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,
44
from bzrlib.recordcounter import RecordCounter
45
from bzrlib.revisiontree import InventoryRevisionTree
46
from bzrlib.testament import Testament
47
from bzrlib.i18n import gettext
53
from bzrlib.decorators import (
58
from bzrlib.inventory import (
65
from bzrlib.repository import (
69
RepositoryFormatMetaDir,
74
from bzrlib.trace import (
79
class VersionedFileRepositoryFormat(RepositoryFormat):
80
"""Base class for all repository formats that are VersionedFiles-based."""
82
supports_full_versioned_files = True
83
supports_versioned_directories = True
84
supports_unreferenced_revisions = True
86
# Should commit add an inventory, or an inventory delta to the repository.
87
_commit_inv_deltas = True
88
# What order should fetch operations request streams in?
89
# The default is unordered as that is the cheapest for an origin to
91
_fetch_order = 'unordered'
92
# Does this repository format use deltas that can be fetched as-deltas ?
93
# (E.g. knits, where the knit deltas can be transplanted intact.
94
# We default to False, which will ensure that enough data to get
95
# a full text out of any fetch stream will be grabbed.
96
_fetch_uses_deltas = False
99
class VersionedFileCommitBuilder(CommitBuilder):
100
"""Commit builder implementation for versioned files based repositories.
103
# this commit builder supports the record_entry_contents interface
104
supports_record_entry_contents = True
106
# the default CommitBuilder does not manage trees whose root is versioned.
107
_versioned_root = False
109
def __init__(self, repository, parents, config, timestamp=None,
110
timezone=None, committer=None, revprops=None,
111
revision_id=None, lossy=False):
112
super(VersionedFileCommitBuilder, self).__init__(repository,
113
parents, config, timestamp, timezone, committer, revprops,
116
basis_id = self.parents[0]
118
basis_id = _mod_revision.NULL_REVISION
119
self.basis_delta_revision = basis_id
120
self.new_inventory = Inventory(None)
121
self._basis_delta = []
122
self.__heads = graph.HeadsCache(repository.get_graph()).heads
123
# memo'd check for no-op commits.
124
self._any_changes = False
125
# API compatibility, older code that used CommitBuilder did not call
126
# .record_delete(), which means the delta that is computed would not be
127
# valid. Callers that will call record_delete() should call
128
# .will_record_deletes() to indicate that.
129
self._recording_deletes = False
131
def will_record_deletes(self):
132
"""Tell the commit builder that deletes are being notified.
134
This enables the accumulation of an inventory delta; for the resulting
135
commit to be valid, deletes against the basis MUST be recorded via
136
builder.record_delete().
138
self._recording_deletes = True
140
def any_changes(self):
141
"""Return True if any entries were changed.
143
This includes merge-only changes. It is the core for the --unchanged
146
:return: True if any changes have occured.
148
return self._any_changes
150
def _ensure_fallback_inventories(self):
151
"""Ensure that appropriate inventories are available.
153
This only applies to repositories that are stacked, and is about
154
enusring the stacking invariants. Namely, that for any revision that is
155
present, we either have all of the file content, or we have the parent
156
inventory and the delta file content.
158
if not self.repository._fallback_repositories:
160
if not self.repository._format.supports_chks:
161
raise errors.BzrError("Cannot commit directly to a stacked branch"
162
" in pre-2a formats. See "
163
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
164
# This is a stacked repo, we need to make sure we have the parent
165
# inventories for the parents.
166
parent_keys = [(p,) for p in self.parents]
167
parent_map = self.repository.inventories._index.get_parent_map(parent_keys)
168
missing_parent_keys = set([pk for pk in parent_keys
169
if pk not in parent_map])
170
fallback_repos = list(reversed(self.repository._fallback_repositories))
171
missing_keys = [('inventories', pk[0])
172
for pk in missing_parent_keys]
174
while missing_keys and fallback_repos:
175
fallback_repo = fallback_repos.pop()
176
source = fallback_repo._get_source(self.repository._format)
177
sink = self.repository._get_sink()
178
stream = source.get_stream_for_missing_keys(missing_keys)
179
missing_keys = sink.insert_stream_without_locking(stream,
180
self.repository._format)
182
raise errors.BzrError('Unable to fill in parent inventories for a'
185
def commit(self, message):
186
"""Make the actual commit.
188
:return: The revision id of the recorded revision.
190
self._validate_unicode_text(message, 'commit message')
191
rev = _mod_revision.Revision(
192
timestamp=self._timestamp,
193
timezone=self._timezone,
194
committer=self._committer,
196
inventory_sha1=self.inv_sha1,
197
revision_id=self._new_revision_id,
198
properties=self._revprops)
199
rev.parent_ids = self.parents
200
self.repository.add_revision(self._new_revision_id, rev,
201
self.new_inventory, self._config)
202
self._ensure_fallback_inventories()
203
self.repository.commit_write_group()
204
return self._new_revision_id
207
"""Abort the commit that is being built.
209
self.repository.abort_write_group()
211
def revision_tree(self):
212
"""Return the tree that was just committed.
214
After calling commit() this can be called to get a
215
RevisionTree representing the newly committed tree. This is
216
preferred to calling Repository.revision_tree() because that may
217
require deserializing the inventory, while we already have a copy in
220
if self.new_inventory is None:
221
self.new_inventory = self.repository.get_inventory(
222
self._new_revision_id)
223
return InventoryRevisionTree(self.repository, self.new_inventory,
224
self._new_revision_id)
226
def finish_inventory(self):
227
"""Tell the builder that the inventory is finished.
229
:return: The inventory id in the repository, which can be used with
230
repository.get_inventory.
232
if self.new_inventory is None:
233
# an inventory delta was accumulated without creating a new
235
basis_id = self.basis_delta_revision
236
# We ignore the 'inventory' returned by add_inventory_by_delta
237
# because self.new_inventory is used to hint to the rest of the
238
# system what code path was taken
239
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
240
basis_id, self._basis_delta, self._new_revision_id,
243
if self.new_inventory.root is None:
244
raise AssertionError('Root entry should be supplied to'
245
' record_entry_contents, as of bzr 0.10.')
246
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
247
self.new_inventory.revision_id = self._new_revision_id
248
self.inv_sha1 = self.repository.add_inventory(
249
self._new_revision_id,
253
return self._new_revision_id
255
def _check_root(self, ie, parent_invs, tree):
256
"""Helper for record_entry_contents.
258
:param ie: An entry being added.
259
:param parent_invs: The inventories of the parent revisions of the
261
:param tree: The tree that is being committed.
263
# In this revision format, root entries have no knit or weave When
264
# serializing out to disk and back in root.revision is always
266
ie.revision = self._new_revision_id
268
def _require_root_change(self, tree):
269
"""Enforce an appropriate root object change.
271
This is called once when record_iter_changes is called, if and only if
272
the root was not in the delta calculated by record_iter_changes.
274
:param tree: The tree which is being committed.
276
if len(self.parents) == 0:
277
raise errors.RootMissing()
278
entry = entry_factory['directory'](tree.path2id(''), '',
280
entry.revision = self._new_revision_id
281
self._basis_delta.append(('', '', entry.file_id, entry))
283
def _get_delta(self, ie, basis_inv, path):
284
"""Get a delta against the basis inventory for ie."""
285
if not basis_inv.has_id(ie.file_id):
287
result = (None, path, ie.file_id, ie)
288
self._basis_delta.append(result)
290
elif ie != basis_inv[ie.file_id]:
292
# TODO: avoid tis id2path call.
293
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
294
self._basis_delta.append(result)
300
def _heads(self, file_id, revision_ids):
301
"""Calculate the graph heads for revision_ids in the graph of file_id.
303
This can use either a per-file graph or a global revision graph as we
304
have an identity relationship between the two graphs.
306
return self.__heads(revision_ids)
308
def get_basis_delta(self):
309
"""Return the complete inventory delta versus the basis inventory.
311
This has been built up with the calls to record_delete and
312
record_entry_contents. The client must have already called
313
will_record_deletes() to indicate that they will be generating a
316
:return: An inventory delta, suitable for use with apply_delta, or
317
Repository.add_inventory_by_delta, etc.
319
if not self._recording_deletes:
320
raise AssertionError("recording deletes not activated.")
321
return self._basis_delta
323
def record_delete(self, path, file_id):
324
"""Record that a delete occured against a basis tree.
326
This is an optional API - when used it adds items to the basis_delta
327
being accumulated by the commit builder. It cannot be called unless the
328
method will_record_deletes() has been called to inform the builder that
329
a delta is being supplied.
331
:param path: The path of the thing deleted.
332
:param file_id: The file id that was deleted.
334
if not self._recording_deletes:
335
raise AssertionError("recording deletes not activated.")
336
delta = (path, None, file_id, None)
337
self._basis_delta.append(delta)
338
self._any_changes = True
341
def record_entry_contents(self, ie, parent_invs, path, tree,
343
"""Record the content of ie from tree into the commit if needed.
345
Side effect: sets ie.revision when unchanged
347
:param ie: An inventory entry present in the commit.
348
:param parent_invs: The inventories of the parent revisions of the
350
:param path: The path the entry is at in the tree.
351
:param tree: The tree which contains this entry and should be used to
353
:param content_summary: Summary data from the tree about the paths
354
content - stat, length, exec, sha/link target. This is only
355
accessed when the entry has a revision of None - that is when it is
356
a candidate to commit.
357
:return: A tuple (change_delta, version_recorded, fs_hash).
358
change_delta is an inventory_delta change for this entry against
359
the basis tree of the commit, or None if no change occured against
361
version_recorded is True if a new version of the entry has been
362
recorded. For instance, committing a merge where a file was only
363
changed on the other side will return (delta, False).
364
fs_hash is either None, or the hash details for the path (currently
365
a tuple of the contents sha1 and the statvalue returned by
366
tree.get_file_with_stat()).
368
if self.new_inventory.root is None:
369
if ie.parent_id is not None:
370
raise errors.RootMissing()
371
self._check_root(ie, parent_invs, tree)
372
if ie.revision is None:
373
kind = content_summary[0]
375
# ie is carried over from a prior commit
377
# XXX: repository specific check for nested tree support goes here - if
378
# the repo doesn't want nested trees we skip it ?
379
if (kind == 'tree-reference' and
380
not self.repository._format.supports_tree_reference):
381
# mismatch between commit builder logic and repository:
382
# this needs the entry creation pushed down into the builder.
383
raise NotImplementedError('Missing repository subtree support.')
384
self.new_inventory.add(ie)
386
# TODO: slow, take it out of the inner loop.
388
basis_inv = parent_invs[0]
390
basis_inv = Inventory(root_id=None)
392
# ie.revision is always None if the InventoryEntry is considered
393
# for committing. We may record the previous parents revision if the
394
# content is actually unchanged against a sole head.
395
if ie.revision is not None:
396
if not self._versioned_root and path == '':
397
# repositories that do not version the root set the root's
398
# revision to the new commit even when no change occurs (more
399
# specifically, they do not record a revision on the root; and
400
# the rev id is assigned to the root during deserialisation -
401
# this masks when a change may have occurred against the basis.
402
# To match this we always issue a delta, because the revision
403
# of the root will always be changing.
404
if basis_inv.has_id(ie.file_id):
405
delta = (basis_inv.id2path(ie.file_id), path,
409
delta = (None, path, ie.file_id, ie)
410
self._basis_delta.append(delta)
411
return delta, False, None
413
# we don't need to commit this, because the caller already
414
# determined that an existing revision of this file is
415
# appropriate. If it's not being considered for committing then
416
# it and all its parents to the root must be unaltered so
417
# no-change against the basis.
418
if ie.revision == self._new_revision_id:
419
raise AssertionError("Impossible situation, a skipped "
420
"inventory entry (%r) claims to be modified in this "
421
"commit (%r).", (ie, self._new_revision_id))
422
return None, False, None
423
# XXX: Friction: parent_candidates should return a list not a dict
424
# so that we don't have to walk the inventories again.
425
parent_candidate_entries = ie.parent_candidates(parent_invs)
426
head_set = self._heads(ie.file_id, parent_candidate_entries.keys())
428
for inv in parent_invs:
429
if inv.has_id(ie.file_id):
430
old_rev = inv[ie.file_id].revision
431
if old_rev in head_set:
432
heads.append(inv[ie.file_id].revision)
433
head_set.remove(inv[ie.file_id].revision)
436
# now we check to see if we need to write a new record to the
438
# We write a new entry unless there is one head to the ancestors, and
439
# the kind-derived content is unchanged.
441
# Cheapest check first: no ancestors, or more the one head in the
442
# ancestors, we write a new node.
446
# There is a single head, look it up for comparison
447
parent_entry = parent_candidate_entries[heads[0]]
448
# if the non-content specific data has changed, we'll be writing a
450
if (parent_entry.parent_id != ie.parent_id or
451
parent_entry.name != ie.name):
453
# now we need to do content specific checks:
455
# if the kind changed the content obviously has
456
if kind != parent_entry.kind:
458
# Stat cache fingerprint feedback for the caller - None as we usually
459
# don't generate one.
462
if content_summary[2] is None:
463
raise ValueError("Files must not have executable = None")
465
# We can't trust a check of the file length because of content
467
if (# if the exec bit has changed we have to store:
468
parent_entry.executable != content_summary[2]):
470
elif parent_entry.text_sha1 == content_summary[3]:
471
# all meta and content is unchanged (using a hash cache
472
# hit to check the sha)
473
ie.revision = parent_entry.revision
474
ie.text_size = parent_entry.text_size
475
ie.text_sha1 = parent_entry.text_sha1
476
ie.executable = parent_entry.executable
477
return self._get_delta(ie, basis_inv, path), False, None
479
# Either there is only a hash change(no hash cache entry,
480
# or same size content change), or there is no change on
482
# Provide the parent's hash to the store layer, so that the
483
# content is unchanged we will not store a new node.
484
nostore_sha = parent_entry.text_sha1
486
# We want to record a new node regardless of the presence or
487
# absence of a content change in the file.
489
ie.executable = content_summary[2]
490
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
492
text = file_obj.read()
496
ie.text_sha1, ie.text_size = self._add_text_to_weave(
497
ie.file_id, text, heads, nostore_sha)
498
# Let the caller know we generated a stat fingerprint.
499
fingerprint = (ie.text_sha1, stat_value)
500
except errors.ExistingContent:
501
# Turns out that the file content was unchanged, and we were
502
# only going to store a new node if it was changed. Carry over
504
ie.revision = parent_entry.revision
505
ie.text_size = parent_entry.text_size
506
ie.text_sha1 = parent_entry.text_sha1
507
ie.executable = parent_entry.executable
508
return self._get_delta(ie, basis_inv, path), False, None
509
elif kind == 'directory':
511
# all data is meta here, nothing specific to directory, so
513
ie.revision = parent_entry.revision
514
return self._get_delta(ie, basis_inv, path), False, None
515
self._add_text_to_weave(ie.file_id, '', heads, None)
516
elif kind == 'symlink':
517
current_link_target = content_summary[3]
519
# symlink target is not generic metadata, check if it has
521
if current_link_target != parent_entry.symlink_target:
524
# unchanged, carry over.
525
ie.revision = parent_entry.revision
526
ie.symlink_target = parent_entry.symlink_target
527
return self._get_delta(ie, basis_inv, path), False, None
528
ie.symlink_target = current_link_target
529
self._add_text_to_weave(ie.file_id, '', heads, None)
530
elif kind == 'tree-reference':
532
if content_summary[3] != parent_entry.reference_revision:
535
# unchanged, carry over.
536
ie.reference_revision = parent_entry.reference_revision
537
ie.revision = parent_entry.revision
538
return self._get_delta(ie, basis_inv, path), False, None
539
ie.reference_revision = content_summary[3]
540
if ie.reference_revision is None:
541
raise AssertionError("invalid content_summary for nested tree: %r"
542
% (content_summary,))
543
self._add_text_to_weave(ie.file_id, '', heads, None)
545
raise NotImplementedError('unknown kind')
546
ie.revision = self._new_revision_id
547
# The initial commit adds a root directory, but this in itself is not
548
# a worthwhile commit.
549
if (self.basis_delta_revision != _mod_revision.NULL_REVISION or
551
self._any_changes = True
552
return self._get_delta(ie, basis_inv, path), True, fingerprint
554
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
555
_entry_factory=entry_factory):
556
"""Record a new tree via iter_changes.
558
:param tree: The tree to obtain text contents from for changed objects.
559
:param basis_revision_id: The revision id of the tree the iter_changes
560
has been generated against. Currently assumed to be the same
561
as self.parents[0] - if it is not, errors may occur.
562
:param iter_changes: An iter_changes iterator with the changes to apply
563
to basis_revision_id. The iterator must not include any items with
564
a current kind of None - missing items must be either filtered out
565
or errored-on before record_iter_changes sees the item.
566
:param _entry_factory: Private method to bind entry_factory locally for
568
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
571
# Create an inventory delta based on deltas between all the parents and
572
# deltas between all the parent inventories. We use inventory delta's
573
# between the inventory objects because iter_changes masks
574
# last-changed-field only changes.
576
# file_id -> change map, change is fileid, paths, changed, versioneds,
577
# parents, names, kinds, executables
579
# {file_id -> revision_id -> inventory entry, for entries in parent
580
# trees that are not parents[0]
584
revtrees = list(self.repository.revision_trees(self.parents))
585
except errors.NoSuchRevision:
586
# one or more ghosts, slow path.
588
for revision_id in self.parents:
590
revtrees.append(self.repository.revision_tree(revision_id))
591
except errors.NoSuchRevision:
593
basis_revision_id = _mod_revision.NULL_REVISION
595
revtrees.append(self.repository.revision_tree(
596
_mod_revision.NULL_REVISION))
597
# The basis inventory from a repository
599
basis_inv = revtrees[0].inventory
601
basis_inv = self.repository.revision_tree(
602
_mod_revision.NULL_REVISION).inventory
603
if len(self.parents) > 0:
604
if basis_revision_id != self.parents[0] and not ghost_basis:
606
"arbitrary basis parents not yet supported with merges")
607
for revtree in revtrees[1:]:
608
for change in revtree.inventory._make_delta(basis_inv):
609
if change[1] is None:
610
# Not present in this parent.
612
if change[2] not in merged_ids:
613
if change[0] is not None:
614
basis_entry = basis_inv[change[2]]
615
merged_ids[change[2]] = [
617
basis_entry.revision,
620
parent_entries[change[2]] = {
622
basis_entry.revision:basis_entry,
624
change[3].revision:change[3],
627
merged_ids[change[2]] = [change[3].revision]
628
parent_entries[change[2]] = {change[3].revision:change[3]}
630
merged_ids[change[2]].append(change[3].revision)
631
parent_entries[change[2]][change[3].revision] = change[3]
634
# Setup the changes from the tree:
635
# changes maps file_id -> (change, [parent revision_ids])
637
for change in iter_changes:
638
# This probably looks up in basis_inv way to much.
639
if change[1][0] is not None:
640
head_candidate = [basis_inv[change[0]].revision]
643
changes[change[0]] = change, merged_ids.get(change[0],
645
unchanged_merged = set(merged_ids) - set(changes)
646
# Extend the changes dict with synthetic changes to record merges of
648
for file_id in unchanged_merged:
649
# Record a merged version of these items that did not change vs the
650
# basis. This can be either identical parallel changes, or a revert
651
# of a specific file after a merge. The recorded content will be
652
# that of the current tree (which is the same as the basis), but
653
# the per-file graph will reflect a merge.
654
# NB:XXX: We are reconstructing path information we had, this
655
# should be preserved instead.
656
# inv delta change: (file_id, (path_in_source, path_in_target),
657
# changed_content, versioned, parent, name, kind,
660
basis_entry = basis_inv[file_id]
661
except errors.NoSuchId:
662
# a change from basis->some_parents but file_id isn't in basis
663
# so was new in the merge, which means it must have changed
664
# from basis -> current, and as it hasn't the add was reverted
665
# by the user. So we discard this change.
669
(basis_inv.id2path(file_id), tree.id2path(file_id)),
671
(basis_entry.parent_id, basis_entry.parent_id),
672
(basis_entry.name, basis_entry.name),
673
(basis_entry.kind, basis_entry.kind),
674
(basis_entry.executable, basis_entry.executable))
675
changes[file_id] = (change, merged_ids[file_id])
676
# changes contains tuples with the change and a set of inventory
677
# candidates for the file.
679
# old_path, new_path, file_id, new_inventory_entry
680
seen_root = False # Is the root in the basis delta?
681
inv_delta = self._basis_delta
682
modified_rev = self._new_revision_id
683
for change, head_candidates in changes.values():
684
if change[3][1]: # versioned in target.
685
# Several things may be happening here:
686
# We may have a fork in the per-file graph
687
# - record a change with the content from tree
688
# We may have a change against < all trees
689
# - carry over the tree that hasn't changed
690
# We may have a change against all trees
691
# - record the change with the content from tree
694
entry = _entry_factory[kind](file_id, change[5][1],
696
head_set = self._heads(change[0], set(head_candidates))
699
for head_candidate in head_candidates:
700
if head_candidate in head_set:
701
heads.append(head_candidate)
702
head_set.remove(head_candidate)
705
# Could be a carry-over situation:
706
parent_entry_revs = parent_entries.get(file_id, None)
707
if parent_entry_revs:
708
parent_entry = parent_entry_revs.get(heads[0], None)
711
if parent_entry is None:
712
# The parent iter_changes was called against is the one
713
# that is the per-file head, so any change is relevant
714
# iter_changes is valid.
715
carry_over_possible = False
717
# could be a carry over situation
718
# A change against the basis may just indicate a merge,
719
# we need to check the content against the source of the
720
# merge to determine if it was changed after the merge
722
if (parent_entry.kind != entry.kind or
723
parent_entry.parent_id != entry.parent_id or
724
parent_entry.name != entry.name):
725
# Metadata common to all entries has changed
726
# against per-file parent
727
carry_over_possible = False
729
carry_over_possible = True
730
# per-type checks for changes against the parent_entry
733
# Cannot be a carry-over situation
734
carry_over_possible = False
735
# Populate the entry in the delta
737
# XXX: There is still a small race here: If someone reverts the content of a file
738
# after iter_changes examines and decides it has changed,
739
# we will unconditionally record a new version even if some
740
# other process reverts it while commit is running (with
741
# the revert happening after iter_changes did its
744
entry.executable = True
746
entry.executable = False
747
if (carry_over_possible and
748
parent_entry.executable == entry.executable):
749
# Check the file length, content hash after reading
751
nostore_sha = parent_entry.text_sha1
754
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
756
text = file_obj.read()
760
entry.text_sha1, entry.text_size = self._add_text_to_weave(
761
file_id, text, heads, nostore_sha)
762
yield file_id, change[1][1], (entry.text_sha1, stat_value)
763
except errors.ExistingContent:
764
# No content change against a carry_over parent
765
# Perhaps this should also yield a fs hash update?
767
entry.text_size = parent_entry.text_size
768
entry.text_sha1 = parent_entry.text_sha1
769
elif kind == 'symlink':
771
entry.symlink_target = tree.get_symlink_target(file_id)
772
if (carry_over_possible and
773
parent_entry.symlink_target == entry.symlink_target):
776
self._add_text_to_weave(change[0], '', heads, None)
777
elif kind == 'directory':
778
if carry_over_possible:
781
# Nothing to set on the entry.
782
# XXX: split into the Root and nonRoot versions.
783
if change[1][1] != '' or self.repository.supports_rich_root():
784
self._add_text_to_weave(change[0], '', heads, None)
785
elif kind == 'tree-reference':
786
if not self.repository._format.supports_tree_reference:
787
# This isn't quite sane as an error, but we shouldn't
788
# ever see this code path in practice: tree's don't
789
# permit references when the repo doesn't support tree
791
raise errors.UnsupportedOperation(tree.add_reference,
793
reference_revision = tree.get_reference_revision(change[0])
794
entry.reference_revision = reference_revision
795
if (carry_over_possible and
796
parent_entry.reference_revision == reference_revision):
799
self._add_text_to_weave(change[0], '', heads, None)
801
raise AssertionError('unknown kind %r' % kind)
803
entry.revision = modified_rev
805
entry.revision = parent_entry.revision
808
new_path = change[1][1]
809
inv_delta.append((change[1][0], new_path, change[0], entry))
812
self.new_inventory = None
813
# The initial commit adds a root directory, but this in itself is not
814
# a worthwhile commit.
815
if ((len(inv_delta) > 0 and basis_revision_id != _mod_revision.NULL_REVISION) or
816
(len(inv_delta) > 1 and basis_revision_id == _mod_revision.NULL_REVISION)):
817
# This should perhaps be guarded by a check that the basis we
818
# commit against is the basis for the commit and if not do a delta
820
self._any_changes = True
822
# housekeeping root entry changes do not affect no-change commits.
823
self._require_root_change(tree)
824
self.basis_delta_revision = basis_revision_id
826
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
827
parent_keys = tuple([(file_id, parent) for parent in parents])
828
return self.repository.texts._add_text(
829
(file_id, self._new_revision_id), parent_keys, new_text,
830
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
833
class VersionedFileRootCommitBuilder(VersionedFileCommitBuilder):
834
"""This commitbuilder actually records the root id"""
836
# the root entry gets versioned properly by this builder.
837
_versioned_root = True
839
def _check_root(self, ie, parent_invs, tree):
840
"""Helper for record_entry_contents.
842
:param ie: An entry being added.
843
:param parent_invs: The inventories of the parent revisions of the
845
:param tree: The tree that is being committed.
848
def _require_root_change(self, tree):
849
"""Enforce an appropriate root object change.
851
This is called once when record_iter_changes is called, if and only if
852
the root was not in the delta calculated by record_iter_changes.
854
:param tree: The tree which is being committed.
856
# versioned roots do not change unless the tree found a change.
859
class VersionedFileRepository(Repository):
860
"""Repository holding history for one or more branches.
862
The repository holds and retrieves historical information including
863
revisions and file history. It's normally accessed only by the Branch,
864
which views a particular line of development through that history.
866
The Repository builds on top of some byte storage facilies (the revisions,
867
signatures, inventories, texts and chk_bytes attributes) and a Transport,
868
which respectively provide byte storage and a means to access the (possibly
871
The byte storage facilities are addressed via tuples, which we refer to
872
as 'keys' throughout the code base. Revision_keys, inventory_keys and
873
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
874
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
875
byte string made up of a hash identifier and a hash value.
876
We use this interface because it allows low friction with the underlying
877
code that implements disk indices, network encoding and other parts of
880
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
881
the serialised revisions for the repository. This can be used to obtain
882
revision graph information or to access raw serialised revisions.
883
The result of trying to insert data into the repository via this store
884
is undefined: it should be considered read-only except for implementors
886
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
887
the serialised signatures for the repository. This can be used to
888
obtain access to raw serialised signatures. The result of trying to
889
insert data into the repository via this store is undefined: it should
890
be considered read-only except for implementors of repositories.
891
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
892
the serialised inventories for the repository. This can be used to
893
obtain unserialised inventories. The result of trying to insert data
894
into the repository via this store is undefined: it should be
895
considered read-only except for implementors of repositories.
896
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
897
texts of files and directories for the repository. This can be used to
898
obtain file texts or file graphs. Note that Repository.iter_file_bytes
899
is usually a better interface for accessing file texts.
