« back to all changes in this revision
Viewing changes to bzrlib/repository.py
- browse files at revision 5691
- compare with another revision
- download diff
- download tarball
- view history from revision 5691
- Committer: Canonical.com Patch Queue Manager
- Date: 2011-02-28 21:14:27 UTC
- mfrom: (5689.1.1 format_gc_chk_sha1_record_725051)
- Revision ID: pqm@pqm.ubuntu.com-20110228211427-bwkkhj6817rkk9ym
(gz) Fix test failure on alpha by correcting format string for
gc_chk_sha1_record (Martin [gz])
gc_chk_sha1_record (Martin [gz])
- files added:
- bzrlib/deprecated_graph.py
- files removed:
- bzrlib/estimate_compressed_size.py
- bzrlib/plugins/changelog_merge
- bzrlib/plugins/changelog_merge/tests
- bzrlib/plugins/weave_fmt
- bzrlib/plugins/weave_fmt/tests
- bzrlib/tests/per_repository_vf
- po
- files renamed:
- bzrlib/plugins/weave_fmt/repository.py => bzrlib/repofmt/weaverepo.py
- bzrlib/commit_signature_commands.py => bzrlib/sign_my_commits.py
- bzrlib/tests/per_repository_vf/helpers.py => bzrlib/tests/per_repository/helpers.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_check_reconcile.py => bzrlib/tests/per_repository/test_check_reconcile.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
- bzrlib/tests/per_repository_vf/test_merge_directive.py => bzrlib/tests/per_repository/test_merge_directive.py
- bzrlib/plugins/weave_fmt/xml4.py => bzrlib/xml4.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
19
import cStringIO
20
import re
20
21
import time
21
22
22
23
from bzrlib import (
23
24
bzrdir,
25
check,
26
chk_map,
24
27
config,
25
28
controldir,
26
29
debug,
30
fetch as _mod_fetch,
31
fifo_cache,
27
32
generate_ids,
33
gpg,
28
34
graph,
35
inventory,
36
inventory_delta,
29
37
lockable_files,
30
38
lockdir,
39
lru_cache,
31
40
osutils,
41
pyutils,
32
42
revision as _mod_revision,
33
testament as _mod_testament,
43
static_tuple,
44
trace,
34
45
tsort,
35
gpg,
46
versionedfile,
36
47
)
37
48
from bzrlib.bundle import serializer
38
from bzrlib.i18n import gettext
49
from bzrlib.revisiontree import RevisionTree
50
from bzrlib.store.versioned import VersionedFileStore
51
from bzrlib.testament import Testament
39
52
""")
40
53
41
54
from bzrlib import (
46
59
)
47
60
from bzrlib.decorators import needs_read_lock, needs_write_lock, only_raises
48
61
from bzrlib.inter import InterObject
62
from bzrlib.inventory import (
63
Inventory,
64
InventoryDirectory,
65
ROOT_ID,
66
entry_factory,
67
)
68
from bzrlib.recordcounter import RecordCounter
49
69
from bzrlib.lock import _RelockDebugMixin, LogicalLockResult
50
70
from bzrlib.trace import (
51
71
log_exception_quietly, note, mutter, mutter_callsite, warning)
72
92
73
93
# all clients should supply tree roots.
74
94
record_root_entry = True
75
# whether this commit builder supports the record_entry_contents interface
76
supports_record_entry_contents = False
95
# the default CommitBuilder does not manage trees whose root is versioned.
96
_versioned_root = False
77
97
78
98
def __init__(self, repository, parents, config, timestamp=None,
79
99
timezone=None, committer=None, revprops=None,
80
revision_id=None, lossy=False):
100
revision_id=None):
81
101
"""Initiate a CommitBuilder.
82
102
83
103
:param repository: Repository to commit to.
84
104
:param parents: Revision ids of the parents of the new revision.
105
:param config: Configuration to use.
85
106
:param timestamp: Optional timestamp recorded for commit.
86
107
:param timezone: Optional timezone for timestamp.
87
108
:param committer: Optional committer to set for commit.
88
109
:param revprops: Optional dictionary of revision properties.
89
110
:param revision_id: Optional revision id.
90
:param lossy: Whether to discard data that can not be natively
91
represented, when pushing to a foreign VCS
92
111
"""
93
112
self._config = config
94
self._lossy = lossy
95
113
96
114
if committer is None:
97
115
self._committer = self._config.username()
100
118
else:
101
119
self._committer = committer
102
120
121
self.new_inventory = Inventory(None)
103
122
self._new_revision_id = revision_id
104
123
self.parents = parents
105
124
self.repository = repository
120
139
self._timezone = int(timezone)
121
140
122
141
self._generate_revision_if_needed()
142
self.__heads = graph.HeadsCache(repository.get_graph()).heads
143
self._basis_delta = []
144
# API compatibility, older code that used CommitBuilder did not call
145
# .record_delete(), which means the delta that is computed would not be
146
# valid. Callers that will call record_delete() should call
147
# .will_record_deletes() to indicate that.
148
self._recording_deletes = False
149
# memo'd check for no-op commits.
150
self._any_changes = False
123
151
124
152
def any_changes(self):
125
153
"""Return True if any entries were changed.
126
154
127
155
This includes merge-only changes. It is the core for the --unchanged
128
156
detection in commit.
129
157
130
158
:return: True if any changes have occured.
131
159
"""
132
raise NotImplementedError(self.any_changes)
160
return self._any_changes
133
161
134
162
def _validate_unicode_text(self, text, context):
135
163
"""Verify things like commit messages don't have bogus characters."""
146
174
self._validate_unicode_text(value,
147
175
'revision property (%s)' % (key,))
148
176
177
def _ensure_fallback_inventories(self):
178
"""Ensure that appropriate inventories are available.
179
180
This only applies to repositories that are stacked, and is about
181
enusring the stacking invariants. Namely, that for any revision that is
182
present, we either have all of the file content, or we have the parent
183
inventory and the delta file content.
184
"""
185
if not self.repository._fallback_repositories:
186
return
187
if not self.repository._format.supports_chks:
188
raise errors.BzrError("Cannot commit directly to a stacked branch"
189
" in pre-2a formats. See "
190
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
191
# This is a stacked repo, we need to make sure we have the parent
192
# inventories for the parents.
193
parent_keys = [(p,) for p in self.parents]
194
parent_map = self.repository.inventories._index.get_parent_map(parent_keys)
195
missing_parent_keys = set([pk for pk in parent_keys
196
if pk not in parent_map])
197
fallback_repos = list(reversed(self.repository._fallback_repositories))
198
missing_keys = [('inventories', pk[0])
199
for pk in missing_parent_keys]
200
resume_tokens = []
201
while missing_keys and fallback_repos:
202
fallback_repo = fallback_repos.pop()
203
source = fallback_repo._get_source(self.repository._format)
204
sink = self.repository._get_sink()
205
stream = source.get_stream_for_missing_keys(missing_keys)
206
missing_keys = sink.insert_stream_without_locking(stream,
207
self.repository._format)
208
if missing_keys:
209
raise errors.BzrError('Unable to fill in parent inventories for a'
210
' stacked branch')
211
149
212
def commit(self, message):
150
213
"""Make the actual commit.
151
214
152
215
:return: The revision id of the recorded revision.
153
216
"""
154
raise NotImplementedError(self.commit)
217
self._validate_unicode_text(message, 'commit message')
218
rev = _mod_revision.Revision(
219
timestamp=self._timestamp,
220
timezone=self._timezone,
221
committer=self._committer,
222
message=message,
223
inventory_sha1=self.inv_sha1,
224
revision_id=self._new_revision_id,
225
properties=self._revprops)
226
rev.parent_ids = self.parents
227
self.repository.add_revision(self._new_revision_id, rev,
228
self.new_inventory, self._config)
229
self._ensure_fallback_inventories()
230
self.repository.commit_write_group()
231
return self._new_revision_id
155
232
156
233
def abort(self):
157
234
"""Abort the commit that is being built.
158
235
"""
159
raise NotImplementedError(self.abort)
236
self.repository.abort_write_group()
160
237
161
238
def revision_tree(self):
162
239
"""Return the tree that was just committed.
163
240
164
After calling commit() this can be called to get a
165
RevisionTree representing the newly committed tree. This is
166
preferred to calling Repository.revision_tree() because that may
167
require deserializing the inventory, while we already have a copy in
241
After calling commit() this can be called to get a RevisionTree
242
representing the newly committed tree. This is preferred to
243
calling Repository.revision_tree() because that may require
244
deserializing the inventory, while we already have a copy in
168
245
memory.
169
246
"""
170
raise NotImplementedError(self.revision_tree)
247
if self.new_inventory is None:
248
self.new_inventory = self.repository.get_inventory(
249
self._new_revision_id)
250
return RevisionTree(self.repository, self.new_inventory,
251
self._new_revision_id)
171
252
172
253
def finish_inventory(self):
173
254
"""Tell the builder that the inventory is finished.
175
256
:return: The inventory id in the repository, which can be used with
176
257
repository.get_inventory.
177
258
"""
178
raise NotImplementedError(self.finish_inventory)
259
if self.new_inventory is None:
260
# an inventory delta was accumulated without creating a new
261
# inventory.
262
basis_id = self.basis_delta_revision
263
# We ignore the 'inventory' returned by add_inventory_by_delta
264
# because self.new_inventory is used to hint to the rest of the
265
# system what code path was taken
266
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
267
basis_id, self._basis_delta, self._new_revision_id,
268
self.parents)
269
else:
270
if self.new_inventory.root is None:
271
raise AssertionError('Root entry should be supplied to'
272
' record_entry_contents, as of bzr 0.10.')
273
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
274
self.new_inventory.revision_id = self._new_revision_id
275
self.inv_sha1 = self.repository.add_inventory(
276
self._new_revision_id,
277
self.new_inventory,
278
self.parents
279
)
280
return self._new_revision_id
179
281
180
282
def _gen_revision_id(self):
181
283
"""Return new revision-id."""
196
298
else:
197
299
self.random_revid = False
198
300
301
def _heads(self, file_id, revision_ids):
302
"""Calculate the graph heads for revision_ids in the graph of file_id.
303
304
This can use either a per-file graph or a global revision graph as we
305
have an identity relationship between the two graphs.
306
"""
307
return self.__heads(revision_ids)
308
309
def _check_root(self, ie, parent_invs, tree):
310
"""Helper for record_entry_contents.
311
312
:param ie: An entry being added.
313
:param parent_invs: The inventories of the parent revisions of the
314
commit.
315
:param tree: The tree that is being committed.
316
"""
317
# In this revision format, root entries have no knit or weave When
318
# serializing out to disk and back in root.revision is always
319
# _new_revision_id
320
ie.revision = self._new_revision_id
321
322
def _require_root_change(self, tree):
323
"""Enforce an appropriate root object change.
324
325
This is called once when record_iter_changes is called, if and only if
326
the root was not in the delta calculated by record_iter_changes.
327
328
:param tree: The tree which is being committed.
329
"""
330
if len(self.parents) == 0:
331
raise errors.RootMissing()
332
entry = entry_factory['directory'](tree.path2id(''), '',
333
None)
334
entry.revision = self._new_revision_id
335
self._basis_delta.append(('', '', entry.file_id, entry))
336
337
def _get_delta(self, ie, basis_inv, path):
338
"""Get a delta against the basis inventory for ie."""
339
if ie.file_id not in basis_inv:
340
# add
341
result = (None, path, ie.file_id, ie)
342
self._basis_delta.append(result)
343
return result
344
elif ie != basis_inv[ie.file_id]:
345
# common but altered
346
# TODO: avoid tis id2path call.
347
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
348
self._basis_delta.append(result)
349
return result
350
else:
351
# common, unaltered
352
return None
353
354
def get_basis_delta(self):
355
"""Return the complete inventory delta versus the basis inventory.
356
357
This has been built up with the calls to record_delete and
358
record_entry_contents. The client must have already called
359
will_record_deletes() to indicate that they will be generating a
360
complete delta.
361
362
:return: An inventory delta, suitable for use with apply_delta, or
363
Repository.add_inventory_by_delta, etc.
364
"""
365
if not self._recording_deletes:
366
raise AssertionError("recording deletes not activated.")
367
return self._basis_delta
368
369
def record_delete(self, path, file_id):
370
"""Record that a delete occured against a basis tree.
371
372
This is an optional API - when used it adds items to the basis_delta
373
being accumulated by the commit builder. It cannot be called unless the
374
method will_record_deletes() has been called to inform the builder that
375
a delta is being supplied.
376
377
:param path: The path of the thing deleted.
378
:param file_id: The file id that was deleted.
379
"""
380
if not self._recording_deletes:
381
raise AssertionError("recording deletes not activated.")
382
delta = (path, None, file_id, None)
383
self._basis_delta.append(delta)
384
self._any_changes = True
385
return delta
386
199
387
def will_record_deletes(self):
200
388
"""Tell the commit builder that deletes are being notified.
201
389
203
391
commit to be valid, deletes against the basis MUST be recorded via
204
392
builder.record_delete().
205
393
"""
206
raise NotImplementedError(self.will_record_deletes)
207
208
def record_iter_changes(self, tree, basis_revision_id, iter_changes):
394
self._recording_deletes = True
395
try:
396
basis_id = self.parents[0]
397
except IndexError:
398
basis_id = _mod_revision.NULL_REVISION
399
self.basis_delta_revision = basis_id
400
401
def record_entry_contents(self, ie, parent_invs, path, tree,
402
content_summary):
403
"""Record the content of ie from tree into the commit if needed.
404
405
Side effect: sets ie.revision when unchanged
406
407
:param ie: An inventory entry present in the commit.
408
:param parent_invs: The inventories of the parent revisions of the
409
commit.
410
:param path: The path the entry is at in the tree.
411
:param tree: The tree which contains this entry and should be used to
412
obtain content.
413
:param content_summary: Summary data from the tree about the paths
414
content - stat, length, exec, sha/link target. This is only
415
accessed when the entry has a revision of None - that is when it is
416
a candidate to commit.
417
:return: A tuple (change_delta, version_recorded, fs_hash).
418
change_delta is an inventory_delta change for this entry against
419
the basis tree of the commit, or None if no change occured against
420
the basis tree.
421
version_recorded is True if a new version of the entry has been
422
recorded. For instance, committing a merge where a file was only
423
changed on the other side will return (delta, False).
424
fs_hash is either None, or the hash details for the path (currently
425
a tuple of the contents sha1 and the statvalue returned by
426
tree.get_file_with_stat()).
427
"""
428
if self.new_inventory.root is None:
429
if ie.parent_id is not None:
430
raise errors.RootMissing()
431
self._check_root(ie, parent_invs, tree)
432
if ie.revision is None:
433
kind = content_summary[0]
434
else:
435
# ie is carried over from a prior commit
436
kind = ie.kind
437
# XXX: repository specific check for nested tree support goes here - if
438
# the repo doesn't want nested trees we skip it ?
439
if (kind == 'tree-reference' and
440
not self.repository._format.supports_tree_reference):
441
# mismatch between commit builder logic and repository:
442
# this needs the entry creation pushed down into the builder.
443
raise NotImplementedError('Missing repository subtree support.')
