1
# Copyright (C) 2005-2011 Canonical Ltd
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.
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.
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
17
"""Repository formats built around versioned files."""
20
from bzrlib.lazy_import import lazy_import
21
lazy_import(globals(), """
34
revision as _mod_revision,
35
serializer as _mod_serializer,
43
from bzrlib.recordcounter import RecordCounter
44
from bzrlib.revisiontree import InventoryRevisionTree
45
from bzrlib.testament import Testament
51
from bzrlib.decorators import (
56
from bzrlib.inventory import (
63
from bzrlib.repository import (
67
MetaDirRepositoryFormat,
72
from bzrlib.trace import (
77
class VersionedFileRepositoryFormat(RepositoryFormat):
78
"""Base class for all repository formats that are VersionedFiles-based."""
80
supports_full_versioned_files = True
82
# Should commit add an inventory, or an inventory delta to the repository.
83
_commit_inv_deltas = True
84
# What order should fetch operations request streams in?
85
# The default is unordered as that is the cheapest for an origin to
87
_fetch_order = 'unordered'
88
# Does this repository format use deltas that can be fetched as-deltas ?
89
# (E.g. knits, where the knit deltas can be transplanted intact.
90
# We default to False, which will ensure that enough data to get
91
# a full text out of any fetch stream will be grabbed.
92
_fetch_uses_deltas = False
95
class VersionedFileCommitBuilder(CommitBuilder):
96
"""Commit builder implementation for versioned files based repositories.
99
# this commit builder supports the record_entry_contents interface
100
supports_record_entry_contents = True
102
# the default CommitBuilder does not manage trees whose root is versioned.
103
_versioned_root = False
105
def __init__(self, repository, parents, config, timestamp=None,
106
timezone=None, committer=None, revprops=None,
107
revision_id=None, lossy=False):
108
super(VersionedFileCommitBuilder, self).__init__(repository,
109
parents, config, timestamp, timezone, committer, revprops,
112
basis_id = self.parents[0]
114
basis_id = _mod_revision.NULL_REVISION
115
self.basis_delta_revision = basis_id
116
self.new_inventory = Inventory(None)
117
self._basis_delta = []
118
self.__heads = graph.HeadsCache(repository.get_graph()).heads
119
# memo'd check for no-op commits.
120
self._any_changes = False
121
# API compatibility, older code that used CommitBuilder did not call
122
# .record_delete(), which means the delta that is computed would not be
123
# valid. Callers that will call record_delete() should call
124
# .will_record_deletes() to indicate that.
125
self._recording_deletes = False
127
def will_record_deletes(self):
128
"""Tell the commit builder that deletes are being notified.
130
This enables the accumulation of an inventory delta; for the resulting
131
commit to be valid, deletes against the basis MUST be recorded via
132
builder.record_delete().
134
self._recording_deletes = True
136
def any_changes(self):
137
"""Return True if any entries were changed.
139
This includes merge-only changes. It is the core for the --unchanged
142
:return: True if any changes have occured.
144
return self._any_changes
146
def _ensure_fallback_inventories(self):
147
"""Ensure that appropriate inventories are available.
149
This only applies to repositories that are stacked, and is about
150
enusring the stacking invariants. Namely, that for any revision that is
151
present, we either have all of the file content, or we have the parent
152
inventory and the delta file content.
154
if not self.repository._fallback_repositories:
156
if not self.repository._format.supports_chks:
157
raise errors.BzrError("Cannot commit directly to a stacked branch"
158
" in pre-2a formats. See "
159
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
160
# This is a stacked repo, we need to make sure we have the parent
161
# inventories for the parents.
162
parent_keys = [(p,) for p in self.parents]
163
parent_map = self.repository.inventories._index.get_parent_map(parent_keys)
164
missing_parent_keys = set([pk for pk in parent_keys
165
if pk not in parent_map])
166
fallback_repos = list(reversed(self.repository._fallback_repositories))
167
missing_keys = [('inventories', pk[0])
168
for pk in missing_parent_keys]
170
while missing_keys and fallback_repos:
171
fallback_repo = fallback_repos.pop()
172
source = fallback_repo._get_source(self.repository._format)
173
sink = self.repository._get_sink()
174
stream = source.get_stream_for_missing_keys(missing_keys)
175
missing_keys = sink.insert_stream_without_locking(stream,
176
self.repository._format)
178
raise errors.BzrError('Unable to fill in parent inventories for a'
181
def commit(self, message):
182
"""Make the actual commit.
184
:return: The revision id of the recorded revision.
186
self._validate_unicode_text(message, 'commit message')
187
rev = _mod_revision.Revision(
188
timestamp=self._timestamp,
189
timezone=self._timezone,
190
committer=self._committer,
192
inventory_sha1=self.inv_sha1,
193
revision_id=self._new_revision_id,
194
properties=self._revprops)
195
rev.parent_ids = self.parents
196
self.repository.add_revision(self._new_revision_id, rev,
197
self.new_inventory, self._config)
198
self._ensure_fallback_inventories()
199
self.repository.commit_write_group()
200
return self._new_revision_id
203
"""Abort the commit that is being built.
205
self.repository.abort_write_group()
207
def revision_tree(self):
208
"""Return the tree that was just committed.
210
After calling commit() this can be called to get a
211
RevisionTree representing the newly committed tree. This is
212
preferred to calling Repository.revision_tree() because that may
213
require deserializing the inventory, while we already have a copy in
216
if self.new_inventory is None:
217
self.new_inventory = self.repository.get_inventory(
218
self._new_revision_id)
219
return InventoryRevisionTree(self.repository, self.new_inventory,
220
self._new_revision_id)
222
def finish_inventory(self):
223
"""Tell the builder that the inventory is finished.
225
:return: The inventory id in the repository, which can be used with
226
repository.get_inventory.
228
if self.new_inventory is None:
229
# an inventory delta was accumulated without creating a new
231
basis_id = self.basis_delta_revision
232
# We ignore the 'inventory' returned by add_inventory_by_delta
233
# because self.new_inventory is used to hint to the rest of the
234
# system what code path was taken
235
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
236
basis_id, self._basis_delta, self._new_revision_id,
239
if self.new_inventory.root is None:
240
raise AssertionError('Root entry should be supplied to'
241
' record_entry_contents, as of bzr 0.10.')
242
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
243
self.new_inventory.revision_id = self._new_revision_id
244
self.inv_sha1 = self.repository.add_inventory(
245
self._new_revision_id,
249
return self._new_revision_id
251
def _check_root(self, ie, parent_invs, tree):
252
"""Helper for record_entry_contents.
254
:param ie: An entry being added.
255
:param parent_invs: The inventories of the parent revisions of the
257
:param tree: The tree that is being committed.
259
# In this revision format, root entries have no knit or weave When
260
# serializing out to disk and back in root.revision is always
262
ie.revision = self._new_revision_id
264
def _require_root_change(self, tree):
265
"""Enforce an appropriate root object change.
267
This is called once when record_iter_changes is called, if and only if
268
the root was not in the delta calculated by record_iter_changes.
270
:param tree: The tree which is being committed.
272
if len(self.parents) == 0:
273
raise errors.RootMissing()
274
entry = entry_factory['directory'](tree.path2id(''), '',
276
entry.revision = self._new_revision_id
277
self._basis_delta.append(('', '', entry.file_id, entry))
279
def _get_delta(self, ie, basis_inv, path):
280
"""Get a delta against the basis inventory for ie."""
281
if ie.file_id not in basis_inv:
283
result = (None, path, ie.file_id, ie)
284
self._basis_delta.append(result)
286
elif ie != basis_inv[ie.file_id]:
288
# TODO: avoid tis id2path call.
289
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
290
self._basis_delta.append(result)
296
def _heads(self, file_id, revision_ids):
297
"""Calculate the graph heads for revision_ids in the graph of file_id.
299
This can use either a per-file graph or a global revision graph as we
300
have an identity relationship between the two graphs.
302
return self.__heads(revision_ids)
304
def get_basis_delta(self):
305
"""Return the complete inventory delta versus the basis inventory.
307
This has been built up with the calls to record_delete and
308
record_entry_contents. The client must have already called
309
will_record_deletes() to indicate that they will be generating a
312
:return: An inventory delta, suitable for use with apply_delta, or
313
Repository.add_inventory_by_delta, etc.
315
if not self._recording_deletes:
316
raise AssertionError("recording deletes not activated.")
317
return self._basis_delta
319
def record_delete(self, path, file_id):
320
"""Record that a delete occured against a basis tree.
322
This is an optional API - when used it adds items to the basis_delta
323
being accumulated by the commit builder. It cannot be called unless the
324
method will_record_deletes() has been called to inform the builder that
325
a delta is being supplied.
327
:param path: The path of the thing deleted.
328
:param file_id: The file id that was deleted.
330
if not self._recording_deletes:
331
raise AssertionError("recording deletes not activated.")
332
delta = (path, None, file_id, None)
333
self._basis_delta.append(delta)
334
self._any_changes = True
337
def record_entry_contents(self, ie, parent_invs, path, tree,
339
"""Record the content of ie from tree into the commit if needed.
341
Side effect: sets ie.revision when unchanged
343
:param ie: An inventory entry present in the commit.
344
:param parent_invs: The inventories of the parent revisions of the
346
:param path: The path the entry is at in the tree.
347
:param tree: The tree which contains this entry and should be used to
349
:param content_summary: Summary data from the tree about the paths
350
content - stat, length, exec, sha/link target. This is only
351
accessed when the entry has a revision of None - that is when it is
352
a candidate to commit.
353
:return: A tuple (change_delta, version_recorded, fs_hash).
354
change_delta is an inventory_delta change for this entry against
355
the basis tree of the commit, or None if no change occured against
357
version_recorded is True if a new version of the entry has been
358
recorded. For instance, committing a merge where a file was only
359
changed on the other side will return (delta, False).
360
fs_hash is either None, or the hash details for the path (currently
361
a tuple of the contents sha1 and the statvalue returned by
362
tree.get_file_with_stat()).
364
if self.new_inventory.root is None:
365
if ie.parent_id is not None:
366
raise errors.RootMissing()
367
self._check_root(ie, parent_invs, tree)
368
if ie.revision is None:
369
kind = content_summary[0]
371
# ie is carried over from a prior commit
373
# XXX: repository specific check for nested tree support goes here - if
374
# the repo doesn't want nested trees we skip it ?
375
if (kind == 'tree-reference' and
376
not self.repository._format.supports_tree_reference):
377
# mismatch between commit builder logic and repository:
378
# this needs the entry creation pushed down into the builder.
379
raise NotImplementedError('Missing repository subtree support.')
380
self.new_inventory.add(ie)
382
# TODO: slow, take it out of the inner loop.
384
basis_inv = parent_invs[0]
386
basis_inv = Inventory(root_id=None)
388
# ie.revision is always None if the InventoryEntry is considered
389
# for committing. We may record the previous parents revision if the
390
# content is actually unchanged against a sole head.
391
if ie.revision is not None:
392
if not self._versioned_root and path == '':
393
# repositories that do not version the root set the root's
394
# revision to the new commit even when no change occurs (more
395
# specifically, they do not record a revision on the root; and
396
# the rev id is assigned to the root during deserialisation -
397
# this masks when a change may have occurred against the basis.
398
# To match this we always issue a delta, because the revision
399
# of the root will always be changing.
400
if ie.file_id in basis_inv:
401
delta = (basis_inv.id2path(ie.file_id), path,
405
delta = (None, path, ie.file_id, ie)
406
self._basis_delta.append(delta)
407
return delta, False, None
409
# we don't need to commit this, because the caller already
410
# determined that an existing revision of this file is
411
# appropriate. If it's not being considered for committing then
412
# it and all its parents to the root must be unaltered so
413
# no-change against the basis.
414
if ie.revision == self._new_revision_id:
415
raise AssertionError("Impossible situation, a skipped "
416
"inventory entry (%r) claims to be modified in this "
417
"commit (%r).", (ie, self._new_revision_id))
418
return None, False, None
419
# XXX: Friction: parent_candidates should return a list not a dict
420
# so that we don't have to walk the inventories again.
421
parent_candiate_entries = ie.parent_candidates(parent_invs)
422
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
424
for inv in parent_invs:
425
if ie.file_id in inv:
426
old_rev = inv[ie.file_id].revision
427
if old_rev in head_set:
428
heads.append(inv[ie.file_id].revision)
429
head_set.remove(inv[ie.file_id].revision)
432
# now we check to see if we need to write a new record to the
434
# We write a new entry unless there is one head to the ancestors, and
435
# the kind-derived content is unchanged.
437
# Cheapest check first: no ancestors, or more the one head in the
438
# ancestors, we write a new node.
442
# There is a single head, look it up for comparison
443
parent_entry = parent_candiate_entries[heads[0]]
444
# if the non-content specific data has changed, we'll be writing a
446
if (parent_entry.parent_id != ie.parent_id or
447
parent_entry.name != ie.name):
449
# now we need to do content specific checks:
451
# if the kind changed the content obviously has
452
if kind != parent_entry.kind:
