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_candidate_entries = ie.parent_candidates(parent_invs)
423
head_set = self._heads(ie.file_id, parent_candidate_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_candidate_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 before 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)
921
# This can raise an exception, so should be done before we lock the
922
# fallback repository.
923
self._check_fallback_repository(repository)
925
# This repository will call fallback.unlock() when we transition to
926
# the unlocked state, so we make sure to increment the lock count
927
repository.lock_read()
928
self._fallback_repositories.append(repository)
929
self.texts.add_fallback_versioned_files(repository.texts)
930
self.inventories.add_fallback_versioned_files(repository.inventories)
931
self.revisions.add_fallback_versioned_files(repository.revisions)
932
self.signatures.add_fallback_versioned_files(repository.signatures)
933
if self.chk_bytes is not None:
934
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
936
@only_raises(errors.LockNotHeld, errors.LockBroken)
938
super(VersionedFileRepository, self).unlock()
939
if self.control_files._lock_count == 0:
940
self._inventory_entry_cache.clear()
942
def add_inventory(self, revision_id, inv, parents):
943
"""Add the inventory inv to the repository as revision_id.
945
:param parents: The revision ids of the parents that revision_id
946
is known to have and are in the repository already.
948
:returns: The validator(which is a sha1 digest, though what is sha'd is
949
repository format specific) of the serialized inventory.
951
if not self.is_in_write_group():
952
raise AssertionError("%r not in write group" % (self,))
953
_mod_revision.check_not_reserved_id(revision_id)
954
if not (inv.revision_id is None or inv.revision_id == revision_id):
955
raise AssertionError(
956
"Mismatch between inventory revision"
957
" id and insertion revid (%r, %r)"
958
% (inv.revision_id, revision_id))
960
raise errors.RootMissing()
961
return self._add_inventory_checked(revision_id, inv, parents)
963
def _add_inventory_checked(self, revision_id, inv, parents):
964
"""Add inv to the repository after checking the inputs.
966
This function can be overridden to allow different inventory styles.
968
:seealso: add_inventory, for the contract.
970
inv_lines = self._serializer.write_inventory_to_lines(inv)
971
return self._inventory_add_lines(revision_id, parents,
972
inv_lines, check_content=False)
974
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
975
parents, basis_inv=None, propagate_caches=False):
976
"""Add a new inventory expressed as a delta against another revision.
978
See the inventory developers documentation for the theory behind
981
:param basis_revision_id: The inventory id the delta was created
982
against. (This does not have to be a direct parent.)
983
:param delta: The inventory delta (see Inventory.apply_delta for
985
:param new_revision_id: The revision id that the inventory is being
987
:param parents: The revision ids of the parents that revision_id is
988
known to have and are in the repository already. These are supplied
989
for repositories that depend on the inventory graph for revision
990
graph access, as well as for those that pun ancestry with delta
992
:param basis_inv: The basis inventory if it is already known,
994
:param propagate_caches: If True, the caches for this inventory are
995
copied to and updated for the result if possible.
997
:returns: (validator, new_inv)
998
The validator(which is a sha1 digest, though what is sha'd is
999
repository format specific) of the serialized inventory, and the
1000
resulting inventory.
1002
if not self.is_in_write_group():
1003
raise AssertionError("%r not in write group" % (self,))
1004
_mod_revision.check_not_reserved_id(new_revision_id)
1005
basis_tree = self.revision_tree(basis_revision_id)
1006
basis_tree.lock_read()
1008
# Note that this mutates the inventory of basis_tree, which not all
1009
# inventory implementations may support: A better idiom would be to
1010
# return a new inventory, but as there is no revision tree cache in
1011
# repository this is safe for now - RBC 20081013
1012
if basis_inv is None:
1013
basis_inv = basis_tree.inventory
1014
basis_inv.apply_delta(delta)
1015
basis_inv.revision_id = new_revision_id
1016
return (self.add_inventory(new_revision_id, basis_inv, parents),
1021
def _inventory_add_lines(self, revision_id, parents, lines,
1022
check_content=True):
1023
"""Store lines in inv_vf and return the sha1 of the inventory."""
1024
parents = [(parent,) for parent in parents]
1025
result = self.inventories.add_lines((revision_id,), parents, lines,
1026
check_content=check_content)[0]
1027
self.inventories._access.flush()
1030
def add_revision(self, revision_id, rev, inv=None, config=None):
1031
"""Add rev to the revision store as revision_id.
1033
:param revision_id: the revision id to use.
1034
:param rev: The revision object.
1035
:param inv: The inventory for the revision. if None, it will be looked
1036
up in the inventory storer
1037
:param config: If None no digital signature will be created.
1038
If supplied its signature_needed method will be used
1039
to determine if a signature should be made.
1041
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1043
_mod_revision.check_not_reserved_id(revision_id)
1044
if config is not None and config.signature_needed():
1046
inv = self.get_inventory(revision_id)
1047
tree = InventoryRevisionTree(self, inv, revision_id)
1048
testament = Testament(rev, tree)
1049
plaintext = testament.as_short_text()
1050
self.store_revision_signature(
1051
gpg.GPGStrategy(config), plaintext, revision_id)
1052
# check inventory present
1053
if not self.inventories.get_parent_map([(revision_id,)]):
1055
raise errors.WeaveRevisionNotPresent(revision_id,
1058
# yes, this is not suitable for adding with ghosts.
1059
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1062
key = (revision_id,)
1063
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1064
self._add_revision(rev)
1066
def _add_revision(self, revision):
1067
text = self._serializer.write_revision_to_string(revision)
1068
key = (revision.revision_id,)
1069
parents = tuple((parent,) for parent in revision.parent_ids)
1070
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1072
def _check_inventories(self, checker):
1073
"""Check the inventories found from the revision scan.
1075
This is responsible for verifying the sha1 of inventories and
1076
creating a pending_keys set that covers data referenced by inventories.
1078
bar = ui.ui_factory.nested_progress_bar()
1080
self._do_check_inventories(checker, bar)
1084
def _do_check_inventories(self, checker, bar):
1085
"""Helper for _check_inventories."""
1087
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1088
kinds = ['chk_bytes', 'texts']
1089
count = len(checker.pending_keys)
1090
bar.update("inventories", 0, 2)
1091
current_keys = checker.pending_keys
1092
checker.pending_keys = {}
1093
# Accumulate current checks.
1094
for key in current_keys:
1095
if key[0] != 'inventories' and key[0] not in kinds:
1096
checker._report_items.append('unknown key type %r' % (key,))
1097
keys[key[0]].add(key[1:])
1098
if keys['inventories']:
1099
# NB: output order *should* be roughly sorted - topo or
1100
# inverse topo depending on repository - either way decent
1101
# to just delta against. However, pre-CHK formats didn't
1102
# try to optimise inventory layout on disk. As such the
1103
# pre-CHK code path does not use inventory deltas.
1105
for record in self.inventories.check(keys=keys['inventories']):
1106
if record.storage_kind == 'absent':
1107
checker._report_items.append(
1108
'Missing inventory {%s}' % (record.key,))
1110
last_object = self._check_record('inventories', record,
1111
checker, last_object,
1112
current_keys[('inventories',) + record.key])
1113
del keys['inventories']
1116
bar.update("texts", 1)
1117
while (checker.pending_keys or keys['chk_bytes']
1119
# Something to check.
1120
current_keys = checker.pending_keys
1121
checker.pending_keys = {}
1122
# Accumulate current checks.
1123
for key in current_keys:
1124
if key[0] not in kinds:
1125
checker._report_items.append('unknown key type %r' % (key,))
1126
keys[key[0]].add(key[1:])
1127
# Check the outermost kind only - inventories || chk_bytes || texts
1131
for record in getattr(self, kind).check(keys=keys[kind]):
1132
if record.storage_kind == 'absent':
1133
checker._report_items.append(
1134
'Missing %s {%s}' % (kind, record.key,))
1136
last_object = self._check_record(kind, record,
1137
checker, last_object, current_keys[(kind,) + record.key])
1141
def _check_record(self, kind, record, checker, last_object, item_data):
1142
"""Check a single text from this repository."""
1143
if kind == 'inventories':
1144
rev_id = record.key[0]
1145
inv = self._deserialise_inventory(rev_id,
1146
record.get_bytes_as('fulltext'))
1147
if last_object is not None:
1148
delta = inv._make_delta(last_object)
1149
for old_path, path, file_id, ie in delta:
1152
ie.check(checker, rev_id, inv)
1154
for path, ie in inv.iter_entries():
1155
ie.check(checker, rev_id, inv)
1156
if self._format.fast_deltas:
1158
elif kind == 'chk_bytes':
1159
# No code written to check chk_bytes for this repo format.
1160
checker._report_items.append(
1161
'unsupported key type chk_bytes for %s' % (record.key,))
1162
elif kind == 'texts':
1163
self._check_text(record, checker, item_data)
1165
checker._report_items.append(
1166
'unknown key type %s for %s' % (kind, record.key))
1168
def _check_text(self, record, checker, item_data):
1169
"""Check a single text."""
1170
# Check it is extractable.
1171
# TODO: check length.
1172
if record.storage_kind == 'chunked':
1173
chunks = record.get_bytes_as(record.storage_kind)
1174
sha1 = osutils.sha_strings(chunks)
1175
length = sum(map(len, chunks))
1177
content = record.get_bytes_as('fulltext')
1178
sha1 = osutils.sha_string(content)
1179
length = len(content)
1180
if item_data and sha1 != item_data[1]:
1181
checker._report_items.append(
1182
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1183
(record.key, sha1, item_data[1], item_data[2]))
1186
def _eliminate_revisions_not_present(self, revision_ids):
1187
"""Check every revision id in revision_ids to see if we have it.
1189
Returns a set of the present revisions.
1192
graph = self.get_graph()
1193
parent_map = graph.get_parent_map(revision_ids)
1194
# The old API returned a list, should this actually be a set?
1195
return parent_map.keys()
1197
def __init__(self, _format, a_bzrdir, control_files):
1198
"""Instantiate a VersionedFileRepository.