444
self.new_inventory.add(ie)
445
446
# TODO: slow, take it out of the inner loop.
447
try:
448
basis_inv = parent_invs[0]
449
except IndexError:
450
basis_inv = Inventory(root_id=None)
451
452
# ie.revision is always None if the InventoryEntry is considered
453
# for committing. We may record the previous parents revision if the
454
# content is actually unchanged against a sole head.
455
if ie.revision is not None:
456
if not self._versioned_root and path == '':
457
# repositories that do not version the root set the root's
458
# revision to the new commit even when no change occurs (more
459
# specifically, they do not record a revision on the root; and
460
# the rev id is assigned to the root during deserialisation -
461
# this masks when a change may have occurred against the basis.
462
# To match this we always issue a delta, because the revision
463
# of the root will always be changing.
464
if ie.file_id in basis_inv:
465
delta = (basis_inv.id2path(ie.file_id), path,
466
ie.file_id, ie)
467
else:
468
# add
469
delta = (None, path, ie.file_id, ie)
470
self._basis_delta.append(delta)
471
return delta, False, None
472
else:
473
# we don't need to commit this, because the caller already
474
# determined that an existing revision of this file is
475
# appropriate. If it's not being considered for committing then
476
# it and all its parents to the root must be unaltered so
477
# no-change against the basis.
478
if ie.revision == self._new_revision_id:
479
raise AssertionError("Impossible situation, a skipped "
480
"inventory entry (%r) claims to be modified in this "
481
"commit (%r).", (ie, self._new_revision_id))
482
return None, False, None
483
# XXX: Friction: parent_candidates should return a list not a dict
484
# so that we don't have to walk the inventories again.
485
parent_candiate_entries = ie.parent_candidates(parent_invs)
486
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
487
heads = []
488
for inv in parent_invs:
489
if ie.file_id in inv:
490
old_rev = inv[ie.file_id].revision
491
if old_rev in head_set:
492
heads.append(inv[ie.file_id].revision)
493
head_set.remove(inv[ie.file_id].revision)
494
495
store = False
496
# now we check to see if we need to write a new record to the
497
# file-graph.
498
# We write a new entry unless there is one head to the ancestors, and
499
# the kind-derived content is unchanged.
500
501
# Cheapest check first: no ancestors, or more the one head in the
502
# ancestors, we write a new node.
503
if len(heads) != 1:
504
store = True
505
if not store:
506
# There is a single head, look it up for comparison
507
parent_entry = parent_candiate_entries[heads[0]]
508
# if the non-content specific data has changed, we'll be writing a
509
# node:
510
if (parent_entry.parent_id != ie.parent_id or
511
parent_entry.name != ie.name):
512
store = True
513
# now we need to do content specific checks:
514
if not store:
515
# if the kind changed the content obviously has
516
if kind != parent_entry.kind:
517
store = True
518
# Stat cache fingerprint feedback for the caller - None as we usually
519
# don't generate one.
520
fingerprint = None
521
if kind == 'file':
522
if content_summary[2] is None:
523
raise ValueError("Files must not have executable = None")
524
if not store:
525
# We can't trust a check of the file length because of content
526
# filtering...
527
if (# if the exec bit has changed we have to store:
528
parent_entry.executable != content_summary[2]):
529
store = True
530
elif parent_entry.text_sha1 == content_summary[3]:
531
# all meta and content is unchanged (using a hash cache
532
# hit to check the sha)
533
ie.revision = parent_entry.revision
534
ie.text_size = parent_entry.text_size
535
ie.text_sha1 = parent_entry.text_sha1
536
ie.executable = parent_entry.executable
537
return self._get_delta(ie, basis_inv, path), False, None
538
else:
539
# Either there is only a hash change(no hash cache entry,
540
# or same size content change), or there is no change on
541
# this file at all.
542
# Provide the parent's hash to the store layer, so that the
543
# content is unchanged we will not store a new node.
544
nostore_sha = parent_entry.text_sha1
545
if store:
546
# We want to record a new node regardless of the presence or
547
# absence of a content change in the file.
548
nostore_sha = None
549
ie.executable = content_summary[2]
550
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
551
try:
552
text = file_obj.read()
553
finally:
554
file_obj.close()
555
try:
556
ie.text_sha1, ie.text_size = self._add_text_to_weave(
557
ie.file_id, text, heads, nostore_sha)
558
# Let the caller know we generated a stat fingerprint.
559
fingerprint = (ie.text_sha1, stat_value)
560
except errors.ExistingContent:
561
# Turns out that the file content was unchanged, and we were
562
# only going to store a new node if it was changed. Carry over
563
# the entry.
564
ie.revision = parent_entry.revision
565
ie.text_size = parent_entry.text_size
566
ie.text_sha1 = parent_entry.text_sha1
567
ie.executable = parent_entry.executable
568
return self._get_delta(ie, basis_inv, path), False, None
569
elif kind == 'directory':
570
if not store:
571
# all data is meta here, nothing specific to directory, so
572
# carry over:
573
ie.revision = parent_entry.revision
574
return self._get_delta(ie, basis_inv, path), False, None
575
self._add_text_to_weave(ie.file_id, '', heads, None)
576
elif kind == 'symlink':
577
current_link_target = content_summary[3]
578
if not store:
579
# symlink target is not generic metadata, check if it has
580
# changed.
581
if current_link_target != parent_entry.symlink_target:
582
store = True
583
if not store:
584
# unchanged, carry over.
585
ie.revision = parent_entry.revision
586
ie.symlink_target = parent_entry.symlink_target
587
return self._get_delta(ie, basis_inv, path), False, None
588
ie.symlink_target = current_link_target
589
self._add_text_to_weave(ie.file_id, '', heads, None)
590
elif kind == 'tree-reference':
591
if not store:
592
if content_summary[3] != parent_entry.reference_revision:
593
store = True
594
if not store:
595
# unchanged, carry over.
596
ie.reference_revision = parent_entry.reference_revision
597
ie.revision = parent_entry.revision
598
return self._get_delta(ie, basis_inv, path), False, None
599
ie.reference_revision = content_summary[3]
600
if ie.reference_revision is None:
601
raise AssertionError("invalid content_summary for nested tree: %r"
602
% (content_summary,))
603
self._add_text_to_weave(ie.file_id, '', heads, None)
604
else:
605
raise NotImplementedError('unknown kind')
606
ie.revision = self._new_revision_id
607
self._any_changes = True
608
return self._get_delta(ie, basis_inv, path), True, fingerprint
609
610
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
611
_entry_factory=entry_factory):
209
612
"""Record a new tree via iter_changes.
210
613
211
614
:param tree: The tree to obtain text contents from for changed objects.
216
619
to basis_revision_id. The iterator must not include any items with
217
620
a current kind of None - missing items must be either filtered out
218
621
or errored-on beefore record_iter_changes sees the item.
622
:param _entry_factory: Private method to bind entry_factory locally for
623
performance.
219
624
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
220
625
tree._observed_sha1.
221
626
"""
222
raise NotImplementedError(self.record_iter_changes)
627
# Create an inventory delta based on deltas between all the parents and
628
# deltas between all the parent inventories. We use inventory delta's
629
# between the inventory objects because iter_changes masks
630
# last-changed-field only changes.
631
# Working data:
632
# file_id -> change map, change is fileid, paths, changed, versioneds,
633
# parents, names, kinds, executables
634
merged_ids = {}
635
# {file_id -> revision_id -> inventory entry, for entries in parent
636
# trees that are not parents[0]
637
parent_entries = {}
638
ghost_basis = False
639
try:
640
revtrees = list(self.repository.revision_trees(self.parents))
641
except errors.NoSuchRevision:
642
# one or more ghosts, slow path.
643
revtrees = []
644
for revision_id in self.parents:
645
try:
646
revtrees.append(self.repository.revision_tree(revision_id))
647
except errors.NoSuchRevision:
648
if not revtrees:
649
basis_revision_id = _mod_revision.NULL_REVISION
650
ghost_basis = True
651
revtrees.append(self.repository.revision_tree(
652
_mod_revision.NULL_REVISION))
653
# The basis inventory from a repository
654
if revtrees:
655
basis_inv = revtrees[0].inventory
656
else:
657
basis_inv = self.repository.revision_tree(
658
_mod_revision.NULL_REVISION).inventory
659
if len(self.parents) > 0:
660
if basis_revision_id != self.parents[0] and not ghost_basis:
661
raise Exception(
662
"arbitrary basis parents not yet supported with merges")
663
for revtree in revtrees[1:]:
664
for change in revtree.inventory._make_delta(basis_inv):
665
if change[1] is None:
666
# Not present in this parent.
667
continue
668
if change[2] not in merged_ids:
669
if change[0] is not None:
670
basis_entry = basis_inv[change[2]]
671
merged_ids[change[2]] = [
672
# basis revid
673
basis_entry.revision,
674
# new tree revid
675
change[3].revision]
676
parent_entries[change[2]] = {
677
# basis parent
678
basis_entry.revision:basis_entry,
679
# this parent
680
change[3].revision:change[3],
681
}
682
else:
683
merged_ids[change[2]] = [change[3].revision]
684
parent_entries[change[2]] = {change[3].revision:change[3]}
685
else:
686
merged_ids[change[2]].append(change[3].revision)
687
parent_entries[change[2]][change[3].revision] = change[3]
688
else:
689
merged_ids = {}
690
# Setup the changes from the tree:
691
# changes maps file_id -> (change, [parent revision_ids])
692
changes= {}
693
for change in iter_changes:
694
# This probably looks up in basis_inv way to much.
695
if change[1][0] is not None:
696
head_candidate = [basis_inv[change[0]].revision]
697
else:
698
head_candidate = []
699
changes[change[0]] = change, merged_ids.get(change[0],
700
head_candidate)
701
unchanged_merged = set(merged_ids) - set(changes)
702
# Extend the changes dict with synthetic changes to record merges of
703
# texts.
704
for file_id in unchanged_merged:
705
# Record a merged version of these items that did not change vs the
706
# basis. This can be either identical parallel changes, or a revert
707
# of a specific file after a merge. The recorded content will be
708
# that of the current tree (which is the same as the basis), but
709
# the per-file graph will reflect a merge.
710
# NB:XXX: We are reconstructing path information we had, this
711
# should be preserved instead.
712
# inv delta change: (file_id, (path_in_source, path_in_target),
713
# changed_content, versioned, parent, name, kind,
714
# executable)
715
try:
716
basis_entry = basis_inv[file_id]
717
except errors.NoSuchId:
718
# a change from basis->some_parents but file_id isn't in basis
719
# so was new in the merge, which means it must have changed
720
# from basis -> current, and as it hasn't the add was reverted
721
# by the user. So we discard this change.
722
pass
723
else:
724
change = (file_id,
725
(basis_inv.id2path(file_id), tree.id2path(file_id)),
726
False, (True, True),
727
(basis_entry.parent_id, basis_entry.parent_id),
728
(basis_entry.name, basis_entry.name),
729
(basis_entry.kind, basis_entry.kind),
730
(basis_entry.executable, basis_entry.executable))
731
changes[file_id] = (change, merged_ids[file_id])
732
# changes contains tuples with the change and a set of inventory
733
# candidates for the file.
734
# inv delta is:
735
# old_path, new_path, file_id, new_inventory_entry
736
seen_root = False # Is the root in the basis delta?
737
inv_delta = self._basis_delta
738
modified_rev = self._new_revision_id
739
for change, head_candidates in changes.values():
740
if change[3][1]: # versioned in target.
741
# Several things may be happening here:
742
# We may have a fork in the per-file graph
743
# - record a change with the content from tree
744
# We may have a change against < all trees
745
# - carry over the tree that hasn't changed
746
# We may have a change against all trees
747
# - record the change with the content from tree
748
kind = change[6][1]
749
file_id = change[0]
750
entry = _entry_factory[kind](file_id, change[5][1],
751
change[4][1])
752
head_set = self._heads(change[0], set(head_candidates))
753
heads = []
754
# Preserve ordering.
755
for head_candidate in head_candidates:
756
if head_candidate in head_set:
757
heads.append(head_candidate)
758
head_set.remove(head_candidate)
759
carried_over = False
760
if len(heads) == 1:
761
# Could be a carry-over situation:
762
parent_entry_revs = parent_entries.get(file_id, None)
763
if parent_entry_revs:
764
parent_entry = parent_entry_revs.get(heads[0], None)
765
else:
766
parent_entry = None
767
if parent_entry is None:
768
# The parent iter_changes was called against is the one
769
# that is the per-file head, so any change is relevant
770
# iter_changes is valid.
771
carry_over_possible = False
772
else:
773
# could be a carry over situation
774
# A change against the basis may just indicate a merge,
775
# we need to check the content against the source of the
776
# merge to determine if it was changed after the merge
777
# or carried over.
778
if (parent_entry.kind != entry.kind or
779
parent_entry.parent_id != entry.parent_id or
780
parent_entry.name != entry.name):
781
# Metadata common to all entries has changed
782
# against per-file parent
783
carry_over_possible = False
784
else:
785
carry_over_possible = True
786
# per-type checks for changes against the parent_entry
787
# are done below.
788
else:
789
# Cannot be a carry-over situation
790
carry_over_possible = False
791
# Populate the entry in the delta
792
if kind == 'file':
793
# XXX: There is still a small race here: If someone reverts the content of a file
794
# after iter_changes examines and decides it has changed,
795
# we will unconditionally record a new version even if some
796
# other process reverts it while commit is running (with
797
# the revert happening after iter_changes did its
798
# examination).
799
if change[7][1]:
800
entry.executable = True
801
else:
802
entry.executable = False
803
if (carry_over_possible and
804
parent_entry.executable == entry.executable):
805
# Check the file length, content hash after reading
806
# the file.
807
nostore_sha = parent_entry.text_sha1
808
else:
809
nostore_sha = None
810
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
811
try:
812
text = file_obj.read()
813
finally:
814
file_obj.close()
815
try:
816
entry.text_sha1, entry.text_size = self._add_text_to_weave(
817
file_id, text, heads, nostore_sha)
818
yield file_id, change[1][1], (entry.text_sha1, stat_value)
819
except errors.ExistingContent:
820
# No content change against a carry_over parent
821
# Perhaps this should also yield a fs hash update?
822
carried_over = True
823
entry.text_size = parent_entry.text_size
824
entry.text_sha1 = parent_entry.text_sha1
825
elif kind == 'symlink':
826
# Wants a path hint?
827
entry.symlink_target = tree.get_symlink_target(file_id)
828
if (carry_over_possible and
829
parent_entry.symlink_target == entry.symlink_target):
830
carried_over = True
831
else:
832
self._add_text_to_weave(change[0], '', heads, None)
833
elif kind == 'directory':
834
if carry_over_possible:
835
carried_over = True
836
else:
837
# Nothing to set on the entry.
838
# XXX: split into the Root and nonRoot versions.
839
if change[1][1] != '' or self.repository.supports_rich_root():
840
self._add_text_to_weave(change[0], '', heads, None)
841
elif kind == 'tree-reference':
842
if not self.repository._format.supports_tree_reference:
843
# This isn't quite sane as an error, but we shouldn't
844
# ever see this code path in practice: tree's don't
845
# permit references when the repo doesn't support tree
846
# references.
847
raise errors.UnsupportedOperation(tree.add_reference,
848
self.repository)
849
reference_revision = tree.get_reference_revision(change[0])
850
entry.reference_revision = reference_revision
851
if (carry_over_possible and
852
parent_entry.reference_revision == reference_revision):
853
carried_over = True
854
else:
855
self._add_text_to_weave(change[0], '', heads, None)
856
else:
857
raise AssertionError('unknown kind %r' % kind)
858
if not carried_over:
859
entry.revision = modified_rev
860
else:
861
entry.revision = parent_entry.revision
862
else:
863
entry = None
864
new_path = change[1][1]
865
inv_delta.append((change[1][0], new_path, change[0], entry))
866
if new_path == '':
867
seen_root = True
868
self.new_inventory = None
869
if len(inv_delta):
870
# This should perhaps be guarded by a check that the basis we
871
# commit against is the basis for the commit and if not do a delta
872
# against the basis.
873
self._any_changes = True
874
if not seen_root:
875
# housekeeping root entry changes do not affect no-change commits.
876
self._require_root_change(tree)
877
self.basis_delta_revision = basis_revision_id
878
879
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
880
parent_keys = tuple([(file_id, parent) for parent in parents])
881
return self.repository.texts._add_text(
882
(file_id, self._new_revision_id), parent_keys, new_text,
883
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
884
885
886
class RootCommitBuilder(CommitBuilder):
887
"""This commitbuilder actually records the root id"""
888
889
# the root entry gets versioned properly by this builder.
890
_versioned_root = True
891
892
def _check_root(self, ie, parent_invs, tree):
893
"""Helper for record_entry_contents.
894
895
:param ie: An entry being added.
896
:param parent_invs: The inventories of the parent revisions of the
897
commit.
898
:param tree: The tree that is being committed.
899
"""
900
901
def _require_root_change(self, tree):
902
"""Enforce an appropriate root object change.
903
904
This is called once when record_iter_changes is called, if and only if
905
the root was not in the delta calculated by record_iter_changes.
906
907
:param tree: The tree which is being committed.
908
"""
909
# versioned roots do not change unless the tree found a change.
223
910
224
911
225
912
class RepositoryWriteLockResult(LogicalLockResult):
250
937
revisions and file history. It's normally accessed only by the Branch,
251
938
which views a particular line of development through that history.
252
939
253
See VersionedFileRepository in bzrlib.vf_repository for the
254
base class for most Bazaar repositories.
940
The Repository builds on top of some byte storage facilies (the revisions,
941
signatures, inventories, texts and chk_bytes attributes) and a Transport,
942
which respectively provide byte storage and a means to access the (possibly
943
remote) disk.
944
945
The byte storage facilities are addressed via tuples, which we refer to
946
as 'keys' throughout the code base. Revision_keys, inventory_keys and
947
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
948
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
949
byte string made up of a hash identifier and a hash value.
950
We use this interface because it allows low friction with the underlying
951
code that implements disk indices, network encoding and other parts of
952
bzrlib.
953
954
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
955
the serialised revisions for the repository. This can be used to obtain
956
revision graph information or to access raw serialised revisions.
957
The result of trying to insert data into the repository via this store
958
is undefined: it should be considered read-only except for implementors
959
of repositories.
