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
81
supports_versioned_directories = True
83
# Should commit add an inventory, or an inventory delta to the repository.
84
_commit_inv_deltas = True
85
# What order should fetch operations request streams in?
86
# The default is unordered as that is the cheapest for an origin to
88
_fetch_order = 'unordered'
89
# Does this repository format use deltas that can be fetched as-deltas ?
90
# (E.g. knits, where the knit deltas can be transplanted intact.
91
# We default to False, which will ensure that enough data to get
92
# a full text out of any fetch stream will be grabbed.
93
_fetch_uses_deltas = False
96
class VersionedFileCommitBuilder(CommitBuilder):
97
"""Commit builder implementation for versioned files based repositories.
100
# this commit builder supports the record_entry_contents interface
101
supports_record_entry_contents = True
103
# the default CommitBuilder does not manage trees whose root is versioned.
104
_versioned_root = False
106
def __init__(self, repository, parents, config, timestamp=None,
107
timezone=None, committer=None, revprops=None,
108
revision_id=None, lossy=False):
109
super(VersionedFileCommitBuilder, self).__init__(repository,
110
parents, config, timestamp, timezone, committer, revprops,
113
basis_id = self.parents[0]
115
basis_id = _mod_revision.NULL_REVISION
116
self.basis_delta_revision = basis_id
117
self.new_inventory = Inventory(None)
118
self._basis_delta = []
119
self.__heads = graph.HeadsCache(repository.get_graph()).heads
120
# memo'd check for no-op commits.
121
self._any_changes = False
122
# API compatibility, older code that used CommitBuilder did not call
123
# .record_delete(), which means the delta that is computed would not be
124
# valid. Callers that will call record_delete() should call
125
# .will_record_deletes() to indicate that.
126
self._recording_deletes = False
128
def will_record_deletes(self):
129
"""Tell the commit builder that deletes are being notified.
131
This enables the accumulation of an inventory delta; for the resulting
132
commit to be valid, deletes against the basis MUST be recorded via
133
builder.record_delete().
135
self._recording_deletes = True
137
def any_changes(self):
138
"""Return True if any entries were changed.
140
This includes merge-only changes. It is the core for the --unchanged
143
:return: True if any changes have occured.
145
return self._any_changes
147
def _ensure_fallback_inventories(self):
148
"""Ensure that appropriate inventories are available.
150
This only applies to repositories that are stacked, and is about
151
enusring the stacking invariants. Namely, that for any revision that is
152
present, we either have all of the file content, or we have the parent
153
inventory and the delta file content.
155
if not self.repository._fallback_repositories:
157
if not self.repository._format.supports_chks:
158
raise errors.BzrError("Cannot commit directly to a stacked branch"
159
" in pre-2a formats. See "
160
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
161
# This is a stacked repo, we need to make sure we have the parent
162
# inventories for the parents.
163
parent_keys = [(p,) for p in self.parents]
164
parent_map = self.repository.inventories._index.get_parent_map(parent_keys)
165
missing_parent_keys = set([pk for pk in parent_keys
166
if pk not in parent_map])
167
fallback_repos = list(reversed(self.repository._fallback_repositories))
168
missing_keys = [('inventories', pk[0])
169
for pk in missing_parent_keys]
171
while missing_keys and fallback_repos:
172
fallback_repo = fallback_repos.pop()
173
source = fallback_repo._get_source(self.repository._format)
174
sink = self.repository._get_sink()
175
stream = source.get_stream_for_missing_keys(missing_keys)
176
missing_keys = sink.insert_stream_without_locking(stream,
177
self.repository._format)
179
raise errors.BzrError('Unable to fill in parent inventories for a'
182
def commit(self, message):
183
"""Make the actual commit.
185
:return: The revision id of the recorded revision.
187
self._validate_unicode_text(message, 'commit message')
188
rev = _mod_revision.Revision(
189
timestamp=self._timestamp,
190
timezone=self._timezone,
191
committer=self._committer,
193
inventory_sha1=self.inv_sha1,
194
revision_id=self._new_revision_id,
195
properties=self._revprops)
196
rev.parent_ids = self.parents
197
self.repository.add_revision(self._new_revision_id, rev,
198
self.new_inventory, self._config)
199
self._ensure_fallback_inventories()
200
self.repository.commit_write_group()
201
return self._new_revision_id
204
"""Abort the commit that is being built.
206
self.repository.abort_write_group()
208
def revision_tree(self):
209
"""Return the tree that was just committed.
211
After calling commit() this can be called to get a
212
RevisionTree representing the newly committed tree. This is
213
preferred to calling Repository.revision_tree() because that may
214
require deserializing the inventory, while we already have a copy in
217
if self.new_inventory is None:
218
self.new_inventory = self.repository.get_inventory(
219
self._new_revision_id)
220
return InventoryRevisionTree(self.repository, self.new_inventory,
221
self._new_revision_id)
223
def finish_inventory(self):
224
"""Tell the builder that the inventory is finished.
226
:return: The inventory id in the repository, which can be used with
227
repository.get_inventory.
229
if self.new_inventory is None:
230
# an inventory delta was accumulated without creating a new
232
basis_id = self.basis_delta_revision
233
# We ignore the 'inventory' returned by add_inventory_by_delta
234
# because self.new_inventory is used to hint to the rest of the
235
# system what code path was taken
236
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
237
basis_id, self._basis_delta, self._new_revision_id,
240
if self.new_inventory.root is None:
241
raise AssertionError('Root entry should be supplied to'
242
' record_entry_contents, as of bzr 0.10.')
243
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
244
self.new_inventory.revision_id = self._new_revision_id
245
self.inv_sha1 = self.repository.add_inventory(
246
self._new_revision_id,
250
return self._new_revision_id
252
def _check_root(self, ie, parent_invs, tree):
253
"""Helper for record_entry_contents.
255
:param ie: An entry being added.
256
:param parent_invs: The inventories of the parent revisions of the
258
:param tree: The tree that is being committed.
260
# In this revision format, root entries have no knit or weave When
261
# serializing out to disk and back in root.revision is always
263
ie.revision = self._new_revision_id
265
def _require_root_change(self, tree):
266
"""Enforce an appropriate root object change.
268
This is called once when record_iter_changes is called, if and only if
269
the root was not in the delta calculated by record_iter_changes.
271
:param tree: The tree which is being committed.
273
if len(self.parents) == 0:
274
raise errors.RootMissing()
275
entry = entry_factory['directory'](tree.path2id(''), '',
277
entry.revision = self._new_revision_id
278
self._basis_delta.append(('', '', entry.file_id, entry))
280
def _get_delta(self, ie, basis_inv, path):
281
"""Get a delta against the basis inventory for ie."""
282
if not basis_inv.has_id(ie.file_id):
284
result = (None, path, ie.file_id, ie)
285
self._basis_delta.append(result)
287
elif ie != basis_inv[ie.file_id]:
289
# TODO: avoid tis id2path call.
290
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
291
self._basis_delta.append(result)
297
def _heads(self, file_id, revision_ids):
298
"""Calculate the graph heads for revision_ids in the graph of file_id.
300
This can use either a per-file graph or a global revision graph as we
301
have an identity relationship between the two graphs.
303
return self.__heads(revision_ids)
305
def get_basis_delta(self):
306
"""Return the complete inventory delta versus the basis inventory.
308
This has been built up with the calls to record_delete and
309
record_entry_contents. The client must have already called
310
will_record_deletes() to indicate that they will be generating a
313
:return: An inventory delta, suitable for use with apply_delta, or
314
Repository.add_inventory_by_delta, etc.
316
if not self._recording_deletes:
317
raise AssertionError("recording deletes not activated.")
318
return self._basis_delta
320
def record_delete(self, path, file_id):
321
"""Record that a delete occured against a basis tree.
323
This is an optional API - when used it adds items to the basis_delta
324
being accumulated by the commit builder. It cannot be called unless the
325
method will_record_deletes() has been called to inform the builder that
326
a delta is being supplied.
328
:param path: The path of the thing deleted.
329
:param file_id: The file id that was deleted.
331
if not self._recording_deletes:
332
raise AssertionError("recording deletes not activated.")
333
delta = (path, None, file_id, None)
334
self._basis_delta.append(delta)
335
self._any_changes = True
338
def record_entry_contents(self, ie, parent_invs, path, tree,
340
"""Record the content of ie from tree into the commit if needed.
342
Side effect: sets ie.revision when unchanged
344
:param ie: An inventory entry present in the commit.
345
:param parent_invs: The inventories of the parent revisions of the
347
:param path: The path the entry is at in the tree.
348
:param tree: The tree which contains this entry and should be used to
350
:param content_summary: Summary data from the tree about the paths
351
content - stat, length, exec, sha/link target. This is only
352
accessed when the entry has a revision of None - that is when it is
353
a candidate to commit.
354
:return: A tuple (change_delta, version_recorded, fs_hash).
355
change_delta is an inventory_delta change for this entry against
356
the basis tree of the commit, or None if no change occured against
358
version_recorded is True if a new version of the entry has been
359
recorded. For instance, committing a merge where a file was only
360
changed on the other side will return (delta, False).
361
fs_hash is either None, or the hash details for the path (currently
362
a tuple of the contents sha1 and the statvalue returned by
363
tree.get_file_with_stat()).
365
if self.new_inventory.root is None:
366
if ie.parent_id is not None:
367
raise errors.RootMissing()
368
self._check_root(ie, parent_invs, tree)
369
if ie.revision is None:
370
kind = content_summary[0]
372
# ie is carried over from a prior commit
374
# XXX: repository specific check for nested tree support goes here - if
375
# the repo doesn't want nested trees we skip it ?
376
if (kind == 'tree-reference' and
377
not self.repository._format.supports_tree_reference):
378
# mismatch between commit builder logic and repository:
379
# this needs the entry creation pushed down into the builder.
380
raise NotImplementedError('Missing repository subtree support.')
381
self.new_inventory.add(ie)
383
# TODO: slow, take it out of the inner loop.
385
basis_inv = parent_invs[0]
387
basis_inv = Inventory(root_id=None)
389
# ie.revision is always None if the InventoryEntry is considered
390
# for committing. We may record the previous parents revision if the
391
# content is actually unchanged against a sole head.
392
if ie.revision is not None:
393
if not self._versioned_root and path == '':
394
# repositories that do not version the root set the root's
395
# revision to the new commit even when no change occurs (more
396
# specifically, they do not record a revision on the root; and
397
# the rev id is assigned to the root during deserialisation -
398
# this masks when a change may have occurred against the basis.
399
# To match this we always issue a delta, because the revision
400
# of the root will always be changing.
401
if basis_inv.has_id(ie.file_id):
402
delta = (basis_inv.id2path(ie.file_id), path,
406
delta = (None, path, ie.file_id, ie)
407
self._basis_delta.append(delta)
408
return delta, False, None
410
# we don't need to commit this, because the caller already
411
# determined that an existing revision of this file is
412
# appropriate. If it's not being considered for committing then
413
# it and all its parents to the root must be unaltered so
414
# no-change against the basis.
415
if ie.revision == self._new_revision_id:
416
raise AssertionError("Impossible situation, a skipped "
417
"inventory entry (%r) claims to be modified in this "
418
"commit (%r).", (ie, self._new_revision_id))
419
return None, False, None
420
# XXX: Friction: parent_candidates should return a list not a dict
421
# so that we don't have to walk the inventories again.
422
parent_candiate_entries = ie.parent_candidates(parent_invs)
423
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
425
for inv in parent_invs:
426
if inv.has_id(ie.file_id):
427
old_rev = inv[ie.file_id].revision
428
if old_rev in head_set:
429
heads.append(inv[ie.file_id].revision)
430
head_set.remove(inv[ie.file_id].revision)
433
# now we check to see if we need to write a new record to the
435
# We write a new entry unless there is one head to the ancestors, and
436
# the kind-derived content is unchanged.
438
# Cheapest check first: no ancestors, or more the one head in the
439
# ancestors, we write a new node.
443
# There is a single head, look it up for comparison
444
parent_entry = parent_candiate_entries[heads[0]]
445
# if the non-content specific data has changed, we'll be writing a
447
if (parent_entry.parent_id != ie.parent_id or
448
parent_entry.name != ie.name):
450
# now we need to do content specific checks:
452
# if the kind changed the content obviously has
453
if kind != parent_entry.kind:
455
# Stat cache fingerprint feedback for the caller - None as we usually
456
# don't generate one.
459
if content_summary[2] is None:
460
raise ValueError("Files must not have executable = None")
462
# We can't trust a check of the file length because of content
464
if (# if the exec bit has changed we have to store:
465
parent_entry.executable != content_summary[2]):
467
elif parent_entry.text_sha1 == content_summary[3]:
