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(), """
32
revision as _mod_revision,
33
serializer as _mod_serializer,
41
from bzrlib.recordcounter import RecordCounter
42
from bzrlib.revisiontree import InventoryRevisionTree
43
from bzrlib.testament import Testament
49
from bzrlib.decorators import (
54
from bzrlib.inventory import (
61
from bzrlib.repository import (
65
MetaDirRepositoryFormat,
70
from bzrlib.trace import (
75
class VersionedFileRepositoryFormat(RepositoryFormat):
76
"""Base class for all repository formats that are VersionedFiles-based."""
78
supports_full_versioned_files = True
80
# Should commit add an inventory, or an inventory delta to the repository.
81
_commit_inv_deltas = True
82
# What order should fetch operations request streams in?
83
# The default is unordered as that is the cheapest for an origin to
85
_fetch_order = 'unordered'
86
# Does this repository format use deltas that can be fetched as-deltas ?
87
# (E.g. knits, where the knit deltas can be transplanted intact.
88
# We default to False, which will ensure that enough data to get
89
# a full text out of any fetch stream will be grabbed.
90
_fetch_uses_deltas = False
93
class VersionedFileCommitBuilder(CommitBuilder):
94
"""Commit builder implementation for versioned files based repositories.
97
# this commit builder supports the record_entry_contents interface
98
supports_record_entry_contents = True
100
# the default CommitBuilder does not manage trees whose root is versioned.
101
_versioned_root = False
103
def __init__(self, repository, parents, config, timestamp=None,
104
timezone=None, committer=None, revprops=None,
105
revision_id=None, lossy=False):
106
super(VersionedFileCommitBuilder, self).__init__(repository,
107
parents, config, timestamp, timezone, committer, revprops,
110
basis_id = self.parents[0]
112
basis_id = _mod_revision.NULL_REVISION
113
self.basis_delta_revision = basis_id
114
self.new_inventory = Inventory(None)
115
self._basis_delta = []
116
self.__heads = graph.HeadsCache(repository.get_graph()).heads
117
# memo'd check for no-op commits.
118
self._any_changes = False
119
# API compatibility, older code that used CommitBuilder did not call
120
# .record_delete(), which means the delta that is computed would not be
121
# valid. Callers that will call record_delete() should call
122
# .will_record_deletes() to indicate that.
123
self._recording_deletes = False
125
def will_record_deletes(self):
126
"""Tell the commit builder that deletes are being notified.
128
This enables the accumulation of an inventory delta; for the resulting
129
commit to be valid, deletes against the basis MUST be recorded via
130
builder.record_delete().
132
self._recording_deletes = True
134
def any_changes(self):
135
"""Return True if any entries were changed.
137
This includes merge-only changes. It is the core for the --unchanged
140
:return: True if any changes have occured.
142
return self._any_changes
144
def _ensure_fallback_inventories(self):
145
"""Ensure that appropriate inventories are available.
147
This only applies to repositories that are stacked, and is about
148
enusring the stacking invariants. Namely, that for any revision that is
149
present, we either have all of the file content, or we have the parent
150
inventory and the delta file content.
152
if not self.repository._fallback_repositories:
154
if not self.repository._format.supports_chks:
155
raise errors.BzrError("Cannot commit directly to a stacked branch"
156
" in pre-2a formats. See "
157
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
158
# This is a stacked repo, we need to make sure we have the parent
159
# inventories for the parents.
160
parent_keys = [(p,) for p in self.parents]
161
parent_map = self.repository.inventories._index.get_parent_map(parent_keys)
162
missing_parent_keys = set([pk for pk in parent_keys
163
if pk not in parent_map])
164
fallback_repos = list(reversed(self.repository._fallback_repositories))
165
missing_keys = [('inventories', pk[0])
166
for pk in missing_parent_keys]
168
while missing_keys and fallback_repos:
169
fallback_repo = fallback_repos.pop()
170
source = fallback_repo._get_source(self.repository._format)
171
sink = self.repository._get_sink()
172
stream = source.get_stream_for_missing_keys(missing_keys)
173
missing_keys = sink.insert_stream_without_locking(stream,
174
self.repository._format)
176
raise errors.BzrError('Unable to fill in parent inventories for a'
179
def commit(self, message):
180
"""Make the actual commit.
182
:return: The revision id of the recorded revision.
184
self._validate_unicode_text(message, 'commit message')
185
rev = _mod_revision.Revision(
186
timestamp=self._timestamp,
187
timezone=self._timezone,
188
committer=self._committer,
190
inventory_sha1=self.inv_sha1,
191
revision_id=self._new_revision_id,
192
properties=self._revprops)
193
rev.parent_ids = self.parents
194
self.repository.add_revision(self._new_revision_id, rev,
195
self.new_inventory, self._config)
196
self._ensure_fallback_inventories()
197
self.repository.commit_write_group()
198
return self._new_revision_id
201
"""Abort the commit that is being built.
203
self.repository.abort_write_group()
205
def revision_tree(self):
206
"""Return the tree that was just committed.
208
After calling commit() this can be called to get a
209
RevisionTree representing the newly committed tree. This is
210
preferred to calling Repository.revision_tree() because that may
211
require deserializing the inventory, while we already have a copy in
214
if self.new_inventory is None:
215
self.new_inventory = self.repository.get_inventory(
216
self._new_revision_id)
217
return InventoryRevisionTree(self.repository, self.new_inventory,
218
self._new_revision_id)
220
def finish_inventory(self):
221
"""Tell the builder that the inventory is finished.
223
:return: The inventory id in the repository, which can be used with
224
repository.get_inventory.
226
if self.new_inventory is None:
227
# an inventory delta was accumulated without creating a new
229
basis_id = self.basis_delta_revision
230
# We ignore the 'inventory' returned by add_inventory_by_delta
231
# because self.new_inventory is used to hint to the rest of the
232
# system what code path was taken
233
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
234
basis_id, self._basis_delta, self._new_revision_id,
237
if self.new_inventory.root is None:
238
raise AssertionError('Root entry should be supplied to'
239
' record_entry_contents, as of bzr 0.10.')
240
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
241
self.new_inventory.revision_id = self._new_revision_id
242
self.inv_sha1 = self.repository.add_inventory(
243
self._new_revision_id,
247
return self._new_revision_id
249
def _check_root(self, ie, parent_invs, tree):
250
"""Helper for record_entry_contents.
252
:param ie: An entry being added.
253
:param parent_invs: The inventories of the parent revisions of the
255
:param tree: The tree that is being committed.
257
# In this revision format, root entries have no knit or weave When
258
# serializing out to disk and back in root.revision is always
260
ie.revision = self._new_revision_id
262
def _require_root_change(self, tree):
263
"""Enforce an appropriate root object change.
265
This is called once when record_iter_changes is called, if and only if
266
the root was not in the delta calculated by record_iter_changes.
268
:param tree: The tree which is being committed.
270
if len(self.parents) == 0:
271
raise errors.RootMissing()
272
entry = entry_factory['directory'](tree.path2id(''), '',
274
entry.revision = self._new_revision_id
275
self._basis_delta.append(('', '', entry.file_id, entry))
277
def _get_delta(self, ie, basis_inv, path):
278
"""Get a delta against the basis inventory for ie."""
279
if ie.file_id not in basis_inv:
281
result = (None, path, ie.file_id, ie)
282
self._basis_delta.append(result)
284
elif ie != basis_inv[ie.file_id]:
286
# TODO: avoid tis id2path call.
287
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
288
self._basis_delta.append(result)
294
def _heads(self, file_id, revision_ids):
295
"""Calculate the graph heads for revision_ids in the graph of file_id.
297
This can use either a per-file graph or a global revision graph as we
298
have an identity relationship between the two graphs.
300
return self.__heads(revision_ids)
302
def get_basis_delta(self):
303
"""Return the complete inventory delta versus the basis inventory.
305
This has been built up with the calls to record_delete and
306
record_entry_contents. The client must have already called
307
will_record_deletes() to indicate that they will be generating a
310
:return: An inventory delta, suitable for use with apply_delta, or
311
Repository.add_inventory_by_delta, etc.
313
if not self._recording_deletes:
314
raise AssertionError("recording deletes not activated.")
315
return self._basis_delta
317
def record_delete(self, path, file_id):
318
"""Record that a delete occured against a basis tree.
320
This is an optional API - when used it adds items to the basis_delta
321
being accumulated by the commit builder. It cannot be called unless the
322
method will_record_deletes() has been called to inform the builder that
323
a delta is being supplied.
325
:param path: The path of the thing deleted.
326
:param file_id: The file id that was deleted.
328
if not self._recording_deletes:
329
raise AssertionError("recording deletes not activated.")
330
delta = (path, None, file_id, None)
331
self._basis_delta.append(delta)
332
self._any_changes = True
335
def record_entry_contents(self, ie, parent_invs, path, tree,
337
"""Record the content of ie from tree into the commit if needed.
339
Side effect: sets ie.revision when unchanged
341
:param ie: An inventory entry present in the commit.
342
:param parent_invs: The inventories of the parent revisions of the
344
:param path: The path the entry is at in the tree.
345
:param tree: The tree which contains this entry and should be used to
347
:param content_summary: Summary data from the tree about the paths
348
content - stat, length, exec, sha/link target. This is only
349
accessed when the entry has a revision of None - that is when it is
350
a candidate to commit.
351
:return: A tuple (change_delta, version_recorded, fs_hash).
352
change_delta is an inventory_delta change for this entry against
353
the basis tree of the commit, or None if no change occured against
355
version_recorded is True if a new version of the entry has been
356
recorded. For instance, committing a merge where a file was only
357
changed on the other side will return (delta, False).
358
fs_hash is either None, or the hash details for the path (currently
359
a tuple of the contents sha1 and the statvalue returned by
360
tree.get_file_with_stat()).
362
if self.new_inventory.root is None:
363
if ie.parent_id is not None:
364
raise errors.RootMissing()
365
self._check_root(ie, parent_invs, tree)
366
if ie.revision is None:
367
kind = content_summary[0]
369
# ie is carried over from a prior commit
371
# XXX: repository specific check for nested tree support goes here - if
372
# the repo doesn't want nested trees we skip it ?
373
if (kind == 'tree-reference' and
374
not self.repository._format.supports_tree_reference):
375
# mismatch between commit builder logic and repository:
376
# this needs the entry creation pushed down into the builder.
377
raise NotImplementedError('Missing repository subtree support.')
378
self.new_inventory.add(ie)
380
# TODO: slow, take it out of the inner loop.
382
basis_inv = parent_invs[0]
384
basis_inv = Inventory(root_id=None)
386
# ie.revision is always None if the InventoryEntry is considered
387
# for committing. We may record the previous parents revision if the
388
# content is actually unchanged against a sole head.
389
if ie.revision is not None:
390
if not self._versioned_root and path == '':
391
# repositories that do not version the root set the root's
392
# revision to the new commit even when no change occurs (more
393
# specifically, they do not record a revision on the root; and
394
# the rev id is assigned to the root during deserialisation -
395
# this masks when a change may have occurred against the basis.
396
# To match this we always issue a delta, because the revision
397
# of the root will always be changing.
398
if ie.file_id in basis_inv:
399
delta = (basis_inv.id2path(ie.file_id), path,
403
delta = (None, path, ie.file_id, ie)
404
self._basis_delta.append(delta)
405
return delta, False, None
407
# we don't need to commit this, because the caller already
408
# determined that an existing revision of this file is
409
# appropriate. If it's not being considered for committing then
410
# it and all its parents to the root must be unaltered so
411
# no-change against the basis.
412
if ie.revision == self._new_revision_id:
413
raise AssertionError("Impossible situation, a skipped "
414
"inventory entry (%r) claims to be modified in this "
415
"commit (%r).", (ie, self._new_revision_id))
416
return None, False, None
417
# XXX: Friction: parent_candidates should return a list not a dict
418
# so that we don't have to walk the inventories again.
419
parent_candiate_entries = ie.parent_candidates(parent_invs)
420
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
422
for inv in parent_invs:
423
if ie.file_id in inv:
424
old_rev = inv[ie.file_id].revision
425
if old_rev in head_set:
426
heads.append(inv[ie.file_id].revision)
427
head_set.remove(inv[ie.file_id].revision)
430
# now we check to see if we need to write a new record to the
432
# We write a new entry unless there is one head to the ancestors, and
433
# the kind-derived content is unchanged.
435
# Cheapest check first: no ancestors, or more the one head in the
436
# ancestors, we write a new node.
440
# There is a single head, look it up for comparison
441
parent_entry = parent_candiate_entries[heads[0]]
442
# if the non-content specific data has changed, we'll be writing a
444
if (parent_entry.parent_id != ie.parent_id or
445
parent_entry.name != ie.name):
447
# now we need to do content specific checks:
449
# if the kind changed the content obviously has
450
if kind != parent_entry.kind:
452
# Stat cache fingerprint feedback for the caller - None as we usually
453
# don't generate one.
456
if content_summary[2] is None:
457
raise ValueError("Files must not have executable = None")
459
# We can't trust a check of the file length because of content
461
if (# if the exec bit has changed we have to store:
462
parent_entry.executable != content_summary[2]):
464
elif parent_entry.text_sha1 == content_summary[3]:
465
# all meta and content is unchanged (using a hash cache
466
# hit to check the sha)
467
ie.revision = parent_entry.revision
468
ie.text_size = parent_entry.text_size
469
ie.text_sha1 = parent_entry.text_sha1
470
ie.executable = parent_entry.executable
471
return self._get_delta(ie, basis_inv, path), False, None
473
# Either there is only a hash change(no hash cache entry,
474
# or same size content change), or there is no change on
476
# Provide the parent's hash to the store layer, so that the
477
# content is unchanged we will not store a new node.
478
nostore_sha = parent_entry.text_sha1
480
# We want to record a new node regardless of the presence or
481
# absence of a content change in the file.
483
ie.executable = content_summary[2]
484
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
486
text = file_obj.read()
490
ie.text_sha1, ie.text_size = self._add_text_to_weave(
491
ie.file_id, text, heads, nostore_sha)
492
# Let the caller know we generated a stat fingerprint.
493
fingerprint = (ie.text_sha1, stat_value)
494
except errors.ExistingContent:
495
# Turns out that the file content was unchanged, and we were
496
# only going to store a new node if it was changed. Carry over
498
ie.revision = parent_entry.revision
499
ie.text_size = parent_entry.text_size
500
ie.text_sha1 = parent_entry.text_sha1
501
ie.executable = parent_entry.executable
502
return self._get_delta(ie, basis_inv, path), False, None
503
elif kind == 'directory':
505
# all data is meta here, nothing specific to directory, so
507
ie.revision = parent_entry.revision
508
return self._get_delta(ie, basis_inv, path), False, None
509
self._add_text_to_weave(ie.file_id, '', heads, None)
510
elif kind == 'symlink':
511
current_link_target = content_summary[3]
513
# symlink target is not generic metadata, check if it has
515
if current_link_target != parent_entry.symlink_target:
518
# unchanged, carry over.
519
ie.revision = parent_entry.revision
520
ie.symlink_target = parent_entry.symlink_target
521
return self._get_delta(ie, basis_inv, path), False, None
522
ie.symlink_target = current_link_target
523
self._add_text_to_weave(ie.file_id, '', heads, None)
524
elif kind == 'tree-reference':
526
if content_summary[3] != parent_entry.reference_revision:
529
# unchanged, carry over.
530
ie.reference_revision = parent_entry.reference_revision
531
ie.revision = parent_entry.revision
532
return self._get_delta(ie, basis_inv, path), False, None
533
ie.reference_revision = content_summary[3]
534
if ie.reference_revision is None:
535
raise AssertionError("invalid content_summary for nested tree: %r"
536
% (content_summary,))
537
self._add_text_to_weave(ie.file_id, '', heads, None)
539
raise NotImplementedError('unknown kind')
540
ie.revision = self._new_revision_id
541
# The initial commit adds a root directory, but this in itself is not
542
# a worthwhile commit.
543
if (self.basis_delta_revision != _mod_revision.NULL_REVISION or
545
self._any_changes = True
546
return self._get_delta(ie, basis_inv, path), True, fingerprint
548
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
549
_entry_factory=entry_factory):
550
"""Record a new tree via iter_changes.
552
:param tree: The tree to obtain text contents from for changed objects.
553
:param basis_revision_id: The revision id of the tree the iter_changes
554
has been generated against. Currently assumed to be the same
555
as self.parents[0] - if it is not, errors may occur.
556
:param iter_changes: An iter_changes iterator with the changes to apply
557
to basis_revision_id. The iterator must not include any items with
558
a current kind of None - missing items must be either filtered out
559
or errored-on beefore record_iter_changes sees the item.
560
:param _entry_factory: Private method to bind entry_factory locally for
562
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
565
# Create an inventory delta based on deltas between all the parents and
566
# deltas between all the parent inventories. We use inventory delta's
567
# between the inventory objects because iter_changes masks
568
# last-changed-field only changes.
570
# file_id -> change map, change is fileid, paths, changed, versioneds,
571
# parents, names, kinds, executables
573
# {file_id -> revision_id -> inventory entry, for entries in parent
574
# trees that are not parents[0]
578
revtrees = list(self.repository.revision_trees(self.parents))
579
except errors.NoSuchRevision:
580
# one or more ghosts, slow path.
582
for revision_id in self.parents:
584
revtrees.append(self.repository.revision_tree(revision_id))
585
except errors.NoSuchRevision:
587
basis_revision_id = _mod_revision.NULL_REVISION
589
revtrees.append(self.repository.revision_tree(
590
_mod_revision.NULL_REVISION))
591
# The basis inventory from a repository
593
basis_inv = revtrees[0].inventory
595
basis_inv = self.repository.revision_tree(
596
_mod_revision.NULL_REVISION).inventory
597
if len(self.parents) > 0:
598
if basis_revision_id != self.parents[0] and not ghost_basis:
600
"arbitrary basis parents not yet supported with merges")
601
for revtree in revtrees[1:]:
602
for change in revtree.inventory._make_delta(basis_inv):
603
if change[1] is None:
604
# Not present in this parent.
606
if change[2] not in merged_ids:
607
if change[0] is not None:
608
basis_entry = basis_inv[change[2]]
609
merged_ids[change[2]] = [
611
basis_entry.revision,
614
parent_entries[change[2]] = {
616
basis_entry.revision:basis_entry,
618
change[3].revision:change[3],
621
merged_ids[change[2]] = [change[3].revision]
622
parent_entries[change[2]] = {change[3].revision:change[3]}
624
merged_ids[change[2]].append(change[3].revision)
625
parent_entries[change[2]][change[3].revision] = change[3]
628
# Setup the changes from the tree:
629
# changes maps file_id -> (change, [parent revision_ids])
631
for change in iter_changes:
632
# This probably looks up in basis_inv way to much.
633
if change[1][0] is not None:
634
head_candidate = [basis_inv[change[0]].revision]
637
changes[change[0]] = change, merged_ids.get(change[0],
639
unchanged_merged = set(merged_ids) - set(changes)
640
# Extend the changes dict with synthetic changes to record merges of
642
for file_id in unchanged_merged:
643
# Record a merged version of these items that did not change vs the
644
# basis. This can be either identical parallel changes, or a revert
645
# of a specific file after a merge. The recorded content will be
646
# that of the current tree (which is the same as the basis), but
647
# the per-file graph will reflect a merge.
648
# NB:XXX: We are reconstructing path information we had, this
649
# should be preserved instead.
650
# inv delta change: (file_id, (path_in_source, path_in_target),
651
# changed_content, versioned, parent, name, kind,
654
basis_entry = basis_inv[file_id]
655
except errors.NoSuchId:
656
# a change from basis->some_parents but file_id isn't in basis
657
# so was new in the merge, which means it must have changed
658
# from basis -> current, and as it hasn't the add was reverted
659
# by the user. So we discard this change.
663
(basis_inv.id2path(file_id), tree.id2path(file_id)),
665
(basis_entry.parent_id, basis_entry.parent_id),
666
(basis_entry.name, basis_entry.name),
667
(basis_entry.kind, basis_entry.kind),
668
(basis_entry.executable, basis_entry.executable))
669
changes[file_id] = (change, merged_ids[file_id])
670
# changes contains tuples with the change and a set of inventory
671
# candidates for the file.
673
# old_path, new_path, file_id, new_inventory_entry
674
seen_root = False # Is the root in the basis delta?
675
inv_delta = self._basis_delta
676
modified_rev = self._new_revision_id
677
for change, head_candidates in changes.values():
678
if change[3][1]: # versioned in target.
679
# Several things may be happening here:
680
# We may have a fork in the per-file graph
681
# - record a change with the content from tree
682
# We may have a change against < all trees
683
# - carry over the tree that hasn't changed
684
# We may have a change against all trees
685
# - record the change with the content from tree
688
entry = _entry_factory[kind](file_id, change[5][1],
690
head_set = self._heads(change[0], set(head_candidates))
693
for head_candidate in head_candidates:
694
if head_candidate in head_set:
695
heads.append(head_candidate)
696
head_set.remove(head_candidate)
699
# Could be a carry-over situation:
700
parent_entry_revs = parent_entries.get(file_id, None)
701
if parent_entry_revs:
702
parent_entry = parent_entry_revs.get(heads[0], None)
705
if parent_entry is None:
706
# The parent iter_changes was called against is the one
707
# that is the per-file head, so any change is relevant
708
# iter_changes is valid.
709
carry_over_possible = False
711
# could be a carry over situation
712
# A change against the basis may just indicate a merge,
713
# we need to check the content against the source of the
714
# merge to determine if it was changed after the merge
716
if (parent_entry.kind != entry.kind or
717
parent_entry.parent_id != entry.parent_id or
718
parent_entry.name != entry.name):
719
# Metadata common to all entries has changed
720
# against per-file parent
721
carry_over_possible = False
723
carry_over_possible = True
724
# per-type checks for changes against the parent_entry
727
# Cannot be a carry-over situation
728
carry_over_possible = False
729
# Populate the entry in the delta
731
# XXX: There is still a small race here: If someone reverts the content of a file
732
# after iter_changes examines and decides it has changed,
733
# we will unconditionally record a new version even if some
734
# other process reverts it while commit is running (with
735
# the revert happening after iter_changes did its
738
entry.executable = True
740
entry.executable = False
741
if (carry_over_possible and
742
parent_entry.executable == entry.executable):
743
# Check the file length, content hash after reading
745
nostore_sha = parent_entry.text_sha1
748
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
750
text = file_obj.read()
754
entry.text_sha1, entry.text_size = self._add_text_to_weave(
755
file_id, text, heads, nostore_sha)
756
yield file_id, change[1][1], (entry.text_sha1, stat_value)
757
except errors.ExistingContent:
758
# No content change against a carry_over parent
759
# Perhaps this should also yield a fs hash update?
761
entry.text_size = parent_entry.text_size
762
entry.text_sha1 = parent_entry.text_sha1
763
elif kind == 'symlink':
765
entry.symlink_target = tree.get_symlink_target(file_id)
766
if (carry_over_possible and
767
parent_entry.symlink_target == entry.symlink_target):
770
self._add_text_to_weave(change[0], '', heads, None)
771
elif kind == 'directory':
772
if carry_over_possible:
775
# Nothing to set on the entry.
776
# XXX: split into the Root and nonRoot versions.
777
if change[1][1] != '' or self.repository.supports_rich_root():
778
self._add_text_to_weave(change[0], '', heads, None)
779
elif kind == 'tree-reference':
780
if not self.repository._format.supports_tree_reference:
781
# This isn't quite sane as an error, but we shouldn't
782
# ever see this code path in practice: tree's don't
783
# permit references when the repo doesn't support tree
785
raise errors.UnsupportedOperation(tree.add_reference,
787
reference_revision = tree.get_reference_revision(change[0])
788
entry.reference_revision = reference_revision
789
if (carry_over_possible and
790
parent_entry.reference_revision == reference_revision):
793
self._add_text_to_weave(change[0], '', heads, None)
795
raise AssertionError('unknown kind %r' % kind)
797
entry.revision = modified_rev
799
entry.revision = parent_entry.revision
802
new_path = change[1][1]
803
inv_delta.append((change[1][0], new_path, change[0], entry))
806
self.new_inventory = None
807
# The initial commit adds a root directory, but this in itself is not
808
# a worthwhile commit.
809
if ((len(inv_delta) > 0 and basis_revision_id != _mod_revision.NULL_REVISION) or
810
(len(inv_delta) > 1 and basis_revision_id == _mod_revision.NULL_REVISION)):
811
# This should perhaps be guarded by a check that the basis we
812
# commit against is the basis for the commit and if not do a delta
814
self._any_changes = True
816
# housekeeping root entry changes do not affect no-change commits.
817
self._require_root_change(tree)
818
self.basis_delta_revision = basis_revision_id
820
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
821
parent_keys = tuple([(file_id, parent) for parent in parents])
822
return self.repository.texts._add_text(
823
(file_id, self._new_revision_id), parent_keys, new_text,
824
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
827
class VersionedFileRootCommitBuilder(VersionedFileCommitBuilder):
828
"""This commitbuilder actually records the root id"""
830
# the root entry gets versioned properly by this builder.
831
_versioned_root = True
833
def _check_root(self, ie, parent_invs, tree):
834
"""Helper for record_entry_contents.
836
:param ie: An entry being added.
837
:param parent_invs: The inventories of the parent revisions of the
839
:param tree: The tree that is being committed.
842
def _require_root_change(self, tree):
843
"""Enforce an appropriate root object change.
845
This is called once when record_iter_changes is called, if and only if
846
the root was not in the delta calculated by record_iter_changes.
848
:param tree: The tree which is being committed.
850
# versioned roots do not change unless the tree found a change.
853
class VersionedFileRepository(Repository):
854
"""Repository holding history for one or more branches.
856
The repository holds and retrieves historical information including
857
revisions and file history. It's normally accessed only by the Branch,
858
which views a particular line of development through that history.
860
The Repository builds on top of some byte storage facilies (the revisions,
861
signatures, inventories, texts and chk_bytes attributes) and a Transport,
862
which respectively provide byte storage and a means to access the (possibly
865
The byte storage facilities are addressed via tuples, which we refer to
866
as 'keys' throughout the code base. Revision_keys, inventory_keys and
867
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
868
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
869
byte string made up of a hash identifier and a hash value.
870
We use this interface because it allows low friction with the underlying
871
code that implements disk indices, network encoding and other parts of
874
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
875
the serialised revisions for the repository. This can be used to obtain
876
revision graph information or to access raw serialised revisions.
