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."""
19
from __future__ import absolute_import
22
from .lazy_import import lazy_import
23
lazy_import(globals(), """
28
config as _mod_config,
37
revision as _mod_revision,
38
serializer as _mod_serializer,
46
from breezy.recordcounter import RecordCounter
47
from breezy.revisiontree import InventoryRevisionTree
48
from breezy.testament import Testament
49
from breezy.i18n import gettext
55
from .decorators import (
60
from .inventory import (
67
from .repository import (
71
RepositoryFormatMetaDir,
87
class VersionedFileRepositoryFormat(RepositoryFormat):
88
"""Base class for all repository formats that are VersionedFiles-based."""
90
supports_full_versioned_files = True
91
supports_versioned_directories = True
92
supports_unreferenced_revisions = True
94
# Should commit add an inventory, or an inventory delta to the repository.
95
_commit_inv_deltas = True
96
# What order should fetch operations request streams in?
97
# The default is unordered as that is the cheapest for an origin to
99
_fetch_order = 'unordered'
100
# Does this repository format use deltas that can be fetched as-deltas ?
101
# (E.g. knits, where the knit deltas can be transplanted intact.
102
# We default to False, which will ensure that enough data to get
103
# a full text out of any fetch stream will be grabbed.
104
_fetch_uses_deltas = False
107
class VersionedFileCommitBuilder(CommitBuilder):
108
"""Commit builder implementation for versioned files based repositories.
111
# this commit builder supports the record_entry_contents interface
112
supports_record_entry_contents = True
114
# the default CommitBuilder does not manage trees whose root is versioned.
115
_versioned_root = False
117
def __init__(self, repository, parents, config_stack, timestamp=None,
118
timezone=None, committer=None, revprops=None,
119
revision_id=None, lossy=False):
120
super(VersionedFileCommitBuilder, self).__init__(repository,
121
parents, config_stack, timestamp, timezone, committer, revprops,
124
basis_id = self.parents[0]
126
basis_id = _mod_revision.NULL_REVISION
127
self.basis_delta_revision = basis_id
128
self.new_inventory = Inventory(None)
129
self._basis_delta = []
130
self.__heads = graph.HeadsCache(repository.get_graph()).heads
131
# memo'd check for no-op commits.
132
self._any_changes = False
133
# API compatibility, older code that used CommitBuilder did not call
134
# .record_delete(), which means the delta that is computed would not be
135
# valid. Callers that will call record_delete() should call
136
# .will_record_deletes() to indicate that.
137
self._recording_deletes = False
139
def will_record_deletes(self):
140
"""Tell the commit builder that deletes are being notified.
142
This enables the accumulation of an inventory delta; for the resulting
143
commit to be valid, deletes against the basis MUST be recorded via
144
builder.record_delete().
146
self._recording_deletes = True
148
def any_changes(self):
149
"""Return True if any entries were changed.
151
This includes merge-only changes. It is the core for the --unchanged
154
:return: True if any changes have occured.
156
return self._any_changes
158
def _ensure_fallback_inventories(self):
159
"""Ensure that appropriate inventories are available.
161
This only applies to repositories that are stacked, and is about
162
enusring the stacking invariants. Namely, that for any revision that is
163
present, we either have all of the file content, or we have the parent
164
inventory and the delta file content.
166
if not self.repository._fallback_repositories:
168
if not self.repository._format.supports_chks:
169
raise errors.BzrError("Cannot commit directly to a stacked branch"
170
" in pre-2a formats. See "
171
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
172
# This is a stacked repo, we need to make sure we have the parent
173
# inventories for the parents.
174
parent_keys = [(p,) for p in self.parents]
175
parent_map = self.repository.inventories._index.get_parent_map(parent_keys)
176
missing_parent_keys = {pk for pk in parent_keys
177
if pk not in parent_map}
178
fallback_repos = list(reversed(self.repository._fallback_repositories))
179
missing_keys = [('inventories', pk[0])
180
for pk in missing_parent_keys]
182
while missing_keys and fallback_repos:
183
fallback_repo = fallback_repos.pop()
184
source = fallback_repo._get_source(self.repository._format)
185
sink = self.repository._get_sink()
186
stream = source.get_stream_for_missing_keys(missing_keys)
187
missing_keys = sink.insert_stream_without_locking(stream,
188
self.repository._format)
190
raise errors.BzrError('Unable to fill in parent inventories for a'
193
def commit(self, message):
194
"""Make the actual commit.
196
:return: The revision id of the recorded revision.
198
self._validate_unicode_text(message, 'commit message')
199
rev = _mod_revision.Revision(
200
timestamp=self._timestamp,
201
timezone=self._timezone,
202
committer=self._committer,
204
inventory_sha1=self.inv_sha1,
205
revision_id=self._new_revision_id,
206
properties=self._revprops)
207
rev.parent_ids = self.parents
208
if self._config_stack.get('create_signatures') == _mod_config.SIGN_ALWAYS:
209
testament = Testament(rev, self.revision_tree())
210
plaintext = testament.as_short_text()
211
self.repository.store_revision_signature(
212
gpg.GPGStrategy(self._config_stack), plaintext,
213
self._new_revision_id)
214
self.repository._add_revision(rev)
215
self._ensure_fallback_inventories()
216
self.repository.commit_write_group()
217
return self._new_revision_id
220
"""Abort the commit that is being built.
222
self.repository.abort_write_group()
224
def revision_tree(self):
225
"""Return the tree that was just committed.
227
After calling commit() this can be called to get a
228
RevisionTree representing the newly committed tree. This is
229
preferred to calling Repository.revision_tree() because that may
230
require deserializing the inventory, while we already have a copy in
233
if self.new_inventory is None:
234
self.new_inventory = self.repository.get_inventory(
235
self._new_revision_id)
236
return InventoryRevisionTree(self.repository, self.new_inventory,
237
self._new_revision_id)
239
def finish_inventory(self):
240
"""Tell the builder that the inventory is finished.
242
:return: The inventory id in the repository, which can be used with
243
repository.get_inventory.
245
if self.new_inventory is None:
246
# an inventory delta was accumulated without creating a new
248
basis_id = self.basis_delta_revision
249
# We ignore the 'inventory' returned by add_inventory_by_delta
250
# because self.new_inventory is used to hint to the rest of the
251
# system what code path was taken
252
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
253
basis_id, self._basis_delta, self._new_revision_id,
256
if self.new_inventory.root is None:
257
raise AssertionError('Root entry should be supplied to'
258
' record_entry_contents, as of bzr 0.10.')
259
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
260
self.new_inventory.revision_id = self._new_revision_id
261
self.inv_sha1 = self.repository.add_inventory(
262
self._new_revision_id,
266
return self._new_revision_id
268
def _check_root(self, ie, parent_invs, tree):
269
"""Helper for record_entry_contents.
271
:param ie: An entry being added.
272
:param parent_invs: The inventories of the parent revisions of the
274
:param tree: The tree that is being committed.
276
# In this revision format, root entries have no knit or weave When
277
# serializing out to disk and back in root.revision is always
279
ie.revision = self._new_revision_id
281
def _require_root_change(self, tree):
282
"""Enforce an appropriate root object change.
284
This is called once when record_iter_changes is called, if and only if
285
the root was not in the delta calculated by record_iter_changes.
287
:param tree: The tree which is being committed.
289
if len(self.parents) == 0:
290
raise errors.RootMissing()
291
entry = entry_factory['directory'](tree.path2id(''), '',
293
entry.revision = self._new_revision_id
294
self._basis_delta.append(('', '', entry.file_id, entry))
296
def _get_delta(self, ie, basis_inv, path):
297
"""Get a delta against the basis inventory for ie."""
298
if not basis_inv.has_id(ie.file_id):
300
result = (None, path, ie.file_id, ie)
301
self._basis_delta.append(result)
303
elif ie != basis_inv[ie.file_id]:
305
# TODO: avoid tis id2path call.
306
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
307
self._basis_delta.append(result)
313
def _heads(self, file_id, revision_ids):
314
"""Calculate the graph heads for revision_ids in the graph of file_id.
316
This can use either a per-file graph or a global revision graph as we
317
have an identity relationship between the two graphs.
319
return self.__heads(revision_ids)
321
def get_basis_delta(self):
322
"""Return the complete inventory delta versus the basis inventory.
324
This has been built up with the calls to record_delete and
325
record_entry_contents. The client must have already called
326
will_record_deletes() to indicate that they will be generating a
329
:return: An inventory delta, suitable for use with apply_delta, or
330
Repository.add_inventory_by_delta, etc.
332
if not self._recording_deletes:
333
raise AssertionError("recording deletes not activated.")
334
return self._basis_delta
336
def record_delete(self, path, file_id):
337
"""Record that a delete occured against a basis tree.
339
This is an optional API - when used it adds items to the basis_delta
340
being accumulated by the commit builder. It cannot be called unless the
341
method will_record_deletes() has been called to inform the builder that
342
a delta is being supplied.
344
:param path: The path of the thing deleted.
345
:param file_id: The file id that was deleted.
347
if not self._recording_deletes:
348
raise AssertionError("recording deletes not activated.")
349
delta = (path, None, file_id, None)
350
self._basis_delta.append(delta)
351
self._any_changes = True
354
def record_entry_contents(self, ie, parent_invs, path, tree,
356
"""Record the content of ie from tree into the commit if needed.
358
Side effect: sets ie.revision when unchanged
360
:param ie: An inventory entry present in the commit.
361
:param parent_invs: The inventories of the parent revisions of the
363
:param path: The path the entry is at in the tree.
364
:param tree: The tree which contains this entry and should be used to
366
:param content_summary: Summary data from the tree about the paths
367
content - stat, length, exec, sha/link target. This is only
368
accessed when the entry has a revision of None - that is when it is
369
a candidate to commit.
370
:return: A tuple (change_delta, version_recorded, fs_hash).
371
change_delta is an inventory_delta change for this entry against
372
the basis tree of the commit, or None if no change occured against
374
version_recorded is True if a new version of the entry has been
375
recorded. For instance, committing a merge where a file was only
376
changed on the other side will return (delta, False).
377
fs_hash is either None, or the hash details for the path (currently
378
a tuple of the contents sha1 and the statvalue returned by
379
tree.get_file_with_stat()).
381
if self.new_inventory.root is None:
382
if ie.parent_id is not None:
383
raise errors.RootMissing()
384
self._check_root(ie, parent_invs, tree)
385
if ie.revision is None:
386
kind = content_summary[0]
388
# ie is carried over from a prior commit
390
# XXX: repository specific check for nested tree support goes here - if
391
# the repo doesn't want nested trees we skip it ?
392
if (kind == 'tree-reference' and
393
not self.repository._format.supports_tree_reference):
394
# mismatch between commit builder logic and repository:
395
# this needs the entry creation pushed down into the builder.
396
raise NotImplementedError('Missing repository subtree support.')
