inconsistencies.
1392
3028
_fetch_reconcile = False
1398
3034
# help), and for fetching when data won't have come from the same
1399
3035
# compressor.
1400
3036
pack_compresses = False
1401
# Does the repository storage understand references to trees?
3037
# Does the repository inventory storage understand references to trees?
1402
3038
supports_tree_reference = None
1403
3039
# Is the format experimental ?
1404
3040
experimental = False
1405
# Does this repository format escape funky characters, or does it create
1406
# files with similar names as the versioned files in its contents on disk
1407
# ?
3041
# Does this repository format escape funky characters, or does it create files with
3042
# similar names as the versioned files in its contents on disk ?
1408
3043
supports_funky_characters = None
1409
3044
# Does this repository format support leaving locks?
1410
3045
supports_leaving_lock = None
1414
3049
supports_revision_signatures = True
1415
3050
# Can the revision graph have incorrect parents?
1416
3051
revision_graph_can_have_wrong_parents = None
1417
# Does this format support rich root data?
1418
rich_root_data = None
1419
# Does this format support explicitly versioned directories?
1420
supports_versioned_directories = None
1421
3052
1422
3053
def __repr__(self):
1423
3054
return "%s()" % self.__class__.__name__
1692
3323
InterRepository.get(other).method_name(parameters).
1693
3324
"""
1694
3325
3326
_walk_to_common_revisions_batch_size = 50
1695
3327
_optimisers = []
1696
3328
"""The available optimised InterRepository types."""
1697
3329
1722
3354
content is copied.
1723
3355
:return: None.
1724
3356
"""
1725
raise NotImplementedError(self.fetch)
3357
ui.ui_factory.warn_experimental_format_fetch(self)
3358
from bzrlib.fetch import RepoFetcher
3359
# See <https://launchpad.net/bugs/456077> asking for a warning here
3360
if self.source._format.network_name() != self.target._format.network_name():
3361
ui.ui_factory.show_user_warning('cross_format_fetch',
3362
from_format=self.source._format,
3363
to_format=self.target._format)
3364
f = RepoFetcher(to_repository=self.target,
3365
from_repository=self.source,
3366
last_revision=revision_id,
3367
fetch_spec=fetch_spec,
3368
find_ghosts=find_ghosts)
3369
3370
def _walk_to_common_revisions(self, revision_ids, if_present_ids=None):
3371
"""Walk out from revision_ids in source to revisions target has.
3372
3373
:param revision_ids: The start point for the search.
3374
:return: A set of revision ids.
3375
"""
3376
target_graph = self.target.get_graph()
3377
revision_ids = frozenset(revision_ids)
3378
if if_present_ids:
3379
all_wanted_revs = revision_ids.union(if_present_ids)
3380
else:
3381
all_wanted_revs = revision_ids
3382
missing_revs = set()
3383
source_graph = self.source.get_graph()
3384
# ensure we don't pay silly lookup costs.
3385
searcher = source_graph._make_breadth_first_searcher(all_wanted_revs)
3386
null_set = frozenset([_mod_revision.NULL_REVISION])
3387
searcher_exhausted = False
3388
while True:
3389
next_revs = set()
3390
ghosts = set()
3391
# Iterate the searcher until we have enough next_revs
3392
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3393
try:
3394
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3395
next_revs.update(next_revs_part)
3396
ghosts.update(ghosts_part)
3397
except StopIteration:
3398
searcher_exhausted = True
3399
break
3400
# If there are ghosts in the source graph, and the caller asked for
3401
# them, make sure that they are present in the target.
3402
# We don't care about other ghosts as we can't fetch them and
3403
# haven't been asked to.
3404
ghosts_to_check = set(revision_ids.intersection(ghosts))
3405
revs_to_get = set(next_revs).union(ghosts_to_check)
3406
if revs_to_get:
3407
have_revs = set(target_graph.get_parent_map(revs_to_get))
3408
# we always have NULL_REVISION present.
3409
have_revs = have_revs.union(null_set)
3410
# Check if the target is missing any ghosts we need.
3411
ghosts_to_check.difference_update(have_revs)
3412
if ghosts_to_check:
3413
# One of the caller's revision_ids is a ghost in both the
3414
# source and the target.
3415
raise errors.NoSuchRevision(
3416
self.source, ghosts_to_check.pop())
3417
missing_revs.update(next_revs - have_revs)
3418
# Because we may have walked past the original stop point, make
3419
# sure everything is stopped
3420
stop_revs = searcher.find_seen_ancestors(have_revs)
3421
searcher.stop_searching_any(stop_revs)
3422
if searcher_exhausted:
3423
break
3424
return searcher.get_result()
1726
3425
1727
3426
@needs_read_lock
1728
3427
def search_missing_revision_ids(self,
1729
3428
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
1730
find_ghosts=True, revision_ids=None, if_present_ids=None,
1731
limit=None):
3429
find_ghosts=True, revision_ids=None, if_present_ids=None):
1732
3430
"""Return the revision ids that source has that target does not.
1733
3431
1734
3432
:param revision_id: only return revision ids included by this
1742
3440
to fetch for tags, which may reference absent revisions.
1743
3441
:param find_ghosts: If True find missing revisions in deep history
1744
3442
rather than just finding the surface difference.
1745
:param limit: Maximum number of revisions to return, topologically
1746
ordered
1747
3443
:return: A bzrlib.graph.SearchResult.