877
The result of trying to insert data into the repository via this store
878
is undefined: it should be considered read-only except for implementors
880
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
881
the serialised signatures for the repository. This can be used to
882
obtain access to raw serialised signatures. The result of trying to
883
insert data into the repository via this store is undefined: it should
884
be considered read-only except for implementors of repositories.
885
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
886
the serialised inventories for the repository. This can be used to
887
obtain unserialised inventories. The result of trying to insert data
888
into the repository via this store is undefined: it should be
889
considered read-only except for implementors of repositories.
890
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
891
texts of files and directories for the repository. This can be used to
892
obtain file texts or file graphs. Note that Repository.iter_file_bytes
893
is usually a better interface for accessing file texts.
894
The result of trying to insert data into the repository via this store
895
is undefined: it should be considered read-only except for implementors
897
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
898
any data the repository chooses to store or have indexed by its hash.
899
The result of trying to insert data into the repository via this store
900
is undefined: it should be considered read-only except for implementors
902
:ivar _transport: Transport for file access to repository, typically
903
pointing to .bzr/repository.
906
# What class to use for a CommitBuilder. Often it's simpler to change this
907
# in a Repository class subclass rather than to override
908
# get_commit_builder.
909
_commit_builder_class = VersionedFileCommitBuilder
911
def add_fallback_repository(self, repository):
912
"""Add a repository to use for looking up data not held locally.
914
:param repository: A repository.
916
if not self._format.supports_external_lookups:
917
raise errors.UnstackableRepositoryFormat(self._format, self.base)
919
# This repository will call fallback.unlock() when we transition to
920
# the unlocked state, so we make sure to increment the lock count
921
repository.lock_read()
922
self._check_fallback_repository(repository)
923
self._fallback_repositories.append(repository)
924
self.texts.add_fallback_versioned_files(repository.texts)
925
self.inventories.add_fallback_versioned_files(repository.inventories)
926
self.revisions.add_fallback_versioned_files(repository.revisions)
927
self.signatures.add_fallback_versioned_files(repository.signatures)
928
if self.chk_bytes is not None:
929
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
931
@only_raises(errors.LockNotHeld, errors.LockBroken)
933
super(VersionedFileRepository, self).unlock()
934
if self.control_files._lock_count == 0:
935
self._inventory_entry_cache.clear()
937
def add_inventory(self, revision_id, inv, parents):
938
"""Add the inventory inv to the repository as revision_id.
940
:param parents: The revision ids of the parents that revision_id
941
is known to have and are in the repository already.
943
:returns: The validator(which is a sha1 digest, though what is sha'd is
944
repository format specific) of the serialized inventory.
946
if not self.is_in_write_group():
947
raise AssertionError("%r not in write group" % (self,))
948
_mod_revision.check_not_reserved_id(revision_id)
949
if not (inv.revision_id is None or inv.revision_id == revision_id):
950
raise AssertionError(
951
"Mismatch between inventory revision"
952
" id and insertion revid (%r, %r)"
953
% (inv.revision_id, revision_id))
955
raise errors.RootMissing()
956
return self._add_inventory_checked(revision_id, inv, parents)
958
def _add_inventory_checked(self, revision_id, inv, parents):
959
"""Add inv to the repository after checking the inputs.
961
This function can be overridden to allow different inventory styles.
963
:seealso: add_inventory, for the contract.
965
inv_lines = self._serializer.write_inventory_to_lines(inv)
966
return self._inventory_add_lines(revision_id, parents,
967
inv_lines, check_content=False)
969
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
970
parents, basis_inv=None, propagate_caches=False):
971
"""Add a new inventory expressed as a delta against another revision.
973
See the inventory developers documentation for the theory behind
976
:param basis_revision_id: The inventory id the delta was created
977
against. (This does not have to be a direct parent.)
978
:param delta: The inventory delta (see Inventory.apply_delta for
980
:param new_revision_id: The revision id that the inventory is being
982
:param parents: The revision ids of the parents that revision_id is
983
known to have and are in the repository already. These are supplied
984
for repositories that depend on the inventory graph for revision
985
graph access, as well as for those that pun ancestry with delta
987
:param basis_inv: The basis inventory if it is already known,
989
:param propagate_caches: If True, the caches for this inventory are
990
copied to and updated for the result if possible.
992
:returns: (validator, new_inv)
993
The validator(which is a sha1 digest, though what is sha'd is
994
repository format specific) of the serialized inventory, and the
997
if not self.is_in_write_group():
998
raise AssertionError("%r not in write group" % (self,))
999
_mod_revision.check_not_reserved_id(new_revision_id)
1000
basis_tree = self.revision_tree(basis_revision_id)
1001
basis_tree.lock_read()
1003
# Note that this mutates the inventory of basis_tree, which not all
1004
# inventory implementations may support: A better idiom would be to
1005
# return a new inventory, but as there is no revision tree cache in
1006
# repository this is safe for now - RBC 20081013
1007
if basis_inv is None:
1008
basis_inv = basis_tree.inventory
1009
basis_inv.apply_delta(delta)
1010
basis_inv.revision_id = new_revision_id
1011
return (self.add_inventory(new_revision_id, basis_inv, parents),
1016
def _inventory_add_lines(self, revision_id, parents, lines,
1017
check_content=True):
1018
"""Store lines in inv_vf and return the sha1 of the inventory."""
1019
parents = [(parent,) for parent in parents]
1020
result = self.inventories.add_lines((revision_id,), parents, lines,
1021
check_content=check_content)[0]
1022
self.inventories._access.flush()
1025
def add_revision(self, revision_id, rev, inv=None, config=None):
1026
"""Add rev to the revision store as revision_id.
1028
:param revision_id: the revision id to use.
1029
:param rev: The revision object.
1030
:param inv: The inventory for the revision. if None, it will be looked
1031
up in the inventory storer
1032
:param config: If None no digital signature will be created.
1033
If supplied its signature_needed method will be used
1034
to determine if a signature should be made.
1036
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1038
_mod_revision.check_not_reserved_id(revision_id)
1039
if config is not None and config.signature_needed():
1041
inv = self.get_inventory(revision_id)
1042
tree = InventoryRevisionTree(self, inv, revision_id)
1043
testament = Testament(rev, tree)
1044
plaintext = testament.as_short_text()
1045
self.store_revision_signature(
1046
gpg.GPGStrategy(config), plaintext, revision_id)
1047
# check inventory present
1048
if not self.inventories.get_parent_map([(revision_id,)]):
1050
raise errors.WeaveRevisionNotPresent(revision_id,
1053
# yes, this is not suitable for adding with ghosts.
1054
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1057
key = (revision_id,)
1058
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1059
self._add_revision(rev)
1061
def _add_revision(self, revision):
1062
text = self._serializer.write_revision_to_string(revision)
1063
key = (revision.revision_id,)
1064
parents = tuple((parent,) for parent in revision.parent_ids)
1065
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1067
def _check_inventories(self, checker):
1068
"""Check the inventories found from the revision scan.
1070
This is responsible for verifying the sha1 of inventories and
1071
creating a pending_keys set that covers data referenced by inventories.
1073
bar = ui.ui_factory.nested_progress_bar()
1075
self._do_check_inventories(checker, bar)
1079
def _do_check_inventories(self, checker, bar):
1080
"""Helper for _check_inventories."""
1082
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1083
kinds = ['chk_bytes', 'texts']
1084
count = len(checker.pending_keys)
1085
bar.update("inventories", 0, 2)
1086
current_keys = checker.pending_keys
1087
checker.pending_keys = {}
1088
# Accumulate current checks.
1089
for key in current_keys:
1090
if key[0] != 'inventories' and key[0] not in kinds:
1091
checker._report_items.append('unknown key type %r' % (key,))
1092
keys[key[0]].add(key[1:])
1093
if keys['inventories']:
1094
# NB: output order *should* be roughly sorted - topo or
1095
# inverse topo depending on repository - either way decent
1096
# to just delta against. However, pre-CHK formats didn't
1097
# try to optimise inventory layout on disk. As such the
1098
# pre-CHK code path does not use inventory deltas.
1100
for record in self.inventories.check(keys=keys['inventories']):
1101
if record.storage_kind == 'absent':
1102
checker._report_items.append(
1103
'Missing inventory {%s}' % (record.key,))
1105
last_object = self._check_record('inventories', record,
1106
checker, last_object,
1107
current_keys[('inventories',) + record.key])
1108
del keys['inventories']
1111
bar.update("texts", 1)
1112
while (checker.pending_keys or keys['chk_bytes']
1114
# Something to check.
1115
current_keys = checker.pending_keys
1116
checker.pending_keys = {}
1117
# Accumulate current checks.
1118
for key in current_keys:
1119
if key[0] not in kinds:
1120
checker._report_items.append('unknown key type %r' % (key,))
1121
keys[key[0]].add(key[1:])
1122
# Check the outermost kind only - inventories || chk_bytes || texts
1126
for record in getattr(self, kind).check(keys=keys[kind]):
1127
if record.storage_kind == 'absent':
1128
checker._report_items.append(
1129
'Missing %s {%s}' % (kind, record.key,))
1131
last_object = self._check_record(kind, record,
1132
checker, last_object, current_keys[(kind,) + record.key])
1136
def _check_record(self, kind, record, checker, last_object, item_data):
1137
"""Check a single text from this repository."""
1138
if kind == 'inventories':
1139
rev_id = record.key[0]
1140
inv = self._deserialise_inventory(rev_id,
1141
record.get_bytes_as('fulltext'))
1142
if last_object is not None:
1143
delta = inv._make_delta(last_object)
1144
for old_path, path, file_id, ie in delta:
1147
ie.check(checker, rev_id, inv)
1149
for path, ie in inv.iter_entries():
1150
ie.check(checker, rev_id, inv)
1151
if self._format.fast_deltas:
1153
elif kind == 'chk_bytes':
1154
# No code written to check chk_bytes for this repo format.
1155
checker._report_items.append(
1156
'unsupported key type chk_bytes for %s' % (record.key,))
1157
elif kind == 'texts':
1158
self._check_text(record, checker, item_data)
1160
checker._report_items.append(
1161
'unknown key type %s for %s' % (kind, record.key))
1163
def _check_text(self, record, checker, item_data):
1164
"""Check a single text."""
1165
# Check it is extractable.
1166
# TODO: check length.
1167
if record.storage_kind == 'chunked':
1168
chunks = record.get_bytes_as(record.storage_kind)
1169
sha1 = osutils.sha_strings(chunks)
1170
length = sum(map(len, chunks))
1172
content = record.get_bytes_as('fulltext')
1173
sha1 = osutils.sha_string(content)
1174
length = len(content)
1175
if item_data and sha1 != item_data[1]:
1176
checker._report_items.append(
1177
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1178
(record.key, sha1, item_data[1], item_data[2]))
1180
def __init__(self, _format, a_bzrdir, control_files):
1181
"""Instantiate a VersionedFileRepository.
1183
:param _format: The format of the repository on disk.
1184
:param a_bzrdir: The BzrDir of the repository.
1185
:param control_files: Control files to use for locking, etc.
1187
# In the future we will have a single api for all stores for
1188
# getting file texts, inventories and revisions, then
1189
# this construct will accept instances of those things.
1190
super(VersionedFileRepository, self).__init__(_format, a_bzrdir,
1193
self._reconcile_does_inventory_gc = True
1194
self._reconcile_fixes_text_parents = False
1195
self._reconcile_backsup_inventory = True
1196
# An InventoryEntry cache, used during deserialization
1197
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1198
# Is it safe to return inventory entries directly from the entry cache,
1199
# rather copying them?
1200
self._safe_to_return_from_cache = False
1203
def gather_stats(self, revid=None, committers=None):
1204
"""See Repository.gather_stats()."""
1205
result = super(VersionedFileRepository, self).gather_stats(revid, committers)
1206
# now gather global repository information
1207
# XXX: This is available for many repos regardless of listability.
1208
if self.user_transport.listable():
1209
# XXX: do we want to __define len__() ?
1210
# Maybe the versionedfiles object should provide a different
1211
# method to get the number of keys.
1212
result['revisions'] = len(self.revisions.keys())
1213
# result['size'] = t
1216
def get_commit_builder(self, branch, parents, config, timestamp=None,
1217
timezone=None, committer=None, revprops=None,
1218
revision_id=None, lossy=False):
1219
"""Obtain a CommitBuilder for this repository.
1221
:param branch: Branch to commit to.
1222
:param parents: Revision ids of the parents of the new revision.
1223
:param config: Configuration to use.
1224
:param timestamp: Optional timestamp recorded for commit.
1225
:param timezone: Optional timezone for timestamp.
1226
:param committer: Optional committer to set for commit.
1227
:param revprops: Optional dictionary of revision properties.
1228
:param revision_id: Optional revision id.
1229
:param lossy: Whether to discard data that can not be natively
1230
represented, when pushing to a foreign VCS
1232
if self._fallback_repositories and not self._format.supports_chks:
1233
raise errors.BzrError("Cannot commit directly to a stacked branch"
1234
" in pre-2a formats. See "
1235
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1236
result = self._commit_builder_class(self, parents, config,
1237
timestamp, timezone, committer, revprops, revision_id,
1239
self.start_write_group()
1242
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1243
"""Return the keys of missing inventory parents for revisions added in
1246
A revision is not complete if the inventory delta for that revision
1247
cannot be calculated. Therefore if the parent inventories of a
1248
revision are not present, the revision is incomplete, and e.g. cannot
1249
be streamed by a smart server. This method finds missing inventory
1250
parents for revisions added in this write group.