397
self.new_inventory.add(ie)
399
# TODO: slow, take it out of the inner loop.
401
basis_inv = parent_invs[0]
403
basis_inv = Inventory(root_id=None)
405
# ie.revision is always None if the InventoryEntry is considered
406
# for committing. We may record the previous parents revision if the
407
# content is actually unchanged against a sole head.
408
if ie.revision is not None:
409
if not self._versioned_root and path == '':
410
# repositories that do not version the root set the root's
411
# revision to the new commit even when no change occurs (more
412
# specifically, they do not record a revision on the root; and
413
# the rev id is assigned to the root during deserialisation -
414
# this masks when a change may have occurred against the basis.
415
# To match this we always issue a delta, because the revision
416
# of the root will always be changing.
417
if basis_inv.has_id(ie.file_id):
418
delta = (basis_inv.id2path(ie.file_id), path,
422
delta = (None, path, ie.file_id, ie)
423
self._basis_delta.append(delta)
424
return delta, False, None
426
# we don't need to commit this, because the caller already
427
# determined that an existing revision of this file is
428
# appropriate. If it's not being considered for committing then
429
# it and all its parents to the root must be unaltered so
430
# no-change against the basis.
431
if ie.revision == self._new_revision_id:
432
raise AssertionError("Impossible situation, a skipped "
433
"inventory entry (%r) claims to be modified in this "
434
"commit (%r).", (ie, self._new_revision_id))
435
return None, False, None
436
# XXX: Friction: parent_candidates should return a list not a dict
437
# so that we don't have to walk the inventories again.
438
parent_candidate_entries = ie.parent_candidates(parent_invs)
439
head_set = self._heads(ie.file_id, parent_candidate_entries)
441
for inv in parent_invs:
442
if inv.has_id(ie.file_id):
443
old_rev = inv[ie.file_id].revision
444
if old_rev in head_set:
445
heads.append(inv[ie.file_id].revision)
446
head_set.remove(inv[ie.file_id].revision)
449
# now we check to see if we need to write a new record to the
451
# We write a new entry unless there is one head to the ancestors, and
452
# the kind-derived content is unchanged.
454
# Cheapest check first: no ancestors, or more the one head in the
455
# ancestors, we write a new node.
459
# There is a single head, look it up for comparison
460
parent_entry = parent_candidate_entries[heads[0]]
461
# if the non-content specific data has changed, we'll be writing a
463
if (parent_entry.parent_id != ie.parent_id or
464
parent_entry.name != ie.name):
466
# now we need to do content specific checks:
468
# if the kind changed the content obviously has
469
if kind != parent_entry.kind:
471
# Stat cache fingerprint feedback for the caller - None as we usually
472
# don't generate one.
475
if content_summary[2] is None:
476
raise ValueError("Files must not have executable = None")
478
# We can't trust a check of the file length because of content
480
if (# if the exec bit has changed we have to store:
481
parent_entry.executable != content_summary[2]):
483
elif parent_entry.text_sha1 == content_summary[3]:
484
# all meta and content is unchanged (using a hash cache
485
# hit to check the sha)
486
ie.revision = parent_entry.revision
487
ie.text_size = parent_entry.text_size
488
ie.text_sha1 = parent_entry.text_sha1
489
ie.executable = parent_entry.executable
490
return self._get_delta(ie, basis_inv, path), False, None
492
# Either there is only a hash change(no hash cache entry,
493
# or same size content change), or there is no change on
495
# Provide the parent's hash to the store layer, so that the
496
# content is unchanged we will not store a new node.
497
nostore_sha = parent_entry.text_sha1
499
# We want to record a new node regardless of the presence or
500
# absence of a content change in the file.
502
ie.executable = content_summary[2]
503
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
505
text = file_obj.read()
509
ie.text_sha1, ie.text_size = self._add_text_to_weave(
510
ie.file_id, text, heads, nostore_sha)
511
# Let the caller know we generated a stat fingerprint.
512
fingerprint = (ie.text_sha1, stat_value)
513
except errors.ExistingContent:
514
# Turns out that the file content was unchanged, and we were
515
# only going to store a new node if it was changed. Carry over
517
ie.revision = parent_entry.revision
518
ie.text_size = parent_entry.text_size
519
ie.text_sha1 = parent_entry.text_sha1
520
ie.executable = parent_entry.executable
521
return self._get_delta(ie, basis_inv, path), False, None
522
elif kind == 'directory':
524
# all data is meta here, nothing specific to directory, so
526
ie.revision = parent_entry.revision
527
return self._get_delta(ie, basis_inv, path), False, None
528
self._add_text_to_weave(ie.file_id, '', heads, None)
529
elif kind == 'symlink':
530
current_link_target = content_summary[3]
532
# symlink target is not generic metadata, check if it has
534
if current_link_target != parent_entry.symlink_target:
537
# unchanged, carry over.
538
ie.revision = parent_entry.revision
539
ie.symlink_target = parent_entry.symlink_target
540
return self._get_delta(ie, basis_inv, path), False, None
541
ie.symlink_target = current_link_target
542
self._add_text_to_weave(ie.file_id, '', heads, None)
543
elif kind == 'tree-reference':
545
if content_summary[3] != parent_entry.reference_revision:
548
# unchanged, carry over.
549
ie.reference_revision = parent_entry.reference_revision
550
ie.revision = parent_entry.revision
551
return self._get_delta(ie, basis_inv, path), False, None
552
ie.reference_revision = content_summary[3]
553
if ie.reference_revision is None:
554
raise AssertionError("invalid content_summary for nested tree: %r"
555
% (content_summary,))
556
self._add_text_to_weave(ie.file_id, '', heads, None)
558
raise NotImplementedError('unknown kind')
559
ie.revision = self._new_revision_id
560
# The initial commit adds a root directory, but this in itself is not
561
# a worthwhile commit.
562
if (self.basis_delta_revision != _mod_revision.NULL_REVISION or
564
self._any_changes = True
565
return self._get_delta(ie, basis_inv, path), True, fingerprint
567
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
568
_entry_factory=entry_factory):
569
"""Record a new tree via iter_changes.
571
:param tree: The tree to obtain text contents from for changed objects.
572
:param basis_revision_id: The revision id of the tree the iter_changes
573
has been generated against. Currently assumed to be the same
574
as self.parents[0] - if it is not, errors may occur.
575
:param iter_changes: An iter_changes iterator with the changes to apply
576
to basis_revision_id. The iterator must not include any items with
577
a current kind of None - missing items must be either filtered out
578
or errored-on before record_iter_changes sees the item.
579
:param _entry_factory: Private method to bind entry_factory locally for
581
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
584
# Create an inventory delta based on deltas between all the parents and
585
# deltas between all the parent inventories. We use inventory delta's
586
# between the inventory objects because iter_changes masks
587
# last-changed-field only changes.
589
# file_id -> change map, change is fileid, paths, changed, versioneds,
590
# parents, names, kinds, executables
592
# {file_id -> revision_id -> inventory entry, for entries in parent
593
# trees that are not parents[0]
597
revtrees = list(self.repository.revision_trees(self.parents))
598
except errors.NoSuchRevision:
599
# one or more ghosts, slow path.
601
for revision_id in self.parents:
603
revtrees.append(self.repository.revision_tree(revision_id))
604
except errors.NoSuchRevision:
606
basis_revision_id = _mod_revision.NULL_REVISION
608
revtrees.append(self.repository.revision_tree(
609
_mod_revision.NULL_REVISION))
610
# The basis inventory from a repository
612
basis_tree = revtrees[0]
614
basis_tree = self.repository.revision_tree(
615
_mod_revision.NULL_REVISION)
616
basis_inv = basis_tree.root_inventory
617
if len(self.parents) > 0:
618
if basis_revision_id != self.parents[0] and not ghost_basis:
620
"arbitrary basis parents not yet supported with merges")
621
for revtree in revtrees[1:]:
622
for change in revtree.root_inventory._make_delta(basis_inv):
623
if change[1] is None:
624
# Not present in this parent.
626
if change[2] not in merged_ids:
627
if change[0] is not None:
628
basis_entry = basis_inv[change[2]]
629
merged_ids[change[2]] = [
631
basis_entry.revision,
634
parent_entries[change[2]] = {
636
basis_entry.revision:basis_entry,
638
change[3].revision:change[3],
641
merged_ids[change[2]] = [change[3].revision]
642
parent_entries[change[2]] = {change[3].revision:change[3]}
644
merged_ids[change[2]].append(change[3].revision)
645
parent_entries[change[2]][change[3].revision] = change[3]
648
# Setup the changes from the tree:
649
# changes maps file_id -> (change, [parent revision_ids])
651
for change in iter_changes:
652
# This probably looks up in basis_inv way to much.
653
if change[1][0] is not None:
654
head_candidate = [basis_inv[change[0]].revision]
657
changes[change[0]] = change, merged_ids.get(change[0],
659
unchanged_merged = set(merged_ids) - set(changes)
660
# Extend the changes dict with synthetic changes to record merges of
662
for file_id in unchanged_merged:
663
# Record a merged version of these items that did not change vs the
664
# basis. This can be either identical parallel changes, or a revert
665
# of a specific file after a merge. The recorded content will be
666
# that of the current tree (which is the same as the basis), but
667
# the per-file graph will reflect a merge.
668
# NB:XXX: We are reconstructing path information we had, this
669
# should be preserved instead.
670
# inv delta change: (file_id, (path_in_source, path_in_target),
671
# changed_content, versioned, parent, name, kind,
674
basis_entry = basis_inv[file_id]
675
except errors.NoSuchId:
676
# a change from basis->some_parents but file_id isn't in basis
677
# so was new in the merge, which means it must have changed
678
# from basis -> current, and as it hasn't the add was reverted
679
# by the user. So we discard this change.
683
(basis_inv.id2path(file_id), tree.id2path(file_id)),
685
(basis_entry.parent_id, basis_entry.parent_id),
686
(basis_entry.name, basis_entry.name),
687
(basis_entry.kind, basis_entry.kind),
688
(basis_entry.executable, basis_entry.executable))
689
changes[file_id] = (change, merged_ids[file_id])
690
# changes contains tuples with the change and a set of inventory
691
# candidates for the file.
693
# old_path, new_path, file_id, new_inventory_entry
694
seen_root = False # Is the root in the basis delta?
695
inv_delta = self._basis_delta
696
modified_rev = self._new_revision_id
697
for change, head_candidates in viewvalues(changes):
698
if change[3][1]: # versioned in target.
699
# Several things may be happening here:
700
# We may have a fork in the per-file graph
701
# - record a change with the content from tree
702
# We may have a change against < all trees
703
# - carry over the tree that hasn't changed
704
# We may have a change against all trees
705
# - record the change with the content from tree
708
entry = _entry_factory[kind](file_id, change[5][1],
710
head_set = self._heads(change[0], set(head_candidates))
713
for head_candidate in head_candidates:
714
if head_candidate in head_set:
715
heads.append(head_candidate)
716
head_set.remove(head_candidate)
719
# Could be a carry-over situation:
720
parent_entry_revs = parent_entries.get(file_id, None)
721
if parent_entry_revs:
722
parent_entry = parent_entry_revs.get(heads[0], None)
725
if parent_entry is None:
726
# The parent iter_changes was called against is the one
727
# that is the per-file head, so any change is relevant
728
# iter_changes is valid.
729
carry_over_possible = False
731
# could be a carry over situation
732
# A change against the basis may just indicate a merge,
733
# we need to check the content against the source of the
734
# merge to determine if it was changed after the merge
736
if (parent_entry.kind != entry.kind or
737
parent_entry.parent_id != entry.parent_id or
738
parent_entry.name != entry.name):
739
# Metadata common to all entries has changed
740
# against per-file parent
741
carry_over_possible = False
743
carry_over_possible = True
744
# per-type checks for changes against the parent_entry
747
# Cannot be a carry-over situation
748
carry_over_possible = False
749
# Populate the entry in the delta
751
# XXX: There is still a small race here: If someone reverts the content of a file
752
# after iter_changes examines and decides it has changed,
753
# we will unconditionally record a new version even if some
754
# other process reverts it while commit is running (with
755
# the revert happening after iter_changes did its
758
entry.executable = True
760
entry.executable = False
761
if (carry_over_possible and
762
parent_entry.executable == entry.executable):
763
# Check the file length, content hash after reading
765
nostore_sha = parent_entry.text_sha1
768
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
770
text = file_obj.read()
774
entry.text_sha1, entry.text_size = self._add_text_to_weave(
775
file_id, text, heads, nostore_sha)
776
yield file_id, change[1][1], (entry.text_sha1, stat_value)
777
except errors.ExistingContent:
778
# No content change against a carry_over parent
779
# Perhaps this should also yield a fs hash update?
781
entry.text_size = parent_entry.text_size
782
entry.text_sha1 = parent_entry.text_sha1
783
elif kind == 'symlink':
785
entry.symlink_target = tree.get_symlink_target(file_id)
786
if (carry_over_possible and
787
parent_entry.symlink_target == entry.symlink_target):
790
self._add_text_to_weave(change[0], '', heads, None)
791
elif kind == 'directory':
792
if carry_over_possible:
795
# Nothing to set on the entry.
796
# XXX: split into the Root and nonRoot versions.
797
if change[1][1] != '' or self.repository.supports_rich_root():
798
self._add_text_to_weave(change[0], '', heads, None)
799
elif kind == 'tree-reference':
800
if not self.repository._format.supports_tree_reference:
801
# This isn't quite sane as an error, but we shouldn't
802
# ever see this code path in practice: tree's don't
803
# permit references when the repo doesn't support tree
805
raise errors.UnsupportedOperation(tree.add_reference,
807
reference_revision = tree.get_reference_revision(change[0])
808
entry.reference_revision = reference_revision
809
if (carry_over_possible and
810
parent_entry.reference_revision == reference_revision):
813
self._add_text_to_weave(change[0], '', heads, None)
815
raise AssertionError('unknown kind %r' % kind)
817
entry.revision = modified_rev
819
entry.revision = parent_entry.revision
822
new_path = change[1][1]
823
inv_delta.append((change[1][0], new_path, change[0], entry))
826
self.new_inventory = None
827
# The initial commit adds a root directory, but this in itself is not
828
# a worthwhile commit.
829
if ((len(inv_delta) > 0 and basis_revision_id != _mod_revision.NULL_REVISION) or
830
(len(inv_delta) > 1 and basis_revision_id == _mod_revision.NULL_REVISION)):
831
# This should perhaps be guarded by a check that the basis we
832
# commit against is the basis for the commit and if not do a delta
834
self._any_changes = True
836
# housekeeping root entry changes do not affect no-change commits.
837
self._require_root_change(tree)
838
self.basis_delta_revision = basis_revision_id
840
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
841
parent_keys = tuple([(file_id, parent) for parent in parents])
842
return self.repository.texts._add_text(
843
(file_id, self._new_revision_id), parent_keys, new_text,
844
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
847
class VersionedFileRootCommitBuilder(VersionedFileCommitBuilder):
848
"""This commitbuilder actually records the root id"""
850
# the root entry gets versioned properly by this builder.
851
_versioned_root = True
853
def _check_root(self, ie, parent_invs, tree):
854
"""Helper for record_entry_contents.
856
:param ie: An entry being added.
857
:param parent_invs: The inventories of the parent revisions of the
859
:param tree: The tree that is being committed.
862
def _require_root_change(self, tree):
863
"""Enforce an appropriate root object change.
865
This is called once when record_iter_changes is called, if and only if
866
the root was not in the delta calculated by record_iter_changes.
868
:param tree: The tree which is being committed.
870
# versioned roots do not change unless the tree found a change.
873
class VersionedFileRepository(Repository):
874
"""Repository holding history for one or more branches.
876
The repository holds and retrieves historical information including
877
revisions and file history. It's normally accessed only by the Branch,
878
which views a particular line of development through that history.
880
The Repository builds on top of some byte storage facilies (the revisions,
881
signatures, inventories, texts and chk_bytes attributes) and a Transport,
882
which respectively provide byte storage and a means to access the (possibly
885
The byte storage facilities are addressed via tuples, which we refer to
886
as 'keys' throughout the code base. Revision_keys, inventory_keys and
887
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
888
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
889
byte string made up of a hash identifier and a hash value.
890
We use this interface because it allows low friction with the underlying
891
code that implements disk indices, network encoding and other parts of
894
:ivar revisions: A breezy.versionedfile.VersionedFiles instance containing
895
the serialised revisions for the repository. This can be used to obtain
896
revision graph information or to access raw serialised revisions.
897
The result of trying to insert data into the repository via this store
898
is undefined: it should be considered read-only except for implementors
900
:ivar signatures: A breezy.versionedfile.VersionedFiles instance containing
901
the serialised signatures for the repository. This can be used to
902
obtain access to raw serialised signatures. The result of trying to
903
insert data into the repository via this store is undefined: it should
904
be considered read-only except for implementors of repositories.
905
:ivar inventories: A breezy.versionedfile.VersionedFiles instance containing
906
the serialised inventories for the repository. This can be used to
907
obtain unserialised inventories. The result of trying to insert data
908
into the repository via this store is undefined: it should be
909
considered read-only except for implementors of repositories.
910
:ivar texts: A breezy.versionedfile.VersionedFiles instance containing the
911
texts of files and directories for the repository. This can be used to
912
obtain file texts or file graphs. Note that Repository.iter_file_bytes
913
is usually a better interface for accessing file texts.
914
The result of trying to insert data into the repository via this store
915
is undefined: it should be considered read-only except for implementors
917
:ivar chk_bytes: A breezy.versionedfile.VersionedFiles instance containing
918
any data the repository chooses to store or have indexed by its hash.
919
The result of trying to insert data into the repository via this store
920
is undefined: it should be considered read-only except for implementors
922
:ivar _transport: Transport for file access to repository, typically
923
pointing to .bzr/repository.
926
# What class to use for a CommitBuilder. Often it's simpler to change this
927
# in a Repository class subclass rather than to override
928
# get_commit_builder.
929
_commit_builder_class = VersionedFileCommitBuilder
931
def add_fallback_repository(self, repository):
932
"""Add a repository to use for looking up data not held locally.
934
:param repository: A repository.
936
if not self._format.supports_external_lookups:
937
raise errors.UnstackableRepositoryFormat(self._format, self.base)
938
# This can raise an exception, so should be done before we lock the
939
# fallback repository.
940
self._check_fallback_repository(repository)
942
# This repository will call fallback.unlock() when we transition to
943
# the unlocked state, so we make sure to increment the lock count
944
repository.lock_read()
945
self._fallback_repositories.append(repository)
946
self.texts.add_fallback_versioned_files(repository.texts)
947
self.inventories.add_fallback_versioned_files(repository.inventories)
948
self.revisions.add_fallback_versioned_files(repository.revisions)
949
self.signatures.add_fallback_versioned_files(repository.signatures)
950
if self.chk_bytes is not None:
951
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
953
@only_raises(errors.LockNotHeld, errors.LockBroken)
955
super(VersionedFileRepository, self).unlock()
956
if self.control_files._lock_count == 0:
957
self._inventory_entry_cache.clear()
959
def add_inventory(self, revision_id, inv, parents):
960
"""Add the inventory inv to the repository as revision_id.
962
:param parents: The revision ids of the parents that revision_id
963
is known to have and are in the repository already.
965
:returns: The validator(which is a sha1 digest, though what is sha'd is
966
repository format specific) of the serialized inventory.
968
if not self.is_in_write_group():
969
raise AssertionError("%r not in write group" % (self,))
970
_mod_revision.check_not_reserved_id(revision_id)
971
if not (inv.revision_id is None or inv.revision_id == revision_id):
972
raise AssertionError(
973
"Mismatch between inventory revision"
974
" id and insertion revid (%r, %r)"
975
% (inv.revision_id, revision_id))
977
raise errors.RootMissing()
978
return self._add_inventory_checked(revision_id, inv, parents)
980
def _add_inventory_checked(self, revision_id, inv, parents):
981
"""Add inv to the repository after checking the inputs.
983
This function can be overridden to allow different inventory styles.
985
:seealso: add_inventory, for the contract.
987
inv_lines = self._serializer.write_inventory_to_lines(inv)
988
return self._inventory_add_lines(revision_id, parents,
989
inv_lines, check_content=False)
991
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
992
parents, basis_inv=None, propagate_caches=False):
993
"""Add a new inventory expressed as a delta against another revision.
995
See the inventory developers documentation for the theory behind
998
:param basis_revision_id: The inventory id the delta was created
999
against. (This does not have to be a direct parent.)
1000
:param delta: The inventory delta (see Inventory.apply_delta for
1002
:param new_revision_id: The revision id that the inventory is being
1004
:param parents: The revision ids of the parents that revision_id is
1005
known to have and are in the repository already. These are supplied
1006
for repositories that depend on the inventory graph for revision
1007
graph access, as well as for those that pun ancestry with delta
1009
:param basis_inv: The basis inventory if it is already known,
1011
:param propagate_caches: If True, the caches for this inventory are
1012
copied to and updated for the result if possible.
1014
:returns: (validator, new_inv)
1015
The validator(which is a sha1 digest, though what is sha'd is
1016
repository format specific) of the serialized inventory, and the
1017
resulting inventory.
1019
if not self.is_in_write_group():
1020
raise AssertionError("%r not in write group" % (self,))
1021
_mod_revision.check_not_reserved_id(new_revision_id)
1022
basis_tree = self.revision_tree(basis_revision_id)
1023
basis_tree.lock_read()
1025
# Note that this mutates the inventory of basis_tree, which not all
1026
# inventory implementations may support: A better idiom would be to
1027
# return a new inventory, but as there is no revision tree cache in
1028
# repository this is safe for now - RBC 20081013
1029
if basis_inv is None:
1030
basis_inv = basis_tree.root_inventory
1031
basis_inv.apply_delta(delta)
1032
basis_inv.revision_id = new_revision_id
1033
return (self.add_inventory(new_revision_id, basis_inv, parents),
1038
def _inventory_add_lines(self, revision_id, parents, lines,
1039
check_content=True):
1040
"""Store lines in inv_vf and return the sha1 of the inventory."""
1041
parents = [(parent,) for parent in parents]
1042
result = self.inventories.add_lines((revision_id,), parents, lines,
1043
check_content=check_content)[0]
1044
self.inventories._access.flush()
1047
def add_revision(self, revision_id, rev, inv=None):
1048
"""Add rev to the revision store as revision_id.
1050
:param revision_id: the revision id to use.
1051
:param rev: The revision object.
1052
:param inv: The inventory for the revision. if None, it will be looked
1053
up in the inventory storer
1055
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1057
_mod_revision.check_not_reserved_id(revision_id)
1058
# check inventory present
1059
if not self.inventories.get_parent_map([(revision_id,)]):
1061
raise errors.WeaveRevisionNotPresent(revision_id,
1064
# yes, this is not suitable for adding with ghosts.
1065
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1068
key = (revision_id,)
1069
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1070
self._add_revision(rev)
1072
def _add_revision(self, revision):
1073
text = self._serializer.write_revision_to_string(revision)
1074
key = (revision.revision_id,)
1075
parents = tuple((parent,) for parent in revision.parent_ids)
1076
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1078
def _check_inventories(self, checker):
1079
"""Check the inventories found from the revision scan.
1081
This is responsible for verifying the sha1 of inventories and
1082
creating a pending_keys set that covers data referenced by inventories.
1084
bar = ui.ui_factory.nested_progress_bar()
1086
self._do_check_inventories(checker, bar)
1090
def _do_check_inventories(self, checker, bar):
1091
"""Helper for _check_inventories."""
1093
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1094
kinds = ['chk_bytes', 'texts']
1095
count = len(checker.pending_keys)
1096
bar.update(gettext("inventories"), 0, 2)
1097
current_keys = checker.pending_keys
1098
checker.pending_keys = {}
1099
# Accumulate current checks.
1100
for key in current_keys:
1101
if key[0] != 'inventories' and key[0] not in kinds:
1102
checker._report_items.append('unknown key type %r' % (key,))
1103
keys[key[0]].add(key[1:])
1104
if keys['inventories']:
1105
# NB: output order *should* be roughly sorted - topo or
1106
# inverse topo depending on repository - either way decent
1107
# to just delta against. However, pre-CHK formats didn't
1108
# try to optimise inventory layout on disk. As such the
1109
# pre-CHK code path does not use inventory deltas.
1111
for record in self.inventories.check(keys=keys['inventories']):
1112
if record.storage_kind == 'absent':
1113
checker._report_items.append(
1114
'Missing inventory {%s}' % (record.key,))
1116
last_object = self._check_record('inventories', record,
1117
checker, last_object,
1118
current_keys[('inventories',) + record.key])
1119
del keys['inventories']
1122
bar.update(gettext("texts"), 1)
1123
while (checker.pending_keys or keys['chk_bytes']
1125
# Something to check.
1126
current_keys = checker.pending_keys
1127
checker.pending_keys = {}
1128
# Accumulate current checks.
1129
for key in current_keys:
1130
if key[0] not in kinds:
1131
checker._report_items.append('unknown key type %r' % (key,))
1132
keys[key[0]].add(key[1:])
1133
# Check the outermost kind only - inventories || chk_bytes || texts
1137
for record in getattr(self, kind).check(keys=keys[kind]):
1138
if record.storage_kind == 'absent':
1139
checker._report_items.append(
1140
'Missing %s {%s}' % (kind, record.key,))
1142
last_object = self._check_record(kind, record,
1143
checker, last_object, current_keys[(kind,) + record.key])
1147
def _check_record(self, kind, record, checker, last_object, item_data):
1148
"""Check a single text from this repository."""