468
# all meta and content is unchanged (using a hash cache
469
# hit to check the sha)
470
ie.revision = parent_entry.revision
471
ie.text_size = parent_entry.text_size
472
ie.text_sha1 = parent_entry.text_sha1
473
ie.executable = parent_entry.executable
474
return self._get_delta(ie, basis_inv, path), False, None
476
# Either there is only a hash change(no hash cache entry,
477
# or same size content change), or there is no change on
479
# Provide the parent's hash to the store layer, so that the
480
# content is unchanged we will not store a new node.
481
nostore_sha = parent_entry.text_sha1
483
# We want to record a new node regardless of the presence or
484
# absence of a content change in the file.
486
ie.executable = content_summary[2]
487
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
489
text = file_obj.read()
493
ie.text_sha1, ie.text_size = self._add_text_to_weave(
494
ie.file_id, text, heads, nostore_sha)
495
# Let the caller know we generated a stat fingerprint.
496
fingerprint = (ie.text_sha1, stat_value)
497
except errors.ExistingContent:
498
# Turns out that the file content was unchanged, and we were
499
# only going to store a new node if it was changed. Carry over
501
ie.revision = parent_entry.revision
502
ie.text_size = parent_entry.text_size
503
ie.text_sha1 = parent_entry.text_sha1
504
ie.executable = parent_entry.executable
505
return self._get_delta(ie, basis_inv, path), False, None
506
elif kind == 'directory':
508
# all data is meta here, nothing specific to directory, so
510
ie.revision = parent_entry.revision
511
return self._get_delta(ie, basis_inv, path), False, None
512
self._add_text_to_weave(ie.file_id, '', heads, None)
513
elif kind == 'symlink':
514
current_link_target = content_summary[3]
516
# symlink target is not generic metadata, check if it has
518
if current_link_target != parent_entry.symlink_target:
521
# unchanged, carry over.
522
ie.revision = parent_entry.revision
523
ie.symlink_target = parent_entry.symlink_target
524
return self._get_delta(ie, basis_inv, path), False, None
525
ie.symlink_target = current_link_target
526
self._add_text_to_weave(ie.file_id, '', heads, None)
527
elif kind == 'tree-reference':
529
if content_summary[3] != parent_entry.reference_revision:
532
# unchanged, carry over.
533
ie.reference_revision = parent_entry.reference_revision
534
ie.revision = parent_entry.revision
535
return self._get_delta(ie, basis_inv, path), False, None
536
ie.reference_revision = content_summary[3]
537
if ie.reference_revision is None:
538
raise AssertionError("invalid content_summary for nested tree: %r"
539
% (content_summary,))
540
self._add_text_to_weave(ie.file_id, '', heads, None)
542
raise NotImplementedError('unknown kind')
543
ie.revision = self._new_revision_id
544
# The initial commit adds a root directory, but this in itself is not
545
# a worthwhile commit.
546
if (self.basis_delta_revision != _mod_revision.NULL_REVISION or
548
self._any_changes = True
549
return self._get_delta(ie, basis_inv, path), True, fingerprint
551
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
552
_entry_factory=entry_factory):
553
"""Record a new tree via iter_changes.
555
:param tree: The tree to obtain text contents from for changed objects.
556
:param basis_revision_id: The revision id of the tree the iter_changes
557
has been generated against. Currently assumed to be the same
558
as self.parents[0] - if it is not, errors may occur.
559
:param iter_changes: An iter_changes iterator with the changes to apply
560
to basis_revision_id. The iterator must not include any items with
561
a current kind of None - missing items must be either filtered out
562
or errored-on beefore record_iter_changes sees the item.
563
:param _entry_factory: Private method to bind entry_factory locally for
565
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
568
# Create an inventory delta based on deltas between all the parents and
569
# deltas between all the parent inventories. We use inventory delta's
570
# between the inventory objects because iter_changes masks
571
# last-changed-field only changes.
573
# file_id -> change map, change is fileid, paths, changed, versioneds,
574
# parents, names, kinds, executables
576
# {file_id -> revision_id -> inventory entry, for entries in parent
577
# trees that are not parents[0]
581
revtrees = list(self.repository.revision_trees(self.parents))
582
except errors.NoSuchRevision:
583
# one or more ghosts, slow path.
585
for revision_id in self.parents:
587
revtrees.append(self.repository.revision_tree(revision_id))
588
except errors.NoSuchRevision:
590
basis_revision_id = _mod_revision.NULL_REVISION
592
revtrees.append(self.repository.revision_tree(
593
_mod_revision.NULL_REVISION))
594
# The basis inventory from a repository
596
basis_inv = revtrees[0].inventory
598
basis_inv = self.repository.revision_tree(
599
_mod_revision.NULL_REVISION).inventory
600
if len(self.parents) > 0:
601
if basis_revision_id != self.parents[0] and not ghost_basis:
603
"arbitrary basis parents not yet supported with merges")
604
for revtree in revtrees[1:]:
605
for change in revtree.inventory._make_delta(basis_inv):
606
if change[1] is None:
607
# Not present in this parent.
609
if change[2] not in merged_ids:
610
if change[0] is not None:
611
basis_entry = basis_inv[change[2]]
612
merged_ids[change[2]] = [
614
basis_entry.revision,
617
parent_entries[change[2]] = {
619
basis_entry.revision:basis_entry,
621
change[3].revision:change[3],
624
merged_ids[change[2]] = [change[3].revision]
625
parent_entries[change[2]] = {change[3].revision:change[3]}
627
merged_ids[change[2]].append(change[3].revision)
628
parent_entries[change[2]][change[3].revision] = change[3]
631
# Setup the changes from the tree:
632
# changes maps file_id -> (change, [parent revision_ids])
634
for change in iter_changes:
635
# This probably looks up in basis_inv way to much.
636
if change[1][0] is not None:
637
head_candidate = [basis_inv[change[0]].revision]
640
changes[change[0]] = change, merged_ids.get(change[0],
642
unchanged_merged = set(merged_ids) - set(changes)
643
# Extend the changes dict with synthetic changes to record merges of
645
for file_id in unchanged_merged:
646
# Record a merged version of these items that did not change vs the
647
# basis. This can be either identical parallel changes, or a revert
648
# of a specific file after a merge. The recorded content will be
649
# that of the current tree (which is the same as the basis), but
650
# the per-file graph will reflect a merge.
651
# NB:XXX: We are reconstructing path information we had, this
652
# should be preserved instead.
653
# inv delta change: (file_id, (path_in_source, path_in_target),
654
# changed_content, versioned, parent, name, kind,
657
basis_entry = basis_inv[file_id]
658
except errors.NoSuchId:
659
# a change from basis->some_parents but file_id isn't in basis
660
# so was new in the merge, which means it must have changed
661
# from basis -> current, and as it hasn't the add was reverted
662
# by the user. So we discard this change.
666
(basis_inv.id2path(file_id), tree.id2path(file_id)),
668
(basis_entry.parent_id, basis_entry.parent_id),
669
(basis_entry.name, basis_entry.name),
670
(basis_entry.kind, basis_entry.kind),
671
(basis_entry.executable, basis_entry.executable))
672
changes[file_id] = (change, merged_ids[file_id])
673
# changes contains tuples with the change and a set of inventory
674
# candidates for the file.
676
# old_path, new_path, file_id, new_inventory_entry
677
seen_root = False # Is the root in the basis delta?
678
inv_delta = self._basis_delta
679
modified_rev = self._new_revision_id
680
for change, head_candidates in changes.values():
681
if change[3][1]: # versioned in target.
682
# Several things may be happening here:
683
# We may have a fork in the per-file graph
684
# - record a change with the content from tree
685
# We may have a change against < all trees
686
# - carry over the tree that hasn't changed
687
# We may have a change against all trees
688
# - record the change with the content from tree
691
entry = _entry_factory[kind](file_id, change[5][1],
693
head_set = self._heads(change[0], set(head_candidates))
696
for head_candidate in head_candidates:
697
if head_candidate in head_set:
698
heads.append(head_candidate)
699
head_set.remove(head_candidate)
702
# Could be a carry-over situation:
703
parent_entry_revs = parent_entries.get(file_id, None)
704
if parent_entry_revs:
705
parent_entry = parent_entry_revs.get(heads[0], None)
708
if parent_entry is None:
709
# The parent iter_changes was called against is the one
710
# that is the per-file head, so any change is relevant
711
# iter_changes is valid.
712
carry_over_possible = False
714
# could be a carry over situation
715
# A change against the basis may just indicate a merge,
716
# we need to check the content against the source of the
717
# merge to determine if it was changed after the merge
719
if (parent_entry.kind != entry.kind or
720
parent_entry.parent_id != entry.parent_id or
721
parent_entry.name != entry.name):
722
# Metadata common to all entries has changed
723
# against per-file parent
724
carry_over_possible = False
726
carry_over_possible = True
727
# per-type checks for changes against the parent_entry
730
# Cannot be a carry-over situation
731
carry_over_possible = False
732
# Populate the entry in the delta
734
# XXX: There is still a small race here: If someone reverts the content of a file
735
# after iter_changes examines and decides it has changed,
736
# we will unconditionally record a new version even if some
737
# other process reverts it while commit is running (with
738
# the revert happening after iter_changes did its
741
entry.executable = True
743
entry.executable = False
744
if (carry_over_possible and
745
parent_entry.executable == entry.executable):
746
# Check the file length, content hash after reading
748
nostore_sha = parent_entry.text_sha1
751
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
753
text = file_obj.read()
757
entry.text_sha1, entry.text_size = self._add_text_to_weave(
758
file_id, text, heads, nostore_sha)
759
yield file_id, change[1][1], (entry.text_sha1, stat_value)
760
except errors.ExistingContent:
761
# No content change against a carry_over parent
762
# Perhaps this should also yield a fs hash update?
764
entry.text_size = parent_entry.text_size
765
entry.text_sha1 = parent_entry.text_sha1
766
elif kind == 'symlink':
768
entry.symlink_target = tree.get_symlink_target(file_id)
769
if (carry_over_possible and
770
parent_entry.symlink_target == entry.symlink_target):
773
self._add_text_to_weave(change[0], '', heads, None)
774
elif kind == 'directory':
775
if carry_over_possible:
778
# Nothing to set on the entry.
779
# XXX: split into the Root and nonRoot versions.
780
if change[1][1] != '' or self.repository.supports_rich_root():
781
self._add_text_to_weave(change[0], '', heads, None)
782
elif kind == 'tree-reference':
783
if not self.repository._format.supports_tree_reference:
784
# This isn't quite sane as an error, but we shouldn't
785
# ever see this code path in practice: tree's don't
786
# permit references when the repo doesn't support tree
788
raise errors.UnsupportedOperation(tree.add_reference,
790
reference_revision = tree.get_reference_revision(change[0])
791
entry.reference_revision = reference_revision
792
if (carry_over_possible and
793
parent_entry.reference_revision == reference_revision):
796
self._add_text_to_weave(change[0], '', heads, None)
798
raise AssertionError('unknown kind %r' % kind)
800
entry.revision = modified_rev
802
entry.revision = parent_entry.revision
805
new_path = change[1][1]
806
inv_delta.append((change[1][0], new_path, change[0], entry))
809
self.new_inventory = None
810
# The initial commit adds a root directory, but this in itself is not
811
# a worthwhile commit.
812
if ((len(inv_delta) > 0 and basis_revision_id != _mod_revision.NULL_REVISION) or
813
(len(inv_delta) > 1 and basis_revision_id == _mod_revision.NULL_REVISION)):
814
# This should perhaps be guarded by a check that the basis we
815
# commit against is the basis for the commit and if not do a delta
817
self._any_changes = True
819
# housekeeping root entry changes do not affect no-change commits.
820
self._require_root_change(tree)
821
self.basis_delta_revision = basis_revision_id
823
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
824
parent_keys = tuple([(file_id, parent) for parent in parents])
825
return self.repository.texts._add_text(
826
(file_id, self._new_revision_id), parent_keys, new_text,
827
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
830
class VersionedFileRootCommitBuilder(VersionedFileCommitBuilder):
831
"""This commitbuilder actually records the root id"""
833
# the root entry gets versioned properly by this builder.
834
_versioned_root = True
836
def _check_root(self, ie, parent_invs, tree):
837
"""Helper for record_entry_contents.
839
:param ie: An entry being added.
840
:param parent_invs: The inventories of the parent revisions of the
842
:param tree: The tree that is being committed.
845
def _require_root_change(self, tree):
846
"""Enforce an appropriate root object change.
848
This is called once when record_iter_changes is called, if and only if
849
the root was not in the delta calculated by record_iter_changes.
851
:param tree: The tree which is being committed.
853
# versioned roots do not change unless the tree found a change.
856
class VersionedFileRepository(Repository):
857
"""Repository holding history for one or more branches.
859
The repository holds and retrieves historical information including
860
revisions and file history. It's normally accessed only by the Branch,
861
which views a particular line of development through that history.
863
The Repository builds on top of some byte storage facilies (the revisions,
864
signatures, inventories, texts and chk_bytes attributes) and a Transport,
865
which respectively provide byte storage and a means to access the (possibly
868
The byte storage facilities are addressed via tuples, which we refer to
869
as 'keys' throughout the code base. Revision_keys, inventory_keys and
870
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
871
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
872
byte string made up of a hash identifier and a hash value.
873
We use this interface because it allows low friction with the underlying
874
code that implements disk indices, network encoding and other parts of
877
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
878
the serialised revisions for the repository. This can be used to obtain
879
revision graph information or to access raw serialised revisions.
880
The result of trying to insert data into the repository via this store
881
is undefined: it should be considered read-only except for implementors
883
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
884
the serialised signatures for the repository. This can be used to
885
obtain access to raw serialised signatures. The result of trying to
886
insert data into the repository via this store is undefined: it should
887
be considered read-only except for implementors of repositories.
888
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
889
the serialised inventories for the repository. This can be used to
890
obtain unserialised inventories. The result of trying to insert data
891
into the repository via this store is undefined: it should be
892
considered read-only except for implementors of repositories.