1252
if not self._format.supports_external_lookups:
1253
# This is only an issue for stacked repositories
1255
if not self.is_in_write_group():
1256
raise AssertionError('not in a write group')
1258
# XXX: We assume that every added revision already has its
1259
# corresponding inventory, so we only check for parent inventories that
1260
# might be missing, rather than all inventories.
1261
parents = set(self.revisions._index.get_missing_parents())
1262
parents.discard(_mod_revision.NULL_REVISION)
1263
unstacked_inventories = self.inventories._index
1264
present_inventories = unstacked_inventories.get_parent_map(
1265
key[-1:] for key in parents)
1266
parents.difference_update(present_inventories)
1267
if len(parents) == 0:
1268
# No missing parent inventories.
1270
if not check_for_missing_texts:
1271
return set(('inventories', rev_id) for (rev_id,) in parents)
1272
# Ok, now we have a list of missing inventories. But these only matter
1273
# if the inventories that reference them are missing some texts they
1274
# appear to introduce.
1275
# XXX: Texts referenced by all added inventories need to be present,
1276
# but at the moment we're only checking for texts referenced by
1277
# inventories at the graph's edge.
1278
key_deps = self.revisions._index._key_dependencies
1279
key_deps.satisfy_refs_for_keys(present_inventories)
1280
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1281
file_ids = self.fileids_altered_by_revision_ids(referrers)
1282
missing_texts = set()
1283
for file_id, version_ids in file_ids.iteritems():
1284
missing_texts.update(
1285
(file_id, version_id) for version_id in version_ids)
1286
present_texts = self.texts.get_parent_map(missing_texts)
1287
missing_texts.difference_update(present_texts)
1288
if not missing_texts:
1289
# No texts are missing, so all revisions and their deltas are
1292
# Alternatively the text versions could be returned as the missing
1293
# keys, but this is likely to be less data.
1294
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1298
def has_revisions(self, revision_ids):
1299
"""Probe to find out the presence of multiple revisions.
1301
:param revision_ids: An iterable of revision_ids.
1302
:return: A set of the revision_ids that were present.
1304
parent_map = self.revisions.get_parent_map(
1305
[(rev_id,) for rev_id in revision_ids])
1307
if _mod_revision.NULL_REVISION in revision_ids:
1308
result.add(_mod_revision.NULL_REVISION)
1309
result.update([key[0] for key in parent_map])
1313
def get_revision_reconcile(self, revision_id):
1314
"""'reconcile' helper routine that allows access to a revision always.
1316
This variant of get_revision does not cross check the weave graph
1317
against the revision one as get_revision does: but it should only
1318
be used by reconcile, or reconcile-alike commands that are correcting
1319
or testing the revision graph.
1321
return self._get_revisions([revision_id])[0]
1324
def get_revisions(self, revision_ids):
1325
"""Get many revisions at once.
1327
Repositories that need to check data on every revision read should
1328
subclass this method.
1330
return self._get_revisions(revision_ids)
1333
def _get_revisions(self, revision_ids):
1334
"""Core work logic to get many revisions without sanity checks."""
1336
for revid, rev in self._iter_revisions(revision_ids):
1338
raise errors.NoSuchRevision(self, revid)
1340
return [revs[revid] for revid in revision_ids]
1342
def _iter_revisions(self, revision_ids):
1343
"""Iterate over revision objects.
1345
:param revision_ids: An iterable of revisions to examine. None may be
1346
passed to request all revisions known to the repository. Note that
1347
not all repositories can find unreferenced revisions; for those
1348
repositories only referenced ones will be returned.
1349
:return: An iterator of (revid, revision) tuples. Absent revisions (
1350
those asked for but not available) are returned as (revid, None).
1352
if revision_ids is None:
1353
revision_ids = self.all_revision_ids()
1355
for rev_id in revision_ids:
1356
if not rev_id or not isinstance(rev_id, basestring):
1357
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1358
keys = [(key,) for key in revision_ids]
1359
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1360
for record in stream:
1361
revid = record.key[0]
1362
if record.storage_kind == 'absent':
1365
text = record.get_bytes_as('fulltext')
1366
rev = self._serializer.read_revision_from_string(text)
1370
def add_signature_text(self, revision_id, signature):
1371
"""Store a signature text for a revision.
1373
:param revision_id: Revision id of the revision
1374
:param signature: Signature text.
1376
self.signatures.add_lines((revision_id,), (),
1377
osutils.split_lines(signature))
1379
def find_text_key_references(self):
1380
"""Find the text key references within the repository.
1382
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1383
to whether they were referred to by the inventory of the
1384
revision_id that they contain. The inventory texts from all present
1385
revision ids are assessed to generate this report.
1387
revision_keys = self.revisions.keys()
1388
w = self.inventories
1389
pb = ui.ui_factory.nested_progress_bar()
1391
return self._serializer._find_text_key_references(
1392
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1396
def _inventory_xml_lines_for_keys(self, keys):
1397
"""Get a line iterator of the sort needed for findind references.
1399
Not relevant for non-xml inventory repositories.
1401
Ghosts in revision_keys are ignored.
1403
:param revision_keys: The revision keys for the inventories to inspect.
1404
:return: An iterator over (inventory line, revid) for the fulltexts of
1405
all of the xml inventories specified by revision_keys.
1407
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1408
for record in stream:
1409
if record.storage_kind != 'absent':
1410
chunks = record.get_bytes_as('chunked')
1411
revid = record.key[-1]
1412
lines = osutils.chunks_to_lines(chunks)
1416
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1418
"""Helper routine for fileids_altered_by_revision_ids.
1420
This performs the translation of xml lines to revision ids.
1422
:param line_iterator: An iterator of lines, origin_version_id
1423
:param revision_keys: The revision ids to filter for. This should be a
1424
set or other type which supports efficient __contains__ lookups, as
1425
the revision key from each parsed line will be looked up in the
1426
revision_keys filter.
1427
:return: a dictionary mapping altered file-ids to an iterable of
1428
revision_ids. Each altered file-ids has the exact revision_ids that
1429
altered it listed explicitly.
1431
seen = set(self._serializer._find_text_key_references(
1432
line_iterator).iterkeys())
1433
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
1434
parent_seen = set(self._serializer._find_text_key_references(
1435
self._inventory_xml_lines_for_keys(parent_keys)))
1436
new_keys = seen - parent_seen
1438
setdefault = result.setdefault
1439
for key in new_keys:
1440
setdefault(key[0], set()).add(key[-1])
1443
def _find_parent_keys_of_revisions(self, revision_keys):
1444
"""Similar to _find_parent_ids_of_revisions, but used with keys.
1446
:param revision_keys: An iterable of revision_keys.
1447
:return: The parents of all revision_keys that are not already in
1450
parent_map = self.revisions.get_parent_map(revision_keys)
1452
map(parent_keys.update, parent_map.itervalues())
1453
parent_keys.difference_update(revision_keys)
1454
parent_keys.discard(_mod_revision.NULL_REVISION)
1457
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
1458
"""Find the file ids and versions affected by revisions.
1460
:param revisions: an iterable containing revision ids.
1461
:param _inv_weave: The inventory weave from this repository or None.
1462
If None, the inventory weave will be opened automatically.
1463
:return: a dictionary mapping altered file-ids to an iterable of
1464
revision_ids. Each altered file-ids has the exact revision_ids that
1465
altered it listed explicitly.
1467
selected_keys = set((revid,) for revid in revision_ids)
1468
w = _inv_weave or self.inventories
1469
return self._find_file_ids_from_xml_inventory_lines(
1470
w.iter_lines_added_or_present_in_keys(
1471
selected_keys, pb=None),
1474
def iter_files_bytes(self, desired_files):
1475
"""Iterate through file versions.
1477
Files will not necessarily be returned in the order they occur in
1478
desired_files. No specific order is guaranteed.
1480
Yields pairs of identifier, bytes_iterator. identifier is an opaque
1481
value supplied by the caller as part of desired_files. It should
1482
uniquely identify the file version in the caller's context. (Examples:
1483
an index number or a TreeTransform trans_id.)
1485
bytes_iterator is an iterable of bytestrings for the file. The
1486
kind of iterable and length of the bytestrings are unspecified, but for
1487
this implementation, it is a list of bytes produced by
1488
VersionedFile.get_record_stream().
1490
:param desired_files: a list of (file_id, revision_id, identifier)
1494
for file_id, revision_id, callable_data in desired_files:
1495
text_keys[(file_id, revision_id)] = callable_data
1496
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
1497
if record.storage_kind == 'absent':
1498
raise errors.RevisionNotPresent(record.key, self)
1499
yield text_keys[record.key], record.get_bytes_as('chunked')
1501
def _generate_text_key_index(self, text_key_references=None,
1503
"""Generate a new text key index for the repository.
1505
This is an expensive function that will take considerable time to run.
1507
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
1508
list of parents, also text keys. When a given key has no parents,
1509
the parents list will be [NULL_REVISION].
1511
# All revisions, to find inventory parents.
1512
if ancestors is None:
1513
graph = self.get_graph()
1514
ancestors = graph.get_parent_map(self.all_revision_ids())
1515
if text_key_references is None:
1516
text_key_references = self.find_text_key_references()
1517
pb = ui.ui_factory.nested_progress_bar()
1519
return self._do_generate_text_key_index(ancestors,
1520
text_key_references, pb)
1524
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
1525
"""Helper for _generate_text_key_index to avoid deep nesting."""
1526
revision_order = tsort.topo_sort(ancestors)
1527
invalid_keys = set()
1529
for revision_id in revision_order:
1530
revision_keys[revision_id] = set()
1531
text_count = len(text_key_references)
1532
# a cache of the text keys to allow reuse; costs a dict of all the
1533
# keys, but saves a 2-tuple for every child of a given key.
1535
for text_key, valid in text_key_references.iteritems():
1537
invalid_keys.add(text_key)
1539
revision_keys[text_key[1]].add(text_key)
1540
text_key_cache[text_key] = text_key
1541
del text_key_references
1543
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
1544
NULL_REVISION = _mod_revision.NULL_REVISION
1545
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
1546
# too small for large or very branchy trees. However, for 55K path
1547
# trees, it would be easy to use too much memory trivially. Ideally we
1548
# could gauge this by looking at available real memory etc, but this is
1549
# always a tricky proposition.
1550
inventory_cache = lru_cache.LRUCache(10)
1551
batch_size = 10 # should be ~150MB on a 55K path tree
1552
batch_count = len(revision_order) / batch_size + 1
1554
pb.update("Calculating text parents", processed_texts, text_count)
1555
for offset in xrange(batch_count):
1556
to_query = revision_order[offset * batch_size:(offset + 1) *
1560
for revision_id in to_query:
1561
parent_ids = ancestors[revision_id]
1562
for text_key in revision_keys[revision_id]:
1563
pb.update("Calculating text parents", processed_texts)
1564
processed_texts += 1
1565
candidate_parents = []
1566
for parent_id in parent_ids:
1567
parent_text_key = (text_key[0], parent_id)
1569
check_parent = parent_text_key not in \
1570
revision_keys[parent_id]
1572
# the parent parent_id is a ghost:
1573
check_parent = False
1574
# truncate the derived graph against this ghost.
1575
parent_text_key = None
1577
# look at the parent commit details inventories to
1578
# determine possible candidates in the per file graph.
1581
inv = inventory_cache[parent_id]
1583
inv = self.revision_tree(parent_id).inventory
1584
inventory_cache[parent_id] = inv
1586
parent_entry = inv[text_key[0]]
1587
except (KeyError, errors.NoSuchId):
1589
if parent_entry is not None:
1591
text_key[0], parent_entry.revision)
1593
parent_text_key = None
1594
if parent_text_key is not None:
1595
candidate_parents.append(
1596
text_key_cache[parent_text_key])
1597
parent_heads = text_graph.heads(candidate_parents)
1598
new_parents = list(parent_heads)
1599
new_parents.sort(key=lambda x:candidate_parents.index(x))
1600
if new_parents == []:
1601
new_parents = [NULL_REVISION]
1602
text_index[text_key] = new_parents
1604
for text_key in invalid_keys:
1605
text_index[text_key] = [NULL_REVISION]
1608
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
1609
"""Get an iterable listing the keys of all the data introduced by a set
1612
The keys will be ordered so that the corresponding items can be safely
1613
fetched and inserted in that order.
1615
:returns: An iterable producing tuples of (knit-kind, file-id,
1616
versions). knit-kind is one of 'file', 'inventory', 'signatures',
1617
'revisions'. file-id is None unless knit-kind is 'file'.