960
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
961
the serialised signatures for the repository. This can be used to
962
obtain access to raw serialised signatures. The result of trying to
963
insert data into the repository via this store is undefined: it should
964
be considered read-only except for implementors of repositories.
965
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
966
the serialised inventories for the repository. This can be used to
967
obtain unserialised inventories. The result of trying to insert data
968
into the repository via this store is undefined: it should be
969
considered read-only except for implementors of repositories.
970
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
971
texts of files and directories for the repository. This can be used to
972
obtain file texts or file graphs. Note that Repository.iter_file_bytes
973
is usually a better interface for accessing file texts.
974
The result of trying to insert data into the repository via this store
975
is undefined: it should be considered read-only except for implementors
976
of repositories.
977
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
978
any data the repository chooses to store or have indexed by its hash.
979
The result of trying to insert data into the repository via this store
980
is undefined: it should be considered read-only except for implementors
981
of repositories.
982
:ivar _transport: Transport for file access to repository, typically
983
pointing to .bzr/repository.
255
984
"""
256
985
986
# What class to use for a CommitBuilder. Often it's simpler to change this
987
# in a Repository class subclass rather than to override
988
# get_commit_builder.
989
_commit_builder_class = CommitBuilder
990
257
991
def abort_write_group(self, suppress_errors=False):
258
992
"""Commit the contents accrued within the current write group.
259
993
281
1015
raise
282
1016
mutter('abort_write_group failed')
283
1017
log_exception_quietly()
284
note(gettext('bzr: ERROR (ignored): %s'), exc)
1018
note('bzr: ERROR (ignored): %s', exc)
285
1019
self._write_group = None
286
1020
287
1021
def _abort_write_group(self):
302
1036
303
1037
:param repository: A repository.
304
1038
"""
305
raise NotImplementedError(self.add_fallback_repository)
1039
if not self._format.supports_external_lookups:
1040
raise errors.UnstackableRepositoryFormat(self._format, self.base)
1041
if self.is_locked():
1042
# This repository will call fallback.unlock() when we transition to
1043
# the unlocked state, so we make sure to increment the lock count
1044
repository.lock_read()
1045
self._check_fallback_repository(repository)
1046
self._fallback_repositories.append(repository)
1047
self.texts.add_fallback_versioned_files(repository.texts)
1048
self.inventories.add_fallback_versioned_files(repository.inventories)
1049
self.revisions.add_fallback_versioned_files(repository.revisions)
1050
self.signatures.add_fallback_versioned_files(repository.signatures)
1051
if self.chk_bytes is not None:
1052
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
306
1053
307
1054
def _check_fallback_repository(self, repository):
308
1055
"""Check that this repository can fallback to repository safely.
313
1060
"""
314
1061
return InterRepository._assert_same_model(self, repository)
315
1062
1063
def add_inventory(self, revision_id, inv, parents):
1064
"""Add the inventory inv to the repository as revision_id.
1065
1066
:param parents: The revision ids of the parents that revision_id
1067
is known to have and are in the repository already.
1068
1069
:returns: The validator(which is a sha1 digest, though what is sha'd is
1070
repository format specific) of the serialized inventory.
1071
"""
1072
if not self.is_in_write_group():
1073
raise AssertionError("%r not in write group" % (self,))
1074
_mod_revision.check_not_reserved_id(revision_id)
1075
if not (inv.revision_id is None or inv.revision_id == revision_id):
1076
raise AssertionError(
1077
"Mismatch between inventory revision"
1078
" id and insertion revid (%r, %r)"
1079
% (inv.revision_id, revision_id))
1080
if inv.root is None:
1081
raise errors.RootMissing()
1082
return self._add_inventory_checked(revision_id, inv, parents)
1083
1084
def _add_inventory_checked(self, revision_id, inv, parents):
1085
"""Add inv to the repository after checking the inputs.
1086
1087
This function can be overridden to allow different inventory styles.
1088
1089
:seealso: add_inventory, for the contract.
1090
"""
1091
inv_lines = self._serializer.write_inventory_to_lines(inv)
1092
return self._inventory_add_lines(revision_id, parents,
1093
inv_lines, check_content=False)
1094
1095
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1096
parents, basis_inv=None, propagate_caches=False):
1097
"""Add a new inventory expressed as a delta against another revision.
1098
1099
See the inventory developers documentation for the theory behind
1100
inventory deltas.
1101
1102
:param basis_revision_id: The inventory id the delta was created
1103
against. (This does not have to be a direct parent.)
1104
:param delta: The inventory delta (see Inventory.apply_delta for
1105
details).
1106
:param new_revision_id: The revision id that the inventory is being
1107
added for.
1108
:param parents: The revision ids of the parents that revision_id is
1109
known to have and are in the repository already. These are supplied
1110
for repositories that depend on the inventory graph for revision
1111
graph access, as well as for those that pun ancestry with delta
1112
compression.
1113
:param basis_inv: The basis inventory if it is already known,
1114
otherwise None.
1115
:param propagate_caches: If True, the caches for this inventory are
1116
copied to and updated for the result if possible.
1117
1118
:returns: (validator, new_inv)
1119
The validator(which is a sha1 digest, though what is sha'd is
1120
repository format specific) of the serialized inventory, and the
1121
resulting inventory.
1122
"""
1123
if not self.is_in_write_group():
1124
raise AssertionError("%r not in write group" % (self,))
1125
_mod_revision.check_not_reserved_id(new_revision_id)
1126
basis_tree = self.revision_tree(basis_revision_id)
1127
basis_tree.lock_read()
1128
try:
1129
# Note that this mutates the inventory of basis_tree, which not all
1130
# inventory implementations may support: A better idiom would be to
1131
# return a new inventory, but as there is no revision tree cache in
1132
# repository this is safe for now - RBC 20081013
1133
if basis_inv is None:
1134
basis_inv = basis_tree.inventory
1135
basis_inv.apply_delta(delta)
1136
basis_inv.revision_id = new_revision_id
1137
return (self.add_inventory(new_revision_id, basis_inv, parents),
1138
basis_inv)
1139
finally:
1140
basis_tree.unlock()
1141
1142
def _inventory_add_lines(self, revision_id, parents, lines,
1143
check_content=True):
1144
"""Store lines in inv_vf and return the sha1 of the inventory."""
1145
parents = [(parent,) for parent in parents]
1146
result = self.inventories.add_lines((revision_id,), parents, lines,
1147
check_content=check_content)[0]
1148
self.inventories._access.flush()
1149
return result
1150
1151
def add_revision(self, revision_id, rev, inv=None, config=None):
1152
"""Add rev to the revision store as revision_id.
1153
1154
:param revision_id: the revision id to use.
1155
:param rev: The revision object.
1156
:param inv: The inventory for the revision. if None, it will be looked
1157
up in the inventory storer
1158
:param config: If None no digital signature will be created.
1159
If supplied its signature_needed method will be used
1160
to determine if a signature should be made.
1161
"""
1162
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1163
# rev.parent_ids?
1164
_mod_revision.check_not_reserved_id(revision_id)
1165
if config is not None and config.signature_needed():
1166
if inv is None:
1167
inv = self.get_inventory(revision_id)
1168
plaintext = Testament(rev, inv).as_short_text()
1169
self.store_revision_signature(
1170
gpg.GPGStrategy(config), plaintext, revision_id)
1171
# check inventory present
1172
if not self.inventories.get_parent_map([(revision_id,)]):
1173
if inv is None:
1174
raise errors.WeaveRevisionNotPresent(revision_id,
1175
self.inventories)
1176
else:
1177
# yes, this is not suitable for adding with ghosts.
1178
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1179
rev.parent_ids)
1180
else:
1181
key = (revision_id,)
1182
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1183
self._add_revision(rev)
1184
1185
def _add_revision(self, revision):
1186
text = self._serializer.write_revision_to_string(revision)
1187
key = (revision.revision_id,)
1188
parents = tuple((parent,) for parent in revision.parent_ids)
1189
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1190
316
1191
def all_revision_ids(self):
317
1192
"""Returns a list of all the revision ids in the repository.
318
1193
341
1216
"""
342
1217
self.control_files.break_lock()
343
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
344
1344
@staticmethod
345
1345
def create(a_bzrdir):
346
1346
"""Construct the current default format repository in a_bzrdir."""
351
1351
352
1352
:param _format: The format of the repository on disk.
353
1353
:param a_bzrdir: The BzrDir of the repository.
354
:param control_files: Control files to use for locking, etc.
355
1354
"""
356
1355
# In the future we will have a single api for all stores for
357
1356
# getting file texts, inventories and revisions, then
361
1360
# the following are part of the public API for Repository:
362
1361
self.bzrdir = a_bzrdir
363
1362
self.control_files = control_files
1363
self._transport = control_files._transport
1364
self.base = self._transport.base
364
1365
# for tests
1366
self._reconcile_does_inventory_gc = True
1367
self._reconcile_fixes_text_parents = False
1368
self._reconcile_backsup_inventory = True
365
1369
self._write_group = None
366
1370
# Additional places to query for data.
367
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
368
1377
369
1378
@property
370
1379
def user_transport(self):
404
1413
"""
405
1414
if self.__class__ is not other.__class__:
406
1415
return False
407
return (self.control_url == other.control_url)
1416
return (self._transport.base == other._transport.base)
408
1417
409
1418
def is_in_write_group(self):
410
1419
"""Return True if there is an open write group.
510
1519
if revid and committers:
511
1520
result['committers'] = 0
512
1521
if revid and revid != _mod_revision.NULL_REVISION:
513
graph = self.get_graph()
514
1522
if committers:
515
1523
all_committers = set()
516
revisions = [r for (r, p) in graph.iter_ancestry([revid])
517
if r != _mod_revision.NULL_REVISION]
518
last_revision = None
1524
revisions = self.get_ancestry(revid)
1525
# pop the leading None
1526
revisions.pop(0)
1527
first_revision = None
519
1528
if not committers:
520
1529
# ignore the revisions in the middle - just grab first and last
521
1530
revisions = revisions[0], revisions[-1]
522
1531
for revision in self.get_revisions(revisions):
523
if not last_revision:
524
last_revision = revision
1532
if not first_revision:
1533
first_revision = revision
525
1534
if committers:
526
1535
all_committers.add(revision.committer)
527
first_revision = revision
1536
last_revision = revision
528
1537
if committers:
529
1538
result['committers'] = len(all_committers)
530
1539
result['firstrev'] = (first_revision.timestamp,
531
1540
first_revision.timezone)
532
1541
result['latestrev'] = (last_revision.timestamp,
533
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
534
1552
return result
535
1553
536
1554
def find_branches(self, using=False):
573
1591
@needs_read_lock
574
1592
def search_missing_revision_ids(self, other,
575
1593
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
576
find_ghosts=True, revision_ids=None, if_present_ids=None,
577
limit=None):
1594
find_ghosts=True, revision_ids=None, if_present_ids=None):
578
1595
"""Return the revision ids that other has that this does not.
579
1596
580
1597
These are returned in topological order.
593
1610
revision_ids = [revision_id]
594
1611
return InterRepository.get(other, self).search_missing_revision_ids(
595
1612
find_ghosts=find_ghosts, revision_ids=revision_ids,
596
if_present_ids=if_present_ids, limit=limit)
1613
if_present_ids=if_present_ids)
597
1614
598
1615
@staticmethod
599
1616
def open(base):
641
1658
def suspend_write_group(self):
642
1659
raise errors.UnsuspendableWriteGroup(self)
643
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
644
1716
def refresh_data(self):
645
1717
"""Re-read any data needed to synchronise with disk.
646
1718
715
1787
716
1788
def get_commit_builder(self, branch, parents, config, timestamp=None,
717
1789
timezone=None, committer=None, revprops=None,
718
revision_id=None, lossy=False):
1790
revision_id=None):
719
1791
"""Obtain a CommitBuilder for this repository.
720
1792
721
1793
:param branch: Branch to commit to.
726
1798
:param committer: Optional committer to set for commit.
727
1799
:param revprops: Optional dictionary of revision properties.
728
1800
:param revision_id: Optional revision id.
729
:param lossy: Whether to discard data that can not be natively
730
represented, when pushing to a foreign VCS
731
1801
"""
732
raise NotImplementedError(self.get_commit_builder)
1802
if self._fallback_repositories and not self._format.supports_chks:
1803
raise errors.BzrError("Cannot commit directly to a stacked branch"
1804
" in pre-2a formats. See "
1805
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1806
result = self._commit_builder_class(self, parents, config,
1807
timestamp, timezone, committer, revprops, revision_id)
1808
self.start_write_group()
1809
return result
733
1810
734
1811
@only_raises(errors.LockNotHeld, errors.LockBroken)
735
1812
def unlock(self):
742
1819
'Must end write groups before releasing write locks.')
743
1820
self.control_files.unlock()
744
1821
if self.control_files._lock_count == 0:
1822
self._inventory_entry_cache.clear()
745
1823
for repo in self._fallback_repositories:
746
1824
repo.unlock()
747
1825
812
1890
dest_repo = a_bzrdir.open_repository()
813
1891
return dest_repo
814
1892
1893
def _get_sink(self):
1894
"""Return a sink for streaming into this repository."""
1895
return StreamSink(self)
1896
1897
def _get_source(self, to_format):
1898
"""Return a source for streaming from this repository."""
1899
return StreamSource(self, to_format)
1900
815
1901
@needs_read_lock
816
1902
def has_revision(self, revision_id):
817
1903
"""True if this repository has a copy of the revision."""
824
1910
:param revision_ids: An iterable of revision_ids.
825
1911
:return: A set of the revision_ids that were present.
826
1912
"""
827
raise NotImplementedError(self.has_revisions)
1913
parent_map = self.revisions.get_parent_map(
1914
[(rev_id,) for rev_id in revision_ids])
1915
result = set()
1916
if _mod_revision.NULL_REVISION in revision_ids:
1917
result.add(_mod_revision.NULL_REVISION)
1918
result.update([key[0] for key in parent_map])
1919
return result
828
1920
829
1921
@needs_read_lock
830
1922
def get_revision(self, revision_id):
831
1923
"""Return the Revision object for a named revision."""
832
1924
return self.get_revisions([revision_id])[0]
833
1925
1926
@needs_read_lock
834
1927
def get_revision_reconcile(self, revision_id):
835
1928
"""'reconcile' helper routine that allows access to a revision always.
836
1929
839
1932
be used by reconcile, or reconcile-alike commands that are correcting
840
1933
or testing the revision graph.
841
1934
"""
842
raise NotImplementedError(self.get_revision_reconcile)
1935
return self._get_revisions([revision_id])[0]
843
1936
1937
@needs_read_lock
844
1938
def get_revisions(self, revision_ids):
845
1939
"""Get many revisions at once.
846
1940
847
1941
Repositories that need to check data on every revision read should
848
1942
subclass this method.
849
1943
"""
850
raise NotImplementedError(self.get_revisions)
1944
return self._get_revisions(revision_ids)
1945
1946
@needs_read_lock
1947
def _get_revisions(self, revision_ids):
1948
"""Core work logic to get many revisions without sanity checks."""
1949
revs = {}
1950
for revid, rev in self._iter_revisions(revision_ids):
1951
if rev is None:
1952
raise errors.NoSuchRevision(self, revid)
1953
revs[revid] = rev
1954
return [revs[revid] for revid in revision_ids]
1955
1956
def _iter_revisions(self, revision_ids):
1957
"""Iterate over revision objects.
1958
1959
:param revision_ids: An iterable of revisions to examine. None may be
1960
passed to request all revisions known to the repository. Note that
1961
not all repositories can find unreferenced revisions; for those
1962
repositories only referenced ones will be returned.
1963
:return: An iterator of (revid, revision) tuples. Absent revisions (
1964
those asked for but not available) are returned as (revid, None).
1965
"""
1966
if revision_ids is None:
1967
revision_ids = self.all_revision_ids()
1968
else:
1969
for rev_id in revision_ids:
1970
if not rev_id or not isinstance(rev_id, basestring):
1971
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1972
keys = [(key,) for key in revision_ids]
1973
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1974
for record in stream:
1975
revid = record.key[0]
1976
if record.storage_kind == 'absent':
1977
yield (revid, None)
1978
else:
1979
text = record.get_bytes_as('fulltext')
1980
rev = self._serializer.read_revision_from_string(text)
1981
yield (revid, rev)
851
1982
852
1983
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
853
1984
"""Produce a generator of revision deltas.
908
2039
signature = gpg_strategy.sign(plaintext)
909
2040
self.add_signature_text(revision_id, signature)
910
2041
2042
@needs_write_lock
911
2043
def add_signature_text(self, revision_id, signature):
912
"""Store a signature text for a revision.
913
914
:param revision_id: Revision id of the revision
915
:param signature: Signature text.
916
"""
917
raise NotImplementedError(self.add_signature_text)
2044
self.signatures.add_lines((revision_id,), (),
2045
osutils.split_lines(signature))
2046
2047
def find_text_key_references(self):
2048
"""Find the text key references within the repository.