893
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
894
texts of files and directories for the repository. This can be used to
895
obtain file texts or file graphs. Note that Repository.iter_file_bytes
896
is usually a better interface for accessing file texts.
897
The result of trying to insert data into the repository via this store
898
is undefined: it should be considered read-only except for implementors
900
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
901
any data the repository chooses to store or have indexed by its hash.
902
The result of trying to insert data into the repository via this store
903
is undefined: it should be considered read-only except for implementors
905
:ivar _transport: Transport for file access to repository, typically
906
pointing to .bzr/repository.
909
# What class to use for a CommitBuilder. Often it's simpler to change this
910
# in a Repository class subclass rather than to override
911
# get_commit_builder.
912
_commit_builder_class = VersionedFileCommitBuilder
914
def add_fallback_repository(self, repository):
915
"""Add a repository to use for looking up data not held locally.
917
:param repository: A repository.
919
if not self._format.supports_external_lookups:
920
raise errors.UnstackableRepositoryFormat(self._format, self.base)
922
# This repository will call fallback.unlock() when we transition to
923
# the unlocked state, so we make sure to increment the lock count
924
repository.lock_read()
925
self._check_fallback_repository(repository)
926
self._fallback_repositories.append(repository)
927
self.texts.add_fallback_versioned_files(repository.texts)
928
self.inventories.add_fallback_versioned_files(repository.inventories)
929
self.revisions.add_fallback_versioned_files(repository.revisions)
930
self.signatures.add_fallback_versioned_files(repository.signatures)
931
if self.chk_bytes is not None:
932
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
934
@only_raises(errors.LockNotHeld, errors.LockBroken)
936
super(VersionedFileRepository, self).unlock()
937
if self.control_files._lock_count == 0:
938
self._inventory_entry_cache.clear()
940
def add_inventory(self, revision_id, inv, parents):
941
"""Add the inventory inv to the repository as revision_id.
943
:param parents: The revision ids of the parents that revision_id
944
is known to have and are in the repository already.
946
:returns: The validator(which is a sha1 digest, though what is sha'd is
947
repository format specific) of the serialized inventory.
949
if not self.is_in_write_group():
950
raise AssertionError("%r not in write group" % (self,))
951
_mod_revision.check_not_reserved_id(revision_id)
952
if not (inv.revision_id is None or inv.revision_id == revision_id):
953
raise AssertionError(
954
"Mismatch between inventory revision"
955
" id and insertion revid (%r, %r)"
956
% (inv.revision_id, revision_id))
958
raise errors.RootMissing()
959
return self._add_inventory_checked(revision_id, inv, parents)
961
def _add_inventory_checked(self, revision_id, inv, parents):
962
"""Add inv to the repository after checking the inputs.
964
This function can be overridden to allow different inventory styles.
966
:seealso: add_inventory, for the contract.
968
inv_lines = self._serializer.write_inventory_to_lines(inv)
969
return self._inventory_add_lines(revision_id, parents,
970
inv_lines, check_content=False)
972
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
973
parents, basis_inv=None, propagate_caches=False):
974
"""Add a new inventory expressed as a delta against another revision.
976
See the inventory developers documentation for the theory behind
979
:param basis_revision_id: The inventory id the delta was created
980
against. (This does not have to be a direct parent.)
981
:param delta: The inventory delta (see Inventory.apply_delta for
983
:param new_revision_id: The revision id that the inventory is being
985
:param parents: The revision ids of the parents that revision_id is
986
known to have and are in the repository already. These are supplied
987
for repositories that depend on the inventory graph for revision
988
graph access, as well as for those that pun ancestry with delta
990
:param basis_inv: The basis inventory if it is already known,
992
:param propagate_caches: If True, the caches for this inventory are
993
copied to and updated for the result if possible.
995
:returns: (validator, new_inv)
996
The validator(which is a sha1 digest, though what is sha'd is
997
repository format specific) of the serialized inventory, and the
1000
if not self.is_in_write_group():
1001
raise AssertionError("%r not in write group" % (self,))
1002
_mod_revision.check_not_reserved_id(new_revision_id)
1003
basis_tree = self.revision_tree(basis_revision_id)
1004
basis_tree.lock_read()
1006
# Note that this mutates the inventory of basis_tree, which not all
1007
# inventory implementations may support: A better idiom would be to
1008
# return a new inventory, but as there is no revision tree cache in
1009
# repository this is safe for now - RBC 20081013
1010
if basis_inv is None:
1011
basis_inv = basis_tree.inventory
1012
basis_inv.apply_delta(delta)
1013
basis_inv.revision_id = new_revision_id
1014
return (self.add_inventory(new_revision_id, basis_inv, parents),
1019
def _inventory_add_lines(self, revision_id, parents, lines,
1020
check_content=True):
1021
"""Store lines in inv_vf and return the sha1 of the inventory."""
1022
parents = [(parent,) for parent in parents]
1023
result = self.inventories.add_lines((revision_id,), parents, lines,
1024
check_content=check_content)[0]
1025
self.inventories._access.flush()
1028
def add_revision(self, revision_id, rev, inv=None, config=None):
1029
"""Add rev to the revision store as revision_id.
1031
:param revision_id: the revision id to use.
1032
:param rev: The revision object.
1033
:param inv: The inventory for the revision. if None, it will be looked
1034
up in the inventory storer
1035
:param config: If None no digital signature will be created.
1036
If supplied its signature_needed method will be used
1037
to determine if a signature should be made.
1039
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1041
_mod_revision.check_not_reserved_id(revision_id)
1042
if config is not None and config.signature_needed():
1044
inv = self.get_inventory(revision_id)
1045
tree = InventoryRevisionTree(self, inv, revision_id)
1046
testament = Testament(rev, tree)
1047
plaintext = testament.as_short_text()
1048
self.store_revision_signature(
1049
gpg.GPGStrategy(config), plaintext, revision_id)
1050
# check inventory present
1051
if not self.inventories.get_parent_map([(revision_id,)]):
1053
raise errors.WeaveRevisionNotPresent(revision_id,
1056
# yes, this is not suitable for adding with ghosts.
1057
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1060
key = (revision_id,)
1061
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1062
self._add_revision(rev)
1064
def _add_revision(self, revision):
1065
text = self._serializer.write_revision_to_string(revision)
1066
key = (revision.revision_id,)
1067
parents = tuple((parent,) for parent in revision.parent_ids)
1068
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1070
def _check_inventories(self, checker):
1071
"""Check the inventories found from the revision scan.
1073
This is responsible for verifying the sha1 of inventories and
1074
creating a pending_keys set that covers data referenced by inventories.
1076
bar = ui.ui_factory.nested_progress_bar()
1078
self._do_check_inventories(checker, bar)
1082
def _do_check_inventories(self, checker, bar):
1083
"""Helper for _check_inventories."""
1085
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1086
kinds = ['chk_bytes', 'texts']
1087
count = len(checker.pending_keys)
1088
bar.update("inventories", 0, 2)
1089
current_keys = checker.pending_keys
1090
checker.pending_keys = {}
1091
# Accumulate current checks.
1092
for key in current_keys:
1093
if key[0] != 'inventories' and key[0] not in kinds:
1094
checker._report_items.append('unknown key type %r' % (key,))
1095
keys[key[0]].add(key[1:])
1096
if keys['inventories']:
1097
# NB: output order *should* be roughly sorted - topo or
1098
# inverse topo depending on repository - either way decent
1099
# to just delta against. However, pre-CHK formats didn't
1100
# try to optimise inventory layout on disk. As such the
1101
# pre-CHK code path does not use inventory deltas.
1103
for record in self.inventories.check(keys=keys['inventories']):
1104
if record.storage_kind == 'absent':
1105
checker._report_items.append(
1106
'Missing inventory {%s}' % (record.key,))
1108
last_object = self._check_record('inventories', record,
1109
checker, last_object,
1110
current_keys[('inventories',) + record.key])
1111
del keys['inventories']
1114
bar.update("texts", 1)
1115
while (checker.pending_keys or keys['chk_bytes']
1117
# Something to check.
1118
current_keys = checker.pending_keys
1119
checker.pending_keys = {}
1120
# Accumulate current checks.
1121
for key in current_keys:
1122
if key[0] not in kinds:
1123
checker._report_items.append('unknown key type %r' % (key,))
1124
keys[key[0]].add(key[1:])
1125
# Check the outermost kind only - inventories || chk_bytes || texts
1129
for record in getattr(self, kind).check(keys=keys[kind]):
1130
if record.storage_kind == 'absent':
1131
checker._report_items.append(
1132
'Missing %s {%s}' % (kind, record.key,))
1134
last_object = self._check_record(kind, record,
1135
checker, last_object, current_keys[(kind,) + record.key])
1139
def _check_record(self, kind, record, checker, last_object, item_data):
1140
"""Check a single text from this repository."""
1141
if kind == 'inventories':
1142
rev_id = record.key[0]
1143
inv = self._deserialise_inventory(rev_id,
1144
record.get_bytes_as('fulltext'))
1145
if last_object is not None:
1146
delta = inv._make_delta(last_object)
1147
for old_path, path, file_id, ie in delta:
1150
ie.check(checker, rev_id, inv)
1152
for path, ie in inv.iter_entries():
1153
ie.check(checker, rev_id, inv)
1154
if self._format.fast_deltas:
1156
elif kind == 'chk_bytes':
1157
# No code written to check chk_bytes for this repo format.
1158
checker._report_items.append(
1159
'unsupported key type chk_bytes for %s' % (record.key,))
1160
elif kind == 'texts':
1161
self._check_text(record, checker, item_data)
1163
checker._report_items.append(
1164
'unknown key type %s for %s' % (kind, record.key))
1166
def _check_text(self, record, checker, item_data):
1167
"""Check a single text."""
1168
# Check it is extractable.
1169
# TODO: check length.
1170
if record.storage_kind == 'chunked':
1171
chunks = record.get_bytes_as(record.storage_kind)
1172
sha1 = osutils.sha_strings(chunks)
1173
length = sum(map(len, chunks))
1175
content = record.get_bytes_as('fulltext')
1176
sha1 = osutils.sha_string(content)
1177
length = len(content)
1178
if item_data and sha1 != item_data[1]:
1179
checker._report_items.append(
1180
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1181
(record.key, sha1, item_data[1], item_data[2]))
1184
def _eliminate_revisions_not_present(self, revision_ids):
1185
"""Check every revision id in revision_ids to see if we have it.
1187
Returns a set of the present revisions.
1190
graph = self.get_graph()
1191
parent_map = graph.get_parent_map(revision_ids)
1192
# The old API returned a list, should this actually be a set?
1193
return parent_map.keys()
1195
def __init__(self, _format, a_bzrdir, control_files):
1196
"""Instantiate a VersionedFileRepository.
1198
:param _format: The format of the repository on disk.
1199
:param a_bzrdir: The BzrDir of the repository.
1200
:param control_files: Control files to use for locking, etc.
1202
# In the future we will have a single api for all stores for
1203
# getting file texts, inventories and revisions, then
1204
# this construct will accept instances of those things.
1205
super(VersionedFileRepository, self).__init__(_format, a_bzrdir,
1208
self._reconcile_does_inventory_gc = True
1209
self._reconcile_fixes_text_parents = False
1210
self._reconcile_backsup_inventory = True
1211
# An InventoryEntry cache, used during deserialization
1212
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1213
# Is it safe to return inventory entries directly from the entry cache,
1214
# rather copying them?
1215
self._safe_to_return_from_cache = False
1218
def gather_stats(self, revid=None, committers=None):
1219
"""See Repository.gather_stats()."""
1220
result = super(VersionedFileRepository, self).gather_stats(revid, committers)
1221
# now gather global repository information
1222
# XXX: This is available for many repos regardless of listability.
1223
if self.user_transport.listable():
1224
# XXX: do we want to __define len__() ?
1225
# Maybe the versionedfiles object should provide a different
1226
# method to get the number of keys.
1227
result['revisions'] = len(self.revisions.keys())
1228
# result['size'] = t
1231
def get_commit_builder(self, branch, parents, config, timestamp=None,
1232
timezone=None, committer=None, revprops=None,
1233
revision_id=None, lossy=False):
1234
"""Obtain a CommitBuilder for this repository.
1236
:param branch: Branch to commit to.
1237
:param parents: Revision ids of the parents of the new revision.
1238
:param config: Configuration to use.
1239
:param timestamp: Optional timestamp recorded for commit.
1240
:param timezone: Optional timezone for timestamp.
1241
:param committer: Optional committer to set for commit.
1242
:param revprops: Optional dictionary of revision properties.
1243
:param revision_id: Optional revision id.
1244
:param lossy: Whether to discard data that can not be natively
1245
represented, when pushing to a foreign VCS
1247
if self._fallback_repositories and not self._format.supports_chks:
1248
raise errors.BzrError("Cannot commit directly to a stacked branch"
1249
" in pre-2a formats. See "
1250
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1251
result = self._commit_builder_class(self, parents, config,
1252
timestamp, timezone, committer, revprops, revision_id,
1254
self.start_write_group()
1257
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1258
"""Return the keys of missing inventory parents for revisions added in
1261
A revision is not complete if the inventory delta for that revision
1262
cannot be calculated. Therefore if the parent inventories of a
1263
revision are not present, the revision is incomplete, and e.g. cannot
1264
be streamed by a smart server. This method finds missing inventory
1265
parents for revisions added in this write group.