1748
3444
"""
1749
raise NotImplementedError(self.search_missing_revision_ids)
3445
if symbol_versioning.deprecated_passed(revision_id):
3446
symbol_versioning.warn(
3447
'search_missing_revision_ids(revision_id=...) was '
3448
'deprecated in 2.4. Use revision_ids=[...] instead.',
3449
DeprecationWarning, stacklevel=2)
3450
if revision_ids is not None:
3451
raise AssertionError(
3452
'revision_ids is mutually exclusive with revision_id')
3453
if revision_id is not None:
3454
revision_ids = [revision_id]
3455
del revision_id
3456
# stop searching at found target revisions.
3457
if not find_ghosts and (revision_ids is not None or if_present_ids is
3458
not None):
3459
return self._walk_to_common_revisions(revision_ids,
3460
if_present_ids=if_present_ids)
3461
# generic, possibly worst case, slow code path.
3462
target_ids = set(self.target.all_revision_ids())
3463
source_ids = self._present_source_revisions_for(
3464
revision_ids, if_present_ids)
3465
result_set = set(source_ids).difference(target_ids)
3466
return self.source.revision_ids_to_search_result(result_set)
3467
3468
def _present_source_revisions_for(self, revision_ids, if_present_ids=None):
3469
"""Returns set of all revisions in ancestry of revision_ids present in
3470
the source repo.
3471
3472
:param revision_ids: if None, all revisions in source are returned.
3473
:param if_present_ids: like revision_ids, but if any/all of these are
3474
absent no error is raised.
3475
"""
3476
if revision_ids is not None or if_present_ids is not None:
3477
# First, ensure all specified revisions exist. Callers expect
3478
# NoSuchRevision when they pass absent revision_ids here.
3479
if revision_ids is None:
3480
revision_ids = set()
3481
if if_present_ids is None:
3482
if_present_ids = set()
3483
revision_ids = set(revision_ids)
3484
if_present_ids = set(if_present_ids)
3485
all_wanted_ids = revision_ids.union(if_present_ids)
3486
graph = self.source.get_graph()
3487
present_revs = set(graph.get_parent_map(all_wanted_ids))
3488
missing = revision_ids.difference(present_revs)
3489
if missing:
3490
raise errors.NoSuchRevision(self.source, missing.pop())
3491
found_ids = all_wanted_ids.intersection(present_revs)
3492
source_ids = [rev_id for (rev_id, parents) in
3493
graph.iter_ancestry(found_ids)
3494
if rev_id != _mod_revision.NULL_REVISION
3495
and parents is not None]
3496
else:
3497
source_ids = self.source.all_revision_ids()
3498
return set(source_ids)
1750
3499
1751
3500
@staticmethod
1752
3501
def _same_model(source, target):
1773
3522
"different serializers")
1774
3523
1775
3524
3525
class InterSameDataRepository(InterRepository):
3526
"""Code for converting between repositories that represent the same data.
3527
3528
Data format and model must match for this to work.
3529
"""
3530
3531
@classmethod
3532
def _get_repo_format_to_test(self):
3533
"""Repository format for testing with.
3534
3535
InterSameData can pull from subtree to subtree and from non-subtree to
3536
non-subtree, so we test this with the richest repository format.
3537
"""
3538
from bzrlib.repofmt import knitrepo
3539
return knitrepo.RepositoryFormatKnit3()
3540
3541
@staticmethod
3542
def is_compatible(source, target):
3543
return InterRepository._same_model(source, target)
3544
3545
3546
class InterDifferingSerializer(InterRepository):
3547
3548
@classmethod
3549
def _get_repo_format_to_test(self):
3550
return None
3551
3552
@staticmethod
3553
def is_compatible(source, target):
3554
"""Be compatible with Knit2 source and Knit3 target"""
3555
# This is redundant with format.check_conversion_target(), however that
3556
# raises an exception, and we just want to say "False" as in we won't
3557
# support converting between these formats.
3558
if 'IDS_never' in debug.debug_flags:
3559
return False
3560
if source.supports_rich_root() and not target.supports_rich_root():
3561
return False
3562
if (source._format.supports_tree_reference
3563
and not target._format.supports_tree_reference):
3564
return False
3565
if target._fallback_repositories and target._format.supports_chks:
3566
# IDS doesn't know how to copy CHKs for the parent inventories it
3567
# adds to stacked repos.
3568
return False
3569
if 'IDS_always' in debug.debug_flags:
3570
return True
3571
# Only use this code path for local source and target. IDS does far
3572
# too much IO (both bandwidth and roundtrips) over a network.
3573
if not source.bzrdir.transport.base.startswith('file:///'):
3574
return False
3575
if not target.bzrdir.transport.base.startswith('file:///'):
3576
return False
3577
return True
3578
3579
def _get_trees(self, revision_ids, cache):
3580
possible_trees = []
3581
for rev_id in revision_ids:
3582
if rev_id in cache:
3583
possible_trees.append((rev_id, cache[rev_id]))
3584
else:
3585
# Not cached, but inventory might be present anyway.
3586
try:
3587
tree = self.source.revision_tree(rev_id)
3588
except errors.NoSuchRevision:
3589
# Nope, parent is ghost.
3590
pass
3591
else:
3592
cache[rev_id] = tree
3593
possible_trees.append((rev_id, tree))
3594
return possible_trees
3595
3596
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3597
"""Get the best delta and base for this revision.
3598
3599
:return: (basis_id, delta)
3600
"""
3601
deltas = []
3602
# Generate deltas against each tree, to find the shortest.
3603
texts_possibly_new_in_tree = set()
3604
for basis_id, basis_tree in possible_trees:
3605
delta = tree.inventory._make_delta(basis_tree.inventory)
3606
for old_path, new_path, file_id, new_entry in delta:
3607
if new_path is None:
3608
# This file_id isn't present in the new rev, so we don't
3609
# care about it.
3610
continue
3611
if not new_path:
3612
# Rich roots are handled elsewhere...
3613
continue
3614
kind = new_entry.kind
3615
if kind != 'directory' and kind != 'file':
3616
# No text record associated with this inventory entry.
3617
continue
3618
# This is a directory or file that has changed somehow.
3619
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3620
deltas.append((len(delta), basis_id, delta))
3621
deltas.sort()
3622
return deltas[0][1:]
3623
3624
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3625
"""Find all parent revisions that are absent, but for which the
3626
inventory is present, and copy those inventories.