454
# Stat cache fingerprint feedback for the caller - None as we usually
455
# don't generate one.
458
if content_summary[2] is None:
459
raise ValueError("Files must not have executable = None")
461
# We can't trust a check of the file length because of content
463
if (# if the exec bit has changed we have to store:
464
parent_entry.executable != content_summary[2]):
466
elif parent_entry.text_sha1 == content_summary[3]:
467
# all meta and content is unchanged (using a hash cache
468
# hit to check the sha)
469
ie.revision = parent_entry.revision
470
ie.text_size = parent_entry.text_size
471
ie.text_sha1 = parent_entry.text_sha1
472
ie.executable = parent_entry.executable
473
return self._get_delta(ie, basis_inv, path), False, None
475
# Either there is only a hash change(no hash cache entry,
476
# or same size content change), or there is no change on
478
# Provide the parent's hash to the store layer, so that the
479
# content is unchanged we will not store a new node.
480
nostore_sha = parent_entry.text_sha1
482
# We want to record a new node regardless of the presence or
483
# absence of a content change in the file.
485
ie.executable = content_summary[2]
486
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
488
text = file_obj.read()
492
ie.text_sha1, ie.text_size = self._add_text_to_weave(
493
ie.file_id, text, heads, nostore_sha)
494
# Let the caller know we generated a stat fingerprint.
495
fingerprint = (ie.text_sha1, stat_value)
496
except errors.ExistingContent:
497
# Turns out that the file content was unchanged, and we were
498
# only going to store a new node if it was changed. Carry over
500
ie.revision = parent_entry.revision
501
ie.text_size = parent_entry.text_size
502
ie.text_sha1 = parent_entry.text_sha1
503
ie.executable = parent_entry.executable
504
return self._get_delta(ie, basis_inv, path), False, None
505
elif kind == 'directory':
507
# all data is meta here, nothing specific to directory, so
509
ie.revision = parent_entry.revision
510
return self._get_delta(ie, basis_inv, path), False, None
511
self._add_text_to_weave(ie.file_id, '', heads, None)
512
elif kind == 'symlink':
513
current_link_target = content_summary[3]
515
# symlink target is not generic metadata, check if it has
517
if current_link_target != parent_entry.symlink_target:
520
# unchanged, carry over.
521
ie.revision = parent_entry.revision
522
ie.symlink_target = parent_entry.symlink_target
523
return self._get_delta(ie, basis_inv, path), False, None
524
ie.symlink_target = current_link_target
525
self._add_text_to_weave(ie.file_id, '', heads, None)
526
elif kind == 'tree-reference':
528
if content_summary[3] != parent_entry.reference_revision:
531
# unchanged, carry over.
532
ie.reference_revision = parent_entry.reference_revision
533
ie.revision = parent_entry.revision
534
return self._get_delta(ie, basis_inv, path), False, None
535
ie.reference_revision = content_summary[3]
536
if ie.reference_revision is None:
537
raise AssertionError("invalid content_summary for nested tree: %r"
538
% (content_summary,))
539
self._add_text_to_weave(ie.file_id, '', heads, None)
541
raise NotImplementedError('unknown kind')
542
ie.revision = self._new_revision_id
543
# The initial commit adds a root directory, but this in itself is not
544
# a worthwhile commit.
545
if (self.basis_delta_revision != _mod_revision.NULL_REVISION or
547
self._any_changes = True
548
return self._get_delta(ie, basis_inv, path), True, fingerprint
550
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
551
_entry_factory=entry_factory):
552
"""Record a new tree via iter_changes.
554
:param tree: The tree to obtain text contents from for changed objects.
555
:param basis_revision_id: The revision id of the tree the iter_changes
556
has been generated against. Currently assumed to be the same
557
as self.parents[0] - if it is not, errors may occur.
558
:param iter_changes: An iter_changes iterator with the changes to apply
559
to basis_revision_id. The iterator must not include any items with
560
a current kind of None - missing items must be either filtered out
561
or errored-on beefore record_iter_changes sees the item.
562
:param _entry_factory: Private method to bind entry_factory locally for
564
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
567
# Create an inventory delta based on deltas between all the parents and
568
# deltas between all the parent inventories. We use inventory delta's
569
# between the inventory objects because iter_changes masks
570
# last-changed-field only changes.
572
# file_id -> change map, change is fileid, paths, changed, versioneds,
573
# parents, names, kinds, executables
575
# {file_id -> revision_id -> inventory entry, for entries in parent
576
# trees that are not parents[0]
580
revtrees = list(self.repository.revision_trees(self.parents))
581
except errors.NoSuchRevision:
582
# one or more ghosts, slow path.
584
for revision_id in self.parents:
586
revtrees.append(self.repository.revision_tree(revision_id))
587
except errors.NoSuchRevision:
589
basis_revision_id = _mod_revision.NULL_REVISION
591
revtrees.append(self.repository.revision_tree(
592
_mod_revision.NULL_REVISION))
593
# The basis inventory from a repository
595
basis_inv = revtrees[0].inventory
597
basis_inv = self.repository.revision_tree(
598
_mod_revision.NULL_REVISION).inventory
599
if len(self.parents) > 0:
600
if basis_revision_id != self.parents[0] and not ghost_basis:
602
"arbitrary basis parents not yet supported with merges")
603
for revtree in revtrees[1:]:
604
for change in revtree.inventory._make_delta(basis_inv):
605
if change[1] is None:
606
# Not present in this parent.
608
if change[2] not in merged_ids:
609
if change[0] is not None:
610
basis_entry = basis_inv[change[2]]
611
merged_ids[change[2]] = [
613
basis_entry.revision,
616
parent_entries[change[2]] = {
618
basis_entry.revision:basis_entry,
620
change[3].revision:change[3],
623
merged_ids[change[2]] = [change[3].revision]
624
parent_entries[change[2]] = {change[3].revision:change[3]}
626
merged_ids[change[2]].append(change[3].revision)
627
parent_entries[change[2]][change[3].revision] = change[3]
630
# Setup the changes from the tree:
631
# changes maps file_id -> (change, [parent revision_ids])
633
for change in iter_changes:
634
# This probably looks up in basis_inv way to much.
635
if change[1][0] is not None:
636
head_candidate = [basis_inv[change[0]].revision]
639
changes[change[0]] = change, merged_ids.get(change[0],
641
unchanged_merged = set(merged_ids) - set(changes)
642
# Extend the changes dict with synthetic changes to record merges of
644
for file_id in unchanged_merged:
645
# Record a merged version of these items that did not change vs the
646
# basis. This can be either identical parallel changes, or a revert
647
# of a specific file after a merge. The recorded content will be
648
# that of the current tree (which is the same as the basis), but
649
# the per-file graph will reflect a merge.
650
# NB:XXX: We are reconstructing path information we had, this
651
# should be preserved instead.
652
# inv delta change: (file_id, (path_in_source, path_in_target),
653
# changed_content, versioned, parent, name, kind,
656
basis_entry = basis_inv[file_id]
657
except errors.NoSuchId:
658
# a change from basis->some_parents but file_id isn't in basis
659
# so was new in the merge, which means it must have changed
660
# from basis -> current, and as it hasn't the add was reverted
661
# by the user. So we discard this change.
665
(basis_inv.id2path(file_id), tree.id2path(file_id)),
667
(basis_entry.parent_id, basis_entry.parent_id),
668
(basis_entry.name, basis_entry.name),
669
(basis_entry.kind, basis_entry.kind),
670
(basis_entry.executable, basis_entry.executable))
671
changes[file_id] = (change, merged_ids[file_id])
672
# changes contains tuples with the change and a set of inventory
673
# candidates for the file.
675
# old_path, new_path, file_id, new_inventory_entry
676
seen_root = False # Is the root in the basis delta?
677
inv_delta = self._basis_delta
678
modified_rev = self._new_revision_id
679
for change, head_candidates in changes.values():
680
if change[3][1]: # versioned in target.
681
# Several things may be happening here:
682
# We may have a fork in the per-file graph
683
# - record a change with the content from tree
684
# We may have a change against < all trees
685
# - carry over the tree that hasn't changed
686
# We may have a change against all trees
687
# - record the change with the content from tree
690
entry = _entry_factory[kind](file_id, change[5][1],
692
head_set = self._heads(change[0], set(head_candidates))
695
for head_candidate in head_candidates:
696
if head_candidate in head_set:
697
heads.append(head_candidate)
698
head_set.remove(head_candidate)
701
# Could be a carry-over situation:
702
parent_entry_revs = parent_entries.get(file_id, None)
703
if parent_entry_revs:
704
parent_entry = parent_entry_revs.get(heads[0], None)
707
if parent_entry is None:
708
# The parent iter_changes was called against is the one
709
# that is the per-file head, so any change is relevant
710
# iter_changes is valid.
711
carry_over_possible = False
713
# could be a carry over situation
714
# A change against the basis may just indicate a merge,
715
# we need to check the content against the source of the
716
# merge to determine if it was changed after the merge
718
if (parent_entry.kind != entry.kind or
719
parent_entry.parent_id != entry.parent_id or
720
parent_entry.name != entry.name):
721
# Metadata common to all entries has changed
722
# against per-file parent
723
carry_over_possible = False
725
carry_over_possible = True
726
# per-type checks for changes against the parent_entry
729
# Cannot be a carry-over situation
730
carry_over_possible = False
731
# Populate the entry in the delta
733
# XXX: There is still a small race here: If someone reverts the content of a file
734
# after iter_changes examines and decides it has changed,
735
# we will unconditionally record a new version even if some
736
# other process reverts it while commit is running (with
737
# the revert happening after iter_changes did its
740
entry.executable = True
742
entry.executable = False
743
if (carry_over_possible and
744
parent_entry.executable == entry.executable):
745
# Check the file length, content hash after reading
747
nostore_sha = parent_entry.text_sha1
750
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
752
text = file_obj.read()
756
entry.text_sha1, entry.text_size = self._add_text_to_weave(
757
file_id, text, heads, nostore_sha)
758
yield file_id, change[1][1], (entry.text_sha1, stat_value)
759
except errors.ExistingContent:
760
# No content change against a carry_over parent
761
# Perhaps this should also yield a fs hash update?
763
entry.text_size = parent_entry.text_size
764
entry.text_sha1 = parent_entry.text_sha1
765
elif kind == 'symlink':
767
entry.symlink_target = tree.get_symlink_target(file_id)
768
if (carry_over_possible and
769
parent_entry.symlink_target == entry.symlink_target):
772
self._add_text_to_weave(change[0], '', heads, None)
773
elif kind == 'directory':
774
if carry_over_possible:
777
# Nothing to set on the entry.
778
# XXX: split into the Root and nonRoot versions.
779
if change[1][1] != '' or self.repository.supports_rich_root():
780
self._add_text_to_weave(change[0], '', heads, None)
781
elif kind == 'tree-reference':
782
if not self.repository._format.supports_tree_reference:
783
# This isn't quite sane as an error, but we shouldn't
784
# ever see this code path in practice: tree's don't
785
# permit references when the repo doesn't support tree
787
raise errors.UnsupportedOperation(tree.add_reference,
789
reference_revision = tree.get_reference_revision(change[0])
790
entry.reference_revision = reference_revision
791
if (carry_over_possible and
792
parent_entry.reference_revision == reference_revision):
795
self._add_text_to_weave(change[0], '', heads, None)
797
raise AssertionError('unknown kind %r' % kind)
799
entry.revision = modified_rev
801
entry.revision = parent_entry.revision
804
new_path = change[1][1]
805
inv_delta.append((change[1][0], new_path, change[0], entry))
808
self.new_inventory = None
809
# The initial commit adds a root directory, but this in itself is not
810
# a worthwhile commit.
811
if ((len(inv_delta) > 0 and basis_revision_id != _mod_revision.NULL_REVISION) or
812
(len(inv_delta) > 1 and basis_revision_id == _mod_revision.NULL_REVISION)):
813
# This should perhaps be guarded by a check that the basis we
814
# commit against is the basis for the commit and if not do a delta
816
self._any_changes = True
818
# housekeeping root entry changes do not affect no-change commits.
819
self._require_root_change(tree)
820
self.basis_delta_revision = basis_revision_id
822
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
823
parent_keys = tuple([(file_id, parent) for parent in parents])
824
return self.repository.texts._add_text(
825
(file_id, self._new_revision_id), parent_keys, new_text,
826
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
829
class VersionedFileRootCommitBuilder(VersionedFileCommitBuilder):
830
"""This commitbuilder actually records the root id"""
832
# the root entry gets versioned properly by this builder.
833
_versioned_root = True
835
def _check_root(self, ie, parent_invs, tree):
836
"""Helper for record_entry_contents.
838
:param ie: An entry being added.
839
:param parent_invs: The inventories of the parent revisions of the
841
:param tree: The tree that is being committed.
844
def _require_root_change(self, tree):
845
"""Enforce an appropriate root object change.
847
This is called once when record_iter_changes is called, if and only if
848
the root was not in the delta calculated by record_iter_changes.
850
:param tree: The tree which is being committed.
852
# versioned roots do not change unless the tree found a change.
855
class VersionedFileRepository(Repository):
856
"""Repository holding history for one or more branches.
858
The repository holds and retrieves historical information including
859
revisions and file history. It's normally accessed only by the Branch,
860
which views a particular line of development through that history.
862
The Repository builds on top of some byte storage facilies (the revisions,
863
signatures, inventories, texts and chk_bytes attributes) and a Transport,
864
which respectively provide byte storage and a means to access the (possibly
867
The byte storage facilities are addressed via tuples, which we refer to
868
as 'keys' throughout the code base. Revision_keys, inventory_keys and
869
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
870
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
871
byte string made up of a hash identifier and a hash value.
