« back to all changes in this revision
Viewing changes to bzrlib/vf_repository.py
- browse files at revision 5848.2.1
- compare with another revision
- download diff
- download tarball
- view history from revision 5848.2.1
- Committer: John Arbash Meinel
- Date: 2011-05-11 11:35:28 UTC
- mto: This revision was merged to the branch mainline in revision 5851.
- Revision ID: john@arbash-meinel.com-20110511113528-qepibuwxicjrbb2h
Break compatibility with python <2.6.
This includes auditing the code for places where we were doing
explicit 'sys.version' checks and removing them as appropriate.
This includes auditing the code for places where we were doing
explicit 'sys.version' checks and removing them as appropriate.
- files added:
- NEWS
- bzrlib/plugins/changelog_merge
- bzrlib/plugins/changelog_merge/tests
- bzrlib/plugins/weave_fmt
- bzrlib/plugins/weave_fmt/tests
- bzrlib/tests/per_bzrdir
- bzrlib/tests/per_repository_vf
- files removed:
- bzrlib/deprecated_graph.py
- files renamed:
- bzrlib/repofmt/weaverepo.py => bzrlib/plugins/weave_fmt/repository.py
- bzrlib/xml4.py => bzrlib/plugins/weave_fmt/xml4.py
- bzrlib/tests/per_repository/helpers.py => bzrlib/tests/per_repository_vf/helpers.py
- bzrlib/tests/per_repository/test_add_inventory_by_delta.py => bzrlib/tests/per_repository_vf/test_add_inventory_by_delta.py
- bzrlib/tests/per_repository/test_check_reconcile.py => bzrlib/tests/per_repository_vf/test_check_reconcile.py
- NEWS => doc/en/release-notes/bzr-2.3.txt
- NEWS-template.txt => doc/en/release-notes/release-template.txt
- files modified:
- .bzrignore
added
removed
bzrlib/vf_repository.py
1
# Copyright (C) 2005-2011 Canonical Ltd
2
#
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
7
#
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
12
#
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
16
17
"""Repository formats built around versioned files."""
18
19
20
from bzrlib.lazy_import import lazy_import
21
lazy_import(globals(), """
22
from bzrlib import (
23
debug,
24
fetch as _mod_fetch,
25
fifo_cache,
26
gpg,
27
graph,
28
inventory_delta,
29
lru_cache,
30
osutils,
31
revision as _mod_revision,
32
serializer as _mod_serializer,
33
static_tuple,
34
tsort,
35
ui,
36
versionedfile,
37
)
38
39
from bzrlib.recordcounter import RecordCounter
40
from bzrlib.revisiontree import InventoryRevisionTree
41
from bzrlib.testament import Testament
42
""")
43
44
from bzrlib import (
45
errors,
46
)
47
from bzrlib.decorators import (
48
needs_read_lock,
49
needs_write_lock,
50
only_raises,
51
)
52
from bzrlib.inventory import (
53
Inventory,
54
InventoryDirectory,
55
ROOT_ID,
56
entry_factory,
57
)
58
59
from bzrlib.repository import (
60
CommitBuilder,
61
InterRepository,
62
MetaDirRepository,
63
MetaDirRepositoryFormat,
64
Repository,
65
RepositoryFormat,
66
)
67
68
from bzrlib.trace import (
69
mutter,
70
)
71
72
73
class VersionedFileRepositoryFormat(RepositoryFormat):
74
"""Base class for all repository formats that are VersionedFiles-based."""
75
76
supports_full_versioned_files = True
77
78
# Should commit add an inventory, or an inventory delta to the repository.
79
_commit_inv_deltas = True
80
# What order should fetch operations request streams in?
81
# The default is unordered as that is the cheapest for an origin to
82
# provide.
83
_fetch_order = 'unordered'
84
# Does this repository format use deltas that can be fetched as-deltas ?
85
# (E.g. knits, where the knit deltas can be transplanted intact.
86
# We default to False, which will ensure that enough data to get
87
# a full text out of any fetch stream will be grabbed.
88
_fetch_uses_deltas = False
89
90
91
class VersionedFileCommitBuilder(CommitBuilder):
92
"""Commit builder implementation for versioned files based repositories.
93
"""
94
95
# this commit builder supports the record_entry_contents interface
96
supports_record_entry_contents = True
97
98
# the default CommitBuilder does not manage trees whose root is versioned.
99
_versioned_root = False
100
101
def __init__(self, repository, parents, config, timestamp=None,
102
timezone=None, committer=None, revprops=None,
103
revision_id=None, lossy=False):
104
super(VersionedFileCommitBuilder, self).__init__(repository,
105
parents, config, timestamp, timezone, committer, revprops,
106
revision_id, lossy)
107
self.new_inventory = Inventory(None)
108
self._basis_delta = []
109
self.__heads = graph.HeadsCache(repository.get_graph()).heads
110
# memo'd check for no-op commits.
111
self._any_changes = False
112
# API compatibility, older code that used CommitBuilder did not call
113
# .record_delete(), which means the delta that is computed would not be
114
# valid. Callers that will call record_delete() should call
115
# .will_record_deletes() to indicate that.
116
self._recording_deletes = False
117
118
def will_record_deletes(self):
119
"""Tell the commit builder that deletes are being notified.
120
121
This enables the accumulation of an inventory delta; for the resulting
122
commit to be valid, deletes against the basis MUST be recorded via
123
builder.record_delete().
124
"""
125
self._recording_deletes = True
126
try:
127
basis_id = self.parents[0]
128
except IndexError:
129
basis_id = _mod_revision.NULL_REVISION
130
self.basis_delta_revision = basis_id
131
132
def any_changes(self):
133
"""Return True if any entries were changed.
134
135
This includes merge-only changes. It is the core for the --unchanged
136
detection in commit.
137
138
:return: True if any changes have occured.
139
"""
140
return self._any_changes
141
142
def _ensure_fallback_inventories(self):
143
"""Ensure that appropriate inventories are available.
144
145
This only applies to repositories that are stacked, and is about
146
enusring the stacking invariants. Namely, that for any revision that is
147
present, we either have all of the file content, or we have the parent
148
inventory and the delta file content.
149
"""
150
if not self.repository._fallback_repositories:
151
return
152
if not self.repository._format.supports_chks:
153
raise errors.BzrError("Cannot commit directly to a stacked branch"
154
" in pre-2a formats. See "
155
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
156
# This is a stacked repo, we need to make sure we have the parent
157
# inventories for the parents.
158
parent_keys = [(p,) for p in self.parents]
159
parent_map = self.repository.inventories._index.get_parent_map(parent_keys)
160
missing_parent_keys = set([pk for pk in parent_keys
161
if pk not in parent_map])
162
fallback_repos = list(reversed(self.repository._fallback_repositories))
163
missing_keys = [('inventories', pk[0])
164
for pk in missing_parent_keys]
165
resume_tokens = []
166
while missing_keys and fallback_repos:
167
fallback_repo = fallback_repos.pop()
168
source = fallback_repo._get_source(self.repository._format)
169
sink = self.repository._get_sink()
170
stream = source.get_stream_for_missing_keys(missing_keys)
171
missing_keys = sink.insert_stream_without_locking(stream,
172
self.repository._format)
173
if missing_keys:
174
raise errors.BzrError('Unable to fill in parent inventories for a'
175
' stacked branch')
176
177
def commit(self, message):
178
"""Make the actual commit.
179
180
:return: The revision id of the recorded revision.
181
"""
182
self._validate_unicode_text(message, 'commit message')
183
rev = _mod_revision.Revision(
184
timestamp=self._timestamp,
185
timezone=self._timezone,
186
committer=self._committer,
187
message=message,
188
inventory_sha1=self.inv_sha1,
189
revision_id=self._new_revision_id,
190
properties=self._revprops)
191
rev.parent_ids = self.parents
192
self.repository.add_revision(self._new_revision_id, rev,
193
self.new_inventory, self._config)
194
self._ensure_fallback_inventories()
195
self.repository.commit_write_group()
196
return self._new_revision_id
197
198
def abort(self):
199
"""Abort the commit that is being built.
200
"""
201
self.repository.abort_write_group()
202
203
def revision_tree(self):
204
"""Return the tree that was just committed.
205
206
After calling commit() this can be called to get a
207
RevisionTree representing the newly committed tree. This is
208
preferred to calling Repository.revision_tree() because that may
209
require deserializing the inventory, while we already have a copy in
210
memory.
211
"""
212
if self.new_inventory is None:
213
self.new_inventory = self.repository.get_inventory(
214
self._new_revision_id)
215
return InventoryRevisionTree(self.repository, self.new_inventory,
216
self._new_revision_id)
217
218
def finish_inventory(self):
219
"""Tell the builder that the inventory is finished.
220
221
:return: The inventory id in the repository, which can be used with
222
repository.get_inventory.
223
"""
224
if self.new_inventory is None:
225
# an inventory delta was accumulated without creating a new
226
# inventory.
227
basis_id = self.basis_delta_revision
228
# We ignore the 'inventory' returned by add_inventory_by_delta
229
# because self.new_inventory is used to hint to the rest of the
230
# system what code path was taken
231
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
232
basis_id, self._basis_delta, self._new_revision_id,
233
self.parents)
234
else:
235
if self.new_inventory.root is None:
236
raise AssertionError('Root entry should be supplied to'
237
' record_entry_contents, as of bzr 0.10.')
238
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
239
self.new_inventory.revision_id = self._new_revision_id
240
self.inv_sha1 = self.repository.add_inventory(
241
self._new_revision_id,
242
self.new_inventory,
243
self.parents
244
)
245
return self._new_revision_id
246
247
def _check_root(self, ie, parent_invs, tree):
248
"""Helper for record_entry_contents.
249
250
:param ie: An entry being added.
251
:param parent_invs: The inventories of the parent revisions of the
252
commit.
253
:param tree: The tree that is being committed.
254
"""
255
# In this revision format, root entries have no knit or weave When
256
# serializing out to disk and back in root.revision is always
257
# _new_revision_id
258
ie.revision = self._new_revision_id
259
260
def _require_root_change(self, tree):
261
"""Enforce an appropriate root object change.
262
263
This is called once when record_iter_changes is called, if and only if
264
the root was not in the delta calculated by record_iter_changes.
265
266
:param tree: The tree which is being committed.
267
"""
268
if len(self.parents) == 0:
269
raise errors.RootMissing()
270
entry = entry_factory['directory'](tree.path2id(''), '',
271
None)
272
entry.revision = self._new_revision_id
273
self._basis_delta.append(('', '', entry.file_id, entry))
274
275
def _get_delta(self, ie, basis_inv, path):
276
"""Get a delta against the basis inventory for ie."""
277
if ie.file_id not in basis_inv:
278
# add
279
result = (None, path, ie.file_id, ie)
280
self._basis_delta.append(result)
281
return result
282
elif ie != basis_inv[ie.file_id]:
283
# common but altered
284
# TODO: avoid tis id2path call.
285
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
286
self._basis_delta.append(result)
287
return result
288
else:
289
# common, unaltered
290
return None
291
292
def _heads(self, file_id, revision_ids):
293
"""Calculate the graph heads for revision_ids in the graph of file_id.
294
295
This can use either a per-file graph or a global revision graph as we
296
have an identity relationship between the two graphs.
297
"""
298
return self.__heads(revision_ids)
299
300
def get_basis_delta(self):
301
"""Return the complete inventory delta versus the basis inventory.
302
303
This has been built up with the calls to record_delete and
304
record_entry_contents. The client must have already called
305
will_record_deletes() to indicate that they will be generating a
306
complete delta.
307
308
:return: An inventory delta, suitable for use with apply_delta, or
309
Repository.add_inventory_by_delta, etc.
310
"""
311
if not self._recording_deletes:
312
raise AssertionError("recording deletes not activated.")
313
return self._basis_delta
314
315
def record_delete(self, path, file_id):
316
"""Record that a delete occured against a basis tree.
317
318
This is an optional API - when used it adds items to the basis_delta
319
being accumulated by the commit builder. It cannot be called unless the
320
method will_record_deletes() has been called to inform the builder that
321
a delta is being supplied.
322
323
:param path: The path of the thing deleted.
324
:param file_id: The file id that was deleted.
325
"""
326
if not self._recording_deletes:
327
raise AssertionError("recording deletes not activated.")
328
delta = (path, None, file_id, None)
329
self._basis_delta.append(delta)
330
self._any_changes = True
331
return delta
332
333
def record_entry_contents(self, ie, parent_invs, path, tree,
334
content_summary):
335
"""Record the content of ie from tree into the commit if needed.
336
337
Side effect: sets ie.revision when unchanged
338
339
:param ie: An inventory entry present in the commit.
340
:param parent_invs: The inventories of the parent revisions of the
341
commit.
342
:param path: The path the entry is at in the tree.
343
:param tree: The tree which contains this entry and should be used to
344
obtain content.
345
:param content_summary: Summary data from the tree about the paths
346
content - stat, length, exec, sha/link target. This is only
347
accessed when the entry has a revision of None - that is when it is
348
a candidate to commit.
349
:return: A tuple (change_delta, version_recorded, fs_hash).
350
change_delta is an inventory_delta change for this entry against
351
the basis tree of the commit, or None if no change occured against
352
the basis tree.
353
version_recorded is True if a new version of the entry has been
354
recorded. For instance, committing a merge where a file was only
355
changed on the other side will return (delta, False).
356
fs_hash is either None, or the hash details for the path (currently
357
a tuple of the contents sha1 and the statvalue returned by
358
tree.get_file_with_stat()).
359
"""
360
if self.new_inventory.root is None:
361
if ie.parent_id is not None:
362
raise errors.RootMissing()
363
self._check_root(ie, parent_invs, tree)
364
if ie.revision is None:
365
kind = content_summary[0]
366
else:
367
# ie is carried over from a prior commit
368
kind = ie.kind
369
# XXX: repository specific check for nested tree support goes here - if
370
# the repo doesn't want nested trees we skip it ?
371
if (kind == 'tree-reference' and
372
not self.repository._format.supports_tree_reference):
373
# mismatch between commit builder logic and repository:
374
# this needs the entry creation pushed down into the builder.
375
raise NotImplementedError('Missing repository subtree support.')