3627
3628
This is necessary to preserve correctness when the source is stacked
3629
without fallbacks configured. (Note that in cases like upgrade the
3630
source may be not have _fallback_repositories even though it is
3631
stacked.)
3632
"""
3633
parent_revs = set()
3634
for parents in parent_map.values():
3635
parent_revs.update(parents)
3636
present_parents = self.source.get_parent_map(parent_revs)
3637
absent_parents = set(parent_revs).difference(present_parents)
3638
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3639
(rev_id,) for rev_id in absent_parents)
3640
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3641
for parent_tree in self.source.revision_trees(parent_inv_ids):
3642
current_revision_id = parent_tree.get_revision_id()
3643
parents_parents_keys = parent_invs_keys_for_stacking[
3644
(current_revision_id,)]
3645
parents_parents = [key[-1] for key in parents_parents_keys]
3646
basis_id = _mod_revision.NULL_REVISION
3647
basis_tree = self.source.revision_tree(basis_id)
3648
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3649
self.target.add_inventory_by_delta(
3650
basis_id, delta, current_revision_id, parents_parents)
3651
cache[current_revision_id] = parent_tree
3652
3653
def _fetch_batch(self, revision_ids, basis_id, cache):
3654
"""Fetch across a few revisions.
3655
3656
:param revision_ids: The revisions to copy
3657
:param basis_id: The revision_id of a tree that must be in cache, used
3658
as a basis for delta when no other base is available
3659
:param cache: A cache of RevisionTrees that we can use.
3660
:return: The revision_id of the last converted tree. The RevisionTree
3661
for it will be in cache
3662
"""
3663
# Walk though all revisions; get inventory deltas, copy referenced
3664
# texts that delta references, insert the delta, revision and
3665
# signature.
3666
root_keys_to_create = set()
3667
text_keys = set()
3668
pending_deltas = []
3669
pending_revisions = []
3670
parent_map = self.source.get_parent_map(revision_ids)
3671
self._fetch_parent_invs_for_stacking(parent_map, cache)
3672
self.source._safe_to_return_from_cache = True
3673
for tree in self.source.revision_trees(revision_ids):
3674
# Find a inventory delta for this revision.
3675
# Find text entries that need to be copied, too.
3676
current_revision_id = tree.get_revision_id()
3677
parent_ids = parent_map.get(current_revision_id, ())
3678
parent_trees = self._get_trees(parent_ids, cache)
3679
possible_trees = list(parent_trees)
3680
if len(possible_trees) == 0:
3681
# There either aren't any parents, or the parents are ghosts,
3682
# so just use the last converted tree.
3683
possible_trees.append((basis_id, cache[basis_id]))
3684
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3685
possible_trees)
3686
revision = self.source.get_revision(current_revision_id)
3687
pending_deltas.append((basis_id, delta,
3688
current_revision_id, revision.parent_ids))
3689
if self._converting_to_rich_root:
3690
self._revision_id_to_root_id[current_revision_id] = \
3691
tree.get_root_id()
3692
# Determine which texts are in present in this revision but not in
3693
# any of the available parents.
3694
texts_possibly_new_in_tree = set()
3695
for old_path, new_path, file_id, entry in delta:
3696
if new_path is None:
3697
# This file_id isn't present in the new rev
3698
continue
3699
if not new_path:
3700
# This is the root
3701
if not self.target.supports_rich_root():
3702
# The target doesn't support rich root, so we don't
3703
# copy
3704
continue
3705
if self._converting_to_rich_root:
3706
# This can't be copied normally, we have to insert
3707
# it specially
3708
root_keys_to_create.add((file_id, entry.revision))
3709
continue
3710
kind = entry.kind
3711
texts_possibly_new_in_tree.add((file_id, entry.revision))
3712
for basis_id, basis_tree in possible_trees:
3713
basis_inv = basis_tree.inventory
3714
for file_key in list(texts_possibly_new_in_tree):
3715
file_id, file_revision = file_key
3716
try:
3717
entry = basis_inv[file_id]
3718
except errors.NoSuchId:
3719
continue
3720
if entry.revision == file_revision:
3721
texts_possibly_new_in_tree.remove(file_key)
3722
text_keys.update(texts_possibly_new_in_tree)
3723
pending_revisions.append(revision)
3724
cache[current_revision_id] = tree
3725
basis_id = current_revision_id
3726
self.source._safe_to_return_from_cache = False
3727
# Copy file texts
3728
from_texts = self.source.texts
3729
to_texts = self.target.texts
3730
if root_keys_to_create:
3731
root_stream = _mod_fetch._new_root_data_stream(
3732
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3733
self.source)
3734
to_texts.insert_record_stream(root_stream)
3735
to_texts.insert_record_stream(from_texts.get_record_stream(
3736
text_keys, self.target._format._fetch_order,
3737
not self.target._format._fetch_uses_deltas))
3738
# insert inventory deltas
3739
for delta in pending_deltas:
3740
self.target.add_inventory_by_delta(*delta)
3741
if self.target._fallback_repositories:
3742
# Make sure this stacked repository has all the parent inventories
3743
# for the new revisions that we are about to insert. We do this
3744
# before adding the revisions so that no revision is added until
3745
# all the inventories it may depend on are added.
3746
# Note that this is overzealous, as we may have fetched these in an
3747
# earlier batch.
3748
parent_ids = set()
3749
revision_ids = set()
3750
for revision in pending_revisions:
3751
revision_ids.add(revision.revision_id)
3752
parent_ids.update(revision.parent_ids)
3753
parent_ids.difference_update(revision_ids)
3754
parent_ids.discard(_mod_revision.NULL_REVISION)
3755
parent_map = self.source.get_parent_map(parent_ids)
3756
# we iterate over parent_map and not parent_ids because we don't
3757
# want to try copying any revision which is a ghost
3758
for parent_tree in self.source.revision_trees(parent_map):
3759
current_revision_id = parent_tree.get_revision_id()
3760
parents_parents = parent_map[current_revision_id]
3761
possible_trees = self._get_trees(parents_parents, cache)
3762
if len(possible_trees) == 0:
3763
# There either aren't any parents, or the parents are
3764
# ghosts, so just use the last converted tree.
3765
possible_trees.append((basis_id, cache[basis_id]))
3766
basis_id, delta = self._get_delta_for_revision(parent_tree,
3767
parents_parents, possible_trees)
3768
self.target.add_inventory_by_delta(
3769
basis_id, delta, current_revision_id, parents_parents)
3770
# insert signatures and revisions
3771
for revision in pending_revisions:
3772
try:
3773
signature = self.source.get_signature_text(
3774
revision.revision_id)
3775
self.target.add_signature_text(revision.revision_id,
3776
signature)
3777
except errors.NoSuchRevision:
3778
pass
3779
self.target.add_revision(revision.revision_id, revision)
3780
return basis_id
3781
3782
def _fetch_all_revisions(self, revision_ids, pb):
3783
"""Fetch everything for the list of revisions.
3784
3785
:param revision_ids: The list of revisions to fetch. Must be in
3786
topological order.
3787
:param pb: A ProgressTask
3788
:return: None
3789
"""
3790
basis_id, basis_tree = self._get_basis(revision_ids[0])
3791
batch_size = 100
3792
cache = lru_cache.LRUCache(100)
3793
cache[basis_id] = basis_tree
3794
del basis_tree # We don't want to hang on to it here
3795
hints = []
3796
a_graph = None
3797
3798
for offset in range(0, len(revision_ids), batch_size):
3799
self.target.start_write_group()
3800
try:
3801
pb.update('Transferring revisions', offset,
3802
len(revision_ids))
3803
batch = revision_ids[offset:offset+batch_size]
3804
basis_id = self._fetch_batch(batch, basis_id, cache)
3805
except:
3806
self.source._safe_to_return_from_cache = False
3807
self.target.abort_write_group()
3808
raise
3809
else:
3810
hint = self.target.commit_write_group()
3811
if hint:
3812
hints.extend(hint)
3813
if hints and self.target._format.pack_compresses:
3814
self.target.pack(hint=hints)
3815
pb.update('Transferring revisions', len(revision_ids),
3816
len(revision_ids))
3817
3818
@needs_write_lock
3819
def fetch(self, revision_id=None, find_ghosts=False,
3820
fetch_spec=None):
3821
"""See InterRepository.fetch()."""
3822
if fetch_spec is not None:
3823
revision_ids = fetch_spec.get_keys()
3824
else:
3825
revision_ids = None
3826
ui.ui_factory.warn_experimental_format_fetch(self)
3827
if (not self.source.supports_rich_root()
3828
and self.target.supports_rich_root()):
3829
self._converting_to_rich_root = True
3830
self._revision_id_to_root_id = {}
3831
else:
3832
self._converting_to_rich_root = False
3833
# See <https://launchpad.net/bugs/456077> asking for a warning here
3834
if self.source._format.network_name() != self.target._format.network_name():
3835
ui.ui_factory.show_user_warning('cross_format_fetch',
3836
from_format=self.source._format,
3837
to_format=self.target._format)
3838
if revision_ids is None:
3839
if revision_id:
3840
search_revision_ids = [revision_id]
3841
else:
3842
search_revision_ids = None
3843
revision_ids = self.target.search_missing_revision_ids(self.source,
3844
revision_ids=search_revision_ids,
3845
find_ghosts=find_ghosts).get_keys()
3846
if not revision_ids:
3847
return 0, 0
3848
revision_ids = tsort.topo_sort(
3849
self.source.get_graph().get_parent_map(revision_ids))
3850
if not revision_ids:
3851
return 0, 0
3852
# Walk though all revisions; get inventory deltas, copy referenced
3853
# texts that delta references, insert the delta, revision and
3854
# signature.
3855
pb = ui.ui_factory.nested_progress_bar()
3856
try:
3857
self._fetch_all_revisions(revision_ids, pb)
3858
finally:
3859
pb.finished()
3860
return len(revision_ids), 0
3861
3862
def _get_basis(self, first_revision_id):
3863
"""Get a revision and tree which exists in the target.
3864
3865
This assumes that first_revision_id is selected for transmission
3866
because all other ancestors are already present. If we can't find an
3867
ancestor we fall back to NULL_REVISION since we know that is safe.
3868
3869
:return: (basis_id, basis_tree)
3870
"""
3871
first_rev = self.source.get_revision(first_revision_id)
3872
try:
3873
basis_id = first_rev.parent_ids[0]
3874
# only valid as a basis if the target has it
3875
self.target.get_revision(basis_id)
3876
# Try to get a basis tree - if it's a ghost it will hit the
3877
# NoSuchRevision case.
3878
basis_tree = self.source.revision_tree(basis_id)
3879
except (IndexError, errors.NoSuchRevision):
3880
basis_id = _mod_revision.NULL_REVISION
3881
basis_tree = self.source.revision_tree(basis_id)
3882
return basis_id, basis_tree
3883
3884
3885
InterRepository.register_optimiser(InterDifferingSerializer)
3886
InterRepository.register_optimiser(InterSameDataRepository)
3887
3888
1776
3889
class CopyConverter(object):
1777
3890
"""A repository conversion tool which just performs a copy of the content.
1778
3891
1821
3934
pb.finished()
1822
3935
1823
3936
3937
class _VersionedFileChecker(object):
3938
3939
def __init__(self, repository, text_key_references=None, ancestors=None):
3940
self.repository = repository
3941
self.text_index = self.repository._generate_text_key_index(
3942
text_key_references=text_key_references, ancestors=ancestors)
3943
3944
def calculate_file_version_parents(self, text_key):
3945
"""Calculate the correct parents for a file version according to
3946
the inventories.
3947
"""
3948
parent_keys = self.text_index[text_key]
3949
if parent_keys == [_mod_revision.NULL_REVISION]:
3950
return ()
3951
return tuple(parent_keys)
3952
3953
def check_file_version_parents(self, texts, progress_bar=None):
3954
"""Check the parents stored in a versioned file are correct.