900
The result of trying to insert data into the repository via this store
901
is undefined: it should be considered read-only except for implementors
903
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
904
any data the repository chooses to store or have indexed by its hash.
905
The result of trying to insert data into the repository via this store
906
is undefined: it should be considered read-only except for implementors
908
:ivar _transport: Transport for file access to repository, typically
909
pointing to .bzr/repository.
912
# What class to use for a CommitBuilder. Often it's simpler to change this
913
# in a Repository class subclass rather than to override
914
# get_commit_builder.
915
_commit_builder_class = VersionedFileCommitBuilder
917
def add_fallback_repository(self, repository):
918
"""Add a repository to use for looking up data not held locally.
920
:param repository: A repository.
922
if not self._format.supports_external_lookups:
923
raise errors.UnstackableRepositoryFormat(self._format, self.base)
924
# This can raise an exception, so should be done before we lock the
925
# fallback repository.
926
self._check_fallback_repository(repository)
928
# This repository will call fallback.unlock() when we transition to
929
# the unlocked state, so we make sure to increment the lock count
930
repository.lock_read()
931
self._fallback_repositories.append(repository)
932
self.texts.add_fallback_versioned_files(repository.texts)
933
self.inventories.add_fallback_versioned_files(repository.inventories)
934
self.revisions.add_fallback_versioned_files(repository.revisions)
935
self.signatures.add_fallback_versioned_files(repository.signatures)
936
if self.chk_bytes is not None:
937
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
939
@only_raises(errors.LockNotHeld, errors.LockBroken)
941
super(VersionedFileRepository, self).unlock()
942
if self.control_files._lock_count == 0:
943
self._inventory_entry_cache.clear()
945
def add_inventory(self, revision_id, inv, parents):
946
"""Add the inventory inv to the repository as revision_id.
948
:param parents: The revision ids of the parents that revision_id
949
is known to have and are in the repository already.
951
:returns: The validator(which is a sha1 digest, though what is sha'd is
952
repository format specific) of the serialized inventory.
954
if not self.is_in_write_group():
955
raise AssertionError("%r not in write group" % (self,))
956
_mod_revision.check_not_reserved_id(revision_id)
957
if not (inv.revision_id is None or inv.revision_id == revision_id):
958
raise AssertionError(
959
"Mismatch between inventory revision"
960
" id and insertion revid (%r, %r)"
961
% (inv.revision_id, revision_id))
963
raise errors.RootMissing()
964
return self._add_inventory_checked(revision_id, inv, parents)
966
def _add_inventory_checked(self, revision_id, inv, parents):
967
"""Add inv to the repository after checking the inputs.
969
This function can be overridden to allow different inventory styles.
971
:seealso: add_inventory, for the contract.
973
inv_lines = self._serializer.write_inventory_to_lines(inv)
974
return self._inventory_add_lines(revision_id, parents,
975
inv_lines, check_content=False)
977
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
978
parents, basis_inv=None, propagate_caches=False):
979
"""Add a new inventory expressed as a delta against another revision.
981
See the inventory developers documentation for the theory behind
984
:param basis_revision_id: The inventory id the delta was created
985
against. (This does not have to be a direct parent.)
986
:param delta: The inventory delta (see Inventory.apply_delta for
988
:param new_revision_id: The revision id that the inventory is being
990
:param parents: The revision ids of the parents that revision_id is
991
known to have and are in the repository already. These are supplied
992
for repositories that depend on the inventory graph for revision
993
graph access, as well as for those that pun ancestry with delta
995
:param basis_inv: The basis inventory if it is already known,
997
:param propagate_caches: If True, the caches for this inventory are
998
copied to and updated for the result if possible.
1000
:returns: (validator, new_inv)
1001
The validator(which is a sha1 digest, though what is sha'd is
1002
repository format specific) of the serialized inventory, and the
1003
resulting inventory.
1005
if not self.is_in_write_group():
1006
raise AssertionError("%r not in write group" % (self,))
1007
_mod_revision.check_not_reserved_id(new_revision_id)
1008
basis_tree = self.revision_tree(basis_revision_id)
1009
basis_tree.lock_read()
1011
# Note that this mutates the inventory of basis_tree, which not all
1012
# inventory implementations may support: A better idiom would be to
1013
# return a new inventory, but as there is no revision tree cache in
1014
# repository this is safe for now - RBC 20081013
1015
if basis_inv is None:
1016
basis_inv = basis_tree.inventory
1017
basis_inv.apply_delta(delta)
1018
basis_inv.revision_id = new_revision_id
1019
return (self.add_inventory(new_revision_id, basis_inv, parents),
1024
def _inventory_add_lines(self, revision_id, parents, lines,
1025
check_content=True):
1026
"""Store lines in inv_vf and return the sha1 of the inventory."""
1027
parents = [(parent,) for parent in parents]
1028
result = self.inventories.add_lines((revision_id,), parents, lines,
1029
check_content=check_content)[0]
1030
self.inventories._access.flush()
1033
def add_revision(self, revision_id, rev, inv=None, config=None):
1034
"""Add rev to the revision store as revision_id.
1036
:param revision_id: the revision id to use.
1037
:param rev: The revision object.
1038
:param inv: The inventory for the revision. if None, it will be looked
1039
up in the inventory storer
1040
:param config: If None no digital signature will be created.
1041
If supplied its signature_needed method will be used
1042
to determine if a signature should be made.
1044
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1046
_mod_revision.check_not_reserved_id(revision_id)
1047
if config is not None and config.signature_needed():
1049
inv = self.get_inventory(revision_id)
1050
tree = InventoryRevisionTree(self, inv, revision_id)
1051
testament = Testament(rev, tree)
1052
plaintext = testament.as_short_text()
1053
self.store_revision_signature(
1054
gpg.GPGStrategy(config), plaintext, revision_id)
1055
# check inventory present
1056
if not self.inventories.get_parent_map([(revision_id,)]):
1058
raise errors.WeaveRevisionNotPresent(revision_id,
1061
# yes, this is not suitable for adding with ghosts.
1062
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1065
key = (revision_id,)
1066
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1067
self._add_revision(rev)
1069
def _add_revision(self, revision):
1070
text = self._serializer.write_revision_to_string(revision)
1071
key = (revision.revision_id,)
1072
parents = tuple((parent,) for parent in revision.parent_ids)
1073
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1075
def _check_inventories(self, checker):
1076
"""Check the inventories found from the revision scan.
1078
This is responsible for verifying the sha1 of inventories and
1079
creating a pending_keys set that covers data referenced by inventories.
1081
bar = ui.ui_factory.nested_progress_bar()
1083
self._do_check_inventories(checker, bar)
1087
def _do_check_inventories(self, checker, bar):
1088
"""Helper for _check_inventories."""
1090
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1091
kinds = ['chk_bytes', 'texts']
1092
count = len(checker.pending_keys)
1093
bar.update(gettext("inventories"), 0, 2)
1094
current_keys = checker.pending_keys
1095
checker.pending_keys = {}
1096
# Accumulate current checks.
1097
for key in current_keys:
1098
if key[0] != 'inventories' and key[0] not in kinds:
1099
checker._report_items.append('unknown key type %r' % (key,))
1100
keys[key[0]].add(key[1:])
1101
if keys['inventories']:
1102
# NB: output order *should* be roughly sorted - topo or
1103
# inverse topo depending on repository - either way decent
1104
# to just delta against. However, pre-CHK formats didn't
1105
# try to optimise inventory layout on disk. As such the
1106
# pre-CHK code path does not use inventory deltas.
1108
for record in self.inventories.check(keys=keys['inventories']):
1109
if record.storage_kind == 'absent':
1110
checker._report_items.append(
1111
'Missing inventory {%s}' % (record.key,))
1113
last_object = self._check_record('inventories', record,
1114
checker, last_object,
1115
current_keys[('inventories',) + record.key])
1116
del keys['inventories']
1119
bar.update(gettext("texts"), 1)
1120
while (checker.pending_keys or keys['chk_bytes']
1122
# Something to check.
1123
current_keys = checker.pending_keys
1124
checker.pending_keys = {}
1125
# Accumulate current checks.
1126
for key in current_keys:
1127
if key[0] not in kinds:
1128
checker._report_items.append('unknown key type %r' % (key,))
1129
keys[key[0]].add(key[1:])
1130
# Check the outermost kind only - inventories || chk_bytes || texts
1134
for record in getattr(self, kind).check(keys=keys[kind]):
1135
if record.storage_kind == 'absent':
1136
checker._report_items.append(
1137
'Missing %s {%s}' % (kind, record.key,))
1139
last_object = self._check_record(kind, record,
1140
checker, last_object, current_keys[(kind,) + record.key])
1144
def _check_record(self, kind, record, checker, last_object, item_data):
1145
"""Check a single text from this repository."""
1146
if kind == 'inventories':
1147
rev_id = record.key[0]
1148
inv = self._deserialise_inventory(rev_id,
1149
record.get_bytes_as('fulltext'))
1150
if last_object is not None:
1151
delta = inv._make_delta(last_object)
1152
for old_path, path, file_id, ie in delta:
1155
ie.check(checker, rev_id, inv)
1157
for path, ie in inv.iter_entries():
1158
ie.check(checker, rev_id, inv)
1159
if self._format.fast_deltas:
1161
elif kind == 'chk_bytes':
1162
# No code written to check chk_bytes for this repo format.
1163
checker._report_items.append(
1164
'unsupported key type chk_bytes for %s' % (record.key,))
1165
elif kind == 'texts':
1166
self._check_text(record, checker, item_data)
1168
checker._report_items.append(
1169
'unknown key type %s for %s' % (kind, record.key))
1171
def _check_text(self, record, checker, item_data):
1172
"""Check a single text."""
1173
# Check it is extractable.
1174
# TODO: check length.
1175
if record.storage_kind == 'chunked':
1176
chunks = record.get_bytes_as(record.storage_kind)
1177
sha1 = osutils.sha_strings(chunks)
1178
length = sum(map(len, chunks))
1180
content = record.get_bytes_as('fulltext')
1181
sha1 = osutils.sha_string(content)
1182
length = len(content)
1183
if item_data and sha1 != item_data[1]:
1184
checker._report_items.append(
1185
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1186
(record.key, sha1, item_data[1], item_data[2]))
1189
def _eliminate_revisions_not_present(self, revision_ids):
1190
"""Check every revision id in revision_ids to see if we have it.
1192
Returns a set of the present revisions.
1195
graph = self.get_graph()
1196
parent_map = graph.get_parent_map(revision_ids)
1197
# The old API returned a list, should this actually be a set?
1198
return parent_map.keys()
1200
def __init__(self, _format, a_bzrdir, control_files):
1201
"""Instantiate a VersionedFileRepository.
1203
:param _format: The format of the repository on disk.
1204
:param controldir: The ControlDir of the repository.
1205
:param control_files: Control files to use for locking, etc.
1207
# In the future we will have a single api for all stores for
1208
# getting file texts, inventories and revisions, then
1209
# this construct will accept instances of those things.
1210
super(VersionedFileRepository, self).__init__(_format, a_bzrdir,
1212
self._transport = control_files._transport
1213
self.base = self._transport.base
1215
self._reconcile_does_inventory_gc = True
1216
self._reconcile_fixes_text_parents = False
1217
self._reconcile_backsup_inventory = True
1218
# An InventoryEntry cache, used during deserialization
1219
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1220
# Is it safe to return inventory entries directly from the entry cache,
1221
# rather copying them?
1222
self._safe_to_return_from_cache = False
1224
def fetch(self, source, revision_id=None, find_ghosts=False,
1226
"""Fetch the content required to construct revision_id from source.
1228
If revision_id is None and fetch_spec is None, then all content is
1231
fetch() may not be used when the repository is in a write group -
1232
either finish the current write group before using fetch, or use
1233
fetch before starting the write group.
1235
:param find_ghosts: Find and copy revisions in the source that are
1236
ghosts in the target (and not reachable directly by walking out to
1237
the first-present revision in target from revision_id).
1238
:param revision_id: If specified, all the content needed for this
1239
revision ID will be copied to the target. Fetch will determine for
1240
itself which content needs to be copied.
1241
:param fetch_spec: If specified, a SearchResult or
1242
PendingAncestryResult that describes which revisions to copy. This
1243
allows copying multiple heads at once. Mutually exclusive with
1246
if fetch_spec is not None and revision_id is not None:
1247
raise AssertionError(
1248
"fetch_spec and revision_id are mutually exclusive.")
1249
if self.is_in_write_group():
1250
raise errors.InternalBzrError(
1251
"May not fetch while in a write group.")
1252
# fast path same-url fetch operations
1253
# TODO: lift out to somewhere common with RemoteRepository
1254
# <https://bugs.launchpad.net/bzr/+bug/401646>
1255
if (self.has_same_location(source)
1256
and fetch_spec is None
1257
and self._has_same_fallbacks(source)):
1258
# check that last_revision is in 'from' and then return a
1260
if (revision_id is not None and
1261
not _mod_revision.is_null(revision_id)):
1262
self.get_revision(revision_id)
1264
inter = InterRepository.get(source, self)
1265
if (fetch_spec is not None and
1266
not getattr(inter, "supports_fetch_spec", False)):
1267
raise errors.UnsupportedOperation(
1268
"fetch_spec not supported for %r" % inter)
1269
return inter.fetch(revision_id=revision_id,
1270
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1273
def gather_stats(self, revid=None, committers=None):
1274
"""See Repository.gather_stats()."""
1275
result = super(VersionedFileRepository, self).gather_stats(revid, committers)
1276
# now gather global repository information
1277
# XXX: This is available for many repos regardless of listability.
1278
if self.user_transport.listable():
1279
# XXX: do we want to __define len__() ?
1280
# Maybe the versionedfiles object should provide a different
1281
# method to get the number of keys.
1282
result['revisions'] = len(self.revisions.keys())
1283
# result['size'] = t
1286
def get_commit_builder(self, branch, parents, config, timestamp=None,
1287
timezone=None, committer=None, revprops=None,
1288
revision_id=None, lossy=False):
1289
"""Obtain a CommitBuilder for this repository.