1149
if kind == 'inventories':
1150
rev_id = record.key[0]
1151
inv = self._deserialise_inventory(rev_id,
1152
record.get_bytes_as('fulltext'))
1153
if last_object is not None:
1154
delta = inv._make_delta(last_object)
1155
for old_path, path, file_id, ie in delta:
1158
ie.check(checker, rev_id, inv)
1160
for path, ie in inv.iter_entries():
1161
ie.check(checker, rev_id, inv)
1162
if self._format.fast_deltas:
1164
elif kind == 'chk_bytes':
1165
# No code written to check chk_bytes for this repo format.
1166
checker._report_items.append(
1167
'unsupported key type chk_bytes for %s' % (record.key,))
1168
elif kind == 'texts':
1169
self._check_text(record, checker, item_data)
1171
checker._report_items.append(
1172
'unknown key type %s for %s' % (kind, record.key))
1174
def _check_text(self, record, checker, item_data):
1175
"""Check a single text."""
1176
# Check it is extractable.
1177
# TODO: check length.
1178
if record.storage_kind == 'chunked':
1179
chunks = record.get_bytes_as(record.storage_kind)
1180
sha1 = osutils.sha_strings(chunks)
1181
length = sum(map(len, chunks))
1183
content = record.get_bytes_as('fulltext')
1184
sha1 = osutils.sha_string(content)
1185
length = len(content)
1186
if item_data and sha1 != item_data[1]:
1187
checker._report_items.append(
1188
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1189
(record.key, sha1, item_data[1], item_data[2]))
1192
def _eliminate_revisions_not_present(self, revision_ids):
1193
"""Check every revision id in revision_ids to see if we have it.
1195
Returns a set of the present revisions.
1198
graph = self.get_graph()
1199
parent_map = graph.get_parent_map(revision_ids)
1200
# The old API returned a list, should this actually be a set?
1201
return list(parent_map)
1203
def __init__(self, _format, a_bzrdir, control_files):
1204
"""Instantiate a VersionedFileRepository.
1206
:param _format: The format of the repository on disk.
1207
:param controldir: The ControlDir of the repository.
1208
:param control_files: Control files to use for locking, etc.
1210
# In the future we will have a single api for all stores for
1211
# getting file texts, inventories and revisions, then
1212
# this construct will accept instances of those things.
1213
super(VersionedFileRepository, self).__init__(_format, a_bzrdir,
1215
self._transport = control_files._transport
1216
self.base = self._transport.base
1218
self._reconcile_does_inventory_gc = True
1219
self._reconcile_fixes_text_parents = False
1220
self._reconcile_backsup_inventory = True
1221
# An InventoryEntry cache, used during deserialization
1222
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1223
# Is it safe to return inventory entries directly from the entry cache,
1224
# rather copying them?
1225
self._safe_to_return_from_cache = False
1227
def fetch(self, source, revision_id=None, find_ghosts=False,
1229
"""Fetch the content required to construct revision_id from source.
1231
If revision_id is None and fetch_spec is None, then all content is
1234
fetch() may not be used when the repository is in a write group -
1235
either finish the current write group before using fetch, or use
1236
fetch before starting the write group.
1238
:param find_ghosts: Find and copy revisions in the source that are
1239
ghosts in the target (and not reachable directly by walking out to
1240
the first-present revision in target from revision_id).
1241
:param revision_id: If specified, all the content needed for this
1242
revision ID will be copied to the target. Fetch will determine for
1243
itself which content needs to be copied.
1244
:param fetch_spec: If specified, a SearchResult or
1245
PendingAncestryResult that describes which revisions to copy. This
1246
allows copying multiple heads at once. Mutually exclusive with
1249
if fetch_spec is not None and revision_id is not None:
1250
raise AssertionError(
1251
"fetch_spec and revision_id are mutually exclusive.")
1252
if self.is_in_write_group():
1253
raise errors.InternalBzrError(
1254
"May not fetch while in a write group.")
1255
# fast path same-url fetch operations
1256
# TODO: lift out to somewhere common with RemoteRepository
1257
# <https://bugs.launchpad.net/bzr/+bug/401646>
1258
if (self.has_same_location(source)
1259
and fetch_spec is None
1260
and self._has_same_fallbacks(source)):
1261
# check that last_revision is in 'from' and then return a
1263
if (revision_id is not None and
1264
not _mod_revision.is_null(revision_id)):
1265
self.get_revision(revision_id)
1267
inter = InterRepository.get(source, self)
1268
if (fetch_spec is not None and
1269
not getattr(inter, "supports_fetch_spec", False)):
1270
raise errors.UnsupportedOperation(
1271
"fetch_spec not supported for %r" % inter)
1272
return inter.fetch(revision_id=revision_id,
1273
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1276
def gather_stats(self, revid=None, committers=None):
1277
"""See Repository.gather_stats()."""
1278
result = super(VersionedFileRepository, self).gather_stats(revid, committers)
1279
# now gather global repository information
1280
# XXX: This is available for many repos regardless of listability.
1281
if self.user_transport.listable():
1282
# XXX: do we want to __define len__() ?
1283
# Maybe the versionedfiles object should provide a different
1284
# method to get the number of keys.
1285
result['revisions'] = len(self.revisions.keys())
1286
# result['size'] = t
1289
def get_commit_builder(self, branch, parents, config_stack, timestamp=None,
1290
timezone=None, committer=None, revprops=None,
1291
revision_id=None, lossy=False):
1292
"""Obtain a CommitBuilder for this repository.
1294
:param branch: Branch to commit to.
1295
:param parents: Revision ids of the parents of the new revision.
1296
:param config_stack: Configuration stack to use.
1297
:param timestamp: Optional timestamp recorded for commit.
1298
:param timezone: Optional timezone for timestamp.
1299
:param committer: Optional committer to set for commit.
1300
:param revprops: Optional dictionary of revision properties.
1301
:param revision_id: Optional revision id.
1302
:param lossy: Whether to discard data that can not be natively
1303
represented, when pushing to a foreign VCS
1305
if self._fallback_repositories and not self._format.supports_chks:
1306
raise errors.BzrError("Cannot commit directly to a stacked branch"
1307
" in pre-2a formats. See "
1308
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1309
result = self._commit_builder_class(self, parents, config_stack,
1310
timestamp, timezone, committer, revprops, revision_id,
1312
self.start_write_group()
1315
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1316
"""Return the keys of missing inventory parents for revisions added in
1319
A revision is not complete if the inventory delta for that revision
1320
cannot be calculated. Therefore if the parent inventories of a
1321
revision are not present, the revision is incomplete, and e.g. cannot
1322
be streamed by a smart server. This method finds missing inventory
1323
parents for revisions added in this write group.
1325
if not self._format.supports_external_lookups:
1326
# This is only an issue for stacked repositories
1328
if not self.is_in_write_group():
1329
raise AssertionError('not in a write group')
1331
# XXX: We assume that every added revision already has its
1332
# corresponding inventory, so we only check for parent inventories that
1333
# might be missing, rather than all inventories.
1334
parents = set(self.revisions._index.get_missing_parents())
1335
parents.discard(_mod_revision.NULL_REVISION)
1336
unstacked_inventories = self.inventories._index
1337
present_inventories = unstacked_inventories.get_parent_map(
1338
key[-1:] for key in parents)
1339
parents.difference_update(present_inventories)
1340
if len(parents) == 0:
1341
# No missing parent inventories.
1343
if not check_for_missing_texts:
1344
return set(('inventories', rev_id) for (rev_id,) in parents)
1345
# Ok, now we have a list of missing inventories. But these only matter
1346
# if the inventories that reference them are missing some texts they
1347
# appear to introduce.
1348
# XXX: Texts referenced by all added inventories need to be present,
1349
# but at the moment we're only checking for texts referenced by
1350
# inventories at the graph's edge.
1351
key_deps = self.revisions._index._key_dependencies
1352
key_deps.satisfy_refs_for_keys(present_inventories)
1353
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1354
file_ids = self.fileids_altered_by_revision_ids(referrers)
1355
missing_texts = set()
1356
for file_id, version_ids in viewitems(file_ids):
1357
missing_texts.update(
1358
(file_id, version_id) for version_id in version_ids)
1359
present_texts = self.texts.get_parent_map(missing_texts)
1360
missing_texts.difference_update(present_texts)
1361
if not missing_texts:
1362
# No texts are missing, so all revisions and their deltas are
1365
# Alternatively the text versions could be returned as the missing
1366
# keys, but this is likely to be less data.
1367
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1371
def has_revisions(self, revision_ids):
1372
"""Probe to find out the presence of multiple revisions.
1374
:param revision_ids: An iterable of revision_ids.
1375
:return: A set of the revision_ids that were present.
1377
parent_map = self.revisions.get_parent_map(
1378
[(rev_id,) for rev_id in revision_ids])
1380
if _mod_revision.NULL_REVISION in revision_ids:
1381
result.add(_mod_revision.NULL_REVISION)
1382
result.update([key[0] for key in parent_map])
1386
def get_revision_reconcile(self, revision_id):
1387
"""'reconcile' helper routine that allows access to a revision always.
1389
This variant of get_revision does not cross check the weave graph
1390
against the revision one as get_revision does: but it should only
1391
be used by reconcile, or reconcile-alike commands that are correcting
1392
or testing the revision graph.
1394
return self._get_revisions([revision_id])[0]
1397
def get_revisions(self, revision_ids):
1398
"""Get many revisions at once.
1400
Repositories that need to check data on every revision read should
1401
subclass this method.
1403
return self._get_revisions(revision_ids)
1406
def _get_revisions(self, revision_ids):
1407
"""Core work logic to get many revisions without sanity checks."""
1409
for revid, rev in self._iter_revisions(revision_ids):
1411
raise errors.NoSuchRevision(self, revid)
1413
return [revs[revid] for revid in revision_ids]
1415
def _iter_revisions(self, revision_ids):
1416
"""Iterate over revision objects.
1418
:param revision_ids: An iterable of revisions to examine. None may be
1419
passed to request all revisions known to the repository. Note that
1420
not all repositories can find unreferenced revisions; for those
1421
repositories only referenced ones will be returned.
1422
:return: An iterator of (revid, revision) tuples. Absent revisions (
1423
those asked for but not available) are returned as (revid, None).
1425
if revision_ids is None:
1426
revision_ids = self.all_revision_ids()
1428
for rev_id in revision_ids:
1429
if not rev_id or not isinstance(rev_id, basestring):
1430
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1431
keys = [(key,) for key in revision_ids]
1432
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1433
for record in stream:
1434
revid = record.key[0]
1435
if record.storage_kind == 'absent':
1438
text = record.get_bytes_as('fulltext')
1439
rev = self._serializer.read_revision_from_string(text)
1443
def add_signature_text(self, revision_id, signature):
1444
"""Store a signature text for a revision.
1446
:param revision_id: Revision id of the revision
1447
:param signature: Signature text.
1449
self.signatures.add_lines((revision_id,), (),
1450
osutils.split_lines(signature))
1452
def find_text_key_references(self):
1453
"""Find the text key references within the repository.
1455
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1456
to whether they were referred to by the inventory of the
1457
revision_id that they contain. The inventory texts from all present
1458
revision ids are assessed to generate this report.
1460
revision_keys = self.revisions.keys()
1461
w = self.inventories
1462
pb = ui.ui_factory.nested_progress_bar()
1464
return self._serializer._find_text_key_references(
1465
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1469
def _inventory_xml_lines_for_keys(self, keys):
1470
"""Get a line iterator of the sort needed for findind references.