2049
2050
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2051
to whether they were referred to by the inventory of the
2052
revision_id that they contain. The inventory texts from all present
2053
revision ids are assessed to generate this report.
2054
"""
2055
revision_keys = self.revisions.keys()
2056
w = self.inventories
2057
pb = ui.ui_factory.nested_progress_bar()
2058
try:
2059
return self._serializer._find_text_key_references(
2060
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
2061
finally:
2062
pb.finished()
2063
2064
def _inventory_xml_lines_for_keys(self, keys):
2065
"""Get a line iterator of the sort needed for findind references.
2066
2067
Not relevant for non-xml inventory repositories.
2068
2069
Ghosts in revision_keys are ignored.
2070
2071
:param revision_keys: The revision keys for the inventories to inspect.
2072
:return: An iterator over (inventory line, revid) for the fulltexts of
2073
all of the xml inventories specified by revision_keys.
2074
"""
2075
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2076
for record in stream:
2077
if record.storage_kind != 'absent':
2078
chunks = record.get_bytes_as('chunked')
2079
revid = record.key[-1]
2080
lines = osutils.chunks_to_lines(chunks)
2081
for line in lines:
2082
yield line, revid
2083
2084
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2085
revision_keys):
2086
"""Helper routine for fileids_altered_by_revision_ids.
2087
2088
This performs the translation of xml lines to revision ids.
2089
2090
:param line_iterator: An iterator of lines, origin_version_id
2091
:param revision_keys: The revision ids to filter for. This should be a
2092
set or other type which supports efficient __contains__ lookups, as
2093
the revision key from each parsed line will be looked up in the
2094
revision_keys filter.
2095
:return: a dictionary mapping altered file-ids to an iterable of
2096
revision_ids. Each altered file-ids has the exact revision_ids that
2097
altered it listed explicitly.
2098
"""
2099
seen = set(self._serializer._find_text_key_references(
2100
line_iterator).iterkeys())
2101
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2102
parent_seen = set(self._serializer._find_text_key_references(
2103
self._inventory_xml_lines_for_keys(parent_keys)))
2104
new_keys = seen - parent_seen
2105
result = {}
2106
setdefault = result.setdefault
2107
for key in new_keys:
2108
setdefault(key[0], set()).add(key[-1])
2109
return result
918
2110
919
2111
def _find_parent_ids_of_revisions(self, revision_ids):
920
2112
"""Find all parent ids that are mentioned in the revision graph.
929
2121
parent_ids.discard(_mod_revision.NULL_REVISION)
930
2122
return parent_ids
931
2123
932
def fileids_altered_by_revision_ids(self, revision_ids):
2124
def _find_parent_keys_of_revisions(self, revision_keys):
2125
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2126
2127
:param revision_keys: An iterable of revision_keys.
2128
:return: The parents of all revision_keys that are not already in
2129
revision_keys
2130
"""
2131
parent_map = self.revisions.get_parent_map(revision_keys)
2132
parent_keys = set()
2133
map(parent_keys.update, parent_map.itervalues())
2134
parent_keys.difference_update(revision_keys)
2135
parent_keys.discard(_mod_revision.NULL_REVISION)
2136
return parent_keys
2137
2138
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
933
2139
"""Find the file ids and versions affected by revisions.
934
2140
935
2141
:param revisions: an iterable containing revision ids.
2142
:param _inv_weave: The inventory weave from this repository or None.
2143
If None, the inventory weave will be opened automatically.
936
2144
:return: a dictionary mapping altered file-ids to an iterable of
937
revision_ids. Each altered file-ids has the exact revision_ids
938
that altered it listed explicitly.
2145
revision_ids. Each altered file-ids has the exact revision_ids that
2146
altered it listed explicitly.
939
2147
"""
940
raise NotImplementedError(self.fileids_altered_by_revision_ids)
2148
selected_keys = set((revid,) for revid in revision_ids)
2149
w = _inv_weave or self.inventories
2150
return self._find_file_ids_from_xml_inventory_lines(
2151
w.iter_lines_added_or_present_in_keys(
2152
selected_keys, pb=None),
2153
selected_keys)
941
2154
942
2155
def iter_files_bytes(self, desired_files):
943
2156
"""Iterate through file versions.
950
2163
uniquely identify the file version in the caller's context. (Examples:
951
2164
an index number or a TreeTransform trans_id.)
952
2165
2166
bytes_iterator is an iterable of bytestrings for the file. The
2167
kind of iterable and length of the bytestrings are unspecified, but for
2168
this implementation, it is a list of bytes produced by
2169
VersionedFile.get_record_stream().
2170
953
2171
:param desired_files: a list of (file_id, revision_id, identifier)
954
2172
triples
955
2173
"""
956
raise NotImplementedError(self.iter_files_bytes)
2174
text_keys = {}
2175
for file_id, revision_id, callable_data in desired_files:
2176
text_keys[(file_id, revision_id)] = callable_data
2177
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2178
if record.storage_kind == 'absent':
2179
raise errors.RevisionNotPresent(record.key, self)
2180
yield text_keys[record.key], record.get_bytes_as('chunked')
2181
2182
def _generate_text_key_index(self, text_key_references=None,
2183
ancestors=None):
2184
"""Generate a new text key index for the repository.
2185
2186
This is an expensive function that will take considerable time to run.
2187
2188
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2189
list of parents, also text keys. When a given key has no parents,
2190
the parents list will be [NULL_REVISION].
2191
"""
2192
# All revisions, to find inventory parents.
2193
if ancestors is None:
2194
graph = self.get_graph()
2195
ancestors = graph.get_parent_map(self.all_revision_ids())
2196
if text_key_references is None:
2197
text_key_references = self.find_text_key_references()
2198
pb = ui.ui_factory.nested_progress_bar()
2199
try:
2200
return self._do_generate_text_key_index(ancestors,
2201
text_key_references, pb)
2202
finally:
2203
pb.finished()
2204
2205
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2206
"""Helper for _generate_text_key_index to avoid deep nesting."""
2207
revision_order = tsort.topo_sort(ancestors)
2208
invalid_keys = set()
2209
revision_keys = {}
2210
for revision_id in revision_order:
2211
revision_keys[revision_id] = set()
2212
text_count = len(text_key_references)
2213
# a cache of the text keys to allow reuse; costs a dict of all the
2214
# keys, but saves a 2-tuple for every child of a given key.
2215
text_key_cache = {}
2216
for text_key, valid in text_key_references.iteritems():
2217
if not valid:
2218
invalid_keys.add(text_key)
2219
else:
2220
revision_keys[text_key[1]].add(text_key)
2221
text_key_cache[text_key] = text_key
2222
del text_key_references
2223
text_index = {}
2224
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2225
NULL_REVISION = _mod_revision.NULL_REVISION
2226
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2227
# too small for large or very branchy trees. However, for 55K path
2228
# trees, it would be easy to use too much memory trivially. Ideally we
2229
# could gauge this by looking at available real memory etc, but this is
2230
# always a tricky proposition.
2231
inventory_cache = lru_cache.LRUCache(10)
2232
batch_size = 10 # should be ~150MB on a 55K path tree
2233
batch_count = len(revision_order) / batch_size + 1
2234
processed_texts = 0
2235
pb.update("Calculating text parents", processed_texts, text_count)
2236
for offset in xrange(batch_count):
2237
to_query = revision_order[offset * batch_size:(offset + 1) *
2238
batch_size]
2239
if not to_query:
2240
break
2241
for revision_id in to_query:
2242
parent_ids = ancestors[revision_id]
2243
for text_key in revision_keys[revision_id]:
2244
pb.update("Calculating text parents", processed_texts)
2245
processed_texts += 1
2246
candidate_parents = []
2247
for parent_id in parent_ids:
2248
parent_text_key = (text_key[0], parent_id)
2249
try:
2250
check_parent = parent_text_key not in \
2251
revision_keys[parent_id]
2252
except KeyError:
2253
# the parent parent_id is a ghost:
2254
check_parent = False
2255
# truncate the derived graph against this ghost.
2256
parent_text_key = None
2257
if check_parent:
2258
# look at the parent commit details inventories to
2259
# determine possible candidates in the per file graph.
2260
# TODO: cache here.
2261
try:
2262
inv = inventory_cache[parent_id]
2263
except KeyError:
2264
inv = self.revision_tree(parent_id).inventory
2265
inventory_cache[parent_id] = inv
2266
try:
2267
parent_entry = inv[text_key[0]]
2268
except (KeyError, errors.NoSuchId):
2269
parent_entry = None
2270
if parent_entry is not None:
2271
parent_text_key = (
2272
text_key[0], parent_entry.revision)
2273
else:
2274
parent_text_key = None
2275
if parent_text_key is not None:
2276
candidate_parents.append(
2277
text_key_cache[parent_text_key])
2278
parent_heads = text_graph.heads(candidate_parents)
2279
new_parents = list(parent_heads)
2280
new_parents.sort(key=lambda x:candidate_parents.index(x))
2281
if new_parents == []:
2282
new_parents = [NULL_REVISION]
2283
text_index[text_key] = new_parents
2284
2285
for text_key in invalid_keys:
2286
text_index[text_key] = [NULL_REVISION]
2287
return text_index
2288
2289
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2290
"""Get an iterable listing the keys of all the data introduced by a set
2291
of revision IDs.
2292
2293
The keys will be ordered so that the corresponding items can be safely
2294
fetched and inserted in that order.
2295
2296
:returns: An iterable producing tuples of (knit-kind, file-id,
2297
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2298
'revisions'. file-id is None unless knit-kind is 'file'.
2299
"""
2300
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2301
yield result
2302
del _files_pb
2303
for result in self._find_non_file_keys_to_fetch(revision_ids):
2304
yield result
2305
2306
def _find_file_keys_to_fetch(self, revision_ids, pb):
2307
# XXX: it's a bit weird to control the inventory weave caching in this
2308
# generator. Ideally the caching would be done in fetch.py I think. Or
2309
# maybe this generator should explicitly have the contract that it
2310
# should not be iterated until the previously yielded item has been
2311
# processed?
2312
inv_w = self.inventories
2313
2314
# file ids that changed
2315
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2316
count = 0
2317
num_file_ids = len(file_ids)
2318
for file_id, altered_versions in file_ids.iteritems():
2319
if pb is not None:
2320
pb.update("Fetch texts", count, num_file_ids)
2321
count += 1
2322
yield ("file", file_id, altered_versions)
2323
2324
def _find_non_file_keys_to_fetch(self, revision_ids):
2325
# inventory
2326
yield ("inventory", None, revision_ids)
2327
2328
# signatures
2329
# XXX: Note ATM no callers actually pay attention to this return
2330
# instead they just use the list of revision ids and ignore
2331
# missing sigs. Consider removing this work entirely
2332
revisions_with_signatures = set(self.signatures.get_parent_map(
2333
[(r,) for r in revision_ids]))
2334
revisions_with_signatures = set(
2335
[r for (r,) in revisions_with_signatures])
2336
revisions_with_signatures.intersection_update(revision_ids)
2337
yield ("signatures", None, revisions_with_signatures)
2338
2339
# revisions
2340
yield ("revisions", None, revision_ids)
2341
2342
@needs_read_lock
2343
def get_inventory(self, revision_id):
2344
"""Get Inventory object by revision id."""
2345
return self.iter_inventories([revision_id]).next()
2346
2347
def iter_inventories(self, revision_ids, ordering=None):
2348
"""Get many inventories by revision_ids.
2349
2350
This will buffer some or all of the texts used in constructing the
2351
inventories in memory, but will only parse a single inventory at a
2352
time.
2353
2354
:param revision_ids: The expected revision ids of the inventories.
2355
:param ordering: optional ordering, e.g. 'topological'. If not
2356
specified, the order of revision_ids will be preserved (by
2357
buffering if necessary).
2358
:return: An iterator of inventories.
2359
"""
2360
if ((None in revision_ids)
2361
or (_mod_revision.NULL_REVISION in revision_ids)):
2362
raise ValueError('cannot get null revision inventory')
2363
return self._iter_inventories(revision_ids, ordering)
2364
2365
def _iter_inventories(self, revision_ids, ordering):
2366
"""single-document based inventory iteration."""
2367
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2368
for text, revision_id in inv_xmls:
2369
yield self._deserialise_inventory(revision_id, text)
2370
2371
def _iter_inventory_xmls(self, revision_ids, ordering):
2372
if ordering is None:
2373
order_as_requested = True
2374
ordering = 'unordered'
2375
else:
2376
order_as_requested = False
2377
keys = [(revision_id,) for revision_id in revision_ids]
2378
if not keys:
2379
return
2380
if order_as_requested:
2381
key_iter = iter(keys)
2382
next_key = key_iter.next()
2383
stream = self.inventories.get_record_stream(keys, ordering, True)
2384
text_chunks = {}
2385
for record in stream:
2386
if record.storage_kind != 'absent':
2387
chunks = record.get_bytes_as('chunked')
2388
if order_as_requested:
2389
text_chunks[record.key] = chunks
2390
else:
2391
yield ''.join(chunks), record.key[-1]
2392
else:
2393
raise errors.NoSuchRevision(self, record.key)
2394
if order_as_requested:
2395
# Yield as many results as we can while preserving order.
2396
while next_key in text_chunks:
2397
chunks = text_chunks.pop(next_key)
2398
yield ''.join(chunks), next_key[-1]
2399
try:
2400
next_key = key_iter.next()
2401
except StopIteration:
2402
# We still want to fully consume the get_record_stream,
2403
# just in case it is not actually finished at this point
2404
next_key = None
2405
break
2406
2407
def _deserialise_inventory(self, revision_id, xml):
2408
"""Transform the xml into an inventory object.
2409
2410
:param revision_id: The expected revision id of the inventory.
2411
:param xml: A serialised inventory.
2412
"""
2413
result = self._serializer.read_inventory_from_string(xml, revision_id,
2414
entry_cache=self._inventory_entry_cache,
2415
return_from_cache=self._safe_to_return_from_cache)
2416
if result.revision_id != revision_id:
2417
raise AssertionError('revision id mismatch %s != %s' % (
2418
result.revision_id, revision_id))
2419
return result
2420
2421
def get_serializer_format(self):
2422
return self._serializer.format_num
2423
2424
@needs_read_lock
2425
def _get_inventory_xml(self, revision_id):
2426
"""Get serialized inventory as a string."""
2427
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2428
try:
2429
text, revision_id = texts.next()
2430
except StopIteration:
2431
raise errors.HistoryMissing(self, 'inventory', revision_id)
2432
return text
957
2433
958
2434
def get_rev_id_for_revno(self, revno, known_pair):
959
2435
"""Return the revision id of a revno, given a later (revno, revid)
990
2466
raise AssertionError('_iter_for_revno returned too much history')
991
2467
return (True, partial_history[-1])
992
2468
993
@symbol_versioning.deprecated_method(symbol_versioning.deprecated_in((2, 4, 0)))
994
2469
def iter_reverse_revision_history(self, revision_id):
995
2470
"""Iterate backwards through revision ids in the lefthand history
996
2471
1033
2508
1034
2509
`revision_id` may be NULL_REVISION for the empty tree revision.
1035
2510
"""
1036
raise NotImplementedError(self.revision_tree)
2511
revision_id = _mod_revision.ensure_null(revision_id)
2512
# TODO: refactor this to use an existing revision object
2513
# so we don't need to read it in twice.
2514
if revision_id == _mod_revision.NULL_REVISION:
2515
return RevisionTree(self, Inventory(root_id=None),
2516
_mod_revision.NULL_REVISION)
2517
else:
2518
inv = self.get_inventory(revision_id)
2519
return RevisionTree(self, inv, revision_id)
1037
2520
1038
2521
def revision_trees(self, revision_ids):
1039
2522
"""Return Trees for revisions in this repository.
1041
2524
:param revision_ids: a sequence of revision-ids;
1042
2525
a revision-id may not be None or 'null:'
1043
2526
"""
1044
raise NotImplementedError(self.revision_trees)
2527
inventories = self.iter_inventories(revision_ids)
2528
for inv in inventories:
2529
yield RevisionTree(self, inv, inv.revision_id)
2530
2531
def _filtered_revision_trees(self, revision_ids, file_ids):
2532
"""Return Tree for a revision on this branch with only some files.
2533
2534
:param revision_ids: a sequence of revision-ids;
2535
a revision-id may not be None or 'null:'
2536
:param file_ids: if not None, the result is filtered
2537
so that only those file-ids, their parents and their
2538
children are included.