1619
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
1622
for result in self._find_non_file_keys_to_fetch(revision_ids):
1625
def _find_file_keys_to_fetch(self, revision_ids, pb):
1626
# XXX: it's a bit weird to control the inventory weave caching in this
1627
# generator. Ideally the caching would be done in fetch.py I think. Or
1628
# maybe this generator should explicitly have the contract that it
1629
# should not be iterated until the previously yielded item has been
1631
inv_w = self.inventories
1633
# file ids that changed
1634
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
1636
num_file_ids = len(file_ids)
1637
for file_id, altered_versions in file_ids.iteritems():
1639
pb.update("Fetch texts", count, num_file_ids)
1641
yield ("file", file_id, altered_versions)
1643
def _find_non_file_keys_to_fetch(self, revision_ids):
1645
yield ("inventory", None, revision_ids)
1648
# XXX: Note ATM no callers actually pay attention to this return
1649
# instead they just use the list of revision ids and ignore
1650
# missing sigs. Consider removing this work entirely
1651
revisions_with_signatures = set(self.signatures.get_parent_map(
1652
[(r,) for r in revision_ids]))
1653
revisions_with_signatures = set(
1654
[r for (r,) in revisions_with_signatures])
1655
revisions_with_signatures.intersection_update(revision_ids)
1656
yield ("signatures", None, revisions_with_signatures)
1659
yield ("revisions", None, revision_ids)
1662
def get_inventory(self, revision_id):
1663
"""Get Inventory object by revision id."""
1664
return self.iter_inventories([revision_id]).next()
1666
def iter_inventories(self, revision_ids, ordering=None):
1667
"""Get many inventories by revision_ids.
1669
This will buffer some or all of the texts used in constructing the
1670
inventories in memory, but will only parse a single inventory at a
1673
:param revision_ids: The expected revision ids of the inventories.
1674
:param ordering: optional ordering, e.g. 'topological'. If not
1675
specified, the order of revision_ids will be preserved (by
1676
buffering if necessary).
1677
:return: An iterator of inventories.
1679
if ((None in revision_ids)
1680
or (_mod_revision.NULL_REVISION in revision_ids)):
1681
raise ValueError('cannot get null revision inventory')
1682
return self._iter_inventories(revision_ids, ordering)
1684
def _iter_inventories(self, revision_ids, ordering):
1685
"""single-document based inventory iteration."""
1686
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
1687
for text, revision_id in inv_xmls:
1688
yield self._deserialise_inventory(revision_id, text)
1690
def _iter_inventory_xmls(self, revision_ids, ordering):
1691
if ordering is None:
1692
order_as_requested = True
1693
ordering = 'unordered'
1695
order_as_requested = False
1696
keys = [(revision_id,) for revision_id in revision_ids]
1699
if order_as_requested:
1700
key_iter = iter(keys)
1701
next_key = key_iter.next()
1702
stream = self.inventories.get_record_stream(keys, ordering, True)
1704
for record in stream:
1705
if record.storage_kind != 'absent':
1706
chunks = record.get_bytes_as('chunked')
1707
if order_as_requested:
1708
text_chunks[record.key] = chunks
1710
yield ''.join(chunks), record.key[-1]
1712
raise errors.NoSuchRevision(self, record.key)
1713
if order_as_requested:
1714
# Yield as many results as we can while preserving order.
1715
while next_key in text_chunks:
1716
chunks = text_chunks.pop(next_key)
1717
yield ''.join(chunks), next_key[-1]
1719
next_key = key_iter.next()
1720
except StopIteration:
1721
# We still want to fully consume the get_record_stream,
1722
# just in case it is not actually finished at this point
1726
def _deserialise_inventory(self, revision_id, xml):
1727
"""Transform the xml into an inventory object.
1729
:param revision_id: The expected revision id of the inventory.
1730
:param xml: A serialised inventory.
1732
result = self._serializer.read_inventory_from_string(xml, revision_id,
1733
entry_cache=self._inventory_entry_cache,
1734
return_from_cache=self._safe_to_return_from_cache)
1735
if result.revision_id != revision_id:
1736
raise AssertionError('revision id mismatch %s != %s' % (
1737
result.revision_id, revision_id))
1740
def get_serializer_format(self):
1741
return self._serializer.format_num
1744
def _get_inventory_xml(self, revision_id):
1745
"""Get serialized inventory as a string."""
1746
texts = self._iter_inventory_xmls([revision_id], 'unordered')
1748
text, revision_id = texts.next()
1749
except StopIteration:
1750
raise errors.HistoryMissing(self, 'inventory', revision_id)
1754
def revision_tree(self, revision_id):
1755
"""Return Tree for a revision on this branch.
1757
`revision_id` may be NULL_REVISION for the empty tree revision.
1759
revision_id = _mod_revision.ensure_null(revision_id)
1760
# TODO: refactor this to use an existing revision object
1761
# so we don't need to read it in twice.
1762
if revision_id == _mod_revision.NULL_REVISION:
1763
return InventoryRevisionTree(self,
1764
Inventory(root_id=None), _mod_revision.NULL_REVISION)
1766
inv = self.get_inventory(revision_id)
1767
return InventoryRevisionTree(self, inv, revision_id)
1769
def revision_trees(self, revision_ids):
1770
"""Return Trees for revisions in this repository.
1772
:param revision_ids: a sequence of revision-ids;
1773
a revision-id may not be None or 'null:'
1775
inventories = self.iter_inventories(revision_ids)
1776
for inv in inventories:
1777
yield InventoryRevisionTree(self, inv, inv.revision_id)
1779
def _filtered_revision_trees(self, revision_ids, file_ids):
1780
"""Return Tree for a revision on this branch with only some files.
1782
:param revision_ids: a sequence of revision-ids;
1783
a revision-id may not be None or 'null:'
1784
:param file_ids: if not None, the result is filtered
1785
so that only those file-ids, their parents and their
1786
children are included.
1788
inventories = self.iter_inventories(revision_ids)
1789
for inv in inventories:
1790
# Should we introduce a FilteredRevisionTree class rather
1791
# than pre-filter the inventory here?
1792
filtered_inv = inv.filter(file_ids)
1793
yield InventoryRevisionTree(self, filtered_inv, filtered_inv.revision_id)
1795
def get_parent_map(self, revision_ids):
1796
"""See graph.StackedParentsProvider.get_parent_map"""
1797
# revisions index works in keys; this just works in revisions
1798
# therefore wrap and unwrap
1801
for revision_id in revision_ids:
1802
if revision_id == _mod_revision.NULL_REVISION:
1803
result[revision_id] = ()
1804
elif revision_id is None:
1805
raise ValueError('get_parent_map(None) is not valid')
1807
query_keys.append((revision_id ,))
1808
for ((revision_id,), parent_keys) in \
1809
self.revisions.get_parent_map(query_keys).iteritems():
1811
result[revision_id] = tuple([parent_revid
1812
for (parent_revid,) in parent_keys])
1814
result[revision_id] = (_mod_revision.NULL_REVISION,)
1818
def get_known_graph_ancestry(self, revision_ids):
1819
"""Return the known graph for a set of revision ids and their ancestors.
1821
st = static_tuple.StaticTuple
1822
revision_keys = [st(r_id).intern() for r_id in revision_ids]
1823
known_graph = self.revisions.get_known_graph_ancestry(revision_keys)
1824
return graph.GraphThunkIdsToKeys(known_graph)
1827
def get_file_graph(self):
1828
"""Return the graph walker for text revisions."""
1829
return graph.Graph(self.texts)
1831
def _get_versioned_file_checker(self, text_key_references=None,
1833
"""Return an object suitable for checking versioned files.
1835
:param text_key_references: if non-None, an already built
1836
dictionary mapping text keys ((fileid, revision_id) tuples)
1837
to whether they were referred to by the inventory of the
1838
revision_id that they contain. If None, this will be
1840
:param ancestors: Optional result from
1841
self.get_graph().get_parent_map(self.all_revision_ids()) if already
1844
return _VersionedFileChecker(self,
1845
text_key_references=text_key_references, ancestors=ancestors)
1848
def has_signature_for_revision_id(self, revision_id):
1849
"""Query for a revision signature for revision_id in the repository."""
1850
if not self.has_revision(revision_id):
1851
raise errors.NoSuchRevision(self, revision_id)
1852
sig_present = (1 == len(
1853
self.signatures.get_parent_map([(revision_id,)])))
1857
def get_signature_text(self, revision_id):
1858
"""Return the text for a signature."""
1859
stream = self.signatures.get_record_stream([(revision_id,)],
1861
record = stream.next()
1862
if record.storage_kind == 'absent':
1863
raise errors.NoSuchRevision(self, revision_id)
1864
return record.get_bytes_as('fulltext')
1867
def _check(self, revision_ids, callback_refs, check_repo):
1868
result = check.VersionedFileCheck(self, check_repo=check_repo)
1869
result.check(callback_refs)
1872
def _find_inconsistent_revision_parents(self, revisions_iterator=None):
1873
"""Find revisions with different parent lists in the revision object
1874
and in the index graph.
1876
:param revisions_iterator: None, or an iterator of (revid,
1877
Revision-or-None). This iterator controls the revisions checked.
1878
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1879
parents-in-revision).
1881
if not self.is_locked():
1882
raise AssertionError()
1884
if revisions_iterator is None:
1885
revisions_iterator = self._iter_revisions(None)
1886
for revid, revision in revisions_iterator:
1887
if revision is None:
1889
parent_map = vf.get_parent_map([(revid,)])
1890
parents_according_to_index = tuple(parent[-1] for parent in
1891
parent_map[(revid,)])
1892
parents_according_to_revision = tuple(revision.parent_ids)
1893
if parents_according_to_index != parents_according_to_revision:
1894
yield (revid, parents_according_to_index,
1895
parents_according_to_revision)
1897
def _check_for_inconsistent_revision_parents(self):
1898
inconsistencies = list(self._find_inconsistent_revision_parents())
1900
raise errors.BzrCheckError(
1901
"Revision knit has inconsistent parents.")
1903
def _get_sink(self):
1904
"""Return a sink for streaming into this repository."""
1905
return StreamSink(self)
1907
def _get_source(self, to_format):
1908
"""Return a source for streaming from this repository."""
1909
return StreamSource(self, to_format)
1912
class MetaDirVersionedFileRepository(MetaDirRepository,
1913
VersionedFileRepository):
1914
"""Repositories in a meta-dir, that work via versioned file objects."""
1916
def __init__(self, _format, a_bzrdir, control_files):
1917
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
1921
class MetaDirVersionedFileRepositoryFormat(MetaDirRepositoryFormat,
1922
VersionedFileRepositoryFormat):
1923
"""Base class for repository formats using versioned files in metadirs."""
1926
class StreamSink(object):
1927
"""An object that can insert a stream into a repository.
1929
This interface handles the complexity of reserialising inventories and
1930
revisions from different formats, and allows unidirectional insertion into
1931
stacked repositories without looking for the missing basis parents
1935
def __init__(self, target_repo):
1936
self.target_repo = target_repo
1938
def insert_stream(self, stream, src_format, resume_tokens):
1939
"""Insert a stream's content into the target repository.
1941
:param src_format: a bzr repository format.
1943
:return: a list of resume tokens and an iterable of keys additional
1944
items required before the insertion can be completed.
1946
self.target_repo.lock_write()
1949
self.target_repo.resume_write_group(resume_tokens)
1952
self.target_repo.start_write_group()
1955
# locked_insert_stream performs a commit|suspend.
1956
missing_keys = self.insert_stream_without_locking(stream,
1957
src_format, is_resume)
1959
# suspend the write group and tell the caller what we is
1960
# missing. We know we can suspend or else we would not have
1961
# entered this code path. (All repositories that can handle
1962
# missing keys can handle suspending a write group).
1963
write_group_tokens = self.target_repo.suspend_write_group()
1964
return write_group_tokens, missing_keys
1965
hint = self.target_repo.commit_write_group()
1966
to_serializer = self.target_repo._format._serializer
1967
src_serializer = src_format._serializer
1968
if (to_serializer != src_serializer and
1969
self.target_repo._format.pack_compresses):
1970
self.target_repo.pack(hint=hint)
1973
self.target_repo.abort_write_group(suppress_errors=True)
1976
self.target_repo.unlock()
1978
def insert_stream_without_locking(self, stream, src_format,
1980
"""Insert a stream's content into the target repository.
1982
This assumes that you already have a locked repository and an active
1985
:param src_format: a bzr repository format.
1986
:param is_resume: Passed down to get_missing_parent_inventories to
1987
indicate if we should be checking for missing texts at the same
1990
:return: A set of keys that are missing.