376
self.new_inventory.add(ie)
377
378
# TODO: slow, take it out of the inner loop.
379
try:
380
basis_inv = parent_invs[0]
381
except IndexError:
382
basis_inv = Inventory(root_id=None)
383
384
# ie.revision is always None if the InventoryEntry is considered
385
# for committing. We may record the previous parents revision if the
386
# content is actually unchanged against a sole head.
387
if ie.revision is not None:
388
if not self._versioned_root and path == '':
389
# repositories that do not version the root set the root's
390
# revision to the new commit even when no change occurs (more
391
# specifically, they do not record a revision on the root; and
392
# the rev id is assigned to the root during deserialisation -
393
# this masks when a change may have occurred against the basis.
394
# To match this we always issue a delta, because the revision
395
# of the root will always be changing.
396
if ie.file_id in basis_inv:
397
delta = (basis_inv.id2path(ie.file_id), path,
398
ie.file_id, ie)
399
else:
400
# add
401
delta = (None, path, ie.file_id, ie)
402
self._basis_delta.append(delta)
403
return delta, False, None
404
else:
405
# we don't need to commit this, because the caller already
406
# determined that an existing revision of this file is
407
# appropriate. If it's not being considered for committing then
408
# it and all its parents to the root must be unaltered so
409
# no-change against the basis.
410
if ie.revision == self._new_revision_id:
411
raise AssertionError("Impossible situation, a skipped "
412
"inventory entry (%r) claims to be modified in this "
413
"commit (%r).", (ie, self._new_revision_id))
414
return None, False, None
415
# XXX: Friction: parent_candidates should return a list not a dict
416
# so that we don't have to walk the inventories again.
417
parent_candiate_entries = ie.parent_candidates(parent_invs)
418
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
419
heads = []
420
for inv in parent_invs:
421
if ie.file_id in inv:
422
old_rev = inv[ie.file_id].revision
423
if old_rev in head_set:
424
heads.append(inv[ie.file_id].revision)
425
head_set.remove(inv[ie.file_id].revision)
426
427
store = False
428
# now we check to see if we need to write a new record to the
429
# file-graph.
430
# We write a new entry unless there is one head to the ancestors, and
431
# the kind-derived content is unchanged.
432
433
# Cheapest check first: no ancestors, or more the one head in the
434
# ancestors, we write a new node.
435
if len(heads) != 1:
436
store = True
437
if not store:
438
# There is a single head, look it up for comparison
439
parent_entry = parent_candiate_entries[heads[0]]
440
# if the non-content specific data has changed, we'll be writing a
441
# node:
442
if (parent_entry.parent_id != ie.parent_id or
443
parent_entry.name != ie.name):
444
store = True
445
# now we need to do content specific checks:
446
if not store:
447
# if the kind changed the content obviously has
448
if kind != parent_entry.kind:
449
store = True
450
# Stat cache fingerprint feedback for the caller - None as we usually
451
# don't generate one.
452
fingerprint = None
453
if kind == 'file':
454
if content_summary[2] is None:
455
raise ValueError("Files must not have executable = None")
456
if not store:
457
# We can't trust a check of the file length because of content
458
# filtering...
459
if (# if the exec bit has changed we have to store:
460
parent_entry.executable != content_summary[2]):
461
store = True
462
elif parent_entry.text_sha1 == content_summary[3]:
463
# all meta and content is unchanged (using a hash cache
464
# hit to check the sha)
465
ie.revision = parent_entry.revision
466
ie.text_size = parent_entry.text_size
467
ie.text_sha1 = parent_entry.text_sha1
468
ie.executable = parent_entry.executable
469
return self._get_delta(ie, basis_inv, path), False, None
470
else:
471
# Either there is only a hash change(no hash cache entry,
472
# or same size content change), or there is no change on
473
# this file at all.
474
# Provide the parent's hash to the store layer, so that the
475
# content is unchanged we will not store a new node.
476
nostore_sha = parent_entry.text_sha1
477
if store:
478
# We want to record a new node regardless of the presence or
479
# absence of a content change in the file.
480
nostore_sha = None
481
ie.executable = content_summary[2]
482
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
483
try:
484
text = file_obj.read()
485
finally:
486
file_obj.close()
487
try:
488
ie.text_sha1, ie.text_size = self._add_text_to_weave(
489
ie.file_id, text, heads, nostore_sha)
490
# Let the caller know we generated a stat fingerprint.
491
fingerprint = (ie.text_sha1, stat_value)
492
except errors.ExistingContent:
493
# Turns out that the file content was unchanged, and we were
494
# only going to store a new node if it was changed. Carry over
495
# the entry.
496
ie.revision = parent_entry.revision
497
ie.text_size = parent_entry.text_size
498
ie.text_sha1 = parent_entry.text_sha1
499
ie.executable = parent_entry.executable
500
return self._get_delta(ie, basis_inv, path), False, None
501
elif kind == 'directory':
502
if not store:
503
# all data is meta here, nothing specific to directory, so
504
# carry over:
505
ie.revision = parent_entry.revision
506
return self._get_delta(ie, basis_inv, path), False, None
507
self._add_text_to_weave(ie.file_id, '', heads, None)
508
elif kind == 'symlink':
509
current_link_target = content_summary[3]
510
if not store:
511
# symlink target is not generic metadata, check if it has
512
# changed.
513
if current_link_target != parent_entry.symlink_target:
514
store = True
515
if not store:
516
# unchanged, carry over.
517
ie.revision = parent_entry.revision
518
ie.symlink_target = parent_entry.symlink_target
519
return self._get_delta(ie, basis_inv, path), False, None
520
ie.symlink_target = current_link_target
521
self._add_text_to_weave(ie.file_id, '', heads, None)
522
elif kind == 'tree-reference':
523
if not store:
524
if content_summary[3] != parent_entry.reference_revision:
525
store = True
526
if not store:
527
# unchanged, carry over.
528
ie.reference_revision = parent_entry.reference_revision
529
ie.revision = parent_entry.revision
530
return self._get_delta(ie, basis_inv, path), False, None
531
ie.reference_revision = content_summary[3]
532
if ie.reference_revision is None:
533
raise AssertionError("invalid content_summary for nested tree: %r"
534
% (content_summary,))
535
self._add_text_to_weave(ie.file_id, '', heads, None)
536
else:
537
raise NotImplementedError('unknown kind')
538
ie.revision = self._new_revision_id
539
self._any_changes = True
540
return self._get_delta(ie, basis_inv, path), True, fingerprint
541
542
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
543
_entry_factory=entry_factory):
544
"""Record a new tree via iter_changes.
545
546
:param tree: The tree to obtain text contents from for changed objects.
547
:param basis_revision_id: The revision id of the tree the iter_changes
548
has been generated against. Currently assumed to be the same
549
as self.parents[0] - if it is not, errors may occur.
550
:param iter_changes: An iter_changes iterator with the changes to apply
551
to basis_revision_id. The iterator must not include any items with
552
a current kind of None - missing items must be either filtered out
553
or errored-on beefore record_iter_changes sees the item.
554
:param _entry_factory: Private method to bind entry_factory locally for
555
performance.
556
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
557
tree._observed_sha1.
558
"""
559
# Create an inventory delta based on deltas between all the parents and
560
# deltas between all the parent inventories. We use inventory delta's
561
# between the inventory objects because iter_changes masks
562
# last-changed-field only changes.
563
# Working data:
564
# file_id -> change map, change is fileid, paths, changed, versioneds,
565
# parents, names, kinds, executables
566
merged_ids = {}
567
# {file_id -> revision_id -> inventory entry, for entries in parent
568
# trees that are not parents[0]
569
parent_entries = {}
570
ghost_basis = False
571
try:
572
revtrees = list(self.repository.revision_trees(self.parents))
573
except errors.NoSuchRevision:
574
# one or more ghosts, slow path.
575
revtrees = []
576
for revision_id in self.parents:
577
try:
578
revtrees.append(self.repository.revision_tree(revision_id))
579
except errors.NoSuchRevision:
580
if not revtrees:
581
basis_revision_id = _mod_revision.NULL_REVISION
582
ghost_basis = True
583
revtrees.append(self.repository.revision_tree(
584
_mod_revision.NULL_REVISION))
585
# The basis inventory from a repository
586
if revtrees:
587
basis_inv = revtrees[0].inventory
588
else:
589
basis_inv = self.repository.revision_tree(
590
_mod_revision.NULL_REVISION).inventory
591
if len(self.parents) > 0:
592
if basis_revision_id != self.parents[0] and not ghost_basis:
593
raise Exception(
594
"arbitrary basis parents not yet supported with merges")
595
for revtree in revtrees[1:]:
596
for change in revtree.inventory._make_delta(basis_inv):
597
if change[1] is None:
598
# Not present in this parent.
599
continue
600
if change[2] not in merged_ids:
601
if change[0] is not None:
602
basis_entry = basis_inv[change[2]]
603
merged_ids[change[2]] = [
604
# basis revid
605
basis_entry.revision,
606
# new tree revid
607
change[3].revision]
608
parent_entries[change[2]] = {
609
# basis parent
610
basis_entry.revision:basis_entry,
611
# this parent
612
change[3].revision:change[3],
613
}
614
else:
615
merged_ids[change[2]] = [change[3].revision]
616
parent_entries[change[2]] = {change[3].revision:change[3]}
617
else:
618
merged_ids[change[2]].append(change[3].revision)
619
parent_entries[change[2]][change[3].revision] = change[3]
620
else:
621
merged_ids = {}
622
# Setup the changes from the tree:
623
# changes maps file_id -> (change, [parent revision_ids])
624
changes= {}
625
for change in iter_changes:
626
# This probably looks up in basis_inv way to much.
627
if change[1][0] is not None:
628
head_candidate = [basis_inv[change[0]].revision]
629
else:
630
head_candidate = []
631
changes[change[0]] = change, merged_ids.get(change[0],
632
head_candidate)
633
unchanged_merged = set(merged_ids) - set(changes)
634
# Extend the changes dict with synthetic changes to record merges of
635
# texts.
636
for file_id in unchanged_merged:
637
# Record a merged version of these items that did not change vs the
638
# basis. This can be either identical parallel changes, or a revert
639
# of a specific file after a merge. The recorded content will be
640
# that of the current tree (which is the same as the basis), but
641
# the per-file graph will reflect a merge.
642
# NB:XXX: We are reconstructing path information we had, this
643
# should be preserved instead.
644
# inv delta change: (file_id, (path_in_source, path_in_target),
645
# changed_content, versioned, parent, name, kind,
646
# executable)
647
try:
648
basis_entry = basis_inv[file_id]
649
except errors.NoSuchId:
650
# a change from basis->some_parents but file_id isn't in basis
651
# so was new in the merge, which means it must have changed
652
# from basis -> current, and as it hasn't the add was reverted
653
# by the user. So we discard this change.
654
pass
655
else:
656
change = (file_id,
657
(basis_inv.id2path(file_id), tree.id2path(file_id)),
658
False, (True, True),
659
(basis_entry.parent_id, basis_entry.parent_id),
660
(basis_entry.name, basis_entry.name),
661
(basis_entry.kind, basis_entry.kind),
662
(basis_entry.executable, basis_entry.executable))
663
changes[file_id] = (change, merged_ids[file_id])
664
# changes contains tuples with the change and a set of inventory
665
# candidates for the file.
666
# inv delta is:
667
# old_path, new_path, file_id, new_inventory_entry
668
seen_root = False # Is the root in the basis delta?
669
inv_delta = self._basis_delta
670
modified_rev = self._new_revision_id
671
for change, head_candidates in changes.values():
672
if change[3][1]: # versioned in target.
673
# Several things may be happening here:
674
# We may have a fork in the per-file graph
675
# - record a change with the content from tree
676
# We may have a change against < all trees
677
# - carry over the tree that hasn't changed
678
# We may have a change against all trees
679
# - record the change with the content from tree
680
kind = change[6][1]
681
file_id = change[0]
682
entry = _entry_factory[kind](file_id, change[5][1],
683
change[4][1])
684
head_set = self._heads(change[0], set(head_candidates))
685
heads = []
686
# Preserve ordering.
687
for head_candidate in head_candidates:
688
if head_candidate in head_set:
689
heads.append(head_candidate)
690
head_set.remove(head_candidate)
691
carried_over = False
692
if len(heads) == 1:
693
# Could be a carry-over situation:
694
parent_entry_revs = parent_entries.get(file_id, None)
695
if parent_entry_revs:
696
parent_entry = parent_entry_revs.get(heads[0], None)
697
else:
698
parent_entry = None
699
if parent_entry is None:
700
# The parent iter_changes was called against is the one
701
# that is the per-file head, so any change is relevant
702
# iter_changes is valid.
703
carry_over_possible = False
704
else:
705
# could be a carry over situation
706
# A change against the basis may just indicate a merge,
707
# we need to check the content against the source of the
708
# merge to determine if it was changed after the merge
709
# or carried over.
710
if (parent_entry.kind != entry.kind or
711
parent_entry.parent_id != entry.parent_id or
712
parent_entry.name != entry.name):
713
# Metadata common to all entries has changed
714
# against per-file parent
715
carry_over_possible = False
716
else:
717
carry_over_possible = True
718
# per-type checks for changes against the parent_entry
719
# are done below.
720
else:
721
# Cannot be a carry-over situation
722
carry_over_possible = False
723
# Populate the entry in the delta
724
if kind == 'file':
725
# XXX: There is still a small race here: If someone reverts the content of a file
726
# after iter_changes examines and decides it has changed,
727
# we will unconditionally record a new version even if some
728
# other process reverts it while commit is running (with
729
# the revert happening after iter_changes did its
730
# examination).
731
if change[7][1]:
732
entry.executable = True
733
else:
734
entry.executable = False
735
if (carry_over_possible and
736
parent_entry.executable == entry.executable):
737
# Check the file length, content hash after reading
738
# the file.
739
nostore_sha = parent_entry.text_sha1
740
else:
741
nostore_sha = None
742
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
743
try:
744
text = file_obj.read()
745
finally:
746
file_obj.close()
747
try:
748
entry.text_sha1, entry.text_size = self._add_text_to_weave(
749
file_id, text, heads, nostore_sha)
750
yield file_id, change[1][1], (entry.text_sha1, stat_value)
751
except errors.ExistingContent:
752
# No content change against a carry_over parent
753
# Perhaps this should also yield a fs hash update?
754
carried_over = True
755
entry.text_size = parent_entry.text_size
756
entry.text_sha1 = parent_entry.text_sha1
757
elif kind == 'symlink':
758
# Wants a path hint?
759
entry.symlink_target = tree.get_symlink_target(file_id)
760
if (carry_over_possible and
761
parent_entry.symlink_target == entry.symlink_target):
762
carried_over = True
763
else:
764
self._add_text_to_weave(change[0], '', heads, None)
765
elif kind == 'directory':
766
if carry_over_possible:
767
carried_over = True
768
else:
769
# Nothing to set on the entry.
770
# XXX: split into the Root and nonRoot versions.
771
if change[1][1] != '' or self.repository.supports_rich_root():
772
self._add_text_to_weave(change[0], '', heads, None)
773
elif kind == 'tree-reference':
774
if not self.repository._format.supports_tree_reference:
775
# This isn't quite sane as an error, but we shouldn't
776
# ever see this code path in practice: tree's don't
777
# permit references when the repo doesn't support tree
778
# references.
779
raise errors.UnsupportedOperation(tree.add_reference,
780
self.repository)
781
reference_revision = tree.get_reference_revision(change[0])
782
entry.reference_revision = reference_revision
783
if (carry_over_possible and
784
parent_entry.reference_revision == reference_revision):
785
carried_over = True
786
else:
787
self._add_text_to_weave(change[0], '', heads, None)
788
else:
789
raise AssertionError('unknown kind %r' % kind)
790
if not carried_over:
791
entry.revision = modified_rev
792
else:
793
entry.revision = parent_entry.revision
794
else:
795
entry = None
796
new_path = change[1][1]
797
inv_delta.append((change[1][0], new_path, change[0], entry))
798
if new_path == '':
799
seen_root = True
800
self.new_inventory = None
801
if len(inv_delta):
802
# This should perhaps be guarded by a check that the basis we
803
# commit against is the basis for the commit and if not do a delta
804
# against the basis.
805
self._any_changes = True
806
if not seen_root:
807
# housekeeping root entry changes do not affect no-change commits.
808
self._require_root_change(tree)
809
self.basis_delta_revision = basis_revision_id
810
811
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
812
parent_keys = tuple([(file_id, parent) for parent in parents])
813
return self.repository.texts._add_text(
814
(file_id, self._new_revision_id), parent_keys, new_text,
815
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
816
817
818
class VersionedFileRootCommitBuilder(VersionedFileCommitBuilder):
819
"""This commitbuilder actually records the root id"""
820
821
# the root entry gets versioned properly by this builder.
822
_versioned_root = True
823
824
def _check_root(self, ie, parent_invs, tree):
825
"""Helper for record_entry_contents.
826
827
:param ie: An entry being added.
828
:param parent_invs: The inventories of the parent revisions of the
829
commit.
830
:param tree: The tree that is being committed.
831
"""
832
833
def _require_root_change(self, tree):
834
"""Enforce an appropriate root object change.
835
836
This is called once when record_iter_changes is called, if and only if
837
the root was not in the delta calculated by record_iter_changes.
838
839
:param tree: The tree which is being committed.
840
"""
841
# versioned roots do not change unless the tree found a change.
842
843
844
class VersionedFileRepository(Repository):
845
"""Repository holding history for one or more branches.
846
847
The repository holds and retrieves historical information including
848
revisions and file history. It's normally accessed only by the Branch,
849
which views a particular line of development through that history.
850
851
The Repository builds on top of some byte storage facilies (the revisions,
852
signatures, inventories, texts and chk_bytes attributes) and a Transport,
853
which respectively provide byte storage and a means to access the (possibly
854
remote) disk.
855
856
The byte storage facilities are addressed via tuples, which we refer to
857
as 'keys' throughout the code base. Revision_keys, inventory_keys and
858
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
859
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
860
byte string made up of a hash identifier and a hash value.
861
We use this interface because it allows low friction with the underlying
862
code that implements disk indices, network encoding and other parts of
863
bzrlib.
864
865
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
866
the serialised revisions for the repository. This can be used to obtain
867
revision graph information or to access raw serialised revisions.
868
The result of trying to insert data into the repository via this store
869
is undefined: it should be considered read-only except for implementors
870
of repositories.
871
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
872
the serialised signatures for the repository. This can be used to
873
obtain access to raw serialised signatures. The result of trying to
874
insert data into the repository via this store is undefined: it should
875
be considered read-only except for implementors of repositories.
876
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
877
the serialised inventories for the repository. This can be used to
878
obtain unserialised inventories. The result of trying to insert data
879
into the repository via this store is undefined: it should be
880
considered read-only except for implementors of repositories.
881
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
882
texts of files and directories for the repository. This can be used to
883
obtain file texts or file graphs. Note that Repository.iter_file_bytes
884
is usually a better interface for accessing file texts.
885
The result of trying to insert data into the repository via this store
886
is undefined: it should be considered read-only except for implementors
887
of repositories.
888
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
889
any data the repository chooses to store or have indexed by its hash.
890
The result of trying to insert data into the repository via this store
891
is undefined: it should be considered read-only except for implementors
892
of repositories.
893
:ivar _transport: Transport for file access to repository, typically
894
pointing to .bzr/repository.
895
"""
896
897
# What class to use for a CommitBuilder. Often it's simpler to change this
898
# in a Repository class subclass rather than to override
899
# get_commit_builder.
900
_commit_builder_class = VersionedFileCommitBuilder
901
902
def add_fallback_repository(self, repository):
903
"""Add a repository to use for looking up data not held locally.
904
905
:param repository: A repository.
906
"""
907
if not self._format.supports_external_lookups:
908
raise errors.UnstackableRepositoryFormat(self._format, self.base)
909
if self.is_locked():
910
# This repository will call fallback.unlock() when we transition to
911
# the unlocked state, so we make sure to increment the lock count
912
repository.lock_read()
913
self._check_fallback_repository(repository)
914
self._fallback_repositories.append(repository)
915
self.texts.add_fallback_versioned_files(repository.texts)
916
self.inventories.add_fallback_versioned_files(repository.inventories)
917
self.revisions.add_fallback_versioned_files(repository.revisions)
918
self.signatures.add_fallback_versioned_files(repository.signatures)
919
if self.chk_bytes is not None:
920
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
921
922
@only_raises(errors.LockNotHeld, errors.LockBroken)
923
def unlock(self):
924
super(VersionedFileRepository, self).unlock()
925
if self.control_files._lock_count == 0:
926
self._inventory_entry_cache.clear()
927
928
def add_inventory(self, revision_id, inv, parents):
929
"""Add the inventory inv to the repository as revision_id.
930
931
:param parents: The revision ids of the parents that revision_id
932
is known to have and are in the repository already.
933
934
:returns: The validator(which is a sha1 digest, though what is sha'd is
935
repository format specific) of the serialized inventory.
936
"""
937
if not self.is_in_write_group():
938
raise AssertionError("%r not in write group" % (self,))
939
_mod_revision.check_not_reserved_id(revision_id)
940
if not (inv.revision_id is None or inv.revision_id == revision_id):
941
raise AssertionError(
942
"Mismatch between inventory revision"
943
" id and insertion revid (%r, %r)"
944
% (inv.revision_id, revision_id))
945
if inv.root is None:
946
raise errors.RootMissing()
947
return self._add_inventory_checked(revision_id, inv, parents)
948
949
def _add_inventory_checked(self, revision_id, inv, parents):
950
"""Add inv to the repository after checking the inputs.
951
952
This function can be overridden to allow different inventory styles.
953
954
:seealso: add_inventory, for the contract.
955
"""
956
inv_lines = self._serializer.write_inventory_to_lines(inv)
957
return self._inventory_add_lines(revision_id, parents,
958
inv_lines, check_content=False)
959
960
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
961
parents, basis_inv=None, propagate_caches=False):
962
"""Add a new inventory expressed as a delta against another revision.
963
964
See the inventory developers documentation for the theory behind
965
inventory deltas.
966
967
:param basis_revision_id: The inventory id the delta was created
968
against. (This does not have to be a direct parent.)
969
:param delta: The inventory delta (see Inventory.apply_delta for
970
details).
971
:param new_revision_id: The revision id that the inventory is being
972
added for.
973
:param parents: The revision ids of the parents that revision_id is
974
known to have and are in the repository already. These are supplied
975
for repositories that depend on the inventory graph for revision
976
graph access, as well as for those that pun ancestry with delta
977
compression.
978
:param basis_inv: The basis inventory if it is already known,
979
otherwise None.
980
:param propagate_caches: If True, the caches for this inventory are
981
copied to and updated for the result if possible.
982
983
:returns: (validator, new_inv)
984
The validator(which is a sha1 digest, though what is sha'd is
985
repository format specific) of the serialized inventory, and the
986
resulting inventory.
987
"""
988
if not self.is_in_write_group():
989
raise AssertionError("%r not in write group" % (self,))
990
_mod_revision.check_not_reserved_id(new_revision_id)
991
basis_tree = self.revision_tree(basis_revision_id)
992
basis_tree.lock_read()
993
try:
994
# Note that this mutates the inventory of basis_tree, which not all
995
# inventory implementations may support: A better idiom would be to
996
# return a new inventory, but as there is no revision tree cache in
997
# repository this is safe for now - RBC 20081013
998
if basis_inv is None:
999
basis_inv = basis_tree.inventory
1000
basis_inv.apply_delta(delta)
1001
basis_inv.revision_id = new_revision_id
1002
return (self.add_inventory(new_revision_id, basis_inv, parents),
1003
basis_inv)
1004
finally:
1005
basis_tree.unlock()
1006
1007
def _inventory_add_lines(self, revision_id, parents, lines,
1008
check_content=True):
1009
"""Store lines in inv_vf and return the sha1 of the inventory."""
1010
parents = [(parent,) for parent in parents]
1011
result = self.inventories.add_lines((revision_id,), parents, lines,
1012
check_content=check_content)[0]
1013
self.inventories._access.flush()
1014
return result
1015
1016
def add_revision(self, revision_id, rev, inv=None, config=None):
1017
"""Add rev to the revision store as revision_id.
1018
1019
:param revision_id: the revision id to use.
1020
:param rev: The revision object.
1021
:param inv: The inventory for the revision. if None, it will be looked
1022
up in the inventory storer
1023
:param config: If None no digital signature will be created.
1024
If supplied its signature_needed method will be used
1025
to determine if a signature should be made.
1026
"""
1027
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1028
# rev.parent_ids?
1029
_mod_revision.check_not_reserved_id(revision_id)
1030
if config is not None and config.signature_needed():
1031
if inv is None:
1032
inv = self.get_inventory(revision_id)
1033
tree = InventoryRevisionTree(self, inv, revision_id)
1034
testament = Testament(rev, tree)
1035
plaintext = testament.as_short_text()
1036
self.store_revision_signature(
1037
gpg.GPGStrategy(config), plaintext, revision_id)
1038
# check inventory present
1039
if not self.inventories.get_parent_map([(revision_id,)]):
1040
if inv is None:
1041
raise errors.WeaveRevisionNotPresent(revision_id,
1042
self.inventories)
1043
else:
1044
# yes, this is not suitable for adding with ghosts.
1045
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1046
rev.parent_ids)
1047
else:
1048
key = (revision_id,)
1049
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1050
self._add_revision(rev)
1051
1052
def _add_revision(self, revision):
1053
text = self._serializer.write_revision_to_string(revision)
1054
key = (revision.revision_id,)
1055
parents = tuple((parent,) for parent in revision.parent_ids)
1056
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1057
1058
def _check_inventories(self, checker):
1059
"""Check the inventories found from the revision scan.
1060
1061
This is responsible for verifying the sha1 of inventories and
1062
creating a pending_keys set that covers data referenced by inventories.
1063
"""
1064
bar = ui.ui_factory.nested_progress_bar()
1065
try:
1066
self._do_check_inventories(checker, bar)
1067
finally:
1068
bar.finished()
1069
1070
def _do_check_inventories(self, checker, bar):
1071
"""Helper for _check_inventories."""
1072
revno = 0
1073
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1074
kinds = ['chk_bytes', 'texts']
1075
count = len(checker.pending_keys)
1076
bar.update("inventories", 0, 2)
1077
current_keys = checker.pending_keys
1078
checker.pending_keys = {}
1079
# Accumulate current checks.
1080
for key in current_keys:
1081
if key[0] != 'inventories' and key[0] not in kinds:
1082
checker._report_items.append('unknown key type %r' % (key,))
1083
keys[key[0]].add(key[1:])
1084
if keys['inventories']:
1085
# NB: output order *should* be roughly sorted - topo or
1086
# inverse topo depending on repository - either way decent
1087
# to just delta against. However, pre-CHK formats didn't
1088
# try to optimise inventory layout on disk. As such the
1089
# pre-CHK code path does not use inventory deltas.
1090
last_object = None
1091
for record in self.inventories.check(keys=keys['inventories']):
1092
if record.storage_kind == 'absent':
1093
checker._report_items.append(
1094
'Missing inventory {%s}' % (record.key,))
1095
else:
1096
last_object = self._check_record('inventories', record,
1097
checker, last_object,
1098
current_keys[('inventories',) + record.key])
1099
del keys['inventories']
1100
else:
1101
return
1102
bar.update("texts", 1)
1103
while (checker.pending_keys or keys['chk_bytes']
1104
or keys['texts']):
1105
# Something to check.
1106
current_keys = checker.pending_keys
1107
checker.pending_keys = {}
1108
# Accumulate current checks.
1109
for key in current_keys:
1110
if key[0] not in kinds:
1111
checker._report_items.append('unknown key type %r' % (key,))
1112
keys[key[0]].add(key[1:])
1113
# Check the outermost kind only - inventories || chk_bytes || texts
1114
for kind in kinds:
1115
if keys[kind]:
1116
last_object = None
1117
for record in getattr(self, kind).check(keys=keys[kind]):
1118
if record.storage_kind == 'absent':
1119
checker._report_items.append(
1120
'Missing %s {%s}' % (kind, record.key,))
1121
else:
1122
last_object = self._check_record(kind, record,
1123
checker, last_object, current_keys[(kind,) + record.key])
1124
keys[kind] = set()
1125
break
1126
1127
def _check_record(self, kind, record, checker, last_object, item_data):
1128
"""Check a single text from this repository."""