2539
"""
2540
inventories = self.iter_inventories(revision_ids)
2541
for inv in inventories:
2542
# Should we introduce a FilteredRevisionTree class rather
2543
# than pre-filter the inventory here?
2544
filtered_inv = inv.filter(file_ids)
2545
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
1045
2546
1046
2547
@needs_read_lock
1047
@symbol_versioning.deprecated_method(
1048
symbol_versioning.deprecated_in((2, 4, 0)))
1049
2548
def get_ancestry(self, revision_id, topo_sorted=True):
1050
2549
"""Return a list of revision-ids integrated by a revision.
1051
2550
1055
2554
1056
2555
This is topologically sorted.
1057
2556
"""
1058
if 'evil' in debug.debug_flags:
1059
mutter_callsite(2, "get_ancestry is linear with history.")
1060
2557
if _mod_revision.is_null(revision_id):
1061
2558
return [None]
1062
2559
if not self.has_revision(revision_id):
1103
2600
1104
2601
def get_parent_map(self, revision_ids):
1105
2602
"""See graph.StackedParentsProvider.get_parent_map"""
1106
raise NotImplementedError(self.get_parent_map)
1107
1108
def _get_parent_map_no_fallbacks(self, revision_ids):
1109
"""Same as Repository.get_parent_map except doesn't query fallbacks."""
1110
2603
# revisions index works in keys; this just works in revisions
1111
2604
# therefore wrap and unwrap
1112
2605
query_keys = []
1118
2611
raise ValueError('get_parent_map(None) is not valid')
1119
2612
else:
1120
2613
query_keys.append((revision_id ,))
1121
vf = self.revisions.without_fallbacks()
1122
2614
for ((revision_id,), parent_keys) in \
1123
vf.get_parent_map(query_keys).iteritems():
2615
self.revisions.get_parent_map(query_keys).iteritems():
1124
2616
if parent_keys:
1125
2617
result[revision_id] = tuple([parent_revid
1126
2618
for (parent_revid,) in parent_keys])
1129
2621
return result
1130
2622
1131
2623
def _make_parents_provider(self):
1132
if not self._format.supports_external_lookups:
1133
return self
1134
return graph.StackedParentsProvider(_LazyListJoin(
1135
[self._make_parents_provider_unstacked()],
1136
self._fallback_repositories))
1137
1138
def _make_parents_provider_unstacked(self):
1139
return graph.CallableToParentsProviderAdapter(
1140
self._get_parent_map_no_fallbacks)
2624
return self
1141
2625
1142
2626
@needs_read_lock
1143
2627
def get_known_graph_ancestry(self, revision_ids):
1144
2628
"""Return the known graph for a set of revision ids and their ancestors.
1145
2629
"""
1146
raise NotImplementedError(self.get_known_graph_ancestry)
1147
1148
def get_file_graph(self):
1149
"""Return the graph walker for files."""
1150
raise NotImplementedError(self.get_file_graph)
2630
st = static_tuple.StaticTuple
2631
revision_keys = [st(r_id).intern() for r_id in revision_ids]
2632
known_graph = self.revisions.get_known_graph_ancestry(revision_keys)
2633
return graph.GraphThunkIdsToKeys(known_graph)
1151
2634
1152
2635
def get_graph(self, other_repository=None):
1153
2636
"""Return the graph walker for this repository format"""
1158
2641
[parents_provider, other_repository._make_parents_provider()])
1159
2642
return graph.Graph(parents_provider)
1160
2643
2644
def _get_versioned_file_checker(self, text_key_references=None,
2645
ancestors=None):
2646
"""Return an object suitable for checking versioned files.
2647
2648
:param text_key_references: if non-None, an already built
2649
dictionary mapping text keys ((fileid, revision_id) tuples)
2650
to whether they were referred to by the inventory of the
2651
revision_id that they contain. If None, this will be
2652
calculated.
2653
:param ancestors: Optional result from
2654
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2655
available.
2656
"""
2657
return _VersionedFileChecker(self,
2658
text_key_references=text_key_references, ancestors=ancestors)
2659
1161
2660
def revision_ids_to_search_result(self, result_set):
1162
2661
"""Convert a set of revision ids to a graph SearchResult."""
1163
2662
result_parents = set()
1189
2688
1190
2689
@needs_write_lock
1191
2690
def sign_revision(self, revision_id, gpg_strategy):
1192
testament = _mod_testament.Testament.from_revision(self, revision_id)
1193
plaintext = testament.as_short_text()
2691
plaintext = Testament.from_revision(self, revision_id).as_short_text()
1194
2692
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
1195
2693
1196
2694
@needs_read_lock
1197
def verify_revision(self, revision_id, gpg_strategy):
1198
"""Verify the signature on a revision.
1199
1200
:param revision_id: the revision to verify
1201
:gpg_strategy: the GPGStrategy object to used
1202
1203
:return: gpg.SIGNATURE_VALID or a failed SIGNATURE_ value
1204
"""
1205
if not self.has_signature_for_revision_id(revision_id):
1206
return gpg.SIGNATURE_NOT_SIGNED, None
1207
signature = self.get_signature_text(revision_id)
1208
1209
testament = _mod_testament.Testament.from_revision(self, revision_id)
1210
plaintext = testament.as_short_text()
1211
1212
return gpg_strategy.verify(signature, plaintext)
1213
1214
2695
def has_signature_for_revision_id(self, revision_id):
1215
2696
"""Query for a revision signature for revision_id in the repository."""
1216
raise NotImplementedError(self.has_signature_for_revision_id)
2697
if not self.has_revision(revision_id):
2698
raise errors.NoSuchRevision(self, revision_id)
2699
sig_present = (1 == len(
2700
self.signatures.get_parent_map([(revision_id,)])))
2701
return sig_present
1217
2702
2703
@needs_read_lock
1218
2704
def get_signature_text(self, revision_id):
1219
2705
"""Return the text for a signature."""
1220
raise NotImplementedError(self.get_signature_text)
2706
stream = self.signatures.get_record_stream([(revision_id,)],
2707
'unordered', True)
2708
record = stream.next()
2709
if record.storage_kind == 'absent':
2710
raise errors.NoSuchRevision(self, revision_id)
2711
return record.get_bytes_as('fulltext')
1221
2712
2713
@needs_read_lock
1222
2714
def check(self, revision_ids=None, callback_refs=None, check_repo=True):
1223
2715
"""Check consistency of all history of given revision_ids.
1224
2716
1232
2724
:param check_repo: If False do not check the repository contents, just
1233
2725
calculate the data callback_refs requires and call them back.
1234
2726
"""
1235
return self._check(revision_ids=revision_ids, callback_refs=callback_refs,
2727
return self._check(revision_ids, callback_refs=callback_refs,
1236
2728
check_repo=check_repo)
1237
2729
1238
def _check(self, revision_ids=None, callback_refs=None, check_repo=True):
1239
raise NotImplementedError(self.check)
2730
def _check(self, revision_ids, callback_refs, check_repo):
2731
result = check.Check(self, check_repo=check_repo)
2732
result.check(callback_refs)
2733
return result
1240
2734
1241
2735
def _warn_if_deprecated(self, branch=None):
1242
if not self._format.is_deprecated():
1243
return
1244
2736
global _deprecation_warning_done
1245
2737
if _deprecation_warning_done:
1246
2738
return
1276
2768
except UnicodeDecodeError:
1277
2769
raise errors.NonAsciiRevisionId(method, self)
1278
2770
2771
def revision_graph_can_have_wrong_parents(self):
2772
"""Is it possible for this repository to have a revision graph with
2773
incorrect parents?
2774
2775
If True, then this repository must also implement
2776
_find_inconsistent_revision_parents so that check and reconcile can
2777
check for inconsistencies before proceeding with other checks that may
2778
depend on the revision index being consistent.
2779
"""
2780
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2781
2782
2783
def install_revision(repository, rev, revision_tree):
2784
"""Install all revision data into a repository."""
2785
install_revisions(repository, [(rev, revision_tree, None)])
2786
2787
2788
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2789
"""Install all revision data into a repository.
2790
2791
Accepts an iterable of revision, tree, signature tuples. The signature
2792
may be None.
2793
"""
2794
repository.start_write_group()
2795
try:
2796
inventory_cache = lru_cache.LRUCache(10)
2797
for n, (revision, revision_tree, signature) in enumerate(iterable):
2798
_install_revision(repository, revision, revision_tree, signature,
2799
inventory_cache)
2800
if pb is not None:
2801
pb.update('Transferring revisions', n + 1, num_revisions)
2802
except:
2803
repository.abort_write_group()
2804
raise
2805
else:
2806
repository.commit_write_group()
2807
2808
2809
def _install_revision(repository, rev, revision_tree, signature,
2810
inventory_cache):
2811
"""Install all revision data into a repository."""
2812
present_parents = []
2813
parent_trees = {}
2814
for p_id in rev.parent_ids:
2815
if repository.has_revision(p_id):
2816
present_parents.append(p_id)
2817
parent_trees[p_id] = repository.revision_tree(p_id)
2818
else:
2819
parent_trees[p_id] = repository.revision_tree(
2820
_mod_revision.NULL_REVISION)
2821
2822
inv = revision_tree.inventory
2823
entries = inv.iter_entries()
2824
# backwards compatibility hack: skip the root id.
2825
if not repository.supports_rich_root():
2826
path, root = entries.next()
2827
if root.revision != rev.revision_id:
2828
raise errors.IncompatibleRevision(repr(repository))
2829
text_keys = {}
2830
for path, ie in entries:
2831
text_keys[(ie.file_id, ie.revision)] = ie
2832
text_parent_map = repository.texts.get_parent_map(text_keys)
2833
missing_texts = set(text_keys) - set(text_parent_map)
2834
# Add the texts that are not already present
2835
for text_key in missing_texts:
2836
ie = text_keys[text_key]
2837
text_parents = []
2838
# FIXME: TODO: The following loop overlaps/duplicates that done by
2839
# commit to determine parents. There is a latent/real bug here where
2840
# the parents inserted are not those commit would do - in particular
2841
# they are not filtered by heads(). RBC, AB
2842
for revision, tree in parent_trees.iteritems():
2843
if ie.file_id not in tree:
2844
continue
2845
parent_id = tree.inventory[ie.file_id].revision
2846
if parent_id in text_parents:
2847
continue
2848
text_parents.append((ie.file_id, parent_id))
2849
lines = revision_tree.get_file(ie.file_id).readlines()
2850
repository.texts.add_lines(text_key, text_parents, lines)
2851
try:
2852
# install the inventory
2853
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2854
# Cache this inventory
2855
inventory_cache[rev.revision_id] = inv
2856
try:
2857
basis_inv = inventory_cache[rev.parent_ids[0]]
2858
except KeyError:
2859
repository.add_inventory(rev.revision_id, inv, present_parents)
2860
else:
2861
delta = inv._make_delta(basis_inv)
2862
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2863
rev.revision_id, present_parents)
2864
else:
2865
repository.add_inventory(rev.revision_id, inv, present_parents)
2866
except errors.RevisionAlreadyPresent:
2867
pass
2868
if signature is not None:
2869
repository.add_signature_text(rev.revision_id, signature)
2870
repository.add_revision(rev.revision_id, rev, inv)
2871
1279
2872
1280
2873
class MetaDirRepository(Repository):
1281
2874
"""Repositories in the new meta-dir layout.
1316
2909
return not self._transport.has('no-working-trees')
1317
2910
1318
2911
2912
class MetaDirVersionedFileRepository(MetaDirRepository):
2913
"""Repositories in a meta-dir, that work via versioned file objects."""
2914
2915
def __init__(self, _format, a_bzrdir, control_files):
2916
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2917
control_files)
2918
2919
1319
2920
class RepositoryFormatRegistry(controldir.ControlComponentFormatRegistry):
1320
2921
"""Repository format registry."""
1321
2922
1386
2987
# Does this format support CHK bytestring lookups. Set to True or False in
1387
2988
# derived classes.
1388
2989
supports_chks = None
2990
# Should commit add an inventory, or an inventory delta to the repository.
2991
_commit_inv_deltas = True
2992
# What order should fetch operations request streams in?
2993
# The default is unordered as that is the cheapest for an origin to
2994
# provide.
2995
_fetch_order = 'unordered'
2996
# Does this repository format use deltas that can be fetched as-deltas ?
2997
# (E.g. knits, where the knit deltas can be transplanted intact.
2998
# We default to False, which will ensure that enough data to get
2999
# a full text out of any fetch stream will be grabbed.
3000
_fetch_uses_deltas = False
1389
3001
# Should fetch trigger a reconcile after the fetch? Only needed for
1390
3002
# some repository formats that can suffer internal inconsistencies.
1391
3003
_fetch_reconcile = False
1397
3009
# help), and for fetching when data won't have come from the same
1398
3010
# compressor.
1399
3011
pack_compresses = False
1400
# Does the repository storage understand references to trees?
3012
# Does the repository inventory storage understand references to trees?
1401
3013
supports_tree_reference = None
1402
3014
# Is the format experimental ?
1403
3015
experimental = False
1404
# Does this repository format escape funky characters, or does it create
1405
# files with similar names as the versioned files in its contents on disk
1406
# ?
1407
supports_funky_characters = None
3016
# Does this repository format escape funky characters, or does it create files with
3017
# similar names as the versioned files in its contents on disk ?
3018
supports_funky_characters = True
1408
3019
# Does this repository format support leaving locks?
1409
supports_leaving_lock = None
1410
# Does this format support the full VersionedFiles interface?
1411
supports_full_versioned_files = None
1412
# Does this format support signing revision signatures?
1413
supports_revision_signatures = True
1414
# Can the revision graph have incorrect parents?
1415
revision_graph_can_have_wrong_parents = None
1416
# Does this format support rich root data?
1417
rich_root_data = None
1418
# Does this format support explicitly versioned directories?
1419
supports_versioned_directories = None
1420
# Can other repositories be nested into one of this format?
1421
supports_nesting_repositories = None
3020
supports_leaving_lock = True
1422
3021
1423
3022
def __repr__(self):
1424
3023
return "%s()" % self.__class__.__name__
1476
3075
"""Return the short description for this format."""
1477
3076
raise NotImplementedError(self.get_format_description)
1478
3077
3078
# TODO: this shouldn't be in the base class, it's specific to things that
3079
# use weaves or knits -- mbp 20070207
3080
def _get_versioned_file_store(self,
3081
name,
3082
transport,
3083
control_files,
3084
prefixed=True,
3085
versionedfile_class=None,
3086
versionedfile_kwargs={},
3087
escaped=False):
3088
if versionedfile_class is None:
3089
versionedfile_class = self._versionedfile_class
3090
weave_transport = control_files._transport.clone(name)
3091
dir_mode = control_files._dir_mode
3092
file_mode = control_files._file_mode
3093
return VersionedFileStore(weave_transport, prefixed=prefixed,
3094
dir_mode=dir_mode,
3095
file_mode=file_mode,
3096
versionedfile_class=versionedfile_class,
3097
versionedfile_kwargs=versionedfile_kwargs,
3098
escaped=escaped)
3099
1479
3100
def initialize(self, a_bzrdir, shared=False):
1480
3101
"""Initialize a repository of this format in a_bzrdir.
1481
3102
1497
3118
"""
1498
3119
return True
1499
3120
1500
def is_deprecated(self):
1501
"""Is this format deprecated?
1502
1503
Deprecated formats may trigger a user-visible warning recommending
1504
the user to upgrade. They are still fully supported.
1505
"""
1506
return False
1507
1508
3121
def network_name(self):
1509
3122
"""A simple byte string uniquely identifying this format for RPC calls.