1291
:param branch: Branch to commit to.
1292
:param parents: Revision ids of the parents of the new revision.
1293
:param config: Configuration to use.
1294
:param timestamp: Optional timestamp recorded for commit.
1295
:param timezone: Optional timezone for timestamp.
1296
:param committer: Optional committer to set for commit.
1297
:param revprops: Optional dictionary of revision properties.
1298
:param revision_id: Optional revision id.
1299
:param lossy: Whether to discard data that can not be natively
1300
represented, when pushing to a foreign VCS
1302
if self._fallback_repositories and not self._format.supports_chks:
1303
raise errors.BzrError("Cannot commit directly to a stacked branch"
1304
" in pre-2a formats. See "
1305
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1306
result = self._commit_builder_class(self, parents, config,
1307
timestamp, timezone, committer, revprops, revision_id,
1309
self.start_write_group()
1312
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1313
"""Return the keys of missing inventory parents for revisions added in
1316
A revision is not complete if the inventory delta for that revision
1317
cannot be calculated. Therefore if the parent inventories of a
1318
revision are not present, the revision is incomplete, and e.g. cannot
1319
be streamed by a smart server. This method finds missing inventory
1320
parents for revisions added in this write group.
1322
if not self._format.supports_external_lookups:
1323
# This is only an issue for stacked repositories
1325
if not self.is_in_write_group():
1326
raise AssertionError('not in a write group')
1328
# XXX: We assume that every added revision already has its
1329
# corresponding inventory, so we only check for parent inventories that
1330
# might be missing, rather than all inventories.
1331
parents = set(self.revisions._index.get_missing_parents())
1332
parents.discard(_mod_revision.NULL_REVISION)
1333
unstacked_inventories = self.inventories._index
1334
present_inventories = unstacked_inventories.get_parent_map(
1335
key[-1:] for key in parents)
1336
parents.difference_update(present_inventories)
1337
if len(parents) == 0:
1338
# No missing parent inventories.
1340
if not check_for_missing_texts:
1341
return set(('inventories', rev_id) for (rev_id,) in parents)
1342
# Ok, now we have a list of missing inventories. But these only matter
1343
# if the inventories that reference them are missing some texts they
1344
# appear to introduce.
1345
# XXX: Texts referenced by all added inventories need to be present,
1346
# but at the moment we're only checking for texts referenced by
1347
# inventories at the graph's edge.
1348
key_deps = self.revisions._index._key_dependencies
1349
key_deps.satisfy_refs_for_keys(present_inventories)
1350
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1351
file_ids = self.fileids_altered_by_revision_ids(referrers)
1352
missing_texts = set()
1353
for file_id, version_ids in file_ids.iteritems():
1354
missing_texts.update(
1355
(file_id, version_id) for version_id in version_ids)
1356
present_texts = self.texts.get_parent_map(missing_texts)
1357
missing_texts.difference_update(present_texts)
1358
if not missing_texts:
1359
# No texts are missing, so all revisions and their deltas are
1362
# Alternatively the text versions could be returned as the missing
1363
# keys, but this is likely to be less data.
1364
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1368
def has_revisions(self, revision_ids):
1369
"""Probe to find out the presence of multiple revisions.
1371
:param revision_ids: An iterable of revision_ids.
1372
:return: A set of the revision_ids that were present.
1374
parent_map = self.revisions.get_parent_map(
1375
[(rev_id,) for rev_id in revision_ids])
1377
if _mod_revision.NULL_REVISION in revision_ids:
1378
result.add(_mod_revision.NULL_REVISION)
1379
result.update([key[0] for key in parent_map])
1383
def get_revision_reconcile(self, revision_id):
1384
"""'reconcile' helper routine that allows access to a revision always.
1386
This variant of get_revision does not cross check the weave graph
1387
against the revision one as get_revision does: but it should only
1388
be used by reconcile, or reconcile-alike commands that are correcting
1389
or testing the revision graph.
1391
return self._get_revisions([revision_id])[0]
1394
def get_revisions(self, revision_ids):
1395
"""Get many revisions at once.
1397
Repositories that need to check data on every revision read should
1398
subclass this method.
1400
return self._get_revisions(revision_ids)
1403
def _get_revisions(self, revision_ids):
1404
"""Core work logic to get many revisions without sanity checks."""
1406
for revid, rev in self._iter_revisions(revision_ids):
1408
raise errors.NoSuchRevision(self, revid)
1410
return [revs[revid] for revid in revision_ids]
1412
def _iter_revisions(self, revision_ids):
1413
"""Iterate over revision objects.
1415
:param revision_ids: An iterable of revisions to examine. None may be
1416
passed to request all revisions known to the repository. Note that
1417
not all repositories can find unreferenced revisions; for those
1418
repositories only referenced ones will be returned.
1419
:return: An iterator of (revid, revision) tuples. Absent revisions (
1420
those asked for but not available) are returned as (revid, None).
1422
if revision_ids is None:
1423
revision_ids = self.all_revision_ids()
1425
for rev_id in revision_ids:
1426
if not rev_id or not isinstance(rev_id, basestring):
1427
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1428
keys = [(key,) for key in revision_ids]
1429
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1430
for record in stream:
1431
revid = record.key[0]
1432
if record.storage_kind == 'absent':
1435
text = record.get_bytes_as('fulltext')
1436
rev = self._serializer.read_revision_from_string(text)
1440
def add_signature_text(self, revision_id, signature):
1441
"""Store a signature text for a revision.
1443
:param revision_id: Revision id of the revision
1444
:param signature: Signature text.
1446
self.signatures.add_lines((revision_id,), (),
1447
osutils.split_lines(signature))
1449
def find_text_key_references(self):
1450
"""Find the text key references within the repository.
1452
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1453
to whether they were referred to by the inventory of the
1454
revision_id that they contain. The inventory texts from all present
1455
revision ids are assessed to generate this report.
1457
revision_keys = self.revisions.keys()
1458
w = self.inventories
1459
pb = ui.ui_factory.nested_progress_bar()
1461
return self._serializer._find_text_key_references(
1462
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1466
def _inventory_xml_lines_for_keys(self, keys):
1467
"""Get a line iterator of the sort needed for findind references.
1469
Not relevant for non-xml inventory repositories.
1471
Ghosts in revision_keys are ignored.
1473
:param revision_keys: The revision keys for the inventories to inspect.
1474
:return: An iterator over (inventory line, revid) for the fulltexts of
1475
all of the xml inventories specified by revision_keys.
1477
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1478
for record in stream:
1479
if record.storage_kind != 'absent':
1480
chunks = record.get_bytes_as('chunked')
1481
revid = record.key[-1]
1482
lines = osutils.chunks_to_lines(chunks)
1486
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1488
"""Helper routine for fileids_altered_by_revision_ids.
1490
This performs the translation of xml lines to revision ids.
1492
:param line_iterator: An iterator of lines, origin_version_id
1493
:param revision_keys: The revision ids to filter for. This should be a
1494
set or other type which supports efficient __contains__ lookups, as
1495
the revision key from each parsed line will be looked up in the
1496
revision_keys filter.
1497
:return: a dictionary mapping altered file-ids to an iterable of
1498
revision_ids. Each altered file-ids has the exact revision_ids that
1499
altered it listed explicitly.
1501
seen = set(self._serializer._find_text_key_references(
1502
line_iterator).iterkeys())
1503
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
1504
parent_seen = set(self._serializer._find_text_key_references(
1505
self._inventory_xml_lines_for_keys(parent_keys)))
1506
new_keys = seen - parent_seen
1508
setdefault = result.setdefault
1509
for key in new_keys:
1510
setdefault(key[0], set()).add(key[-1])
1513
def _find_parent_keys_of_revisions(self, revision_keys):
1514
"""Similar to _find_parent_ids_of_revisions, but used with keys.
1516
:param revision_keys: An iterable of revision_keys.
1517
:return: The parents of all revision_keys that are not already in
1520
parent_map = self.revisions.get_parent_map(revision_keys)
1522
map(parent_keys.update, parent_map.itervalues())
1523
parent_keys.difference_update(revision_keys)
1524
parent_keys.discard(_mod_revision.NULL_REVISION)
1527
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
1528
"""Find the file ids and versions affected by revisions.
1530
:param revisions: an iterable containing revision ids.
1531
:param _inv_weave: The inventory weave from this repository or None.
1532
If None, the inventory weave will be opened automatically.
1533
:return: a dictionary mapping altered file-ids to an iterable of
1534
revision_ids. Each altered file-ids has the exact revision_ids that
1535
altered it listed explicitly.
1537
selected_keys = set((revid,) for revid in revision_ids)
1538
w = _inv_weave or self.inventories
1539
return self._find_file_ids_from_xml_inventory_lines(
1540
w.iter_lines_added_or_present_in_keys(
1541
selected_keys, pb=None),
1544
def iter_files_bytes(self, desired_files):
1545
"""Iterate through file versions.
1547
Files will not necessarily be returned in the order they occur in
1548
desired_files. No specific order is guaranteed.
1550
Yields pairs of identifier, bytes_iterator. identifier is an opaque
1551
value supplied by the caller as part of desired_files. It should
1552
uniquely identify the file version in the caller's context. (Examples:
1553
an index number or a TreeTransform trans_id.)
1555
bytes_iterator is an iterable of bytestrings for the file. The
1556
kind of iterable and length of the bytestrings are unspecified, but for
1557
this implementation, it is a list of bytes produced by
1558
VersionedFile.get_record_stream().
1560
:param desired_files: a list of (file_id, revision_id, identifier)
1564
for file_id, revision_id, callable_data in desired_files:
1565
text_keys[(file_id, revision_id)] = callable_data
1566
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
1567
if record.storage_kind == 'absent':
1568
raise errors.RevisionNotPresent(record.key[1], record.key[0])
1569
yield text_keys[record.key], record.get_bytes_as('chunked')
1571
def _generate_text_key_index(self, text_key_references=None,
1573
"""Generate a new text key index for the repository.
1575
This is an expensive function that will take considerable time to run.
1577
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
1578
list of parents, also text keys. When a given key has no parents,
1579
the parents list will be [NULL_REVISION].
1581
# All revisions, to find inventory parents.
1582
if ancestors is None:
1583
graph = self.get_graph()
1584
ancestors = graph.get_parent_map(self.all_revision_ids())
1585
if text_key_references is None:
1586
text_key_references = self.find_text_key_references()
1587
pb = ui.ui_factory.nested_progress_bar()
1589
return self._do_generate_text_key_index(ancestors,
1590
text_key_references, pb)
1594
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
1595
"""Helper for _generate_text_key_index to avoid deep nesting."""
1596
revision_order = tsort.topo_sort(ancestors)
1597
invalid_keys = set()
1599
for revision_id in revision_order:
1600
revision_keys[revision_id] = set()
1601
text_count = len(text_key_references)
1602
# a cache of the text keys to allow reuse; costs a dict of all the
1603
# keys, but saves a 2-tuple for every child of a given key.
1605
for text_key, valid in text_key_references.iteritems():
1607
invalid_keys.add(text_key)
1609
revision_keys[text_key[1]].add(text_key)
1610
text_key_cache[text_key] = text_key
1611
del text_key_references
1613
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
1614
NULL_REVISION = _mod_revision.NULL_REVISION
1615
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
1616
# too small for large or very branchy trees. However, for 55K path
1617
# trees, it would be easy to use too much memory trivially. Ideally we
1618
# could gauge this by looking at available real memory etc, but this is
1619
# always a tricky proposition.
1620
inventory_cache = lru_cache.LRUCache(10)
1621
batch_size = 10 # should be ~150MB on a 55K path tree
1622
batch_count = len(revision_order) / batch_size + 1
1624
pb.update(gettext("Calculating text parents"), processed_texts, text_count)
1625
for offset in xrange(batch_count):
1626
to_query = revision_order[offset * batch_size:(offset + 1) *
1630
for revision_id in to_query:
1631
parent_ids = ancestors[revision_id]
1632
for text_key in revision_keys[revision_id]:
1633
pb.update(gettext("Calculating text parents"), processed_texts)
1634
processed_texts += 1
1635
candidate_parents = []
1636
for parent_id in parent_ids:
1637
parent_text_key = (text_key[0], parent_id)
1639
check_parent = parent_text_key not in \
1640
revision_keys[parent_id]
1642
# the parent parent_id is a ghost:
1643
check_parent = False
1644
# truncate the derived graph against this ghost.
1645
parent_text_key = None
1647
# look at the parent commit details inventories to
1648
# determine possible candidates in the per file graph.
1651
inv = inventory_cache[parent_id]
1653
inv = self.revision_tree(parent_id).inventory
1654
inventory_cache[parent_id] = inv
1656
parent_entry = inv[text_key[0]]
1657
except (KeyError, errors.NoSuchId):
1659
if parent_entry is not None:
1661
text_key[0], parent_entry.revision)
1663
parent_text_key = None
1664
if parent_text_key is not None:
1665
candidate_parents.append(
1666
text_key_cache[parent_text_key])
1667
parent_heads = text_graph.heads(candidate_parents)
1668
new_parents = list(parent_heads)
1669
new_parents.sort(key=lambda x:candidate_parents.index(x))
1670
if new_parents == []:
1671
new_parents = [NULL_REVISION]
1672
text_index[text_key] = new_parents
1674
for text_key in invalid_keys:
1675
text_index[text_key] = [NULL_REVISION]
1678
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
1679
"""Get an iterable listing the keys of all the data introduced by a set
1682
The keys will be ordered so that the corresponding items can be safely
1683
fetched and inserted in that order.