1129
if kind == 'inventories':
1130
rev_id = record.key[0]
1131
inv = self._deserialise_inventory(rev_id,
1132
record.get_bytes_as('fulltext'))
1133
if last_object is not None:
1134
delta = inv._make_delta(last_object)
1135
for old_path, path, file_id, ie in delta:
1136
if ie is None:
1137
continue
1138
ie.check(checker, rev_id, inv)
1139
else:
1140
for path, ie in inv.iter_entries():
1141
ie.check(checker, rev_id, inv)
1142
if self._format.fast_deltas:
1143
return inv
1144
elif kind == 'chk_bytes':
1145
# No code written to check chk_bytes for this repo format.
1146
checker._report_items.append(
1147
'unsupported key type chk_bytes for %s' % (record.key,))
1148
elif kind == 'texts':
1149
self._check_text(record, checker, item_data)
1150
else:
1151
checker._report_items.append(
1152
'unknown key type %s for %s' % (kind, record.key))
1153
1154
def _check_text(self, record, checker, item_data):
1155
"""Check a single text."""
1156
# Check it is extractable.
1157
# TODO: check length.
1158
if record.storage_kind == 'chunked':
1159
chunks = record.get_bytes_as(record.storage_kind)
1160
sha1 = osutils.sha_strings(chunks)
1161
length = sum(map(len, chunks))
1162
else:
1163
content = record.get_bytes_as('fulltext')
1164
sha1 = osutils.sha_string(content)
1165
length = len(content)
1166
if item_data and sha1 != item_data[1]:
1167
checker._report_items.append(
1168
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1169
(record.key, sha1, item_data[1], item_data[2]))
1170
1171
def __init__(self, _format, a_bzrdir, control_files):
1172
"""Instantiate a VersionedFileRepository.
1173
1174
:param _format: The format of the repository on disk.
1175
:param a_bzrdir: The BzrDir of the repository.
1176
:param control_files: Control files to use for locking, etc.
1177
"""
1178
# In the future we will have a single api for all stores for
1179
# getting file texts, inventories and revisions, then
1180
# this construct will accept instances of those things.
1181
super(VersionedFileRepository, self).__init__(_format, a_bzrdir,
1182
control_files)
1183
# for tests
1184
self._reconcile_does_inventory_gc = True
1185
self._reconcile_fixes_text_parents = False
1186
self._reconcile_backsup_inventory = True
1187
# An InventoryEntry cache, used during deserialization
1188
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1189
# Is it safe to return inventory entries directly from the entry cache,
1190
# rather copying them?
1191
self._safe_to_return_from_cache = False
1192
1193
@needs_read_lock
1194
def gather_stats(self, revid=None, committers=None):
1195
"""See Repository.gather_stats()."""
1196
result = super(VersionedFileRepository, self).gather_stats(revid, committers)
1197
# now gather global repository information
1198
# XXX: This is available for many repos regardless of listability.
1199
if self.user_transport.listable():
1200
# XXX: do we want to __define len__() ?
1201
# Maybe the versionedfiles object should provide a different
1202
# method to get the number of keys.
1203
result['revisions'] = len(self.revisions.keys())
1204
# result['size'] = t
1205
return result
1206
1207
1208
def get_commit_builder(self, branch, parents, config, timestamp=None,
1209
timezone=None, committer=None, revprops=None,
1210
revision_id=None, lossy=False):
1211
"""Obtain a CommitBuilder for this repository.
1212
1213
:param branch: Branch to commit to.
1214
:param parents: Revision ids of the parents of the new revision.
1215
:param config: Configuration to use.
1216
:param timestamp: Optional timestamp recorded for commit.
1217
:param timezone: Optional timezone for timestamp.
1218
:param committer: Optional committer to set for commit.
1219
:param revprops: Optional dictionary of revision properties.
1220
:param revision_id: Optional revision id.
1221
:param lossy: Whether to discard data that can not be natively
1222
represented, when pushing to a foreign VCS
1223
"""
1224
if self._fallback_repositories and not self._format.supports_chks:
1225
raise errors.BzrError("Cannot commit directly to a stacked branch"
1226
" in pre-2a formats. See "
1227
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1228
result = self._commit_builder_class(self, parents, config,
1229
timestamp, timezone, committer, revprops, revision_id,
1230
lossy)
1231
self.start_write_group()
1232
return result
1233
1234
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1235
"""Return the keys of missing inventory parents for revisions added in
1236
this write group.
1237
1238
A revision is not complete if the inventory delta for that revision
1239
cannot be calculated. Therefore if the parent inventories of a
1240
revision are not present, the revision is incomplete, and e.g. cannot
1241
be streamed by a smart server. This method finds missing inventory
1242
parents for revisions added in this write group.
1243
"""
1244
if not self._format.supports_external_lookups:
1245
# This is only an issue for stacked repositories
1246
return set()
1247
if not self.is_in_write_group():
1248
raise AssertionError('not in a write group')
1249
1250
# XXX: We assume that every added revision already has its
1251
# corresponding inventory, so we only check for parent inventories that
1252
# might be missing, rather than all inventories.
1253
parents = set(self.revisions._index.get_missing_parents())
1254
parents.discard(_mod_revision.NULL_REVISION)
1255
unstacked_inventories = self.inventories._index
1256
present_inventories = unstacked_inventories.get_parent_map(
1257
key[-1:] for key in parents)
1258
parents.difference_update(present_inventories)
1259
if len(parents) == 0:
1260
# No missing parent inventories.
1261
return set()
1262
if not check_for_missing_texts:
1263
return set(('inventories', rev_id) for (rev_id,) in parents)
1264
# Ok, now we have a list of missing inventories. But these only matter
1265
# if the inventories that reference them are missing some texts they
1266
# appear to introduce.
1267
# XXX: Texts referenced by all added inventories need to be present,
1268
# but at the moment we're only checking for texts referenced by
1269
# inventories at the graph's edge.
1270
key_deps = self.revisions._index._key_dependencies
1271
key_deps.satisfy_refs_for_keys(present_inventories)
1272
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1273
file_ids = self.fileids_altered_by_revision_ids(referrers)
1274
missing_texts = set()
1275
for file_id, version_ids in file_ids.iteritems():
1276
missing_texts.update(
1277
(file_id, version_id) for version_id in version_ids)
1278
present_texts = self.texts.get_parent_map(missing_texts)
1279
missing_texts.difference_update(present_texts)
1280
if not missing_texts:
1281
# No texts are missing, so all revisions and their deltas are
1282
# reconstructable.
1283
return set()
1284
# Alternatively the text versions could be returned as the missing
1285
# keys, but this is likely to be less data.
1286
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1287
return missing_keys
1288
1289
@needs_read_lock
1290
def has_revisions(self, revision_ids):
1291
"""Probe to find out the presence of multiple revisions.
1292
1293
:param revision_ids: An iterable of revision_ids.
1294
:return: A set of the revision_ids that were present.
1295
"""
1296
parent_map = self.revisions.get_parent_map(
1297
[(rev_id,) for rev_id in revision_ids])
1298
result = set()
1299
if _mod_revision.NULL_REVISION in revision_ids:
1300
result.add(_mod_revision.NULL_REVISION)
1301
result.update([key[0] for key in parent_map])
1302
return result
1303
1304
@needs_read_lock
1305
def get_revision_reconcile(self, revision_id):
1306
"""'reconcile' helper routine that allows access to a revision always.
1307
1308
This variant of get_revision does not cross check the weave graph
1309
against the revision one as get_revision does: but it should only
1310
be used by reconcile, or reconcile-alike commands that are correcting
1311
or testing the revision graph.
1312
"""
1313
return self._get_revisions([revision_id])[0]
1314
1315
@needs_read_lock
1316
def get_revisions(self, revision_ids):
1317
"""Get many revisions at once.
1318
1319
Repositories that need to check data on every revision read should
1320
subclass this method.
1321
"""
1322
return self._get_revisions(revision_ids)
1323
1324
@needs_read_lock
1325
def _get_revisions(self, revision_ids):
1326
"""Core work logic to get many revisions without sanity checks."""
1327
revs = {}
1328
for revid, rev in self._iter_revisions(revision_ids):
1329
if rev is None:
1330
raise errors.NoSuchRevision(self, revid)
1331
revs[revid] = rev
1332
return [revs[revid] for revid in revision_ids]
1333
1334
def _iter_revisions(self, revision_ids):
1335
"""Iterate over revision objects.
1336
1337
:param revision_ids: An iterable of revisions to examine. None may be
1338
passed to request all revisions known to the repository. Note that
1339
not all repositories can find unreferenced revisions; for those
1340
repositories only referenced ones will be returned.
1341
:return: An iterator of (revid, revision) tuples. Absent revisions (
1342
those asked for but not available) are returned as (revid, None).
1343
"""
1344
if revision_ids is None:
1345
revision_ids = self.all_revision_ids()
1346
else:
1347
for rev_id in revision_ids:
1348
if not rev_id or not isinstance(rev_id, basestring):
1349
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1350
keys = [(key,) for key in revision_ids]
1351
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1352
for record in stream:
1353
revid = record.key[0]
1354
if record.storage_kind == 'absent':
1355
yield (revid, None)
1356
else:
1357
text = record.get_bytes_as('fulltext')
1358
rev = self._serializer.read_revision_from_string(text)
1359
yield (revid, rev)
1360
1361
@needs_write_lock
1362
def add_signature_text(self, revision_id, signature):
1363
"""Store a signature text for a revision.
1364
1365
:param revision_id: Revision id of the revision
1366
:param signature: Signature text.
1367
"""
1368
self.signatures.add_lines((revision_id,), (),
1369
osutils.split_lines(signature))
1370
1371
def find_text_key_references(self):
1372
"""Find the text key references within the repository.
1373
1374
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1375
to whether they were referred to by the inventory of the
1376
revision_id that they contain. The inventory texts from all present
1377
revision ids are assessed to generate this report.
1378
"""
1379
revision_keys = self.revisions.keys()
1380
w = self.inventories
1381
pb = ui.ui_factory.nested_progress_bar()
1382
try:
1383
return self._serializer._find_text_key_references(
1384
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1385
finally:
1386
pb.finished()
1387
1388
def _inventory_xml_lines_for_keys(self, keys):
1389
"""Get a line iterator of the sort needed for findind references.
1390
1391
Not relevant for non-xml inventory repositories.
1392
1393
Ghosts in revision_keys are ignored.
1394
1395
:param revision_keys: The revision keys for the inventories to inspect.
1396
:return: An iterator over (inventory line, revid) for the fulltexts of
1397
all of the xml inventories specified by revision_keys.
1398
"""
1399
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1400
for record in stream:
1401
if record.storage_kind != 'absent':
1402
chunks = record.get_bytes_as('chunked')
1403
revid = record.key[-1]
1404
lines = osutils.chunks_to_lines(chunks)
1405
for line in lines:
1406
yield line, revid
1407
1408
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1409
revision_keys):
1410
"""Helper routine for fileids_altered_by_revision_ids.
1411
1412
This performs the translation of xml lines to revision ids.
1413
1414
:param line_iterator: An iterator of lines, origin_version_id
1415
:param revision_keys: The revision ids to filter for. This should be a
1416
set or other type which supports efficient __contains__ lookups, as
1417
the revision key from each parsed line will be looked up in the
1418
revision_keys filter.
1419
:return: a dictionary mapping altered file-ids to an iterable of
1420
revision_ids. Each altered file-ids has the exact revision_ids that
1421
altered it listed explicitly.
1422
"""
1423
seen = set(self._serializer._find_text_key_references(
1424
line_iterator).iterkeys())
1425
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
1426
parent_seen = set(self._serializer._find_text_key_references(
1427
self._inventory_xml_lines_for_keys(parent_keys)))
1428
new_keys = seen - parent_seen
1429
result = {}
1430
setdefault = result.setdefault
1431
for key in new_keys:
1432
setdefault(key[0], set()).add(key[-1])
1433
return result
1434
1435
def _find_parent_keys_of_revisions(self, revision_keys):
1436
"""Similar to _find_parent_ids_of_revisions, but used with keys.
1437
1438
:param revision_keys: An iterable of revision_keys.
1439
:return: The parents of all revision_keys that are not already in
1440
revision_keys
1441
"""
1442
parent_map = self.revisions.get_parent_map(revision_keys)
1443
parent_keys = set()
1444
map(parent_keys.update, parent_map.itervalues())
1445
parent_keys.difference_update(revision_keys)
1446
parent_keys.discard(_mod_revision.NULL_REVISION)
1447
return parent_keys
1448
1449
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
1450
"""Find the file ids and versions affected by revisions.
1451
1452
:param revisions: an iterable containing revision ids.
1453
:param _inv_weave: The inventory weave from this repository or None.
1454
If None, the inventory weave will be opened automatically.
1455
:return: a dictionary mapping altered file-ids to an iterable of
1456
revision_ids. Each altered file-ids has the exact revision_ids that
1457
altered it listed explicitly.
1458
"""
1459
selected_keys = set((revid,) for revid in revision_ids)
1460
w = _inv_weave or self.inventories
1461
return self._find_file_ids_from_xml_inventory_lines(
1462
w.iter_lines_added_or_present_in_keys(
1463
selected_keys, pb=None),
1464
selected_keys)
1465
1466
def iter_files_bytes(self, desired_files):
1467
"""Iterate through file versions.
1468
1469
Files will not necessarily be returned in the order they occur in
1470
desired_files. No specific order is guaranteed.
1471
1472
Yields pairs of identifier, bytes_iterator. identifier is an opaque
1473
value supplied by the caller as part of desired_files. It should
1474
uniquely identify the file version in the caller's context. (Examples:
1475
an index number or a TreeTransform trans_id.)
1476
1477
bytes_iterator is an iterable of bytestrings for the file. The
1478
kind of iterable and length of the bytestrings are unspecified, but for
1479
this implementation, it is a list of bytes produced by
1480
VersionedFile.get_record_stream().
1481
1482
:param desired_files: a list of (file_id, revision_id, identifier)
1483
triples
1484
"""
1485
text_keys = {}
1486
for file_id, revision_id, callable_data in desired_files:
1487
text_keys[(file_id, revision_id)] = callable_data
1488
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
1489
if record.storage_kind == 'absent':
1490
raise errors.RevisionNotPresent(record.key, self)
1491
yield text_keys[record.key], record.get_bytes_as('chunked')
1492
1493
def _generate_text_key_index(self, text_key_references=None,
1494
ancestors=None):
1495
"""Generate a new text key index for the repository.