1200
:param _format: The format of the repository on disk.
1201
:param a_bzrdir: The BzrDir of the repository.
1202
:param control_files: Control files to use for locking, etc.
1204
# In the future we will have a single api for all stores for
1205
# getting file texts, inventories and revisions, then
1206
# this construct will accept instances of those things.
1207
super(VersionedFileRepository, self).__init__(_format, a_bzrdir,
1210
self._reconcile_does_inventory_gc = True
1211
self._reconcile_fixes_text_parents = False
1212
self._reconcile_backsup_inventory = True
1213
# An InventoryEntry cache, used during deserialization
1214
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1215
# Is it safe to return inventory entries directly from the entry cache,
1216
# rather copying them?
1217
self._safe_to_return_from_cache = False
1220
def gather_stats(self, revid=None, committers=None):
1221
"""See Repository.gather_stats()."""
1222
result = super(VersionedFileRepository, self).gather_stats(revid, committers)
1223
# now gather global repository information
1224
# XXX: This is available for many repos regardless of listability.
1225
if self.user_transport.listable():
1226
# XXX: do we want to __define len__() ?
1227
# Maybe the versionedfiles object should provide a different
1228
# method to get the number of keys.
1229
result['revisions'] = len(self.revisions.keys())
1230
# result['size'] = t
1233
def get_commit_builder(self, branch, parents, config, timestamp=None,
1234
timezone=None, committer=None, revprops=None,
1235
revision_id=None, lossy=False):
1236
"""Obtain a CommitBuilder for this repository.
1238
:param branch: Branch to commit to.
1239
:param parents: Revision ids of the parents of the new revision.
1240
:param config: Configuration to use.
1241
:param timestamp: Optional timestamp recorded for commit.
1242
:param timezone: Optional timezone for timestamp.
1243
:param committer: Optional committer to set for commit.
1244
:param revprops: Optional dictionary of revision properties.
1245
:param revision_id: Optional revision id.
1246
:param lossy: Whether to discard data that can not be natively
1247
represented, when pushing to a foreign VCS
1249
if self._fallback_repositories and not self._format.supports_chks:
1250
raise errors.BzrError("Cannot commit directly to a stacked branch"
1251
" in pre-2a formats. See "
1252
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1253
result = self._commit_builder_class(self, parents, config,
1254
timestamp, timezone, committer, revprops, revision_id,
1256
self.start_write_group()
1259
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1260
"""Return the keys of missing inventory parents for revisions added in
1263
A revision is not complete if the inventory delta for that revision
1264
cannot be calculated. Therefore if the parent inventories of a
1265
revision are not present, the revision is incomplete, and e.g. cannot
1266
be streamed by a smart server. This method finds missing inventory
1267
parents for revisions added in this write group.
1269
if not self._format.supports_external_lookups:
1270
# This is only an issue for stacked repositories
1272
if not self.is_in_write_group():
1273
raise AssertionError('not in a write group')
1275
# XXX: We assume that every added revision already has its
1276
# corresponding inventory, so we only check for parent inventories that
1277
# might be missing, rather than all inventories.
1278
parents = set(self.revisions._index.get_missing_parents())
1279
parents.discard(_mod_revision.NULL_REVISION)
1280
unstacked_inventories = self.inventories._index
1281
present_inventories = unstacked_inventories.get_parent_map(
1282
key[-1:] for key in parents)
1283
parents.difference_update(present_inventories)
1284
if len(parents) == 0:
1285
# No missing parent inventories.
1287
if not check_for_missing_texts:
1288
return set(('inventories', rev_id) for (rev_id,) in parents)
1289
# Ok, now we have a list of missing inventories. But these only matter
1290
# if the inventories that reference them are missing some texts they
1291
# appear to introduce.
1292
# XXX: Texts referenced by all added inventories need to be present,
1293
# but at the moment we're only checking for texts referenced by
1294
# inventories at the graph's edge.
1295
key_deps = self.revisions._index._key_dependencies
1296
key_deps.satisfy_refs_for_keys(present_inventories)
1297
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1298
file_ids = self.fileids_altered_by_revision_ids(referrers)
1299
missing_texts = set()
1300
for file_id, version_ids in file_ids.iteritems():
1301
missing_texts.update(
1302
(file_id, version_id) for version_id in version_ids)
1303
present_texts = self.texts.get_parent_map(missing_texts)
1304
missing_texts.difference_update(present_texts)
1305
if not missing_texts:
1306
# No texts are missing, so all revisions and their deltas are
1309
# Alternatively the text versions could be returned as the missing
1310
# keys, but this is likely to be less data.
1311
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1315
def has_revisions(self, revision_ids):
1316
"""Probe to find out the presence of multiple revisions.
1318
:param revision_ids: An iterable of revision_ids.
1319
:return: A set of the revision_ids that were present.
1321
parent_map = self.revisions.get_parent_map(
1322
[(rev_id,) for rev_id in revision_ids])
1324
if _mod_revision.NULL_REVISION in revision_ids:
1325
result.add(_mod_revision.NULL_REVISION)
1326
result.update([key[0] for key in parent_map])
1330
def get_revision_reconcile(self, revision_id):
1331
"""'reconcile' helper routine that allows access to a revision always.
1333
This variant of get_revision does not cross check the weave graph
1334
against the revision one as get_revision does: but it should only
1335
be used by reconcile, or reconcile-alike commands that are correcting
1336
or testing the revision graph.
1338
return self._get_revisions([revision_id])[0]
1341
def get_revisions(self, revision_ids):
1342
"""Get many revisions at once.
1344
Repositories that need to check data on every revision read should
1345
subclass this method.
1347
return self._get_revisions(revision_ids)
1350
def _get_revisions(self, revision_ids):
1351
"""Core work logic to get many revisions without sanity checks."""
1353
for revid, rev in self._iter_revisions(revision_ids):
1355
raise errors.NoSuchRevision(self, revid)
1357
return [revs[revid] for revid in revision_ids]
1359
def _iter_revisions(self, revision_ids):
1360
"""Iterate over revision objects.
1362
:param revision_ids: An iterable of revisions to examine. None may be
1363
passed to request all revisions known to the repository. Note that
1364
not all repositories can find unreferenced revisions; for those
1365
repositories only referenced ones will be returned.
1366
:return: An iterator of (revid, revision) tuples. Absent revisions (
1367
those asked for but not available) are returned as (revid, None).
1369
if revision_ids is None:
1370
revision_ids = self.all_revision_ids()
1372
for rev_id in revision_ids:
1373
if not rev_id or not isinstance(rev_id, basestring):
1374
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1375
keys = [(key,) for key in revision_ids]
1376
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1377
for record in stream:
1378
revid = record.key[0]
1379
if record.storage_kind == 'absent':
1382
text = record.get_bytes_as('fulltext')
1383
rev = self._serializer.read_revision_from_string(text)
1387
def add_signature_text(self, revision_id, signature):
1388
"""Store a signature text for a revision.
1390
:param revision_id: Revision id of the revision
1391
:param signature: Signature text.
1393
self.signatures.add_lines((revision_id,), (),
1394
osutils.split_lines(signature))
1396
def find_text_key_references(self):
1397
"""Find the text key references within the repository.
1399
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1400
to whether they were referred to by the inventory of the
1401
revision_id that they contain. The inventory texts from all present
1402
revision ids are assessed to generate this report.
1404
revision_keys = self.revisions.keys()
1405
w = self.inventories
1406
pb = ui.ui_factory.nested_progress_bar()
1408
return self._serializer._find_text_key_references(
1409
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1413
def _inventory_xml_lines_for_keys(self, keys):
1414
"""Get a line iterator of the sort needed for findind references.
1416
Not relevant for non-xml inventory repositories.
1418
Ghosts in revision_keys are ignored.
1420
:param revision_keys: The revision keys for the inventories to inspect.
1421
:return: An iterator over (inventory line, revid) for the fulltexts of
1422
all of the xml inventories specified by revision_keys.
1424
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1425
for record in stream:
1426
if record.storage_kind != 'absent':
1427
chunks = record.get_bytes_as('chunked')
1428
revid = record.key[-1]
1429
lines = osutils.chunks_to_lines(chunks)
1433
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1435
"""Helper routine for fileids_altered_by_revision_ids.
1437
This performs the translation of xml lines to revision ids.
1439
:param line_iterator: An iterator of lines, origin_version_id
1440
:param revision_keys: The revision ids to filter for. This should be a
1441
set or other type which supports efficient __contains__ lookups, as
1442
the revision key from each parsed line will be looked up in the
1443
revision_keys filter.
1444
:return: a dictionary mapping altered file-ids to an iterable of
1445
revision_ids. Each altered file-ids has the exact revision_ids that
1446
altered it listed explicitly.
1448
seen = set(self._serializer._find_text_key_references(
1449
line_iterator).iterkeys())
1450
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
1451
parent_seen = set(self._serializer._find_text_key_references(
1452
self._inventory_xml_lines_for_keys(parent_keys)))
1453
new_keys = seen - parent_seen
1455
setdefault = result.setdefault
1456
for key in new_keys:
1457
setdefault(key[0], set()).add(key[-1])
1460
def _find_parent_keys_of_revisions(self, revision_keys):
1461
"""Similar to _find_parent_ids_of_revisions, but used with keys.
1463
:param revision_keys: An iterable of revision_keys.
1464
:return: The parents of all revision_keys that are not already in
1467
parent_map = self.revisions.get_parent_map(revision_keys)
1469
map(parent_keys.update, parent_map.itervalues())
1470
parent_keys.difference_update(revision_keys)
1471
parent_keys.discard(_mod_revision.NULL_REVISION)
1474
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
1475
"""Find the file ids and versions affected by revisions.
1477
:param revisions: an iterable containing revision ids.
1478
:param _inv_weave: The inventory weave from this repository or None.
1479
If None, the inventory weave will be opened automatically.