1685
:returns: An iterable producing tuples of (knit-kind, file-id,
1686
versions). knit-kind is one of 'file', 'inventory', 'signatures',
1687
'revisions'. file-id is None unless knit-kind is 'file'.
1689
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
1692
for result in self._find_non_file_keys_to_fetch(revision_ids):
1695
def _find_file_keys_to_fetch(self, revision_ids, pb):
1696
# XXX: it's a bit weird to control the inventory weave caching in this
1697
# generator. Ideally the caching would be done in fetch.py I think. Or
1698
# maybe this generator should explicitly have the contract that it
1699
# should not be iterated until the previously yielded item has been
1701
inv_w = self.inventories
1703
# file ids that changed
1704
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
1706
num_file_ids = len(file_ids)
1707
for file_id, altered_versions in file_ids.iteritems():
1709
pb.update(gettext("Fetch texts"), count, num_file_ids)
1711
yield ("file", file_id, altered_versions)
1713
def _find_non_file_keys_to_fetch(self, revision_ids):
1715
yield ("inventory", None, revision_ids)
1718
# XXX: Note ATM no callers actually pay attention to this return
1719
# instead they just use the list of revision ids and ignore
1720
# missing sigs. Consider removing this work entirely
1721
revisions_with_signatures = set(self.signatures.get_parent_map(
1722
[(r,) for r in revision_ids]))
1723
revisions_with_signatures = set(
1724
[r for (r,) in revisions_with_signatures])
1725
revisions_with_signatures.intersection_update(revision_ids)
1726
yield ("signatures", None, revisions_with_signatures)
1729
yield ("revisions", None, revision_ids)
1732
def get_inventory(self, revision_id):
1733
"""Get Inventory object by revision id."""
1734
return self.iter_inventories([revision_id]).next()
1736
def iter_inventories(self, revision_ids, ordering=None):
1737
"""Get many inventories by revision_ids.
1739
This will buffer some or all of the texts used in constructing the
1740
inventories in memory, but will only parse a single inventory at a
1743
:param revision_ids: The expected revision ids of the inventories.
1744
:param ordering: optional ordering, e.g. 'topological'. If not
1745
specified, the order of revision_ids will be preserved (by
1746
buffering if necessary).
1747
:return: An iterator of inventories.
1749
if ((None in revision_ids)
1750
or (_mod_revision.NULL_REVISION in revision_ids)):
1751
raise ValueError('cannot get null revision inventory')
1752
for inv, revid in self._iter_inventories(revision_ids, ordering):
1754
raise errors.NoSuchRevision(self, revid)
1757
def _iter_inventories(self, revision_ids, ordering):
1758
"""single-document based inventory iteration."""
1759
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
1760
for text, revision_id in inv_xmls:
1762
yield None, revision_id
1764
yield self._deserialise_inventory(revision_id, text), revision_id
1766
def _iter_inventory_xmls(self, revision_ids, ordering):
1767
if ordering is None:
1768
order_as_requested = True
1769
ordering = 'unordered'
1771
order_as_requested = False
1772
keys = [(revision_id,) for revision_id in revision_ids]
1775
if order_as_requested:
1776
key_iter = iter(keys)
1777
next_key = key_iter.next()
1778
stream = self.inventories.get_record_stream(keys, ordering, True)
1780
for record in stream:
1781
if record.storage_kind != 'absent':
1782
chunks = record.get_bytes_as('chunked')
1783
if order_as_requested:
1784
text_chunks[record.key] = chunks
1786
yield ''.join(chunks), record.key[-1]
1788
yield None, record.key[-1]
1789
if order_as_requested:
1790
# Yield as many results as we can while preserving order.
1791
while next_key in text_chunks:
1792
chunks = text_chunks.pop(next_key)
1793
yield ''.join(chunks), next_key[-1]
1795
next_key = key_iter.next()
1796
except StopIteration:
1797
# We still want to fully consume the get_record_stream,
1798
# just in case it is not actually finished at this point
1802
def _deserialise_inventory(self, revision_id, xml):
1803
"""Transform the xml into an inventory object.
1805
:param revision_id: The expected revision id of the inventory.
1806
:param xml: A serialised inventory.
1808
result = self._serializer.read_inventory_from_string(xml, revision_id,
1809
entry_cache=self._inventory_entry_cache,
1810
return_from_cache=self._safe_to_return_from_cache)
1811
if result.revision_id != revision_id:
1812
raise AssertionError('revision id mismatch %s != %s' % (
1813
result.revision_id, revision_id))
1816
def get_serializer_format(self):
1817
return self._serializer.format_num
1820
def _get_inventory_xml(self, revision_id):
1821
"""Get serialized inventory as a string."""
1822
texts = self._iter_inventory_xmls([revision_id], 'unordered')
1823
text, revision_id = texts.next()
1825
raise errors.NoSuchRevision(self, revision_id)
1829
def revision_tree(self, revision_id):
1830
"""Return Tree for a revision on this branch.
1832
`revision_id` may be NULL_REVISION for the empty tree revision.
1834
revision_id = _mod_revision.ensure_null(revision_id)
1835
# TODO: refactor this to use an existing revision object
1836
# so we don't need to read it in twice.
1837
if revision_id == _mod_revision.NULL_REVISION:
1838
return InventoryRevisionTree(self,
1839
Inventory(root_id=None), _mod_revision.NULL_REVISION)
1841
inv = self.get_inventory(revision_id)
1842
return InventoryRevisionTree(self, inv, revision_id)
1844
def revision_trees(self, revision_ids):
1845
"""Return Trees for revisions in this repository.
1847
:param revision_ids: a sequence of revision-ids;
1848
a revision-id may not be None or 'null:'
1850
inventories = self.iter_inventories(revision_ids)
1851
for inv in inventories:
1852
yield InventoryRevisionTree(self, inv, inv.revision_id)
1854
def _filtered_revision_trees(self, revision_ids, file_ids):
1855
"""Return Tree for a revision on this branch with only some files.
1857
:param revision_ids: a sequence of revision-ids;
1858
a revision-id may not be None or 'null:'
1859
:param file_ids: if not None, the result is filtered
1860
so that only those file-ids, their parents and their
1861
children are included.
1863
inventories = self.iter_inventories(revision_ids)
1864
for inv in inventories:
1865
# Should we introduce a FilteredRevisionTree class rather
1866
# than pre-filter the inventory here?
1867
filtered_inv = inv.filter(file_ids)
1868
yield InventoryRevisionTree(self, filtered_inv, filtered_inv.revision_id)
1870
def get_parent_map(self, revision_ids):
1871
"""See graph.StackedParentsProvider.get_parent_map"""
1872
# revisions index works in keys; this just works in revisions
1873
# therefore wrap and unwrap
1876
for revision_id in revision_ids:
1877
if revision_id == _mod_revision.NULL_REVISION:
1878
result[revision_id] = ()
1879
elif revision_id is None:
1880
raise ValueError('get_parent_map(None) is not valid')
1882
query_keys.append((revision_id ,))
1883
for ((revision_id,), parent_keys) in \
1884
self.revisions.get_parent_map(query_keys).iteritems():
1886
result[revision_id] = tuple([parent_revid
1887
for (parent_revid,) in parent_keys])
1889
result[revision_id] = (_mod_revision.NULL_REVISION,)
1893
def get_known_graph_ancestry(self, revision_ids):
1894
"""Return the known graph for a set of revision ids and their ancestors.
1896
st = static_tuple.StaticTuple
1897
revision_keys = [st(r_id).intern() for r_id in revision_ids]
1898
known_graph = self.revisions.get_known_graph_ancestry(revision_keys)
1899
return graph.GraphThunkIdsToKeys(known_graph)
1902
def get_file_graph(self):
1903
"""Return the graph walker for text revisions."""
1904
return graph.Graph(self.texts)
1906
def revision_ids_to_search_result(self, result_set):
1907
"""Convert a set of revision ids to a graph SearchResult."""
1908
result_parents = set()
1909
for parents in self.get_graph().get_parent_map(
1910
result_set).itervalues():
1911
result_parents.update(parents)
1912
included_keys = result_set.intersection(result_parents)
1913
start_keys = result_set.difference(included_keys)
1914
exclude_keys = result_parents.difference(result_set)
1915
result = vf_search.SearchResult(start_keys, exclude_keys,
1916
len(result_set), result_set)
1919
def _get_versioned_file_checker(self, text_key_references=None,
1921
"""Return an object suitable for checking versioned files.
1923
:param text_key_references: if non-None, an already built
1924
dictionary mapping text keys ((fileid, revision_id) tuples)
1925
to whether they were referred to by the inventory of the
1926
revision_id that they contain. If None, this will be
1928
:param ancestors: Optional result from
1929
self.get_graph().get_parent_map(self.all_revision_ids()) if already
1932
return _VersionedFileChecker(self,
1933
text_key_references=text_key_references, ancestors=ancestors)
1936
def has_signature_for_revision_id(self, revision_id):
1937
"""Query for a revision signature for revision_id in the repository."""
1938
if not self.has_revision(revision_id):
1939
raise errors.NoSuchRevision(self, revision_id)
1940
sig_present = (1 == len(
1941
self.signatures.get_parent_map([(revision_id,)])))
1945
def get_signature_text(self, revision_id):
1946
"""Return the text for a signature."""
1947
stream = self.signatures.get_record_stream([(revision_id,)],
1949
record = stream.next()
1950
if record.storage_kind == 'absent':
1951
raise errors.NoSuchRevision(self, revision_id)
1952
return record.get_bytes_as('fulltext')
1955
def _check(self, revision_ids, callback_refs, check_repo):
1956
result = check.VersionedFileCheck(self, check_repo=check_repo)
1957
result.check(callback_refs)
1960
def _find_inconsistent_revision_parents(self, revisions_iterator=None):
1961
"""Find revisions with different parent lists in the revision object
1962
and in the index graph.
1964
:param revisions_iterator: None, or an iterator of (revid,
1965
Revision-or-None). This iterator controls the revisions checked.
1966
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1967
parents-in-revision).
1969
if not self.is_locked():
1970
raise AssertionError()
1972
if revisions_iterator is None:
1973
revisions_iterator = self._iter_revisions(None)
1974
for revid, revision in revisions_iterator:
1975
if revision is None:
1977
parent_map = vf.get_parent_map([(revid,)])
1978
parents_according_to_index = tuple(parent[-1] for parent in
1979
parent_map[(revid,)])
1980
parents_according_to_revision = tuple(revision.parent_ids)
1981
if parents_according_to_index != parents_according_to_revision:
1982
yield (revid, parents_according_to_index,
1983
parents_according_to_revision)
1985
def _check_for_inconsistent_revision_parents(self):
1986
inconsistencies = list(self._find_inconsistent_revision_parents())
1988
raise errors.BzrCheckError(
1989
"Revision knit has inconsistent parents.")
1991
def _get_sink(self):
1992
"""Return a sink for streaming into this repository."""
1993
return StreamSink(self)
1995
def _get_source(self, to_format):
1996
"""Return a source for streaming from this repository."""
1997
return StreamSource(self, to_format)
2000
class MetaDirVersionedFileRepository(MetaDirRepository,
2001
VersionedFileRepository):
2002
"""Repositories in a meta-dir, that work via versioned file objects."""
2004
def __init__(self, _format, a_bzrdir, control_files):
2005
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2009
class MetaDirVersionedFileRepositoryFormat(RepositoryFormatMetaDir,
2010
VersionedFileRepositoryFormat):
2011
"""Base class for repository formats using versioned files in metadirs."""
2014
class StreamSink(object):
2015
"""An object that can insert a stream into a repository.
2017
This interface handles the complexity of reserialising inventories and
2018
revisions from different formats, and allows unidirectional insertion into
2019
stacked repositories without looking for the missing basis parents
2023
def __init__(self, target_repo):
2024
self.target_repo = target_repo
2026
def insert_stream(self, stream, src_format, resume_tokens):
2027
"""Insert a stream's content into the target repository.
2029
:param src_format: a bzr repository format.
2031
:return: a list of resume tokens and an iterable of keys additional
2032
items required before the insertion can be completed.
2034
self.target_repo.lock_write()
2037
self.target_repo.resume_write_group(resume_tokens)
2040
self.target_repo.start_write_group()
2043
# locked_insert_stream performs a commit|suspend.
2044
missing_keys = self.insert_stream_without_locking(stream,
2045
src_format, is_resume)
2047
# suspend the write group and tell the caller what we is
2048
# missing. We know we can suspend or else we would not have
2049
# entered this code path. (All repositories that can handle
2050
# missing keys can handle suspending a write group).
2051
write_group_tokens = self.target_repo.suspend_write_group()
2052
return write_group_tokens, missing_keys
2053
hint = self.target_repo.commit_write_group()
2054
to_serializer = self.target_repo._format._serializer
2055
src_serializer = src_format._serializer
2056
if (to_serializer != src_serializer and
2057
self.target_repo._format.pack_compresses):
2058
self.target_repo.pack(hint=hint)
2061
self.target_repo.abort_write_group(suppress_errors=True)
2064
self.target_repo.unlock()
2066
def insert_stream_without_locking(self, stream, src_format,
2068
"""Insert a stream's content into the target repository.
2070
This assumes that you already have a locked repository and an active
2073
:param src_format: a bzr repository format.
2074
:param is_resume: Passed down to get_missing_parent_inventories to
2075
indicate if we should be checking for missing texts at the same
2078
:return: A set of keys that are missing.