1992
if not self.target_repo.is_write_locked():
1993
raise errors.ObjectNotLocked(self)
1994
if not self.target_repo.is_in_write_group():
1995
raise errors.BzrError('you must already be in a write group')
1996
to_serializer = self.target_repo._format._serializer
1997
src_serializer = src_format._serializer
1999
if to_serializer == src_serializer:
2000
# If serializers match and the target is a pack repository, set the
2001
# write cache size on the new pack. This avoids poor performance
2002
# on transports where append is unbuffered (such as
2003
# RemoteTransport). This is safe to do because nothing should read
2004
# back from the target repository while a stream with matching
2005
# serialization is being inserted.
2006
# The exception is that a delta record from the source that should
2007
# be a fulltext may need to be expanded by the target (see
2008
# test_fetch_revisions_with_deltas_into_pack); but we take care to
2009
# explicitly flush any buffered writes first in that rare case.
2011
new_pack = self.target_repo._pack_collection._new_pack
2012
except AttributeError:
2013
# Not a pack repository
2016
new_pack.set_write_cache_size(1024*1024)
2017
for substream_type, substream in stream:
2018
if 'stream' in debug.debug_flags:
2019
mutter('inserting substream: %s', substream_type)
2020
if substream_type == 'texts':
2021
self.target_repo.texts.insert_record_stream(substream)
2022
elif substream_type == 'inventories':
2023
if src_serializer == to_serializer:
2024
self.target_repo.inventories.insert_record_stream(
2027
self._extract_and_insert_inventories(
2028
substream, src_serializer)
2029
elif substream_type == 'inventory-deltas':
2030
self._extract_and_insert_inventory_deltas(
2031
substream, src_serializer)
2032
elif substream_type == 'chk_bytes':
2033
# XXX: This doesn't support conversions, as it assumes the
2034
# conversion was done in the fetch code.
2035
self.target_repo.chk_bytes.insert_record_stream(substream)
2036
elif substream_type == 'revisions':
2037
# This may fallback to extract-and-insert more often than
2038
# required if the serializers are different only in terms of
2040
if src_serializer == to_serializer:
2041
self.target_repo.revisions.insert_record_stream(substream)
2043
self._extract_and_insert_revisions(substream,
2045
elif substream_type == 'signatures':
2046
self.target_repo.signatures.insert_record_stream(substream)
2048
raise AssertionError('kaboom! %s' % (substream_type,))
2049
# Done inserting data, and the missing_keys calculations will try to
2050
# read back from the inserted data, so flush the writes to the new pack
2051
# (if this is pack format).
2052
if new_pack is not None:
2053
new_pack._write_data('', flush=True)
2054
# Find all the new revisions (including ones from resume_tokens)
2055
missing_keys = self.target_repo.get_missing_parent_inventories(
2056
check_for_missing_texts=is_resume)
2058
for prefix, versioned_file in (
2059
('texts', self.target_repo.texts),
2060
('inventories', self.target_repo.inventories),
2061
('revisions', self.target_repo.revisions),
2062
('signatures', self.target_repo.signatures),
2063
('chk_bytes', self.target_repo.chk_bytes),
2065
if versioned_file is None:
2067
# TODO: key is often going to be a StaticTuple object
2068
# I don't believe we can define a method by which
2069
# (prefix,) + StaticTuple will work, though we could
2070
# define a StaticTuple.sq_concat that would allow you to
2071
# pass in either a tuple or a StaticTuple as the second
2072
# object, so instead we could have:
2073
# StaticTuple(prefix) + key here...
2074
missing_keys.update((prefix,) + key for key in
2075
versioned_file.get_missing_compression_parent_keys())
2076
except NotImplementedError:
2077
# cannot even attempt suspending, and missing would have failed
2078
# during stream insertion.
2079
missing_keys = set()
2082
def _extract_and_insert_inventory_deltas(self, substream, serializer):
2083
target_rich_root = self.target_repo._format.rich_root_data
2084
target_tree_refs = self.target_repo._format.supports_tree_reference
2085
for record in substream:
2086
# Insert the delta directly
2087
inventory_delta_bytes = record.get_bytes_as('fulltext')
2088
deserialiser = inventory_delta.InventoryDeltaDeserializer()
2090
parse_result = deserialiser.parse_text_bytes(
2091
inventory_delta_bytes)
2092
except inventory_delta.IncompatibleInventoryDelta, err:
2093
mutter("Incompatible delta: %s", err.msg)
2094
raise errors.IncompatibleRevision(self.target_repo._format)
2095
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
2096
revision_id = new_id
2097
parents = [key[0] for key in record.parents]
2098
self.target_repo.add_inventory_by_delta(
2099
basis_id, inv_delta, revision_id, parents)
2101
def _extract_and_insert_inventories(self, substream, serializer,
2103
"""Generate a new inventory versionedfile in target, converting data.
2105
The inventory is retrieved from the source, (deserializing it), and
2106
stored in the target (reserializing it in a different format).
2108
target_rich_root = self.target_repo._format.rich_root_data
2109
target_tree_refs = self.target_repo._format.supports_tree_reference
2110
for record in substream:
2111
# It's not a delta, so it must be a fulltext in the source
2112
# serializer's format.
2113
bytes = record.get_bytes_as('fulltext')
2114
revision_id = record.key[0]
2115
inv = serializer.read_inventory_from_string(bytes, revision_id)
2116
parents = [key[0] for key in record.parents]
2117
self.target_repo.add_inventory(revision_id, inv, parents)
2118
# No need to keep holding this full inv in memory when the rest of
2119
# the substream is likely to be all deltas.
2122
def _extract_and_insert_revisions(self, substream, serializer):
2123
for record in substream:
2124
bytes = record.get_bytes_as('fulltext')
2125
revision_id = record.key[0]
2126
rev = serializer.read_revision_from_string(bytes)
2127
if rev.revision_id != revision_id:
2128
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
2129
self.target_repo.add_revision(revision_id, rev)
2132
if self.target_repo._format._fetch_reconcile:
2133
self.target_repo.reconcile()
2136
class StreamSource(object):
2137
"""A source of a stream for fetching between repositories."""
2139
def __init__(self, from_repository, to_format):
2140
"""Create a StreamSource streaming from from_repository."""
2141
self.from_repository = from_repository
2142
self.to_format = to_format
2143
self._record_counter = RecordCounter()
2145
def delta_on_metadata(self):
2146
"""Return True if delta's are permitted on metadata streams.
2148
That is on revisions and signatures.
2150
src_serializer = self.from_repository._format._serializer
2151
target_serializer = self.to_format._serializer
2152
return (self.to_format._fetch_uses_deltas and
2153
src_serializer == target_serializer)
2155
def _fetch_revision_texts(self, revs):
2156
# fetch signatures first and then the revision texts
2157
# may need to be a InterRevisionStore call here.
2158
from_sf = self.from_repository.signatures
2159
# A missing signature is just skipped.
2160
keys = [(rev_id,) for rev_id in revs]
2161
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
2163
self.to_format._fetch_order,
2164
not self.to_format._fetch_uses_deltas))
2165
# If a revision has a delta, this is actually expanded inside the
2166
# insert_record_stream code now, which is an alternate fix for
2168
from_rf = self.from_repository.revisions
2169
revisions = from_rf.get_record_stream(
2171
self.to_format._fetch_order,
2172
not self.delta_on_metadata())
2173
return [('signatures', signatures), ('revisions', revisions)]
2175
def _generate_root_texts(self, revs):
2176
"""This will be called by get_stream between fetching weave texts and
2177
fetching the inventory weave.
2179
if self._rich_root_upgrade():
2180
return _mod_fetch.Inter1and2Helper(
2181
self.from_repository).generate_root_texts(revs)
2185
def get_stream(self, search):
2187
revs = search.get_keys()
2188
graph = self.from_repository.get_graph()
2189
revs = tsort.topo_sort(graph.get_parent_map(revs))
2190
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
2192
for knit_kind, file_id, revisions in data_to_fetch:
2193
if knit_kind != phase:
2195
# Make a new progress bar for this phase
2196
if knit_kind == "file":
2197
# Accumulate file texts
2198
text_keys.extend([(file_id, revision) for revision in
2200
elif knit_kind == "inventory":
2201
# Now copy the file texts.
2202
from_texts = self.from_repository.texts
2203
yield ('texts', from_texts.get_record_stream(
2204
text_keys, self.to_format._fetch_order,
2205
not self.to_format._fetch_uses_deltas))
2206
# Cause an error if a text occurs after we have done the
2209
# Before we process the inventory we generate the root
2210
# texts (if necessary) so that the inventories references
2212
for _ in self._generate_root_texts(revs):
2214
# we fetch only the referenced inventories because we do not
2215
# know for unselected inventories whether all their required
2216
# texts are present in the other repository - it could be
2218
for info in self._get_inventory_stream(revs):
2220
elif knit_kind == "signatures":
2221
# Nothing to do here; this will be taken care of when
2222
# _fetch_revision_texts happens.
2224
elif knit_kind == "revisions":
2225
for record in self._fetch_revision_texts(revs):
2228
raise AssertionError("Unknown knit kind %r" % knit_kind)
2230
def get_stream_for_missing_keys(self, missing_keys):
2231
# missing keys can only occur when we are byte copying and not
2232
# translating (because translation means we don't send
2233
# unreconstructable deltas ever).
2235
keys['texts'] = set()
2236
keys['revisions'] = set()
2237
keys['inventories'] = set()
2238
keys['chk_bytes'] = set()
2239
keys['signatures'] = set()
2240
for key in missing_keys:
2241
keys[key[0]].add(key[1:])
2242
if len(keys['revisions']):
2243
# If we allowed copying revisions at this point, we could end up
2244
# copying a revision without copying its required texts: a
2245
# violation of the requirements for repository integrity.
2246
raise AssertionError(
2247
'cannot copy revisions to fill in missing deltas %s' % (
2248
keys['revisions'],))
2249
for substream_kind, keys in keys.iteritems():
2250
vf = getattr(self.from_repository, substream_kind)
2251
if vf is None and keys:
2252
raise AssertionError(
2253
"cannot fill in keys for a versioned file we don't"
2254
" have: %s needs %s" % (substream_kind, keys))
2256
# No need to stream something we don't have
2258
if substream_kind == 'inventories':
2259
# Some missing keys are genuinely ghosts, filter those out.
2260
present = self.from_repository.inventories.get_parent_map(keys)
2261
revs = [key[0] for key in present]
2262
# Get the inventory stream more-or-less as we do for the
2263
# original stream; there's no reason to assume that records
2264
# direct from the source will be suitable for the sink. (Think
2265
# e.g. 2a -> 1.9-rich-root).
2266
for info in self._get_inventory_stream(revs, missing=True):
2270
# Ask for full texts always so that we don't need more round trips
2271
# after this stream.
2272
# Some of the missing keys are genuinely ghosts, so filter absent
2273
# records. The Sink is responsible for doing another check to
2274
# ensure that ghosts don't introduce missing data for future
2276
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
2277
self.to_format._fetch_order, True))
2278
yield substream_kind, stream
2280
def inventory_fetch_order(self):
2281
if self._rich_root_upgrade():
2282
return 'topological'
2284
return self.to_format._fetch_order
2286
def _rich_root_upgrade(self):
2287
return (not self.from_repository._format.rich_root_data and
2288
self.to_format.rich_root_data)
2290
def _get_inventory_stream(self, revision_ids, missing=False):
2291
from_format = self.from_repository._format
2292
if (from_format.supports_chks and self.to_format.supports_chks and
2293
from_format.network_name() == self.to_format.network_name()):
2294
raise AssertionError(
2295
"this case should be handled by GroupCHKStreamSource")
2296
elif 'forceinvdeltas' in debug.debug_flags:
2297
return self._get_convertable_inventory_stream(revision_ids,
2298
delta_versus_null=missing)
2299
elif from_format.network_name() == self.to_format.network_name():
2301
return self._get_simple_inventory_stream(revision_ids,
2303
elif (not from_format.supports_chks and not self.to_format.supports_chks
2304
and from_format._serializer == self.to_format._serializer):
2305
# Essentially the same format.
2306
return self._get_simple_inventory_stream(revision_ids,
2309
# Any time we switch serializations, we want to use an
2310
# inventory-delta based approach.
2311
return self._get_convertable_inventory_stream(revision_ids,
2312
delta_versus_null=missing)
2314
def _get_simple_inventory_stream(self, revision_ids, missing=False):
2315
# NB: This currently reopens the inventory weave in source;
2316
# using a single stream interface instead would avoid this.
2317
from_weave = self.from_repository.inventories
2319
delta_closure = True
2321
delta_closure = not self.delta_on_metadata()
2322
yield ('inventories', from_weave.get_record_stream(
2323
[(rev_id,) for rev_id in revision_ids],
2324
self.inventory_fetch_order(), delta_closure))
2326
def _get_convertable_inventory_stream(self, revision_ids,
2327
delta_versus_null=False):
2328
# The two formats are sufficiently different that there is no fast
2329
# path, so we need to send just inventorydeltas, which any
2330
# sufficiently modern client can insert into any repository.
2331
# The StreamSink code expects to be able to
2332
# convert on the target, so we need to put bytes-on-the-wire that can
2333
# be converted. That means inventory deltas (if the remote is <1.19,
2334
# RemoteStreamSink will fallback to VFS to insert the deltas).
2335
yield ('inventory-deltas',
2336
self._stream_invs_as_deltas(revision_ids,
2337
delta_versus_null=delta_versus_null))
2339
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
2340
"""Return a stream of inventory-deltas for the given rev ids.
2342
:param revision_ids: The list of inventories to transmit
2343
:param delta_versus_null: Don't try to find a minimal delta for this
2344
entry, instead compute the delta versus the NULL_REVISION. This
2345
effectively streams a complete inventory. Used for stuff like
2346
filling in missing parents, etc.