1510
3123
1550
3163
supports_tree_reference = False
1551
3164
supports_external_lookups = False
1552
3165
supports_leaving_lock = True
1553
supports_nesting_repositories = True
1554
3166
1555
3167
@property
1556
3168
def _matchingbzrdir(self):
1594
3206
return self.get_format_string()
1595
3207
1596
3208
3209
# Pre-0.8 formats that don't have a disk format string (because they are
3210
# versioned by the matching control directory). We use the control directories
3211
# disk format string as a key for the network_name because they meet the
3212
# constraints (simple string, unique, immutable).
3213
network_format_registry.register_lazy(
3214
"Bazaar-NG branch, format 5\n",
3215
'bzrlib.repofmt.weaverepo',
3216
'RepositoryFormat5',
3217
)
3218
network_format_registry.register_lazy(
3219
"Bazaar-NG branch, format 6\n",
3220
'bzrlib.repofmt.weaverepo',
3221
'RepositoryFormat6',
3222
)
3223
3224
format_registry.register_extra_lazy(
3225
'bzrlib.repofmt.weaverepo',
3226
'RepositoryFormat4')
3227
format_registry.register_extra_lazy(
3228
'bzrlib.repofmt.weaverepo',
3229
'RepositoryFormat5')
3230
format_registry.register_extra_lazy(
3231
'bzrlib.repofmt.weaverepo',
3232
'RepositoryFormat6')
3233
1597
3234
# formats which have no format string are not discoverable or independently
1598
3235
# creatable on disk, so are not registered in format_registry. They're
1599
# all in bzrlib.repofmt.knitreponow. When an instance of one of these is
3236
# all in bzrlib.repofmt.weaverepo now. When an instance of one of these is
1600
3237
# needed, it's constructed directly by the BzrDir. Non-native formats where
1601
3238
# the repository is not separately opened are similar.
1602
3239
1603
3240
format_registry.register_lazy(
3241
'Bazaar-NG Repository format 7',
3242
'bzrlib.repofmt.weaverepo',
3243
'RepositoryFormat7'
3244
)
3245
3246
format_registry.register_lazy(
1604
3247
'Bazaar-NG Knit Repository Format 1',
1605
3248
'bzrlib.repofmt.knitrepo',
1606
3249
'RepositoryFormatKnit1',
1623
3266
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
1624
3267
format_registry.register_lazy(
1625
3268
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
1626
'bzrlib.repofmt.knitpack_repo',
3269
'bzrlib.repofmt.pack_repo',
1627
3270
'RepositoryFormatKnitPack1',
1628
3271
)
1629
3272
format_registry.register_lazy(
1630
3273
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
1631
'bzrlib.repofmt.knitpack_repo',
3274
'bzrlib.repofmt.pack_repo',
1632
3275
'RepositoryFormatKnitPack3',
1633
3276
)
1634
3277
format_registry.register_lazy(
1635
3278
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
1636
'bzrlib.repofmt.knitpack_repo',
3279
'bzrlib.repofmt.pack_repo',
1637
3280
'RepositoryFormatKnitPack4',
1638
3281
)
1639
3282
format_registry.register_lazy(
1640
3283
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
1641
'bzrlib.repofmt.knitpack_repo',
3284
'bzrlib.repofmt.pack_repo',
1642
3285
'RepositoryFormatKnitPack5',
1643
3286
)
1644
3287
format_registry.register_lazy(
1645
3288
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
1646
'bzrlib.repofmt.knitpack_repo',
3289
'bzrlib.repofmt.pack_repo',
1647
3290
'RepositoryFormatKnitPack5RichRoot',
1648
3291
)
1649
3292
format_registry.register_lazy(
1650
3293
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
1651
'bzrlib.repofmt.knitpack_repo',
3294
'bzrlib.repofmt.pack_repo',
1652
3295
'RepositoryFormatKnitPack5RichRootBroken',
1653
3296
)
1654
3297
format_registry.register_lazy(
1655
3298
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
1656
'bzrlib.repofmt.knitpack_repo',
3299
'bzrlib.repofmt.pack_repo',
1657
3300
'RepositoryFormatKnitPack6',
1658
3301
)
1659
3302
format_registry.register_lazy(
1660
3303
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
1661
'bzrlib.repofmt.knitpack_repo',
3304
'bzrlib.repofmt.pack_repo',
1662
3305
'RepositoryFormatKnitPack6RichRoot',
1663
3306
)
1664
3307
format_registry.register_lazy(
1672
3315
format_registry.register_lazy(
1673
3316
("Bazaar development format 2 with subtree support "
1674
3317
"(needs bzr.dev from before 1.8)\n"),
1675
'bzrlib.repofmt.knitpack_repo',
3318
'bzrlib.repofmt.pack_repo',
1676
3319
'RepositoryFormatPackDevelopment2Subtree',
1677
3320
)
1678
3321
format_registry.register_lazy(
1694
3337
InterRepository.get(other).method_name(parameters).
1695
3338
"""
1696
3339
3340
_walk_to_common_revisions_batch_size = 50
1697
3341
_optimisers = []
1698
3342
"""The available optimised InterRepository types."""
1699
3343
1724
3368
content is copied.
1725
3369
:return: None.
1726
3370
"""
1727
raise NotImplementedError(self.fetch)
3371
ui.ui_factory.warn_experimental_format_fetch(self)
3372
from bzrlib.fetch import RepoFetcher
3373
# See <https://launchpad.net/bugs/456077> asking for a warning here
3374
if self.source._format.network_name() != self.target._format.network_name():
3375
ui.ui_factory.show_user_warning('cross_format_fetch',
3376
from_format=self.source._format,
3377
to_format=self.target._format)
3378
f = RepoFetcher(to_repository=self.target,
3379
from_repository=self.source,
3380
last_revision=revision_id,
3381
fetch_spec=fetch_spec,
3382
find_ghosts=find_ghosts)
3383
3384
def _walk_to_common_revisions(self, revision_ids, if_present_ids=None):
3385
"""Walk out from revision_ids in source to revisions target has.
3386
3387
:param revision_ids: The start point for the search.
3388
:return: A set of revision ids.
3389
"""
3390
target_graph = self.target.get_graph()
3391
revision_ids = frozenset(revision_ids)
3392
if if_present_ids:
3393
all_wanted_revs = revision_ids.union(if_present_ids)
3394
else:
3395
all_wanted_revs = revision_ids
3396
missing_revs = set()
3397
source_graph = self.source.get_graph()
3398
# ensure we don't pay silly lookup costs.
3399
searcher = source_graph._make_breadth_first_searcher(all_wanted_revs)
3400
null_set = frozenset([_mod_revision.NULL_REVISION])
3401
searcher_exhausted = False
3402
while True:
3403
next_revs = set()
3404
ghosts = set()
3405
# Iterate the searcher until we have enough next_revs
3406
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3407
try:
3408
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3409
next_revs.update(next_revs_part)
3410
ghosts.update(ghosts_part)
3411
except StopIteration:
3412
searcher_exhausted = True
3413
break
3414
# If there are ghosts in the source graph, and the caller asked for
3415
# them, make sure that they are present in the target.
3416
# We don't care about other ghosts as we can't fetch them and
3417
# haven't been asked to.
3418
ghosts_to_check = set(revision_ids.intersection(ghosts))
3419
revs_to_get = set(next_revs).union(ghosts_to_check)
3420
if revs_to_get:
3421
have_revs = set(target_graph.get_parent_map(revs_to_get))
3422
# we always have NULL_REVISION present.
3423
have_revs = have_revs.union(null_set)
3424
# Check if the target is missing any ghosts we need.
3425
ghosts_to_check.difference_update(have_revs)
3426
if ghosts_to_check:
3427
# One of the caller's revision_ids is a ghost in both the
3428
# source and the target.
3429
raise errors.NoSuchRevision(
3430
self.source, ghosts_to_check.pop())
3431
missing_revs.update(next_revs - have_revs)
3432
# Because we may have walked past the original stop point, make
3433
# sure everything is stopped
3434
stop_revs = searcher.find_seen_ancestors(have_revs)
3435
searcher.stop_searching_any(stop_revs)
3436
if searcher_exhausted:
3437
break
3438
return searcher.get_result()
1728
3439
1729
3440
@needs_read_lock
1730
3441
def search_missing_revision_ids(self,
1731
3442
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
1732
find_ghosts=True, revision_ids=None, if_present_ids=None,
1733
limit=None):
3443
find_ghosts=True, revision_ids=None, if_present_ids=None):
1734
3444
"""Return the revision ids that source has that target does not.
1735
3445
1736
3446
:param revision_id: only return revision ids included by this
1744
3454
to fetch for tags, which may reference absent revisions.
1745
3455
:param find_ghosts: If True find missing revisions in deep history
1746
3456
rather than just finding the surface difference.
1747
:param limit: Maximum number of revisions to return, topologically
1748
ordered
1749
3457
:return: A bzrlib.graph.SearchResult.
1750
3458
"""
1751
raise NotImplementedError(self.search_missing_revision_ids)
3459
if symbol_versioning.deprecated_passed(revision_id):
3460
symbol_versioning.warn(
3461
'search_missing_revision_ids(revision_id=...) was '
3462
'deprecated in 2.4. Use revision_ids=[...] instead.',
3463
DeprecationWarning, stacklevel=2)
3464
if revision_ids is not None:
3465
raise AssertionError(
3466
'revision_ids is mutually exclusive with revision_id')
3467
if revision_id is not None:
3468
revision_ids = [revision_id]
3469
del revision_id
3470
# stop searching at found target revisions.
3471
if not find_ghosts and (revision_ids is not None or if_present_ids is
3472
not None):
3473
return self._walk_to_common_revisions(revision_ids,
3474
if_present_ids=if_present_ids)
3475
# generic, possibly worst case, slow code path.
3476
target_ids = set(self.target.all_revision_ids())
3477
source_ids = self._present_source_revisions_for(
3478
revision_ids, if_present_ids)
3479
result_set = set(source_ids).difference(target_ids)
3480
return self.source.revision_ids_to_search_result(result_set)
3481
3482
def _present_source_revisions_for(self, revision_ids, if_present_ids=None):
3483
"""Returns set of all revisions in ancestry of revision_ids present in
3484
the source repo.
3485
3486
:param revision_ids: if None, all revisions in source are returned.
3487
:param if_present_ids: like revision_ids, but if any/all of these are
3488
absent no error is raised.
3489
"""
3490
if revision_ids is not None or if_present_ids is not None:
3491
# First, ensure all specified revisions exist. Callers expect
3492
# NoSuchRevision when they pass absent revision_ids here.
3493
if revision_ids is None:
3494
revision_ids = set()
3495
if if_present_ids is None:
3496
if_present_ids = set()
3497
revision_ids = set(revision_ids)
3498
if_present_ids = set(if_present_ids)
3499
all_wanted_ids = revision_ids.union(if_present_ids)
3500
graph = self.source.get_graph()
3501
present_revs = set(graph.get_parent_map(all_wanted_ids))
3502
missing = revision_ids.difference(present_revs)
3503
if missing:
3504
raise errors.NoSuchRevision(self.source, missing.pop())
3505
found_ids = all_wanted_ids.intersection(present_revs)
3506
source_ids = [rev_id for (rev_id, parents) in
3507
graph.iter_ancestry(found_ids)
3508
if rev_id != _mod_revision.NULL_REVISION
3509
and parents is not None]
3510
else:
3511
source_ids = self.source.all_revision_ids()
3512
return set(source_ids)
1752
3513
1753
3514
@staticmethod
1754
3515
def _same_model(source, target):
1775
3536
"different serializers")
1776
3537
1777
3538
3539
class InterSameDataRepository(InterRepository):
3540
"""Code for converting between repositories that represent the same data.
3541
3542
Data format and model must match for this to work.
3543
"""
3544
3545
@classmethod
3546
def _get_repo_format_to_test(self):
3547
"""Repository format for testing with.
3548
3549
InterSameData can pull from subtree to subtree and from non-subtree to
3550
non-subtree, so we test this with the richest repository format.
3551
"""
3552
from bzrlib.repofmt import knitrepo
3553
return knitrepo.RepositoryFormatKnit3()
3554
3555
@staticmethod
3556
def is_compatible(source, target):
3557
return InterRepository._same_model(source, target)
3558
3559
3560
class InterDifferingSerializer(InterRepository):
3561
3562
@classmethod
3563
def _get_repo_format_to_test(self):
3564
return None
3565
3566
@staticmethod
3567
def is_compatible(source, target):
3568
"""Be compatible with Knit2 source and Knit3 target"""
3569
# This is redundant with format.check_conversion_target(), however that
3570
# raises an exception, and we just want to say "False" as in we won't
3571
# support converting between these formats.
3572
if 'IDS_never' in debug.debug_flags:
3573
return False
3574
if source.supports_rich_root() and not target.supports_rich_root():
3575
return False
3576
if (source._format.supports_tree_reference
3577
and not target._format.supports_tree_reference):
3578
return False
3579
if target._fallback_repositories and target._format.supports_chks:
3580
# IDS doesn't know how to copy CHKs for the parent inventories it
3581
# adds to stacked repos.
3582
return False
3583
if 'IDS_always' in debug.debug_flags:
3584
return True
3585
# Only use this code path for local source and target. IDS does far
3586
# too much IO (both bandwidth and roundtrips) over a network.
3587
if not source.bzrdir.transport.base.startswith('file:///'):
3588
return False
3589
if not target.bzrdir.transport.base.startswith('file:///'):
3590
return False
3591
return True
3592
3593
def _get_trees(self, revision_ids, cache):
3594
possible_trees = []
3595
for rev_id in revision_ids:
3596
if rev_id in cache:
3597
possible_trees.append((rev_id, cache[rev_id]))
3598
else:
3599
# Not cached, but inventory might be present anyway.
3600
try:
3601
tree = self.source.revision_tree(rev_id)
3602
except errors.NoSuchRevision:
3603
# Nope, parent is ghost.
3604
pass
3605
else:
3606
cache[rev_id] = tree
3607
possible_trees.append((rev_id, tree))
3608
return possible_trees
3609
3610
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3611
"""Get the best delta and base for this revision.
3612
3613
:return: (basis_id, delta)
3614
"""
3615
deltas = []
3616
# Generate deltas against each tree, to find the shortest.
3617
texts_possibly_new_in_tree = set()
3618
for basis_id, basis_tree in possible_trees:
3619
delta = tree.inventory._make_delta(basis_tree.inventory)
3620
for old_path, new_path, file_id, new_entry in delta:
3621
if new_path is None:
3622
# This file_id isn't present in the new rev, so we don't
3623
# care about it.
3624
continue
3625
if not new_path:
3626
# Rich roots are handled elsewhere...
3627
continue
3628
kind = new_entry.kind
3629
if kind != 'directory' and kind != 'file':
3630
# No text record associated with this inventory entry.
3631
continue
3632
# This is a directory or file that has changed somehow.
3633
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3634
deltas.append((len(delta), basis_id, delta))
3635
deltas.sort()
3636
return deltas[0][1:]
3637
3638
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3639
"""Find all parent revisions that are absent, but for which the
3640
inventory is present, and copy those inventories.
3641
3642
This is necessary to preserve correctness when the source is stacked
3643
without fallbacks configured. (Note that in cases like upgrade the
3644
source may be not have _fallback_repositories even though it is
3645
stacked.)
3646
"""
3647
parent_revs = set()
3648
for parents in parent_map.values():
3649
parent_revs.update(parents)
3650
present_parents = self.source.get_parent_map(parent_revs)
3651
absent_parents = set(parent_revs).difference(present_parents)
3652
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3653
(rev_id,) for rev_id in absent_parents)
3654
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3655
for parent_tree in self.source.revision_trees(parent_inv_ids):
3656
current_revision_id = parent_tree.get_revision_id()
3657
parents_parents_keys = parent_invs_keys_for_stacking[
3658
(current_revision_id,)]
3659
parents_parents = [key[-1] for key in parents_parents_keys]
3660
basis_id = _mod_revision.NULL_REVISION
3661
basis_tree = self.source.revision_tree(basis_id)
3662
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3663
self.target.add_inventory_by_delta(
3664
basis_id, delta, current_revision_id, parents_parents)
3665
cache[current_revision_id] = parent_tree
3666
3667
def _fetch_batch(self, revision_ids, basis_id, cache):
3668
"""Fetch across a few revisions.
3669
3670
:param revision_ids: The revisions to copy
3671
:param basis_id: The revision_id of a tree that must be in cache, used
3672
as a basis for delta when no other base is available
3673
:param cache: A cache of RevisionTrees that we can use.