1267
if not self._format.supports_external_lookups:
1268
# This is only an issue for stacked repositories
1270
if not self.is_in_write_group():
1271
raise AssertionError('not in a write group')
1273
# XXX: We assume that every added revision already has its
1274
# corresponding inventory, so we only check for parent inventories that
1275
# might be missing, rather than all inventories.
1276
parents = set(self.revisions._index.get_missing_parents())
1277
parents.discard(_mod_revision.NULL_REVISION)
1278
unstacked_inventories = self.inventories._index
1279
present_inventories = unstacked_inventories.get_parent_map(
1280
key[-1:] for key in parents)
1281
parents.difference_update(present_inventories)
1282
if len(parents) == 0:
1283
# No missing parent inventories.
1285
if not check_for_missing_texts:
1286
return set(('inventories', rev_id) for (rev_id,) in parents)
1287
# Ok, now we have a list of missing inventories. But these only matter
1288
# if the inventories that reference them are missing some texts they
1289
# appear to introduce.
1290
# XXX: Texts referenced by all added inventories need to be present,
1291
# but at the moment we're only checking for texts referenced by
1292
# inventories at the graph's edge.
1293
key_deps = self.revisions._index._key_dependencies
1294
key_deps.satisfy_refs_for_keys(present_inventories)
1295
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1296
file_ids = self.fileids_altered_by_revision_ids(referrers)
1297
missing_texts = set()
1298
for file_id, version_ids in file_ids.iteritems():
1299
missing_texts.update(
1300
(file_id, version_id) for version_id in version_ids)
1301
present_texts = self.texts.get_parent_map(missing_texts)
1302
missing_texts.difference_update(present_texts)
1303
if not missing_texts:
1304
# No texts are missing, so all revisions and their deltas are
1307
# Alternatively the text versions could be returned as the missing
1308
# keys, but this is likely to be less data.
1309
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1313
def has_revisions(self, revision_ids):
1314
"""Probe to find out the presence of multiple revisions.
1316
:param revision_ids: An iterable of revision_ids.
1317
:return: A set of the revision_ids that were present.
1319
parent_map = self.revisions.get_parent_map(
1320
[(rev_id,) for rev_id in revision_ids])
1322
if _mod_revision.NULL_REVISION in revision_ids:
1323
result.add(_mod_revision.NULL_REVISION)
1324
result.update([key[0] for key in parent_map])
1328
def get_revision_reconcile(self, revision_id):
1329
"""'reconcile' helper routine that allows access to a revision always.
1331
This variant of get_revision does not cross check the weave graph
1332
against the revision one as get_revision does: but it should only
1333
be used by reconcile, or reconcile-alike commands that are correcting
1334
or testing the revision graph.
1336
return self._get_revisions([revision_id])[0]
1339
def get_revisions(self, revision_ids):
1340
"""Get many revisions at once.
1342
Repositories that need to check data on every revision read should
1343
subclass this method.
1345
return self._get_revisions(revision_ids)
1348
def _get_revisions(self, revision_ids):
1349
"""Core work logic to get many revisions without sanity checks."""
1351
for revid, rev in self._iter_revisions(revision_ids):
1353
raise errors.NoSuchRevision(self, revid)
1355
return [revs[revid] for revid in revision_ids]
1357
def _iter_revisions(self, revision_ids):
1358
"""Iterate over revision objects.
1360
:param revision_ids: An iterable of revisions to examine. None may be
1361
passed to request all revisions known to the repository. Note that
1362
not all repositories can find unreferenced revisions; for those
1363
repositories only referenced ones will be returned.
1364
:return: An iterator of (revid, revision) tuples. Absent revisions (
1365
those asked for but not available) are returned as (revid, None).
1367
if revision_ids is None:
1368
revision_ids = self.all_revision_ids()
1370
for rev_id in revision_ids:
1371
if not rev_id or not isinstance(rev_id, basestring):
1372
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1373
keys = [(key,) for key in revision_ids]
1374
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1375
for record in stream:
1376
revid = record.key[0]
1377
if record.storage_kind == 'absent':
1380
text = record.get_bytes_as('fulltext')
1381
rev = self._serializer.read_revision_from_string(text)
1385
def add_signature_text(self, revision_id, signature):
1386
"""Store a signature text for a revision.
1388
:param revision_id: Revision id of the revision
1389
:param signature: Signature text.
1391
self.signatures.add_lines((revision_id,), (),
1392
osutils.split_lines(signature))
1394
def find_text_key_references(self):
1395
"""Find the text key references within the repository.
1397
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1398
to whether they were referred to by the inventory of the
1399
revision_id that they contain. The inventory texts from all present
1400
revision ids are assessed to generate this report.
1402
revision_keys = self.revisions.keys()
1403
w = self.inventories
1404
pb = ui.ui_factory.nested_progress_bar()
1406
return self._serializer._find_text_key_references(
1407
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1411
def _inventory_xml_lines_for_keys(self, keys):
1412
"""Get a line iterator of the sort needed for findind references.
1414
Not relevant for non-xml inventory repositories.
1416
Ghosts in revision_keys are ignored.
1418
:param revision_keys: The revision keys for the inventories to inspect.
1419
:return: An iterator over (inventory line, revid) for the fulltexts of
1420
all of the xml inventories specified by revision_keys.
1422
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1423
for record in stream:
1424
if record.storage_kind != 'absent':
1425
chunks = record.get_bytes_as('chunked')
1426
revid = record.key[-1]
1427
lines = osutils.chunks_to_lines(chunks)
1431
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1433
"""Helper routine for fileids_altered_by_revision_ids.
1435
This performs the translation of xml lines to revision ids.
1437
:param line_iterator: An iterator of lines, origin_version_id
1438
:param revision_keys: The revision ids to filter for. This should be a
1439
set or other type which supports efficient __contains__ lookups, as
1440
the revision key from each parsed line will be looked up in the
1441
revision_keys filter.
1442
:return: a dictionary mapping altered file-ids to an iterable of
1443
revision_ids. Each altered file-ids has the exact revision_ids that
1444
altered it listed explicitly.
1446
seen = set(self._serializer._find_text_key_references(
1447
line_iterator).iterkeys())
1448
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
1449
parent_seen = set(self._serializer._find_text_key_references(
1450
self._inventory_xml_lines_for_keys(parent_keys)))
1451
new_keys = seen - parent_seen
1453
setdefault = result.setdefault
1454
for key in new_keys:
1455
setdefault(key[0], set()).add(key[-1])
1458
def _find_parent_keys_of_revisions(self, revision_keys):
1459
"""Similar to _find_parent_ids_of_revisions, but used with keys.
1461
:param revision_keys: An iterable of revision_keys.
1462
:return: The parents of all revision_keys that are not already in
1465
parent_map = self.revisions.get_parent_map(revision_keys)
1467
map(parent_keys.update, parent_map.itervalues())
1468
parent_keys.difference_update(revision_keys)
1469
parent_keys.discard(_mod_revision.NULL_REVISION)
1472
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
1473
"""Find the file ids and versions affected by revisions.
1475
:param revisions: an iterable containing revision ids.
1476
:param _inv_weave: The inventory weave from this repository or None.
1477
If None, the inventory weave will be opened automatically.
1478
:return: a dictionary mapping altered file-ids to an iterable of
1479
revision_ids. Each altered file-ids has the exact revision_ids that
1480
altered it listed explicitly.
1482
selected_keys = set((revid,) for revid in revision_ids)
1483
w = _inv_weave or self.inventories
1484
return self._find_file_ids_from_xml_inventory_lines(
1485
w.iter_lines_added_or_present_in_keys(
1486
selected_keys, pb=None),
1489
def iter_files_bytes(self, desired_files):
1490
"""Iterate through file versions.
1492
Files will not necessarily be returned in the order they occur in
1493
desired_files. No specific order is guaranteed.
1495
Yields pairs of identifier, bytes_iterator. identifier is an opaque
1496
value supplied by the caller as part of desired_files. It should
1497
uniquely identify the file version in the caller's context. (Examples:
1498
an index number or a TreeTransform trans_id.)
1500
bytes_iterator is an iterable of bytestrings for the file. The
1501
kind of iterable and length of the bytestrings are unspecified, but for
1502
this implementation, it is a list of bytes produced by
1503
VersionedFile.get_record_stream().
1505
:param desired_files: a list of (file_id, revision_id, identifier)
1509
for file_id, revision_id, callable_data in desired_files:
1510
text_keys[(file_id, revision_id)] = callable_data
1511
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
1512
if record.storage_kind == 'absent':
1513
raise errors.RevisionNotPresent(record.key, self)
1514
yield text_keys[record.key], record.get_bytes_as('chunked')
1516
def _generate_text_key_index(self, text_key_references=None,
1518
"""Generate a new text key index for the repository.
1520
This is an expensive function that will take considerable time to run.
1522
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
1523
list of parents, also text keys. When a given key has no parents,
1524
the parents list will be [NULL_REVISION].
1526
# All revisions, to find inventory parents.
1527
if ancestors is None:
1528
graph = self.get_graph()
1529
ancestors = graph.get_parent_map(self.all_revision_ids())
1530
if text_key_references is None:
1531
text_key_references = self.find_text_key_references()
1532
pb = ui.ui_factory.nested_progress_bar()
1534
return self._do_generate_text_key_index(ancestors,
1535
text_key_references, pb)
1539
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
1540
"""Helper for _generate_text_key_index to avoid deep nesting."""
1541
revision_order = tsort.topo_sort(ancestors)
1542
invalid_keys = set()
1544
for revision_id in revision_order:
1545
revision_keys[revision_id] = set()
1546
text_count = len(text_key_references)
1547
# a cache of the text keys to allow reuse; costs a dict of all the
1548
# keys, but saves a 2-tuple for every child of a given key.
1550
for text_key, valid in text_key_references.iteritems():
1552
invalid_keys.add(text_key)
1554
revision_keys[text_key[1]].add(text_key)
1555
text_key_cache[text_key] = text_key
1556
del text_key_references
1558
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
1559
NULL_REVISION = _mod_revision.NULL_REVISION
1560
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
1561
# too small for large or very branchy trees. However, for 55K path
1562
# trees, it would be easy to use too much memory trivially. Ideally we
1563
# could gauge this by looking at available real memory etc, but this is
1564
# always a tricky proposition.
1565
inventory_cache = lru_cache.LRUCache(10)
1566
batch_size = 10 # should be ~150MB on a 55K path tree
1567
batch_count = len(revision_order) / batch_size + 1
1569
pb.update("Calculating text parents", processed_texts, text_count)
1570
for offset in xrange(batch_count):
1571
to_query = revision_order[offset * batch_size:(offset + 1) *
1575
for revision_id in to_query:
1576
parent_ids = ancestors[revision_id]
1577
for text_key in revision_keys[revision_id]:
1578
pb.update("Calculating text parents", processed_texts)
1579
processed_texts += 1
1580
candidate_parents = []
1581
for parent_id in parent_ids:
1582
parent_text_key = (text_key[0], parent_id)
1584
check_parent = parent_text_key not in \
1585
revision_keys[parent_id]
1587
# the parent parent_id is a ghost:
1588
check_parent = False
1589
# truncate the derived graph against this ghost.
1590
parent_text_key = None
1592
# look at the parent commit details inventories to
1593
# determine possible candidates in the per file graph.
1596
inv = inventory_cache[parent_id]
1598
inv = self.revision_tree(parent_id).inventory
1599
inventory_cache[parent_id] = inv
1601
parent_entry = inv[text_key[0]]
1602
except (KeyError, errors.NoSuchId):
1604
if parent_entry is not None:
1606
text_key[0], parent_entry.revision)
1608
parent_text_key = None
1609
if parent_text_key is not None:
1610
candidate_parents.append(
1611
text_key_cache[parent_text_key])
1612
parent_heads = text_graph.heads(candidate_parents)
1613
new_parents = list(parent_heads)
1614
new_parents.sort(key=lambda x:candidate_parents.index(x))
1615
if new_parents == []:
1616
new_parents = [NULL_REVISION]
1617
text_index[text_key] = new_parents
1619
for text_key in invalid_keys:
1620
text_index[text_key] = [NULL_REVISION]
1623
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
1624
"""Get an iterable listing the keys of all the data introduced by a set
1627
The keys will be ordered so that the corresponding items can be safely
1628
fetched and inserted in that order.
1630
:returns: An iterable producing tuples of (knit-kind, file-id,
1631
versions). knit-kind is one of 'file', 'inventory', 'signatures',
1632
'revisions'. file-id is None unless knit-kind is 'file'.