3955
3956
It also detects file versions that are not referenced by their
3957
corresponding revision's inventory.
3958
3959
:returns: A tuple of (wrong_parents, dangling_file_versions).
3960
wrong_parents is a dict mapping {revision_id: (stored_parents,
3961
correct_parents)} for each revision_id where the stored parents
3962
are not correct. dangling_file_versions is a set of (file_id,
3963
revision_id) tuples for versions that are present in this versioned
3964
file, but not used by the corresponding inventory.
3965
"""
3966
local_progress = None
3967
if progress_bar is None:
3968
local_progress = ui.ui_factory.nested_progress_bar()
3969
progress_bar = local_progress
3970
try:
3971
return self._check_file_version_parents(texts, progress_bar)
3972
finally:
3973
if local_progress:
3974
local_progress.finished()
3975
3976
def _check_file_version_parents(self, texts, progress_bar):
3977
"""See check_file_version_parents."""
3978
wrong_parents = {}
3979
self.file_ids = set([file_id for file_id, _ in
3980
self.text_index.iterkeys()])
3981
# text keys is now grouped by file_id
3982
n_versions = len(self.text_index)
3983
progress_bar.update('loading text store', 0, n_versions)
3984
parent_map = self.repository.texts.get_parent_map(self.text_index)
3985
# On unlistable transports this could well be empty/error...
3986
text_keys = self.repository.texts.keys()
3987
unused_keys = frozenset(text_keys) - set(self.text_index)
3988
for num, key in enumerate(self.text_index.iterkeys()):
3989
progress_bar.update('checking text graph', num, n_versions)
3990
correct_parents = self.calculate_file_version_parents(key)
3991
try:
3992
knit_parents = parent_map[key]
3993
except errors.RevisionNotPresent:
3994
# Missing text!
3995
knit_parents = None
3996
if correct_parents != knit_parents:
3997
wrong_parents[key] = (knit_parents, correct_parents)
3998
return wrong_parents, unused_keys
3999
4000
1824
4001
def _strip_NULL_ghosts(revision_graph):
1825
4002
"""Also don't use this. more compatibility code for unmigrated clients."""
1826
4003
# Filter ghosts, and null:
1832
4009
return revision_graph
1833
4010
1834
4011
4012
class StreamSink(object):
4013
"""An object that can insert a stream into a repository.
4014
4015
This interface handles the complexity of reserialising inventories and
4016
revisions from different formats, and allows unidirectional insertion into
4017
stacked repositories without looking for the missing basis parents
4018
beforehand.
4019
"""
4020
4021
def __init__(self, target_repo):
4022
self.target_repo = target_repo
4023
4024
def insert_stream(self, stream, src_format, resume_tokens):
4025
"""Insert a stream's content into the target repository.
4026
4027
:param src_format: a bzr repository format.
4028
4029
:return: a list of resume tokens and an iterable of keys additional
4030
items required before the insertion can be completed.
4031
"""
4032
self.target_repo.lock_write()
4033
try:
4034
if resume_tokens:
4035
self.target_repo.resume_write_group(resume_tokens)
4036
is_resume = True
4037
else:
4038
self.target_repo.start_write_group()
4039
is_resume = False
4040
try:
4041
# locked_insert_stream performs a commit|suspend.
4042
missing_keys = self.insert_stream_without_locking(stream,
4043
src_format, is_resume)
4044
if missing_keys:
4045
# suspend the write group and tell the caller what we is
4046
# missing. We know we can suspend or else we would not have
4047
# entered this code path. (All repositories that can handle
4048
# missing keys can handle suspending a write group).
4049
write_group_tokens = self.target_repo.suspend_write_group()
4050
return write_group_tokens, missing_keys
4051
hint = self.target_repo.commit_write_group()
4052
to_serializer = self.target_repo._format._serializer
4053
src_serializer = src_format._serializer
4054
if (to_serializer != src_serializer and
4055
self.target_repo._format.pack_compresses):
4056
self.target_repo.pack(hint=hint)
4057
return [], set()
4058
except:
4059
self.target_repo.abort_write_group(suppress_errors=True)
4060
raise
4061
finally:
4062
self.target_repo.unlock()
4063
4064
def insert_stream_without_locking(self, stream, src_format,
4065
is_resume=False):
4066
"""Insert a stream's content into the target repository.
4067
4068
This assumes that you already have a locked repository and an active
4069
write group.
4070
4071
:param src_format: a bzr repository format.
4072
:param is_resume: Passed down to get_missing_parent_inventories to
4073
indicate if we should be checking for missing texts at the same
4074
time.
4075
4076
:return: A set of keys that are missing.
4077
"""
4078
if not self.target_repo.is_write_locked():
4079
raise errors.ObjectNotLocked(self)
4080
if not self.target_repo.is_in_write_group():
4081
raise errors.BzrError('you must already be in a write group')
4082
to_serializer = self.target_repo._format._serializer
4083
src_serializer = src_format._serializer
4084
new_pack = None
4085
if to_serializer == src_serializer:
4086
# If serializers match and the target is a pack repository, set the
4087
# write cache size on the new pack. This avoids poor performance
4088
# on transports where append is unbuffered (such as
4089
# RemoteTransport). This is safe to do because nothing should read
4090
# back from the target repository while a stream with matching
4091
# serialization is being inserted.
4092
# The exception is that a delta record from the source that should
4093
# be a fulltext may need to be expanded by the target (see
4094
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4095
# explicitly flush any buffered writes first in that rare case.
4096
try:
4097
new_pack = self.target_repo._pack_collection._new_pack
4098
except AttributeError:
4099
# Not a pack repository
4100
pass
4101
else:
4102
new_pack.set_write_cache_size(1024*1024)
4103
for substream_type, substream in stream:
4104
if 'stream' in debug.debug_flags:
4105
mutter('inserting substream: %s', substream_type)
4106
if substream_type == 'texts':
4107
self.target_repo.texts.insert_record_stream(substream)
4108
elif substream_type == 'inventories':
4109
if src_serializer == to_serializer:
4110
self.target_repo.inventories.insert_record_stream(
4111
substream)
4112
else:
4113
self._extract_and_insert_inventories(
4114
substream, src_serializer)
4115
elif substream_type == 'inventory-deltas':
4116
self._extract_and_insert_inventory_deltas(
4117
substream, src_serializer)
4118
elif substream_type == 'chk_bytes':
4119
# XXX: This doesn't support conversions, as it assumes the
4120
# conversion was done in the fetch code.
4121
self.target_repo.chk_bytes.insert_record_stream(substream)
4122
elif substream_type == 'revisions':
4123
# This may fallback to extract-and-insert more often than
4124
# required if the serializers are different only in terms of
4125
# the inventory.
4126
if src_serializer == to_serializer:
4127
self.target_repo.revisions.insert_record_stream(substream)
4128
else:
4129
self._extract_and_insert_revisions(substream,
4130
src_serializer)
4131
elif substream_type == 'signatures':
4132
self.target_repo.signatures.insert_record_stream(substream)
4133
else:
4134
raise AssertionError('kaboom! %s' % (substream_type,))
4135
# Done inserting data, and the missing_keys calculations will try to
4136
# read back from the inserted data, so flush the writes to the new pack
4137
# (if this is pack format).
4138
if new_pack is not None:
4139
new_pack._write_data('', flush=True)
4140
# Find all the new revisions (including ones from resume_tokens)
4141
missing_keys = self.target_repo.get_missing_parent_inventories(
4142
check_for_missing_texts=is_resume)
4143
try:
4144
for prefix, versioned_file in (
4145
('texts', self.target_repo.texts),
4146
('inventories', self.target_repo.inventories),
4147
('revisions', self.target_repo.revisions),
4148
('signatures', self.target_repo.signatures),
4149
('chk_bytes', self.target_repo.chk_bytes),
4150
):
4151
if versioned_file is None:
4152
continue
4153
# TODO: key is often going to be a StaticTuple object
4154
# I don't believe we can define a method by which
4155
# (prefix,) + StaticTuple will work, though we could
4156
# define a StaticTuple.sq_concat that would allow you to
4157
# pass in either a tuple or a StaticTuple as the second
4158
# object, so instead we could have:
4159
# StaticTuple(prefix) + key here...
4160
missing_keys.update((prefix,) + key for key in
4161
versioned_file.get_missing_compression_parent_keys())
4162
except NotImplementedError:
4163
# cannot even attempt suspending, and missing would have failed
4164
# during stream insertion.
4165
missing_keys = set()
4166
return missing_keys
4167
4168
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4169
target_rich_root = self.target_repo._format.rich_root_data
4170
target_tree_refs = self.target_repo._format.supports_tree_reference
4171
for record in substream:
4172
# Insert the delta directly
4173
inventory_delta_bytes = record.get_bytes_as('fulltext')
4174
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4175
try:
4176
parse_result = deserialiser.parse_text_bytes(
4177
inventory_delta_bytes)
4178
except inventory_delta.IncompatibleInventoryDelta, err:
4179
mutter("Incompatible delta: %s", err.msg)
4180
raise errors.IncompatibleRevision(self.target_repo._format)
4181
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4182
revision_id = new_id
4183
parents = [key[0] for key in record.parents]
4184
self.target_repo.add_inventory_by_delta(
4185
basis_id, inv_delta, revision_id, parents)
4186
4187
def _extract_and_insert_inventories(self, substream, serializer,
4188
parse_delta=None):
4189
"""Generate a new inventory versionedfile in target, converting data.
4190
4191
The inventory is retrieved from the source, (deserializing it), and
4192
stored in the target (reserializing it in a different format).
4193
"""
4194
target_rich_root = self.target_repo._format.rich_root_data
4195
target_tree_refs = self.target_repo._format.supports_tree_reference
4196
for record in substream:
4197
# It's not a delta, so it must be a fulltext in the source
4198
# serializer's format.
4199
bytes = record.get_bytes_as('fulltext')
4200
revision_id = record.key[0]
4201
inv = serializer.read_inventory_from_string(bytes, revision_id)
4202
parents = [key[0] for key in record.parents]
4203
self.target_repo.add_inventory(revision_id, inv, parents)
4204
# No need to keep holding this full inv in memory when the rest of
4205
# the substream is likely to be all deltas.
4206
del inv
4207
4208
def _extract_and_insert_revisions(self, substream, serializer):
4209
for record in substream:
4210
bytes = record.get_bytes_as('fulltext')
4211
revision_id = record.key[0]
4212
rev = serializer.read_revision_from_string(bytes)
4213
if rev.revision_id != revision_id:
4214
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4215
self.target_repo.add_revision(revision_id, rev)
4216
4217
def finished(self):
4218
if self.target_repo._format._fetch_reconcile:
4219
self.target_repo.reconcile()
4220
4221
4222
class StreamSource(object):
4223
"""A source of a stream for fetching between repositories."""
4224
4225
def __init__(self, from_repository, to_format):
4226
"""Create a StreamSource streaming from from_repository."""
4227
self.from_repository = from_repository
4228
self.to_format = to_format
4229
self._record_counter = RecordCounter()
4230
4231
def delta_on_metadata(self):
4232
"""Return True if delta's are permitted on metadata streams.
4233
4234
That is on revisions and signatures.