872
We use this interface because it allows low friction with the underlying
873
code that implements disk indices, network encoding and other parts of
876
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
877
the serialised revisions for the repository. This can be used to obtain
878
revision graph information or to access raw serialised revisions.
879
The result of trying to insert data into the repository via this store
880
is undefined: it should be considered read-only except for implementors
882
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
883
the serialised signatures for the repository. This can be used to
884
obtain access to raw serialised signatures. The result of trying to
885
insert data into the repository via this store is undefined: it should
886
be considered read-only except for implementors of repositories.
887
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
888
the serialised inventories for the repository. This can be used to
889
obtain unserialised inventories. The result of trying to insert data
890
into the repository via this store is undefined: it should be
891
considered read-only except for implementors of repositories.
892
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
893
texts of files and directories for the repository. This can be used to
894
obtain file texts or file graphs. Note that Repository.iter_file_bytes
895
is usually a better interface for accessing file texts.
896
The result of trying to insert data into the repository via this store
897
is undefined: it should be considered read-only except for implementors
899
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
900
any data the repository chooses to store or have indexed by its hash.
901
The result of trying to insert data into the repository via this store
902
is undefined: it should be considered read-only except for implementors
904
:ivar _transport: Transport for file access to repository, typically
905
pointing to .bzr/repository.
908
# What class to use for a CommitBuilder. Often it's simpler to change this
909
# in a Repository class subclass rather than to override
910
# get_commit_builder.
911
_commit_builder_class = VersionedFileCommitBuilder
913
def add_fallback_repository(self, repository):
914
"""Add a repository to use for looking up data not held locally.
916
:param repository: A repository.
918
if not self._format.supports_external_lookups:
919
raise errors.UnstackableRepositoryFormat(self._format, self.base)
921
# This repository will call fallback.unlock() when we transition to
922
# the unlocked state, so we make sure to increment the lock count
923
repository.lock_read()
924
self._check_fallback_repository(repository)
925
self._fallback_repositories.append(repository)
926
self.texts.add_fallback_versioned_files(repository.texts)
927
self.inventories.add_fallback_versioned_files(repository.inventories)
928
self.revisions.add_fallback_versioned_files(repository.revisions)
929
self.signatures.add_fallback_versioned_files(repository.signatures)
930
if self.chk_bytes is not None:
931
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
933
@only_raises(errors.LockNotHeld, errors.LockBroken)
935
super(VersionedFileRepository, self).unlock()
936
if self.control_files._lock_count == 0:
937
self._inventory_entry_cache.clear()
939
def add_inventory(self, revision_id, inv, parents):
940
"""Add the inventory inv to the repository as revision_id.
942
:param parents: The revision ids of the parents that revision_id
943
is known to have and are in the repository already.
945
:returns: The validator(which is a sha1 digest, though what is sha'd is
946
repository format specific) of the serialized inventory.
948
if not self.is_in_write_group():
949
raise AssertionError("%r not in write group" % (self,))
950
_mod_revision.check_not_reserved_id(revision_id)
951
if not (inv.revision_id is None or inv.revision_id == revision_id):
952
raise AssertionError(
953
"Mismatch between inventory revision"
954
" id and insertion revid (%r, %r)"
955
% (inv.revision_id, revision_id))
957
raise errors.RootMissing()
958
return self._add_inventory_checked(revision_id, inv, parents)
960
def _add_inventory_checked(self, revision_id, inv, parents):
961
"""Add inv to the repository after checking the inputs.
963
This function can be overridden to allow different inventory styles.
965
:seealso: add_inventory, for the contract.
967
inv_lines = self._serializer.write_inventory_to_lines(inv)
968
return self._inventory_add_lines(revision_id, parents,
969
inv_lines, check_content=False)
971
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
972
parents, basis_inv=None, propagate_caches=False):
973
"""Add a new inventory expressed as a delta against another revision.
975
See the inventory developers documentation for the theory behind
978
:param basis_revision_id: The inventory id the delta was created
979
against. (This does not have to be a direct parent.)
980
:param delta: The inventory delta (see Inventory.apply_delta for
982
:param new_revision_id: The revision id that the inventory is being
984
:param parents: The revision ids of the parents that revision_id is
985
known to have and are in the repository already. These are supplied
986
for repositories that depend on the inventory graph for revision
987
graph access, as well as for those that pun ancestry with delta
989
:param basis_inv: The basis inventory if it is already known,
991
:param propagate_caches: If True, the caches for this inventory are
992
copied to and updated for the result if possible.
994
:returns: (validator, new_inv)
995
The validator(which is a sha1 digest, though what is sha'd is
996
repository format specific) of the serialized inventory, and the
999
if not self.is_in_write_group():
1000
raise AssertionError("%r not in write group" % (self,))
1001
_mod_revision.check_not_reserved_id(new_revision_id)
1002
basis_tree = self.revision_tree(basis_revision_id)
1003
basis_tree.lock_read()
1005
# Note that this mutates the inventory of basis_tree, which not all
1006
# inventory implementations may support: A better idiom would be to
1007
# return a new inventory, but as there is no revision tree cache in
1008
# repository this is safe for now - RBC 20081013
1009
if basis_inv is None:
1010
basis_inv = basis_tree.inventory
1011
basis_inv.apply_delta(delta)
1012
basis_inv.revision_id = new_revision_id
1013
return (self.add_inventory(new_revision_id, basis_inv, parents),
1018
def _inventory_add_lines(self, revision_id, parents, lines,
1019
check_content=True):
1020
"""Store lines in inv_vf and return the sha1 of the inventory."""
1021
parents = [(parent,) for parent in parents]
1022
result = self.inventories.add_lines((revision_id,), parents, lines,
1023
check_content=check_content)[0]
1024
self.inventories._access.flush()
1027
def add_revision(self, revision_id, rev, inv=None, config=None):
1028
"""Add rev to the revision store as revision_id.
1030
:param revision_id: the revision id to use.
1031
:param rev: The revision object.
1032
:param inv: The inventory for the revision. if None, it will be looked
1033
up in the inventory storer
1034
:param config: If None no digital signature will be created.
1035
If supplied its signature_needed method will be used
1036
to determine if a signature should be made.
1038
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1040
_mod_revision.check_not_reserved_id(revision_id)
1041
if config is not None and config.signature_needed():
1043
inv = self.get_inventory(revision_id)
1044
tree = InventoryRevisionTree(self, inv, revision_id)
1045
testament = Testament(rev, tree)
1046
plaintext = testament.as_short_text()
1047
self.store_revision_signature(
1048
gpg.GPGStrategy(config), plaintext, revision_id)
1049
# check inventory present
1050
if not self.inventories.get_parent_map([(revision_id,)]):
1052
raise errors.WeaveRevisionNotPresent(revision_id,
1055
# yes, this is not suitable for adding with ghosts.
1056
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1059
key = (revision_id,)
1060
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1061
self._add_revision(rev)
1063
def _add_revision(self, revision):
1064
text = self._serializer.write_revision_to_string(revision)
1065
key = (revision.revision_id,)
1066
parents = tuple((parent,) for parent in revision.parent_ids)
1067
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1069
def _check_inventories(self, checker):
1070
"""Check the inventories found from the revision scan.
1072
This is responsible for verifying the sha1 of inventories and
1073
creating a pending_keys set that covers data referenced by inventories.
1075
bar = ui.ui_factory.nested_progress_bar()
1077
self._do_check_inventories(checker, bar)
1081
def _do_check_inventories(self, checker, bar):
1082
"""Helper for _check_inventories."""
1084
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1085
kinds = ['chk_bytes', 'texts']
1086
count = len(checker.pending_keys)
1087
bar.update("inventories", 0, 2)
1088
current_keys = checker.pending_keys
1089
checker.pending_keys = {}
1090
# Accumulate current checks.
1091
for key in current_keys:
1092
if key[0] != 'inventories' and key[0] not in kinds:
1093
checker._report_items.append('unknown key type %r' % (key,))
1094
keys[key[0]].add(key[1:])
1095
if keys['inventories']:
1096
# NB: output order *should* be roughly sorted - topo or
1097
# inverse topo depending on repository - either way decent
1098
# to just delta against. However, pre-CHK formats didn't
1099
# try to optimise inventory layout on disk. As such the
1100
# pre-CHK code path does not use inventory deltas.
1102
for record in self.inventories.check(keys=keys['inventories']):
1103
if record.storage_kind == 'absent':
1104
checker._report_items.append(
1105
'Missing inventory {%s}' % (record.key,))
1107
last_object = self._check_record('inventories', record,
1108
checker, last_object,
1109
current_keys[('inventories',) + record.key])
1110
del keys['inventories']
1113
bar.update("texts", 1)
1114
while (checker.pending_keys or keys['chk_bytes']
1116
# Something to check.
1117
current_keys = checker.pending_keys
1118
checker.pending_keys = {}
1119
# Accumulate current checks.
1120
for key in current_keys:
1121
if key[0] not in kinds:
1122
checker._report_items.append('unknown key type %r' % (key,))
1123
keys[key[0]].add(key[1:])
1124
# Check the outermost kind only - inventories || chk_bytes || texts
1128
for record in getattr(self, kind).check(keys=keys[kind]):
1129
if record.storage_kind == 'absent':
1130
checker._report_items.append(
1131
'Missing %s {%s}' % (kind, record.key,))
1133
last_object = self._check_record(kind, record,
1134
checker, last_object, current_keys[(kind,) + record.key])
1138
def _check_record(self, kind, record, checker, last_object, item_data):
1139
"""Check a single text from this repository."""
1140
if kind == 'inventories':
1141
rev_id = record.key[0]
1142
inv = self._deserialise_inventory(rev_id,
1143
record.get_bytes_as('fulltext'))
1144
if last_object is not None:
1145
delta = inv._make_delta(last_object)
1146
for old_path, path, file_id, ie in delta:
1149
ie.check(checker, rev_id, inv)
1151
for path, ie in inv.iter_entries():
1152
ie.check(checker, rev_id, inv)
1153
if self._format.fast_deltas:
1155
elif kind == 'chk_bytes':
1156
# No code written to check chk_bytes for this repo format.
1157
checker._report_items.append(
1158
'unsupported key type chk_bytes for %s' % (record.key,))
1159
elif kind == 'texts':
1160
self._check_text(record, checker, item_data)
1162
checker._report_items.append(
1163
'unknown key type %s for %s' % (kind, record.key))
1165
def _check_text(self, record, checker, item_data):
1166
"""Check a single text."""
1167
# Check it is extractable.
1168
# TODO: check length.
1169
if record.storage_kind == 'chunked':
1170
chunks = record.get_bytes_as(record.storage_kind)
1171
sha1 = osutils.sha_strings(chunks)
1172
length = sum(map(len, chunks))
1174
content = record.get_bytes_as('fulltext')
1175
sha1 = osutils.sha_string(content)
1176
length = len(content)
1177
if item_data and sha1 != item_data[1]:
1178
checker._report_items.append(
1179
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1180
(record.key, sha1, item_data[1], item_data[2]))
1182
def __init__(self, _format, a_bzrdir, control_files):
1183
"""Instantiate a VersionedFileRepository.
1185
:param _format: The format of the repository on disk.
1186
:param a_bzrdir: The BzrDir of the repository.
1187
:param control_files: Control files to use for locking, etc.
1189
# In the future we will have a single api for all stores for
1190
# getting file texts, inventories and revisions, then
1191
# this construct will accept instances of those things.
1192
super(VersionedFileRepository, self).__init__(_format, a_bzrdir,
1195
self._reconcile_does_inventory_gc = True
1196
self._reconcile_fixes_text_parents = False
1197
self._reconcile_backsup_inventory = True
1198
# An InventoryEntry cache, used during deserialization
1199
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1200
# Is it safe to return inventory entries directly from the entry cache,
1201
# rather copying them?
1202
self._safe_to_return_from_cache = False
1205
def gather_stats(self, revid=None, committers=None):
1206
"""See Repository.gather_stats()."""
1207
result = super(VersionedFileRepository, self).gather_stats(revid, committers)
1208
# now gather global repository information
1209
# XXX: This is available for many repos regardless of listability.
1210
if self.user_transport.listable():
1211
# XXX: do we want to __define len__() ?
1212
# Maybe the versionedfiles object should provide a different
1213
# method to get the number of keys.
1214
result['revisions'] = len(self.revisions.keys())
1215
# result['size'] = t
1218
def get_commit_builder(self, branch, parents, config, timestamp=None,
1219
timezone=None, committer=None, revprops=None,
1220
revision_id=None, lossy=False):
1221
"""Obtain a CommitBuilder for this repository.
1223
:param branch: Branch to commit to.
1224
:param parents: Revision ids of the parents of the new revision.
1225
:param config: Configuration to use.
1226
:param timestamp: Optional timestamp recorded for commit.
1227
:param timezone: Optional timezone for timestamp.
1228
:param committer: Optional committer to set for commit.
1229
:param revprops: Optional dictionary of revision properties.
1230
:param revision_id: Optional revision id.
1231
:param lossy: Whether to discard data that can not be natively
1232
represented, when pushing to a foreign VCS
1234
if self._fallback_repositories and not self._format.supports_chks:
1235
raise errors.BzrError("Cannot commit directly to a stacked branch"
1236
" in pre-2a formats. See "
1237
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1238
result = self._commit_builder_class(self, parents, config,
1239
timestamp, timezone, committer, revprops, revision_id,
1241
self.start_write_group()
1244
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1245
"""Return the keys of missing inventory parents for revisions added in
1248
A revision is not complete if the inventory delta for that revision
1249
cannot be calculated. Therefore if the parent inventories of a
1250
revision are not present, the revision is incomplete, and e.g. cannot
1251
be streamed by a smart server. This method finds missing inventory
1252
parents for revisions added in this write group.