1480
:return: a dictionary mapping altered file-ids to an iterable of
1481
revision_ids. Each altered file-ids has the exact revision_ids that
1482
altered it listed explicitly.
1484
selected_keys = set((revid,) for revid in revision_ids)
1485
w = _inv_weave or self.inventories
1486
return self._find_file_ids_from_xml_inventory_lines(
1487
w.iter_lines_added_or_present_in_keys(
1488
selected_keys, pb=None),
1491
def iter_files_bytes(self, desired_files):
1492
"""Iterate through file versions.
1494
Files will not necessarily be returned in the order they occur in
1495
desired_files. No specific order is guaranteed.
1497
Yields pairs of identifier, bytes_iterator. identifier is an opaque
1498
value supplied by the caller as part of desired_files. It should
1499
uniquely identify the file version in the caller's context. (Examples:
1500
an index number or a TreeTransform trans_id.)
1502
bytes_iterator is an iterable of bytestrings for the file. The
1503
kind of iterable and length of the bytestrings are unspecified, but for
1504
this implementation, it is a list of bytes produced by
1505
VersionedFile.get_record_stream().
1507
:param desired_files: a list of (file_id, revision_id, identifier)
1511
for file_id, revision_id, callable_data in desired_files:
1512
text_keys[(file_id, revision_id)] = callable_data
1513
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
1514
if record.storage_kind == 'absent':
1515
raise errors.RevisionNotPresent(record.key, self)
1516
yield text_keys[record.key], record.get_bytes_as('chunked')
1518
def _generate_text_key_index(self, text_key_references=None,
1520
"""Generate a new text key index for the repository.
1522
This is an expensive function that will take considerable time to run.
1524
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
1525
list of parents, also text keys. When a given key has no parents,
1526
the parents list will be [NULL_REVISION].
1528
# All revisions, to find inventory parents.
1529
if ancestors is None:
1530
graph = self.get_graph()
1531
ancestors = graph.get_parent_map(self.all_revision_ids())
1532
if text_key_references is None:
1533
text_key_references = self.find_text_key_references()
1534
pb = ui.ui_factory.nested_progress_bar()
1536
return self._do_generate_text_key_index(ancestors,
1537
text_key_references, pb)
1541
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
1542
"""Helper for _generate_text_key_index to avoid deep nesting."""
1543
revision_order = tsort.topo_sort(ancestors)
1544
invalid_keys = set()
1546
for revision_id in revision_order:
1547
revision_keys[revision_id] = set()
1548
text_count = len(text_key_references)
1549
# a cache of the text keys to allow reuse; costs a dict of all the
1550
# keys, but saves a 2-tuple for every child of a given key.
1552
for text_key, valid in text_key_references.iteritems():
1554
invalid_keys.add(text_key)
1556
revision_keys[text_key[1]].add(text_key)
1557
text_key_cache[text_key] = text_key
1558
del text_key_references
1560
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
1561
NULL_REVISION = _mod_revision.NULL_REVISION
1562
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
1563
# too small for large or very branchy trees. However, for 55K path
1564
# trees, it would be easy to use too much memory trivially. Ideally we
1565
# could gauge this by looking at available real memory etc, but this is
1566
# always a tricky proposition.
1567
inventory_cache = lru_cache.LRUCache(10)
1568
batch_size = 10 # should be ~150MB on a 55K path tree
1569
batch_count = len(revision_order) / batch_size + 1
1571
pb.update("Calculating text parents", processed_texts, text_count)
1572
for offset in xrange(batch_count):
1573
to_query = revision_order[offset * batch_size:(offset + 1) *
1577
for revision_id in to_query:
1578
parent_ids = ancestors[revision_id]
1579
for text_key in revision_keys[revision_id]:
1580
pb.update("Calculating text parents", processed_texts)
1581
processed_texts += 1
1582
candidate_parents = []
1583
for parent_id in parent_ids:
1584
parent_text_key = (text_key[0], parent_id)
1586
check_parent = parent_text_key not in \
1587
revision_keys[parent_id]
1589
# the parent parent_id is a ghost:
1590
check_parent = False
1591
# truncate the derived graph against this ghost.
1592
parent_text_key = None
1594
# look at the parent commit details inventories to
1595
# determine possible candidates in the per file graph.
1598
inv = inventory_cache[parent_id]
1600
inv = self.revision_tree(parent_id).inventory
1601
inventory_cache[parent_id] = inv
1603
parent_entry = inv[text_key[0]]
1604
except (KeyError, errors.NoSuchId):
1606
if parent_entry is not None:
1608
text_key[0], parent_entry.revision)
1610
parent_text_key = None
1611
if parent_text_key is not None:
1612
candidate_parents.append(
1613
text_key_cache[parent_text_key])
1614
parent_heads = text_graph.heads(candidate_parents)
1615
new_parents = list(parent_heads)
1616
new_parents.sort(key=lambda x:candidate_parents.index(x))
1617
if new_parents == []:
1618
new_parents = [NULL_REVISION]
1619
text_index[text_key] = new_parents
1621
for text_key in invalid_keys:
1622
text_index[text_key] = [NULL_REVISION]
1625
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
1626
"""Get an iterable listing the keys of all the data introduced by a set
1629
The keys will be ordered so that the corresponding items can be safely
1630
fetched and inserted in that order.
1632
:returns: An iterable producing tuples of (knit-kind, file-id,
1633
versions). knit-kind is one of 'file', 'inventory', 'signatures',
1634
'revisions'. file-id is None unless knit-kind is 'file'.
1636
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
1639
for result in self._find_non_file_keys_to_fetch(revision_ids):
1642
def _find_file_keys_to_fetch(self, revision_ids, pb):
1643
# XXX: it's a bit weird to control the inventory weave caching in this
1644
# generator. Ideally the caching would be done in fetch.py I think. Or
1645
# maybe this generator should explicitly have the contract that it
1646
# should not be iterated until the previously yielded item has been
1648
inv_w = self.inventories
1650
# file ids that changed
1651
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
1653
num_file_ids = len(file_ids)
1654
for file_id, altered_versions in file_ids.iteritems():
1656
pb.update("Fetch texts", count, num_file_ids)
1658
yield ("file", file_id, altered_versions)
1660
def _find_non_file_keys_to_fetch(self, revision_ids):
1662
yield ("inventory", None, revision_ids)
1665
# XXX: Note ATM no callers actually pay attention to this return
1666
# instead they just use the list of revision ids and ignore
1667
# missing sigs. Consider removing this work entirely
1668
revisions_with_signatures = set(self.signatures.get_parent_map(
1669
[(r,) for r in revision_ids]))
1670
revisions_with_signatures = set(
1671
[r for (r,) in revisions_with_signatures])
1672
revisions_with_signatures.intersection_update(revision_ids)
1673
yield ("signatures", None, revisions_with_signatures)
1676
yield ("revisions", None, revision_ids)
1679
def get_inventory(self, revision_id):
1680
"""Get Inventory object by revision id."""
1681
return self.iter_inventories([revision_id]).next()
1683
def iter_inventories(self, revision_ids, ordering=None):
1684
"""Get many inventories by revision_ids.
1686
This will buffer some or all of the texts used in constructing the
1687
inventories in memory, but will only parse a single inventory at a
1690
:param revision_ids: The expected revision ids of the inventories.
1691
:param ordering: optional ordering, e.g. 'topological'. If not
1692
specified, the order of revision_ids will be preserved (by
1693
buffering if necessary).
1694
:return: An iterator of inventories.
1696
if ((None in revision_ids)
1697
or (_mod_revision.NULL_REVISION in revision_ids)):
1698
raise ValueError('cannot get null revision inventory')
1699
return self._iter_inventories(revision_ids, ordering)
1701
def _iter_inventories(self, revision_ids, ordering):
1702
"""single-document based inventory iteration."""
1703
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
1704
for text, revision_id in inv_xmls:
1705
yield self._deserialise_inventory(revision_id, text)
1707
def _iter_inventory_xmls(self, revision_ids, ordering):
1708
if ordering is None:
1709
order_as_requested = True
1710
ordering = 'unordered'
1712
order_as_requested = False
1713
keys = [(revision_id,) for revision_id in revision_ids]
1716
if order_as_requested:
1717
key_iter = iter(keys)
1718
next_key = key_iter.next()
1719
stream = self.inventories.get_record_stream(keys, ordering, True)
1721
for record in stream:
1722
if record.storage_kind != 'absent':
1723
chunks = record.get_bytes_as('chunked')
1724
if order_as_requested:
1725
text_chunks[record.key] = chunks
1727
yield ''.join(chunks), record.key[-1]
1729
raise errors.NoSuchRevision(self, record.key)
1730
if order_as_requested:
1731
# Yield as many results as we can while preserving order.
1732
while next_key in text_chunks:
1733
chunks = text_chunks.pop(next_key)
1734
yield ''.join(chunks), next_key[-1]
1736
next_key = key_iter.next()
1737
except StopIteration:
1738
# We still want to fully consume the get_record_stream,
1739
# just in case it is not actually finished at this point
1743
def _deserialise_inventory(self, revision_id, xml):
1744
"""Transform the xml into an inventory object.
1746
:param revision_id: The expected revision id of the inventory.
1747
:param xml: A serialised inventory.
1749
result = self._serializer.read_inventory_from_string(xml, revision_id,
1750
entry_cache=self._inventory_entry_cache,
1751
return_from_cache=self._safe_to_return_from_cache)
1752
if result.revision_id != revision_id:
1753
raise AssertionError('revision id mismatch %s != %s' % (
1754
result.revision_id, revision_id))
1757
def get_serializer_format(self):
1758
return self._serializer.format_num
1761
def _get_inventory_xml(self, revision_id):
1762
"""Get serialized inventory as a string."""
1763
texts = self._iter_inventory_xmls([revision_id], 'unordered')
1765
text, revision_id = texts.next()
1766
except StopIteration:
1767
raise errors.HistoryMissing(self, 'inventory', revision_id)
1771
def revision_tree(self, revision_id):
1772
"""Return Tree for a revision on this branch.