1496
1497
This is an expensive function that will take considerable time to run.
1498
1499
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
1500
list of parents, also text keys. When a given key has no parents,
1501
the parents list will be [NULL_REVISION].
1502
"""
1503
# All revisions, to find inventory parents.
1504
if ancestors is None:
1505
graph = self.get_graph()
1506
ancestors = graph.get_parent_map(self.all_revision_ids())
1507
if text_key_references is None:
1508
text_key_references = self.find_text_key_references()
1509
pb = ui.ui_factory.nested_progress_bar()
1510
try:
1511
return self._do_generate_text_key_index(ancestors,
1512
text_key_references, pb)
1513
finally:
1514
pb.finished()
1515
1516
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
1517
"""Helper for _generate_text_key_index to avoid deep nesting."""
1518
revision_order = tsort.topo_sort(ancestors)
1519
invalid_keys = set()
1520
revision_keys = {}
1521
for revision_id in revision_order:
1522
revision_keys[revision_id] = set()
1523
text_count = len(text_key_references)
1524
# a cache of the text keys to allow reuse; costs a dict of all the
1525
# keys, but saves a 2-tuple for every child of a given key.
1526
text_key_cache = {}
1527
for text_key, valid in text_key_references.iteritems():
1528
if not valid:
1529
invalid_keys.add(text_key)
1530
else:
1531
revision_keys[text_key[1]].add(text_key)
1532
text_key_cache[text_key] = text_key
1533
del text_key_references
1534
text_index = {}
1535
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
1536
NULL_REVISION = _mod_revision.NULL_REVISION
1537
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
1538
# too small for large or very branchy trees. However, for 55K path
1539
# trees, it would be easy to use too much memory trivially. Ideally we
1540
# could gauge this by looking at available real memory etc, but this is
1541
# always a tricky proposition.
1542
inventory_cache = lru_cache.LRUCache(10)
1543
batch_size = 10 # should be ~150MB on a 55K path tree
1544
batch_count = len(revision_order) / batch_size + 1
1545
processed_texts = 0
1546
pb.update("Calculating text parents", processed_texts, text_count)
1547
for offset in xrange(batch_count):
1548
to_query = revision_order[offset * batch_size:(offset + 1) *
1549
batch_size]
1550
if not to_query:
1551
break
1552
for revision_id in to_query:
1553
parent_ids = ancestors[revision_id]
1554
for text_key in revision_keys[revision_id]:
1555
pb.update("Calculating text parents", processed_texts)
1556
processed_texts += 1
1557
candidate_parents = []
1558
for parent_id in parent_ids:
1559
parent_text_key = (text_key[0], parent_id)
1560
try:
1561
check_parent = parent_text_key not in \
1562
revision_keys[parent_id]
1563
except KeyError:
1564
# the parent parent_id is a ghost:
1565
check_parent = False
1566
# truncate the derived graph against this ghost.
1567
parent_text_key = None
1568
if check_parent:
1569
# look at the parent commit details inventories to
1570
# determine possible candidates in the per file graph.
1571
# TODO: cache here.
1572
try:
1573
inv = inventory_cache[parent_id]
1574
except KeyError:
1575
inv = self.revision_tree(parent_id).inventory
1576
inventory_cache[parent_id] = inv
1577
try:
1578
parent_entry = inv[text_key[0]]
1579
except (KeyError, errors.NoSuchId):
1580
parent_entry = None
1581
if parent_entry is not None:
1582
parent_text_key = (
1583
text_key[0], parent_entry.revision)
1584
else:
1585
parent_text_key = None
1586
if parent_text_key is not None:
1587
candidate_parents.append(
1588
text_key_cache[parent_text_key])
1589
parent_heads = text_graph.heads(candidate_parents)
1590
new_parents = list(parent_heads)
1591
new_parents.sort(key=lambda x:candidate_parents.index(x))
1592
if new_parents == []:
1593
new_parents = [NULL_REVISION]
1594
text_index[text_key] = new_parents
1595
1596
for text_key in invalid_keys:
1597
text_index[text_key] = [NULL_REVISION]
1598
return text_index
1599
1600
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
1601
"""Get an iterable listing the keys of all the data introduced by a set
1602
of revision IDs.
1603
1604
The keys will be ordered so that the corresponding items can be safely
1605
fetched and inserted in that order.
1606
1607
:returns: An iterable producing tuples of (knit-kind, file-id,
1608
versions). knit-kind is one of 'file', 'inventory', 'signatures',
1609
'revisions'. file-id is None unless knit-kind is 'file'.
1610
"""
1611
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
1612
yield result
1613
del _files_pb
1614
for result in self._find_non_file_keys_to_fetch(revision_ids):
1615
yield result
1616
1617
def _find_file_keys_to_fetch(self, revision_ids, pb):
1618
# XXX: it's a bit weird to control the inventory weave caching in this
1619
# generator. Ideally the caching would be done in fetch.py I think. Or
1620
# maybe this generator should explicitly have the contract that it
1621
# should not be iterated until the previously yielded item has been
1622
# processed?
1623
inv_w = self.inventories
1624
1625
# file ids that changed
1626
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
1627
count = 0
1628
num_file_ids = len(file_ids)
1629
for file_id, altered_versions in file_ids.iteritems():
1630
if pb is not None:
1631
pb.update("Fetch texts", count, num_file_ids)
1632
count += 1
1633
yield ("file", file_id, altered_versions)
1634
1635
def _find_non_file_keys_to_fetch(self, revision_ids):
1636
# inventory
1637
yield ("inventory", None, revision_ids)
1638
1639
# signatures
1640
# XXX: Note ATM no callers actually pay attention to this return
1641
# instead they just use the list of revision ids and ignore
1642
# missing sigs. Consider removing this work entirely
1643
revisions_with_signatures = set(self.signatures.get_parent_map(
1644
[(r,) for r in revision_ids]))
1645
revisions_with_signatures = set(
1646
[r for (r,) in revisions_with_signatures])
1647
revisions_with_signatures.intersection_update(revision_ids)
1648
yield ("signatures", None, revisions_with_signatures)
1649
1650
# revisions
1651
yield ("revisions", None, revision_ids)
1652
1653
@needs_read_lock
1654
def get_inventory(self, revision_id):
1655
"""Get Inventory object by revision id."""
1656
return self.iter_inventories([revision_id]).next()
1657
1658
def iter_inventories(self, revision_ids, ordering=None):
1659
"""Get many inventories by revision_ids.
1660
1661
This will buffer some or all of the texts used in constructing the
1662
inventories in memory, but will only parse a single inventory at a
1663
time.
1664
1665
:param revision_ids: The expected revision ids of the inventories.
1666
:param ordering: optional ordering, e.g. 'topological'. If not
1667
specified, the order of revision_ids will be preserved (by
1668
buffering if necessary).
1669
:return: An iterator of inventories.
1670
"""
1671
if ((None in revision_ids)
1672
or (_mod_revision.NULL_REVISION in revision_ids)):
1673
raise ValueError('cannot get null revision inventory')
1674
return self._iter_inventories(revision_ids, ordering)
1675
1676
def _iter_inventories(self, revision_ids, ordering):
1677
"""single-document based inventory iteration."""
1678
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
1679
for text, revision_id in inv_xmls:
1680
yield self._deserialise_inventory(revision_id, text)
1681
1682
def _iter_inventory_xmls(self, revision_ids, ordering):
1683
if ordering is None:
1684
order_as_requested = True
1685
ordering = 'unordered'
1686
else:
1687
order_as_requested = False
1688
keys = [(revision_id,) for revision_id in revision_ids]
1689
if not keys:
1690
return
1691
if order_as_requested:
1692
key_iter = iter(keys)
1693
next_key = key_iter.next()
1694
stream = self.inventories.get_record_stream(keys, ordering, True)
1695
text_chunks = {}
1696
for record in stream:
1697
if record.storage_kind != 'absent':
1698
chunks = record.get_bytes_as('chunked')
1699
if order_as_requested:
1700
text_chunks[record.key] = chunks
1701
else:
1702
yield ''.join(chunks), record.key[-1]
1703
else:
1704
raise errors.NoSuchRevision(self, record.key)
1705
if order_as_requested:
1706
# Yield as many results as we can while preserving order.
1707
while next_key in text_chunks:
1708
chunks = text_chunks.pop(next_key)
1709
yield ''.join(chunks), next_key[-1]
1710
try:
1711
next_key = key_iter.next()
1712
except StopIteration:
1713
# We still want to fully consume the get_record_stream,
1714
# just in case it is not actually finished at this point
1715
next_key = None
1716
break
1717
1718
def _deserialise_inventory(self, revision_id, xml):
1719
"""Transform the xml into an inventory object.
1720
1721
:param revision_id: The expected revision id of the inventory.
1722
:param xml: A serialised inventory.
1723
"""
1724
result = self._serializer.read_inventory_from_string(xml, revision_id,
1725
entry_cache=self._inventory_entry_cache,
1726
return_from_cache=self._safe_to_return_from_cache)
1727
if result.revision_id != revision_id:
1728
raise AssertionError('revision id mismatch %s != %s' % (
1729
result.revision_id, revision_id))
1730
return result
1731
1732
def get_serializer_format(self):
1733
return self._serializer.format_num
1734
1735
@needs_read_lock
1736
def _get_inventory_xml(self, revision_id):
1737
"""Get serialized inventory as a string."""
1738
texts = self._iter_inventory_xmls([revision_id], 'unordered')
1739
try:
1740
text, revision_id = texts.next()
1741
except StopIteration:
1742
raise errors.HistoryMissing(self, 'inventory', revision_id)
1743
return text
1744
1745
@needs_read_lock
1746
def revision_tree(self, revision_id):
1747
"""Return Tree for a revision on this branch.
1748
1749
`revision_id` may be NULL_REVISION for the empty tree revision.
1750
"""
1751
revision_id = _mod_revision.ensure_null(revision_id)
1752
# TODO: refactor this to use an existing revision object
1753
# so we don't need to read it in twice.
1754
if revision_id == _mod_revision.NULL_REVISION:
1755
return InventoryRevisionTree(self,
1756
Inventory(root_id=None), _mod_revision.NULL_REVISION)
1757
else:
1758
inv = self.get_inventory(revision_id)
1759
return InventoryRevisionTree(self, inv, revision_id)
1760
1761
def revision_trees(self, revision_ids):
1762
"""Return Trees for revisions in this repository.
1763
1764
:param revision_ids: a sequence of revision-ids;
1765
a revision-id may not be None or 'null:'
1766
"""
1767
inventories = self.iter_inventories(revision_ids)
1768
for inv in inventories:
1769
yield InventoryRevisionTree(self, inv, inv.revision_id)
1770
1771
def _filtered_revision_trees(self, revision_ids, file_ids):
1772
"""Return Tree for a revision on this branch with only some files.
1773
1774
:param revision_ids: a sequence of revision-ids;
1775
a revision-id may not be None or 'null:'
1776
:param file_ids: if not None, the result is filtered
1777
so that only those file-ids, their parents and their
1778
children are included.
1779
"""
1780
inventories = self.iter_inventories(revision_ids)
1781
for inv in inventories:
1782
# Should we introduce a FilteredRevisionTree class rather
1783
# than pre-filter the inventory here?
1784
filtered_inv = inv.filter(file_ids)
1785
yield InventoryRevisionTree(self, filtered_inv, filtered_inv.revision_id)
1786
1787
def get_parent_map(self, revision_ids):
1788
"""See graph.StackedParentsProvider.get_parent_map"""
1789
# revisions index works in keys; this just works in revisions
1790
# therefore wrap and unwrap
1791
query_keys = []
1792
result = {}
1793
for revision_id in revision_ids:
1794
if revision_id == _mod_revision.NULL_REVISION:
1795
result[revision_id] = ()
1796
elif revision_id is None:
1797
raise ValueError('get_parent_map(None) is not valid')
1798
else:
1799
query_keys.append((revision_id ,))
1800
for ((revision_id,), parent_keys) in \
1801
self.revisions.get_parent_map(query_keys).iteritems():
1802
if parent_keys:
1803
result[revision_id] = tuple([parent_revid
1804
for (parent_revid,) in parent_keys])
1805
else:
1806
result[revision_id] = (_mod_revision.NULL_REVISION,)
1807
return result
1808
1809
@needs_read_lock
1810
def get_known_graph_ancestry(self, revision_ids):
1811
"""Return the known graph for a set of revision ids and their ancestors.
1812
"""
1813
st = static_tuple.StaticTuple
1814
revision_keys = [st(r_id).intern() for r_id in revision_ids]
1815
known_graph = self.revisions.get_known_graph_ancestry(revision_keys)
1816
return graph.GraphThunkIdsToKeys(known_graph)
1817
1818
def _get_versioned_file_checker(self, text_key_references=None,
1819
ancestors=None):
1820
"""Return an object suitable for checking versioned files.
1821
1822
:param text_key_references: if non-None, an already built
1823
dictionary mapping text keys ((fileid, revision_id) tuples)
1824
to whether they were referred to by the inventory of the
1825
revision_id that they contain. If None, this will be
1826
calculated.
1827
:param ancestors: Optional result from
1828
self.get_graph().get_parent_map(self.all_revision_ids()) if already
1829
available.
1830
"""
1831
return _VersionedFileChecker(self,
1832
text_key_references=text_key_references, ancestors=ancestors)
1833
1834
@needs_read_lock
1835
def has_signature_for_revision_id(self, revision_id):
1836
"""Query for a revision signature for revision_id in the repository."""
1837
if not self.has_revision(revision_id):
1838
raise errors.NoSuchRevision(self, revision_id)
1839
sig_present = (1 == len(
1840
self.signatures.get_parent_map([(revision_id,)])))
1841
return sig_present
1842
1843
@needs_read_lock
1844
def get_signature_text(self, revision_id):
1845
"""Return the text for a signature."""
1846
stream = self.signatures.get_record_stream([(revision_id,)],
1847
'unordered', True)
1848
record = stream.next()
1849
if record.storage_kind == 'absent':
1850
raise errors.NoSuchRevision(self, revision_id)
1851
return record.get_bytes_as('fulltext')
1852
1853
def _find_inconsistent_revision_parents(self, revisions_iterator=None):
1854
"""Find revisions with different parent lists in the revision object
1855
and in the index graph.