1254
if not self._format.supports_external_lookups:
1255
# This is only an issue for stacked repositories
1257
if not self.is_in_write_group():
1258
raise AssertionError('not in a write group')
1260
# XXX: We assume that every added revision already has its
1261
# corresponding inventory, so we only check for parent inventories that
1262
# might be missing, rather than all inventories.
1263
parents = set(self.revisions._index.get_missing_parents())
1264
parents.discard(_mod_revision.NULL_REVISION)
1265
unstacked_inventories = self.inventories._index
1266
present_inventories = unstacked_inventories.get_parent_map(
1267
key[-1:] for key in parents)
1268
parents.difference_update(present_inventories)
1269
if len(parents) == 0:
1270
# No missing parent inventories.
1272
if not check_for_missing_texts:
1273
return set(('inventories', rev_id) for (rev_id,) in parents)
1274
# Ok, now we have a list of missing inventories. But these only matter
1275
# if the inventories that reference them are missing some texts they
1276
# appear to introduce.
1277
# XXX: Texts referenced by all added inventories need to be present,
1278
# but at the moment we're only checking for texts referenced by
1279
# inventories at the graph's edge.
1280
key_deps = self.revisions._index._key_dependencies
1281
key_deps.satisfy_refs_for_keys(present_inventories)
1282
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1283
file_ids = self.fileids_altered_by_revision_ids(referrers)
1284
missing_texts = set()
1285
for file_id, version_ids in file_ids.iteritems():
1286
missing_texts.update(
1287
(file_id, version_id) for version_id in version_ids)
1288
present_texts = self.texts.get_parent_map(missing_texts)
1289
missing_texts.difference_update(present_texts)
1290
if not missing_texts:
1291
# No texts are missing, so all revisions and their deltas are
1294
# Alternatively the text versions could be returned as the missing
1295
# keys, but this is likely to be less data.
1296
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1300
def has_revisions(self, revision_ids):
1301
"""Probe to find out the presence of multiple revisions.
1303
:param revision_ids: An iterable of revision_ids.
1304
:return: A set of the revision_ids that were present.
1306
parent_map = self.revisions.get_parent_map(
1307
[(rev_id,) for rev_id in revision_ids])
1309
if _mod_revision.NULL_REVISION in revision_ids:
1310
result.add(_mod_revision.NULL_REVISION)
1311
result.update([key[0] for key in parent_map])
1315
def get_revision_reconcile(self, revision_id):
1316
"""'reconcile' helper routine that allows access to a revision always.
1318
This variant of get_revision does not cross check the weave graph
1319
against the revision one as get_revision does: but it should only
1320
be used by reconcile, or reconcile-alike commands that are correcting
1321
or testing the revision graph.
1323
return self._get_revisions([revision_id])[0]
1326
def get_revisions(self, revision_ids):
1327
"""Get many revisions at once.
1329
Repositories that need to check data on every revision read should
1330
subclass this method.
1332
return self._get_revisions(revision_ids)
1335
def _get_revisions(self, revision_ids):
1336
"""Core work logic to get many revisions without sanity checks."""
1338
for revid, rev in self._iter_revisions(revision_ids):
1340
raise errors.NoSuchRevision(self, revid)
1342
return [revs[revid] for revid in revision_ids]
1344
def _iter_revisions(self, revision_ids):
1345
"""Iterate over revision objects.
1347
:param revision_ids: An iterable of revisions to examine. None may be
1348
passed to request all revisions known to the repository. Note that
1349
not all repositories can find unreferenced revisions; for those
1350
repositories only referenced ones will be returned.
1351
:return: An iterator of (revid, revision) tuples. Absent revisions (
1352
those asked for but not available) are returned as (revid, None).
1354
if revision_ids is None:
1355
revision_ids = self.all_revision_ids()
1357
for rev_id in revision_ids:
1358
if not rev_id or not isinstance(rev_id, basestring):
1359
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1360
keys = [(key,) for key in revision_ids]
1361
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1362
for record in stream:
1363
revid = record.key[0]
1364
if record.storage_kind == 'absent':
1367
text = record.get_bytes_as('fulltext')
1368
rev = self._serializer.read_revision_from_string(text)
1372
def add_signature_text(self, revision_id, signature):
1373
"""Store a signature text for a revision.
1375
:param revision_id: Revision id of the revision
1376
:param signature: Signature text.
1378
self.signatures.add_lines((revision_id,), (),
1379
osutils.split_lines(signature))
1381
def find_text_key_references(self):
1382
"""Find the text key references within the repository.
1384
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1385
to whether they were referred to by the inventory of the
1386
revision_id that they contain. The inventory texts from all present
1387
revision ids are assessed to generate this report.
1389
revision_keys = self.revisions.keys()
1390
w = self.inventories
1391
pb = ui.ui_factory.nested_progress_bar()
1393
return self._serializer._find_text_key_references(
1394
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1398
def _inventory_xml_lines_for_keys(self, keys):
1399
"""Get a line iterator of the sort needed for findind references.
1401
Not relevant for non-xml inventory repositories.
1403
Ghosts in revision_keys are ignored.
1405
:param revision_keys: The revision keys for the inventories to inspect.
1406
:return: An iterator over (inventory line, revid) for the fulltexts of
1407
all of the xml inventories specified by revision_keys.
1409
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1410
for record in stream:
1411
if record.storage_kind != 'absent':
1412
chunks = record.get_bytes_as('chunked')
1413
revid = record.key[-1]
1414
lines = osutils.chunks_to_lines(chunks)
1418
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1420
"""Helper routine for fileids_altered_by_revision_ids.
1422
This performs the translation of xml lines to revision ids.
1424
:param line_iterator: An iterator of lines, origin_version_id
1425
:param revision_keys: The revision ids to filter for. This should be a
1426
set or other type which supports efficient __contains__ lookups, as
1427
the revision key from each parsed line will be looked up in the
1428
revision_keys filter.
1429
:return: a dictionary mapping altered file-ids to an iterable of
1430
revision_ids. Each altered file-ids has the exact revision_ids that
1431
altered it listed explicitly.
1433
seen = set(self._serializer._find_text_key_references(
1434
line_iterator).iterkeys())
1435
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
1436
parent_seen = set(self._serializer._find_text_key_references(
1437
self._inventory_xml_lines_for_keys(parent_keys)))
1438
new_keys = seen - parent_seen
1440
setdefault = result.setdefault
1441
for key in new_keys:
1442
setdefault(key[0], set()).add(key[-1])
1445
def _find_parent_keys_of_revisions(self, revision_keys):
1446
"""Similar to _find_parent_ids_of_revisions, but used with keys.
1448
:param revision_keys: An iterable of revision_keys.
1449
:return: The parents of all revision_keys that are not already in
1452
parent_map = self.revisions.get_parent_map(revision_keys)
1454
map(parent_keys.update, parent_map.itervalues())
1455
parent_keys.difference_update(revision_keys)
1456
parent_keys.discard(_mod_revision.NULL_REVISION)
1459
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
1460
"""Find the file ids and versions affected by revisions.
1462
:param revisions: an iterable containing revision ids.
1463
:param _inv_weave: The inventory weave from this repository or None.
1464
If None, the inventory weave will be opened automatically.
1465
:return: a dictionary mapping altered file-ids to an iterable of
1466
revision_ids. Each altered file-ids has the exact revision_ids that
1467
altered it listed explicitly.
1469
selected_keys = set((revid,) for revid in revision_ids)
1470
w = _inv_weave or self.inventories
1471
return self._find_file_ids_from_xml_inventory_lines(
1472
w.iter_lines_added_or_present_in_keys(
1473
selected_keys, pb=None),
1476
def iter_files_bytes(self, desired_files):
1477
"""Iterate through file versions.
1479
Files will not necessarily be returned in the order they occur in
1480
desired_files. No specific order is guaranteed.
1482
Yields pairs of identifier, bytes_iterator. identifier is an opaque
1483
value supplied by the caller as part of desired_files. It should
1484
uniquely identify the file version in the caller's context. (Examples:
1485
an index number or a TreeTransform trans_id.)
1487
bytes_iterator is an iterable of bytestrings for the file. The
1488
kind of iterable and length of the bytestrings are unspecified, but for
1489
this implementation, it is a list of bytes produced by
1490
VersionedFile.get_record_stream().
1492
:param desired_files: a list of (file_id, revision_id, identifier)
1496
for file_id, revision_id, callable_data in desired_files:
1497
text_keys[(file_id, revision_id)] = callable_data
1498
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
1499
if record.storage_kind == 'absent':
1500
raise errors.RevisionNotPresent(record.key, self)
1501
yield text_keys[record.key], record.get_bytes_as('chunked')
1503
def _generate_text_key_index(self, text_key_references=None,
1505
"""Generate a new text key index for the repository.
1507
This is an expensive function that will take considerable time to run.
1509
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
1510
list of parents, also text keys. When a given key has no parents,
1511
the parents list will be [NULL_REVISION].
1513
# All revisions, to find inventory parents.
1514
if ancestors is None:
1515
graph = self.get_graph()
1516
ancestors = graph.get_parent_map(self.all_revision_ids())
1517
if text_key_references is None:
1518
text_key_references = self.find_text_key_references()
1519
pb = ui.ui_factory.nested_progress_bar()
1521
return self._do_generate_text_key_index(ancestors,
1522
text_key_references, pb)
1526
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
1527
"""Helper for _generate_text_key_index to avoid deep nesting."""
1528
revision_order = tsort.topo_sort(ancestors)
1529
invalid_keys = set()
1531
for revision_id in revision_order:
1532
revision_keys[revision_id] = set()
1533
text_count = len(text_key_references)
1534
# a cache of the text keys to allow reuse; costs a dict of all the
1535
# keys, but saves a 2-tuple for every child of a given key.
1537
for text_key, valid in text_key_references.iteritems():
1539
invalid_keys.add(text_key)
1541
revision_keys[text_key[1]].add(text_key)
1542
text_key_cache[text_key] = text_key
1543
del text_key_references
1545
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
1546
NULL_REVISION = _mod_revision.NULL_REVISION
1547
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
1548
# too small for large or very branchy trees. However, for 55K path
1549
# trees, it would be easy to use too much memory trivially. Ideally we
1550
# could gauge this by looking at available real memory etc, but this is
1551
# always a tricky proposition.
1552
inventory_cache = lru_cache.LRUCache(10)
1553
batch_size = 10 # should be ~150MB on a 55K path tree
1554
batch_count = len(revision_order) / batch_size + 1
1556
pb.update("Calculating text parents", processed_texts, text_count)
1557
for offset in xrange(batch_count):
1558
to_query = revision_order[offset * batch_size:(offset + 1) *
1562
for revision_id in to_query:
1563
parent_ids = ancestors[revision_id]
1564
for text_key in revision_keys[revision_id]:
1565
pb.update("Calculating text parents", processed_texts)
1566
processed_texts += 1
1567
candidate_parents = []
1568
for parent_id in parent_ids:
1569
parent_text_key = (text_key[0], parent_id)
1571
check_parent = parent_text_key not in \
1572
revision_keys[parent_id]
1574
# the parent parent_id is a ghost:
1575
check_parent = False
1576
# truncate the derived graph against this ghost.
1577
parent_text_key = None
1579
# look at the parent commit details inventories to
1580
# determine possible candidates in the per file graph.
1583
inv = inventory_cache[parent_id]
1585
inv = self.revision_tree(parent_id).inventory
1586
inventory_cache[parent_id] = inv
1588
parent_entry = inv[text_key[0]]
1589
except (KeyError, errors.NoSuchId):
1591
if parent_entry is not None:
1593
text_key[0], parent_entry.revision)
1595
parent_text_key = None
1596
if parent_text_key is not None:
1597
candidate_parents.append(
1598
text_key_cache[parent_text_key])
1599
parent_heads = text_graph.heads(candidate_parents)
1600
new_parents = list(parent_heads)
1601
new_parents.sort(key=lambda x:candidate_parents.index(x))
1602
if new_parents == []:
1603
new_parents = [NULL_REVISION]
1604
text_index[text_key] = new_parents
1606
for text_key in invalid_keys:
1607
text_index[text_key] = [NULL_REVISION]
1610
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
1611
"""Get an iterable listing the keys of all the data introduced by a set
1614
The keys will be ordered so that the corresponding items can be safely
1615
fetched and inserted in that order.
1617
:returns: An iterable producing tuples of (knit-kind, file-id,
1618
versions). knit-kind is one of 'file', 'inventory', 'signatures',
1619
'revisions'. file-id is None unless knit-kind is 'file'.