2348
from_repo = self.from_repository
2349
revision_keys = [(rev_id,) for rev_id in revision_ids]
2350
parent_map = from_repo.inventories.get_parent_map(revision_keys)
2351
# XXX: possibly repos could implement a more efficient iter_inv_deltas
2353
inventories = self.from_repository.iter_inventories(
2354
revision_ids, 'topological')
2355
format = from_repo._format
2356
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
2357
inventory_cache = lru_cache.LRUCache(50)
2358
null_inventory = from_repo.revision_tree(
2359
_mod_revision.NULL_REVISION).inventory
2360
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
2361
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
2362
# repo back into a non-rich-root repo ought to be allowed)
2363
serializer = inventory_delta.InventoryDeltaSerializer(
2364
versioned_root=format.rich_root_data,
2365
tree_references=format.supports_tree_reference)
2366
for inv in inventories:
2367
key = (inv.revision_id,)
2368
parent_keys = parent_map.get(key, ())
2370
if not delta_versus_null and parent_keys:
2371
# The caller did not ask for complete inventories and we have
2372
# some parents that we can delta against. Make a delta against
2373
# each parent so that we can find the smallest.
2374
parent_ids = [parent_key[0] for parent_key in parent_keys]
2375
for parent_id in parent_ids:
2376
if parent_id not in invs_sent_so_far:
2377
# We don't know that the remote side has this basis, so
2380
if parent_id == _mod_revision.NULL_REVISION:
2381
parent_inv = null_inventory
2383
parent_inv = inventory_cache.get(parent_id, None)
2384
if parent_inv is None:
2385
parent_inv = from_repo.get_inventory(parent_id)
2386
candidate_delta = inv._make_delta(parent_inv)
2387
if (delta is None or
2388
len(delta) > len(candidate_delta)):
2389
delta = candidate_delta
2390
basis_id = parent_id
2392
# Either none of the parents ended up being suitable, or we
2393
# were asked to delta against NULL
2394
basis_id = _mod_revision.NULL_REVISION
2395
delta = inv._make_delta(null_inventory)
2396
invs_sent_so_far.add(inv.revision_id)
2397
inventory_cache[inv.revision_id] = inv
2398
delta_serialized = ''.join(
2399
serializer.delta_to_lines(basis_id, key[-1], delta))
2400
yield versionedfile.FulltextContentFactory(
2401
key, parent_keys, None, delta_serialized)
2404
class _VersionedFileChecker(object):
2406
def __init__(self, repository, text_key_references=None, ancestors=None):
2407
self.repository = repository
2408
self.text_index = self.repository._generate_text_key_index(
2409
text_key_references=text_key_references, ancestors=ancestors)
2411
def calculate_file_version_parents(self, text_key):
2412
"""Calculate the correct parents for a file version according to
2415
parent_keys = self.text_index[text_key]
2416
if parent_keys == [_mod_revision.NULL_REVISION]:
2418
return tuple(parent_keys)
2420
def check_file_version_parents(self, texts, progress_bar=None):
2421
"""Check the parents stored in a versioned file are correct.
2423
It also detects file versions that are not referenced by their
2424
corresponding revision's inventory.
2426
:returns: A tuple of (wrong_parents, dangling_file_versions).
2427
wrong_parents is a dict mapping {revision_id: (stored_parents,
2428
correct_parents)} for each revision_id where the stored parents
2429
are not correct. dangling_file_versions is a set of (file_id,
2430
revision_id) tuples for versions that are present in this versioned
2431
file, but not used by the corresponding inventory.
2433
local_progress = None
2434
if progress_bar is None:
2435
local_progress = ui.ui_factory.nested_progress_bar()
2436
progress_bar = local_progress
2438
return self._check_file_version_parents(texts, progress_bar)
2441
local_progress.finished()
2443
def _check_file_version_parents(self, texts, progress_bar):
2444
"""See check_file_version_parents."""
2446
self.file_ids = set([file_id for file_id, _ in
2447
self.text_index.iterkeys()])
2448
# text keys is now grouped by file_id
2449
n_versions = len(self.text_index)
2450
progress_bar.update('loading text store', 0, n_versions)
2451
parent_map = self.repository.texts.get_parent_map(self.text_index)
2452
# On unlistable transports this could well be empty/error...
2453
text_keys = self.repository.texts.keys()
2454
unused_keys = frozenset(text_keys) - set(self.text_index)
2455
for num, key in enumerate(self.text_index.iterkeys()):
2456
progress_bar.update('checking text graph', num, n_versions)
2457
correct_parents = self.calculate_file_version_parents(key)
2459
knit_parents = parent_map[key]
2460
except errors.RevisionNotPresent:
2463
if correct_parents != knit_parents:
2464
wrong_parents[key] = (knit_parents, correct_parents)
2465
return wrong_parents, unused_keys
2468
class InterVersionedFileRepository(InterRepository):
2470
_walk_to_common_revisions_batch_size = 50
2473
def fetch(self, revision_id=None, find_ghosts=False,
2475
"""Fetch the content required to construct revision_id.
2477
The content is copied from self.source to self.target.
2479
:param revision_id: if None all content is copied, if NULL_REVISION no
2483
ui.ui_factory.warn_experimental_format_fetch(self)
2484
from bzrlib.fetch import RepoFetcher
2485
# See <https://launchpad.net/bugs/456077> asking for a warning here
2486
if self.source._format.network_name() != self.target._format.network_name():
2487
ui.ui_factory.show_user_warning('cross_format_fetch',
2488
from_format=self.source._format,
2489
to_format=self.target._format)
2490
f = RepoFetcher(to_repository=self.target,
2491
from_repository=self.source,
2492
last_revision=revision_id,
2493
fetch_spec=fetch_spec,
2494
find_ghosts=find_ghosts)
2496
def _walk_to_common_revisions(self, revision_ids, if_present_ids=None):
2497
"""Walk out from revision_ids in source to revisions target has.
2499
:param revision_ids: The start point for the search.
2500
:return: A set of revision ids.
2502
target_graph = self.target.get_graph()
2503
revision_ids = frozenset(revision_ids)
2505
all_wanted_revs = revision_ids.union(if_present_ids)
2507
all_wanted_revs = revision_ids
2508
missing_revs = set()
2509
source_graph = self.source.get_graph()
2510
# ensure we don't pay silly lookup costs.
2511
searcher = source_graph._make_breadth_first_searcher(all_wanted_revs)
2512
null_set = frozenset([_mod_revision.NULL_REVISION])
2513
searcher_exhausted = False
2517
# Iterate the searcher until we have enough next_revs
2518
while len(next_revs) < self._walk_to_common_revisions_batch_size:
2520
next_revs_part, ghosts_part = searcher.next_with_ghosts()
2521
next_revs.update(next_revs_part)
2522
ghosts.update(ghosts_part)
2523
except StopIteration:
2524
searcher_exhausted = True
2526
# If there are ghosts in the source graph, and the caller asked for
2527
# them, make sure that they are present in the target.
2528
# We don't care about other ghosts as we can't fetch them and
2529
# haven't been asked to.
2530
ghosts_to_check = set(revision_ids.intersection(ghosts))
2531
revs_to_get = set(next_revs).union(ghosts_to_check)
2533
have_revs = set(target_graph.get_parent_map(revs_to_get))
2534
# we always have NULL_REVISION present.
2535
have_revs = have_revs.union(null_set)
2536
# Check if the target is missing any ghosts we need.
2537
ghosts_to_check.difference_update(have_revs)
2539
# One of the caller's revision_ids is a ghost in both the
2540
# source and the target.
2541
raise errors.NoSuchRevision(
2542
self.source, ghosts_to_check.pop())
2543
missing_revs.update(next_revs - have_revs)
2544
# Because we may have walked past the original stop point, make
2545
# sure everything is stopped
2546
stop_revs = searcher.find_seen_ancestors(have_revs)
2547
searcher.stop_searching_any(stop_revs)
2548
if searcher_exhausted:
2550
return searcher.get_result()
2553
def search_missing_revision_ids(self,
2554
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
2555
find_ghosts=True, revision_ids=None, if_present_ids=None):
2556
"""Return the revision ids that source has that target does not.
2558
:param revision_id: only return revision ids included by this
2560
:param revision_ids: return revision ids included by these
2561
revision_ids. NoSuchRevision will be raised if any of these
2562
revisions are not present.
2563
:param if_present_ids: like revision_ids, but will not cause
2564
NoSuchRevision if any of these are absent, instead they will simply
2565
not be in the result. This is useful for e.g. finding revisions
2566
to fetch for tags, which may reference absent revisions.
2567
:param find_ghosts: If True find missing revisions in deep history
2568
rather than just finding the surface difference.
2569
:return: A bzrlib.graph.SearchResult.
2571
if symbol_versioning.deprecated_passed(revision_id):
2572
symbol_versioning.warn(
2573
'search_missing_revision_ids(revision_id=...) was '
2574
'deprecated in 2.4. Use revision_ids=[...] instead.',
2575
DeprecationWarning, stacklevel=2)
2576
if revision_ids is not None:
2577
raise AssertionError(
2578
'revision_ids is mutually exclusive with revision_id')
2579
if revision_id is not None:
2580
revision_ids = [revision_id]
2582
# stop searching at found target revisions.
2583
if not find_ghosts and (revision_ids is not None or if_present_ids is
2585
return self._walk_to_common_revisions(revision_ids,
2586
if_present_ids=if_present_ids)
2587
# generic, possibly worst case, slow code path.
2588
target_ids = set(self.target.all_revision_ids())
2589
source_ids = self._present_source_revisions_for(
2590
revision_ids, if_present_ids)
2591
result_set = set(source_ids).difference(target_ids)
2592
return self.source.revision_ids_to_search_result(result_set)
2594
def _present_source_revisions_for(self, revision_ids, if_present_ids=None):
2595
"""Returns set of all revisions in ancestry of revision_ids present in
2598
:param revision_ids: if None, all revisions in source are returned.
2599
:param if_present_ids: like revision_ids, but if any/all of these are
2600
absent no error is raised.
2602
if revision_ids is not None or if_present_ids is not None:
2603
# First, ensure all specified revisions exist. Callers expect
2604
# NoSuchRevision when they pass absent revision_ids here.
2605
if revision_ids is None:
2606
revision_ids = set()
2607
if if_present_ids is None:
2608
if_present_ids = set()
2609
revision_ids = set(revision_ids)
2610
if_present_ids = set(if_present_ids)
2611
all_wanted_ids = revision_ids.union(if_present_ids)
2612
graph = self.source.get_graph()
2613
present_revs = set(graph.get_parent_map(all_wanted_ids))
2614
missing = revision_ids.difference(present_revs)
2616
raise errors.NoSuchRevision(self.source, missing.pop())
2617
found_ids = all_wanted_ids.intersection(present_revs)
2618
source_ids = [rev_id for (rev_id, parents) in
2619
graph.iter_ancestry(found_ids)
2620
if rev_id != _mod_revision.NULL_REVISION
2621
and parents is not None]
2623
source_ids = self.source.all_revision_ids()
2624
return set(source_ids)
2627
def _get_repo_format_to_test(self):
2631
def is_compatible(cls, source, target):
2632
# The default implementation is compatible with everything
2633
return (source._format.supports_full_versioned_files and
2634
target._format.supports_full_versioned_files)
2637
class InterDifferingSerializer(InterVersionedFileRepository):
2640
def _get_repo_format_to_test(self):
2644
def is_compatible(source, target):
2645
if not source._format.supports_full_versioned_files:
2647
if not target._format.supports_full_versioned_files:
2649
# This is redundant with format.check_conversion_target(), however that
2650
# raises an exception, and we just want to say "False" as in we won't
2651
# support converting between these formats.
2652
if 'IDS_never' in debug.debug_flags:
2654
if source.supports_rich_root() and not target.supports_rich_root():
2656
if (source._format.supports_tree_reference
2657
and not target._format.supports_tree_reference):
2659
if target._fallback_repositories and target._format.supports_chks:
2660
# IDS doesn't know how to copy CHKs for the parent inventories it
2661
# adds to stacked repos.
2663
if 'IDS_always' in debug.debug_flags:
2665
# Only use this code path for local source and target. IDS does far
2666
# too much IO (both bandwidth and roundtrips) over a network.
2667
if not source.bzrdir.transport.base.startswith('file:///'):
2669
if not target.bzrdir.transport.base.startswith('file:///'):
2673
def _get_trees(self, revision_ids, cache):
2675
for rev_id in revision_ids:
2677
possible_trees.append((rev_id, cache[rev_id]))
2679
# Not cached, but inventory might be present anyway.
2681
tree = self.source.revision_tree(rev_id)
2682
except errors.NoSuchRevision:
2683
# Nope, parent is ghost.
2686
cache[rev_id] = tree
2687
possible_trees.append((rev_id, tree))
2688
return possible_trees
2690
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
2691
"""Get the best delta and base for this revision.
2693
:return: (basis_id, delta)
2696
# Generate deltas against each tree, to find the shortest.
2697
texts_possibly_new_in_tree = set()
2698
for basis_id, basis_tree in possible_trees:
2699
delta = tree.inventory._make_delta(basis_tree.inventory)
2700
for old_path, new_path, file_id, new_entry in delta:
2701
if new_path is None:
2702
# This file_id isn't present in the new rev, so we don't
2706
# Rich roots are handled elsewhere...
2708
kind = new_entry.kind
2709
if kind != 'directory' and kind != 'file':
2710
# No text record associated with this inventory entry.
2712
# This is a directory or file that has changed somehow.
2713
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
2714
deltas.append((len(delta), basis_id, delta))
2716
return deltas[0][1:]
2718
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
2719
"""Find all parent revisions that are absent, but for which the
2720
inventory is present, and copy those inventories.