1856
1857
:param revisions_iterator: None, or an iterator of (revid,
1858
Revision-or-None). This iterator controls the revisions checked.
1859
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1860
parents-in-revision).
1861
"""
1862
if not self.is_locked():
1863
raise AssertionError()
1864
vf = self.revisions
1865
if revisions_iterator is None:
1866
revisions_iterator = self._iter_revisions(None)
1867
for revid, revision in revisions_iterator:
1868
if revision is None:
1869
pass
1870
parent_map = vf.get_parent_map([(revid,)])
1871
parents_according_to_index = tuple(parent[-1] for parent in
1872
parent_map[(revid,)])
1873
parents_according_to_revision = tuple(revision.parent_ids)
1874
if parents_according_to_index != parents_according_to_revision:
1875
yield (revid, parents_according_to_index,
1876
parents_according_to_revision)
1877
1878
def _check_for_inconsistent_revision_parents(self):
1879
inconsistencies = list(self._find_inconsistent_revision_parents())
1880
if inconsistencies:
1881
raise errors.BzrCheckError(
1882
"Revision knit has inconsistent parents.")
1883
1884
def _get_sink(self):
1885
"""Return a sink for streaming into this repository."""
1886
return StreamSink(self)
1887
1888
def _get_source(self, to_format):
1889
"""Return a source for streaming from this repository."""
1890
return StreamSource(self, to_format)
1891
1892
1893
class MetaDirVersionedFileRepository(MetaDirRepository,
1894
VersionedFileRepository):
1895
"""Repositories in a meta-dir, that work via versioned file objects."""
1896
1897
def __init__(self, _format, a_bzrdir, control_files):
1898
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
1899
control_files)
1900
1901
1902
class MetaDirVersionedFileRepositoryFormat(MetaDirRepositoryFormat,
1903
VersionedFileRepositoryFormat):
1904
"""Base class for repository formats using versioned files in metadirs."""
1905
1906
1907
class StreamSink(object):
1908
"""An object that can insert a stream into a repository.
1909
1910
This interface handles the complexity of reserialising inventories and
1911
revisions from different formats, and allows unidirectional insertion into
1912
stacked repositories without looking for the missing basis parents
1913
beforehand.
1914
"""
1915
1916
def __init__(self, target_repo):
1917
self.target_repo = target_repo
1918
1919
def insert_stream(self, stream, src_format, resume_tokens):
1920
"""Insert a stream's content into the target repository.
1921
1922
:param src_format: a bzr repository format.
1923
1924
:return: a list of resume tokens and an iterable of keys additional
1925
items required before the insertion can be completed.
1926
"""
1927
self.target_repo.lock_write()
1928
try:
1929
if resume_tokens:
1930
self.target_repo.resume_write_group(resume_tokens)
1931
is_resume = True
1932
else:
1933
self.target_repo.start_write_group()
1934
is_resume = False
1935
try:
1936
# locked_insert_stream performs a commit|suspend.
1937
missing_keys = self.insert_stream_without_locking(stream,
1938
src_format, is_resume)
1939
if missing_keys:
1940
# suspend the write group and tell the caller what we is
1941
# missing. We know we can suspend or else we would not have
1942
# entered this code path. (All repositories that can handle
1943
# missing keys can handle suspending a write group).
1944
write_group_tokens = self.target_repo.suspend_write_group()
1945
return write_group_tokens, missing_keys
1946
hint = self.target_repo.commit_write_group()
1947
to_serializer = self.target_repo._format._serializer
1948
src_serializer = src_format._serializer
1949
if (to_serializer != src_serializer and
1950
self.target_repo._format.pack_compresses):
1951
self.target_repo.pack(hint=hint)
1952
return [], set()
1953
except:
1954
self.target_repo.abort_write_group(suppress_errors=True)
1955
raise
1956
finally:
1957
self.target_repo.unlock()
1958
1959
def insert_stream_without_locking(self, stream, src_format,
1960
is_resume=False):
1961
"""Insert a stream's content into the target repository.
1962
1963
This assumes that you already have a locked repository and an active
1964
write group.
1965
1966
:param src_format: a bzr repository format.
1967
:param is_resume: Passed down to get_missing_parent_inventories to
1968
indicate if we should be checking for missing texts at the same
1969
time.
1970
1971
:return: A set of keys that are missing.
1972
"""
1973
if not self.target_repo.is_write_locked():
1974
raise errors.ObjectNotLocked(self)
1975
if not self.target_repo.is_in_write_group():
1976
raise errors.BzrError('you must already be in a write group')
1977
to_serializer = self.target_repo._format._serializer
1978
src_serializer = src_format._serializer
1979
new_pack = None
1980
if to_serializer == src_serializer:
1981
# If serializers match and the target is a pack repository, set the
1982
# write cache size on the new pack. This avoids poor performance
1983
# on transports where append is unbuffered (such as
1984
# RemoteTransport). This is safe to do because nothing should read
1985
# back from the target repository while a stream with matching
1986
# serialization is being inserted.
1987
# The exception is that a delta record from the source that should
1988
# be a fulltext may need to be expanded by the target (see
1989
# test_fetch_revisions_with_deltas_into_pack); but we take care to
1990
# explicitly flush any buffered writes first in that rare case.
1991
try:
1992
new_pack = self.target_repo._pack_collection._new_pack
1993
except AttributeError:
1994
# Not a pack repository
1995
pass
1996
else:
1997
new_pack.set_write_cache_size(1024*1024)
1998
for substream_type, substream in stream:
1999
if 'stream' in debug.debug_flags:
2000
mutter('inserting substream: %s', substream_type)
2001
if substream_type == 'texts':
2002
self.target_repo.texts.insert_record_stream(substream)
2003
elif substream_type == 'inventories':
2004
if src_serializer == to_serializer:
2005
self.target_repo.inventories.insert_record_stream(
2006
substream)
2007
else:
2008
self._extract_and_insert_inventories(
2009
substream, src_serializer)
2010
elif substream_type == 'inventory-deltas':
2011
self._extract_and_insert_inventory_deltas(
2012
substream, src_serializer)
2013
elif substream_type == 'chk_bytes':
2014
# XXX: This doesn't support conversions, as it assumes the
2015
# conversion was done in the fetch code.
2016
self.target_repo.chk_bytes.insert_record_stream(substream)
2017
elif substream_type == 'revisions':
2018
# This may fallback to extract-and-insert more often than
2019
# required if the serializers are different only in terms of
2020
# the inventory.
2021
if src_serializer == to_serializer:
2022
self.target_repo.revisions.insert_record_stream(substream)
2023
else:
2024
self._extract_and_insert_revisions(substream,
2025
src_serializer)
2026
elif substream_type == 'signatures':
2027
self.target_repo.signatures.insert_record_stream(substream)
2028
else:
2029
raise AssertionError('kaboom! %s' % (substream_type,))
2030
# Done inserting data, and the missing_keys calculations will try to
2031
# read back from the inserted data, so flush the writes to the new pack
2032
# (if this is pack format).
2033
if new_pack is not None:
2034
new_pack._write_data('', flush=True)
2035
# Find all the new revisions (including ones from resume_tokens)
2036
missing_keys = self.target_repo.get_missing_parent_inventories(
2037
check_for_missing_texts=is_resume)
2038
try:
2039
for prefix, versioned_file in (
2040
('texts', self.target_repo.texts),
2041
('inventories', self.target_repo.inventories),
2042
('revisions', self.target_repo.revisions),
2043
('signatures', self.target_repo.signatures),
2044
('chk_bytes', self.target_repo.chk_bytes),
2045
):
2046
if versioned_file is None:
2047
continue
2048
# TODO: key is often going to be a StaticTuple object
2049
# I don't believe we can define a method by which
2050
# (prefix,) + StaticTuple will work, though we could
2051
# define a StaticTuple.sq_concat that would allow you to
2052
# pass in either a tuple or a StaticTuple as the second
2053
# object, so instead we could have:
2054
# StaticTuple(prefix) + key here...
2055
missing_keys.update((prefix,) + key for key in
2056
versioned_file.get_missing_compression_parent_keys())
2057
except NotImplementedError:
2058
# cannot even attempt suspending, and missing would have failed
2059
# during stream insertion.
2060
missing_keys = set()
2061
return missing_keys
2062
2063
def _extract_and_insert_inventory_deltas(self, substream, serializer):
2064
target_rich_root = self.target_repo._format.rich_root_data
2065
target_tree_refs = self.target_repo._format.supports_tree_reference
2066
for record in substream:
2067
# Insert the delta directly
2068
inventory_delta_bytes = record.get_bytes_as('fulltext')
2069
deserialiser = inventory_delta.InventoryDeltaDeserializer()
2070
try:
2071
parse_result = deserialiser.parse_text_bytes(
2072
inventory_delta_bytes)
2073
except inventory_delta.IncompatibleInventoryDelta, err:
2074
mutter("Incompatible delta: %s", err.msg)
2075
raise errors.IncompatibleRevision(self.target_repo._format)
2076
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
2077
revision_id = new_id
2078
parents = [key[0] for key in record.parents]
2079
self.target_repo.add_inventory_by_delta(
2080
basis_id, inv_delta, revision_id, parents)
2081
2082
def _extract_and_insert_inventories(self, substream, serializer,
2083
parse_delta=None):
2084
"""Generate a new inventory versionedfile in target, converting data.
2085
2086
The inventory is retrieved from the source, (deserializing it), and
2087
stored in the target (reserializing it in a different format).
2088
"""
2089
target_rich_root = self.target_repo._format.rich_root_data
2090
target_tree_refs = self.target_repo._format.supports_tree_reference
2091
for record in substream:
2092
# It's not a delta, so it must be a fulltext in the source
2093
# serializer's format.
2094
bytes = record.get_bytes_as('fulltext')
2095
revision_id = record.key[0]
2096
inv = serializer.read_inventory_from_string(bytes, revision_id)
2097
parents = [key[0] for key in record.parents]
2098
self.target_repo.add_inventory(revision_id, inv, parents)
2099
# No need to keep holding this full inv in memory when the rest of
2100
# the substream is likely to be all deltas.
2101
del inv
2102
2103
def _extract_and_insert_revisions(self, substream, serializer):
2104
for record in substream:
2105
bytes = record.get_bytes_as('fulltext')
2106
revision_id = record.key[0]
2107
rev = serializer.read_revision_from_string(bytes)
2108
if rev.revision_id != revision_id:
2109
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
2110
self.target_repo.add_revision(revision_id, rev)
2111
2112
def finished(self):
2113
if self.target_repo._format._fetch_reconcile:
2114
self.target_repo.reconcile()
2115
2116
2117
class StreamSource(object):
2118
"""A source of a stream for fetching between repositories."""
2119
2120
def __init__(self, from_repository, to_format):
2121
"""Create a StreamSource streaming from from_repository."""
2122
self.from_repository = from_repository
2123
self.to_format = to_format
2124
self._record_counter = RecordCounter()
2125
2126
def delta_on_metadata(self):
2127
"""Return True if delta's are permitted on metadata streams.
2128
2129
That is on revisions and signatures.
2130
"""
2131
src_serializer = self.from_repository._format._serializer
2132
target_serializer = self.to_format._serializer
2133
return (self.to_format._fetch_uses_deltas and
2134
src_serializer == target_serializer)
2135
2136
def _fetch_revision_texts(self, revs):
2137
# fetch signatures first and then the revision texts
2138
# may need to be a InterRevisionStore call here.
2139
from_sf = self.from_repository.signatures
2140
# A missing signature is just skipped.
2141
keys = [(rev_id,) for rev_id in revs]
2142
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
2143
keys,
2144
self.to_format._fetch_order,
2145
not self.to_format._fetch_uses_deltas))
2146
# If a revision has a delta, this is actually expanded inside the
2147
# insert_record_stream code now, which is an alternate fix for
2148
# bug #261339
2149
from_rf = self.from_repository.revisions
2150
revisions = from_rf.get_record_stream(
2151
keys,
2152
self.to_format._fetch_order,
2153
not self.delta_on_metadata())
2154
return [('signatures', signatures), ('revisions', revisions)]
2155
2156
def _generate_root_texts(self, revs):
2157
"""This will be called by get_stream between fetching weave texts and
2158
fetching the inventory weave.