1621
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
1624
for result in self._find_non_file_keys_to_fetch(revision_ids):
1627
def _find_file_keys_to_fetch(self, revision_ids, pb):
1628
# XXX: it's a bit weird to control the inventory weave caching in this
1629
# generator. Ideally the caching would be done in fetch.py I think. Or
1630
# maybe this generator should explicitly have the contract that it
1631
# should not be iterated until the previously yielded item has been
1633
inv_w = self.inventories
1635
# file ids that changed
1636
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
1638
num_file_ids = len(file_ids)
1639
for file_id, altered_versions in file_ids.iteritems():
1641
pb.update("Fetch texts", count, num_file_ids)
1643
yield ("file", file_id, altered_versions)
1645
def _find_non_file_keys_to_fetch(self, revision_ids):
1647
yield ("inventory", None, revision_ids)
1650
# XXX: Note ATM no callers actually pay attention to this return
1651
# instead they just use the list of revision ids and ignore
1652
# missing sigs. Consider removing this work entirely
1653
revisions_with_signatures = set(self.signatures.get_parent_map(
1654
[(r,) for r in revision_ids]))
1655
revisions_with_signatures = set(
1656
[r for (r,) in revisions_with_signatures])
1657
revisions_with_signatures.intersection_update(revision_ids)
1658
yield ("signatures", None, revisions_with_signatures)
1661
yield ("revisions", None, revision_ids)
1664
def get_inventory(self, revision_id):
1665
"""Get Inventory object by revision id."""
1666
return self.iter_inventories([revision_id]).next()
1668
def iter_inventories(self, revision_ids, ordering=None):
1669
"""Get many inventories by revision_ids.
1671
This will buffer some or all of the texts used in constructing the
1672
inventories in memory, but will only parse a single inventory at a
1675
:param revision_ids: The expected revision ids of the inventories.
1676
:param ordering: optional ordering, e.g. 'topological'. If not
1677
specified, the order of revision_ids will be preserved (by
1678
buffering if necessary).
1679
:return: An iterator of inventories.
1681
if ((None in revision_ids)
1682
or (_mod_revision.NULL_REVISION in revision_ids)):
1683
raise ValueError('cannot get null revision inventory')
1684
return self._iter_inventories(revision_ids, ordering)
1686
def _iter_inventories(self, revision_ids, ordering):
1687
"""single-document based inventory iteration."""
1688
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
1689
for text, revision_id in inv_xmls:
1690
yield self._deserialise_inventory(revision_id, text)
1692
def _iter_inventory_xmls(self, revision_ids, ordering):
1693
if ordering is None:
1694
order_as_requested = True
1695
ordering = 'unordered'
1697
order_as_requested = False
1698
keys = [(revision_id,) for revision_id in revision_ids]
1701
if order_as_requested:
1702
key_iter = iter(keys)
1703
next_key = key_iter.next()
1704
stream = self.inventories.get_record_stream(keys, ordering, True)
1706
for record in stream:
1707
if record.storage_kind != 'absent':
1708
chunks = record.get_bytes_as('chunked')
1709
if order_as_requested:
1710
text_chunks[record.key] = chunks
1712
yield ''.join(chunks), record.key[-1]
1714
raise errors.NoSuchRevision(self, record.key)
1715
if order_as_requested:
1716
# Yield as many results as we can while preserving order.
1717
while next_key in text_chunks:
1718
chunks = text_chunks.pop(next_key)
1719
yield ''.join(chunks), next_key[-1]
1721
next_key = key_iter.next()
1722
except StopIteration:
1723
# We still want to fully consume the get_record_stream,
1724
# just in case it is not actually finished at this point
1728
def _deserialise_inventory(self, revision_id, xml):
1729
"""Transform the xml into an inventory object.
1731
:param revision_id: The expected revision id of the inventory.
1732
:param xml: A serialised inventory.
1734
result = self._serializer.read_inventory_from_string(xml, revision_id,
1735
entry_cache=self._inventory_entry_cache,
1736
return_from_cache=self._safe_to_return_from_cache)
1737
if result.revision_id != revision_id:
1738
raise AssertionError('revision id mismatch %s != %s' % (
1739
result.revision_id, revision_id))
1742
def get_serializer_format(self):
1743
return self._serializer.format_num
1746
def _get_inventory_xml(self, revision_id):
1747
"""Get serialized inventory as a string."""
1748
texts = self._iter_inventory_xmls([revision_id], 'unordered')
1750
text, revision_id = texts.next()
1751
except StopIteration:
1752
raise errors.HistoryMissing(self, 'inventory', revision_id)
1756
def revision_tree(self, revision_id):
1757
"""Return Tree for a revision on this branch.
1759
`revision_id` may be NULL_REVISION for the empty tree revision.
1761
revision_id = _mod_revision.ensure_null(revision_id)
1762
# TODO: refactor this to use an existing revision object
1763
# so we don't need to read it in twice.
1764
if revision_id == _mod_revision.NULL_REVISION:
1765
return InventoryRevisionTree(self,
1766
Inventory(root_id=None), _mod_revision.NULL_REVISION)
1768
inv = self.get_inventory(revision_id)
1769
return InventoryRevisionTree(self, inv, revision_id)
1771
def revision_trees(self, revision_ids):
1772
"""Return Trees for revisions in this repository.
1774
:param revision_ids: a sequence of revision-ids;
1775
a revision-id may not be None or 'null:'
1777
inventories = self.iter_inventories(revision_ids)
1778
for inv in inventories:
1779
yield InventoryRevisionTree(self, inv, inv.revision_id)
1781
def _filtered_revision_trees(self, revision_ids, file_ids):
1782
"""Return Tree for a revision on this branch with only some files.
1784
:param revision_ids: a sequence of revision-ids;
1785
a revision-id may not be None or 'null:'
1786
:param file_ids: if not None, the result is filtered
1787
so that only those file-ids, their parents and their
1788
children are included.
1790
inventories = self.iter_inventories(revision_ids)
1791
for inv in inventories:
1792
# Should we introduce a FilteredRevisionTree class rather
1793
# than pre-filter the inventory here?
1794
filtered_inv = inv.filter(file_ids)
1795
yield InventoryRevisionTree(self, filtered_inv, filtered_inv.revision_id)
1797
def get_parent_map(self, revision_ids):
1798
"""See graph.StackedParentsProvider.get_parent_map"""
1799
# revisions index works in keys; this just works in revisions
1800
# therefore wrap and unwrap
1803
for revision_id in revision_ids:
1804
if revision_id == _mod_revision.NULL_REVISION:
1805
result[revision_id] = ()
1806
elif revision_id is None:
1807
raise ValueError('get_parent_map(None) is not valid')
1809
query_keys.append((revision_id ,))
1810
for ((revision_id,), parent_keys) in \
1811
self.revisions.get_parent_map(query_keys).iteritems():
1813
result[revision_id] = tuple([parent_revid
1814
for (parent_revid,) in parent_keys])
1816
result[revision_id] = (_mod_revision.NULL_REVISION,)
1820
def get_known_graph_ancestry(self, revision_ids):
1821
"""Return the known graph for a set of revision ids and their ancestors.
1823
st = static_tuple.StaticTuple
1824
revision_keys = [st(r_id).intern() for r_id in revision_ids]
1825
known_graph = self.revisions.get_known_graph_ancestry(revision_keys)
1826
return graph.GraphThunkIdsToKeys(known_graph)
1829
def get_file_graph(self):
1830
"""Return the graph walker for text revisions."""
1831
return graph.Graph(self.texts)
1833
def _get_versioned_file_checker(self, text_key_references=None,
1835
"""Return an object suitable for checking versioned files.
1837
:param text_key_references: if non-None, an already built
1838
dictionary mapping text keys ((fileid, revision_id) tuples)
1839
to whether they were referred to by the inventory of the
1840
revision_id that they contain. If None, this will be
1842
:param ancestors: Optional result from
1843
self.get_graph().get_parent_map(self.all_revision_ids()) if already
1846
return _VersionedFileChecker(self,
1847
text_key_references=text_key_references, ancestors=ancestors)
1850
def has_signature_for_revision_id(self, revision_id):
1851
"""Query for a revision signature for revision_id in the repository."""
1852
if not self.has_revision(revision_id):
1853
raise errors.NoSuchRevision(self, revision_id)
1854
sig_present = (1 == len(
1855
self.signatures.get_parent_map([(revision_id,)])))
1859
def get_signature_text(self, revision_id):
1860
"""Return the text for a signature."""
1861
stream = self.signatures.get_record_stream([(revision_id,)],
1863
record = stream.next()
1864
if record.storage_kind == 'absent':
1865
raise errors.NoSuchRevision(self, revision_id)
1866
return record.get_bytes_as('fulltext')
1869
def _check(self, revision_ids, callback_refs, check_repo):
1870
result = check.VersionedFileCheck(self, check_repo=check_repo)
1871
result.check(callback_refs)
1874
def _find_inconsistent_revision_parents(self, revisions_iterator=None):
1875
"""Find revisions with different parent lists in the revision object
1876
and in the index graph.
1878
:param revisions_iterator: None, or an iterator of (revid,
1879
Revision-or-None). This iterator controls the revisions checked.
1880
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1881
parents-in-revision).
1883
if not self.is_locked():
1884
raise AssertionError()
1886
if revisions_iterator is None:
1887
revisions_iterator = self._iter_revisions(None)
1888
for revid, revision in revisions_iterator:
1889
if revision is None:
1891
parent_map = vf.get_parent_map([(revid,)])
1892
parents_according_to_index = tuple(parent[-1] for parent in
1893
parent_map[(revid,)])
1894
parents_according_to_revision = tuple(revision.parent_ids)
1895
if parents_according_to_index != parents_according_to_revision:
1896
yield (revid, parents_according_to_index,
1897
parents_according_to_revision)
1899
def _check_for_inconsistent_revision_parents(self):
1900
inconsistencies = list(self._find_inconsistent_revision_parents())
1902
raise errors.BzrCheckError(
1903
"Revision knit has inconsistent parents.")
1905
def _get_sink(self):
1906
"""Return a sink for streaming into this repository."""
1907
return StreamSink(self)
1909
def _get_source(self, to_format):
1910
"""Return a source for streaming from this repository."""
1911
return StreamSource(self, to_format)
1914
class MetaDirVersionedFileRepository(MetaDirRepository,
1915
VersionedFileRepository):
1916
"""Repositories in a meta-dir, that work via versioned file objects."""
1918
def __init__(self, _format, a_bzrdir, control_files):
1919
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
1923
class MetaDirVersionedFileRepositoryFormat(MetaDirRepositoryFormat,
1924
VersionedFileRepositoryFormat):
1925
"""Base class for repository formats using versioned files in metadirs."""
1928
class StreamSink(object):
1929
"""An object that can insert a stream into a repository.
1931
This interface handles the complexity of reserialising inventories and
1932
revisions from different formats, and allows unidirectional insertion into
1933
stacked repositories without looking for the missing basis parents
1937
def __init__(self, target_repo):
1938
self.target_repo = target_repo
1940
def insert_stream(self, stream, src_format, resume_tokens):
1941
"""Insert a stream's content into the target repository.
1943
:param src_format: a bzr repository format.
1945
:return: a list of resume tokens and an iterable of keys additional
1946
items required before the insertion can be completed.
1948
self.target_repo.lock_write()
1951
self.target_repo.resume_write_group(resume_tokens)
1954
self.target_repo.start_write_group()
1957
# locked_insert_stream performs a commit|suspend.
1958
missing_keys = self.insert_stream_without_locking(stream,
1959
src_format, is_resume)
1961
# suspend the write group and tell the caller what we is
1962
# missing. We know we can suspend or else we would not have
1963
# entered this code path. (All repositories that can handle
1964
# missing keys can handle suspending a write group).
1965
write_group_tokens = self.target_repo.suspend_write_group()
1966
return write_group_tokens, missing_keys
1967
hint = self.target_repo.commit_write_group()
1968
to_serializer = self.target_repo._format._serializer
1969
src_serializer = src_format._serializer
1970
if (to_serializer != src_serializer and
1971
self.target_repo._format.pack_compresses):
1972
self.target_repo.pack(hint=hint)
1975
self.target_repo.abort_write_group(suppress_errors=True)
1978
self.target_repo.unlock()
1980
def insert_stream_without_locking(self, stream, src_format,
1982
"""Insert a stream's content into the target repository.
1984
This assumes that you already have a locked repository and an active
1987
:param src_format: a bzr repository format.
1988
:param is_resume: Passed down to get_missing_parent_inventories to
1989
indicate if we should be checking for missing texts at the same
1992
:return: A set of keys that are missing.