4235
"""
4236
src_serializer = self.from_repository._format._serializer
4237
target_serializer = self.to_format._serializer
4238
return (self.to_format._fetch_uses_deltas and
4239
src_serializer == target_serializer)
4240
4241
def _fetch_revision_texts(self, revs):
4242
# fetch signatures first and then the revision texts
4243
# may need to be a InterRevisionStore call here.
4244
from_sf = self.from_repository.signatures
4245
# A missing signature is just skipped.
4246
keys = [(rev_id,) for rev_id in revs]
4247
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4248
keys,
4249
self.to_format._fetch_order,
4250
not self.to_format._fetch_uses_deltas))
4251
# If a revision has a delta, this is actually expanded inside the
4252
# insert_record_stream code now, which is an alternate fix for
4253
# bug #261339
4254
from_rf = self.from_repository.revisions
4255
revisions = from_rf.get_record_stream(
4256
keys,
4257
self.to_format._fetch_order,
4258
not self.delta_on_metadata())
4259
return [('signatures', signatures), ('revisions', revisions)]
4260
4261
def _generate_root_texts(self, revs):
4262
"""This will be called by get_stream between fetching weave texts and
4263
fetching the inventory weave.
4264
"""
4265
if self._rich_root_upgrade():
4266
return _mod_fetch.Inter1and2Helper(
4267
self.from_repository).generate_root_texts(revs)
4268
else:
4269
return []
4270
4271
def get_stream(self, search):
4272
phase = 'file'
4273
revs = search.get_keys()
4274
graph = self.from_repository.get_graph()
4275
revs = tsort.topo_sort(graph.get_parent_map(revs))
4276
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4277
text_keys = []
4278
for knit_kind, file_id, revisions in data_to_fetch:
4279
if knit_kind != phase:
4280
phase = knit_kind
4281
# Make a new progress bar for this phase
4282
if knit_kind == "file":
4283
# Accumulate file texts
4284
text_keys.extend([(file_id, revision) for revision in
4285
revisions])
4286
elif knit_kind == "inventory":
4287
# Now copy the file texts.
4288
from_texts = self.from_repository.texts
4289
yield ('texts', from_texts.get_record_stream(
4290
text_keys, self.to_format._fetch_order,
4291
not self.to_format._fetch_uses_deltas))
4292
# Cause an error if a text occurs after we have done the
4293
# copy.
4294
text_keys = None
4295
# Before we process the inventory we generate the root
4296
# texts (if necessary) so that the inventories references
4297
# will be valid.
4298
for _ in self._generate_root_texts(revs):
4299
yield _
4300
# we fetch only the referenced inventories because we do not
4301
# know for unselected inventories whether all their required
4302
# texts are present in the other repository - it could be
4303
# corrupt.
4304
for info in self._get_inventory_stream(revs):
4305
yield info
4306
elif knit_kind == "signatures":
4307
# Nothing to do here; this will be taken care of when
4308
# _fetch_revision_texts happens.
4309
pass
4310
elif knit_kind == "revisions":
4311
for record in self._fetch_revision_texts(revs):
4312
yield record
4313
else:
4314
raise AssertionError("Unknown knit kind %r" % knit_kind)
4315
4316
def get_stream_for_missing_keys(self, missing_keys):
4317
# missing keys can only occur when we are byte copying and not
4318
# translating (because translation means we don't send
4319
# unreconstructable deltas ever).
4320
keys = {}
4321
keys['texts'] = set()
4322
keys['revisions'] = set()
4323
keys['inventories'] = set()
4324
keys['chk_bytes'] = set()
4325
keys['signatures'] = set()
4326
for key in missing_keys:
4327
keys[key[0]].add(key[1:])
4328
if len(keys['revisions']):
4329
# If we allowed copying revisions at this point, we could end up
4330
# copying a revision without copying its required texts: a
4331
# violation of the requirements for repository integrity.
4332
raise AssertionError(
4333
'cannot copy revisions to fill in missing deltas %s' % (
4334
keys['revisions'],))
4335
for substream_kind, keys in keys.iteritems():
4336
vf = getattr(self.from_repository, substream_kind)
4337
if vf is None and keys:
4338
raise AssertionError(
4339
"cannot fill in keys for a versioned file we don't"
4340
" have: %s needs %s" % (substream_kind, keys))
4341
if not keys:
4342
# No need to stream something we don't have
4343
continue
4344
if substream_kind == 'inventories':
4345
# Some missing keys are genuinely ghosts, filter those out.
4346
present = self.from_repository.inventories.get_parent_map(keys)
4347
revs = [key[0] for key in present]
4348
# Get the inventory stream more-or-less as we do for the
4349
# original stream; there's no reason to assume that records
4350
# direct from the source will be suitable for the sink. (Think
4351
# e.g. 2a -> 1.9-rich-root).
4352
for info in self._get_inventory_stream(revs, missing=True):
4353
yield info
4354
continue
4355
4356
# Ask for full texts always so that we don't need more round trips
4357
# after this stream.
4358
# Some of the missing keys are genuinely ghosts, so filter absent
4359
# records. The Sink is responsible for doing another check to
4360
# ensure that ghosts don't introduce missing data for future
4361
# fetches.
4362
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4363
self.to_format._fetch_order, True))
4364
yield substream_kind, stream
4365
4366
def inventory_fetch_order(self):
4367
if self._rich_root_upgrade():
4368
return 'topological'
4369
else:
4370
return self.to_format._fetch_order
4371
4372
def _rich_root_upgrade(self):
4373
return (not self.from_repository._format.rich_root_data and
4374
self.to_format.rich_root_data)
4375
4376
def _get_inventory_stream(self, revision_ids, missing=False):
4377
from_format = self.from_repository._format
4378
if (from_format.supports_chks and self.to_format.supports_chks and
4379
from_format.network_name() == self.to_format.network_name()):
4380
raise AssertionError(
4381
"this case should be handled by GroupCHKStreamSource")
4382
elif 'forceinvdeltas' in debug.debug_flags:
4383
return self._get_convertable_inventory_stream(revision_ids,
4384
delta_versus_null=missing)
4385
elif from_format.network_name() == self.to_format.network_name():
4386
# Same format.
4387
return self._get_simple_inventory_stream(revision_ids,
4388
missing=missing)
4389
elif (not from_format.supports_chks and not self.to_format.supports_chks
4390
and from_format._serializer == self.to_format._serializer):