1774
`revision_id` may be NULL_REVISION for the empty tree revision.
1776
revision_id = _mod_revision.ensure_null(revision_id)
1777
# TODO: refactor this to use an existing revision object
1778
# so we don't need to read it in twice.
1779
if revision_id == _mod_revision.NULL_REVISION:
1780
return InventoryRevisionTree(self,
1781
Inventory(root_id=None), _mod_revision.NULL_REVISION)
1783
inv = self.get_inventory(revision_id)
1784
return InventoryRevisionTree(self, inv, revision_id)
1786
def revision_trees(self, revision_ids):
1787
"""Return Trees for revisions in this repository.
1789
:param revision_ids: a sequence of revision-ids;
1790
a revision-id may not be None or 'null:'
1792
inventories = self.iter_inventories(revision_ids)
1793
for inv in inventories:
1794
yield InventoryRevisionTree(self, inv, inv.revision_id)
1796
def _filtered_revision_trees(self, revision_ids, file_ids):
1797
"""Return Tree for a revision on this branch with only some files.
1799
:param revision_ids: a sequence of revision-ids;
1800
a revision-id may not be None or 'null:'
1801
:param file_ids: if not None, the result is filtered
1802
so that only those file-ids, their parents and their
1803
children are included.
1805
inventories = self.iter_inventories(revision_ids)
1806
for inv in inventories:
1807
# Should we introduce a FilteredRevisionTree class rather
1808
# than pre-filter the inventory here?
1809
filtered_inv = inv.filter(file_ids)
1810
yield InventoryRevisionTree(self, filtered_inv, filtered_inv.revision_id)
1812
def get_parent_map(self, revision_ids):
1813
"""See graph.StackedParentsProvider.get_parent_map"""
1814
# revisions index works in keys; this just works in revisions
1815
# therefore wrap and unwrap
1818
for revision_id in revision_ids:
1819
if revision_id == _mod_revision.NULL_REVISION:
1820
result[revision_id] = ()
1821
elif revision_id is None:
1822
raise ValueError('get_parent_map(None) is not valid')
1824
query_keys.append((revision_id ,))
1825
for ((revision_id,), parent_keys) in \
1826
self.revisions.get_parent_map(query_keys).iteritems():
1828
result[revision_id] = tuple([parent_revid
1829
for (parent_revid,) in parent_keys])
1831
result[revision_id] = (_mod_revision.NULL_REVISION,)
1835
def get_known_graph_ancestry(self, revision_ids):
1836
"""Return the known graph for a set of revision ids and their ancestors.
1838
st = static_tuple.StaticTuple
1839
revision_keys = [st(r_id).intern() for r_id in revision_ids]
1840
known_graph = self.revisions.get_known_graph_ancestry(revision_keys)
1841
return graph.GraphThunkIdsToKeys(known_graph)
1844
def get_file_graph(self):
1845
"""Return the graph walker for text revisions."""
1846
return graph.Graph(self.texts)
1848
def _get_versioned_file_checker(self, text_key_references=None,
1850
"""Return an object suitable for checking versioned files.
1852
:param text_key_references: if non-None, an already built
1853
dictionary mapping text keys ((fileid, revision_id) tuples)
1854
to whether they were referred to by the inventory of the
1855
revision_id that they contain. If None, this will be
1857
:param ancestors: Optional result from
1858
self.get_graph().get_parent_map(self.all_revision_ids()) if already
1861
return _VersionedFileChecker(self,
1862
text_key_references=text_key_references, ancestors=ancestors)
1865
def has_signature_for_revision_id(self, revision_id):
1866
"""Query for a revision signature for revision_id in the repository."""
1867
if not self.has_revision(revision_id):
1868
raise errors.NoSuchRevision(self, revision_id)
1869
sig_present = (1 == len(
1870
self.signatures.get_parent_map([(revision_id,)])))
1874
def get_signature_text(self, revision_id):
1875
"""Return the text for a signature."""
1876
stream = self.signatures.get_record_stream([(revision_id,)],
1878
record = stream.next()
1879
if record.storage_kind == 'absent':
1880
raise errors.NoSuchRevision(self, revision_id)
1881
return record.get_bytes_as('fulltext')
1884
def _check(self, revision_ids, callback_refs, check_repo):
1885
result = check.VersionedFileCheck(self, check_repo=check_repo)
1886
result.check(callback_refs)
1889
def _find_inconsistent_revision_parents(self, revisions_iterator=None):
1890
"""Find revisions with different parent lists in the revision object
1891
and in the index graph.
1893
:param revisions_iterator: None, or an iterator of (revid,
1894
Revision-or-None). This iterator controls the revisions checked.
1895
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1896
parents-in-revision).
1898
if not self.is_locked():
1899
raise AssertionError()
1901
if revisions_iterator is None:
1902
revisions_iterator = self._iter_revisions(None)
1903
for revid, revision in revisions_iterator:
1904
if revision is None:
1906
parent_map = vf.get_parent_map([(revid,)])
1907
parents_according_to_index = tuple(parent[-1] for parent in
1908
parent_map[(revid,)])
1909
parents_according_to_revision = tuple(revision.parent_ids)
1910
if parents_according_to_index != parents_according_to_revision:
1911
yield (revid, parents_according_to_index,
1912
parents_according_to_revision)
1914
def _check_for_inconsistent_revision_parents(self):
1915
inconsistencies = list(self._find_inconsistent_revision_parents())
1917
raise errors.BzrCheckError(
1918
"Revision knit has inconsistent parents.")
1920
def _get_sink(self):
1921
"""Return a sink for streaming into this repository."""
1922
return StreamSink(self)
1924
def _get_source(self, to_format):
1925
"""Return a source for streaming from this repository."""
1926
return StreamSource(self, to_format)
1929
class MetaDirVersionedFileRepository(MetaDirRepository,
1930
VersionedFileRepository):
1931
"""Repositories in a meta-dir, that work via versioned file objects."""
1933
def __init__(self, _format, a_bzrdir, control_files):
1934
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
1938
class MetaDirVersionedFileRepositoryFormat(MetaDirRepositoryFormat,
1939
VersionedFileRepositoryFormat):
1940
"""Base class for repository formats using versioned files in metadirs."""
1943
class StreamSink(object):
1944
"""An object that can insert a stream into a repository.
1946
This interface handles the complexity of reserialising inventories and
1947
revisions from different formats, and allows unidirectional insertion into
1948
stacked repositories without looking for the missing basis parents
1952
def __init__(self, target_repo):
1953
self.target_repo = target_repo
1955
def insert_stream(self, stream, src_format, resume_tokens):
1956
"""Insert a stream's content into the target repository.
1958
:param src_format: a bzr repository format.
1960
:return: a list of resume tokens and an iterable of keys additional
1961
items required before the insertion can be completed.
1963
self.target_repo.lock_write()
1966
self.target_repo.resume_write_group(resume_tokens)
1969
self.target_repo.start_write_group()
1972
# locked_insert_stream performs a commit|suspend.
1973
missing_keys = self.insert_stream_without_locking(stream,
1974
src_format, is_resume)
1976
# suspend the write group and tell the caller what we is
1977
# missing. We know we can suspend or else we would not have
1978
# entered this code path. (All repositories that can handle
1979
# missing keys can handle suspending a write group).
1980
write_group_tokens = self.target_repo.suspend_write_group()
1981
return write_group_tokens, missing_keys
1982
hint = self.target_repo.commit_write_group()
1983
to_serializer = self.target_repo._format._serializer
1984
src_serializer = src_format._serializer
1985
if (to_serializer != src_serializer and
1986
self.target_repo._format.pack_compresses):
1987
self.target_repo.pack(hint=hint)
1990
self.target_repo.abort_write_group(suppress_errors=True)
1993
self.target_repo.unlock()
1995
def insert_stream_without_locking(self, stream, src_format,
1997
"""Insert a stream's content into the target repository.
1999
This assumes that you already have a locked repository and an active
2002
:param src_format: a bzr repository format.
2003
:param is_resume: Passed down to get_missing_parent_inventories to
2004
indicate if we should be checking for missing texts at the same
2007
:return: A set of keys that are missing.
2009
if not self.target_repo.is_write_locked():
2010
raise errors.ObjectNotLocked(self)
2011
if not self.target_repo.is_in_write_group():
2012
raise errors.BzrError('you must already be in a write group')
2013
to_serializer = self.target_repo._format._serializer
2014
src_serializer = src_format._serializer
2016
if to_serializer == src_serializer:
2017
# If serializers match and the target is a pack repository, set the
2018
# write cache size on the new pack. This avoids poor performance
2019
# on transports where append is unbuffered (such as
2020
# RemoteTransport). This is safe to do because nothing should read
2021
# back from the target repository while a stream with matching
2022
# serialization is being inserted.
2023
# The exception is that a delta record from the source that should
2024
# be a fulltext may need to be expanded by the target (see
2025
# test_fetch_revisions_with_deltas_into_pack); but we take care to
2026
# explicitly flush any buffered writes first in that rare case.
2028
new_pack = self.target_repo._pack_collection._new_pack
2029
except AttributeError:
2030
# Not a pack repository
2033
new_pack.set_write_cache_size(1024*1024)
2034
for substream_type, substream in stream:
2035
if 'stream' in debug.debug_flags:
2036
mutter('inserting substream: %s', substream_type)
2037
if substream_type == 'texts':
2038
self.target_repo.texts.insert_record_stream(substream)
2039
elif substream_type == 'inventories':
2040
if src_serializer == to_serializer:
2041
self.target_repo.inventories.insert_record_stream(
2044
self._extract_and_insert_inventories(
2045
substream, src_serializer)
2046
elif substream_type == 'inventory-deltas':
2047
self._extract_and_insert_inventory_deltas(
2048
substream, src_serializer)
2049
elif substream_type == 'chk_bytes':