1634
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
1637
for result in self._find_non_file_keys_to_fetch(revision_ids):
1640
def _find_file_keys_to_fetch(self, revision_ids, pb):
1641
# XXX: it's a bit weird to control the inventory weave caching in this
1642
# generator. Ideally the caching would be done in fetch.py I think. Or
1643
# maybe this generator should explicitly have the contract that it
1644
# should not be iterated until the previously yielded item has been
1646
inv_w = self.inventories
1648
# file ids that changed
1649
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
1651
num_file_ids = len(file_ids)
1652
for file_id, altered_versions in file_ids.iteritems():
1654
pb.update("Fetch texts", count, num_file_ids)
1656
yield ("file", file_id, altered_versions)
1658
def _find_non_file_keys_to_fetch(self, revision_ids):
1660
yield ("inventory", None, revision_ids)
1663
# XXX: Note ATM no callers actually pay attention to this return
1664
# instead they just use the list of revision ids and ignore
1665
# missing sigs. Consider removing this work entirely
1666
revisions_with_signatures = set(self.signatures.get_parent_map(
1667
[(r,) for r in revision_ids]))
1668
revisions_with_signatures = set(
1669
[r for (r,) in revisions_with_signatures])
1670
revisions_with_signatures.intersection_update(revision_ids)
1671
yield ("signatures", None, revisions_with_signatures)
1674
yield ("revisions", None, revision_ids)
1677
def get_inventory(self, revision_id):
1678
"""Get Inventory object by revision id."""
1679
return self.iter_inventories([revision_id]).next()
1681
def iter_inventories(self, revision_ids, ordering=None):
1682
"""Get many inventories by revision_ids.
1684
This will buffer some or all of the texts used in constructing the
1685
inventories in memory, but will only parse a single inventory at a
1688
:param revision_ids: The expected revision ids of the inventories.
1689
:param ordering: optional ordering, e.g. 'topological'. If not
1690
specified, the order of revision_ids will be preserved (by
1691
buffering if necessary).
1692
:return: An iterator of inventories.
1694
if ((None in revision_ids)
1695
or (_mod_revision.NULL_REVISION in revision_ids)):
1696
raise ValueError('cannot get null revision inventory')
1697
return self._iter_inventories(revision_ids, ordering)
1699
def _iter_inventories(self, revision_ids, ordering):
1700
"""single-document based inventory iteration."""
1701
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
1702
for text, revision_id in inv_xmls:
1703
yield self._deserialise_inventory(revision_id, text)
1705
def _iter_inventory_xmls(self, revision_ids, ordering):
1706
if ordering is None:
1707
order_as_requested = True
1708
ordering = 'unordered'
1710
order_as_requested = False
1711
keys = [(revision_id,) for revision_id in revision_ids]
1714
if order_as_requested:
1715
key_iter = iter(keys)
1716
next_key = key_iter.next()
1717
stream = self.inventories.get_record_stream(keys, ordering, True)
1719
for record in stream:
1720
if record.storage_kind != 'absent':
1721
chunks = record.get_bytes_as('chunked')
1722
if order_as_requested:
1723
text_chunks[record.key] = chunks
1725
yield ''.join(chunks), record.key[-1]
1727
raise errors.NoSuchRevision(self, record.key)
1728
if order_as_requested:
1729
# Yield as many results as we can while preserving order.
1730
while next_key in text_chunks:
1731
chunks = text_chunks.pop(next_key)
1732
yield ''.join(chunks), next_key[-1]
1734
next_key = key_iter.next()
1735
except StopIteration:
1736
# We still want to fully consume the get_record_stream,
1737
# just in case it is not actually finished at this point
1741
def _deserialise_inventory(self, revision_id, xml):
1742
"""Transform the xml into an inventory object.
1744
:param revision_id: The expected revision id of the inventory.
1745
:param xml: A serialised inventory.
1747
result = self._serializer.read_inventory_from_string(xml, revision_id,
1748
entry_cache=self._inventory_entry_cache,
1749
return_from_cache=self._safe_to_return_from_cache)
1750
if result.revision_id != revision_id:
1751
raise AssertionError('revision id mismatch %s != %s' % (
1752
result.revision_id, revision_id))
1755
def get_serializer_format(self):
1756
return self._serializer.format_num
1759
def _get_inventory_xml(self, revision_id):
1760
"""Get serialized inventory as a string."""
1761
texts = self._iter_inventory_xmls([revision_id], 'unordered')
1763
text, revision_id = texts.next()
1764
except StopIteration:
1765
raise errors.HistoryMissing(self, 'inventory', revision_id)
1769
def revision_tree(self, revision_id):
1770
"""Return Tree for a revision on this branch.
1772
`revision_id` may be NULL_REVISION for the empty tree revision.
1774
revision_id = _mod_revision.ensure_null(revision_id)
1775
# TODO: refactor this to use an existing revision object
1776
# so we don't need to read it in twice.
1777
if revision_id == _mod_revision.NULL_REVISION:
1778
return InventoryRevisionTree(self,
1779
Inventory(root_id=None), _mod_revision.NULL_REVISION)
1781
inv = self.get_inventory(revision_id)
1782
return InventoryRevisionTree(self, inv, revision_id)
1784
def revision_trees(self, revision_ids):
1785
"""Return Trees for revisions in this repository.
1787
:param revision_ids: a sequence of revision-ids;
1788
a revision-id may not be None or 'null:'
1790
inventories = self.iter_inventories(revision_ids)
1791
for inv in inventories:
1792
yield InventoryRevisionTree(self, inv, inv.revision_id)
1794
def _filtered_revision_trees(self, revision_ids, file_ids):
1795
"""Return Tree for a revision on this branch with only some files.
1797
:param revision_ids: a sequence of revision-ids;
1798
a revision-id may not be None or 'null:'
1799
:param file_ids: if not None, the result is filtered
1800
so that only those file-ids, their parents and their
1801
children are included.
1803
inventories = self.iter_inventories(revision_ids)
1804
for inv in inventories:
1805
# Should we introduce a FilteredRevisionTree class rather
1806
# than pre-filter the inventory here?
1807
filtered_inv = inv.filter(file_ids)
1808
yield InventoryRevisionTree(self, filtered_inv, filtered_inv.revision_id)
1810
def get_parent_map(self, revision_ids):
1811
"""See graph.StackedParentsProvider.get_parent_map"""
1812
# revisions index works in keys; this just works in revisions
1813
# therefore wrap and unwrap
1816
for revision_id in revision_ids:
1817
if revision_id == _mod_revision.NULL_REVISION:
1818
result[revision_id] = ()
1819
elif revision_id is None:
1820
raise ValueError('get_parent_map(None) is not valid')
1822
query_keys.append((revision_id ,))
1823
for ((revision_id,), parent_keys) in \
1824
self.revisions.get_parent_map(query_keys).iteritems():
1826
result[revision_id] = tuple([parent_revid
1827
for (parent_revid,) in parent_keys])
1829
result[revision_id] = (_mod_revision.NULL_REVISION,)
1833
def get_known_graph_ancestry(self, revision_ids):
1834
"""Return the known graph for a set of revision ids and their ancestors.
1836
st = static_tuple.StaticTuple
1837
revision_keys = [st(r_id).intern() for r_id in revision_ids]
1838
known_graph = self.revisions.get_known_graph_ancestry(revision_keys)
1839
return graph.GraphThunkIdsToKeys(known_graph)
1842
def get_file_graph(self):
1843
"""Return the graph walker for text revisions."""
1844
return graph.Graph(self.texts)
1846
def _get_versioned_file_checker(self, text_key_references=None,
1848
"""Return an object suitable for checking versioned files.
1850
:param text_key_references: if non-None, an already built
1851
dictionary mapping text keys ((fileid, revision_id) tuples)
1852
to whether they were referred to by the inventory of the
1853
revision_id that they contain. If None, this will be
1855
:param ancestors: Optional result from
1856
self.get_graph().get_parent_map(self.all_revision_ids()) if already
1859
return _VersionedFileChecker(self,
1860
text_key_references=text_key_references, ancestors=ancestors)
1863
def has_signature_for_revision_id(self, revision_id):
1864
"""Query for a revision signature for revision_id in the repository."""
1865
if not self.has_revision(revision_id):
1866
raise errors.NoSuchRevision(self, revision_id)
1867
sig_present = (1 == len(
1868
self.signatures.get_parent_map([(revision_id,)])))
1872
def get_signature_text(self, revision_id):
1873
"""Return the text for a signature."""
1874
stream = self.signatures.get_record_stream([(revision_id,)],
1876
record = stream.next()
1877
if record.storage_kind == 'absent':
1878
raise errors.NoSuchRevision(self, revision_id)
1879
return record.get_bytes_as('fulltext')
1882
def _check(self, revision_ids, callback_refs, check_repo):
1883
result = check.VersionedFileCheck(self, check_repo=check_repo)
1884
result.check(callback_refs)
1887
def _find_inconsistent_revision_parents(self, revisions_iterator=None):
1888
"""Find revisions with different parent lists in the revision object
1889
and in the index graph.
1891
:param revisions_iterator: None, or an iterator of (revid,
1892
Revision-or-None). This iterator controls the revisions checked.
1893
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1894
parents-in-revision).
1896
if not self.is_locked():
1897
raise AssertionError()
1899
if revisions_iterator is None:
1900
revisions_iterator = self._iter_revisions(None)
1901
for revid, revision in revisions_iterator:
1902
if revision is None:
1904
parent_map = vf.get_parent_map([(revid,)])
1905
parents_according_to_index = tuple(parent[-1] for parent in
1906
parent_map[(revid,)])
1907
parents_according_to_revision = tuple(revision.parent_ids)
1908
if parents_according_to_index != parents_according_to_revision:
1909
yield (revid, parents_according_to_index,
1910
parents_according_to_revision)
1912
def _check_for_inconsistent_revision_parents(self):
1913
inconsistencies = list(self._find_inconsistent_revision_parents())
1915
raise errors.BzrCheckError(
1916
"Revision knit has inconsistent parents.")
1918
def _get_sink(self):
1919
"""Return a sink for streaming into this repository."""
1920
return StreamSink(self)
1922
def _get_source(self, to_format):
1923
"""Return a source for streaming from this repository."""
1924
return StreamSource(self, to_format)
1927
class MetaDirVersionedFileRepository(MetaDirRepository,
1928
VersionedFileRepository):
1929
"""Repositories in a meta-dir, that work via versioned file objects."""
1931
def __init__(self, _format, a_bzrdir, control_files):
1932
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
1936
class MetaDirVersionedFileRepositoryFormat(MetaDirRepositoryFormat,
1937
VersionedFileRepositoryFormat):
1938
"""Base class for repository formats using versioned files in metadirs."""
1941
class StreamSink(object):
1942
"""An object that can insert a stream into a repository.
1944
This interface handles the complexity of reserialising inventories and
1945
revisions from different formats, and allows unidirectional insertion into
1946
stacked repositories without looking for the missing basis parents
1950
def __init__(self, target_repo):
1951
self.target_repo = target_repo
1953
def insert_stream(self, stream, src_format, resume_tokens):
1954
"""Insert a stream's content into the target repository.
1956
:param src_format: a bzr repository format.
1958
:return: a list of resume tokens and an iterable of keys additional
1959
items required before the insertion can be completed.
1961
self.target_repo.lock_write()
1964
self.target_repo.resume_write_group(resume_tokens)
1967
self.target_repo.start_write_group()
1970
# locked_insert_stream performs a commit|suspend.
1971
missing_keys = self.insert_stream_without_locking(stream,
1972
src_format, is_resume)
1974
# suspend the write group and tell the caller what we is
1975
# missing. We know we can suspend or else we would not have
1976
# entered this code path. (All repositories that can handle
1977
# missing keys can handle suspending a write group).
1978
write_group_tokens = self.target_repo.suspend_write_group()
1979
return write_group_tokens, missing_keys
1980
hint = self.target_repo.commit_write_group()
1981
to_serializer = self.target_repo._format._serializer
1982
src_serializer = src_format._serializer
1983
if (to_serializer != src_serializer and
1984
self.target_repo._format.pack_compresses):
1985
self.target_repo.pack(hint=hint)
1988
self.target_repo.abort_write_group(suppress_errors=True)
1991
self.target_repo.unlock()
1993
def insert_stream_without_locking(self, stream, src_format,
1995
"""Insert a stream's content into the target repository.
1997
This assumes that you already have a locked repository and an active
2000
:param src_format: a bzr repository format.
2001
:param is_resume: Passed down to get_missing_parent_inventories to
2002
indicate if we should be checking for missing texts at the same
2005
:return: A set of keys that are missing.