2159
"""
2160
if self._rich_root_upgrade():
2161
return _mod_fetch.Inter1and2Helper(
2162
self.from_repository).generate_root_texts(revs)
2163
else:
2164
return []
2165
2166
def get_stream(self, search):
2167
phase = 'file'
2168
revs = search.get_keys()
2169
graph = self.from_repository.get_graph()
2170
revs = tsort.topo_sort(graph.get_parent_map(revs))
2171
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
2172
text_keys = []
2173
for knit_kind, file_id, revisions in data_to_fetch:
2174
if knit_kind != phase:
2175
phase = knit_kind
2176
# Make a new progress bar for this phase
2177
if knit_kind == "file":
2178
# Accumulate file texts
2179
text_keys.extend([(file_id, revision) for revision in
2180
revisions])
2181
elif knit_kind == "inventory":
2182
# Now copy the file texts.
2183
from_texts = self.from_repository.texts
2184
yield ('texts', from_texts.get_record_stream(
2185
text_keys, self.to_format._fetch_order,
2186
not self.to_format._fetch_uses_deltas))
2187
# Cause an error if a text occurs after we have done the
2188
# copy.
2189
text_keys = None
2190
# Before we process the inventory we generate the root
2191
# texts (if necessary) so that the inventories references
2192
# will be valid.
2193
for _ in self._generate_root_texts(revs):
2194
yield _
2195
# we fetch only the referenced inventories because we do not
2196
# know for unselected inventories whether all their required
2197
# texts are present in the other repository - it could be
2198
# corrupt.
2199
for info in self._get_inventory_stream(revs):
2200
yield info
2201
elif knit_kind == "signatures":
2202
# Nothing to do here; this will be taken care of when
2203
# _fetch_revision_texts happens.
2204
pass
2205
elif knit_kind == "revisions":
2206
for record in self._fetch_revision_texts(revs):
2207
yield record
2208
else:
2209
raise AssertionError("Unknown knit kind %r" % knit_kind)
2210
2211
def get_stream_for_missing_keys(self, missing_keys):
2212
# missing keys can only occur when we are byte copying and not
2213
# translating (because translation means we don't send
2214
# unreconstructable deltas ever).
2215
keys = {}
2216
keys['texts'] = set()
2217
keys['revisions'] = set()
2218
keys['inventories'] = set()
2219
keys['chk_bytes'] = set()
2220
keys['signatures'] = set()
2221
for key in missing_keys:
2222
keys[key[0]].add(key[1:])
2223
if len(keys['revisions']):
2224
# If we allowed copying revisions at this point, we could end up
2225
# copying a revision without copying its required texts: a
2226
# violation of the requirements for repository integrity.
2227
raise AssertionError(
2228
'cannot copy revisions to fill in missing deltas %s' % (
2229
keys['revisions'],))
2230
for substream_kind, keys in keys.iteritems():
2231
vf = getattr(self.from_repository, substream_kind)
2232
if vf is None and keys:
2233
raise AssertionError(
2234
"cannot fill in keys for a versioned file we don't"
2235
" have: %s needs %s" % (substream_kind, keys))
2236
if not keys:
2237
# No need to stream something we don't have
2238
continue
2239
if substream_kind == 'inventories':
2240
# Some missing keys are genuinely ghosts, filter those out.
2241
present = self.from_repository.inventories.get_parent_map(keys)
2242
revs = [key[0] for key in present]
2243
# Get the inventory stream more-or-less as we do for the
2244
# original stream; there's no reason to assume that records
2245
# direct from the source will be suitable for the sink. (Think
2246
# e.g. 2a -> 1.9-rich-root).
2247
for info in self._get_inventory_stream(revs, missing=True):
2248
yield info
2249
continue
2250
2251
# Ask for full texts always so that we don't need more round trips
2252
# after this stream.
2253
# Some of the missing keys are genuinely ghosts, so filter absent
2254
# records. The Sink is responsible for doing another check to
2255
# ensure that ghosts don't introduce missing data for future
2256
# fetches.
2257
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
2258
self.to_format._fetch_order, True))
2259
yield substream_kind, stream
2260
2261
def inventory_fetch_order(self):
2262
if self._rich_root_upgrade():
2263
return 'topological'
2264
else:
2265
return self.to_format._fetch_order
2266
2267
def _rich_root_upgrade(self):
2268
return (not self.from_repository._format.rich_root_data and
2269
self.to_format.rich_root_data)
2270
2271
def _get_inventory_stream(self, revision_ids, missing=False):
2272
from_format = self.from_repository._format
2273
if (from_format.supports_chks and self.to_format.supports_chks and
2274
from_format.network_name() == self.to_format.network_name()):
2275
raise AssertionError(
2276
"this case should be handled by GroupCHKStreamSource")
2277
elif 'forceinvdeltas' in debug.debug_flags:
2278
return self._get_convertable_inventory_stream(revision_ids,
2279
delta_versus_null=missing)
2280
elif from_format.network_name() == self.to_format.network_name():
2281
# Same format.
2282
return self._get_simple_inventory_stream(revision_ids,
2283
missing=missing)
2284
elif (not from_format.supports_chks and not self.to_format.supports_chks
2285
and from_format._serializer == self.to_format._serializer):
2286
# Essentially the same format.
2287
return self._get_simple_inventory_stream(revision_ids,
2288
missing=missing)
2289
else:
2290
# Any time we switch serializations, we want to use an
2291
# inventory-delta based approach.
2292
return self._get_convertable_inventory_stream(revision_ids,
2293
delta_versus_null=missing)
2294
2295
def _get_simple_inventory_stream(self, revision_ids, missing=False):
2296
# NB: This currently reopens the inventory weave in source;
2297
# using a single stream interface instead would avoid this.
2298
from_weave = self.from_repository.inventories
2299
if missing:
2300
delta_closure = True
2301
else:
2302
delta_closure = not self.delta_on_metadata()
2303
yield ('inventories', from_weave.get_record_stream(
2304
[(rev_id,) for rev_id in revision_ids],
2305
self.inventory_fetch_order(), delta_closure))
2306
2307
def _get_convertable_inventory_stream(self, revision_ids,
2308
delta_versus_null=False):
2309
# The two formats are sufficiently different that there is no fast
2310
# path, so we need to send just inventorydeltas, which any
2311
# sufficiently modern client can insert into any repository.
2312
# The StreamSink code expects to be able to
2313
# convert on the target, so we need to put bytes-on-the-wire that can
2314
# be converted. That means inventory deltas (if the remote is <1.19,
2315
# RemoteStreamSink will fallback to VFS to insert the deltas).
2316
yield ('inventory-deltas',
2317
self._stream_invs_as_deltas(revision_ids,
2318
delta_versus_null=delta_versus_null))
2319
2320
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
2321
"""Return a stream of inventory-deltas for the given rev ids.
2322
2323
:param revision_ids: The list of inventories to transmit
2324
:param delta_versus_null: Don't try to find a minimal delta for this
2325
entry, instead compute the delta versus the NULL_REVISION. This
2326
effectively streams a complete inventory. Used for stuff like
2327
filling in missing parents, etc.
2328
"""
2329
from_repo = self.from_repository
2330
revision_keys = [(rev_id,) for rev_id in revision_ids]
2331
parent_map = from_repo.inventories.get_parent_map(revision_keys)
2332
# XXX: possibly repos could implement a more efficient iter_inv_deltas
2333
# method...
2334
inventories = self.from_repository.iter_inventories(
2335
revision_ids, 'topological')
2336
format = from_repo._format
2337
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
2338
inventory_cache = lru_cache.LRUCache(50)
2339
null_inventory = from_repo.revision_tree(
2340
_mod_revision.NULL_REVISION).inventory
2341
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
2342
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
2343
# repo back into a non-rich-root repo ought to be allowed)
2344
serializer = inventory_delta.InventoryDeltaSerializer(
2345
versioned_root=format.rich_root_data,
2346
tree_references=format.supports_tree_reference)
2347
for inv in inventories:
2348
key = (inv.revision_id,)
2349
parent_keys = parent_map.get(key, ())
2350
delta = None
2351
if not delta_versus_null and parent_keys:
2352
# The caller did not ask for complete inventories and we have
2353
# some parents that we can delta against. Make a delta against
2354
# each parent so that we can find the smallest.
2355
parent_ids = [parent_key[0] for parent_key in parent_keys]
2356
for parent_id in parent_ids:
2357
if parent_id not in invs_sent_so_far:
2358
# We don't know that the remote side has this basis, so
2359
# we can't use it.
2360
continue
2361
if parent_id == _mod_revision.NULL_REVISION:
2362
parent_inv = null_inventory
2363
else:
2364
parent_inv = inventory_cache.get(parent_id, None)
2365
if parent_inv is None:
2366
parent_inv = from_repo.get_inventory(parent_id)
2367
candidate_delta = inv._make_delta(parent_inv)
2368
if (delta is None or
2369
len(delta) > len(candidate_delta)):
2370
delta = candidate_delta
2371
basis_id = parent_id
2372
if delta is None:
2373
# Either none of the parents ended up being suitable, or we
2374
# were asked to delta against NULL
2375
basis_id = _mod_revision.NULL_REVISION
2376
delta = inv._make_delta(null_inventory)
2377
invs_sent_so_far.add(inv.revision_id)
2378
inventory_cache[inv.revision_id] = inv
2379
delta_serialized = ''.join(
2380
serializer.delta_to_lines(basis_id, key[-1], delta))
2381
yield versionedfile.FulltextContentFactory(
2382
key, parent_keys, None, delta_serialized)
2383
2384
2385
class _VersionedFileChecker(object):
2386
2387
def __init__(self, repository, text_key_references=None, ancestors=None):
2388
self.repository = repository
2389
self.text_index = self.repository._generate_text_key_index(
2390
text_key_references=text_key_references, ancestors=ancestors)
2391
2392
def calculate_file_version_parents(self, text_key):
2393
"""Calculate the correct parents for a file version according to
2394
the inventories.
2395
"""
2396
parent_keys = self.text_index[text_key]
2397
if parent_keys == [_mod_revision.NULL_REVISION]:
2398
return ()
2399
return tuple(parent_keys)
2400
2401
def check_file_version_parents(self, texts, progress_bar=None):
2402
"""Check the parents stored in a versioned file are correct.
2403
2404
It also detects file versions that are not referenced by their
2405
corresponding revision's inventory.
2406
2407
:returns: A tuple of (wrong_parents, dangling_file_versions).
2408
wrong_parents is a dict mapping {revision_id: (stored_parents,
2409
correct_parents)} for each revision_id where the stored parents
2410
are not correct. dangling_file_versions is a set of (file_id,
2411
revision_id) tuples for versions that are present in this versioned
2412
file, but not used by the corresponding inventory.
2413
"""
2414
local_progress = None
2415
if progress_bar is None:
2416
local_progress = ui.ui_factory.nested_progress_bar()
2417
progress_bar = local_progress
2418
try:
2419
return self._check_file_version_parents(texts, progress_bar)
2420
finally:
2421
if local_progress:
2422
local_progress.finished()
2423
2424
def _check_file_version_parents(self, texts, progress_bar):
2425
"""See check_file_version_parents."""
2426
wrong_parents = {}
2427
self.file_ids = set([file_id for file_id, _ in
2428
self.text_index.iterkeys()])
2429
# text keys is now grouped by file_id
2430
n_versions = len(self.text_index)
2431
progress_bar.update('loading text store', 0, n_versions)
2432
parent_map = self.repository.texts.get_parent_map(self.text_index)
2433
# On unlistable transports this could well be empty/error...
2434
text_keys = self.repository.texts.keys()
2435
unused_keys = frozenset(text_keys) - set(self.text_index)
2436
for num, key in enumerate(self.text_index.iterkeys()):
2437
progress_bar.update('checking text graph', num, n_versions)
2438
correct_parents = self.calculate_file_version_parents(key)
2439
try:
2440
knit_parents = parent_map[key]
2441
except errors.RevisionNotPresent:
2442
# Missing text!
2443
knit_parents = None
2444
if correct_parents != knit_parents:
2445
wrong_parents[key] = (knit_parents, correct_parents)
2446
return wrong_parents, unused_keys
2447
2448
2449
class InterDifferingSerializer(InterRepository):
2450
2451
@classmethod
2452
def _get_repo_format_to_test(self):
2453
return None
2454
2455
@staticmethod
2456
def is_compatible(source, target):
2457
if not source._format.supports_full_versioned_files:
2458
return False
2459
if not target._format.supports_full_versioned_files:
2460
return False
2461
# This is redundant with format.check_conversion_target(), however that
2462
# raises an exception, and we just want to say "False" as in we won't
2463
# support converting between these formats.
2464
if 'IDS_never' in debug.debug_flags:
2465
return False
2466
if source.supports_rich_root() and not target.supports_rich_root():
2467
return False
2468
if (source._format.supports_tree_reference
2469
and not target._format.supports_tree_reference):
2470
return False
2471
if target._fallback_repositories and target._format.supports_chks:
2472
# IDS doesn't know how to copy CHKs for the parent inventories it
2473
# adds to stacked repos.
2474
return False
2475
if 'IDS_always' in debug.debug_flags:
2476
return True
2477
# Only use this code path for local source and target. IDS does far
2478
# too much IO (both bandwidth and roundtrips) over a network.
2479
if not source.bzrdir.transport.base.startswith('file:///'):
2480
return False
2481
if not target.bzrdir.transport.base.startswith('file:///'):
2482
return False
2483
return True
2484
2485
def _get_trees(self, revision_ids, cache):
2486
possible_trees = []
2487
for rev_id in revision_ids:
2488
if rev_id in cache:
2489
possible_trees.append((rev_id, cache[rev_id]))
2490
else:
2491
# Not cached, but inventory might be present anyway.
2492
try:
2493
tree = self.source.revision_tree(rev_id)
2494
except errors.NoSuchRevision:
2495
# Nope, parent is ghost.
2496
pass
2497
else:
2498
cache[rev_id] = tree
2499
possible_trees.append((rev_id, tree))
2500
return possible_trees
2501
2502
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
2503
"""Get the best delta and base for this revision.
2504
2505
:return: (basis_id, delta)
2506
"""
2507
deltas = []
2508
# Generate deltas against each tree, to find the shortest.
2509
texts_possibly_new_in_tree = set()
2510
for basis_id, basis_tree in possible_trees:
2511
delta = tree.inventory._make_delta(basis_tree.inventory)
2512
for old_path, new_path, file_id, new_entry in delta:
2513
if new_path is None:
2514
# This file_id isn't present in the new rev, so we don't
2515
# care about it.
2516
continue
2517
if not new_path:
2518
# Rich roots are handled elsewhere...
2519
continue
2520
kind = new_entry.kind
2521
if kind != 'directory' and kind != 'file':
2522
# No text record associated with this inventory entry.
2523
continue
2524
# This is a directory or file that has changed somehow.
2525
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
2526
deltas.append((len(delta), basis_id, delta))
2527
deltas.sort()
2528
return deltas[0][1:]
2529
2530
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
2531
"""Find all parent revisions that are absent, but for which the
2532
inventory is present, and copy those inventories.
2533
2534
This is necessary to preserve correctness when the source is stacked
2535
without fallbacks configured. (Note that in cases like upgrade the
2536
source may be not have _fallback_repositories even though it is
2537
stacked.)
2538
"""
2539
parent_revs = set()
2540
for parents in parent_map.values():
2541
parent_revs.update(parents)
2542
present_parents = self.source.get_parent_map(parent_revs)
2543
absent_parents = set(parent_revs).difference(present_parents)
2544
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
2545
(rev_id,) for rev_id in absent_parents)
2546
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
2547
for parent_tree in self.source.revision_trees(parent_inv_ids):
2548
current_revision_id = parent_tree.get_revision_id()
2549
parents_parents_keys = parent_invs_keys_for_stacking[
2550
(current_revision_id,)]
2551
parents_parents = [key[-1] for key in parents_parents_keys]
2552
basis_id = _mod_revision.NULL_REVISION
2553
basis_tree = self.source.revision_tree(basis_id)
2554
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
2555
self.target.add_inventory_by_delta(
2556
basis_id, delta, current_revision_id, parents_parents)
2557
cache[current_revision_id] = parent_tree
2558
2559
def _fetch_batch(self, revision_ids, basis_id, cache):
2560
"""Fetch across a few revisions.