2050
# XXX: This doesn't support conversions, as it assumes the
2051
# conversion was done in the fetch code.
2052
self.target_repo.chk_bytes.insert_record_stream(substream)
2053
elif substream_type == 'revisions':
2054
# This may fallback to extract-and-insert more often than
2055
# required if the serializers are different only in terms of
2057
if src_serializer == to_serializer:
2058
self.target_repo.revisions.insert_record_stream(substream)
2060
self._extract_and_insert_revisions(substream,
2062
elif substream_type == 'signatures':
2063
self.target_repo.signatures.insert_record_stream(substream)
2065
raise AssertionError('kaboom! %s' % (substream_type,))
2066
# Done inserting data, and the missing_keys calculations will try to
2067
# read back from the inserted data, so flush the writes to the new pack
2068
# (if this is pack format).
2069
if new_pack is not None:
2070
new_pack._write_data('', flush=True)
2071
# Find all the new revisions (including ones from resume_tokens)
2072
missing_keys = self.target_repo.get_missing_parent_inventories(
2073
check_for_missing_texts=is_resume)
2075
for prefix, versioned_file in (
2076
('texts', self.target_repo.texts),
2077
('inventories', self.target_repo.inventories),
2078
('revisions', self.target_repo.revisions),
2079
('signatures', self.target_repo.signatures),
2080
('chk_bytes', self.target_repo.chk_bytes),
2082
if versioned_file is None:
2084
# TODO: key is often going to be a StaticTuple object
2085
# I don't believe we can define a method by which
2086
# (prefix,) + StaticTuple will work, though we could
2087
# define a StaticTuple.sq_concat that would allow you to
2088
# pass in either a tuple or a StaticTuple as the second
2089
# object, so instead we could have:
2090
# StaticTuple(prefix) + key here...
2091
missing_keys.update((prefix,) + key for key in
2092
versioned_file.get_missing_compression_parent_keys())
2093
except NotImplementedError:
2094
# cannot even attempt suspending, and missing would have failed
2095
# during stream insertion.
2096
missing_keys = set()
2099
def _extract_and_insert_inventory_deltas(self, substream, serializer):
2100
target_rich_root = self.target_repo._format.rich_root_data
2101
target_tree_refs = self.target_repo._format.supports_tree_reference
2102
for record in substream:
2103
# Insert the delta directly
2104
inventory_delta_bytes = record.get_bytes_as('fulltext')
2105
deserialiser = inventory_delta.InventoryDeltaDeserializer()
2107
parse_result = deserialiser.parse_text_bytes(
2108
inventory_delta_bytes)
2109
except inventory_delta.IncompatibleInventoryDelta, err:
2110
mutter("Incompatible delta: %s", err.msg)
2111
raise errors.IncompatibleRevision(self.target_repo._format)
2112
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
2113
revision_id = new_id
2114
parents = [key[0] for key in record.parents]
2115
self.target_repo.add_inventory_by_delta(
2116
basis_id, inv_delta, revision_id, parents)
2118
def _extract_and_insert_inventories(self, substream, serializer,
2120
"""Generate a new inventory versionedfile in target, converting data.
2122
The inventory is retrieved from the source, (deserializing it), and
2123
stored in the target (reserializing it in a different format).
2125
target_rich_root = self.target_repo._format.rich_root_data
2126
target_tree_refs = self.target_repo._format.supports_tree_reference
2127
for record in substream:
2128
# It's not a delta, so it must be a fulltext in the source
2129
# serializer's format.
2130
bytes = record.get_bytes_as('fulltext')
2131
revision_id = record.key[0]
2132
inv = serializer.read_inventory_from_string(bytes, revision_id)
2133
parents = [key[0] for key in record.parents]
2134
self.target_repo.add_inventory(revision_id, inv, parents)
2135
# No need to keep holding this full inv in memory when the rest of
2136
# the substream is likely to be all deltas.
2139
def _extract_and_insert_revisions(self, substream, serializer):
2140
for record in substream:
2141
bytes = record.get_bytes_as('fulltext')
2142
revision_id = record.key[0]
2143
rev = serializer.read_revision_from_string(bytes)
2144
if rev.revision_id != revision_id:
2145
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
2146
self.target_repo.add_revision(revision_id, rev)
2149
if self.target_repo._format._fetch_reconcile:
2150
self.target_repo.reconcile()
2153
class StreamSource(object):
2154
"""A source of a stream for fetching between repositories."""
2156
def __init__(self, from_repository, to_format):
2157
"""Create a StreamSource streaming from from_repository."""
2158
self.from_repository = from_repository
2159
self.to_format = to_format
2160
self._record_counter = RecordCounter()
2162
def delta_on_metadata(self):
2163
"""Return True if delta's are permitted on metadata streams.
2165
That is on revisions and signatures.
2167
src_serializer = self.from_repository._format._serializer
2168
target_serializer = self.to_format._serializer
2169
return (self.to_format._fetch_uses_deltas and
2170
src_serializer == target_serializer)
2172
def _fetch_revision_texts(self, revs):
2173
# fetch signatures first and then the revision texts
2174
# may need to be a InterRevisionStore call here.
2175
from_sf = self.from_repository.signatures
2176
# A missing signature is just skipped.
2177
keys = [(rev_id,) for rev_id in revs]
2178
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
2180
self.to_format._fetch_order,
2181
not self.to_format._fetch_uses_deltas))
2182
# If a revision has a delta, this is actually expanded inside the
2183
# insert_record_stream code now, which is an alternate fix for
2185
from_rf = self.from_repository.revisions
2186
revisions = from_rf.get_record_stream(
2188
self.to_format._fetch_order,
2189
not self.delta_on_metadata())
2190
return [('signatures', signatures), ('revisions', revisions)]
2192
def _generate_root_texts(self, revs):
2193
"""This will be called by get_stream between fetching weave texts and
2194
fetching the inventory weave.
2196
if self._rich_root_upgrade():
2197
return _mod_fetch.Inter1and2Helper(
2198
self.from_repository).generate_root_texts(revs)
2202
def get_stream(self, search):
2204
revs = search.get_keys()
2205
graph = self.from_repository.get_graph()
2206
revs = tsort.topo_sort(graph.get_parent_map(revs))
2207
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
2209
for knit_kind, file_id, revisions in data_to_fetch:
2210
if knit_kind != phase:
2212
# Make a new progress bar for this phase
2213
if knit_kind == "file":
2214
# Accumulate file texts
2215
text_keys.extend([(file_id, revision) for revision in
2217
elif knit_kind == "inventory":
2218
# Now copy the file texts.
2219
from_texts = self.from_repository.texts
2220
yield ('texts', from_texts.get_record_stream(
2221
text_keys, self.to_format._fetch_order,
2222
not self.to_format._fetch_uses_deltas))
2223
# Cause an error if a text occurs after we have done the
2226
# Before we process the inventory we generate the root
2227
# texts (if necessary) so that the inventories references
2229
for _ in self._generate_root_texts(revs):
2231
# we fetch only the referenced inventories because we do not
2232
# know for unselected inventories whether all their required
2233
# texts are present in the other repository - it could be
2235
for info in self._get_inventory_stream(revs):
2237
elif knit_kind == "signatures":
2238
# Nothing to do here; this will be taken care of when
2239
# _fetch_revision_texts happens.
2241
elif knit_kind == "revisions":
2242
for record in self._fetch_revision_texts(revs):
2245
raise AssertionError("Unknown knit kind %r" % knit_kind)
2247
def get_stream_for_missing_keys(self, missing_keys):
2248
# missing keys can only occur when we are byte copying and not
2249
# translating (because translation means we don't send
2250
# unreconstructable deltas ever).
2252
keys['texts'] = set()
2253
keys['revisions'] = set()
2254
keys['inventories'] = set()
2255
keys['chk_bytes'] = set()
2256
keys['signatures'] = set()
2257
for key in missing_keys:
2258
keys[key[0]].add(key[1:])
2259
if len(keys['revisions']):
2260
# If we allowed copying revisions at this point, we could end up
2261
# copying a revision without copying its required texts: a
2262
# violation of the requirements for repository integrity.
2263
raise AssertionError(
2264
'cannot copy revisions to fill in missing deltas %s' % (
2265
keys['revisions'],))
2266
for substream_kind, keys in keys.iteritems():
2267
vf = getattr(self.from_repository, substream_kind)
2268
if vf is None and keys:
2269
raise AssertionError(
2270
"cannot fill in keys for a versioned file we don't"
2271
" have: %s needs %s" % (substream_kind, keys))
2273
# No need to stream something we don't have
2275
if substream_kind == 'inventories':
2276
# Some missing keys are genuinely ghosts, filter those out.
2277
present = self.from_repository.inventories.get_parent_map(keys)
2278
revs = [key[0] for key in present]
2279
# Get the inventory stream more-or-less as we do for the
2280
# original stream; there's no reason to assume that records
2281
# direct from the source will be suitable for the sink. (Think
2282
# e.g. 2a -> 1.9-rich-root).
2283
for info in self._get_inventory_stream(revs, missing=True):
2287
# Ask for full texts always so that we don't need more round trips
2288
# after this stream.
2289
# Some of the missing keys are genuinely ghosts, so filter absent
2290
# records. The Sink is responsible for doing another check to
2291
# ensure that ghosts don't introduce missing data for future
2293
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
2294
self.to_format._fetch_order, True))
2295
yield substream_kind, stream
2297
def inventory_fetch_order(self):
2298
if self._rich_root_upgrade():
2299
return 'topological'
2301
return self.to_format._fetch_order
2303
def _rich_root_upgrade(self):
2304
return (not self.from_repository._format.rich_root_data and
2305
self.to_format.rich_root_data)
2307
def _get_inventory_stream(self, revision_ids, missing=False):
2308
from_format = self.from_repository._format
2309
if (from_format.supports_chks and self.to_format.supports_chks and
2310
from_format.network_name() == self.to_format.network_name()):
2311
raise AssertionError(
2312
"this case should be handled by GroupCHKStreamSource")
2313
elif 'forceinvdeltas' in debug.debug_flags:
2314
return self._get_convertable_inventory_stream(revision_ids,
2315
delta_versus_null=missing)
2316
elif from_format.network_name() == self.to_format.network_name():
2318
return self._get_simple_inventory_stream(revision_ids,
2320
elif (not from_format.supports_chks and not self.to_format.supports_chks
2321
and from_format._serializer == self.to_format._serializer):