3674
:return: The revision_id of the last converted tree. The RevisionTree
3675
for it will be in cache
3676
"""
3677
# Walk though all revisions; get inventory deltas, copy referenced
3678
# texts that delta references, insert the delta, revision and
3679
# signature.
3680
root_keys_to_create = set()
3681
text_keys = set()
3682
pending_deltas = []
3683
pending_revisions = []
3684
parent_map = self.source.get_parent_map(revision_ids)
3685
self._fetch_parent_invs_for_stacking(parent_map, cache)
3686
self.source._safe_to_return_from_cache = True
3687
for tree in self.source.revision_trees(revision_ids):
3688
# Find a inventory delta for this revision.
3689
# Find text entries that need to be copied, too.
3690
current_revision_id = tree.get_revision_id()
3691
parent_ids = parent_map.get(current_revision_id, ())
3692
parent_trees = self._get_trees(parent_ids, cache)
3693
possible_trees = list(parent_trees)
3694
if len(possible_trees) == 0:
3695
# There either aren't any parents, or the parents are ghosts,
3696
# so just use the last converted tree.
3697
possible_trees.append((basis_id, cache[basis_id]))
3698
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3699
possible_trees)
3700
revision = self.source.get_revision(current_revision_id)
3701
pending_deltas.append((basis_id, delta,
3702
current_revision_id, revision.parent_ids))
3703
if self._converting_to_rich_root:
3704
self._revision_id_to_root_id[current_revision_id] = \
3705
tree.get_root_id()
3706
# Determine which texts are in present in this revision but not in
3707
# any of the available parents.
3708
texts_possibly_new_in_tree = set()
3709
for old_path, new_path, file_id, entry in delta:
3710
if new_path is None:
3711
# This file_id isn't present in the new rev
3712
continue
3713
if not new_path:
3714
# This is the root
3715
if not self.target.supports_rich_root():
3716
# The target doesn't support rich root, so we don't
3717
# copy
3718
continue
3719
if self._converting_to_rich_root:
3720
# This can't be copied normally, we have to insert
3721
# it specially
3722
root_keys_to_create.add((file_id, entry.revision))
3723
continue
3724
kind = entry.kind
3725
texts_possibly_new_in_tree.add((file_id, entry.revision))
3726
for basis_id, basis_tree in possible_trees:
3727
basis_inv = basis_tree.inventory
3728
for file_key in list(texts_possibly_new_in_tree):
3729
file_id, file_revision = file_key
3730
try:
3731
entry = basis_inv[file_id]
3732
except errors.NoSuchId:
3733
continue
3734
if entry.revision == file_revision:
3735
texts_possibly_new_in_tree.remove(file_key)
3736
text_keys.update(texts_possibly_new_in_tree)
3737
pending_revisions.append(revision)
3738
cache[current_revision_id] = tree
3739
basis_id = current_revision_id
3740
self.source._safe_to_return_from_cache = False
3741
# Copy file texts
3742
from_texts = self.source.texts
3743
to_texts = self.target.texts
3744
if root_keys_to_create:
3745
root_stream = _mod_fetch._new_root_data_stream(
3746
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3747
self.source)
3748
to_texts.insert_record_stream(root_stream)
3749
to_texts.insert_record_stream(from_texts.get_record_stream(
3750
text_keys, self.target._format._fetch_order,
3751
not self.target._format._fetch_uses_deltas))
3752
# insert inventory deltas
3753
for delta in pending_deltas:
3754
self.target.add_inventory_by_delta(*delta)
3755
if self.target._fallback_repositories:
3756
# Make sure this stacked repository has all the parent inventories
3757
# for the new revisions that we are about to insert. We do this
3758
# before adding the revisions so that no revision is added until
3759
# all the inventories it may depend on are added.
3760
# Note that this is overzealous, as we may have fetched these in an
3761
# earlier batch.
3762
parent_ids = set()
3763
revision_ids = set()
3764
for revision in pending_revisions:
3765
revision_ids.add(revision.revision_id)
3766
parent_ids.update(revision.parent_ids)
3767
parent_ids.difference_update(revision_ids)
3768
parent_ids.discard(_mod_revision.NULL_REVISION)
3769
parent_map = self.source.get_parent_map(parent_ids)
3770
# we iterate over parent_map and not parent_ids because we don't
3771
# want to try copying any revision which is a ghost
3772
for parent_tree in self.source.revision_trees(parent_map):
3773
current_revision_id = parent_tree.get_revision_id()
3774
parents_parents = parent_map[current_revision_id]
3775
possible_trees = self._get_trees(parents_parents, cache)
3776
if len(possible_trees) == 0:
3777
# There either aren't any parents, or the parents are
3778
# ghosts, so just use the last converted tree.
3779
possible_trees.append((basis_id, cache[basis_id]))
3780
basis_id, delta = self._get_delta_for_revision(parent_tree,
3781
parents_parents, possible_trees)
3782
self.target.add_inventory_by_delta(
3783
basis_id, delta, current_revision_id, parents_parents)
3784
# insert signatures and revisions
3785
for revision in pending_revisions:
3786
try:
3787
signature = self.source.get_signature_text(
3788
revision.revision_id)
3789
self.target.add_signature_text(revision.revision_id,
3790
signature)
3791
except errors.NoSuchRevision:
3792
pass
3793
self.target.add_revision(revision.revision_id, revision)
3794
return basis_id
3795
3796
def _fetch_all_revisions(self, revision_ids, pb):
3797
"""Fetch everything for the list of revisions.
3798
3799
:param revision_ids: The list of revisions to fetch. Must be in
3800
topological order.
3801
:param pb: A ProgressTask
3802
:return: None
3803
"""
3804
basis_id, basis_tree = self._get_basis(revision_ids[0])
3805
batch_size = 100
3806
cache = lru_cache.LRUCache(100)
3807
cache[basis_id] = basis_tree
3808
del basis_tree # We don't want to hang on to it here
3809
hints = []
3810
a_graph = None
3811
3812
for offset in range(0, len(revision_ids), batch_size):
3813
self.target.start_write_group()
3814
try:
3815
pb.update('Transferring revisions', offset,
3816
len(revision_ids))
3817
batch = revision_ids[offset:offset+batch_size]
3818
basis_id = self._fetch_batch(batch, basis_id, cache)
3819
except:
3820
self.source._safe_to_return_from_cache = False
3821
self.target.abort_write_group()
3822
raise
3823
else:
3824
hint = self.target.commit_write_group()
3825
if hint:
3826
hints.extend(hint)
3827
if hints and self.target._format.pack_compresses:
3828
self.target.pack(hint=hints)
3829
pb.update('Transferring revisions', len(revision_ids),
3830
len(revision_ids))
3831
3832
@needs_write_lock
3833
def fetch(self, revision_id=None, find_ghosts=False,
3834
fetch_spec=None):
3835
"""See InterRepository.fetch()."""
3836
if fetch_spec is not None:
3837
revision_ids = fetch_spec.get_keys()
3838
else:
3839
revision_ids = None
3840
ui.ui_factory.warn_experimental_format_fetch(self)
3841
if (not self.source.supports_rich_root()
3842
and self.target.supports_rich_root()):
3843
self._converting_to_rich_root = True
3844
self._revision_id_to_root_id = {}
3845
else:
3846
self._converting_to_rich_root = False
3847
# See <https://launchpad.net/bugs/456077> asking for a warning here
3848
if self.source._format.network_name() != self.target._format.network_name():
3849
ui.ui_factory.show_user_warning('cross_format_fetch',
3850
from_format=self.source._format,
3851
to_format=self.target._format)
3852
if revision_ids is None:
3853
if revision_id:
3854
search_revision_ids = [revision_id]
3855
else:
3856
search_revision_ids = None
3857
revision_ids = self.target.search_missing_revision_ids(self.source,
3858
revision_ids=search_revision_ids,
3859
find_ghosts=find_ghosts).get_keys()
3860
if not revision_ids:
3861
return 0, 0
3862
revision_ids = tsort.topo_sort(
3863
self.source.get_graph().get_parent_map(revision_ids))
3864
if not revision_ids:
3865
return 0, 0
3866
# Walk though all revisions; get inventory deltas, copy referenced
3867
# texts that delta references, insert the delta, revision and
3868
# signature.
3869
pb = ui.ui_factory.nested_progress_bar()
3870
try:
3871
self._fetch_all_revisions(revision_ids, pb)
3872
finally:
3873
pb.finished()
3874
return len(revision_ids), 0
3875
3876
def _get_basis(self, first_revision_id):
3877
"""Get a revision and tree which exists in the target.
3878
3879
This assumes that first_revision_id is selected for transmission
3880
because all other ancestors are already present. If we can't find an
3881
ancestor we fall back to NULL_REVISION since we know that is safe.
3882
3883
:return: (basis_id, basis_tree)
3884
"""
3885
first_rev = self.source.get_revision(first_revision_id)
3886
try:
3887
basis_id = first_rev.parent_ids[0]
3888
# only valid as a basis if the target has it
3889
self.target.get_revision(basis_id)
3890
# Try to get a basis tree - if it's a ghost it will hit the
3891
# NoSuchRevision case.
3892
basis_tree = self.source.revision_tree(basis_id)
3893
except (IndexError, errors.NoSuchRevision):
3894
basis_id = _mod_revision.NULL_REVISION
3895
basis_tree = self.source.revision_tree(basis_id)
3896
return basis_id, basis_tree
3897
3898
3899
InterRepository.register_optimiser(InterDifferingSerializer)
3900
InterRepository.register_optimiser(InterSameDataRepository)
3901
3902
1778
3903
class CopyConverter(object):
1779
3904
"""A repository conversion tool which just performs a copy of the content.
1780
3905
1801
3926
# trigger an assertion if not such
1802
3927
repo._format.get_format_string()
1803
3928
self.repo_dir = repo.bzrdir
1804
pb.update(gettext('Moving repository to repository.backup'))
3929
pb.update('Moving repository to repository.backup')
1805
3930
self.repo_dir.transport.move('repository', 'repository.backup')
1806
3931
backup_transport = self.repo_dir.transport.clone('repository.backup')
1807
3932
repo._format.check_conversion_target(self.target_format)
1808
3933
self.source_repo = repo._format.open(self.repo_dir,
1809
3934
_found=True,
1810
3935
_override_transport=backup_transport)
1811
pb.update(gettext('Creating new repository'))
3936
pb.update('Creating new repository')
1812
3937
converted = self.target_format.initialize(self.repo_dir,
1813
3938
self.source_repo.is_shared())
1814
3939
converted.lock_write()
1815
3940
try:
1816
pb.update(gettext('Copying content'))
3941
pb.update('Copying content')
1817
3942
self.source_repo.copy_content_into(converted)
1818
3943
finally:
1819
3944
converted.unlock()
1820
pb.update(gettext('Deleting old repository content'))
3945
pb.update('Deleting old repository content')
1821
3946
self.repo_dir.transport.delete_tree('repository.backup')
1822
ui.ui_factory.note(gettext('repository converted'))
3947
ui.ui_factory.note('repository converted')
1823
3948
pb.finished()
1824
3949
1825
3950
3951
class _VersionedFileChecker(object):
3952
3953
def __init__(self, repository, text_key_references=None, ancestors=None):
3954
self.repository = repository
3955
self.text_index = self.repository._generate_text_key_index(
3956
text_key_references=text_key_references, ancestors=ancestors)
3957
3958
def calculate_file_version_parents(self, text_key):
3959
"""Calculate the correct parents for a file version according to
3960
the inventories.
3961
"""
3962
parent_keys = self.text_index[text_key]
3963
if parent_keys == [_mod_revision.NULL_REVISION]:
3964
return ()
3965
return tuple(parent_keys)
3966
3967
def check_file_version_parents(self, texts, progress_bar=None):
3968
"""Check the parents stored in a versioned file are correct.
3969
3970
It also detects file versions that are not referenced by their
3971
corresponding revision's inventory.
3972
3973
:returns: A tuple of (wrong_parents, dangling_file_versions).
3974
wrong_parents is a dict mapping {revision_id: (stored_parents,
3975
correct_parents)} for each revision_id where the stored parents
3976
are not correct. dangling_file_versions is a set of (file_id,
3977
revision_id) tuples for versions that are present in this versioned
3978
file, but not used by the corresponding inventory.
3979
"""
3980
local_progress = None
3981
if progress_bar is None:
3982
local_progress = ui.ui_factory.nested_progress_bar()
3983
progress_bar = local_progress
3984
try:
3985
return self._check_file_version_parents(texts, progress_bar)
3986
finally:
3987
if local_progress:
3988
local_progress.finished()
3989
3990
def _check_file_version_parents(self, texts, progress_bar):
3991
"""See check_file_version_parents."""
3992
wrong_parents = {}
3993
self.file_ids = set([file_id for file_id, _ in
3994
self.text_index.iterkeys()])
3995
# text keys is now grouped by file_id
3996
n_versions = len(self.text_index)
3997
progress_bar.update('loading text store', 0, n_versions)
3998
parent_map = self.repository.texts.get_parent_map(self.text_index)
3999
# On unlistable transports this could well be empty/error...
4000
text_keys = self.repository.texts.keys()
4001
unused_keys = frozenset(text_keys) - set(self.text_index)
4002
for num, key in enumerate(self.text_index.iterkeys()):
4003
progress_bar.update('checking text graph', num, n_versions)
4004
correct_parents = self.calculate_file_version_parents(key)
4005
try:
4006
knit_parents = parent_map[key]
4007
except errors.RevisionNotPresent:
4008
# Missing text!
4009
knit_parents = None
4010
if correct_parents != knit_parents:
4011
wrong_parents[key] = (knit_parents, correct_parents)
4012
return wrong_parents, unused_keys
4013
4014
4015
def _old_get_graph(repository, revision_id):
4016
"""DO NOT USE. That is all. I'm serious."""
4017
graph = repository.get_graph()
4018
revision_graph = dict(((key, value) for key, value in
4019
graph.iter_ancestry([revision_id]) if value is not None))
4020
return _strip_NULL_ghosts(revision_graph)
4021
4022
1826
4023
def _strip_NULL_ghosts(revision_graph):
1827
4024
"""Also don't use this. more compatibility code for unmigrated clients."""
1828
4025
# Filter ghosts, and null:
1834
4031
return revision_graph
1835
4032
1836
4033
4034
class StreamSink(object):
4035
"""An object that can insert a stream into a repository.
4036
4037
This interface handles the complexity of reserialising inventories and
4038
revisions from different formats, and allows unidirectional insertion into
4039
stacked repositories without looking for the missing basis parents
4040
beforehand.
4041
"""
4042
4043
def __init__(self, target_repo):
4044
self.target_repo = target_repo
4045
4046
def insert_stream(self, stream, src_format, resume_tokens):
4047
"""Insert a stream's content into the target repository.
4048
4049
:param src_format: a bzr repository format.
4050
4051
:return: a list of resume tokens and an iterable of keys additional
4052
items required before the insertion can be completed.
4053
"""
4054
self.target_repo.lock_write()
4055
try:
4056
if resume_tokens:
4057
self.target_repo.resume_write_group(resume_tokens)
4058
is_resume = True
4059
else:
4060
self.target_repo.start_write_group()
4061
is_resume = False
4062
try:
4063
# locked_insert_stream performs a commit|suspend.
4064
missing_keys = self.insert_stream_without_locking(stream,
4065
src_format, is_resume)
4066
if missing_keys:
4067
# suspend the write group and tell the caller what we is
4068
# missing. We know we can suspend or else we would not have
4069
# entered this code path. (All repositories that can handle
4070
# missing keys can handle suspending a write group).
4071
write_group_tokens = self.target_repo.suspend_write_group()
4072
return write_group_tokens, missing_keys
4073
hint = self.target_repo.commit_write_group()
4074
to_serializer = self.target_repo._format._serializer
4075
src_serializer = src_format._serializer
4076
if (to_serializer != src_serializer and
4077
self.target_repo._format.pack_compresses):
4078
self.target_repo.pack(hint=hint)
4079
return [], set()
4080
except:
4081
self.target_repo.abort_write_group(suppress_errors=True)
4082
raise
4083
finally:
4084
self.target_repo.unlock()
4085
4086
def insert_stream_without_locking(self, stream, src_format,
4087
is_resume=False):
4088
"""Insert a stream's content into the target repository.
4089
4090
This assumes that you already have a locked repository and an active
4091
write group.
4092
4093
:param src_format: a bzr repository format.
4094
:param is_resume: Passed down to get_missing_parent_inventories to
4095
indicate if we should be checking for missing texts at the same
4096
time.
4097
4098
:return: A set of keys that are missing.