2007
if not self.target_repo.is_write_locked():
2008
raise errors.ObjectNotLocked(self)
2009
if not self.target_repo.is_in_write_group():
2010
raise errors.BzrError('you must already be in a write group')
2011
to_serializer = self.target_repo._format._serializer
2012
src_serializer = src_format._serializer
2014
if to_serializer == src_serializer:
2015
# If serializers match and the target is a pack repository, set the
2016
# write cache size on the new pack. This avoids poor performance
2017
# on transports where append is unbuffered (such as
2018
# RemoteTransport). This is safe to do because nothing should read
2019
# back from the target repository while a stream with matching
2020
# serialization is being inserted.
2021
# The exception is that a delta record from the source that should
2022
# be a fulltext may need to be expanded by the target (see
2023
# test_fetch_revisions_with_deltas_into_pack); but we take care to
2024
# explicitly flush any buffered writes first in that rare case.
2026
new_pack = self.target_repo._pack_collection._new_pack
2027
except AttributeError:
2028
# Not a pack repository
2031
new_pack.set_write_cache_size(1024*1024)
2032
for substream_type, substream in stream:
2033
if 'stream' in debug.debug_flags:
2034
mutter('inserting substream: %s', substream_type)
2035
if substream_type == 'texts':
2036
self.target_repo.texts.insert_record_stream(substream)
2037
elif substream_type == 'inventories':
2038
if src_serializer == to_serializer:
2039
self.target_repo.inventories.insert_record_stream(
2042
self._extract_and_insert_inventories(
2043
substream, src_serializer)
2044
elif substream_type == 'inventory-deltas':
2045
self._extract_and_insert_inventory_deltas(
2046
substream, src_serializer)
2047
elif substream_type == 'chk_bytes':
2048
# XXX: This doesn't support conversions, as it assumes the
2049
# conversion was done in the fetch code.
2050
self.target_repo.chk_bytes.insert_record_stream(substream)
2051
elif substream_type == 'revisions':
2052
# This may fallback to extract-and-insert more often than
2053
# required if the serializers are different only in terms of
2055
if src_serializer == to_serializer:
2056
self.target_repo.revisions.insert_record_stream(substream)
2058
self._extract_and_insert_revisions(substream,
2060
elif substream_type == 'signatures':
2061
self.target_repo.signatures.insert_record_stream(substream)
2063
raise AssertionError('kaboom! %s' % (substream_type,))
2064
# Done inserting data, and the missing_keys calculations will try to
2065
# read back from the inserted data, so flush the writes to the new pack
2066
# (if this is pack format).
2067
if new_pack is not None:
2068
new_pack._write_data('', flush=True)
2069
# Find all the new revisions (including ones from resume_tokens)
2070
missing_keys = self.target_repo.get_missing_parent_inventories(
2071
check_for_missing_texts=is_resume)
2073
for prefix, versioned_file in (
2074
('texts', self.target_repo.texts),
2075
('inventories', self.target_repo.inventories),
2076
('revisions', self.target_repo.revisions),
2077
('signatures', self.target_repo.signatures),
2078
('chk_bytes', self.target_repo.chk_bytes),
2080
if versioned_file is None:
2082
# TODO: key is often going to be a StaticTuple object
2083
# I don't believe we can define a method by which
2084
# (prefix,) + StaticTuple will work, though we could
2085
# define a StaticTuple.sq_concat that would allow you to
2086
# pass in either a tuple or a StaticTuple as the second
2087
# object, so instead we could have:
2088
# StaticTuple(prefix) + key here...
2089
missing_keys.update((prefix,) + key for key in
2090
versioned_file.get_missing_compression_parent_keys())
2091
except NotImplementedError:
2092
# cannot even attempt suspending, and missing would have failed
2093
# during stream insertion.
2094
missing_keys = set()
2097
def _extract_and_insert_inventory_deltas(self, substream, serializer):
2098
target_rich_root = self.target_repo._format.rich_root_data
2099
target_tree_refs = self.target_repo._format.supports_tree_reference
2100
for record in substream:
2101
# Insert the delta directly
2102
inventory_delta_bytes = record.get_bytes_as('fulltext')
2103
deserialiser = inventory_delta.InventoryDeltaDeserializer()
2105
parse_result = deserialiser.parse_text_bytes(
2106
inventory_delta_bytes)
2107
except inventory_delta.IncompatibleInventoryDelta, err:
2108
mutter("Incompatible delta: %s", err.msg)
2109
raise errors.IncompatibleRevision(self.target_repo._format)
2110
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
2111
revision_id = new_id
2112
parents = [key[0] for key in record.parents]
2113
self.target_repo.add_inventory_by_delta(
2114
basis_id, inv_delta, revision_id, parents)
2116
def _extract_and_insert_inventories(self, substream, serializer,
2118
"""Generate a new inventory versionedfile in target, converting data.
2120
The inventory is retrieved from the source, (deserializing it), and
2121
stored in the target (reserializing it in a different format).
2123
target_rich_root = self.target_repo._format.rich_root_data
2124
target_tree_refs = self.target_repo._format.supports_tree_reference
2125
for record in substream:
2126
# It's not a delta, so it must be a fulltext in the source
2127
# serializer's format.
2128
bytes = record.get_bytes_as('fulltext')
2129
revision_id = record.key[0]
2130
inv = serializer.read_inventory_from_string(bytes, revision_id)
2131
parents = [key[0] for key in record.parents]
2132
self.target_repo.add_inventory(revision_id, inv, parents)
2133
# No need to keep holding this full inv in memory when the rest of
2134
# the substream is likely to be all deltas.
2137
def _extract_and_insert_revisions(self, substream, serializer):
2138
for record in substream:
2139
bytes = record.get_bytes_as('fulltext')
2140
revision_id = record.key[0]
2141
rev = serializer.read_revision_from_string(bytes)
2142
if rev.revision_id != revision_id:
2143
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
2144
self.target_repo.add_revision(revision_id, rev)
2147
if self.target_repo._format._fetch_reconcile:
2148
self.target_repo.reconcile()
2151
class StreamSource(object):
2152
"""A source of a stream for fetching between repositories."""
2154
def __init__(self, from_repository, to_format):
2155
"""Create a StreamSource streaming from from_repository."""
2156
self.from_repository = from_repository
2157
self.to_format = to_format
2158
self._record_counter = RecordCounter()
2160
def delta_on_metadata(self):
2161
"""Return True if delta's are permitted on metadata streams.
2163
That is on revisions and signatures.
2165
src_serializer = self.from_repository._format._serializer
2166
target_serializer = self.to_format._serializer
2167
return (self.to_format._fetch_uses_deltas and
2168
src_serializer == target_serializer)
2170
def _fetch_revision_texts(self, revs):
2171
# fetch signatures first and then the revision texts
2172
# may need to be a InterRevisionStore call here.
2173
from_sf = self.from_repository.signatures
2174
# A missing signature is just skipped.
2175
keys = [(rev_id,) for rev_id in revs]
2176
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
2178
self.to_format._fetch_order,
2179
not self.to_format._fetch_uses_deltas))
2180
# If a revision has a delta, this is actually expanded inside the
2181
# insert_record_stream code now, which is an alternate fix for
2183
from_rf = self.from_repository.revisions
2184
revisions = from_rf.get_record_stream(
2186
self.to_format._fetch_order,
2187
not self.delta_on_metadata())
2188
return [('signatures', signatures), ('revisions', revisions)]
2190
def _generate_root_texts(self, revs):
2191
"""This will be called by get_stream between fetching weave texts and
2192
fetching the inventory weave.
2194
if self._rich_root_upgrade():
2195
return _mod_fetch.Inter1and2Helper(
2196
self.from_repository).generate_root_texts(revs)
2200
def get_stream(self, search):
2202
revs = search.get_keys()
2203
graph = self.from_repository.get_graph()
2204
revs = tsort.topo_sort(graph.get_parent_map(revs))
2205
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
2207
for knit_kind, file_id, revisions in data_to_fetch:
2208
if knit_kind != phase:
2210
# Make a new progress bar for this phase
2211
if knit_kind == "file":
2212
# Accumulate file texts
2213
text_keys.extend([(file_id, revision) for revision in
2215
elif knit_kind == "inventory":
2216
# Now copy the file texts.
2217
from_texts = self.from_repository.texts
2218
yield ('texts', from_texts.get_record_stream(
2219
text_keys, self.to_format._fetch_order,
2220
not self.to_format._fetch_uses_deltas))
2221
# Cause an error if a text occurs after we have done the
2224
# Before we process the inventory we generate the root
2225
# texts (if necessary) so that the inventories references
2227
for _ in self._generate_root_texts(revs):
2229
# we fetch only the referenced inventories because we do not
2230
# know for unselected inventories whether all their required
2231
# texts are present in the other repository - it could be
2233
for info in self._get_inventory_stream(revs):
2235
elif knit_kind == "signatures":
2236
# Nothing to do here; this will be taken care of when
2237
# _fetch_revision_texts happens.
2239
elif knit_kind == "revisions":
2240
for record in self._fetch_revision_texts(revs):
2243
raise AssertionError("Unknown knit kind %r" % knit_kind)
2245
def get_stream_for_missing_keys(self, missing_keys):
2246
# missing keys can only occur when we are byte copying and not
2247
# translating (because translation means we don't send
2248
# unreconstructable deltas ever).
2250
keys['texts'] = set()
2251
keys['revisions'] = set()
2252
keys['inventories'] = set()
2253
keys['chk_bytes'] = set()
2254
keys['signatures'] = set()
2255
for key in missing_keys:
2256
keys[key[0]].add(key[1:])
2257
if len(keys['revisions']):
2258
# If we allowed copying revisions at this point, we could end up
2259
# copying a revision without copying its required texts: a
2260
# violation of the requirements for repository integrity.
2261
raise AssertionError(
2262
'cannot copy revisions to fill in missing deltas %s' % (
2263
keys['revisions'],))
2264
for substream_kind, keys in keys.iteritems():
2265
vf = getattr(self.from_repository, substream_kind)
2266
if vf is None and keys:
2267
raise AssertionError(
2268
"cannot fill in keys for a versioned file we don't"
2269
" have: %s needs %s" % (substream_kind, keys))
2271
# No need to stream something we don't have
2273
if substream_kind == 'inventories':
2274
# Some missing keys are genuinely ghosts, filter those out.
2275
present = self.from_repository.inventories.get_parent_map(keys)
2276
revs = [key[0] for key in present]
2277
# Get the inventory stream more-or-less as we do for the
2278
# original stream; there's no reason to assume that records
2279
# direct from the source will be suitable for the sink. (Think
2280
# e.g. 2a -> 1.9-rich-root).
2281
for info in self._get_inventory_stream(revs, missing=True):
2285
# Ask for full texts always so that we don't need more round trips
2286
# after this stream.
2287
# Some of the missing keys are genuinely ghosts, so filter absent
2288
# records. The Sink is responsible for doing another check to
2289
# ensure that ghosts don't introduce missing data for future
2291
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
2292
self.to_format._fetch_order, True))
2293
yield substream_kind, stream
2295
def inventory_fetch_order(self):
2296
if self._rich_root_upgrade():
2297
return 'topological'
2299
return self.to_format._fetch_order
2301
def _rich_root_upgrade(self):
2302
return (not self.from_repository._format.rich_root_data and
2303
self.to_format.rich_root_data)
2305
def _get_inventory_stream(self, revision_ids, missing=False):
2306
from_format = self.from_repository._format
2307
if (from_format.supports_chks and self.to_format.supports_chks and
2308
from_format.network_name() == self.to_format.network_name()):
2309
raise AssertionError(
2310
"this case should be handled by GroupCHKStreamSource")
2311
elif 'forceinvdeltas' in debug.debug_flags:
2312
return self._get_convertable_inventory_stream(revision_ids,
2313
delta_versus_null=missing)
2314
elif from_format.network_name() == self.to_format.network_name():
2316
return self._get_simple_inventory_stream(revision_ids,
2318
elif (not from_format.supports_chks and not self.to_format.supports_chks
2319
and from_format._serializer == self.to_format._serializer):
2320
# Essentially the same format.
2321
return self._get_simple_inventory_stream(revision_ids,
2324
# Any time we switch serializations, we want to use an
2325
# inventory-delta based approach.
2326
return self._get_convertable_inventory_stream(revision_ids,
2327
delta_versus_null=missing)
2329
def _get_simple_inventory_stream(self, revision_ids, missing=False):
2330
# NB: This currently reopens the inventory weave in source;
2331
# using a single stream interface instead would avoid this.
2332
from_weave = self.from_repository.inventories
2334
delta_closure = True
2336
delta_closure = not self.delta_on_metadata()
2337
yield ('inventories', from_weave.get_record_stream(
2338
[(rev_id,) for rev_id in revision_ids],
2339
self.inventory_fetch_order(), delta_closure))
2341
def _get_convertable_inventory_stream(self, revision_ids,
2342
delta_versus_null=False):
2343
# The two formats are sufficiently different that there is no fast
2344
# path, so we need to send just inventorydeltas, which any
2345
# sufficiently modern client can insert into any repository.
2346
# The StreamSink code expects to be able to
2347
# convert on the target, so we need to put bytes-on-the-wire that can
2348
# be converted. That means inventory deltas (if the remote is <1.19,
2349
# RemoteStreamSink will fallback to VFS to insert the deltas).
2350
yield ('inventory-deltas',
2351
self._stream_invs_as_deltas(revision_ids,
2352
delta_versus_null=delta_versus_null))
2354
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
2355
"""Return a stream of inventory-deltas for the given rev ids.
2357
:param revision_ids: The list of inventories to transmit
2358
:param delta_versus_null: Don't try to find a minimal delta for this
2359
entry, instead compute the delta versus the NULL_REVISION. This
2360
effectively streams a complete inventory. Used for stuff like
2361
filling in missing parents, etc.