2322
# Essentially the same format.
2323
return self._get_simple_inventory_stream(revision_ids,
2326
# Any time we switch serializations, we want to use an
2327
# inventory-delta based approach.
2328
return self._get_convertable_inventory_stream(revision_ids,
2329
delta_versus_null=missing)
2331
def _get_simple_inventory_stream(self, revision_ids, missing=False):
2332
# NB: This currently reopens the inventory weave in source;
2333
# using a single stream interface instead would avoid this.
2334
from_weave = self.from_repository.inventories
2336
delta_closure = True
2338
delta_closure = not self.delta_on_metadata()
2339
yield ('inventories', from_weave.get_record_stream(
2340
[(rev_id,) for rev_id in revision_ids],
2341
self.inventory_fetch_order(), delta_closure))
2343
def _get_convertable_inventory_stream(self, revision_ids,
2344
delta_versus_null=False):
2345
# The two formats are sufficiently different that there is no fast
2346
# path, so we need to send just inventorydeltas, which any
2347
# sufficiently modern client can insert into any repository.
2348
# The StreamSink code expects to be able to
2349
# convert on the target, so we need to put bytes-on-the-wire that can
2350
# be converted. That means inventory deltas (if the remote is <1.19,
2351
# RemoteStreamSink will fallback to VFS to insert the deltas).
2352
yield ('inventory-deltas',
2353
self._stream_invs_as_deltas(revision_ids,
2354
delta_versus_null=delta_versus_null))
2356
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
2357
"""Return a stream of inventory-deltas for the given rev ids.
2359
:param revision_ids: The list of inventories to transmit
2360
:param delta_versus_null: Don't try to find a minimal delta for this
2361
entry, instead compute the delta versus the NULL_REVISION. This
2362
effectively streams a complete inventory. Used for stuff like
2363
filling in missing parents, etc.
2365
from_repo = self.from_repository
2366
revision_keys = [(rev_id,) for rev_id in revision_ids]
2367
parent_map = from_repo.inventories.get_parent_map(revision_keys)
2368
# XXX: possibly repos could implement a more efficient iter_inv_deltas
2370
inventories = self.from_repository.iter_inventories(
2371
revision_ids, 'topological')
2372
format = from_repo._format
2373
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
2374
inventory_cache = lru_cache.LRUCache(50)
2375
null_inventory = from_repo.revision_tree(
2376
_mod_revision.NULL_REVISION).inventory
2377
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
2378
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
2379
# repo back into a non-rich-root repo ought to be allowed)
2380
serializer = inventory_delta.InventoryDeltaSerializer(
2381
versioned_root=format.rich_root_data,
2382
tree_references=format.supports_tree_reference)
2383
for inv in inventories:
2384
key = (inv.revision_id,)
2385
parent_keys = parent_map.get(key, ())
2387
if not delta_versus_null and parent_keys:
2388
# The caller did not ask for complete inventories and we have
2389
# some parents that we can delta against. Make a delta against
2390
# each parent so that we can find the smallest.
2391
parent_ids = [parent_key[0] for parent_key in parent_keys]
2392
for parent_id in parent_ids:
2393
if parent_id not in invs_sent_so_far:
2394
# We don't know that the remote side has this basis, so
2397
if parent_id == _mod_revision.NULL_REVISION:
2398
parent_inv = null_inventory
2400
parent_inv = inventory_cache.get(parent_id, None)
2401
if parent_inv is None:
2402
parent_inv = from_repo.get_inventory(parent_id)
2403
candidate_delta = inv._make_delta(parent_inv)
2404
if (delta is None or
2405
len(delta) > len(candidate_delta)):
2406
delta = candidate_delta
2407
basis_id = parent_id
2409
# Either none of the parents ended up being suitable, or we
2410
# were asked to delta against NULL
2411
basis_id = _mod_revision.NULL_REVISION
2412
delta = inv._make_delta(null_inventory)
2413
invs_sent_so_far.add(inv.revision_id)
2414
inventory_cache[inv.revision_id] = inv
2415
delta_serialized = ''.join(
2416
serializer.delta_to_lines(basis_id, key[-1], delta))
2417
yield versionedfile.FulltextContentFactory(
2418
key, parent_keys, None, delta_serialized)
2421
class _VersionedFileChecker(object):
2423
def __init__(self, repository, text_key_references=None, ancestors=None):
2424
self.repository = repository
2425
self.text_index = self.repository._generate_text_key_index(
2426
text_key_references=text_key_references, ancestors=ancestors)
2428
def calculate_file_version_parents(self, text_key):
2429
"""Calculate the correct parents for a file version according to
2432
parent_keys = self.text_index[text_key]
2433
if parent_keys == [_mod_revision.NULL_REVISION]:
2435
return tuple(parent_keys)
2437
def check_file_version_parents(self, texts, progress_bar=None):
2438
"""Check the parents stored in a versioned file are correct.
2440
It also detects file versions that are not referenced by their
2441
corresponding revision's inventory.
2443
:returns: A tuple of (wrong_parents, dangling_file_versions).
2444
wrong_parents is a dict mapping {revision_id: (stored_parents,
2445
correct_parents)} for each revision_id where the stored parents
2446
are not correct. dangling_file_versions is a set of (file_id,
2447
revision_id) tuples for versions that are present in this versioned
2448
file, but not used by the corresponding inventory.
2450
local_progress = None
2451
if progress_bar is None:
2452
local_progress = ui.ui_factory.nested_progress_bar()
2453
progress_bar = local_progress
2455
return self._check_file_version_parents(texts, progress_bar)
2458
local_progress.finished()
2460
def _check_file_version_parents(self, texts, progress_bar):
2461
"""See check_file_version_parents."""
2463
self.file_ids = set([file_id for file_id, _ in
2464
self.text_index.iterkeys()])
2465
# text keys is now grouped by file_id
2466
n_versions = len(self.text_index)
2467
progress_bar.update('loading text store', 0, n_versions)
2468
parent_map = self.repository.texts.get_parent_map(self.text_index)
2469
# On unlistable transports this could well be empty/error...
2470
text_keys = self.repository.texts.keys()
2471
unused_keys = frozenset(text_keys) - set(self.text_index)
2472
for num, key in enumerate(self.text_index.iterkeys()):
2473
progress_bar.update('checking text graph', num, n_versions)
2474
correct_parents = self.calculate_file_version_parents(key)
2476
knit_parents = parent_map[key]
2477
except errors.RevisionNotPresent:
2480
if correct_parents != knit_parents:
2481
wrong_parents[key] = (knit_parents, correct_parents)
2482
return wrong_parents, unused_keys
2485
class InterVersionedFileRepository(InterRepository):
2487
_walk_to_common_revisions_batch_size = 50
2490
def fetch(self, revision_id=None, find_ghosts=False,
2492
"""Fetch the content required to construct revision_id.
2494
The content is copied from self.source to self.target.
2496
:param revision_id: if None all content is copied, if NULL_REVISION no
2500
if self.target._format.experimental:
2501
ui.ui_factory.show_user_warning('experimental_format_fetch',
2502
from_format=self.source._format,
2503
to_format=self.target._format)
2504
from bzrlib.fetch import RepoFetcher
2505
# See <https://launchpad.net/bugs/456077> asking for a warning here
2506
if self.source._format.network_name() != self.target._format.network_name():
2507
ui.ui_factory.show_user_warning('cross_format_fetch',
2508
from_format=self.source._format,
2509
to_format=self.target._format)
2510
f = RepoFetcher(to_repository=self.target,
2511
from_repository=self.source,
2512
last_revision=revision_id,
2513
fetch_spec=fetch_spec,
2514
find_ghosts=find_ghosts)
2516
def _walk_to_common_revisions(self, revision_ids, if_present_ids=None):
2517
"""Walk out from revision_ids in source to revisions target has.
2519
:param revision_ids: The start point for the search.
2520
:return: A set of revision ids.
2522
target_graph = self.target.get_graph()
2523
revision_ids = frozenset(revision_ids)
2525
all_wanted_revs = revision_ids.union(if_present_ids)
2527
all_wanted_revs = revision_ids
2528
missing_revs = set()
2529
source_graph = self.source.get_graph()
2530
# ensure we don't pay silly lookup costs.
2531
searcher = source_graph._make_breadth_first_searcher(all_wanted_revs)
2532
null_set = frozenset([_mod_revision.NULL_REVISION])
2533
searcher_exhausted = False
2537
# Iterate the searcher until we have enough next_revs
2538
while len(next_revs) < self._walk_to_common_revisions_batch_size:
2540
next_revs_part, ghosts_part = searcher.next_with_ghosts()
2541
next_revs.update(next_revs_part)
2542
ghosts.update(ghosts_part)
2543
except StopIteration:
2544
searcher_exhausted = True
2546
# If there are ghosts in the source graph, and the caller asked for
2547
# them, make sure that they are present in the target.
2548
# We don't care about other ghosts as we can't fetch them and
2549
# haven't been asked to.
2550
ghosts_to_check = set(revision_ids.intersection(ghosts))
2551
revs_to_get = set(next_revs).union(ghosts_to_check)
2553
have_revs = set(target_graph.get_parent_map(revs_to_get))
2554
# we always have NULL_REVISION present.
2555
have_revs = have_revs.union(null_set)
2556
# Check if the target is missing any ghosts we need.
2557
ghosts_to_check.difference_update(have_revs)
2559
# One of the caller's revision_ids is a ghost in both the
2560
# source and the target.
2561
raise errors.NoSuchRevision(
2562
self.source, ghosts_to_check.pop())
2563
missing_revs.update(next_revs - have_revs)
2564
# Because we may have walked past the original stop point, make
2565
# sure everything is stopped
2566
stop_revs = searcher.find_seen_ancestors(have_revs)
2567
searcher.stop_searching_any(stop_revs)
2568
if searcher_exhausted:
2570
return searcher.get_result()
2573
def search_missing_revision_ids(self,
2574
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
2575
find_ghosts=True, revision_ids=None, if_present_ids=None,
2577
"""Return the revision ids that source has that target does not.