1994
if not self.target_repo.is_write_locked():
1995
raise errors.ObjectNotLocked(self)
1996
if not self.target_repo.is_in_write_group():
1997
raise errors.BzrError('you must already be in a write group')
1998
to_serializer = self.target_repo._format._serializer
1999
src_serializer = src_format._serializer
2001
if to_serializer == src_serializer:
2002
# If serializers match and the target is a pack repository, set the
2003
# write cache size on the new pack. This avoids poor performance
2004
# on transports where append is unbuffered (such as
2005
# RemoteTransport). This is safe to do because nothing should read
2006
# back from the target repository while a stream with matching
2007
# serialization is being inserted.
2008
# The exception is that a delta record from the source that should
2009
# be a fulltext may need to be expanded by the target (see
2010
# test_fetch_revisions_with_deltas_into_pack); but we take care to
2011
# explicitly flush any buffered writes first in that rare case.
2013
new_pack = self.target_repo._pack_collection._new_pack
2014
except AttributeError:
2015
# Not a pack repository
2018
new_pack.set_write_cache_size(1024*1024)
2019
for substream_type, substream in stream:
2020
if 'stream' in debug.debug_flags:
2021
mutter('inserting substream: %s', substream_type)
2022
if substream_type == 'texts':
2023
self.target_repo.texts.insert_record_stream(substream)
2024
elif substream_type == 'inventories':
2025
if src_serializer == to_serializer:
2026
self.target_repo.inventories.insert_record_stream(
2029
self._extract_and_insert_inventories(
2030
substream, src_serializer)
2031
elif substream_type == 'inventory-deltas':
2032
self._extract_and_insert_inventory_deltas(
2033
substream, src_serializer)
2034
elif substream_type == 'chk_bytes':
2035
# XXX: This doesn't support conversions, as it assumes the
2036
# conversion was done in the fetch code.
2037
self.target_repo.chk_bytes.insert_record_stream(substream)
2038
elif substream_type == 'revisions':
2039
# This may fallback to extract-and-insert more often than
2040
# required if the serializers are different only in terms of
2042
if src_serializer == to_serializer:
2043
self.target_repo.revisions.insert_record_stream(substream)
2045
self._extract_and_insert_revisions(substream,
2047
elif substream_type == 'signatures':
2048
self.target_repo.signatures.insert_record_stream(substream)
2050
raise AssertionError('kaboom! %s' % (substream_type,))
2051
# Done inserting data, and the missing_keys calculations will try to
2052
# read back from the inserted data, so flush the writes to the new pack
2053
# (if this is pack format).
2054
if new_pack is not None:
2055
new_pack._write_data('', flush=True)
2056
# Find all the new revisions (including ones from resume_tokens)
2057
missing_keys = self.target_repo.get_missing_parent_inventories(
2058
check_for_missing_texts=is_resume)
2060
for prefix, versioned_file in (
2061
('texts', self.target_repo.texts),
2062
('inventories', self.target_repo.inventories),
2063
('revisions', self.target_repo.revisions),
2064
('signatures', self.target_repo.signatures),
2065
('chk_bytes', self.target_repo.chk_bytes),
2067
if versioned_file is None:
2069
# TODO: key is often going to be a StaticTuple object
2070
# I don't believe we can define a method by which
2071
# (prefix,) + StaticTuple will work, though we could
2072
# define a StaticTuple.sq_concat that would allow you to
2073
# pass in either a tuple or a StaticTuple as the second
2074
# object, so instead we could have:
2075
# StaticTuple(prefix) + key here...
2076
missing_keys.update((prefix,) + key for key in
2077
versioned_file.get_missing_compression_parent_keys())
2078
except NotImplementedError:
2079
# cannot even attempt suspending, and missing would have failed
2080
# during stream insertion.
2081
missing_keys = set()
2084
def _extract_and_insert_inventory_deltas(self, substream, serializer):
2085
target_rich_root = self.target_repo._format.rich_root_data
2086
target_tree_refs = self.target_repo._format.supports_tree_reference
2087
for record in substream:
2088
# Insert the delta directly
2089
inventory_delta_bytes = record.get_bytes_as('fulltext')
2090
deserialiser = inventory_delta.InventoryDeltaDeserializer()
2092
parse_result = deserialiser.parse_text_bytes(
2093
inventory_delta_bytes)
2094
except inventory_delta.IncompatibleInventoryDelta, err:
2095
mutter("Incompatible delta: %s", err.msg)
2096
raise errors.IncompatibleRevision(self.target_repo._format)
2097
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
2098
revision_id = new_id
2099
parents = [key[0] for key in record.parents]
2100
self.target_repo.add_inventory_by_delta(
2101
basis_id, inv_delta, revision_id, parents)
2103
def _extract_and_insert_inventories(self, substream, serializer,
2105
"""Generate a new inventory versionedfile in target, converting data.
2107
The inventory is retrieved from the source, (deserializing it), and
2108
stored in the target (reserializing it in a different format).
2110
target_rich_root = self.target_repo._format.rich_root_data
2111
target_tree_refs = self.target_repo._format.supports_tree_reference
2112
for record in substream:
2113
# It's not a delta, so it must be a fulltext in the source
2114
# serializer's format.
2115
bytes = record.get_bytes_as('fulltext')
2116
revision_id = record.key[0]
2117
inv = serializer.read_inventory_from_string(bytes, revision_id)
2118
parents = [key[0] for key in record.parents]
2119
self.target_repo.add_inventory(revision_id, inv, parents)
2120
# No need to keep holding this full inv in memory when the rest of
2121
# the substream is likely to be all deltas.
2124
def _extract_and_insert_revisions(self, substream, serializer):
2125
for record in substream:
2126
bytes = record.get_bytes_as('fulltext')
2127
revision_id = record.key[0]
2128
rev = serializer.read_revision_from_string(bytes)
2129
if rev.revision_id != revision_id:
2130
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
2131
self.target_repo.add_revision(revision_id, rev)
2134
if self.target_repo._format._fetch_reconcile:
2135
self.target_repo.reconcile()
2138
class StreamSource(object):
2139
"""A source of a stream for fetching between repositories."""
2141
def __init__(self, from_repository, to_format):
2142
"""Create a StreamSource streaming from from_repository."""
2143
self.from_repository = from_repository
2144
self.to_format = to_format
2145
self._record_counter = RecordCounter()
2147
def delta_on_metadata(self):
2148
"""Return True if delta's are permitted on metadata streams.
2150
That is on revisions and signatures.
2152
src_serializer = self.from_repository._format._serializer
2153
target_serializer = self.to_format._serializer
2154
return (self.to_format._fetch_uses_deltas and
2155
src_serializer == target_serializer)
2157
def _fetch_revision_texts(self, revs):
2158
# fetch signatures first and then the revision texts
2159
# may need to be a InterRevisionStore call here.
2160
from_sf = self.from_repository.signatures
2161
# A missing signature is just skipped.
2162
keys = [(rev_id,) for rev_id in revs]
2163
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
2165
self.to_format._fetch_order,
2166
not self.to_format._fetch_uses_deltas))
2167
# If a revision has a delta, this is actually expanded inside the
2168
# insert_record_stream code now, which is an alternate fix for
2170
from_rf = self.from_repository.revisions
2171
revisions = from_rf.get_record_stream(
2173
self.to_format._fetch_order,
2174
not self.delta_on_metadata())
2175
return [('signatures', signatures), ('revisions', revisions)]
2177
def _generate_root_texts(self, revs):
2178
"""This will be called by get_stream between fetching weave texts and
2179
fetching the inventory weave.
2181
if self._rich_root_upgrade():
2182
return _mod_fetch.Inter1and2Helper(
2183
self.from_repository).generate_root_texts(revs)
2187
def get_stream(self, search):
2189
revs = search.get_keys()
2190
graph = self.from_repository.get_graph()
2191
revs = tsort.topo_sort(graph.get_parent_map(revs))
2192
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
2194
for knit_kind, file_id, revisions in data_to_fetch:
2195
if knit_kind != phase:
2197
# Make a new progress bar for this phase
2198
if knit_kind == "file":
2199
# Accumulate file texts
2200
text_keys.extend([(file_id, revision) for revision in
2202
elif knit_kind == "inventory":
2203
# Now copy the file texts.
2204
from_texts = self.from_repository.texts
2205
yield ('texts', from_texts.get_record_stream(
2206
text_keys, self.to_format._fetch_order,
2207
not self.to_format._fetch_uses_deltas))
2208
# Cause an error if a text occurs after we have done the
2211
# Before we process the inventory we generate the root
2212
# texts (if necessary) so that the inventories references
2214
for _ in self._generate_root_texts(revs):
2216
# we fetch only the referenced inventories because we do not
2217
# know for unselected inventories whether all their required
2218
# texts are present in the other repository - it could be
2220
for info in self._get_inventory_stream(revs):
2222
elif knit_kind == "signatures":
2223
# Nothing to do here; this will be taken care of when
2224
# _fetch_revision_texts happens.
2226
elif knit_kind == "revisions":
2227
for record in self._fetch_revision_texts(revs):
2230
raise AssertionError("Unknown knit kind %r" % knit_kind)
2232
def get_stream_for_missing_keys(self, missing_keys):
2233
# missing keys can only occur when we are byte copying and not
2234
# translating (because translation means we don't send
2235
# unreconstructable deltas ever).
2237
keys['texts'] = set()
2238
keys['revisions'] = set()
2239
keys['inventories'] = set()
2240
keys['chk_bytes'] = set()
2241
keys['signatures'] = set()
2242
for key in missing_keys:
2243
keys[key[0]].add(key[1:])
2244
if len(keys['revisions']):
2245
# If we allowed copying revisions at this point, we could end up
2246
# copying a revision without copying its required texts: a
2247
# violation of the requirements for repository integrity.
2248
raise AssertionError(
2249
'cannot copy revisions to fill in missing deltas %s' % (
2250
keys['revisions'],))
2251
for substream_kind, keys in keys.iteritems():
2252
vf = getattr(self.from_repository, substream_kind)
2253
if vf is None and keys:
2254
raise AssertionError(
2255
"cannot fill in keys for a versioned file we don't"
2256
" have: %s needs %s" % (substream_kind, keys))
2258
# No need to stream something we don't have
2260
if substream_kind == 'inventories':
2261
# Some missing keys are genuinely ghosts, filter those out.
2262
present = self.from_repository.inventories.get_parent_map(keys)
2263
revs = [key[0] for key in present]
2264
# Get the inventory stream more-or-less as we do for the
2265
# original stream; there's no reason to assume that records
2266
# direct from the source will be suitable for the sink. (Think
2267
# e.g. 2a -> 1.9-rich-root).
2268
for info in self._get_inventory_stream(revs, missing=True):
2272
# Ask for full texts always so that we don't need more round trips
2273
# after this stream.
2274
# Some of the missing keys are genuinely ghosts, so filter absent
2275
# records. The Sink is responsible for doing another check to
2276
# ensure that ghosts don't introduce missing data for future
2278
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
2279
self.to_format._fetch_order, True))
2280
yield substream_kind, stream
2282
def inventory_fetch_order(self):
2283
if self._rich_root_upgrade():
2284
return 'topological'
2286
return self.to_format._fetch_order
2288
def _rich_root_upgrade(self):
2289
return (not self.from_repository._format.rich_root_data and
2290
self.to_format.rich_root_data)
2292
def _get_inventory_stream(self, revision_ids, missing=False):
2293
from_format = self.from_repository._format
2294
if (from_format.supports_chks and self.to_format.supports_chks and
2295
from_format.network_name() == self.to_format.network_name()):
2296
raise AssertionError(
2297
"this case should be handled by GroupCHKStreamSource")
2298
elif 'forceinvdeltas' in debug.debug_flags:
2299
return self._get_convertable_inventory_stream(revision_ids,
2300
delta_versus_null=missing)
2301
elif from_format.network_name() == self.to_format.network_name():
2303
return self._get_simple_inventory_stream(revision_ids,
2305
elif (not from_format.supports_chks and not self.to_format.supports_chks
2306
and from_format._serializer == self.to_format._serializer):
2307
# Essentially the same format.
2308
return self._get_simple_inventory_stream(revision_ids,
2311
# Any time we switch serializations, we want to use an
2312
# inventory-delta based approach.
2313
return self._get_convertable_inventory_stream(revision_ids,
2314
delta_versus_null=missing)
2316
def _get_simple_inventory_stream(self, revision_ids, missing=False):
2317
# NB: This currently reopens the inventory weave in source;
2318
# using a single stream interface instead would avoid this.
2319
from_weave = self.from_repository.inventories
2321
delta_closure = True
2323
delta_closure = not self.delta_on_metadata()
2324
yield ('inventories', from_weave.get_record_stream(
2325
[(rev_id,) for rev_id in revision_ids],
2326
self.inventory_fetch_order(), delta_closure))
2328
def _get_convertable_inventory_stream(self, revision_ids,
2329
delta_versus_null=False):
2330
# The two formats are sufficiently different that there is no fast
2331
# path, so we need to send just inventorydeltas, which any
2332
# sufficiently modern client can insert into any repository.
2333
# The StreamSink code expects to be able to
2334
# convert on the target, so we need to put bytes-on-the-wire that can
2335
# be converted. That means inventory deltas (if the remote is <1.19,
2336
# RemoteStreamSink will fallback to VFS to insert the deltas).
2337
yield ('inventory-deltas',
2338
self._stream_invs_as_deltas(revision_ids,
2339
delta_versus_null=delta_versus_null))
2341
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
2342
"""Return a stream of inventory-deltas for the given rev ids.
2344
:param revision_ids: The list of inventories to transmit
2345
:param delta_versus_null: Don't try to find a minimal delta for this
2346
entry, instead compute the delta versus the NULL_REVISION. This
2347
effectively streams a complete inventory. Used for stuff like
2348
filling in missing parents, etc.