1472
Not relevant for non-xml inventory repositories.
1474
Ghosts in revision_keys are ignored.
1476
:param revision_keys: The revision keys for the inventories to inspect.
1477
:return: An iterator over (inventory line, revid) for the fulltexts of
1478
all of the xml inventories specified by revision_keys.
1480
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1481
for record in stream:
1482
if record.storage_kind != 'absent':
1483
chunks = record.get_bytes_as('chunked')
1484
revid = record.key[-1]
1485
lines = osutils.chunks_to_lines(chunks)
1489
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1491
"""Helper routine for fileids_altered_by_revision_ids.
1493
This performs the translation of xml lines to revision ids.
1495
:param line_iterator: An iterator of lines, origin_version_id
1496
:param revision_keys: The revision ids to filter for. This should be a
1497
set or other type which supports efficient __contains__ lookups, as
1498
the revision key from each parsed line will be looked up in the
1499
revision_keys filter.
1500
:return: a dictionary mapping altered file-ids to an iterable of
1501
revision_ids. Each altered file-ids has the exact revision_ids that
1502
altered it listed explicitly.
1504
seen = set(self._serializer._find_text_key_references(line_iterator))
1505
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
1506
parent_seen = set(self._serializer._find_text_key_references(
1507
self._inventory_xml_lines_for_keys(parent_keys)))
1508
new_keys = seen - parent_seen
1510
setdefault = result.setdefault
1511
for key in new_keys:
1512
setdefault(key[0], set()).add(key[-1])
1515
def _find_parent_keys_of_revisions(self, revision_keys):
1516
"""Similar to _find_parent_ids_of_revisions, but used with keys.
1518
:param revision_keys: An iterable of revision_keys.
1519
:return: The parents of all revision_keys that are not already in
1522
parent_map = self.revisions.get_parent_map(revision_keys)
1523
parent_keys = set(itertools.chain.from_iterable(
1524
viewvalues(parent_map)))
1525
parent_keys.difference_update(revision_keys)
1526
parent_keys.discard(_mod_revision.NULL_REVISION)
1529
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
1530
"""Find the file ids and versions affected by revisions.
1532
:param revisions: an iterable containing revision ids.
1533
:param _inv_weave: The inventory weave from this repository or None.
1534
If None, the inventory weave will be opened automatically.
1535
:return: a dictionary mapping altered file-ids to an iterable of
1536
revision_ids. Each altered file-ids has the exact revision_ids that
1537
altered it listed explicitly.
1539
selected_keys = set((revid,) for revid in revision_ids)
1540
w = _inv_weave or self.inventories
1541
return self._find_file_ids_from_xml_inventory_lines(
1542
w.iter_lines_added_or_present_in_keys(
1543
selected_keys, pb=None),
1546
def iter_files_bytes(self, desired_files):
1547
"""Iterate through file versions.
1549
Files will not necessarily be returned in the order they occur in
1550
desired_files. No specific order is guaranteed.
1552
Yields pairs of identifier, bytes_iterator. identifier is an opaque
1553
value supplied by the caller as part of desired_files. It should
1554
uniquely identify the file version in the caller's context. (Examples:
1555
an index number or a TreeTransform trans_id.)
1557
bytes_iterator is an iterable of bytestrings for the file. The
1558
kind of iterable and length of the bytestrings are unspecified, but for
1559
this implementation, it is a list of bytes produced by
1560
VersionedFile.get_record_stream().
1562
:param desired_files: a list of (file_id, revision_id, identifier)
1566
for file_id, revision_id, callable_data in desired_files:
1567
text_keys[(file_id, revision_id)] = callable_data
1568
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
1569
if record.storage_kind == 'absent':
1570
raise errors.RevisionNotPresent(record.key[1], record.key[0])
1571
yield text_keys[record.key], record.get_bytes_as('chunked')
1573
def _generate_text_key_index(self, text_key_references=None,
1575
"""Generate a new text key index for the repository.
1577
This is an expensive function that will take considerable time to run.
1579
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
1580
list of parents, also text keys. When a given key has no parents,
1581
the parents list will be [NULL_REVISION].
1583
# All revisions, to find inventory parents.
1584
if ancestors is None:
1585
graph = self.get_graph()
1586
ancestors = graph.get_parent_map(self.all_revision_ids())
1587
if text_key_references is None:
1588
text_key_references = self.find_text_key_references()
1589
pb = ui.ui_factory.nested_progress_bar()
1591
return self._do_generate_text_key_index(ancestors,
1592
text_key_references, pb)
1596
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
1597
"""Helper for _generate_text_key_index to avoid deep nesting."""
1598
revision_order = tsort.topo_sort(ancestors)
1599
invalid_keys = set()
1601
for revision_id in revision_order:
1602
revision_keys[revision_id] = set()
1603
text_count = len(text_key_references)
1604
# a cache of the text keys to allow reuse; costs a dict of all the
1605
# keys, but saves a 2-tuple for every child of a given key.
1607
for text_key, valid in viewitems(text_key_references):
1609
invalid_keys.add(text_key)
1611
revision_keys[text_key[1]].add(text_key)
1612
text_key_cache[text_key] = text_key
1613
del text_key_references
1615
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
1616
NULL_REVISION = _mod_revision.NULL_REVISION
1617
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
1618
# too small for large or very branchy trees. However, for 55K path
1619
# trees, it would be easy to use too much memory trivially. Ideally we
1620
# could gauge this by looking at available real memory etc, but this is
1621
# always a tricky proposition.
1622
inventory_cache = lru_cache.LRUCache(10)
1623
batch_size = 10 # should be ~150MB on a 55K path tree
1624
batch_count = len(revision_order) / batch_size + 1
1626
pb.update(gettext("Calculating text parents"), processed_texts, text_count)
1627
for offset in range(batch_count):
1628
to_query = revision_order[offset * batch_size:(offset + 1) *
1632
for revision_id in to_query:
1633
parent_ids = ancestors[revision_id]
1634
for text_key in revision_keys[revision_id]:
1635
pb.update(gettext("Calculating text parents"), processed_texts)
1636
processed_texts += 1
1637
candidate_parents = []
1638
for parent_id in parent_ids:
1639
parent_text_key = (text_key[0], parent_id)
1641
check_parent = parent_text_key not in \
1642
revision_keys[parent_id]
1644
# the parent parent_id is a ghost:
1645
check_parent = False
1646
# truncate the derived graph against this ghost.
1647
parent_text_key = None
1649
# look at the parent commit details inventories to
1650
# determine possible candidates in the per file graph.
1653
inv = inventory_cache[parent_id]
1655
inv = self.revision_tree(parent_id).root_inventory
1656
inventory_cache[parent_id] = inv
1658
parent_entry = inv[text_key[0]]
1659
except (KeyError, errors.NoSuchId):
1661
if parent_entry is not None:
1663
text_key[0], parent_entry.revision)
1665
parent_text_key = None
1666
if parent_text_key is not None:
1667
candidate_parents.append(
1668
text_key_cache[parent_text_key])
1669
parent_heads = text_graph.heads(candidate_parents)
1670
new_parents = list(parent_heads)
1671
new_parents.sort(key=lambda x:candidate_parents.index(x))
1672
if new_parents == []:
1673
new_parents = [NULL_REVISION]
1674
text_index[text_key] = new_parents
1676
for text_key in invalid_keys:
1677
text_index[text_key] = [NULL_REVISION]
1680
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
1681
"""Get an iterable listing the keys of all the data introduced by a set
1684
The keys will be ordered so that the corresponding items can be safely
1685
fetched and inserted in that order.
1687
:returns: An iterable producing tuples of (knit-kind, file-id,
1688
versions). knit-kind is one of 'file', 'inventory', 'signatures',
1689
'revisions'. file-id is None unless knit-kind is 'file'.
1691
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
1694
for result in self._find_non_file_keys_to_fetch(revision_ids):
1697
def _find_file_keys_to_fetch(self, revision_ids, pb):
1698
# XXX: it's a bit weird to control the inventory weave caching in this
1699
# generator. Ideally the caching would be done in fetch.py I think. Or
1700
# maybe this generator should explicitly have the contract that it
1701
# should not be iterated until the previously yielded item has been
1703
inv_w = self.inventories
1705
# file ids that changed
1706
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
1708
num_file_ids = len(file_ids)
1709
for file_id, altered_versions in viewitems(file_ids):
1711
pb.update(gettext("Fetch texts"), count, num_file_ids)
1713
yield ("file", file_id, altered_versions)
1715
def _find_non_file_keys_to_fetch(self, revision_ids):
1717
yield ("inventory", None, revision_ids)
1720
# XXX: Note ATM no callers actually pay attention to this return
1721
# instead they just use the list of revision ids and ignore
1722
# missing sigs. Consider removing this work entirely
1723
revisions_with_signatures = set(self.signatures.get_parent_map(
1724
[(r,) for r in revision_ids]))
1725
revisions_with_signatures = {r for (r,) in revisions_with_signatures}
1726
revisions_with_signatures.intersection_update(revision_ids)
1727
yield ("signatures", None, revisions_with_signatures)
1730
yield ("revisions", None, revision_ids)
1733
def get_inventory(self, revision_id):
1734
"""Get Inventory object by revision id."""
1735
return next(self.iter_inventories([revision_id]))
1737
def iter_inventories(self, revision_ids, ordering=None):
1738
"""Get many inventories by revision_ids.
1740
This will buffer some or all of the texts used in constructing the
1741
inventories in memory, but will only parse a single inventory at a
1744
:param revision_ids: The expected revision ids of the inventories.
1745
:param ordering: optional ordering, e.g. 'topological'. If not
1746
specified, the order of revision_ids will be preserved (by
1747
buffering if necessary).
1748
:return: An iterator of inventories.
1750
if ((None in revision_ids)
1751
or (_mod_revision.NULL_REVISION in revision_ids)):
1752
raise ValueError('cannot get null revision inventory')
1753
for inv, revid in self._iter_inventories(revision_ids, ordering):
1755
raise errors.NoSuchRevision(self, revid)
1758
def _iter_inventories(self, revision_ids, ordering):
1759
"""single-document based inventory iteration."""
1760
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
1761
for text, revision_id in inv_xmls:
1763
yield None, revision_id
1765
yield self._deserialise_inventory(revision_id, text), revision_id
1767
def _iter_inventory_xmls(self, revision_ids, ordering):
1768
if ordering is None:
1769
order_as_requested = True
1770
ordering = 'unordered'
1772
order_as_requested = False
1773
keys = [(revision_id,) for revision_id in revision_ids]
1776
if order_as_requested:
1777
key_iter = iter(keys)
1778
next_key = next(key_iter)
1779
stream = self.inventories.get_record_stream(keys, ordering, True)
1781
for record in stream:
1782
if record.storage_kind != 'absent':
1783
chunks = record.get_bytes_as('chunked')
1784
if order_as_requested:
1785
text_chunks[record.key] = chunks
1787
yield ''.join(chunks), record.key[-1]
1789
yield None, record.key[-1]
1790
if order_as_requested:
1791
# Yield as many results as we can while preserving order.
1792
while next_key in text_chunks:
1793
chunks = text_chunks.pop(next_key)
1794
yield ''.join(chunks), next_key[-1]
1796
next_key = next(key_iter)
1797
except StopIteration:
1798
# We still want to fully consume the get_record_stream,
1799
# just in case it is not actually finished at this point
1803
def _deserialise_inventory(self, revision_id, xml):
1804
"""Transform the xml into an inventory object.
1806
:param revision_id: The expected revision id of the inventory.
1807
:param xml: A serialised inventory.
1809
result = self._serializer.read_inventory_from_string(xml, revision_id,
1810
entry_cache=self._inventory_entry_cache,
1811
return_from_cache=self._safe_to_return_from_cache)
1812
if result.revision_id != revision_id:
1813
raise AssertionError('revision id mismatch %s != %s' % (
1814
result.revision_id, revision_id))
1817
def get_serializer_format(self):
1818
return self._serializer.format_num
1821
def _get_inventory_xml(self, revision_id):
1822
"""Get serialized inventory as a string."""