4099
"""
4100
if not self.target_repo.is_write_locked():
4101
raise errors.ObjectNotLocked(self)
4102
if not self.target_repo.is_in_write_group():
4103
raise errors.BzrError('you must already be in a write group')
4104
to_serializer = self.target_repo._format._serializer
4105
src_serializer = src_format._serializer
4106
new_pack = None
4107
if to_serializer == src_serializer:
4108
# If serializers match and the target is a pack repository, set the
4109
# write cache size on the new pack. This avoids poor performance
4110
# on transports where append is unbuffered (such as
4111
# RemoteTransport). This is safe to do because nothing should read
4112
# back from the target repository while a stream with matching
4113
# serialization is being inserted.
4114
# The exception is that a delta record from the source that should
4115
# be a fulltext may need to be expanded by the target (see
4116
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4117
# explicitly flush any buffered writes first in that rare case.
4118
try:
4119
new_pack = self.target_repo._pack_collection._new_pack
4120
except AttributeError:
4121
# Not a pack repository
4122
pass
4123
else:
4124
new_pack.set_write_cache_size(1024*1024)
4125
for substream_type, substream in stream:
4126
if 'stream' in debug.debug_flags:
4127
mutter('inserting substream: %s', substream_type)
4128
if substream_type == 'texts':
4129
self.target_repo.texts.insert_record_stream(substream)
4130
elif substream_type == 'inventories':
4131
if src_serializer == to_serializer:
4132
self.target_repo.inventories.insert_record_stream(
4133
substream)
4134
else:
4135
self._extract_and_insert_inventories(
4136
substream, src_serializer)
4137
elif substream_type == 'inventory-deltas':
4138
self._extract_and_insert_inventory_deltas(
4139
substream, src_serializer)
4140
elif substream_type == 'chk_bytes':
4141
# XXX: This doesn't support conversions, as it assumes the
4142
# conversion was done in the fetch code.
4143
self.target_repo.chk_bytes.insert_record_stream(substream)
4144
elif substream_type == 'revisions':
4145
# This may fallback to extract-and-insert more often than
4146
# required if the serializers are different only in terms of
4147
# the inventory.
4148
if src_serializer == to_serializer:
4149
self.target_repo.revisions.insert_record_stream(substream)
4150
else:
4151
self._extract_and_insert_revisions(substream,
4152
src_serializer)
4153
elif substream_type == 'signatures':
4154
self.target_repo.signatures.insert_record_stream(substream)
4155
else:
4156
raise AssertionError('kaboom! %s' % (substream_type,))
4157
# Done inserting data, and the missing_keys calculations will try to
4158
# read back from the inserted data, so flush the writes to the new pack
4159
# (if this is pack format).
4160
if new_pack is not None:
4161
new_pack._write_data('', flush=True)
4162
# Find all the new revisions (including ones from resume_tokens)
4163
missing_keys = self.target_repo.get_missing_parent_inventories(
4164
check_for_missing_texts=is_resume)
4165
try:
4166
for prefix, versioned_file in (
4167
('texts', self.target_repo.texts),
4168
('inventories', self.target_repo.inventories),
4169
('revisions', self.target_repo.revisions),
4170
('signatures', self.target_repo.signatures),
4171
('chk_bytes', self.target_repo.chk_bytes),
4172
):
4173
if versioned_file is None:
4174
continue
4175
# TODO: key is often going to be a StaticTuple object
4176
# I don't believe we can define a method by which
4177
# (prefix,) + StaticTuple will work, though we could
4178
# define a StaticTuple.sq_concat that would allow you to
4179
# pass in either a tuple or a StaticTuple as the second
4180
# object, so instead we could have:
4181
# StaticTuple(prefix) + key here...
4182
missing_keys.update((prefix,) + key for key in
4183
versioned_file.get_missing_compression_parent_keys())
4184
except NotImplementedError:
4185
# cannot even attempt suspending, and missing would have failed
4186
# during stream insertion.
4187
missing_keys = set()
4188
return missing_keys
4189
4190
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4191
target_rich_root = self.target_repo._format.rich_root_data
4192
target_tree_refs = self.target_repo._format.supports_tree_reference
4193
for record in substream:
4194
# Insert the delta directly
4195
inventory_delta_bytes = record.get_bytes_as('fulltext')
4196
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4197
try:
4198
parse_result = deserialiser.parse_text_bytes(
4199
inventory_delta_bytes)
4200
except inventory_delta.IncompatibleInventoryDelta, err:
4201
trace.mutter("Incompatible delta: %s", err.msg)
4202
raise errors.IncompatibleRevision(self.target_repo._format)
4203
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4204
revision_id = new_id
4205
parents = [key[0] for key in record.parents]
4206
self.target_repo.add_inventory_by_delta(
4207
basis_id, inv_delta, revision_id, parents)
4208
4209
def _extract_and_insert_inventories(self, substream, serializer,
4210
parse_delta=None):
4211
"""Generate a new inventory versionedfile in target, converting data.
4212
4213
The inventory is retrieved from the source, (deserializing it), and
4214
stored in the target (reserializing it in a different format).
4215
"""
4216
target_rich_root = self.target_repo._format.rich_root_data
4217
target_tree_refs = self.target_repo._format.supports_tree_reference
4218
for record in substream:
4219
# It's not a delta, so it must be a fulltext in the source
4220
# serializer's format.
4221
bytes = record.get_bytes_as('fulltext')
4222
revision_id = record.key[0]
4223
inv = serializer.read_inventory_from_string(bytes, revision_id)
4224
parents = [key[0] for key in record.parents]
4225
self.target_repo.add_inventory(revision_id, inv, parents)
4226
# No need to keep holding this full inv in memory when the rest of
4227
# the substream is likely to be all deltas.
4228
del inv
4229
4230
def _extract_and_insert_revisions(self, substream, serializer):
4231
for record in substream:
4232
bytes = record.get_bytes_as('fulltext')
4233
revision_id = record.key[0]
4234
rev = serializer.read_revision_from_string(bytes)
4235
if rev.revision_id != revision_id:
4236
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4237
self.target_repo.add_revision(revision_id, rev)
4238
4239
def finished(self):
4240
if self.target_repo._format._fetch_reconcile:
4241
self.target_repo.reconcile()
4242
4243
4244
class StreamSource(object):
4245
"""A source of a stream for fetching between repositories."""
4246
4247
def __init__(self, from_repository, to_format):
4248
"""Create a StreamSource streaming from from_repository."""
4249
self.from_repository = from_repository
4250
self.to_format = to_format
4251
self._record_counter = RecordCounter()
4252
4253
def delta_on_metadata(self):
4254
"""Return True if delta's are permitted on metadata streams.
4255
4256
That is on revisions and signatures.
4257
"""
4258
src_serializer = self.from_repository._format._serializer
4259
target_serializer = self.to_format._serializer
4260
return (self.to_format._fetch_uses_deltas and
4261
src_serializer == target_serializer)
4262
4263
def _fetch_revision_texts(self, revs):
4264
# fetch signatures first and then the revision texts
4265
# may need to be a InterRevisionStore call here.
4266
from_sf = self.from_repository.signatures
4267
# A missing signature is just skipped.
4268
keys = [(rev_id,) for rev_id in revs]
4269
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4270
keys,
4271
self.to_format._fetch_order,
4272
not self.to_format._fetch_uses_deltas))
4273
# If a revision has a delta, this is actually expanded inside the
4274
# insert_record_stream code now, which is an alternate fix for
4275
# bug #261339
4276
from_rf = self.from_repository.revisions
4277
revisions = from_rf.get_record_stream(
4278
keys,
4279
self.to_format._fetch_order,
4280
not self.delta_on_metadata())
4281
return [('signatures', signatures), ('revisions', revisions)]
4282
4283
def _generate_root_texts(self, revs):
4284
"""This will be called by get_stream between fetching weave texts and
4285
fetching the inventory weave.
4286
"""
4287
if self._rich_root_upgrade():
4288
return _mod_fetch.Inter1and2Helper(
4289
self.from_repository).generate_root_texts(revs)
4290
else:
4291
return []
4292
4293
def get_stream(self, search):
4294
phase = 'file'
4295
revs = search.get_keys()
4296
graph = self.from_repository.get_graph()
4297
revs = tsort.topo_sort(graph.get_parent_map(revs))
4298
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4299
text_keys = []
4300
for knit_kind, file_id, revisions in data_to_fetch:
4301
if knit_kind != phase:
4302
phase = knit_kind
4303
# Make a new progress bar for this phase
4304
if knit_kind == "file":
4305
# Accumulate file texts
4306
text_keys.extend([(file_id, revision) for revision in
4307
revisions])
4308
elif knit_kind == "inventory":
4309
# Now copy the file texts.
4310
from_texts = self.from_repository.texts
4311
yield ('texts', from_texts.get_record_stream(
4312
text_keys, self.to_format._fetch_order,
4313
not self.to_format._fetch_uses_deltas))
4314
# Cause an error if a text occurs after we have done the
4315
# copy.
4316
text_keys = None
4317
# Before we process the inventory we generate the root
4318
# texts (if necessary) so that the inventories references
4319
# will be valid.
4320
for _ in self._generate_root_texts(revs):
4321
yield _
4322
# we fetch only the referenced inventories because we do not
4323
# know for unselected inventories whether all their required
4324
# texts are present in the other repository - it could be
4325
# corrupt.
4326
for info in self._get_inventory_stream(revs):
4327
yield info
4328
elif knit_kind == "signatures":
4329
# Nothing to do here; this will be taken care of when
4330
# _fetch_revision_texts happens.
4331
pass
4332
elif knit_kind == "revisions":
4333
for record in self._fetch_revision_texts(revs):
4334
yield record
4335
else:
4336
raise AssertionError("Unknown knit kind %r" % knit_kind)
4337
4338
def get_stream_for_missing_keys(self, missing_keys):
4339
# missing keys can only occur when we are byte copying and not
4340
# translating (because translation means we don't send
4341
# unreconstructable deltas ever).
4342
keys = {}
4343
keys['texts'] = set()
4344
keys['revisions'] = set()
4345
keys['inventories'] = set()
4346
keys['chk_bytes'] = set()
4347
keys['signatures'] = set()
4348
for key in missing_keys:
4349
keys[key[0]].add(key[1:])
4350
if len(keys['revisions']):
4351
# If we allowed copying revisions at this point, we could end up
4352
# copying a revision without copying its required texts: a
4353
# violation of the requirements for repository integrity.
4354
raise AssertionError(
4355
'cannot copy revisions to fill in missing deltas %s' % (
4356
keys['revisions'],))
4357
for substream_kind, keys in keys.iteritems():
4358
vf = getattr(self.from_repository, substream_kind)
4359
if vf is None and keys:
4360
raise AssertionError(
4361
"cannot fill in keys for a versioned file we don't"
4362
" have: %s needs %s" % (substream_kind, keys))
4363
if not keys:
4364
# No need to stream something we don't have
4365
continue
4366
if substream_kind == 'inventories':
4367
# Some missing keys are genuinely ghosts, filter those out.
4368
present = self.from_repository.inventories.get_parent_map(keys)
4369
revs = [key[0] for key in present]
4370
# Get the inventory stream more-or-less as we do for the
4371
# original stream; there's no reason to assume that records
4372
# direct from the source will be suitable for the sink. (Think
4373
# e.g. 2a -> 1.9-rich-root).
4374
for info in self._get_inventory_stream(revs, missing=True):
4375
yield info
4376
continue
4377
4378
# Ask for full texts always so that we don't need more round trips
4379
# after this stream.
4380
# Some of the missing keys are genuinely ghosts, so filter absent
4381
# records. The Sink is responsible for doing another check to
4382
# ensure that ghosts don't introduce missing data for future
4383
# fetches.
4384
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4385
self.to_format._fetch_order, True))
4386
yield substream_kind, stream
4387
4388
def inventory_fetch_order(self):
4389
if self._rich_root_upgrade():
4390
return 'topological'
4391
else:
4392
return self.to_format._fetch_order
4393
4394
def _rich_root_upgrade(self):
4395
return (not self.from_repository._format.rich_root_data and
4396
self.to_format.rich_root_data)
4397
4398
def _get_inventory_stream(self, revision_ids, missing=False):
4399
from_format = self.from_repository._format
4400
if (from_format.supports_chks and self.to_format.supports_chks and
4401
from_format.network_name() == self.to_format.network_name()):
4402
raise AssertionError(
4403
"this case should be handled by GroupCHKStreamSource")
4404
elif 'forceinvdeltas' in debug.debug_flags:
4405
return self._get_convertable_inventory_stream(revision_ids,
4406
delta_versus_null=missing)
4407
elif from_format.network_name() == self.to_format.network_name():
4408
# Same format.
4409
return self._get_simple_inventory_stream(revision_ids,
4410
missing=missing)
4411
elif (not from_format.supports_chks and not self.to_format.supports_chks
4412
and from_format._serializer == self.to_format._serializer):
4413
# Essentially the same format.
4414
return self._get_simple_inventory_stream(revision_ids,
4415
missing=missing)
4416
else:
4417
# Any time we switch serializations, we want to use an
4418
# inventory-delta based approach.
4419
return self._get_convertable_inventory_stream(revision_ids,
4420
delta_versus_null=missing)
4421
4422
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4423
# NB: This currently reopens the inventory weave in source;
4424
# using a single stream interface instead would avoid this.
4425
from_weave = self.from_repository.inventories
4426
if missing:
4427
delta_closure = True
4428
else:
4429
delta_closure = not self.delta_on_metadata()
4430
yield ('inventories', from_weave.get_record_stream(
4431
[(rev_id,) for rev_id in revision_ids],
4432
self.inventory_fetch_order(), delta_closure))
4433
4434
def _get_convertable_inventory_stream(self, revision_ids,
4435
delta_versus_null=False):
4436
# The two formats are sufficiently different that there is no fast
4437
# path, so we need to send just inventorydeltas, which any
4438
# sufficiently modern client can insert into any repository.
4439
# The StreamSink code expects to be able to
4440
# convert on the target, so we need to put bytes-on-the-wire that can
4441
# be converted. That means inventory deltas (if the remote is <1.19,
4442
# RemoteStreamSink will fallback to VFS to insert the deltas).
4443
yield ('inventory-deltas',
4444
self._stream_invs_as_deltas(revision_ids,
4445
delta_versus_null=delta_versus_null))
4446
4447
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4448
"""Return a stream of inventory-deltas for the given rev ids.
4449
4450
:param revision_ids: The list of inventories to transmit
4451
:param delta_versus_null: Don't try to find a minimal delta for this
4452
entry, instead compute the delta versus the NULL_REVISION. This
4453
effectively streams a complete inventory. Used for stuff like
4454
filling in missing parents, etc.
4455
"""
4456
from_repo = self.from_repository
4457
revision_keys = [(rev_id,) for rev_id in revision_ids]
4458
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4459
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4460
# method...
4461
inventories = self.from_repository.iter_inventories(
4462
revision_ids, 'topological')
4463
format = from_repo._format
4464
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4465
inventory_cache = lru_cache.LRUCache(50)
4466
null_inventory = from_repo.revision_tree(
4467
_mod_revision.NULL_REVISION).inventory
4468
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4469
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4470
# repo back into a non-rich-root repo ought to be allowed)
4471
serializer = inventory_delta.InventoryDeltaSerializer(
4472
versioned_root=format.rich_root_data,
4473
tree_references=format.supports_tree_reference)
4474
for inv in inventories:
4475
key = (inv.revision_id,)
4476
parent_keys = parent_map.get(key, ())
4477
delta = None
4478
if not delta_versus_null and parent_keys:
4479
# The caller did not ask for complete inventories and we have
4480
# some parents that we can delta against. Make a delta against
4481
# each parent so that we can find the smallest.
4482
parent_ids = [parent_key[0] for parent_key in parent_keys]
4483
for parent_id in parent_ids:
4484
if parent_id not in invs_sent_so_far:
4485
# We don't know that the remote side has this basis, so
4486
# we can't use it.
4487
continue
4488
if parent_id == _mod_revision.NULL_REVISION:
4489
parent_inv = null_inventory
4490
else:
4491
parent_inv = inventory_cache.get(parent_id, None)
4492
if parent_inv is None:
4493
parent_inv = from_repo.get_inventory(parent_id)
4494
candidate_delta = inv._make_delta(parent_inv)
4495
if (delta is None or
4496
len(delta) > len(candidate_delta)):
4497
delta = candidate_delta
4498
basis_id = parent_id
4499
if delta is None:
4500
# Either none of the parents ended up being suitable, or we
4501
# were asked to delta against NULL
4502
basis_id = _mod_revision.NULL_REVISION
4503
delta = inv._make_delta(null_inventory)
4504
invs_sent_so_far.add(inv.revision_id)
4505
inventory_cache[inv.revision_id] = inv
4506
delta_serialized = ''.join(
4507
serializer.delta_to_lines(basis_id, key[-1], delta))
4508
yield versionedfile.FulltextContentFactory(
4509
key, parent_keys, None, delta_serialized)
4510
4511
1837
4512
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
1838
4513
stop_revision=None):
1839
4514
"""Extend the partial history to include a given index
1849
4524
it is encountered, history extension will stop.
1850
4525
"""
1851
4526
start_revision = partial_history_cache[-1]
1852
graph = repo.get_graph()
1853
iterator = graph.iter_lefthand_ancestry(start_revision,
1854
(_mod_revision.NULL_REVISION,))
4527
iterator = repo.iter_reverse_revision_history(start_revision)
1855
4528
try:
1856
# skip the last revision in the list
4529
#skip the last revision in the list
1857
4530
iterator.next()
1858
4531
while True:
1859
4532
if (stop_index is not None and
1868
4541
return
1869
4542
1870
4543
1871
class _LazyListJoin(object):
1872
"""An iterable yielding the contents of many lists as one list.
1873
1874
Each iterator made from this will reflect the current contents of the lists
1875
at the time the iterator is made.
1876
1877
This is used by Repository's _make_parents_provider implementation so that
1878
it is safe to do::
1879
1880
pp = repo._make_parents_provider() # uses a list of fallback repos
1881
pp.add_fallback_repository(other_repo) # appends to that list
1882
result = pp.get_parent_map(...)
1883
# The result will include revs from other_repo
1884
"""
1885
1886
def __init__(self, *list_parts):
1887
self.list_parts = list_parts
1888
1889
def __iter__(self):
1890
full_list = []
1891
for list_part in self.list_parts:
1892
full_list.extend(list_part)
1893
return iter(full_list)
1894
1895
def __repr__(self):
1896
return "%s.%s(%s)" % (self.__module__, self.__class__.__name__,
1897
self.list_parts)
4544
Loggerhead is a web-based interface for Breezy
Version: 2.0.1