2350
from_repo = self.from_repository
2351
revision_keys = [(rev_id,) for rev_id in revision_ids]
2352
parent_map = from_repo.inventories.get_parent_map(revision_keys)
2353
# XXX: possibly repos could implement a more efficient iter_inv_deltas
2355
inventories = self.from_repository.iter_inventories(
2356
revision_ids, 'topological')
2357
format = from_repo._format
2358
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
2359
inventory_cache = lru_cache.LRUCache(50)
2360
null_inventory = from_repo.revision_tree(
2361
_mod_revision.NULL_REVISION).inventory
2362
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
2363
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
2364
# repo back into a non-rich-root repo ought to be allowed)
2365
serializer = inventory_delta.InventoryDeltaSerializer(
2366
versioned_root=format.rich_root_data,
2367
tree_references=format.supports_tree_reference)
2368
for inv in inventories:
2369
key = (inv.revision_id,)
2370
parent_keys = parent_map.get(key, ())
2372
if not delta_versus_null and parent_keys:
2373
# The caller did not ask for complete inventories and we have
2374
# some parents that we can delta against. Make a delta against
2375
# each parent so that we can find the smallest.
2376
parent_ids = [parent_key[0] for parent_key in parent_keys]
2377
for parent_id in parent_ids:
2378
if parent_id not in invs_sent_so_far:
2379
# We don't know that the remote side has this basis, so
2382
if parent_id == _mod_revision.NULL_REVISION:
2383
parent_inv = null_inventory
2385
parent_inv = inventory_cache.get(parent_id, None)
2386
if parent_inv is None:
2387
parent_inv = from_repo.get_inventory(parent_id)
2388
candidate_delta = inv._make_delta(parent_inv)
2389
if (delta is None or
2390
len(delta) > len(candidate_delta)):
2391
delta = candidate_delta
2392
basis_id = parent_id
2394
# Either none of the parents ended up being suitable, or we
2395
# were asked to delta against NULL
2396
basis_id = _mod_revision.NULL_REVISION
2397
delta = inv._make_delta(null_inventory)
2398
invs_sent_so_far.add(inv.revision_id)
2399
inventory_cache[inv.revision_id] = inv
2400
delta_serialized = ''.join(
2401
serializer.delta_to_lines(basis_id, key[-1], delta))
2402
yield versionedfile.FulltextContentFactory(
2403
key, parent_keys, None, delta_serialized)
2406
class _VersionedFileChecker(object):
2408
def __init__(self, repository, text_key_references=None, ancestors=None):
2409
self.repository = repository
2410
self.text_index = self.repository._generate_text_key_index(
2411
text_key_references=text_key_references, ancestors=ancestors)
2413
def calculate_file_version_parents(self, text_key):
2414
"""Calculate the correct parents for a file version according to
2417
parent_keys = self.text_index[text_key]
2418
if parent_keys == [_mod_revision.NULL_REVISION]:
2420
return tuple(parent_keys)
2422
def check_file_version_parents(self, texts, progress_bar=None):
2423
"""Check the parents stored in a versioned file are correct.
2425
It also detects file versions that are not referenced by their
2426
corresponding revision's inventory.
2428
:returns: A tuple of (wrong_parents, dangling_file_versions).
2429
wrong_parents is a dict mapping {revision_id: (stored_parents,
2430
correct_parents)} for each revision_id where the stored parents
2431
are not correct. dangling_file_versions is a set of (file_id,
2432
revision_id) tuples for versions that are present in this versioned
2433
file, but not used by the corresponding inventory.
2435
local_progress = None
2436
if progress_bar is None:
2437
local_progress = ui.ui_factory.nested_progress_bar()
2438
progress_bar = local_progress
2440
return self._check_file_version_parents(texts, progress_bar)
2443
local_progress.finished()
2445
def _check_file_version_parents(self, texts, progress_bar):
2446
"""See check_file_version_parents."""
2448
self.file_ids = set([file_id for file_id, _ in
2449
self.text_index.iterkeys()])
2450
# text keys is now grouped by file_id
2451
n_versions = len(self.text_index)
2452
progress_bar.update('loading text store', 0, n_versions)
2453
parent_map = self.repository.texts.get_parent_map(self.text_index)
2454
# On unlistable transports this could well be empty/error...
2455
text_keys = self.repository.texts.keys()
2456
unused_keys = frozenset(text_keys) - set(self.text_index)
2457
for num, key in enumerate(self.text_index.iterkeys()):
2458
progress_bar.update('checking text graph', num, n_versions)
2459
correct_parents = self.calculate_file_version_parents(key)
2461
knit_parents = parent_map[key]
2462
except errors.RevisionNotPresent:
2465
if correct_parents != knit_parents:
2466
wrong_parents[key] = (knit_parents, correct_parents)
2467
return wrong_parents, unused_keys
2470
class InterVersionedFileRepository(InterRepository):
2472
_walk_to_common_revisions_batch_size = 50
2475
def fetch(self, revision_id=None, find_ghosts=False,
2477
"""Fetch the content required to construct revision_id.
2479
The content is copied from self.source to self.target.
2481
:param revision_id: if None all content is copied, if NULL_REVISION no
2485
ui.ui_factory.warn_experimental_format_fetch(self)
2486
from bzrlib.fetch import RepoFetcher
2487
# See <https://launchpad.net/bugs/456077> asking for a warning here
2488
if self.source._format.network_name() != self.target._format.network_name():
2489
ui.ui_factory.show_user_warning('cross_format_fetch',
2490
from_format=self.source._format,
2491
to_format=self.target._format)
2492
f = RepoFetcher(to_repository=self.target,
2493
from_repository=self.source,
2494
last_revision=revision_id,
2495
fetch_spec=fetch_spec,
2496
find_ghosts=find_ghosts)
2498
def _walk_to_common_revisions(self, revision_ids, if_present_ids=None):
2499
"""Walk out from revision_ids in source to revisions target has.
2501
:param revision_ids: The start point for the search.
2502
:return: A set of revision ids.
2504
target_graph = self.target.get_graph()
2505
revision_ids = frozenset(revision_ids)
2507
all_wanted_revs = revision_ids.union(if_present_ids)
2509
all_wanted_revs = revision_ids
2510
missing_revs = set()
2511
source_graph = self.source.get_graph()
2512
# ensure we don't pay silly lookup costs.
2513
searcher = source_graph._make_breadth_first_searcher(all_wanted_revs)
2514
null_set = frozenset([_mod_revision.NULL_REVISION])
2515
searcher_exhausted = False
2519
# Iterate the searcher until we have enough next_revs
2520
while len(next_revs) < self._walk_to_common_revisions_batch_size:
2522
next_revs_part, ghosts_part = searcher.next_with_ghosts()
2523
next_revs.update(next_revs_part)
2524
ghosts.update(ghosts_part)
2525
except StopIteration:
2526
searcher_exhausted = True
2528
# If there are ghosts in the source graph, and the caller asked for
2529
# them, make sure that they are present in the target.
2530
# We don't care about other ghosts as we can't fetch them and
2531
# haven't been asked to.
2532
ghosts_to_check = set(revision_ids.intersection(ghosts))
2533
revs_to_get = set(next_revs).union(ghosts_to_check)
2535
have_revs = set(target_graph.get_parent_map(revs_to_get))
2536
# we always have NULL_REVISION present.
2537
have_revs = have_revs.union(null_set)
2538
# Check if the target is missing any ghosts we need.
2539
ghosts_to_check.difference_update(have_revs)
2541
# One of the caller's revision_ids is a ghost in both the
2542
# source and the target.
2543
raise errors.NoSuchRevision(
2544
self.source, ghosts_to_check.pop())
2545
missing_revs.update(next_revs - have_revs)
2546
# Because we may have walked past the original stop point, make
2547
# sure everything is stopped
2548
stop_revs = searcher.find_seen_ancestors(have_revs)
2549
searcher.stop_searching_any(stop_revs)
2550
if searcher_exhausted:
2552
return searcher.get_result()
2555
def search_missing_revision_ids(self,
2556
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
2557
find_ghosts=True, revision_ids=None, if_present_ids=None,
2559
"""Return the revision ids that source has that target does not.
2561
:param revision_id: only return revision ids included by this
2563
:param revision_ids: return revision ids included by these
2564
revision_ids. NoSuchRevision will be raised if any of these
2565
revisions are not present.
2566
:param if_present_ids: like revision_ids, but will not cause
2567
NoSuchRevision if any of these are absent, instead they will simply
2568
not be in the result. This is useful for e.g. finding revisions
2569
to fetch for tags, which may reference absent revisions.
2570
:param find_ghosts: If True find missing revisions in deep history
2571
rather than just finding the surface difference.
2572
:return: A bzrlib.graph.SearchResult.
2574
if symbol_versioning.deprecated_passed(revision_id):
2575
symbol_versioning.warn(
2576
'search_missing_revision_ids(revision_id=...) was '
2577
'deprecated in 2.4. Use revision_ids=[...] instead.',
2578
DeprecationWarning, stacklevel=2)
2579
if revision_ids is not None:
2580
raise AssertionError(
2581
'revision_ids is mutually exclusive with revision_id')
2582
if revision_id is not None:
2583
revision_ids = [revision_id]
2585
# stop searching at found target revisions.
2586
if not find_ghosts and (revision_ids is not None or if_present_ids is
2588
result = self._walk_to_common_revisions(revision_ids,
2589
if_present_ids=if_present_ids)
2592
result_set = result.get_keys()
2594
# generic, possibly worst case, slow code path.
2595
target_ids = set(self.target.all_revision_ids())
2596
source_ids = self._present_source_revisions_for(
2597
revision_ids, if_present_ids)
2598
result_set = set(source_ids).difference(target_ids)
2599
if limit is not None:
2600
topo_ordered = self.source.get_graph().iter_topo_order(result_set)
2601
result_set = set(itertools.islice(topo_ordered, limit))
2602
return self.source.revision_ids_to_search_result(result_set)
2604
def _present_source_revisions_for(self, revision_ids, if_present_ids=None):
2605
"""Returns set of all revisions in ancestry of revision_ids present in
2608
:param revision_ids: if None, all revisions in source are returned.
2609
:param if_present_ids: like revision_ids, but if any/all of these are
2610
absent no error is raised.
2612
if revision_ids is not None or if_present_ids is not None:
2613
# First, ensure all specified revisions exist. Callers expect
2614
# NoSuchRevision when they pass absent revision_ids here.
2615
if revision_ids is None:
2616
revision_ids = set()
2617
if if_present_ids is None:
2618
if_present_ids = set()
2619
revision_ids = set(revision_ids)
2620
if_present_ids = set(if_present_ids)
2621
all_wanted_ids = revision_ids.union(if_present_ids)
2622
graph = self.source.get_graph()
2623
present_revs = set(graph.get_parent_map(all_wanted_ids))
2624
missing = revision_ids.difference(present_revs)
2626
raise errors.NoSuchRevision(self.source, missing.pop())
2627
found_ids = all_wanted_ids.intersection(present_revs)
2628
source_ids = [rev_id for (rev_id, parents) in
2629
graph.iter_ancestry(found_ids)
2630
if rev_id != _mod_revision.NULL_REVISION
2631
and parents is not None]
2633
source_ids = self.source.all_revision_ids()
2634
return set(source_ids)
2637
def _get_repo_format_to_test(self):
2641
def is_compatible(cls, source, target):
2642
# The default implementation is compatible with everything
2643
return (source._format.supports_full_versioned_files and
2644
target._format.supports_full_versioned_files)
2647
class InterDifferingSerializer(InterVersionedFileRepository):
2650
def _get_repo_format_to_test(self):
2654
def is_compatible(source, target):
2655
if not source._format.supports_full_versioned_files:
2657
if not target._format.supports_full_versioned_files:
2659
# This is redundant with format.check_conversion_target(), however that
2660
# raises an exception, and we just want to say "False" as in we won't
2661
# support converting between these formats.
2662
if 'IDS_never' in debug.debug_flags:
2664
if source.supports_rich_root() and not target.supports_rich_root():
2666
if (source._format.supports_tree_reference
2667
and not target._format.supports_tree_reference):
2669
if target._fallback_repositories and target._format.supports_chks:
2670
# IDS doesn't know how to copy CHKs for the parent inventories it
2671
# adds to stacked repos.
2673
if 'IDS_always' in debug.debug_flags:
2675
# Only use this code path for local source and target. IDS does far
2676
# too much IO (both bandwidth and roundtrips) over a network.
2677
if not source.bzrdir.transport.base.startswith('file:///'):
2679
if not target.bzrdir.transport.base.startswith('file:///'):
2683
def _get_trees(self, revision_ids, cache):
2685
for rev_id in revision_ids:
2687
possible_trees.append((rev_id, cache[rev_id]))
2689
# Not cached, but inventory might be present anyway.
2691
tree = self.source.revision_tree(rev_id)
2692
except errors.NoSuchRevision:
2693
# Nope, parent is ghost.
2696
cache[rev_id] = tree
2697
possible_trees.append((rev_id, tree))
2698
return possible_trees
2700
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
2701
"""Get the best delta and base for this revision.
2703
:return: (basis_id, delta)
2706
# Generate deltas against each tree, to find the shortest.
2707
texts_possibly_new_in_tree = set()
2708
for basis_id, basis_tree in possible_trees:
2709
delta = tree.inventory._make_delta(basis_tree.inventory)
2710
for old_path, new_path, file_id, new_entry in delta:
2711
if new_path is None:
2712
# This file_id isn't present in the new rev, so we don't
2716
# Rich roots are handled elsewhere...
2718
kind = new_entry.kind
2719
if kind != 'directory' and kind != 'file':
2720
# No text record associated with this inventory entry.
2722
# This is a directory or file that has changed somehow.
2723
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
2724
deltas.append((len(delta), basis_id, delta))
2726
return deltas[0][1:]
2728
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
2729
"""Find all parent revisions that are absent, but for which the
2730
inventory is present, and copy those inventories.