2579
:param revision_id: only return revision ids included by this
2581
:param revision_ids: return revision ids included by these
2582
revision_ids. NoSuchRevision will be raised if any of these
2583
revisions are not present.
2584
:param if_present_ids: like revision_ids, but will not cause
2585
NoSuchRevision if any of these are absent, instead they will simply
2586
not be in the result. This is useful for e.g. finding revisions
2587
to fetch for tags, which may reference absent revisions.
2588
:param find_ghosts: If True find missing revisions in deep history
2589
rather than just finding the surface difference.
2590
:return: A bzrlib.graph.SearchResult.
2592
if symbol_versioning.deprecated_passed(revision_id):
2593
symbol_versioning.warn(
2594
'search_missing_revision_ids(revision_id=...) was '
2595
'deprecated in 2.4. Use revision_ids=[...] instead.',
2596
DeprecationWarning, stacklevel=2)
2597
if revision_ids is not None:
2598
raise AssertionError(
2599
'revision_ids is mutually exclusive with revision_id')
2600
if revision_id is not None:
2601
revision_ids = [revision_id]
2603
# stop searching at found target revisions.
2604
if not find_ghosts and (revision_ids is not None or if_present_ids is
2606
result = self._walk_to_common_revisions(revision_ids,
2607
if_present_ids=if_present_ids)
2610
result_set = result.get_keys()
2612
# generic, possibly worst case, slow code path.
2613
target_ids = set(self.target.all_revision_ids())
2614
source_ids = self._present_source_revisions_for(
2615
revision_ids, if_present_ids)
2616
result_set = set(source_ids).difference(target_ids)
2617
if limit is not None:
2618
topo_ordered = self.source.get_graph().iter_topo_order(result_set)
2619
result_set = set(itertools.islice(topo_ordered, limit))
2620
return self.source.revision_ids_to_search_result(result_set)
2622
def _present_source_revisions_for(self, revision_ids, if_present_ids=None):
2623
"""Returns set of all revisions in ancestry of revision_ids present in
2626
:param revision_ids: if None, all revisions in source are returned.
2627
:param if_present_ids: like revision_ids, but if any/all of these are
2628
absent no error is raised.
2630
if revision_ids is not None or if_present_ids is not None:
2631
# First, ensure all specified revisions exist. Callers expect
2632
# NoSuchRevision when they pass absent revision_ids here.
2633
if revision_ids is None:
2634
revision_ids = set()
2635
if if_present_ids is None:
2636
if_present_ids = set()
2637
revision_ids = set(revision_ids)
2638
if_present_ids = set(if_present_ids)
2639
all_wanted_ids = revision_ids.union(if_present_ids)
2640
graph = self.source.get_graph()
2641
present_revs = set(graph.get_parent_map(all_wanted_ids))
2642
missing = revision_ids.difference(present_revs)
2644
raise errors.NoSuchRevision(self.source, missing.pop())
2645
found_ids = all_wanted_ids.intersection(present_revs)
2646
source_ids = [rev_id for (rev_id, parents) in
2647
graph.iter_ancestry(found_ids)
2648
if rev_id != _mod_revision.NULL_REVISION
2649
and parents is not None]
2651
source_ids = self.source.all_revision_ids()
2652
return set(source_ids)
2655
def _get_repo_format_to_test(self):
2659
def is_compatible(cls, source, target):
2660
# The default implementation is compatible with everything
2661
return (source._format.supports_full_versioned_files and
2662
target._format.supports_full_versioned_files)
2665
class InterDifferingSerializer(InterVersionedFileRepository):
2668
def _get_repo_format_to_test(self):
2672
def is_compatible(source, target):
2673
if not source._format.supports_full_versioned_files:
2675
if not target._format.supports_full_versioned_files:
2677
# This is redundant with format.check_conversion_target(), however that
2678
# raises an exception, and we just want to say "False" as in we won't
2679
# support converting between these formats.
2680
if 'IDS_never' in debug.debug_flags:
2682
if source.supports_rich_root() and not target.supports_rich_root():
2684
if (source._format.supports_tree_reference
2685
and not target._format.supports_tree_reference):
2687
if target._fallback_repositories and target._format.supports_chks:
2688
# IDS doesn't know how to copy CHKs for the parent inventories it
2689
# adds to stacked repos.
2691
if 'IDS_always' in debug.debug_flags:
2693
# Only use this code path for local source and target. IDS does far
2694
# too much IO (both bandwidth and roundtrips) over a network.
2695
if not source.bzrdir.transport.base.startswith('file:///'):
2697
if not target.bzrdir.transport.base.startswith('file:///'):
2701
def _get_trees(self, revision_ids, cache):
2703
for rev_id in revision_ids:
2705
possible_trees.append((rev_id, cache[rev_id]))
2707
# Not cached, but inventory might be present anyway.
2709
tree = self.source.revision_tree(rev_id)
2710
except errors.NoSuchRevision:
2711
# Nope, parent is ghost.
2714
cache[rev_id] = tree
2715
possible_trees.append((rev_id, tree))
2716
return possible_trees
2718
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
2719
"""Get the best delta and base for this revision.
2721
:return: (basis_id, delta)
2724
# Generate deltas against each tree, to find the shortest.
2725
texts_possibly_new_in_tree = set()
2726
for basis_id, basis_tree in possible_trees:
2727
delta = tree.inventory._make_delta(basis_tree.inventory)
2728
for old_path, new_path, file_id, new_entry in delta:
2729
if new_path is None:
2730
# This file_id isn't present in the new rev, so we don't
2734
# Rich roots are handled elsewhere...
2736
kind = new_entry.kind
2737
if kind != 'directory' and kind != 'file':
2738
# No text record associated with this inventory entry.
2740
# This is a directory or file that has changed somehow.
2741
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
2742
deltas.append((len(delta), basis_id, delta))
2744
return deltas[0][1:]
2746
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
2747
"""Find all parent revisions that are absent, but for which the
2748
inventory is present, and copy those inventories.
2750
This is necessary to preserve correctness when the source is stacked
2751
without fallbacks configured. (Note that in cases like upgrade the
2752
source may be not have _fallback_repositories even though it is
2756
for parents in parent_map.values():
2757
parent_revs.update(parents)
2758
present_parents = self.source.get_parent_map(parent_revs)
2759
absent_parents = set(parent_revs).difference(present_parents)
2760
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
2761
(rev_id,) for rev_id in absent_parents)
2762
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
2763
for parent_tree in self.source.revision_trees(parent_inv_ids):
2764
current_revision_id = parent_tree.get_revision_id()
2765
parents_parents_keys = parent_invs_keys_for_stacking[
2766
(current_revision_id,)]
2767
parents_parents = [key[-1] for key in parents_parents_keys]
2768
basis_id = _mod_revision.NULL_REVISION
2769
basis_tree = self.source.revision_tree(basis_id)
2770
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
2771
self.target.add_inventory_by_delta(
2772
basis_id, delta, current_revision_id, parents_parents)
2773
cache[current_revision_id] = parent_tree
2775
def _fetch_batch(self, revision_ids, basis_id, cache):
2776
"""Fetch across a few revisions.
2778
:param revision_ids: The revisions to copy
2779
:param basis_id: The revision_id of a tree that must be in cache, used
2780
as a basis for delta when no other base is available
2781
:param cache: A cache of RevisionTrees that we can use.
2782
:return: The revision_id of the last converted tree. The RevisionTree
2783
for it will be in cache
2785
# Walk though all revisions; get inventory deltas, copy referenced
2786
# texts that delta references, insert the delta, revision and
2788
root_keys_to_create = set()
2791
pending_revisions = []
2792
parent_map = self.source.get_parent_map(revision_ids)
2793
self._fetch_parent_invs_for_stacking(parent_map, cache)
2794
self.source._safe_to_return_from_cache = True
2795
for tree in self.source.revision_trees(revision_ids):
2796
# Find a inventory delta for this revision.
2797
# Find text entries that need to be copied, too.
2798
current_revision_id = tree.get_revision_id()
2799
parent_ids = parent_map.get(current_revision_id, ())
2800
parent_trees = self._get_trees(parent_ids, cache)
2801
possible_trees = list(parent_trees)
2802
if len(possible_trees) == 0:
2803
# There either aren't any parents, or the parents are ghosts,
2804
# so just use the last converted tree.
2805
possible_trees.append((basis_id, cache[basis_id]))
2806
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
2808
revision = self.source.get_revision(current_revision_id)
2809
pending_deltas.append((basis_id, delta,
2810
current_revision_id, revision.parent_ids))
2811
if self._converting_to_rich_root:
2812
self._revision_id_to_root_id[current_revision_id] = \
2814
# Determine which texts are in present in this revision but not in
2815
# any of the available parents.
2816
texts_possibly_new_in_tree = set()
2817
for old_path, new_path, file_id, entry in delta:
2818
if new_path is None:
2819
# This file_id isn't present in the new rev
2823
if not self.target.supports_rich_root():
2824
# The target doesn't support rich root, so we don't
2827
if self._converting_to_rich_root:
2828
# This can't be copied normally, we have to insert
2830
root_keys_to_create.add((file_id, entry.revision))
2833
texts_possibly_new_in_tree.add((file_id, entry.revision))
2834
for basis_id, basis_tree in possible_trees:
2835
basis_inv = basis_tree.inventory
2836
for file_key in list(texts_possibly_new_in_tree):
2837
file_id, file_revision = file_key
2839
entry = basis_inv[file_id]
2840
except errors.NoSuchId:
2842
if entry.revision == file_revision:
2843
texts_possibly_new_in_tree.remove(file_key)
2844
text_keys.update(texts_possibly_new_in_tree)
2845
pending_revisions.append(revision)
2846
cache[current_revision_id] = tree
2847
basis_id = current_revision_id
2848
self.source._safe_to_return_from_cache = False
2850
from_texts = self.source.texts
2851
to_texts = self.target.texts
2852
if root_keys_to_create:
2853
root_stream = _mod_fetch._new_root_data_stream(
2854
root_keys_to_create, self._revision_id_to_root_id, parent_map,
2856
to_texts.insert_record_stream(root_stream)
2857
to_texts.insert_record_stream(from_texts.get_record_stream(
2858
text_keys, self.target._format._fetch_order,
2859
not self.target._format._fetch_uses_deltas))
2860
# insert inventory deltas
2861
for delta in pending_deltas:
2862
self.target.add_inventory_by_delta(*delta)
2863
if self.target._fallback_repositories:
2864
# Make sure this stacked repository has all the parent inventories
2865
# for the new revisions that we are about to insert. We do this
2866
# before adding the revisions so that no revision is added until
2867
# all the inventories it may depend on are added.
2868
# Note that this is overzealous, as we may have fetched these in an
2871
revision_ids = set()
2872
for revision in pending_revisions:
2873
revision_ids.add(revision.revision_id)
2874
parent_ids.update(revision.parent_ids)
2875
parent_ids.difference_update(revision_ids)
2876
parent_ids.discard(_mod_revision.NULL_REVISION)
2877
parent_map = self.source.get_parent_map(parent_ids)
2878
# we iterate over parent_map and not parent_ids because we don't
2879
# want to try copying any revision which is a ghost
2880
for parent_tree in self.source.revision_trees(parent_map):
2881
current_revision_id = parent_tree.get_revision_id()
2882
parents_parents = parent_map[current_revision_id]
2883
possible_trees = self._get_trees(parents_parents, cache)
2884
if len(possible_trees) == 0:
2885
# There either aren't any parents, or the parents are
2886
# ghosts, so just use the last converted tree.
2887
possible_trees.append((basis_id, cache[basis_id]))
2888
basis_id, delta = self._get_delta_for_revision(parent_tree,
2889
parents_parents, possible_trees)
2890
self.target.add_inventory_by_delta(
2891
basis_id, delta, current_revision_id, parents_parents)
2892
# insert signatures and revisions
2893
for revision in pending_revisions:
2895
signature = self.source.get_signature_text(
2896
revision.revision_id)
2897
self.target.add_signature_text(revision.revision_id,
2899
except errors.NoSuchRevision:
2901
self.target.add_revision(revision.revision_id, revision)
2904
def _fetch_all_revisions(self, revision_ids, pb):
2905
"""Fetch everything for the list of revisions.
2907
:param revision_ids: The list of revisions to fetch. Must be in
2909
:param pb: A ProgressTask
2912
basis_id, basis_tree = self._get_basis(revision_ids[0])
2914
cache = lru_cache.LRUCache(100)
2915
cache[basis_id] = basis_tree
2916
del basis_tree # We don't want to hang on to it here
2920
for offset in range(0, len(revision_ids), batch_size):
2921
self.target.start_write_group()
2923
pb.update('Transferring revisions', offset,
2925
batch = revision_ids[offset:offset+batch_size]
2926
basis_id = self._fetch_batch(batch, basis_id, cache)
2928
self.source._safe_to_return_from_cache = False
2929
self.target.abort_write_group()
2932
hint = self.target.commit_write_group()
2935
if hints and self.target._format.pack_compresses:
2936
self.target.pack(hint=hints)
2937
pb.update('Transferring revisions', len(revision_ids),
2941
def fetch(self, revision_id=None, find_ghosts=False,
2943
"""See InterRepository.fetch()."""
2944
if fetch_spec is not None:
2945
revision_ids = fetch_spec.get_keys()
2948
if self.source._format.experimental:
2949
ui.ui_factory.show_user_warning('experimental_format_fetch',
2950
from_format=self.source._format,
2951
to_format=self.target._format)
2952
if (not self.source.supports_rich_root()
2953
and self.target.supports_rich_root()):
2954
self._converting_to_rich_root = True
2955
self._revision_id_to_root_id = {}
2957
self._converting_to_rich_root = False
2958
# See <https://launchpad.net/bugs/456077> asking for a warning here
2959
if self.source._format.network_name() != self.target._format.network_name():
2960
ui.ui_factory.show_user_warning('cross_format_fetch',
2961
from_format=self.source._format,
2962
to_format=self.target._format)
2963
if revision_ids is None:
2965
search_revision_ids = [revision_id]
2967
search_revision_ids = None
2968
revision_ids = self.target.search_missing_revision_ids(self.source,
2969
revision_ids=search_revision_ids,
2970
find_ghosts=find_ghosts).get_keys()
2971
if not revision_ids:
2973
revision_ids = tsort.topo_sort(
2974
self.source.get_graph().get_parent_map(revision_ids))
2975
if not revision_ids:
2977
# Walk though all revisions; get inventory deltas, copy referenced
2978
# texts that delta references, insert the delta, revision and
2980
pb = ui.ui_factory.nested_progress_bar()
2982
self._fetch_all_revisions(revision_ids, pb)
2985
return len(revision_ids), 0
2987
def _get_basis(self, first_revision_id):
2988
"""Get a revision and tree which exists in the target.
2990
This assumes that first_revision_id is selected for transmission
2991
because all other ancestors are already present. If we can't find an
2992
ancestor we fall back to NULL_REVISION since we know that is safe.
2994
:return: (basis_id, basis_tree)
2996
first_rev = self.source.get_revision(first_revision_id)
2998
basis_id = first_rev.parent_ids[0]
2999
# only valid as a basis if the target has it
3000
self.target.get_revision(basis_id)
3001
# Try to get a basis tree - if it's a ghost it will hit the
3002
# NoSuchRevision case.
3003
basis_tree = self.source.revision_tree(basis_id)
3004
except (IndexError, errors.NoSuchRevision):
3005
basis_id = _mod_revision.NULL_REVISION
3006
basis_tree = self.source.revision_tree(basis_id)
3007
return basis_id, basis_tree
3010
class InterSameDataRepository(InterVersionedFileRepository):
3011
"""Code for converting between repositories that represent the same data.
3013
Data format and model must match for this to work.
3017
def _get_repo_format_to_test(self):
3018
"""Repository format for testing with.
3020
InterSameData can pull from subtree to subtree and from non-subtree to
3021
non-subtree, so we test this with the richest repository format.
3023
from bzrlib.repofmt import knitrepo
3024
return knitrepo.RepositoryFormatKnit3()
3027
def is_compatible(source, target):
3029
InterRepository._same_model(source, target) and
3030
source._format.supports_full_versioned_files and
3031
target._format.supports_full_versioned_files)
3034
InterRepository.register_optimiser(InterVersionedFileRepository)
3035
InterRepository.register_optimiser(InterDifferingSerializer)
3036
InterRepository.register_optimiser(InterSameDataRepository)
3039
def install_revisions(repository, iterable, num_revisions=None, pb=None):
3040
"""Install all revision data into a repository.
3042
Accepts an iterable of revision, tree, signature tuples. The signature
3045
repository.start_write_group()
3047
inventory_cache = lru_cache.LRUCache(10)
3048
for n, (revision, revision_tree, signature) in enumerate(iterable):
3049
_install_revision(repository, revision, revision_tree, signature,
3052
pb.update('Transferring revisions', n + 1, num_revisions)
3054
repository.abort_write_group()
3057
repository.commit_write_group()
3060
def _install_revision(repository, rev, revision_tree, signature,
3062
"""Install all revision data into a repository."""
3063
present_parents = []
3065
for p_id in rev.parent_ids:
3066
if repository.has_revision(p_id):
3067
present_parents.append(p_id)
3068
parent_trees[p_id] = repository.revision_tree(p_id)
3070
parent_trees[p_id] = repository.revision_tree(
3071
_mod_revision.NULL_REVISION)
3073
inv = revision_tree.inventory
3074
entries = inv.iter_entries()
3075
# backwards compatibility hack: skip the root id.
3076
if not repository.supports_rich_root():
3077
path, root = entries.next()
3078
if root.revision != rev.revision_id:
3079
raise errors.IncompatibleRevision(repr(repository))
3081
for path, ie in entries:
3082
text_keys[(ie.file_id, ie.revision)] = ie
3083
text_parent_map = repository.texts.get_parent_map(text_keys)
3084
missing_texts = set(text_keys) - set(text_parent_map)
3085
# Add the texts that are not already present
3086
for text_key in missing_texts:
3087
ie = text_keys[text_key]
3089
# FIXME: TODO: The following loop overlaps/duplicates that done by
3090
# commit to determine parents. There is a latent/real bug here where
3091
# the parents inserted are not those commit would do - in particular
3092
# they are not filtered by heads(). RBC, AB
3093
for revision, tree in parent_trees.iteritems():
3094
if not tree.has_id(ie.file_id):
3096
parent_id = tree.get_file_revision(ie.file_id)
3097
if parent_id in text_parents:
3099
text_parents.append((ie.file_id, parent_id))
3100
lines = revision_tree.get_file(ie.file_id).readlines()
3101
repository.texts.add_lines(text_key, text_parents, lines)
3103
# install the inventory
3104
if repository._format._commit_inv_deltas and len(rev.parent_ids):
3105
# Cache this inventory
3106
inventory_cache[rev.revision_id] = inv
3108
basis_inv = inventory_cache[rev.parent_ids[0]]
3110
repository.add_inventory(rev.revision_id, inv, present_parents)
3112
delta = inv._make_delta(basis_inv)
3113
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
3114
rev.revision_id, present_parents)
3116
repository.add_inventory(rev.revision_id, inv, present_parents)
3117
except errors.RevisionAlreadyPresent:
3119
if signature is not None:
3120
repository.add_signature_text(rev.revision_id, signature)
3121
repository.add_revision(rev.revision_id, rev, inv)
3124
def install_revision(repository, rev, revision_tree):
3125
"""Install all revision data into a repository."""
3126
install_revisions(repository, [(rev, revision_tree, None)])