2363
from_repo = self.from_repository
2364
revision_keys = [(rev_id,) for rev_id in revision_ids]
2365
parent_map = from_repo.inventories.get_parent_map(revision_keys)
2366
# XXX: possibly repos could implement a more efficient iter_inv_deltas
2368
inventories = self.from_repository.iter_inventories(
2369
revision_ids, 'topological')
2370
format = from_repo._format
2371
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
2372
inventory_cache = lru_cache.LRUCache(50)
2373
null_inventory = from_repo.revision_tree(
2374
_mod_revision.NULL_REVISION).inventory
2375
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
2376
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
2377
# repo back into a non-rich-root repo ought to be allowed)
2378
serializer = inventory_delta.InventoryDeltaSerializer(
2379
versioned_root=format.rich_root_data,
2380
tree_references=format.supports_tree_reference)
2381
for inv in inventories:
2382
key = (inv.revision_id,)
2383
parent_keys = parent_map.get(key, ())
2385
if not delta_versus_null and parent_keys:
2386
# The caller did not ask for complete inventories and we have
2387
# some parents that we can delta against. Make a delta against
2388
# each parent so that we can find the smallest.
2389
parent_ids = [parent_key[0] for parent_key in parent_keys]
2390
for parent_id in parent_ids:
2391
if parent_id not in invs_sent_so_far:
2392
# We don't know that the remote side has this basis, so
2395
if parent_id == _mod_revision.NULL_REVISION:
2396
parent_inv = null_inventory
2398
parent_inv = inventory_cache.get(parent_id, None)
2399
if parent_inv is None:
2400
parent_inv = from_repo.get_inventory(parent_id)
2401
candidate_delta = inv._make_delta(parent_inv)
2402
if (delta is None or
2403
len(delta) > len(candidate_delta)):
2404
delta = candidate_delta
2405
basis_id = parent_id
2407
# Either none of the parents ended up being suitable, or we
2408
# were asked to delta against NULL
2409
basis_id = _mod_revision.NULL_REVISION
2410
delta = inv._make_delta(null_inventory)
2411
invs_sent_so_far.add(inv.revision_id)
2412
inventory_cache[inv.revision_id] = inv
2413
delta_serialized = ''.join(
2414
serializer.delta_to_lines(basis_id, key[-1], delta))
2415
yield versionedfile.FulltextContentFactory(
2416
key, parent_keys, None, delta_serialized)
2419
class _VersionedFileChecker(object):
2421
def __init__(self, repository, text_key_references=None, ancestors=None):
2422
self.repository = repository
2423
self.text_index = self.repository._generate_text_key_index(
2424
text_key_references=text_key_references, ancestors=ancestors)
2426
def calculate_file_version_parents(self, text_key):
2427
"""Calculate the correct parents for a file version according to
2430
parent_keys = self.text_index[text_key]
2431
if parent_keys == [_mod_revision.NULL_REVISION]:
2433
return tuple(parent_keys)
2435
def check_file_version_parents(self, texts, progress_bar=None):
2436
"""Check the parents stored in a versioned file are correct.
2438
It also detects file versions that are not referenced by their
2439
corresponding revision's inventory.
2441
:returns: A tuple of (wrong_parents, dangling_file_versions).
2442
wrong_parents is a dict mapping {revision_id: (stored_parents,
2443
correct_parents)} for each revision_id where the stored parents
2444
are not correct. dangling_file_versions is a set of (file_id,
2445
revision_id) tuples for versions that are present in this versioned
2446
file, but not used by the corresponding inventory.
2448
local_progress = None
2449
if progress_bar is None:
2450
local_progress = ui.ui_factory.nested_progress_bar()
2451
progress_bar = local_progress
2453
return self._check_file_version_parents(texts, progress_bar)
2456
local_progress.finished()
2458
def _check_file_version_parents(self, texts, progress_bar):
2459
"""See check_file_version_parents."""
2461
self.file_ids = set([file_id for file_id, _ in
2462
self.text_index.iterkeys()])
2463
# text keys is now grouped by file_id
2464
n_versions = len(self.text_index)
2465
progress_bar.update('loading text store', 0, n_versions)
2466
parent_map = self.repository.texts.get_parent_map(self.text_index)
2467
# On unlistable transports this could well be empty/error...
2468
text_keys = self.repository.texts.keys()
2469
unused_keys = frozenset(text_keys) - set(self.text_index)
2470
for num, key in enumerate(self.text_index.iterkeys()):
2471
progress_bar.update('checking text graph', num, n_versions)
2472
correct_parents = self.calculate_file_version_parents(key)
2474
knit_parents = parent_map[key]
2475
except errors.RevisionNotPresent:
2478
if correct_parents != knit_parents:
2479
wrong_parents[key] = (knit_parents, correct_parents)
2480
return wrong_parents, unused_keys
2483
class InterVersionedFileRepository(InterRepository):
2485
_walk_to_common_revisions_batch_size = 50
2488
def fetch(self, revision_id=None, find_ghosts=False,
2490
"""Fetch the content required to construct revision_id.
2492
The content is copied from self.source to self.target.
2494
:param revision_id: if None all content is copied, if NULL_REVISION no
2498
ui.ui_factory.warn_experimental_format_fetch(self)
2499
from bzrlib.fetch import RepoFetcher
2500
# See <https://launchpad.net/bugs/456077> asking for a warning here
2501
if self.source._format.network_name() != self.target._format.network_name():
2502
ui.ui_factory.show_user_warning('cross_format_fetch',
2503
from_format=self.source._format,
2504
to_format=self.target._format)
2505
f = RepoFetcher(to_repository=self.target,
2506
from_repository=self.source,
2507
last_revision=revision_id,
2508
fetch_spec=fetch_spec,
2509
find_ghosts=find_ghosts)
2511
def _walk_to_common_revisions(self, revision_ids, if_present_ids=None):
2512
"""Walk out from revision_ids in source to revisions target has.
2514
:param revision_ids: The start point for the search.
2515
:return: A set of revision ids.
2517
target_graph = self.target.get_graph()
2518
revision_ids = frozenset(revision_ids)
2520
all_wanted_revs = revision_ids.union(if_present_ids)
2522
all_wanted_revs = revision_ids
2523
missing_revs = set()
2524
source_graph = self.source.get_graph()
2525
# ensure we don't pay silly lookup costs.
2526
searcher = source_graph._make_breadth_first_searcher(all_wanted_revs)
2527
null_set = frozenset([_mod_revision.NULL_REVISION])
2528
searcher_exhausted = False
2532
# Iterate the searcher until we have enough next_revs
2533
while len(next_revs) < self._walk_to_common_revisions_batch_size:
2535
next_revs_part, ghosts_part = searcher.next_with_ghosts()
2536
next_revs.update(next_revs_part)
2537
ghosts.update(ghosts_part)
2538
except StopIteration:
2539
searcher_exhausted = True
2541
# If there are ghosts in the source graph, and the caller asked for
2542
# them, make sure that they are present in the target.
2543
# We don't care about other ghosts as we can't fetch them and
2544
# haven't been asked to.
2545
ghosts_to_check = set(revision_ids.intersection(ghosts))
2546
revs_to_get = set(next_revs).union(ghosts_to_check)
2548
have_revs = set(target_graph.get_parent_map(revs_to_get))
2549
# we always have NULL_REVISION present.
2550
have_revs = have_revs.union(null_set)
2551
# Check if the target is missing any ghosts we need.
2552
ghosts_to_check.difference_update(have_revs)
2554
# One of the caller's revision_ids is a ghost in both the
2555
# source and the target.
2556
raise errors.NoSuchRevision(
2557
self.source, ghosts_to_check.pop())
2558
missing_revs.update(next_revs - have_revs)
2559
# Because we may have walked past the original stop point, make
2560
# sure everything is stopped
2561
stop_revs = searcher.find_seen_ancestors(have_revs)
2562
searcher.stop_searching_any(stop_revs)
2563
if searcher_exhausted:
2565
return searcher.get_result()
2568
def search_missing_revision_ids(self,
2569
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
2570
find_ghosts=True, revision_ids=None, if_present_ids=None,
2572
"""Return the revision ids that source has that target does not.
2574
:param revision_id: only return revision ids included by this
2576
:param revision_ids: return revision ids included by these
2577
revision_ids. NoSuchRevision will be raised if any of these
2578
revisions are not present.
2579
:param if_present_ids: like revision_ids, but will not cause
2580
NoSuchRevision if any of these are absent, instead they will simply
2581
not be in the result. This is useful for e.g. finding revisions
2582
to fetch for tags, which may reference absent revisions.
2583
:param find_ghosts: If True find missing revisions in deep history
2584
rather than just finding the surface difference.
2585
:return: A bzrlib.graph.SearchResult.
2587
if symbol_versioning.deprecated_passed(revision_id):
2588
symbol_versioning.warn(
2589
'search_missing_revision_ids(revision_id=...) was '
2590
'deprecated in 2.4. Use revision_ids=[...] instead.',
2591
DeprecationWarning, stacklevel=2)
2592
if revision_ids is not None:
2593
raise AssertionError(
2594
'revision_ids is mutually exclusive with revision_id')
2595
if revision_id is not None:
2596
revision_ids = [revision_id]
2598
# stop searching at found target revisions.
2599
if not find_ghosts and (revision_ids is not None or if_present_ids is
2601
result = self._walk_to_common_revisions(revision_ids,
2602
if_present_ids=if_present_ids)
2605
result_set = result.get_keys()
2607
# generic, possibly worst case, slow code path.
2608
target_ids = set(self.target.all_revision_ids())
2609
source_ids = self._present_source_revisions_for(
2610
revision_ids, if_present_ids)
2611
result_set = set(source_ids).difference(target_ids)
2612
if limit is not None:
2613
topo_ordered = self.source.get_graph().iter_topo_order(result_set)
2614
result_set = set(itertools.islice(topo_ordered, limit))
2615
return self.source.revision_ids_to_search_result(result_set)
2617
def _present_source_revisions_for(self, revision_ids, if_present_ids=None):
2618
"""Returns set of all revisions in ancestry of revision_ids present in
2621
:param revision_ids: if None, all revisions in source are returned.
2622
:param if_present_ids: like revision_ids, but if any/all of these are
2623
absent no error is raised.
2625
if revision_ids is not None or if_present_ids is not None:
2626
# First, ensure all specified revisions exist. Callers expect
2627
# NoSuchRevision when they pass absent revision_ids here.
2628
if revision_ids is None:
2629
revision_ids = set()
2630
if if_present_ids is None:
2631
if_present_ids = set()
2632
revision_ids = set(revision_ids)
2633
if_present_ids = set(if_present_ids)
2634
all_wanted_ids = revision_ids.union(if_present_ids)
2635
graph = self.source.get_graph()
2636
present_revs = set(graph.get_parent_map(all_wanted_ids))
2637
missing = revision_ids.difference(present_revs)
2639
raise errors.NoSuchRevision(self.source, missing.pop())
2640
found_ids = all_wanted_ids.intersection(present_revs)
2641
source_ids = [rev_id for (rev_id, parents) in
2642
graph.iter_ancestry(found_ids)
2643
if rev_id != _mod_revision.NULL_REVISION
2644
and parents is not None]
2646
source_ids = self.source.all_revision_ids()
2647
return set(source_ids)
2650
def _get_repo_format_to_test(self):
2654
def is_compatible(cls, source, target):
2655
# The default implementation is compatible with everything
2656
return (source._format.supports_full_versioned_files and
2657
target._format.supports_full_versioned_files)
2660
class InterDifferingSerializer(InterVersionedFileRepository):
2663
def _get_repo_format_to_test(self):
2667
def is_compatible(source, target):
2668
if not source._format.supports_full_versioned_files:
2670
if not target._format.supports_full_versioned_files:
2672
# This is redundant with format.check_conversion_target(), however that
2673
# raises an exception, and we just want to say "False" as in we won't
2674
# support converting between these formats.
2675
if 'IDS_never' in debug.debug_flags:
2677
if source.supports_rich_root() and not target.supports_rich_root():
2679
if (source._format.supports_tree_reference
2680
and not target._format.supports_tree_reference):
2682
if target._fallback_repositories and target._format.supports_chks:
2683
# IDS doesn't know how to copy CHKs for the parent inventories it
2684
# adds to stacked repos.
2686
if 'IDS_always' in debug.debug_flags:
2688
# Only use this code path for local source and target. IDS does far
2689
# too much IO (both bandwidth and roundtrips) over a network.
2690
if not source.bzrdir.transport.base.startswith('file:///'):
2692
if not target.bzrdir.transport.base.startswith('file:///'):
2696
def _get_trees(self, revision_ids, cache):
2698
for rev_id in revision_ids:
2700
possible_trees.append((rev_id, cache[rev_id]))
2702
# Not cached, but inventory might be present anyway.
2704
tree = self.source.revision_tree(rev_id)
2705
except errors.NoSuchRevision:
2706
# Nope, parent is ghost.
2709
cache[rev_id] = tree
2710
possible_trees.append((rev_id, tree))
2711
return possible_trees
2713
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
2714
"""Get the best delta and base for this revision.
2716
:return: (basis_id, delta)
2719
# Generate deltas against each tree, to find the shortest.
2720
texts_possibly_new_in_tree = set()
2721
for basis_id, basis_tree in possible_trees:
2722
delta = tree.inventory._make_delta(basis_tree.inventory)
2723
for old_path, new_path, file_id, new_entry in delta:
2724
if new_path is None:
2725
# This file_id isn't present in the new rev, so we don't
2729
# Rich roots are handled elsewhere...
2731
kind = new_entry.kind
2732
if kind != 'directory' and kind != 'file':
2733
# No text record associated with this inventory entry.
2735
# This is a directory or file that has changed somehow.
2736
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
2737
deltas.append((len(delta), basis_id, delta))
2739
return deltas[0][1:]
2741
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
2742
"""Find all parent revisions that are absent, but for which the
2743
inventory is present, and copy those inventories.