1823
texts = self._iter_inventory_xmls([revision_id], 'unordered')
1824
text, revision_id = next(texts)
1826
raise errors.NoSuchRevision(self, revision_id)
1830
def revision_tree(self, revision_id):
1831
"""Return Tree for a revision on this branch.
1833
`revision_id` may be NULL_REVISION for the empty tree revision.
1835
revision_id = _mod_revision.ensure_null(revision_id)
1836
# TODO: refactor this to use an existing revision object
1837
# so we don't need to read it in twice.
1838
if revision_id == _mod_revision.NULL_REVISION:
1839
return InventoryRevisionTree(self,
1840
Inventory(root_id=None), _mod_revision.NULL_REVISION)
1842
inv = self.get_inventory(revision_id)
1843
return InventoryRevisionTree(self, inv, revision_id)
1845
def revision_trees(self, revision_ids):
1846
"""Return Trees for revisions in this repository.
1848
:param revision_ids: a sequence of revision-ids;
1849
a revision-id may not be None or 'null:'
1851
inventories = self.iter_inventories(revision_ids)
1852
for inv in inventories:
1853
yield InventoryRevisionTree(self, inv, inv.revision_id)
1855
def _filtered_revision_trees(self, revision_ids, file_ids):
1856
"""Return Tree for a revision on this branch with only some files.
1858
:param revision_ids: a sequence of revision-ids;
1859
a revision-id may not be None or 'null:'
1860
:param file_ids: if not None, the result is filtered
1861
so that only those file-ids, their parents and their
1862
children are included.
1864
inventories = self.iter_inventories(revision_ids)
1865
for inv in inventories:
1866
# Should we introduce a FilteredRevisionTree class rather
1867
# than pre-filter the inventory here?
1868
filtered_inv = inv.filter(file_ids)
1869
yield InventoryRevisionTree(self, filtered_inv, filtered_inv.revision_id)
1871
def get_parent_map(self, revision_ids):
1872
"""See graph.StackedParentsProvider.get_parent_map"""
1873
# revisions index works in keys; this just works in revisions
1874
# therefore wrap and unwrap
1877
for revision_id in revision_ids:
1878
if revision_id == _mod_revision.NULL_REVISION:
1879
result[revision_id] = ()
1880
elif revision_id is None:
1881
raise ValueError('get_parent_map(None) is not valid')
1883
query_keys.append((revision_id ,))
1884
for (revision_id,), parent_keys in viewitems(
1885
self.revisions.get_parent_map(query_keys)):
1887
result[revision_id] = tuple([parent_revid
1888
for (parent_revid,) in parent_keys])
1890
result[revision_id] = (_mod_revision.NULL_REVISION,)
1894
def get_known_graph_ancestry(self, revision_ids):
1895
"""Return the known graph for a set of revision ids and their ancestors.
1897
st = static_tuple.StaticTuple
1898
revision_keys = [st(r_id).intern() for r_id in revision_ids]
1899
known_graph = self.revisions.get_known_graph_ancestry(revision_keys)
1900
return graph.GraphThunkIdsToKeys(known_graph)
1903
def get_file_graph(self):
1904
"""Return the graph walker for text revisions."""
1905
return graph.Graph(self.texts)
1907
def revision_ids_to_search_result(self, result_set):
1908
"""Convert a set of revision ids to a graph SearchResult."""
1909
result_parents = set(itertools.chain.from_iterable(viewvalues(
1910
self.get_graph().get_parent_map(result_set))))
1911
included_keys = result_set.intersection(result_parents)
1912
start_keys = result_set.difference(included_keys)
1913
exclude_keys = result_parents.difference(result_set)
1914
result = vf_search.SearchResult(start_keys, exclude_keys,
1915
len(result_set), result_set)
1918
def _get_versioned_file_checker(self, text_key_references=None,
1920
"""Return an object suitable for checking versioned files.
1922
:param text_key_references: if non-None, an already built
1923
dictionary mapping text keys ((fileid, revision_id) tuples)
1924
to whether they were referred to by the inventory of the
1925
revision_id that they contain. If None, this will be
1927
:param ancestors: Optional result from
1928
self.get_graph().get_parent_map(self.all_revision_ids()) if already
1931
return _VersionedFileChecker(self,
1932
text_key_references=text_key_references, ancestors=ancestors)
1935
def has_signature_for_revision_id(self, revision_id):
1936
"""Query for a revision signature for revision_id in the repository."""
1937
if not self.has_revision(revision_id):
1938
raise errors.NoSuchRevision(self, revision_id)
1939
sig_present = (1 == len(
1940
self.signatures.get_parent_map([(revision_id,)])))
1944
def get_signature_text(self, revision_id):
1945
"""Return the text for a signature."""
1946
stream = self.signatures.get_record_stream([(revision_id,)],
1948
record = next(stream)
1949
if record.storage_kind == 'absent':
1950
raise errors.NoSuchRevision(self, revision_id)
1951
return record.get_bytes_as('fulltext')
1954
def _check(self, revision_ids, callback_refs, check_repo):
1955
result = check.VersionedFileCheck(self, check_repo=check_repo)
1956
result.check(callback_refs)
1959
def _find_inconsistent_revision_parents(self, revisions_iterator=None):
1960
"""Find revisions with different parent lists in the revision object
1961
and in the index graph.
1963
:param revisions_iterator: None, or an iterator of (revid,
1964
Revision-or-None). This iterator controls the revisions checked.
1965
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1966
parents-in-revision).
1968
if not self.is_locked():
1969
raise AssertionError()
1971
if revisions_iterator is None:
1972
revisions_iterator = self._iter_revisions(None)
1973
for revid, revision in revisions_iterator:
1974
if revision is None:
1976
parent_map = vf.get_parent_map([(revid,)])
1977
parents_according_to_index = tuple(parent[-1] for parent in
1978
parent_map[(revid,)])
1979
parents_according_to_revision = tuple(revision.parent_ids)
1980
if parents_according_to_index != parents_according_to_revision:
1981
yield (revid, parents_according_to_index,
1982
parents_according_to_revision)
1984
def _check_for_inconsistent_revision_parents(self):
1985
inconsistencies = list(self._find_inconsistent_revision_parents())
1987
raise errors.BzrCheckError(
1988
"Revision knit has inconsistent parents.")
1990
def _get_sink(self):
1991
"""Return a sink for streaming into this repository."""
1992
return StreamSink(self)
1994
def _get_source(self, to_format):
1995
"""Return a source for streaming from this repository."""
1996
return StreamSource(self, to_format)
1999
class MetaDirVersionedFileRepository(MetaDirRepository,
2000
VersionedFileRepository):
2001
"""Repositories in a meta-dir, that work via versioned file objects."""
2003
def __init__(self, _format, a_bzrdir, control_files):
2004
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2008
class MetaDirVersionedFileRepositoryFormat(RepositoryFormatMetaDir,
2009
VersionedFileRepositoryFormat):
2010
"""Base class for repository formats using versioned files in metadirs."""
2013
class StreamSink(object):
2014
"""An object that can insert a stream into a repository.
2016
This interface handles the complexity of reserialising inventories and
2017
revisions from different formats, and allows unidirectional insertion into
2018
stacked repositories without looking for the missing basis parents
2022
def __init__(self, target_repo):
2023
self.target_repo = target_repo
2025
def insert_stream(self, stream, src_format, resume_tokens):
2026
"""Insert a stream's content into the target repository.
2028
:param src_format: a bzr repository format.
2030
:return: a list of resume tokens and an iterable of keys additional
2031
items required before the insertion can be completed.
2033
self.target_repo.lock_write()
2036
self.target_repo.resume_write_group(resume_tokens)
2039
self.target_repo.start_write_group()
2042
# locked_insert_stream performs a commit|suspend.
2043
missing_keys = self.insert_stream_without_locking(stream,
2044
src_format, is_resume)
2046
# suspend the write group and tell the caller what we is
2047
# missing. We know we can suspend or else we would not have
2048
# entered this code path. (All repositories that can handle
2049
# missing keys can handle suspending a write group).
2050
write_group_tokens = self.target_repo.suspend_write_group()
2051
return write_group_tokens, missing_keys
2052
hint = self.target_repo.commit_write_group()
2053
to_serializer = self.target_repo._format._serializer
2054
src_serializer = src_format._serializer
2055
if (to_serializer != src_serializer and
2056
self.target_repo._format.pack_compresses):
2057
self.target_repo.pack(hint=hint)
2060
self.target_repo.abort_write_group(suppress_errors=True)
2063
self.target_repo.unlock()
2065
def insert_stream_without_locking(self, stream, src_format,
2067
"""Insert a stream's content into the target repository.
2069
This assumes that you already have a locked repository and an active
2072
:param src_format: a bzr repository format.
2073
:param is_resume: Passed down to get_missing_parent_inventories to
2074
indicate if we should be checking for missing texts at the same
2077
:return: A set of keys that are missing.
2079
if not self.target_repo.is_write_locked():
2080
raise errors.ObjectNotLocked(self)
2081
if not self.target_repo.is_in_write_group():
2082
raise errors.BzrError('you must already be in a write group')
2083
to_serializer = self.target_repo._format._serializer
2084
src_serializer = src_format._serializer
2086
if to_serializer == src_serializer:
2087
# If serializers match and the target is a pack repository, set the
2088
# write cache size on the new pack. This avoids poor performance
2089
# on transports where append is unbuffered (such as
2090
# RemoteTransport). This is safe to do because nothing should read
2091
# back from the target repository while a stream with matching
2092
# serialization is being inserted.
2093
# The exception is that a delta record from the source that should
2094
# be a fulltext may need to be expanded by the target (see
2095
# test_fetch_revisions_with_deltas_into_pack); but we take care to
2096
# explicitly flush any buffered writes first in that rare case.
2098
new_pack = self.target_repo._pack_collection._new_pack
2099
except AttributeError:
2100
# Not a pack repository
2103
new_pack.set_write_cache_size(1024*1024)
2104
for substream_type, substream in stream:
2105
if 'stream' in debug.debug_flags:
2106
mutter('inserting substream: %s', substream_type)
2107
if substream_type == 'texts':
2108
self.target_repo.texts.insert_record_stream(substream)
2109
elif substream_type == 'inventories':
2110
if src_serializer == to_serializer:
2111
self.target_repo.inventories.insert_record_stream(
2114
self._extract_and_insert_inventories(
2115
substream, src_serializer)
2116
elif substream_type == 'inventory-deltas':
2117
self._extract_and_insert_inventory_deltas(
2118
substream, src_serializer)
2119
elif substream_type == 'chk_bytes':
2120
# XXX: This doesn't support conversions, as it assumes the
2121
# conversion was done in the fetch code.
2122
self.target_repo.chk_bytes.insert_record_stream(substream)
2123
elif substream_type == 'revisions':
2124
# This may fallback to extract-and-insert more often than
2125
# required if the serializers are different only in terms of
2127
if src_serializer == to_serializer:
2128
self.target_repo.revisions.insert_record_stream(substream)
2130
self._extract_and_insert_revisions(substream,
2132
elif substream_type == 'signatures':
2133
self.target_repo.signatures.insert_record_stream(substream)
2135
raise AssertionError('kaboom! %s' % (substream_type,))
2136
# Done inserting data, and the missing_keys calculations will try to
2137
# read back from the inserted data, so flush the writes to the new pack
2138
# (if this is pack format).
2139
if new_pack is not None:
2140
new_pack._write_data('', flush=True)
2141
# Find all the new revisions (including ones from resume_tokens)
2142
missing_keys = self.target_repo.get_missing_parent_inventories(
2143
check_for_missing_texts=is_resume)
2145
for prefix, versioned_file in (
2146
('texts', self.target_repo.texts),
2147
('inventories', self.target_repo.inventories),
2148
('revisions', self.target_repo.revisions),
2149
('signatures', self.target_repo.signatures),
2150
('chk_bytes', self.target_repo.chk_bytes),
2152
if versioned_file is None:
2154
# TODO: key is often going to be a StaticTuple object
2155
# I don't believe we can define a method by which
2156
# (prefix,) + StaticTuple will work, though we could
2157
# define a StaticTuple.sq_concat that would allow you to
2158
# pass in either a tuple or a StaticTuple as the second
2159
# object, so instead we could have:
2160
# StaticTuple(prefix) + key here...
2161
missing_keys.update((prefix,) + key for key in
2162
versioned_file.get_missing_compression_parent_keys())
2163
except NotImplementedError:
2164
# cannot even attempt suspending, and missing would have failed
2165
# during stream insertion.
2166
missing_keys = set()
2169
def _extract_and_insert_inventory_deltas(self, substream, serializer):
2170
target_rich_root = self.target_repo._format.rich_root_data
2171
target_tree_refs = self.target_repo._format.supports_tree_reference
2172
for record in substream:
2173
# Insert the delta directly
2174
inventory_delta_bytes = record.get_bytes_as('fulltext')
2175
deserialiser = inventory_delta.InventoryDeltaDeserializer()
2177
parse_result = deserialiser.parse_text_bytes(
2178
inventory_delta_bytes)
2179
except inventory_delta.IncompatibleInventoryDelta as err:
2180
mutter("Incompatible delta: %s", err.msg)
2181
raise errors.IncompatibleRevision(self.target_repo._format)
2182
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
2183
revision_id = new_id
2184
parents = [key[0] for key in record.parents]
2185
self.target_repo.add_inventory_by_delta(
2186
basis_id, inv_delta, revision_id, parents)
2188
def _extract_and_insert_inventories(self, substream, serializer,
2190
"""Generate a new inventory versionedfile in target, converting data.