2080
if not self.target_repo.is_write_locked():
2081
raise errors.ObjectNotLocked(self)
2082
if not self.target_repo.is_in_write_group():
2083
raise errors.BzrError('you must already be in a write group')
2084
to_serializer = self.target_repo._format._serializer
2085
src_serializer = src_format._serializer
2087
if to_serializer == src_serializer:
2088
# If serializers match and the target is a pack repository, set the
2089
# write cache size on the new pack. This avoids poor performance
2090
# on transports where append is unbuffered (such as
2091
# RemoteTransport). This is safe to do because nothing should read
2092
# back from the target repository while a stream with matching
2093
# serialization is being inserted.
2094
# The exception is that a delta record from the source that should
2095
# be a fulltext may need to be expanded by the target (see
2096
# test_fetch_revisions_with_deltas_into_pack); but we take care to
2097
# explicitly flush any buffered writes first in that rare case.
2099
new_pack = self.target_repo._pack_collection._new_pack
2100
except AttributeError:
2101
# Not a pack repository
2104
new_pack.set_write_cache_size(1024*1024)
2105
for substream_type, substream in stream:
2106
if 'stream' in debug.debug_flags:
2107
mutter('inserting substream: %s', substream_type)
2108
if substream_type == 'texts':
2109
self.target_repo.texts.insert_record_stream(substream)
2110
elif substream_type == 'inventories':
2111
if src_serializer == to_serializer:
2112
self.target_repo.inventories.insert_record_stream(
2115
self._extract_and_insert_inventories(
2116
substream, src_serializer)
2117
elif substream_type == 'inventory-deltas':
2118
self._extract_and_insert_inventory_deltas(
2119
substream, src_serializer)
2120
elif substream_type == 'chk_bytes':
2121
# XXX: This doesn't support conversions, as it assumes the
2122
# conversion was done in the fetch code.
2123
self.target_repo.chk_bytes.insert_record_stream(substream)
2124
elif substream_type == 'revisions':
2125
# This may fallback to extract-and-insert more often than
2126
# required if the serializers are different only in terms of
2128
if src_serializer == to_serializer:
2129
self.target_repo.revisions.insert_record_stream(substream)
2131
self._extract_and_insert_revisions(substream,
2133
elif substream_type == 'signatures':
2134
self.target_repo.signatures.insert_record_stream(substream)
2136
raise AssertionError('kaboom! %s' % (substream_type,))
2137
# Done inserting data, and the missing_keys calculations will try to
2138
# read back from the inserted data, so flush the writes to the new pack
2139
# (if this is pack format).
2140
if new_pack is not None:
2141
new_pack._write_data('', flush=True)
2142
# Find all the new revisions (including ones from resume_tokens)
2143
missing_keys = self.target_repo.get_missing_parent_inventories(
2144
check_for_missing_texts=is_resume)
2146
for prefix, versioned_file in (
2147
('texts', self.target_repo.texts),
2148
('inventories', self.target_repo.inventories),
2149
('revisions', self.target_repo.revisions),
2150
('signatures', self.target_repo.signatures),
2151
('chk_bytes', self.target_repo.chk_bytes),
2153
if versioned_file is None:
2155
# TODO: key is often going to be a StaticTuple object
2156
# I don't believe we can define a method by which
2157
# (prefix,) + StaticTuple will work, though we could
2158
# define a StaticTuple.sq_concat that would allow you to
2159
# pass in either a tuple or a StaticTuple as the second
2160
# object, so instead we could have:
2161
# StaticTuple(prefix) + key here...
2162
missing_keys.update((prefix,) + key for key in
2163
versioned_file.get_missing_compression_parent_keys())
2164
except NotImplementedError:
2165
# cannot even attempt suspending, and missing would have failed
2166
# during stream insertion.
2167
missing_keys = set()
2170
def _extract_and_insert_inventory_deltas(self, substream, serializer):
2171
target_rich_root = self.target_repo._format.rich_root_data
2172
target_tree_refs = self.target_repo._format.supports_tree_reference
2173
for record in substream:
2174
# Insert the delta directly
2175
inventory_delta_bytes = record.get_bytes_as('fulltext')
2176
deserialiser = inventory_delta.InventoryDeltaDeserializer()
2178
parse_result = deserialiser.parse_text_bytes(
2179
inventory_delta_bytes)
2180
except inventory_delta.IncompatibleInventoryDelta, err:
2181
mutter("Incompatible delta: %s", err.msg)
2182
raise errors.IncompatibleRevision(self.target_repo._format)
2183
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
2184
revision_id = new_id
2185
parents = [key[0] for key in record.parents]
2186
self.target_repo.add_inventory_by_delta(
2187
basis_id, inv_delta, revision_id, parents)
2189
def _extract_and_insert_inventories(self, substream, serializer,
2191
"""Generate a new inventory versionedfile in target, converting data.
2193
The inventory is retrieved from the source, (deserializing it), and
2194
stored in the target (reserializing it in a different format).
2196
target_rich_root = self.target_repo._format.rich_root_data
2197
target_tree_refs = self.target_repo._format.supports_tree_reference
2198
for record in substream:
2199
# It's not a delta, so it must be a fulltext in the source
2200
# serializer's format.
2201
bytes = record.get_bytes_as('fulltext')
2202
revision_id = record.key[0]
2203
inv = serializer.read_inventory_from_string(bytes, revision_id)
2204
parents = [key[0] for key in record.parents]
2205
self.target_repo.add_inventory(revision_id, inv, parents)
2206
# No need to keep holding this full inv in memory when the rest of
2207
# the substream is likely to be all deltas.
2210
def _extract_and_insert_revisions(self, substream, serializer):
2211
for record in substream:
2212
bytes = record.get_bytes_as('fulltext')
2213
revision_id = record.key[0]
2214
rev = serializer.read_revision_from_string(bytes)
2215
if rev.revision_id != revision_id:
2216
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
2217
self.target_repo.add_revision(revision_id, rev)
2220
if self.target_repo._format._fetch_reconcile:
2221
self.target_repo.reconcile()
2224
class StreamSource(object):
2225
"""A source of a stream for fetching between repositories."""
2227
def __init__(self, from_repository, to_format):
2228
"""Create a StreamSource streaming from from_repository."""
2229
self.from_repository = from_repository
2230
self.to_format = to_format
2231
self._record_counter = RecordCounter()
2233
def delta_on_metadata(self):
2234
"""Return True if delta's are permitted on metadata streams.
2236
That is on revisions and signatures.
2238
src_serializer = self.from_repository._format._serializer
2239
target_serializer = self.to_format._serializer
2240
return (self.to_format._fetch_uses_deltas and
2241
src_serializer == target_serializer)
2243
def _fetch_revision_texts(self, revs):
2244
# fetch signatures first and then the revision texts
2245
# may need to be a InterRevisionStore call here.
2246
from_sf = self.from_repository.signatures
2247
# A missing signature is just skipped.
2248
keys = [(rev_id,) for rev_id in revs]
2249
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
2251
self.to_format._fetch_order,
2252
not self.to_format._fetch_uses_deltas))
2253
# If a revision has a delta, this is actually expanded inside the
2254
# insert_record_stream code now, which is an alternate fix for
2256
from_rf = self.from_repository.revisions
2257
revisions = from_rf.get_record_stream(
2259
self.to_format._fetch_order,
2260
not self.delta_on_metadata())
2261
return [('signatures', signatures), ('revisions', revisions)]
2263
def _generate_root_texts(self, revs):
2264
"""This will be called by get_stream between fetching weave texts and
2265
fetching the inventory weave.
2267
if self._rich_root_upgrade():
2268
return _mod_fetch.Inter1and2Helper(
2269
self.from_repository).generate_root_texts(revs)
2273
def get_stream(self, search):
2275
revs = search.get_keys()
2276
graph = self.from_repository.get_graph()
2277
revs = tsort.topo_sort(graph.get_parent_map(revs))
2278
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
2280
for knit_kind, file_id, revisions in data_to_fetch:
2281
if knit_kind != phase:
2283
# Make a new progress bar for this phase
2284
if knit_kind == "file":
2285
# Accumulate file texts
2286
text_keys.extend([(file_id, revision) for revision in
2288
elif knit_kind == "inventory":
2289
# Now copy the file texts.
2290
from_texts = self.from_repository.texts
2291
yield ('texts', from_texts.get_record_stream(
2292
text_keys, self.to_format._fetch_order,
2293
not self.to_format._fetch_uses_deltas))
2294
# Cause an error if a text occurs after we have done the
2297
# Before we process the inventory we generate the root
2298
# texts (if necessary) so that the inventories references
2300
for _ in self._generate_root_texts(revs):
2302
# we fetch only the referenced inventories because we do not
2303
# know for unselected inventories whether all their required
2304
# texts are present in the other repository - it could be
2306
for info in self._get_inventory_stream(revs):
2308
elif knit_kind == "signatures":
2309
# Nothing to do here; this will be taken care of when
2310
# _fetch_revision_texts happens.
2312
elif knit_kind == "revisions":
2313
for record in self._fetch_revision_texts(revs):
2316
raise AssertionError("Unknown knit kind %r" % knit_kind)
2318
def get_stream_for_missing_keys(self, missing_keys):
2319
# missing keys can only occur when we are byte copying and not
2320
# translating (because translation means we don't send
2321
# unreconstructable deltas ever).
2323
keys['texts'] = set()
2324
keys['revisions'] = set()
2325
keys['inventories'] = set()
2326
keys['chk_bytes'] = set()
2327
keys['signatures'] = set()
2328
for key in missing_keys:
2329
keys[key[0]].add(key[1:])
2330
if len(keys['revisions']):
2331
# If we allowed copying revisions at this point, we could end up
2332
# copying a revision without copying its required texts: a
2333
# violation of the requirements for repository integrity.
2334
raise AssertionError(
2335
'cannot copy revisions to fill in missing deltas %s' % (
2336
keys['revisions'],))
2337
for substream_kind, keys in keys.iteritems():
2338
vf = getattr(self.from_repository, substream_kind)
2339
if vf is None and keys:
2340
raise AssertionError(
2341
"cannot fill in keys for a versioned file we don't"
2342
" have: %s needs %s" % (substream_kind, keys))
2344
# No need to stream something we don't have
2346
if substream_kind == 'inventories':
2347
# Some missing keys are genuinely ghosts, filter those out.
2348
present = self.from_repository.inventories.get_parent_map(keys)
2349
revs = [key[0] for key in present]
2350
# Get the inventory stream more-or-less as we do for the
2351
# original stream; there's no reason to assume that records
2352
# direct from the source will be suitable for the sink. (Think
2353
# e.g. 2a -> 1.9-rich-root).
2354
for info in self._get_inventory_stream(revs, missing=True):
2358
# Ask for full texts always so that we don't need more round trips
2359
# after this stream.
2360
# Some of the missing keys are genuinely ghosts, so filter absent
2361
# records. The Sink is responsible for doing another check to
2362
# ensure that ghosts don't introduce missing data for future
2364
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
2365
self.to_format._fetch_order, True))
2366
yield substream_kind, stream
2368
def inventory_fetch_order(self):
2369
if self._rich_root_upgrade():
2370
return 'topological'
2372
return self.to_format._fetch_order
2374
def _rich_root_upgrade(self):
2375
return (not self.from_repository._format.rich_root_data and
2376
self.to_format.rich_root_data)
2378
def _get_inventory_stream(self, revision_ids, missing=False):
2379
from_format = self.from_repository._format
2380
if (from_format.supports_chks and self.to_format.supports_chks and
2381
from_format.network_name() == self.to_format.network_name()):
2382
raise AssertionError(
2383
"this case should be handled by GroupCHKStreamSource")
2384
elif 'forceinvdeltas' in debug.debug_flags:
2385
return self._get_convertable_inventory_stream(revision_ids,
2386
delta_versus_null=missing)
2387
elif from_format.network_name() == self.to_format.network_name():
2389
return self._get_simple_inventory_stream(revision_ids,
2391
elif (not from_format.supports_chks and not self.to_format.supports_chks
2392
and from_format._serializer == self.to_format._serializer):
2393
# Essentially the same format.
2394
return self._get_simple_inventory_stream(revision_ids,
2397
# Any time we switch serializations, we want to use an
2398
# inventory-delta based approach.
2399
return self._get_convertable_inventory_stream(revision_ids,
2400
delta_versus_null=missing)
2402
def _get_simple_inventory_stream(self, revision_ids, missing=False):
2403
# NB: This currently reopens the inventory weave in source;
2404
# using a single stream interface instead would avoid this.
2405
from_weave = self.from_repository.inventories
2407
delta_closure = True
2409
delta_closure = not self.delta_on_metadata()
2410
yield ('inventories', from_weave.get_record_stream(
2411
[(rev_id,) for rev_id in revision_ids],
2412
self.inventory_fetch_order(), delta_closure))
2414
def _get_convertable_inventory_stream(self, revision_ids,
2415
delta_versus_null=False):
2416
# The two formats are sufficiently different that there is no fast
2417
# path, so we need to send just inventorydeltas, which any
2418
# sufficiently modern client can insert into any repository.
2419
# The StreamSink code expects to be able to
2420
# convert on the target, so we need to put bytes-on-the-wire that can
2421
# be converted. That means inventory deltas (if the remote is <1.19,
2422
# RemoteStreamSink will fallback to VFS to insert the deltas).
2423
yield ('inventory-deltas',
2424
self._stream_invs_as_deltas(revision_ids,
2425
delta_versus_null=delta_versus_null))
2427
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
2428
"""Return a stream of inventory-deltas for the given rev ids.
2430
:param revision_ids: The list of inventories to transmit
2431
:param delta_versus_null: Don't try to find a minimal delta for this
2432
entry, instead compute the delta versus the NULL_REVISION. This
2433
effectively streams a complete inventory. Used for stuff like
2434
filling in missing parents, etc.