2561
2562
:param revision_ids: The revisions to copy
2563
:param basis_id: The revision_id of a tree that must be in cache, used
2564
as a basis for delta when no other base is available
2565
:param cache: A cache of RevisionTrees that we can use.
2566
:return: The revision_id of the last converted tree. The RevisionTree
2567
for it will be in cache
2568
"""
2569
# Walk though all revisions; get inventory deltas, copy referenced
2570
# texts that delta references, insert the delta, revision and
2571
# signature.
2572
root_keys_to_create = set()
2573
text_keys = set()
2574
pending_deltas = []
2575
pending_revisions = []
2576
parent_map = self.source.get_parent_map(revision_ids)
2577
self._fetch_parent_invs_for_stacking(parent_map, cache)
2578
self.source._safe_to_return_from_cache = True
2579
for tree in self.source.revision_trees(revision_ids):
2580
# Find a inventory delta for this revision.
2581
# Find text entries that need to be copied, too.
2582
current_revision_id = tree.get_revision_id()
2583
parent_ids = parent_map.get(current_revision_id, ())
2584
parent_trees = self._get_trees(parent_ids, cache)
2585
possible_trees = list(parent_trees)
2586
if len(possible_trees) == 0:
2587
# There either aren't any parents, or the parents are ghosts,
2588
# so just use the last converted tree.
2589
possible_trees.append((basis_id, cache[basis_id]))
2590
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
2591
possible_trees)
2592
revision = self.source.get_revision(current_revision_id)
2593
pending_deltas.append((basis_id, delta,
2594
current_revision_id, revision.parent_ids))
2595
if self._converting_to_rich_root:
2596
self._revision_id_to_root_id[current_revision_id] = \
2597
tree.get_root_id()
2598
# Determine which texts are in present in this revision but not in
2599
# any of the available parents.
2600
texts_possibly_new_in_tree = set()
2601
for old_path, new_path, file_id, entry in delta:
2602
if new_path is None:
2603
# This file_id isn't present in the new rev
2604
continue
2605
if not new_path:
2606
# This is the root
2607
if not self.target.supports_rich_root():
2608
# The target doesn't support rich root, so we don't
2609
# copy
2610
continue
2611
if self._converting_to_rich_root:
2612
# This can't be copied normally, we have to insert
2613
# it specially
2614
root_keys_to_create.add((file_id, entry.revision))
2615
continue
2616
kind = entry.kind
2617
texts_possibly_new_in_tree.add((file_id, entry.revision))
2618
for basis_id, basis_tree in possible_trees:
2619
basis_inv = basis_tree.inventory
2620
for file_key in list(texts_possibly_new_in_tree):
2621
file_id, file_revision = file_key
2622
try:
2623
entry = basis_inv[file_id]
2624
except errors.NoSuchId:
2625
continue
2626
if entry.revision == file_revision:
2627
texts_possibly_new_in_tree.remove(file_key)
2628
text_keys.update(texts_possibly_new_in_tree)
2629
pending_revisions.append(revision)
2630
cache[current_revision_id] = tree
2631
basis_id = current_revision_id
2632
self.source._safe_to_return_from_cache = False
2633
# Copy file texts
2634
from_texts = self.source.texts
2635
to_texts = self.target.texts
2636
if root_keys_to_create:
2637
root_stream = _mod_fetch._new_root_data_stream(
2638
root_keys_to_create, self._revision_id_to_root_id, parent_map,
2639
self.source)
2640
to_texts.insert_record_stream(root_stream)
2641
to_texts.insert_record_stream(from_texts.get_record_stream(
2642
text_keys, self.target._format._fetch_order,
2643
not self.target._format._fetch_uses_deltas))
2644
# insert inventory deltas
2645
for delta in pending_deltas:
2646
self.target.add_inventory_by_delta(*delta)
2647
if self.target._fallback_repositories:
2648
# Make sure this stacked repository has all the parent inventories
2649
# for the new revisions that we are about to insert. We do this
2650
# before adding the revisions so that no revision is added until
2651
# all the inventories it may depend on are added.
2652
# Note that this is overzealous, as we may have fetched these in an
2653
# earlier batch.
2654
parent_ids = set()
2655
revision_ids = set()
2656
for revision in pending_revisions:
2657
revision_ids.add(revision.revision_id)
2658
parent_ids.update(revision.parent_ids)
2659
parent_ids.difference_update(revision_ids)
2660
parent_ids.discard(_mod_revision.NULL_REVISION)
2661
parent_map = self.source.get_parent_map(parent_ids)
2662
# we iterate over parent_map and not parent_ids because we don't
2663
# want to try copying any revision which is a ghost
2664
for parent_tree in self.source.revision_trees(parent_map):
2665
current_revision_id = parent_tree.get_revision_id()
2666
parents_parents = parent_map[current_revision_id]
2667
possible_trees = self._get_trees(parents_parents, cache)
2668
if len(possible_trees) == 0:
2669
# There either aren't any parents, or the parents are
2670
# ghosts, so just use the last converted tree.
2671
possible_trees.append((basis_id, cache[basis_id]))
2672
basis_id, delta = self._get_delta_for_revision(parent_tree,
2673
parents_parents, possible_trees)
2674
self.target.add_inventory_by_delta(
2675
basis_id, delta, current_revision_id, parents_parents)
2676
# insert signatures and revisions
2677
for revision in pending_revisions:
2678
try:
2679
signature = self.source.get_signature_text(
2680
revision.revision_id)
2681
self.target.add_signature_text(revision.revision_id,
2682
signature)
2683
except errors.NoSuchRevision:
2684
pass
2685
self.target.add_revision(revision.revision_id, revision)
2686
return basis_id
2687
2688
def _fetch_all_revisions(self, revision_ids, pb):
2689
"""Fetch everything for the list of revisions.
2690
2691
:param revision_ids: The list of revisions to fetch. Must be in
2692
topological order.
2693
:param pb: A ProgressTask
2694
:return: None
2695
"""
2696
basis_id, basis_tree = self._get_basis(revision_ids[0])
2697
batch_size = 100
2698
cache = lru_cache.LRUCache(100)
2699
cache[basis_id] = basis_tree
2700
del basis_tree # We don't want to hang on to it here
2701
hints = []
2702
a_graph = None
2703
2704
for offset in range(0, len(revision_ids), batch_size):
2705
self.target.start_write_group()
2706
try:
2707
pb.update('Transferring revisions', offset,
2708
len(revision_ids))
2709
batch = revision_ids[offset:offset+batch_size]
2710
basis_id = self._fetch_batch(batch, basis_id, cache)
2711
except:
2712
self.source._safe_to_return_from_cache = False
2713
self.target.abort_write_group()
2714
raise
2715
else:
2716
hint = self.target.commit_write_group()
2717
if hint:
2718
hints.extend(hint)
2719
if hints and self.target._format.pack_compresses:
2720
self.target.pack(hint=hints)
2721
pb.update('Transferring revisions', len(revision_ids),
2722
len(revision_ids))
2723
2724
@needs_write_lock
2725
def fetch(self, revision_id=None, find_ghosts=False,
2726
fetch_spec=None):
2727
"""See InterRepository.fetch()."""
2728
if fetch_spec is not None:
2729
revision_ids = fetch_spec.get_keys()
2730
else:
2731
revision_ids = None
2732
ui.ui_factory.warn_experimental_format_fetch(self)
2733
if (not self.source.supports_rich_root()
2734
and self.target.supports_rich_root()):
2735
self._converting_to_rich_root = True
2736
self._revision_id_to_root_id = {}
2737
else:
2738
self._converting_to_rich_root = False
2739
# See <https://launchpad.net/bugs/456077> asking for a warning here
2740
if self.source._format.network_name() != self.target._format.network_name():
2741
ui.ui_factory.show_user_warning('cross_format_fetch',
2742
from_format=self.source._format,
2743
to_format=self.target._format)
2744
if revision_ids is None:
2745
if revision_id:
2746
search_revision_ids = [revision_id]
2747
else:
2748
search_revision_ids = None
2749
revision_ids = self.target.search_missing_revision_ids(self.source,
2750
revision_ids=search_revision_ids,
2751
find_ghosts=find_ghosts).get_keys()
2752
if not revision_ids:
2753
return 0, 0
2754
revision_ids = tsort.topo_sort(
2755
self.source.get_graph().get_parent_map(revision_ids))
2756
if not revision_ids:
2757
return 0, 0
2758
# Walk though all revisions; get inventory deltas, copy referenced
2759
# texts that delta references, insert the delta, revision and
2760
# signature.
2761
pb = ui.ui_factory.nested_progress_bar()
2762
try:
2763
self._fetch_all_revisions(revision_ids, pb)
2764
finally:
2765
pb.finished()
2766
return len(revision_ids), 0
2767
2768
def _get_basis(self, first_revision_id):
2769
"""Get a revision and tree which exists in the target.
2770
2771
This assumes that first_revision_id is selected for transmission
2772
because all other ancestors are already present. If we can't find an
2773
ancestor we fall back to NULL_REVISION since we know that is safe.
2774
2775
:return: (basis_id, basis_tree)
2776
"""
2777
first_rev = self.source.get_revision(first_revision_id)
2778
try:
2779
basis_id = first_rev.parent_ids[0]
2780
# only valid as a basis if the target has it
2781
self.target.get_revision(basis_id)
2782
# Try to get a basis tree - if it's a ghost it will hit the
2783
# NoSuchRevision case.
2784
basis_tree = self.source.revision_tree(basis_id)
2785
except (IndexError, errors.NoSuchRevision):
2786
basis_id = _mod_revision.NULL_REVISION
2787
basis_tree = self.source.revision_tree(basis_id)
2788
return basis_id, basis_tree
2789
2790
2791
class InterSameDataRepository(InterRepository):
2792
"""Code for converting between repositories that represent the same data.
2793
2794
Data format and model must match for this to work.
2795
"""
2796
2797
@classmethod
2798
def _get_repo_format_to_test(self):
2799
"""Repository format for testing with.
2800
2801
InterSameData can pull from subtree to subtree and from non-subtree to
2802
non-subtree, so we test this with the richest repository format.
2803
"""
2804
from bzrlib.repofmt import knitrepo
2805
return knitrepo.RepositoryFormatKnit3()
2806
2807
@staticmethod
2808
def is_compatible(source, target):
2809
return (
2810
InterRepository._same_model(source, target) and
2811
source._format.supports_full_versioned_files and
2812
target._format.supports_full_versioned_files)
2813
2814
2815
InterRepository.register_optimiser(InterDifferingSerializer)
2816
InterRepository.register_optimiser(InterSameDataRepository)
2817
2818
2819
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2820
"""Install all revision data into a repository.
2821
2822
Accepts an iterable of revision, tree, signature tuples. The signature
2823
may be None.
2824
"""
2825
repository.start_write_group()
2826
try:
2827
inventory_cache = lru_cache.LRUCache(10)
2828
for n, (revision, revision_tree, signature) in enumerate(iterable):
2829
_install_revision(repository, revision, revision_tree, signature,
2830
inventory_cache)
2831
if pb is not None:
2832
pb.update('Transferring revisions', n + 1, num_revisions)
2833
except:
2834
repository.abort_write_group()
2835
raise
2836
else:
2837
repository.commit_write_group()
2838
2839
2840
def _install_revision(repository, rev, revision_tree, signature,
2841
inventory_cache):
2842
"""Install all revision data into a repository."""
2843
present_parents = []
2844
parent_trees = {}
2845
for p_id in rev.parent_ids:
2846
if repository.has_revision(p_id):
2847
present_parents.append(p_id)
2848
parent_trees[p_id] = repository.revision_tree(p_id)
2849
else:
2850
parent_trees[p_id] = repository.revision_tree(
2851
_mod_revision.NULL_REVISION)
2852
2853
inv = revision_tree.inventory
2854
entries = inv.iter_entries()
2855
# backwards compatibility hack: skip the root id.
2856
if not repository.supports_rich_root():
2857
path, root = entries.next()
2858
if root.revision != rev.revision_id:
2859
raise errors.IncompatibleRevision(repr(repository))
2860
text_keys = {}
2861
for path, ie in entries:
2862
text_keys[(ie.file_id, ie.revision)] = ie
2863
text_parent_map = repository.texts.get_parent_map(text_keys)
2864
missing_texts = set(text_keys) - set(text_parent_map)
2865
# Add the texts that are not already present
2866
for text_key in missing_texts:
2867
ie = text_keys[text_key]
2868
text_parents = []
2869
# FIXME: TODO: The following loop overlaps/duplicates that done by
2870
# commit to determine parents. There is a latent/real bug here where
2871
# the parents inserted are not those commit would do - in particular
2872
# they are not filtered by heads(). RBC, AB
2873
for revision, tree in parent_trees.iteritems():
2874
if ie.file_id not in tree:
2875
continue
2876
parent_id = tree.get_file_revision(ie.file_id)
2877
if parent_id in text_parents:
2878
continue
2879
text_parents.append((ie.file_id, parent_id))
2880
lines = revision_tree.get_file(ie.file_id).readlines()
2881
repository.texts.add_lines(text_key, text_parents, lines)
2882
try:
2883
# install the inventory
2884
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2885
# Cache this inventory
2886
inventory_cache[rev.revision_id] = inv
2887
try:
2888
basis_inv = inventory_cache[rev.parent_ids[0]]
2889
except KeyError:
2890
repository.add_inventory(rev.revision_id, inv, present_parents)
2891
else:
2892
delta = inv._make_delta(basis_inv)
2893
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2894
rev.revision_id, present_parents)
2895
else:
2896
repository.add_inventory(rev.revision_id, inv, present_parents)
2897
except errors.RevisionAlreadyPresent:
2898
pass
2899
if signature is not None:
2900
repository.add_signature_text(rev.revision_id, signature)
2901
repository.add_revision(rev.revision_id, rev, inv)
2902
2903
2904
def install_revision(repository, rev, revision_tree):
2905
"""Install all revision data into a repository."""
2906
install_revisions(repository, [(rev, revision_tree, None)])
Loggerhead is a web-based interface for Breezy
Version: 2.0.1