2192
The inventory is retrieved from the source, (deserializing it), and
2193
stored in the target (reserializing it in a different format).
2195
target_rich_root = self.target_repo._format.rich_root_data
2196
target_tree_refs = self.target_repo._format.supports_tree_reference
2197
for record in substream:
2198
# It's not a delta, so it must be a fulltext in the source
2199
# serializer's format.
2200
bytes = record.get_bytes_as('fulltext')
2201
revision_id = record.key[0]
2202
inv = serializer.read_inventory_from_string(bytes, revision_id)
2203
parents = [key[0] for key in record.parents]
2204
self.target_repo.add_inventory(revision_id, inv, parents)
2205
# No need to keep holding this full inv in memory when the rest of
2206
# the substream is likely to be all deltas.
2209
def _extract_and_insert_revisions(self, substream, serializer):
2210
for record in substream:
2211
bytes = record.get_bytes_as('fulltext')
2212
revision_id = record.key[0]
2213
rev = serializer.read_revision_from_string(bytes)
2214
if rev.revision_id != revision_id:
2215
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
2216
self.target_repo.add_revision(revision_id, rev)
2219
if self.target_repo._format._fetch_reconcile:
2220
self.target_repo.reconcile()
2223
class StreamSource(object):
2224
"""A source of a stream for fetching between repositories."""
2226
def __init__(self, from_repository, to_format):
2227
"""Create a StreamSource streaming from from_repository."""
2228
self.from_repository = from_repository
2229
self.to_format = to_format
2230
self._record_counter = RecordCounter()
2232
def delta_on_metadata(self):
2233
"""Return True if delta's are permitted on metadata streams.
2235
That is on revisions and signatures.
2237
src_serializer = self.from_repository._format._serializer
2238
target_serializer = self.to_format._serializer
2239
return (self.to_format._fetch_uses_deltas and
2240
src_serializer == target_serializer)
2242
def _fetch_revision_texts(self, revs):
2243
# fetch signatures first and then the revision texts
2244
# may need to be a InterRevisionStore call here.
2245
from_sf = self.from_repository.signatures
2246
# A missing signature is just skipped.
2247
keys = [(rev_id,) for rev_id in revs]
2248
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
2250
self.to_format._fetch_order,
2251
not self.to_format._fetch_uses_deltas))
2252
# If a revision has a delta, this is actually expanded inside the
2253
# insert_record_stream code now, which is an alternate fix for
2255
from_rf = self.from_repository.revisions
2256
revisions = from_rf.get_record_stream(
2258
self.to_format._fetch_order,
2259
not self.delta_on_metadata())
2260
return [('signatures', signatures), ('revisions', revisions)]
2262
def _generate_root_texts(self, revs):
2263
"""This will be called by get_stream between fetching weave texts and
2264
fetching the inventory weave.
2266
if self._rich_root_upgrade():
2267
return _mod_fetch.Inter1and2Helper(
2268
self.from_repository).generate_root_texts(revs)
2272
def get_stream(self, search):
2274
revs = search.get_keys()
2275
graph = self.from_repository.get_graph()
2276
revs = tsort.topo_sort(graph.get_parent_map(revs))
2277
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
2279
for knit_kind, file_id, revisions in data_to_fetch:
2280
if knit_kind != phase:
2282
# Make a new progress bar for this phase
2283
if knit_kind == "file":
2284
# Accumulate file texts
2285
text_keys.extend([(file_id, revision) for revision in
2287
elif knit_kind == "inventory":
2288
# Now copy the file texts.
2289
from_texts = self.from_repository.texts
2290
yield ('texts', from_texts.get_record_stream(
2291
text_keys, self.to_format._fetch_order,
2292
not self.to_format._fetch_uses_deltas))
2293
# Cause an error if a text occurs after we have done the
2296
# Before we process the inventory we generate the root
2297
# texts (if necessary) so that the inventories references
2299
for _ in self._generate_root_texts(revs):
2301
# we fetch only the referenced inventories because we do not
2302
# know for unselected inventories whether all their required
2303
# texts are present in the other repository - it could be
2305
for info in self._get_inventory_stream(revs):
2307
elif knit_kind == "signatures":
2308
# Nothing to do here; this will be taken care of when
2309
# _fetch_revision_texts happens.
2311
elif knit_kind == "revisions":
2312
for record in self._fetch_revision_texts(revs):
2315
raise AssertionError("Unknown knit kind %r" % knit_kind)
2317
def get_stream_for_missing_keys(self, missing_keys):
2318
# missing keys can only occur when we are byte copying and not
2319
# translating (because translation means we don't send
2320
# unreconstructable deltas ever).
2322
keys['texts'] = set()
2323
keys['revisions'] = set()
2324
keys['inventories'] = set()
2325
keys['chk_bytes'] = set()
2326
keys['signatures'] = set()
2327
for key in missing_keys:
2328
keys[key[0]].add(key[1:])
2329
if len(keys['revisions']):
2330
# If we allowed copying revisions at this point, we could end up
2331
# copying a revision without copying its required texts: a
2332
# violation of the requirements for repository integrity.
2333
raise AssertionError(
2334
'cannot copy revisions to fill in missing deltas %s' % (
2335
keys['revisions'],))
2336
for substream_kind, keys in viewitems(keys):
2337
vf = getattr(self.from_repository, substream_kind)
2338
if vf is None and keys:
2339
raise AssertionError(
2340
"cannot fill in keys for a versioned file we don't"
2341
" have: %s needs %s" % (substream_kind, keys))
2343
# No need to stream something we don't have
2345
if substream_kind == 'inventories':
2346
# Some missing keys are genuinely ghosts, filter those out.
2347
present = self.from_repository.inventories.get_parent_map(keys)
2348
revs = [key[0] for key in present]
2349
# Get the inventory stream more-or-less as we do for the
2350
# original stream; there's no reason to assume that records
2351
# direct from the source will be suitable for the sink. (Think
2352
# e.g. 2a -> 1.9-rich-root).
2353
for info in self._get_inventory_stream(revs, missing=True):
2357
# Ask for full texts always so that we don't need more round trips
2358
# after this stream.
2359
# Some of the missing keys are genuinely ghosts, so filter absent
2360
# records. The Sink is responsible for doing another check to
2361
# ensure that ghosts don't introduce missing data for future
2363
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
2364
self.to_format._fetch_order, True))
2365
yield substream_kind, stream
2367
def inventory_fetch_order(self):
2368
if self._rich_root_upgrade():
2369
return 'topological'
2371
return self.to_format._fetch_order
2373
def _rich_root_upgrade(self):
2374
return (not self.from_repository._format.rich_root_data and
2375
self.to_format.rich_root_data)
2377
def _get_inventory_stream(self, revision_ids, missing=False):
2378
from_format = self.from_repository._format
2379
if (from_format.supports_chks and self.to_format.supports_chks and
2380
from_format.network_name() == self.to_format.network_name()):
2381
raise AssertionError(
2382
"this case should be handled by GroupCHKStreamSource")
2383
elif 'forceinvdeltas' in debug.debug_flags:
2384
return self._get_convertable_inventory_stream(revision_ids,
2385
delta_versus_null=missing)
2386
elif from_format.network_name() == self.to_format.network_name():
2388
return self._get_simple_inventory_stream(revision_ids,
2390
elif (not from_format.supports_chks and not self.to_format.supports_chks
2391
and from_format._serializer == self.to_format._serializer):
2392
# Essentially the same format.
2393
return self._get_simple_inventory_stream(revision_ids,
2396
# Any time we switch serializations, we want to use an
2397
# inventory-delta based approach.
2398
return self._get_convertable_inventory_stream(revision_ids,
2399
delta_versus_null=missing)
2401
def _get_simple_inventory_stream(self, revision_ids, missing=False):
2402
# NB: This currently reopens the inventory weave in source;
2403
# using a single stream interface instead would avoid this.
2404
from_weave = self.from_repository.inventories
2406
delta_closure = True
2408
delta_closure = not self.delta_on_metadata()
2409
yield ('inventories', from_weave.get_record_stream(
2410
[(rev_id,) for rev_id in revision_ids],
2411
self.inventory_fetch_order(), delta_closure))
2413
def _get_convertable_inventory_stream(self, revision_ids,
2414
delta_versus_null=False):
2415
# The two formats are sufficiently different that there is no fast
2416
# path, so we need to send just inventorydeltas, which any
2417
# sufficiently modern client can insert into any repository.
2418
# The StreamSink code expects to be able to
2419
# convert on the target, so we need to put bytes-on-the-wire that can
2420
# be converted. That means inventory deltas (if the remote is <1.19,
2421
# RemoteStreamSink will fallback to VFS to insert the deltas).
2422
yield ('inventory-deltas',
2423
self._stream_invs_as_deltas(revision_ids,
2424
delta_versus_null=delta_versus_null))
2426
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
2427
"""Return a stream of inventory-deltas for the given rev ids.
2429
:param revision_ids: The list of inventories to transmit
2430
:param delta_versus_null: Don't try to find a minimal delta for this
2431
entry, instead compute the delta versus the NULL_REVISION. This
2432
effectively streams a complete inventory. Used for stuff like
2433
filling in missing parents, etc.