2436
from_repo = self.from_repository
2437
revision_keys = [(rev_id,) for rev_id in revision_ids]
2438
parent_map = from_repo.inventories.get_parent_map(revision_keys)
2439
# XXX: possibly repos could implement a more efficient iter_inv_deltas
2441
inventories = self.from_repository.iter_inventories(
2442
revision_ids, 'topological')
2443
format = from_repo._format
2444
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
2445
inventory_cache = lru_cache.LRUCache(50)
2446
null_inventory = from_repo.revision_tree(
2447
_mod_revision.NULL_REVISION).inventory
2448
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
2449
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
2450
# repo back into a non-rich-root repo ought to be allowed)
2451
serializer = inventory_delta.InventoryDeltaSerializer(
2452
versioned_root=format.rich_root_data,
2453
tree_references=format.supports_tree_reference)
2454
for inv in inventories:
2455
key = (inv.revision_id,)
2456
parent_keys = parent_map.get(key, ())
2458
if not delta_versus_null and parent_keys:
2459
# The caller did not ask for complete inventories and we have
2460
# some parents that we can delta against. Make a delta against
2461
# each parent so that we can find the smallest.
2462
parent_ids = [parent_key[0] for parent_key in parent_keys]
2463
for parent_id in parent_ids:
2464
if parent_id not in invs_sent_so_far:
2465
# We don't know that the remote side has this basis, so
2468
if parent_id == _mod_revision.NULL_REVISION:
2469
parent_inv = null_inventory
2471
parent_inv = inventory_cache.get(parent_id, None)
2472
if parent_inv is None:
2473
parent_inv = from_repo.get_inventory(parent_id)
2474
candidate_delta = inv._make_delta(parent_inv)
2475
if (delta is None or
2476
len(delta) > len(candidate_delta)):
2477
delta = candidate_delta
2478
basis_id = parent_id
2480
# Either none of the parents ended up being suitable, or we
2481
# were asked to delta against NULL
2482
basis_id = _mod_revision.NULL_REVISION
2483
delta = inv._make_delta(null_inventory)
2484
invs_sent_so_far.add(inv.revision_id)
2485
inventory_cache[inv.revision_id] = inv
2486
delta_serialized = ''.join(
2487
serializer.delta_to_lines(basis_id, key[-1], delta))
2488
yield versionedfile.FulltextContentFactory(
2489
key, parent_keys, None, delta_serialized)
2492
class _VersionedFileChecker(object):
2494
def __init__(self, repository, text_key_references=None, ancestors=None):
2495
self.repository = repository
2496
self.text_index = self.repository._generate_text_key_index(
2497
text_key_references=text_key_references, ancestors=ancestors)
2499
def calculate_file_version_parents(self, text_key):
2500
"""Calculate the correct parents for a file version according to
2503
parent_keys = self.text_index[text_key]
2504
if parent_keys == [_mod_revision.NULL_REVISION]:
2506
return tuple(parent_keys)
2508
def check_file_version_parents(self, texts, progress_bar=None):
2509
"""Check the parents stored in a versioned file are correct.
2511
It also detects file versions that are not referenced by their
2512
corresponding revision's inventory.
2514
:returns: A tuple of (wrong_parents, dangling_file_versions).
2515
wrong_parents is a dict mapping {revision_id: (stored_parents,
2516
correct_parents)} for each revision_id where the stored parents
2517
are not correct. dangling_file_versions is a set of (file_id,
2518
revision_id) tuples for versions that are present in this versioned
2519
file, but not used by the corresponding inventory.
2521
local_progress = None
2522
if progress_bar is None:
2523
local_progress = ui.ui_factory.nested_progress_bar()
2524
progress_bar = local_progress
2526
return self._check_file_version_parents(texts, progress_bar)
2529
local_progress.finished()
2531
def _check_file_version_parents(self, texts, progress_bar):
2532
"""See check_file_version_parents."""
2534
self.file_ids = set([file_id for file_id, _ in
2535
self.text_index.iterkeys()])
2536
# text keys is now grouped by file_id
2537
n_versions = len(self.text_index)
2538
progress_bar.update(gettext('loading text store'), 0, n_versions)
2539
parent_map = self.repository.texts.get_parent_map(self.text_index)
2540
# On unlistable transports this could well be empty/error...
2541
text_keys = self.repository.texts.keys()
2542
unused_keys = frozenset(text_keys) - set(self.text_index)
2543
for num, key in enumerate(self.text_index.iterkeys()):
2544
progress_bar.update(gettext('checking text graph'), num, n_versions)
2545
correct_parents = self.calculate_file_version_parents(key)
2547
knit_parents = parent_map[key]
2548
except errors.RevisionNotPresent:
2551
if correct_parents != knit_parents:
2552
wrong_parents[key] = (knit_parents, correct_parents)
2553
return wrong_parents, unused_keys
2556
class InterVersionedFileRepository(InterRepository):
2558
_walk_to_common_revisions_batch_size = 50
2560
supports_fetch_spec = True
2563
def fetch(self, revision_id=None, find_ghosts=False,
2565
"""Fetch the content required to construct revision_id.
2567
The content is copied from self.source to self.target.
2569
:param revision_id: if None all content is copied, if NULL_REVISION no
2573
if self.target._format.experimental:
2574
ui.ui_factory.show_user_warning('experimental_format_fetch',
2575
from_format=self.source._format,
2576
to_format=self.target._format)
2577
from bzrlib.fetch import RepoFetcher
2578
# See <https://launchpad.net/bugs/456077> asking for a warning here
2579
if self.source._format.network_name() != self.target._format.network_name():
2580
ui.ui_factory.show_user_warning('cross_format_fetch',
2581
from_format=self.source._format,
2582
to_format=self.target._format)
2583
f = RepoFetcher(to_repository=self.target,
2584
from_repository=self.source,
2585
last_revision=revision_id,
2586
fetch_spec=fetch_spec,
2587
find_ghosts=find_ghosts)
2589
def _walk_to_common_revisions(self, revision_ids, if_present_ids=None):
2590
"""Walk out from revision_ids in source to revisions target has.
2592
:param revision_ids: The start point for the search.
2593
:return: A set of revision ids.
2595
target_graph = self.target.get_graph()
2596
revision_ids = frozenset(revision_ids)
2598
all_wanted_revs = revision_ids.union(if_present_ids)
2600
all_wanted_revs = revision_ids
2601
missing_revs = set()
2602
source_graph = self.source.get_graph()
2603
# ensure we don't pay silly lookup costs.
2604
searcher = source_graph._make_breadth_first_searcher(all_wanted_revs)
2605
null_set = frozenset([_mod_revision.NULL_REVISION])
2606
searcher_exhausted = False
2610
# Iterate the searcher until we have enough next_revs
2611
while len(next_revs) < self._walk_to_common_revisions_batch_size:
2613
next_revs_part, ghosts_part = searcher.next_with_ghosts()
2614
next_revs.update(next_revs_part)
2615
ghosts.update(ghosts_part)
2616
except StopIteration:
2617
searcher_exhausted = True
2619
# If there are ghosts in the source graph, and the caller asked for
2620
# them, make sure that they are present in the target.
2621
# We don't care about other ghosts as we can't fetch them and
2622
# haven't been asked to.
2623
ghosts_to_check = set(revision_ids.intersection(ghosts))
2624
revs_to_get = set(next_revs).union(ghosts_to_check)
2626
have_revs = set(target_graph.get_parent_map(revs_to_get))
2627
# we always have NULL_REVISION present.
2628
have_revs = have_revs.union(null_set)
2629
# Check if the target is missing any ghosts we need.
2630
ghosts_to_check.difference_update(have_revs)
2632
# One of the caller's revision_ids is a ghost in both the
2633
# source and the target.
2634
raise errors.NoSuchRevision(
2635
self.source, ghosts_to_check.pop())
2636
missing_revs.update(next_revs - have_revs)
2637
# Because we may have walked past the original stop point, make
2638
# sure everything is stopped
2639
stop_revs = searcher.find_seen_ancestors(have_revs)
2640
searcher.stop_searching_any(stop_revs)
2641
if searcher_exhausted:
2643
(started_keys, excludes, included_keys) = searcher.get_state()
2644
return vf_search.SearchResult(started_keys, excludes,
2645
len(included_keys), included_keys)
2648
def search_missing_revision_ids(self,
2649
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
2650
find_ghosts=True, revision_ids=None, if_present_ids=None,
2652
"""Return the revision ids that source has that target does not.
2654
:param revision_id: only return revision ids included by this
2656
:param revision_ids: return revision ids included by these
2657
revision_ids. NoSuchRevision will be raised if any of these
2658
revisions are not present.
2659
:param if_present_ids: like revision_ids, but will not cause
2660
NoSuchRevision if any of these are absent, instead they will simply
2661
not be in the result. This is useful for e.g. finding revisions
2662
to fetch for tags, which may reference absent revisions.
2663
:param find_ghosts: If True find missing revisions in deep history
2664
rather than just finding the surface difference.
2665
:return: A bzrlib.graph.SearchResult.
2667
if symbol_versioning.deprecated_passed(revision_id):
2668
symbol_versioning.warn(
2669
'search_missing_revision_ids(revision_id=...) was '
2670
'deprecated in 2.4. Use revision_ids=[...] instead.',
2671
DeprecationWarning, stacklevel=2)
2672
if revision_ids is not None:
2673
raise AssertionError(
2674
'revision_ids is mutually exclusive with revision_id')
2675
if revision_id is not None:
2676
revision_ids = [revision_id]
2678
# stop searching at found target revisions.
2679
if not find_ghosts and (revision_ids is not None or if_present_ids is
2681
result = self._walk_to_common_revisions(revision_ids,
2682
if_present_ids=if_present_ids)
2685
result_set = result.get_keys()
2687
# generic, possibly worst case, slow code path.
2688
target_ids = set(self.target.all_revision_ids())
2689
source_ids = self._present_source_revisions_for(
2690
revision_ids, if_present_ids)
2691
result_set = set(source_ids).difference(target_ids)
2692
if limit is not None:
2693
topo_ordered = self.source.get_graph().iter_topo_order(result_set)
2694
result_set = set(itertools.islice(topo_ordered, limit))
2695
return self.source.revision_ids_to_search_result(result_set)
2697
def _present_source_revisions_for(self, revision_ids, if_present_ids=None):
2698
"""Returns set of all revisions in ancestry of revision_ids present in
2701
:param revision_ids: if None, all revisions in source are returned.
2702
:param if_present_ids: like revision_ids, but if any/all of these are
2703
absent no error is raised.
2705
if revision_ids is not None or if_present_ids is not None:
2706
# First, ensure all specified revisions exist. Callers expect
2707
# NoSuchRevision when they pass absent revision_ids here.
2708
if revision_ids is None:
2709
revision_ids = set()
2710
if if_present_ids is None:
2711
if_present_ids = set()
2712
revision_ids = set(revision_ids)
2713
if_present_ids = set(if_present_ids)
2714
all_wanted_ids = revision_ids.union(if_present_ids)
2715
graph = self.source.get_graph()
2716
present_revs = set(graph.get_parent_map(all_wanted_ids))
2717
missing = revision_ids.difference(present_revs)
2719
raise errors.NoSuchRevision(self.source, missing.pop())
2720
found_ids = all_wanted_ids.intersection(present_revs)
2721
source_ids = [rev_id for (rev_id, parents) in
2722
graph.iter_ancestry(found_ids)
2723
if rev_id != _mod_revision.NULL_REVISION
2724
and parents is not None]
2726
source_ids = self.source.all_revision_ids()
2727
return set(source_ids)
2730
def _get_repo_format_to_test(self):
2734
def is_compatible(cls, source, target):
2735
# The default implementation is compatible with everything
2736
return (source._format.supports_full_versioned_files and
2737
target._format.supports_full_versioned_files)
2740
class InterDifferingSerializer(InterVersionedFileRepository):
2743
def _get_repo_format_to_test(self):
2747
def is_compatible(source, target):
2748
if not source._format.supports_full_versioned_files:
2750
if not target._format.supports_full_versioned_files:
2752
# This is redundant with format.check_conversion_target(), however that
2753
# raises an exception, and we just want to say "False" as in we won't
2754
# support converting between these formats.
2755
if 'IDS_never' in debug.debug_flags:
2757
if source.supports_rich_root() and not target.supports_rich_root():
2759
if (source._format.supports_tree_reference
2760
and not target._format.supports_tree_reference):
2762
if target._fallback_repositories and target._format.supports_chks:
2763
# IDS doesn't know how to copy CHKs for the parent inventories it
2764
# adds to stacked repos.
2766
if 'IDS_always' in debug.debug_flags:
2768
# Only use this code path for local source and target. IDS does far
2769
# too much IO (both bandwidth and roundtrips) over a network.
2770
if not source.bzrdir.transport.base.startswith('file:///'):
2772
if not target.bzrdir.transport.base.startswith('file:///'):
2776
def _get_trees(self, revision_ids, cache):
2778
for rev_id in revision_ids:
2780
possible_trees.append((rev_id, cache[rev_id]))
2782
# Not cached, but inventory might be present anyway.
2784
tree = self.source.revision_tree(rev_id)
2785
except errors.NoSuchRevision:
2786
# Nope, parent is ghost.
2789
cache[rev_id] = tree
2790
possible_trees.append((rev_id, tree))
2791
return possible_trees
2793
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
2794
"""Get the best delta and base for this revision.
2796
:return: (basis_id, delta)
2799
# Generate deltas against each tree, to find the shortest.
2800
texts_possibly_new_in_tree = set()
2801
for basis_id, basis_tree in possible_trees:
2802
delta = tree.inventory._make_delta(basis_tree.inventory)
2803
for old_path, new_path, file_id, new_entry in delta:
2804
if new_path is None:
2805
# This file_id isn't present in the new rev, so we don't
2809
# Rich roots are handled elsewhere...
2811
kind = new_entry.kind
2812
if kind != 'directory' and kind != 'file':
2813
# No text record associated with this inventory entry.
2815
# This is a directory or file that has changed somehow.
2816
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
2817
deltas.append((len(delta), basis_id, delta))
2819
return deltas[0][1:]
2821
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
2822
"""Find all parent revisions that are absent, but for which the
2823
inventory is present, and copy those inventories.