2745
This is necessary to preserve correctness when the source is stacked
2746
without fallbacks configured. (Note that in cases like upgrade the
2747
source may be not have _fallback_repositories even though it is
2751
for parents in parent_map.values():
2752
parent_revs.update(parents)
2753
present_parents = self.source.get_parent_map(parent_revs)
2754
absent_parents = set(parent_revs).difference(present_parents)
2755
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
2756
(rev_id,) for rev_id in absent_parents)
2757
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
2758
for parent_tree in self.source.revision_trees(parent_inv_ids):
2759
current_revision_id = parent_tree.get_revision_id()
2760
parents_parents_keys = parent_invs_keys_for_stacking[
2761
(current_revision_id,)]
2762
parents_parents = [key[-1] for key in parents_parents_keys]
2763
basis_id = _mod_revision.NULL_REVISION
2764
basis_tree = self.source.revision_tree(basis_id)
2765
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
2766
self.target.add_inventory_by_delta(
2767
basis_id, delta, current_revision_id, parents_parents)
2768
cache[current_revision_id] = parent_tree
2770
def _fetch_batch(self, revision_ids, basis_id, cache):
2771
"""Fetch across a few revisions.
2773
:param revision_ids: The revisions to copy
2774
:param basis_id: The revision_id of a tree that must be in cache, used
2775
as a basis for delta when no other base is available
2776
:param cache: A cache of RevisionTrees that we can use.
2777
:return: The revision_id of the last converted tree. The RevisionTree
2778
for it will be in cache
2780
# Walk though all revisions; get inventory deltas, copy referenced
2781
# texts that delta references, insert the delta, revision and
2783
root_keys_to_create = set()
2786
pending_revisions = []
2787
parent_map = self.source.get_parent_map(revision_ids)
2788
self._fetch_parent_invs_for_stacking(parent_map, cache)
2789
self.source._safe_to_return_from_cache = True
2790
for tree in self.source.revision_trees(revision_ids):
2791
# Find a inventory delta for this revision.
2792
# Find text entries that need to be copied, too.
2793
current_revision_id = tree.get_revision_id()
2794
parent_ids = parent_map.get(current_revision_id, ())
2795
parent_trees = self._get_trees(parent_ids, cache)
2796
possible_trees = list(parent_trees)
2797
if len(possible_trees) == 0:
2798
# There either aren't any parents, or the parents are ghosts,
2799
# so just use the last converted tree.
2800
possible_trees.append((basis_id, cache[basis_id]))
2801
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
2803
revision = self.source.get_revision(current_revision_id)
2804
pending_deltas.append((basis_id, delta,
2805
current_revision_id, revision.parent_ids))
2806
if self._converting_to_rich_root:
2807
self._revision_id_to_root_id[current_revision_id] = \
2809
# Determine which texts are in present in this revision but not in
2810
# any of the available parents.
2811
texts_possibly_new_in_tree = set()
2812
for old_path, new_path, file_id, entry in delta:
2813
if new_path is None:
2814
# This file_id isn't present in the new rev
2818
if not self.target.supports_rich_root():
2819
# The target doesn't support rich root, so we don't
2822
if self._converting_to_rich_root:
2823
# This can't be copied normally, we have to insert
2825
root_keys_to_create.add((file_id, entry.revision))
2828
texts_possibly_new_in_tree.add((file_id, entry.revision))
2829
for basis_id, basis_tree in possible_trees:
2830
basis_inv = basis_tree.inventory
2831
for file_key in list(texts_possibly_new_in_tree):
2832
file_id, file_revision = file_key
2834
entry = basis_inv[file_id]
2835
except errors.NoSuchId:
2837
if entry.revision == file_revision:
2838
texts_possibly_new_in_tree.remove(file_key)
2839
text_keys.update(texts_possibly_new_in_tree)
2840
pending_revisions.append(revision)
2841
cache[current_revision_id] = tree
2842
basis_id = current_revision_id
2843
self.source._safe_to_return_from_cache = False
2845
from_texts = self.source.texts
2846
to_texts = self.target.texts
2847
if root_keys_to_create:
2848
root_stream = _mod_fetch._new_root_data_stream(
2849
root_keys_to_create, self._revision_id_to_root_id, parent_map,
2851
to_texts.insert_record_stream(root_stream)
2852
to_texts.insert_record_stream(from_texts.get_record_stream(
2853
text_keys, self.target._format._fetch_order,
2854
not self.target._format._fetch_uses_deltas))
2855
# insert inventory deltas
2856
for delta in pending_deltas:
2857
self.target.add_inventory_by_delta(*delta)
2858
if self.target._fallback_repositories:
2859
# Make sure this stacked repository has all the parent inventories
2860
# for the new revisions that we are about to insert. We do this
2861
# before adding the revisions so that no revision is added until
2862
# all the inventories it may depend on are added.
2863
# Note that this is overzealous, as we may have fetched these in an
2866
revision_ids = set()
2867
for revision in pending_revisions:
2868
revision_ids.add(revision.revision_id)
2869
parent_ids.update(revision.parent_ids)
2870
parent_ids.difference_update(revision_ids)
2871
parent_ids.discard(_mod_revision.NULL_REVISION)
2872
parent_map = self.source.get_parent_map(parent_ids)
2873
# we iterate over parent_map and not parent_ids because we don't
2874
# want to try copying any revision which is a ghost
2875
for parent_tree in self.source.revision_trees(parent_map):
2876
current_revision_id = parent_tree.get_revision_id()
2877
parents_parents = parent_map[current_revision_id]
2878
possible_trees = self._get_trees(parents_parents, cache)
2879
if len(possible_trees) == 0:
2880
# There either aren't any parents, or the parents are
2881
# ghosts, so just use the last converted tree.
2882
possible_trees.append((basis_id, cache[basis_id]))
2883
basis_id, delta = self._get_delta_for_revision(parent_tree,
2884
parents_parents, possible_trees)
2885
self.target.add_inventory_by_delta(
2886
basis_id, delta, current_revision_id, parents_parents)
2887
# insert signatures and revisions
2888
for revision in pending_revisions:
2890
signature = self.source.get_signature_text(
2891
revision.revision_id)
2892
self.target.add_signature_text(revision.revision_id,
2894
except errors.NoSuchRevision:
2896
self.target.add_revision(revision.revision_id, revision)
2899
def _fetch_all_revisions(self, revision_ids, pb):
2900
"""Fetch everything for the list of revisions.
2902
:param revision_ids: The list of revisions to fetch. Must be in
2904
:param pb: A ProgressTask
2907
basis_id, basis_tree = self._get_basis(revision_ids[0])
2909
cache = lru_cache.LRUCache(100)
2910
cache[basis_id] = basis_tree
2911
del basis_tree # We don't want to hang on to it here
2915
for offset in range(0, len(revision_ids), batch_size):
2916
self.target.start_write_group()
2918
pb.update('Transferring revisions', offset,
2920
batch = revision_ids[offset:offset+batch_size]
2921
basis_id = self._fetch_batch(batch, basis_id, cache)
2923
self.source._safe_to_return_from_cache = False
2924
self.target.abort_write_group()
2927
hint = self.target.commit_write_group()
2930
if hints and self.target._format.pack_compresses:
2931
self.target.pack(hint=hints)
2932
pb.update('Transferring revisions', len(revision_ids),
2936
def fetch(self, revision_id=None, find_ghosts=False,
2938
"""See InterRepository.fetch()."""
2939
if fetch_spec is not None:
2940
revision_ids = fetch_spec.get_keys()
2943
ui.ui_factory.warn_experimental_format_fetch(self)
2944
if (not self.source.supports_rich_root()
2945
and self.target.supports_rich_root()):
2946
self._converting_to_rich_root = True
2947
self._revision_id_to_root_id = {}
2949
self._converting_to_rich_root = False
2950
# See <https://launchpad.net/bugs/456077> asking for a warning here
2951
if self.source._format.network_name() != self.target._format.network_name():
2952
ui.ui_factory.show_user_warning('cross_format_fetch',
2953
from_format=self.source._format,
2954
to_format=self.target._format)
2955
if revision_ids is None:
2957
search_revision_ids = [revision_id]
2959
search_revision_ids = None
2960
revision_ids = self.target.search_missing_revision_ids(self.source,
2961
revision_ids=search_revision_ids,
2962
find_ghosts=find_ghosts).get_keys()
2963
if not revision_ids:
2965
revision_ids = tsort.topo_sort(
2966
self.source.get_graph().get_parent_map(revision_ids))
2967
if not revision_ids:
2969
# Walk though all revisions; get inventory deltas, copy referenced
2970
# texts that delta references, insert the delta, revision and
2972
pb = ui.ui_factory.nested_progress_bar()
2974
self._fetch_all_revisions(revision_ids, pb)
2977
return len(revision_ids), 0
2979
def _get_basis(self, first_revision_id):
2980
"""Get a revision and tree which exists in the target.
2982
This assumes that first_revision_id is selected for transmission
2983
because all other ancestors are already present. If we can't find an
2984
ancestor we fall back to NULL_REVISION since we know that is safe.
2986
:return: (basis_id, basis_tree)
2988
first_rev = self.source.get_revision(first_revision_id)
2990
basis_id = first_rev.parent_ids[0]
2991
# only valid as a basis if the target has it
2992
self.target.get_revision(basis_id)
2993
# Try to get a basis tree - if it's a ghost it will hit the
2994
# NoSuchRevision case.
2995
basis_tree = self.source.revision_tree(basis_id)
2996
except (IndexError, errors.NoSuchRevision):
2997
basis_id = _mod_revision.NULL_REVISION
2998
basis_tree = self.source.revision_tree(basis_id)
2999
return basis_id, basis_tree
3002
class InterSameDataRepository(InterVersionedFileRepository):
3003
"""Code for converting between repositories that represent the same data.
3005
Data format and model must match for this to work.
3009
def _get_repo_format_to_test(self):
3010
"""Repository format for testing with.
3012
InterSameData can pull from subtree to subtree and from non-subtree to
3013
non-subtree, so we test this with the richest repository format.
3015
from bzrlib.repofmt import knitrepo
3016
return knitrepo.RepositoryFormatKnit3()
3019
def is_compatible(source, target):
3021
InterRepository._same_model(source, target) and
3022
source._format.supports_full_versioned_files and
3023
target._format.supports_full_versioned_files)
3026
InterRepository.register_optimiser(InterVersionedFileRepository)
3027
InterRepository.register_optimiser(InterDifferingSerializer)
3028
InterRepository.register_optimiser(InterSameDataRepository)
3031
def install_revisions(repository, iterable, num_revisions=None, pb=None):
3032
"""Install all revision data into a repository.
3034
Accepts an iterable of revision, tree, signature tuples. The signature
3037
repository.start_write_group()
3039
inventory_cache = lru_cache.LRUCache(10)
3040
for n, (revision, revision_tree, signature) in enumerate(iterable):
3041
_install_revision(repository, revision, revision_tree, signature,
3044
pb.update('Transferring revisions', n + 1, num_revisions)
3046
repository.abort_write_group()
3049
repository.commit_write_group()
3052
def _install_revision(repository, rev, revision_tree, signature,
3054
"""Install all revision data into a repository."""
3055
present_parents = []
3057
for p_id in rev.parent_ids:
3058
if repository.has_revision(p_id):
3059
present_parents.append(p_id)
3060
parent_trees[p_id] = repository.revision_tree(p_id)
3062
parent_trees[p_id] = repository.revision_tree(
3063
_mod_revision.NULL_REVISION)
3065
inv = revision_tree.inventory
3066
entries = inv.iter_entries()
3067
# backwards compatibility hack: skip the root id.
3068
if not repository.supports_rich_root():
3069
path, root = entries.next()
3070
if root.revision != rev.revision_id:
3071
raise errors.IncompatibleRevision(repr(repository))
3073
for path, ie in entries:
3074
text_keys[(ie.file_id, ie.revision)] = ie
3075
text_parent_map = repository.texts.get_parent_map(text_keys)
3076
missing_texts = set(text_keys) - set(text_parent_map)
3077
# Add the texts that are not already present
3078
for text_key in missing_texts:
3079
ie = text_keys[text_key]
3081
# FIXME: TODO: The following loop overlaps/duplicates that done by
3082
# commit to determine parents. There is a latent/real bug here where
3083
# the parents inserted are not those commit would do - in particular
3084
# they are not filtered by heads(). RBC, AB
3085
for revision, tree in parent_trees.iteritems():
3086
if not tree.has_id(ie.file_id):
3088
parent_id = tree.get_file_revision(ie.file_id)
3089
if parent_id in text_parents:
3091
text_parents.append((ie.file_id, parent_id))
3092
lines = revision_tree.get_file(ie.file_id).readlines()
3093
repository.texts.add_lines(text_key, text_parents, lines)
3095
# install the inventory
3096
if repository._format._commit_inv_deltas and len(rev.parent_ids):
3097
# Cache this inventory
3098
inventory_cache[rev.revision_id] = inv
3100
basis_inv = inventory_cache[rev.parent_ids[0]]
3102
repository.add_inventory(rev.revision_id, inv, present_parents)
3104
delta = inv._make_delta(basis_inv)
3105
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
3106
rev.revision_id, present_parents)
3108
repository.add_inventory(rev.revision_id, inv, present_parents)
3109
except errors.RevisionAlreadyPresent:
3111
if signature is not None:
3112
repository.add_signature_text(rev.revision_id, signature)
3113
repository.add_revision(rev.revision_id, rev, inv)
3116
def install_revision(repository, rev, revision_tree):
3117
"""Install all revision data into a repository."""
3118
install_revisions(repository, [(rev, revision_tree, None)])
added
removed
bzrlib/vf_repository.py