4391
# Essentially the same format.
4392
return self._get_simple_inventory_stream(revision_ids,
4393
missing=missing)
4394
else:
4395
# Any time we switch serializations, we want to use an
4396
# inventory-delta based approach.
4397
return self._get_convertable_inventory_stream(revision_ids,
4398
delta_versus_null=missing)
4399
4400
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4401
# NB: This currently reopens the inventory weave in source;
4402
# using a single stream interface instead would avoid this.
4403
from_weave = self.from_repository.inventories
4404
if missing:
4405
delta_closure = True
4406
else:
4407
delta_closure = not self.delta_on_metadata()
4408
yield ('inventories', from_weave.get_record_stream(
4409
[(rev_id,) for rev_id in revision_ids],
4410
self.inventory_fetch_order(), delta_closure))
4411
4412
def _get_convertable_inventory_stream(self, revision_ids,
4413
delta_versus_null=False):
4414
# The two formats are sufficiently different that there is no fast
4415
# path, so we need to send just inventorydeltas, which any
4416
# sufficiently modern client can insert into any repository.
4417
# The StreamSink code expects to be able to
4418
# convert on the target, so we need to put bytes-on-the-wire that can
4419
# be converted. That means inventory deltas (if the remote is <1.19,
4420
# RemoteStreamSink will fallback to VFS to insert the deltas).
4421
yield ('inventory-deltas',
4422
self._stream_invs_as_deltas(revision_ids,
4423
delta_versus_null=delta_versus_null))
4424
4425
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4426
"""Return a stream of inventory-deltas for the given rev ids.
4427
4428
:param revision_ids: The list of inventories to transmit
4429
:param delta_versus_null: Don't try to find a minimal delta for this
4430
entry, instead compute the delta versus the NULL_REVISION. This
4431
effectively streams a complete inventory. Used for stuff like
4432
filling in missing parents, etc.
4433
"""
4434
from_repo = self.from_repository
4435
revision_keys = [(rev_id,) for rev_id in revision_ids]
4436
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4437
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4438
# method...
4439
inventories = self.from_repository.iter_inventories(
4440
revision_ids, 'topological')
4441
format = from_repo._format
4442
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4443
inventory_cache = lru_cache.LRUCache(50)
4444
null_inventory = from_repo.revision_tree(
4445
_mod_revision.NULL_REVISION).inventory
4446
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4447
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4448
# repo back into a non-rich-root repo ought to be allowed)
4449
serializer = inventory_delta.InventoryDeltaSerializer(
4450
versioned_root=format.rich_root_data,
4451
tree_references=format.supports_tree_reference)
4452
for inv in inventories:
4453
key = (inv.revision_id,)
4454
parent_keys = parent_map.get(key, ())
4455
delta = None
4456
if not delta_versus_null and parent_keys:
4457
# The caller did not ask for complete inventories and we have
4458
# some parents that we can delta against. Make a delta against
4459
# each parent so that we can find the smallest.
4460
parent_ids = [parent_key[0] for parent_key in parent_keys]
4461
for parent_id in parent_ids:
4462
if parent_id not in invs_sent_so_far:
4463
# We don't know that the remote side has this basis, so
4464
# we can't use it.
4465
continue
4466
if parent_id == _mod_revision.NULL_REVISION:
4467
parent_inv = null_inventory
4468
else:
4469
parent_inv = inventory_cache.get(parent_id, None)
4470
if parent_inv is None:
4471
parent_inv = from_repo.get_inventory(parent_id)
4472
candidate_delta = inv._make_delta(parent_inv)
4473
if (delta is None or
4474
len(delta) > len(candidate_delta)):
4475
delta = candidate_delta
4476
basis_id = parent_id
4477
if delta is None:
4478
# Either none of the parents ended up being suitable, or we
4479
# were asked to delta against NULL
4480
basis_id = _mod_revision.NULL_REVISION
4481
delta = inv._make_delta(null_inventory)
4482
invs_sent_so_far.add(inv.revision_id)
4483
inventory_cache[inv.revision_id] = inv
4484
delta_serialized = ''.join(
4485
serializer.delta_to_lines(basis_id, key[-1], delta))
4486
yield versionedfile.FulltextContentFactory(
4487
key, parent_keys, None, delta_serialized)
4488
4489
1835
4490
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
1836
4491
stop_revision=None):
1837
4492
"""Extend the partial history to include a given index
1847
4502
it is encountered, history extension will stop.
1848
4503
"""
1849
4504
start_revision = partial_history_cache[-1]
1850
graph = repo.get_graph()
1851
iterator = graph.iter_lefthand_ancestry(start_revision,
1852
(_mod_revision.NULL_REVISION,))
4505
iterator = repo.iter_reverse_revision_history(start_revision)
1853
4506
try:
1854
# skip the last revision in the list
4507
#skip the last revision in the list
1855
4508
iterator.next()
1856
4509
while True:
1857
4510
if (stop_index is not None and
1864
4517
except StopIteration:
1865
4518
# No more history
1866
4519
return
1867
1868
1869
class _LazyListJoin(object):
1870
"""An iterable yielding the contents of many lists as one list.
1871
1872
Each iterator made from this will reflect the current contents of the lists
1873
at the time the iterator is made.
1874
1875
This is used by Repository's _make_parents_provider implementation so that
1876
it is safe to do::
1877
1878
pp = repo._make_parents_provider() # uses a list of fallback repos
1879
pp.add_fallback_repository(other_repo) # appends to that list
1880
result = pp.get_parent_map(...)
1881
# The result will include revs from other_repo
1882
"""
1883
1884
def __init__(self, *list_parts):
1885
self.list_parts = list_parts
1886
1887
def __iter__(self):
1888
full_list = []
1889
for list_part in self.list_parts:
1890
full_list.extend(list_part)
1891
return iter(full_list)
Loggerhead is a web-based interface for Breezy
Version: 2.0.1