2825
This is necessary to preserve correctness when the source is stacked
2826
without fallbacks configured. (Note that in cases like upgrade the
2827
source may be not have _fallback_repositories even though it is
2831
for parents in parent_map.values():
2832
parent_revs.update(parents)
2833
present_parents = self.source.get_parent_map(parent_revs)
2834
absent_parents = set(parent_revs).difference(present_parents)
2835
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
2836
(rev_id,) for rev_id in absent_parents)
2837
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
2838
for parent_tree in self.source.revision_trees(parent_inv_ids):
2839
current_revision_id = parent_tree.get_revision_id()
2840
parents_parents_keys = parent_invs_keys_for_stacking[
2841
(current_revision_id,)]
2842
parents_parents = [key[-1] for key in parents_parents_keys]
2843
basis_id = _mod_revision.NULL_REVISION
2844
basis_tree = self.source.revision_tree(basis_id)
2845
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
2846
self.target.add_inventory_by_delta(
2847
basis_id, delta, current_revision_id, parents_parents)
2848
cache[current_revision_id] = parent_tree
2850
def _fetch_batch(self, revision_ids, basis_id, cache):
2851
"""Fetch across a few revisions.
2853
:param revision_ids: The revisions to copy
2854
:param basis_id: The revision_id of a tree that must be in cache, used
2855
as a basis for delta when no other base is available
2856
:param cache: A cache of RevisionTrees that we can use.
2857
:return: The revision_id of the last converted tree. The RevisionTree
2858
for it will be in cache
2860
# Walk though all revisions; get inventory deltas, copy referenced
2861
# texts that delta references, insert the delta, revision and
2863
root_keys_to_create = set()
2866
pending_revisions = []
2867
parent_map = self.source.get_parent_map(revision_ids)
2868
self._fetch_parent_invs_for_stacking(parent_map, cache)
2869
self.source._safe_to_return_from_cache = True
2870
for tree in self.source.revision_trees(revision_ids):
2871
# Find a inventory delta for this revision.
2872
# Find text entries that need to be copied, too.
2873
current_revision_id = tree.get_revision_id()
2874
parent_ids = parent_map.get(current_revision_id, ())
2875
parent_trees = self._get_trees(parent_ids, cache)
2876
possible_trees = list(parent_trees)
2877
if len(possible_trees) == 0:
2878
# There either aren't any parents, or the parents are ghosts,
2879
# so just use the last converted tree.
2880
possible_trees.append((basis_id, cache[basis_id]))
2881
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
2883
revision = self.source.get_revision(current_revision_id)
2884
pending_deltas.append((basis_id, delta,
2885
current_revision_id, revision.parent_ids))
2886
if self._converting_to_rich_root:
2887
self._revision_id_to_root_id[current_revision_id] = \
2889
# Determine which texts are in present in this revision but not in
2890
# any of the available parents.
2891
texts_possibly_new_in_tree = set()
2892
for old_path, new_path, file_id, entry in delta:
2893
if new_path is None:
2894
# This file_id isn't present in the new rev
2898
if not self.target.supports_rich_root():
2899
# The target doesn't support rich root, so we don't
2902
if self._converting_to_rich_root:
2903
# This can't be copied normally, we have to insert
2905
root_keys_to_create.add((file_id, entry.revision))
2908
texts_possibly_new_in_tree.add((file_id, entry.revision))
2909
for basis_id, basis_tree in possible_trees:
2910
basis_inv = basis_tree.inventory
2911
for file_key in list(texts_possibly_new_in_tree):
2912
file_id, file_revision = file_key
2914
entry = basis_inv[file_id]
2915
except errors.NoSuchId:
2917
if entry.revision == file_revision:
2918
texts_possibly_new_in_tree.remove(file_key)
2919
text_keys.update(texts_possibly_new_in_tree)
2920
pending_revisions.append(revision)
2921
cache[current_revision_id] = tree
2922
basis_id = current_revision_id
2923
self.source._safe_to_return_from_cache = False
2925
from_texts = self.source.texts
2926
to_texts = self.target.texts
2927
if root_keys_to_create:
2928
root_stream = _mod_fetch._new_root_data_stream(
2929
root_keys_to_create, self._revision_id_to_root_id, parent_map,
2931
to_texts.insert_record_stream(root_stream)
2932
to_texts.insert_record_stream(from_texts.get_record_stream(
2933
text_keys, self.target._format._fetch_order,
2934
not self.target._format._fetch_uses_deltas))
2935
# insert inventory deltas
2936
for delta in pending_deltas:
2937
self.target.add_inventory_by_delta(*delta)
2938
if self.target._fallback_repositories:
2939
# Make sure this stacked repository has all the parent inventories
2940
# for the new revisions that we are about to insert. We do this
2941
# before adding the revisions so that no revision is added until
2942
# all the inventories it may depend on are added.
2943
# Note that this is overzealous, as we may have fetched these in an
2946
revision_ids = set()
2947
for revision in pending_revisions:
2948
revision_ids.add(revision.revision_id)
2949
parent_ids.update(revision.parent_ids)
2950
parent_ids.difference_update(revision_ids)
2951
parent_ids.discard(_mod_revision.NULL_REVISION)
2952
parent_map = self.source.get_parent_map(parent_ids)
2953
# we iterate over parent_map and not parent_ids because we don't
2954
# want to try copying any revision which is a ghost
2955
for parent_tree in self.source.revision_trees(parent_map):
2956
current_revision_id = parent_tree.get_revision_id()
2957
parents_parents = parent_map[current_revision_id]
2958
possible_trees = self._get_trees(parents_parents, cache)
2959
if len(possible_trees) == 0:
2960
# There either aren't any parents, or the parents are
2961
# ghosts, so just use the last converted tree.
2962
possible_trees.append((basis_id, cache[basis_id]))
2963
basis_id, delta = self._get_delta_for_revision(parent_tree,
2964
parents_parents, possible_trees)
2965
self.target.add_inventory_by_delta(
2966
basis_id, delta, current_revision_id, parents_parents)
2967
# insert signatures and revisions
2968
for revision in pending_revisions:
2970
signature = self.source.get_signature_text(
2971
revision.revision_id)
2972
self.target.add_signature_text(revision.revision_id,
2974
except errors.NoSuchRevision:
2976
self.target.add_revision(revision.revision_id, revision)
2979
def _fetch_all_revisions(self, revision_ids, pb):
2980
"""Fetch everything for the list of revisions.
2982
:param revision_ids: The list of revisions to fetch. Must be in
2984
:param pb: A ProgressTask
2987
basis_id, basis_tree = self._get_basis(revision_ids[0])
2989
cache = lru_cache.LRUCache(100)
2990
cache[basis_id] = basis_tree
2991
del basis_tree # We don't want to hang on to it here
2995
for offset in range(0, len(revision_ids), batch_size):
2996
self.target.start_write_group()
2998
pb.update(gettext('Transferring revisions'), offset,
3000
batch = revision_ids[offset:offset+batch_size]
3001
basis_id = self._fetch_batch(batch, basis_id, cache)
3003
self.source._safe_to_return_from_cache = False
3004
self.target.abort_write_group()
3007
hint = self.target.commit_write_group()
3010
if hints and self.target._format.pack_compresses:
3011
self.target.pack(hint=hints)
3012
pb.update(gettext('Transferring revisions'), len(revision_ids),
3016
def fetch(self, revision_id=None, find_ghosts=False,
3018
"""See InterRepository.fetch()."""
3019
if fetch_spec is not None:
3020
revision_ids = fetch_spec.get_keys()
3023
if self.source._format.experimental:
3024
ui.ui_factory.show_user_warning('experimental_format_fetch',
3025
from_format=self.source._format,
3026
to_format=self.target._format)
3027
if (not self.source.supports_rich_root()
3028
and self.target.supports_rich_root()):
3029
self._converting_to_rich_root = True
3030
self._revision_id_to_root_id = {}
3032
self._converting_to_rich_root = False
3033
# See <https://launchpad.net/bugs/456077> asking for a warning here
3034
if self.source._format.network_name() != self.target._format.network_name():
3035
ui.ui_factory.show_user_warning('cross_format_fetch',
3036
from_format=self.source._format,
3037
to_format=self.target._format)
3038
if revision_ids is None:
3040
search_revision_ids = [revision_id]
3042
search_revision_ids = None
3043
revision_ids = self.target.search_missing_revision_ids(self.source,
3044
revision_ids=search_revision_ids,
3045
find_ghosts=find_ghosts).get_keys()
3046
if not revision_ids:
3048
revision_ids = tsort.topo_sort(
3049
self.source.get_graph().get_parent_map(revision_ids))
3050
if not revision_ids:
3052
# Walk though all revisions; get inventory deltas, copy referenced
3053
# texts that delta references, insert the delta, revision and
3055
pb = ui.ui_factory.nested_progress_bar()
3057
self._fetch_all_revisions(revision_ids, pb)
3060
return len(revision_ids), 0
3062
def _get_basis(self, first_revision_id):
3063
"""Get a revision and tree which exists in the target.
3065
This assumes that first_revision_id is selected for transmission
3066
because all other ancestors are already present. If we can't find an
3067
ancestor we fall back to NULL_REVISION since we know that is safe.
3069
:return: (basis_id, basis_tree)
3071
first_rev = self.source.get_revision(first_revision_id)
3073
basis_id = first_rev.parent_ids[0]
3074
# only valid as a basis if the target has it
3075
self.target.get_revision(basis_id)
3076
# Try to get a basis tree - if it's a ghost it will hit the
3077
# NoSuchRevision case.
3078
basis_tree = self.source.revision_tree(basis_id)
3079
except (IndexError, errors.NoSuchRevision):
3080
basis_id = _mod_revision.NULL_REVISION
3081
basis_tree = self.source.revision_tree(basis_id)
3082
return basis_id, basis_tree
3085
class InterSameDataRepository(InterVersionedFileRepository):
3086
"""Code for converting between repositories that represent the same data.
3088
Data format and model must match for this to work.
3092
def _get_repo_format_to_test(self):
3093
"""Repository format for testing with.
3095
InterSameData can pull from subtree to subtree and from non-subtree to
3096
non-subtree, so we test this with the richest repository format.
3098
from bzrlib.repofmt import knitrepo
3099
return knitrepo.RepositoryFormatKnit3()
3102
def is_compatible(source, target):
3104
InterRepository._same_model(source, target) and
3105
source._format.supports_full_versioned_files and
3106
target._format.supports_full_versioned_files)
3109
InterRepository.register_optimiser(InterVersionedFileRepository)
3110
InterRepository.register_optimiser(InterDifferingSerializer)
3111
InterRepository.register_optimiser(InterSameDataRepository)
3114
def install_revisions(repository, iterable, num_revisions=None, pb=None):
3115
"""Install all revision data into a repository.
3117
Accepts an iterable of revision, tree, signature tuples. The signature
3120
repository.start_write_group()
3122
inventory_cache = lru_cache.LRUCache(10)
3123
for n, (revision, revision_tree, signature) in enumerate(iterable):
3124
_install_revision(repository, revision, revision_tree, signature,
3127
pb.update(gettext('Transferring revisions'), n + 1, num_revisions)
3129
repository.abort_write_group()
3132
repository.commit_write_group()
3135
def _install_revision(repository, rev, revision_tree, signature,
3137
"""Install all revision data into a repository."""
3138
present_parents = []
3140
for p_id in rev.parent_ids:
3141
if repository.has_revision(p_id):
3142
present_parents.append(p_id)
3143
parent_trees[p_id] = repository.revision_tree(p_id)
3145
parent_trees[p_id] = repository.revision_tree(
3146
_mod_revision.NULL_REVISION)
3148
inv = revision_tree.inventory
3149
entries = inv.iter_entries()
3150
# backwards compatibility hack: skip the root id.
3151
if not repository.supports_rich_root():
3152
path, root = entries.next()
3153
if root.revision != rev.revision_id:
3154
raise errors.IncompatibleRevision(repr(repository))
3156
for path, ie in entries:
3157
text_keys[(ie.file_id, ie.revision)] = ie
3158
text_parent_map = repository.texts.get_parent_map(text_keys)
3159
missing_texts = set(text_keys) - set(text_parent_map)
3160
# Add the texts that are not already present
3161
for text_key in missing_texts:
3162
ie = text_keys[text_key]
3164
# FIXME: TODO: The following loop overlaps/duplicates that done by
3165
# commit to determine parents. There is a latent/real bug here where
3166
# the parents inserted are not those commit would do - in particular
3167
# they are not filtered by heads(). RBC, AB
3168
for revision, tree in parent_trees.iteritems():
3169
if not tree.has_id(ie.file_id):
3171
parent_id = tree.get_file_revision(ie.file_id)
3172
if parent_id in text_parents:
3174
text_parents.append((ie.file_id, parent_id))
3175
lines = revision_tree.get_file(ie.file_id).readlines()
3176
repository.texts.add_lines(text_key, text_parents, lines)
3178
# install the inventory
3179
if repository._format._commit_inv_deltas and len(rev.parent_ids):
3180
# Cache this inventory
3181
inventory_cache[rev.revision_id] = inv
3183
basis_inv = inventory_cache[rev.parent_ids[0]]
3185
repository.add_inventory(rev.revision_id, inv, present_parents)
3187
delta = inv._make_delta(basis_inv)
3188
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
3189
rev.revision_id, present_parents)
3191
repository.add_inventory(rev.revision_id, inv, present_parents)
3192
except errors.RevisionAlreadyPresent:
3194
if signature is not None:
3195
repository.add_signature_text(rev.revision_id, signature)
3196
repository.add_revision(rev.revision_id, rev, inv)
3199
def install_revision(repository, rev, revision_tree):
3200
"""Install all revision data into a repository."""
3201
install_revisions(repository, [(rev, revision_tree, None)])
added
removed
bzrlib/vf_repository.py