2722
This is necessary to preserve correctness when the source is stacked
2723
without fallbacks configured. (Note that in cases like upgrade the
2724
source may be not have _fallback_repositories even though it is
2728
for parents in parent_map.values():
2729
parent_revs.update(parents)
2730
present_parents = self.source.get_parent_map(parent_revs)
2731
absent_parents = set(parent_revs).difference(present_parents)
2732
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
2733
(rev_id,) for rev_id in absent_parents)
2734
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
2735
for parent_tree in self.source.revision_trees(parent_inv_ids):
2736
current_revision_id = parent_tree.get_revision_id()
2737
parents_parents_keys = parent_invs_keys_for_stacking[
2738
(current_revision_id,)]
2739
parents_parents = [key[-1] for key in parents_parents_keys]
2740
basis_id = _mod_revision.NULL_REVISION
2741
basis_tree = self.source.revision_tree(basis_id)
2742
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
2743
self.target.add_inventory_by_delta(
2744
basis_id, delta, current_revision_id, parents_parents)
2745
cache[current_revision_id] = parent_tree
2747
def _fetch_batch(self, revision_ids, basis_id, cache):
2748
"""Fetch across a few revisions.
2750
:param revision_ids: The revisions to copy
2751
:param basis_id: The revision_id of a tree that must be in cache, used
2752
as a basis for delta when no other base is available
2753
:param cache: A cache of RevisionTrees that we can use.
2754
:return: The revision_id of the last converted tree. The RevisionTree
2755
for it will be in cache
2757
# Walk though all revisions; get inventory deltas, copy referenced
2758
# texts that delta references, insert the delta, revision and
2760
root_keys_to_create = set()
2763
pending_revisions = []
2764
parent_map = self.source.get_parent_map(revision_ids)
2765
self._fetch_parent_invs_for_stacking(parent_map, cache)
2766
self.source._safe_to_return_from_cache = True
2767
for tree in self.source.revision_trees(revision_ids):
2768
# Find a inventory delta for this revision.
2769
# Find text entries that need to be copied, too.
2770
current_revision_id = tree.get_revision_id()
2771
parent_ids = parent_map.get(current_revision_id, ())
2772
parent_trees = self._get_trees(parent_ids, cache)
2773
possible_trees = list(parent_trees)
2774
if len(possible_trees) == 0:
2775
# There either aren't any parents, or the parents are ghosts,
2776
# so just use the last converted tree.
2777
possible_trees.append((basis_id, cache[basis_id]))
2778
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
2780
revision = self.source.get_revision(current_revision_id)
2781
pending_deltas.append((basis_id, delta,
2782
current_revision_id, revision.parent_ids))
2783
if self._converting_to_rich_root:
2784
self._revision_id_to_root_id[current_revision_id] = \
2786
# Determine which texts are in present in this revision but not in
2787
# any of the available parents.
2788
texts_possibly_new_in_tree = set()
2789
for old_path, new_path, file_id, entry in delta:
2790
if new_path is None:
2791
# This file_id isn't present in the new rev
2795
if not self.target.supports_rich_root():
2796
# The target doesn't support rich root, so we don't
2799
if self._converting_to_rich_root:
2800
# This can't be copied normally, we have to insert
2802
root_keys_to_create.add((file_id, entry.revision))
2805
texts_possibly_new_in_tree.add((file_id, entry.revision))
2806
for basis_id, basis_tree in possible_trees:
2807
basis_inv = basis_tree.inventory
2808
for file_key in list(texts_possibly_new_in_tree):
2809
file_id, file_revision = file_key
2811
entry = basis_inv[file_id]
2812
except errors.NoSuchId:
2814
if entry.revision == file_revision:
2815
texts_possibly_new_in_tree.remove(file_key)
2816
text_keys.update(texts_possibly_new_in_tree)
2817
pending_revisions.append(revision)
2818
cache[current_revision_id] = tree
2819
basis_id = current_revision_id
2820
self.source._safe_to_return_from_cache = False
2822
from_texts = self.source.texts
2823
to_texts = self.target.texts
2824
if root_keys_to_create:
2825
root_stream = _mod_fetch._new_root_data_stream(
2826
root_keys_to_create, self._revision_id_to_root_id, parent_map,
2828
to_texts.insert_record_stream(root_stream)
2829
to_texts.insert_record_stream(from_texts.get_record_stream(
2830
text_keys, self.target._format._fetch_order,
2831
not self.target._format._fetch_uses_deltas))
2832
# insert inventory deltas
2833
for delta in pending_deltas:
2834
self.target.add_inventory_by_delta(*delta)
2835
if self.target._fallback_repositories:
2836
# Make sure this stacked repository has all the parent inventories
2837
# for the new revisions that we are about to insert. We do this
2838
# before adding the revisions so that no revision is added until
2839
# all the inventories it may depend on are added.
2840
# Note that this is overzealous, as we may have fetched these in an
2843
revision_ids = set()
2844
for revision in pending_revisions:
2845
revision_ids.add(revision.revision_id)
2846
parent_ids.update(revision.parent_ids)
2847
parent_ids.difference_update(revision_ids)
2848
parent_ids.discard(_mod_revision.NULL_REVISION)
2849
parent_map = self.source.get_parent_map(parent_ids)
2850
# we iterate over parent_map and not parent_ids because we don't
2851
# want to try copying any revision which is a ghost
2852
for parent_tree in self.source.revision_trees(parent_map):
2853
current_revision_id = parent_tree.get_revision_id()
2854
parents_parents = parent_map[current_revision_id]
2855
possible_trees = self._get_trees(parents_parents, cache)
2856
if len(possible_trees) == 0:
2857
# There either aren't any parents, or the parents are
2858
# ghosts, so just use the last converted tree.
2859
possible_trees.append((basis_id, cache[basis_id]))
2860
basis_id, delta = self._get_delta_for_revision(parent_tree,
2861
parents_parents, possible_trees)
2862
self.target.add_inventory_by_delta(
2863
basis_id, delta, current_revision_id, parents_parents)
2864
# insert signatures and revisions
2865
for revision in pending_revisions:
2867
signature = self.source.get_signature_text(
2868
revision.revision_id)
2869
self.target.add_signature_text(revision.revision_id,
2871
except errors.NoSuchRevision:
2873
self.target.add_revision(revision.revision_id, revision)
2876
def _fetch_all_revisions(self, revision_ids, pb):
2877
"""Fetch everything for the list of revisions.
2879
:param revision_ids: The list of revisions to fetch. Must be in
2881
:param pb: A ProgressTask
2884
basis_id, basis_tree = self._get_basis(revision_ids[0])
2886
cache = lru_cache.LRUCache(100)
2887
cache[basis_id] = basis_tree
2888
del basis_tree # We don't want to hang on to it here
2892
for offset in range(0, len(revision_ids), batch_size):
2893
self.target.start_write_group()
2895
pb.update('Transferring revisions', offset,
2897
batch = revision_ids[offset:offset+batch_size]
2898
basis_id = self._fetch_batch(batch, basis_id, cache)
2900
self.source._safe_to_return_from_cache = False
2901
self.target.abort_write_group()
2904
hint = self.target.commit_write_group()
2907
if hints and self.target._format.pack_compresses:
2908
self.target.pack(hint=hints)
2909
pb.update('Transferring revisions', len(revision_ids),
2913
def fetch(self, revision_id=None, find_ghosts=False,
2915
"""See InterRepository.fetch()."""
2916
if fetch_spec is not None:
2917
revision_ids = fetch_spec.get_keys()
2920
ui.ui_factory.warn_experimental_format_fetch(self)
2921
if (not self.source.supports_rich_root()
2922
and self.target.supports_rich_root()):
2923
self._converting_to_rich_root = True
2924
self._revision_id_to_root_id = {}
2926
self._converting_to_rich_root = False
2927
# See <https://launchpad.net/bugs/456077> asking for a warning here
2928
if self.source._format.network_name() != self.target._format.network_name():
2929
ui.ui_factory.show_user_warning('cross_format_fetch',
2930
from_format=self.source._format,
2931
to_format=self.target._format)
2932
if revision_ids is None:
2934
search_revision_ids = [revision_id]
2936
search_revision_ids = None
2937
revision_ids = self.target.search_missing_revision_ids(self.source,
2938
revision_ids=search_revision_ids,
2939
find_ghosts=find_ghosts).get_keys()
2940
if not revision_ids:
2942
revision_ids = tsort.topo_sort(
2943
self.source.get_graph().get_parent_map(revision_ids))
2944
if not revision_ids:
2946
# Walk though all revisions; get inventory deltas, copy referenced
2947
# texts that delta references, insert the delta, revision and
2949
pb = ui.ui_factory.nested_progress_bar()
2951
self._fetch_all_revisions(revision_ids, pb)
2954
return len(revision_ids), 0
2956
def _get_basis(self, first_revision_id):
2957
"""Get a revision and tree which exists in the target.
2959
This assumes that first_revision_id is selected for transmission
2960
because all other ancestors are already present. If we can't find an
2961
ancestor we fall back to NULL_REVISION since we know that is safe.
2963
:return: (basis_id, basis_tree)
2965
first_rev = self.source.get_revision(first_revision_id)
2967
basis_id = first_rev.parent_ids[0]
2968
# only valid as a basis if the target has it
2969
self.target.get_revision(basis_id)
2970
# Try to get a basis tree - if it's a ghost it will hit the
2971
# NoSuchRevision case.
2972
basis_tree = self.source.revision_tree(basis_id)
2973
except (IndexError, errors.NoSuchRevision):
2974
basis_id = _mod_revision.NULL_REVISION
2975
basis_tree = self.source.revision_tree(basis_id)
2976
return basis_id, basis_tree
2979
class InterSameDataRepository(InterVersionedFileRepository):
2980
"""Code for converting between repositories that represent the same data.
2982
Data format and model must match for this to work.
2986
def _get_repo_format_to_test(self):
2987
"""Repository format for testing with.
2989
InterSameData can pull from subtree to subtree and from non-subtree to
2990
non-subtree, so we test this with the richest repository format.
2992
from bzrlib.repofmt import knitrepo
2993
return knitrepo.RepositoryFormatKnit3()
2996
def is_compatible(source, target):
2998
InterRepository._same_model(source, target) and
2999
source._format.supports_full_versioned_files and
3000
target._format.supports_full_versioned_files)
3003
InterRepository.register_optimiser(InterVersionedFileRepository)
3004
InterRepository.register_optimiser(InterDifferingSerializer)
3005
InterRepository.register_optimiser(InterSameDataRepository)
3008
def install_revisions(repository, iterable, num_revisions=None, pb=None):
3009
"""Install all revision data into a repository.
3011
Accepts an iterable of revision, tree, signature tuples. The signature
3014
repository.start_write_group()
3016
inventory_cache = lru_cache.LRUCache(10)
3017
for n, (revision, revision_tree, signature) in enumerate(iterable):
3018
_install_revision(repository, revision, revision_tree, signature,
3021
pb.update('Transferring revisions', n + 1, num_revisions)
3023
repository.abort_write_group()
3026
repository.commit_write_group()
3029
def _install_revision(repository, rev, revision_tree, signature,
3031
"""Install all revision data into a repository."""
3032
present_parents = []
3034
for p_id in rev.parent_ids:
3035
if repository.has_revision(p_id):
3036
present_parents.append(p_id)
3037
parent_trees[p_id] = repository.revision_tree(p_id)
3039
parent_trees[p_id] = repository.revision_tree(
3040
_mod_revision.NULL_REVISION)
3042
inv = revision_tree.inventory
3043
entries = inv.iter_entries()
3044
# backwards compatibility hack: skip the root id.
3045
if not repository.supports_rich_root():
3046
path, root = entries.next()
3047
if root.revision != rev.revision_id:
3048
raise errors.IncompatibleRevision(repr(repository))
3050
for path, ie in entries:
3051
text_keys[(ie.file_id, ie.revision)] = ie
3052
text_parent_map = repository.texts.get_parent_map(text_keys)
3053
missing_texts = set(text_keys) - set(text_parent_map)
3054
# Add the texts that are not already present
3055
for text_key in missing_texts:
3056
ie = text_keys[text_key]
3058
# FIXME: TODO: The following loop overlaps/duplicates that done by
3059
# commit to determine parents. There is a latent/real bug here where
3060
# the parents inserted are not those commit would do - in particular
3061
# they are not filtered by heads(). RBC, AB
3062
for revision, tree in parent_trees.iteritems():
3063
if ie.file_id not in tree:
3065
parent_id = tree.get_file_revision(ie.file_id)
3066
if parent_id in text_parents:
3068
text_parents.append((ie.file_id, parent_id))
3069
lines = revision_tree.get_file(ie.file_id).readlines()
3070
repository.texts.add_lines(text_key, text_parents, lines)
3072
# install the inventory
3073
if repository._format._commit_inv_deltas and len(rev.parent_ids):
3074
# Cache this inventory
3075
inventory_cache[rev.revision_id] = inv
3077
basis_inv = inventory_cache[rev.parent_ids[0]]
3079
repository.add_inventory(rev.revision_id, inv, present_parents)
3081
delta = inv._make_delta(basis_inv)
3082
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
3083
rev.revision_id, present_parents)
3085
repository.add_inventory(rev.revision_id, inv, present_parents)
3086
except errors.RevisionAlreadyPresent:
3088
if signature is not None:
3089
repository.add_signature_text(rev.revision_id, signature)
3090
repository.add_revision(rev.revision_id, rev, inv)
3093
def install_revision(repository, rev, revision_tree):
3094
"""Install all revision data into a repository."""
3095
install_revisions(repository, [(rev, revision_tree, None)])
added
removed
bzrlib/vf_repository.py