2732
This is necessary to preserve correctness when the source is stacked
2733
without fallbacks configured. (Note that in cases like upgrade the
2734
source may be not have _fallback_repositories even though it is
2738
for parents in parent_map.values():
2739
parent_revs.update(parents)
2740
present_parents = self.source.get_parent_map(parent_revs)
2741
absent_parents = set(parent_revs).difference(present_parents)
2742
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
2743
(rev_id,) for rev_id in absent_parents)
2744
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
2745
for parent_tree in self.source.revision_trees(parent_inv_ids):
2746
current_revision_id = parent_tree.get_revision_id()
2747
parents_parents_keys = parent_invs_keys_for_stacking[
2748
(current_revision_id,)]
2749
parents_parents = [key[-1] for key in parents_parents_keys]
2750
basis_id = _mod_revision.NULL_REVISION
2751
basis_tree = self.source.revision_tree(basis_id)
2752
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
2753
self.target.add_inventory_by_delta(
2754
basis_id, delta, current_revision_id, parents_parents)
2755
cache[current_revision_id] = parent_tree
2757
def _fetch_batch(self, revision_ids, basis_id, cache):
2758
"""Fetch across a few revisions.
2760
:param revision_ids: The revisions to copy
2761
:param basis_id: The revision_id of a tree that must be in cache, used
2762
as a basis for delta when no other base is available
2763
:param cache: A cache of RevisionTrees that we can use.
2764
:return: The revision_id of the last converted tree. The RevisionTree
2765
for it will be in cache
2767
# Walk though all revisions; get inventory deltas, copy referenced
2768
# texts that delta references, insert the delta, revision and
2770
root_keys_to_create = set()
2773
pending_revisions = []
2774
parent_map = self.source.get_parent_map(revision_ids)
2775
self._fetch_parent_invs_for_stacking(parent_map, cache)
2776
self.source._safe_to_return_from_cache = True
2777
for tree in self.source.revision_trees(revision_ids):
2778
# Find a inventory delta for this revision.
2779
# Find text entries that need to be copied, too.
2780
current_revision_id = tree.get_revision_id()
2781
parent_ids = parent_map.get(current_revision_id, ())
2782
parent_trees = self._get_trees(parent_ids, cache)
2783
possible_trees = list(parent_trees)
2784
if len(possible_trees) == 0:
2785
# There either aren't any parents, or the parents are ghosts,
2786
# so just use the last converted tree.
2787
possible_trees.append((basis_id, cache[basis_id]))
2788
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
2790
revision = self.source.get_revision(current_revision_id)
2791
pending_deltas.append((basis_id, delta,
2792
current_revision_id, revision.parent_ids))
2793
if self._converting_to_rich_root:
2794
self._revision_id_to_root_id[current_revision_id] = \
2796
# Determine which texts are in present in this revision but not in
2797
# any of the available parents.
2798
texts_possibly_new_in_tree = set()
2799
for old_path, new_path, file_id, entry in delta:
2800
if new_path is None:
2801
# This file_id isn't present in the new rev
2805
if not self.target.supports_rich_root():
2806
# The target doesn't support rich root, so we don't
2809
if self._converting_to_rich_root:
2810
# This can't be copied normally, we have to insert
2812
root_keys_to_create.add((file_id, entry.revision))
2815
texts_possibly_new_in_tree.add((file_id, entry.revision))
2816
for basis_id, basis_tree in possible_trees:
2817
basis_inv = basis_tree.inventory
2818
for file_key in list(texts_possibly_new_in_tree):
2819
file_id, file_revision = file_key
2821
entry = basis_inv[file_id]
2822
except errors.NoSuchId:
2824
if entry.revision == file_revision:
2825
texts_possibly_new_in_tree.remove(file_key)
2826
text_keys.update(texts_possibly_new_in_tree)
2827
pending_revisions.append(revision)
2828
cache[current_revision_id] = tree
2829
basis_id = current_revision_id
2830
self.source._safe_to_return_from_cache = False
2832
from_texts = self.source.texts
2833
to_texts = self.target.texts
2834
if root_keys_to_create:
2835
root_stream = _mod_fetch._new_root_data_stream(
2836
root_keys_to_create, self._revision_id_to_root_id, parent_map,
2838
to_texts.insert_record_stream(root_stream)
2839
to_texts.insert_record_stream(from_texts.get_record_stream(
2840
text_keys, self.target._format._fetch_order,
2841
not self.target._format._fetch_uses_deltas))
2842
# insert inventory deltas
2843
for delta in pending_deltas:
2844
self.target.add_inventory_by_delta(*delta)
2845
if self.target._fallback_repositories:
2846
# Make sure this stacked repository has all the parent inventories
2847
# for the new revisions that we are about to insert. We do this
2848
# before adding the revisions so that no revision is added until
2849
# all the inventories it may depend on are added.
2850
# Note that this is overzealous, as we may have fetched these in an
2853
revision_ids = set()
2854
for revision in pending_revisions:
2855
revision_ids.add(revision.revision_id)
2856
parent_ids.update(revision.parent_ids)
2857
parent_ids.difference_update(revision_ids)
2858
parent_ids.discard(_mod_revision.NULL_REVISION)
2859
parent_map = self.source.get_parent_map(parent_ids)
2860
# we iterate over parent_map and not parent_ids because we don't
2861
# want to try copying any revision which is a ghost
2862
for parent_tree in self.source.revision_trees(parent_map):
2863
current_revision_id = parent_tree.get_revision_id()
2864
parents_parents = parent_map[current_revision_id]
2865
possible_trees = self._get_trees(parents_parents, cache)
2866
if len(possible_trees) == 0:
2867
# There either aren't any parents, or the parents are
2868
# ghosts, so just use the last converted tree.
2869
possible_trees.append((basis_id, cache[basis_id]))
2870
basis_id, delta = self._get_delta_for_revision(parent_tree,
2871
parents_parents, possible_trees)
2872
self.target.add_inventory_by_delta(
2873
basis_id, delta, current_revision_id, parents_parents)
2874
# insert signatures and revisions
2875
for revision in pending_revisions:
2877
signature = self.source.get_signature_text(
2878
revision.revision_id)
2879
self.target.add_signature_text(revision.revision_id,
2881
except errors.NoSuchRevision:
2883
self.target.add_revision(revision.revision_id, revision)
2886
def _fetch_all_revisions(self, revision_ids, pb):
2887
"""Fetch everything for the list of revisions.
2889
:param revision_ids: The list of revisions to fetch. Must be in
2891
:param pb: A ProgressTask
2894
basis_id, basis_tree = self._get_basis(revision_ids[0])
2896
cache = lru_cache.LRUCache(100)
2897
cache[basis_id] = basis_tree
2898
del basis_tree # We don't want to hang on to it here
2902
for offset in range(0, len(revision_ids), batch_size):
2903
self.target.start_write_group()
2905
pb.update('Transferring revisions', offset,
2907
batch = revision_ids[offset:offset+batch_size]
2908
basis_id = self._fetch_batch(batch, basis_id, cache)
2910
self.source._safe_to_return_from_cache = False
2911
self.target.abort_write_group()
2914
hint = self.target.commit_write_group()
2917
if hints and self.target._format.pack_compresses:
2918
self.target.pack(hint=hints)
2919
pb.update('Transferring revisions', len(revision_ids),
2923
def fetch(self, revision_id=None, find_ghosts=False,
2925
"""See InterRepository.fetch()."""
2926
if fetch_spec is not None:
2927
revision_ids = fetch_spec.get_keys()
2930
ui.ui_factory.warn_experimental_format_fetch(self)
2931
if (not self.source.supports_rich_root()
2932
and self.target.supports_rich_root()):
2933
self._converting_to_rich_root = True
2934
self._revision_id_to_root_id = {}
2936
self._converting_to_rich_root = False
2937
# See <https://launchpad.net/bugs/456077> asking for a warning here
2938
if self.source._format.network_name() != self.target._format.network_name():
2939
ui.ui_factory.show_user_warning('cross_format_fetch',
2940
from_format=self.source._format,
2941
to_format=self.target._format)
2942
if revision_ids is None:
2944
search_revision_ids = [revision_id]
2946
search_revision_ids = None
2947
revision_ids = self.target.search_missing_revision_ids(self.source,
2948
revision_ids=search_revision_ids,
2949
find_ghosts=find_ghosts).get_keys()
2950
if not revision_ids:
2952
revision_ids = tsort.topo_sort(
2953
self.source.get_graph().get_parent_map(revision_ids))
2954
if not revision_ids:
2956
# Walk though all revisions; get inventory deltas, copy referenced
2957
# texts that delta references, insert the delta, revision and
2959
pb = ui.ui_factory.nested_progress_bar()
2961
self._fetch_all_revisions(revision_ids, pb)
2964
return len(revision_ids), 0
2966
def _get_basis(self, first_revision_id):
2967
"""Get a revision and tree which exists in the target.
2969
This assumes that first_revision_id is selected for transmission
2970
because all other ancestors are already present. If we can't find an
2971
ancestor we fall back to NULL_REVISION since we know that is safe.
2973
:return: (basis_id, basis_tree)
2975
first_rev = self.source.get_revision(first_revision_id)
2977
basis_id = first_rev.parent_ids[0]
2978
# only valid as a basis if the target has it
2979
self.target.get_revision(basis_id)
2980
# Try to get a basis tree - if it's a ghost it will hit the
2981
# NoSuchRevision case.
2982
basis_tree = self.source.revision_tree(basis_id)
2983
except (IndexError, errors.NoSuchRevision):
2984
basis_id = _mod_revision.NULL_REVISION
2985
basis_tree = self.source.revision_tree(basis_id)
2986
return basis_id, basis_tree
2989
class InterSameDataRepository(InterVersionedFileRepository):
2990
"""Code for converting between repositories that represent the same data.
2992
Data format and model must match for this to work.
2996
def _get_repo_format_to_test(self):
2997
"""Repository format for testing with.
2999
InterSameData can pull from subtree to subtree and from non-subtree to
3000
non-subtree, so we test this with the richest repository format.
3002
from bzrlib.repofmt import knitrepo
3003
return knitrepo.RepositoryFormatKnit3()
3006
def is_compatible(source, target):
3008
InterRepository._same_model(source, target) and
3009
source._format.supports_full_versioned_files and
3010
target._format.supports_full_versioned_files)
3013
InterRepository.register_optimiser(InterVersionedFileRepository)
3014
InterRepository.register_optimiser(InterDifferingSerializer)
3015
InterRepository.register_optimiser(InterSameDataRepository)
3018
def install_revisions(repository, iterable, num_revisions=None, pb=None):
3019
"""Install all revision data into a repository.
3021
Accepts an iterable of revision, tree, signature tuples. The signature
3024
repository.start_write_group()
3026
inventory_cache = lru_cache.LRUCache(10)
3027
for n, (revision, revision_tree, signature) in enumerate(iterable):
3028
_install_revision(repository, revision, revision_tree, signature,
3031
pb.update('Transferring revisions', n + 1, num_revisions)
3033
repository.abort_write_group()
3036
repository.commit_write_group()
3039
def _install_revision(repository, rev, revision_tree, signature,
3041
"""Install all revision data into a repository."""
3042
present_parents = []
3044
for p_id in rev.parent_ids:
3045
if repository.has_revision(p_id):
3046
present_parents.append(p_id)
3047
parent_trees[p_id] = repository.revision_tree(p_id)
3049
parent_trees[p_id] = repository.revision_tree(
3050
_mod_revision.NULL_REVISION)
3052
inv = revision_tree.inventory
3053
entries = inv.iter_entries()
3054
# backwards compatibility hack: skip the root id.
3055
if not repository.supports_rich_root():
3056
path, root = entries.next()
3057
if root.revision != rev.revision_id:
3058
raise errors.IncompatibleRevision(repr(repository))
3060
for path, ie in entries:
3061
text_keys[(ie.file_id, ie.revision)] = ie
3062
text_parent_map = repository.texts.get_parent_map(text_keys)
3063
missing_texts = set(text_keys) - set(text_parent_map)
3064
# Add the texts that are not already present
3065
for text_key in missing_texts:
3066
ie = text_keys[text_key]
3068
# FIXME: TODO: The following loop overlaps/duplicates that done by
3069
# commit to determine parents. There is a latent/real bug here where
3070
# the parents inserted are not those commit would do - in particular
3071
# they are not filtered by heads(). RBC, AB
3072
for revision, tree in parent_trees.iteritems():
3073
if ie.file_id not in tree:
3075
parent_id = tree.get_file_revision(ie.file_id)
3076
if parent_id in text_parents:
3078
text_parents.append((ie.file_id, parent_id))
3079
lines = revision_tree.get_file(ie.file_id).readlines()
3080
repository.texts.add_lines(text_key, text_parents, lines)
3082
# install the inventory
3083
if repository._format._commit_inv_deltas and len(rev.parent_ids):
3084
# Cache this inventory
3085
inventory_cache[rev.revision_id] = inv
3087
basis_inv = inventory_cache[rev.parent_ids[0]]
3089
repository.add_inventory(rev.revision_id, inv, present_parents)
3091
delta = inv._make_delta(basis_inv)
3092
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
3093
rev.revision_id, present_parents)
3095
repository.add_inventory(rev.revision_id, inv, present_parents)
3096
except errors.RevisionAlreadyPresent:
3098
if signature is not None:
3099
repository.add_signature_text(rev.revision_id, signature)
3100
repository.add_revision(rev.revision_id, rev, inv)
3103
def install_revision(repository, rev, revision_tree):
3104
"""Install all revision data into a repository."""
3105
install_revisions(repository, [(rev, revision_tree, None)])
added
removed
bzrlib/vf_repository.py