2435
from_repo = self.from_repository
2436
revision_keys = [(rev_id,) for rev_id in revision_ids]
2437
parent_map = from_repo.inventories.get_parent_map(revision_keys)
2438
# XXX: possibly repos could implement a more efficient iter_inv_deltas
2440
inventories = self.from_repository.iter_inventories(
2441
revision_ids, 'topological')
2442
format = from_repo._format
2443
invs_sent_so_far = {_mod_revision.NULL_REVISION}
2444
inventory_cache = lru_cache.LRUCache(50)
2445
null_inventory = from_repo.revision_tree(
2446
_mod_revision.NULL_REVISION).root_inventory
2447
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
2448
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
2449
# repo back into a non-rich-root repo ought to be allowed)
2450
serializer = inventory_delta.InventoryDeltaSerializer(
2451
versioned_root=format.rich_root_data,
2452
tree_references=format.supports_tree_reference)
2453
for inv in inventories:
2454
key = (inv.revision_id,)
2455
parent_keys = parent_map.get(key, ())
2457
if not delta_versus_null and parent_keys:
2458
# The caller did not ask for complete inventories and we have
2459
# some parents that we can delta against. Make a delta against
2460
# each parent so that we can find the smallest.
2461
parent_ids = [parent_key[0] for parent_key in parent_keys]
2462
for parent_id in parent_ids:
2463
if parent_id not in invs_sent_so_far:
2464
# We don't know that the remote side has this basis, so
2467
if parent_id == _mod_revision.NULL_REVISION:
2468
parent_inv = null_inventory
2470
parent_inv = inventory_cache.get(parent_id, None)
2471
if parent_inv is None:
2472
parent_inv = from_repo.get_inventory(parent_id)
2473
candidate_delta = inv._make_delta(parent_inv)
2474
if (delta is None or
2475
len(delta) > len(candidate_delta)):
2476
delta = candidate_delta
2477
basis_id = parent_id
2479
# Either none of the parents ended up being suitable, or we
2480
# were asked to delta against NULL
2481
basis_id = _mod_revision.NULL_REVISION
2482
delta = inv._make_delta(null_inventory)
2483
invs_sent_so_far.add(inv.revision_id)
2484
inventory_cache[inv.revision_id] = inv
2485
delta_serialized = ''.join(
2486
serializer.delta_to_lines(basis_id, key[-1], delta))
2487
yield versionedfile.FulltextContentFactory(
2488
key, parent_keys, None, delta_serialized)
2491
class _VersionedFileChecker(object):
2493
def __init__(self, repository, text_key_references=None, ancestors=None):
2494
self.repository = repository
2495
self.text_index = self.repository._generate_text_key_index(
2496
text_key_references=text_key_references, ancestors=ancestors)
2498
def calculate_file_version_parents(self, text_key):
2499
"""Calculate the correct parents for a file version according to
2502
parent_keys = self.text_index[text_key]
2503
if parent_keys == [_mod_revision.NULL_REVISION]:
2505
return tuple(parent_keys)
2507
def check_file_version_parents(self, texts, progress_bar=None):
2508
"""Check the parents stored in a versioned file are correct.
2510
It also detects file versions that are not referenced by their
2511
corresponding revision's inventory.
2513
:returns: A tuple of (wrong_parents, dangling_file_versions).
2514
wrong_parents is a dict mapping {revision_id: (stored_parents,
2515
correct_parents)} for each revision_id where the stored parents
2516
are not correct. dangling_file_versions is a set of (file_id,
2517
revision_id) tuples for versions that are present in this versioned
2518
file, but not used by the corresponding inventory.
2520
local_progress = None
2521
if progress_bar is None:
2522
local_progress = ui.ui_factory.nested_progress_bar()
2523
progress_bar = local_progress
2525
return self._check_file_version_parents(texts, progress_bar)
2528
local_progress.finished()
2530
def _check_file_version_parents(self, texts, progress_bar):
2531
"""See check_file_version_parents."""
2533
self.file_ids = {file_id for file_id, _ in self.text_index}
2534
# text keys is now grouped by file_id
2535
n_versions = len(self.text_index)
2536
progress_bar.update(gettext('loading text store'), 0, n_versions)
2537
parent_map = self.repository.texts.get_parent_map(self.text_index)
2538
# On unlistable transports this could well be empty/error...
2539
text_keys = self.repository.texts.keys()
2540
unused_keys = frozenset(text_keys) - set(self.text_index)
2541
for num, key in enumerate(self.text_index):
2542
progress_bar.update(gettext('checking text graph'), num, n_versions)
2543
correct_parents = self.calculate_file_version_parents(key)
2545
knit_parents = parent_map[key]
2546
except errors.RevisionNotPresent:
2549
if correct_parents != knit_parents:
2550
wrong_parents[key] = (knit_parents, correct_parents)
2551
return wrong_parents, unused_keys
2554
class InterVersionedFileRepository(InterRepository):
2556
_walk_to_common_revisions_batch_size = 50
2558
supports_fetch_spec = True
2561
def fetch(self, revision_id=None, find_ghosts=False,
2563
"""Fetch the content required to construct revision_id.
2565
The content is copied from self.source to self.target.
2567
:param revision_id: if None all content is copied, if NULL_REVISION no
2571
if self.target._format.experimental:
2572
ui.ui_factory.show_user_warning('experimental_format_fetch',
2573
from_format=self.source._format,
2574
to_format=self.target._format)
2575
from breezy.fetch import RepoFetcher
2576
# See <https://launchpad.net/bugs/456077> asking for a warning here
2577
if self.source._format.network_name() != self.target._format.network_name():
2578
ui.ui_factory.show_user_warning('cross_format_fetch',
2579
from_format=self.source._format,
2580
to_format=self.target._format)
2581
f = RepoFetcher(to_repository=self.target,
2582
from_repository=self.source,
2583
last_revision=revision_id,
2584
fetch_spec=fetch_spec,
2585
find_ghosts=find_ghosts)
2587
def _walk_to_common_revisions(self, revision_ids, if_present_ids=None):
2588
"""Walk out from revision_ids in source to revisions target has.
2590
:param revision_ids: The start point for the search.
2591
:return: A set of revision ids.
2593
target_graph = self.target.get_graph()
2594
revision_ids = frozenset(revision_ids)
2596
all_wanted_revs = revision_ids.union(if_present_ids)
2598
all_wanted_revs = revision_ids
2599
missing_revs = set()
2600
source_graph = self.source.get_graph()
2601
# ensure we don't pay silly lookup costs.
2602
searcher = source_graph._make_breadth_first_searcher(all_wanted_revs)
2603
null_set = frozenset([_mod_revision.NULL_REVISION])
2604
searcher_exhausted = False
2608
# Iterate the searcher until we have enough next_revs
2609
while len(next_revs) < self._walk_to_common_revisions_batch_size:
2611
next_revs_part, ghosts_part = searcher.next_with_ghosts()
2612
next_revs.update(next_revs_part)
2613
ghosts.update(ghosts_part)
2614
except StopIteration:
2615
searcher_exhausted = True
2617
# If there are ghosts in the source graph, and the caller asked for
2618
# them, make sure that they are present in the target.
2619
# We don't care about other ghosts as we can't fetch them and
2620
# haven't been asked to.
2621
ghosts_to_check = set(revision_ids.intersection(ghosts))
2622
revs_to_get = set(next_revs).union(ghosts_to_check)
2624
have_revs = set(target_graph.get_parent_map(revs_to_get))
2625
# we always have NULL_REVISION present.
2626
have_revs = have_revs.union(null_set)
2627
# Check if the target is missing any ghosts we need.
2628
ghosts_to_check.difference_update(have_revs)
2630
# One of the caller's revision_ids is a ghost in both the
2631
# source and the target.
2632
raise errors.NoSuchRevision(
2633
self.source, ghosts_to_check.pop())
2634
missing_revs.update(next_revs - have_revs)
2635
# Because we may have walked past the original stop point, make
2636
# sure everything is stopped
2637
stop_revs = searcher.find_seen_ancestors(have_revs)
2638
searcher.stop_searching_any(stop_revs)
2639
if searcher_exhausted:
2641
(started_keys, excludes, included_keys) = searcher.get_state()
2642
return vf_search.SearchResult(started_keys, excludes,
2643
len(included_keys), included_keys)
2646
def search_missing_revision_ids(self,
2647
find_ghosts=True, revision_ids=None, if_present_ids=None,
2649
"""Return the revision ids that source has that target does not.
2651
:param revision_ids: return revision ids included by these
2652
revision_ids. NoSuchRevision will be raised if any of these
2653
revisions are not present.
2654
:param if_present_ids: like revision_ids, but will not cause
2655
NoSuchRevision if any of these are absent, instead they will simply
2656
not be in the result. This is useful for e.g. finding revisions
2657
to fetch for tags, which may reference absent revisions.
2658
:param find_ghosts: If True find missing revisions in deep history
2659
rather than just finding the surface difference.
2660
:return: A breezy.graph.SearchResult.
2662
# stop searching at found target revisions.
2663
if not find_ghosts and (revision_ids is not None or if_present_ids is
2665
result = self._walk_to_common_revisions(revision_ids,
2666
if_present_ids=if_present_ids)
2669
result_set = result.get_keys()
2671
# generic, possibly worst case, slow code path.
2672
target_ids = set(self.target.all_revision_ids())
2673
source_ids = self._present_source_revisions_for(
2674
revision_ids, if_present_ids)
2675
result_set = set(source_ids).difference(target_ids)
2676
if limit is not None:
2677
topo_ordered = self.source.get_graph().iter_topo_order(result_set)
2678
result_set = set(itertools.islice(topo_ordered, limit))
2679
return self.source.revision_ids_to_search_result(result_set)
2681
def _present_source_revisions_for(self, revision_ids, if_present_ids=None):
2682
"""Returns set of all revisions in ancestry of revision_ids present in
2685
:param revision_ids: if None, all revisions in source are returned.
2686
:param if_present_ids: like revision_ids, but if any/all of these are
2687
absent no error is raised.
2689
if revision_ids is not None or if_present_ids is not None:
2690
# First, ensure all specified revisions exist. Callers expect
2691
# NoSuchRevision when they pass absent revision_ids here.
2692
if revision_ids is None:
2693
revision_ids = set()
2694
if if_present_ids is None:
2695
if_present_ids = set()
2696
revision_ids = set(revision_ids)
2697
if_present_ids = set(if_present_ids)
2698
all_wanted_ids = revision_ids.union(if_present_ids)
2699
graph = self.source.get_graph()
2700
present_revs = set(graph.get_parent_map(all_wanted_ids))
2701
missing = revision_ids.difference(present_revs)
2703
raise errors.NoSuchRevision(self.source, missing.pop())
2704
found_ids = all_wanted_ids.intersection(present_revs)
2705
source_ids = [rev_id for (rev_id, parents) in
2706
graph.iter_ancestry(found_ids)
2707
if rev_id != _mod_revision.NULL_REVISION
2708
and parents is not None]
2710
source_ids = self.source.all_revision_ids()
2711
return set(source_ids)
2714
def _get_repo_format_to_test(self):
2718
def is_compatible(cls, source, target):
2719
# The default implementation is compatible with everything
2720
return (source._format.supports_full_versioned_files and
2721
target._format.supports_full_versioned_files)
2724
class InterDifferingSerializer(InterVersionedFileRepository):
2727
def _get_repo_format_to_test(self):
2731
def is_compatible(source, target):
2732
if not source._format.supports_full_versioned_files:
2734
if not target._format.supports_full_versioned_files:
2736
# This is redundant with format.check_conversion_target(), however that
2737
# raises an exception, and we just want to say "False" as in we won't
2738
# support converting between these formats.
2739
if 'IDS_never' in debug.debug_flags:
2741
if source.supports_rich_root() and not target.supports_rich_root():
2743
if (source._format.supports_tree_reference
2744
and not target._format.supports_tree_reference):
2746
if target._fallback_repositories and target._format.supports_chks:
2747
# IDS doesn't know how to copy CHKs for the parent inventories it
2748
# adds to stacked repos.
2750
if 'IDS_always' in debug.debug_flags:
2752
# Only use this code path for local source and target. IDS does far
2753
# too much IO (both bandwidth and roundtrips) over a network.
2754
if not source.bzrdir.transport.base.startswith('file:///'):
2756
if not target.bzrdir.transport.base.startswith('file:///'):
2760
def _get_trees(self, revision_ids, cache):
2762
for rev_id in revision_ids:
2764
possible_trees.append((rev_id, cache[rev_id]))
2766
# Not cached, but inventory might be present anyway.
2768
tree = self.source.revision_tree(rev_id)
2769
except errors.NoSuchRevision:
2770
# Nope, parent is ghost.
2773
cache[rev_id] = tree
2774
possible_trees.append((rev_id, tree))
2775
return possible_trees
2777
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
2778
"""Get the best delta and base for this revision.
2780
:return: (basis_id, delta)
2783
# Generate deltas against each tree, to find the shortest.
2784
# FIXME: Support nested trees
2785
texts_possibly_new_in_tree = set()
2786
for basis_id, basis_tree in possible_trees:
2787
delta = tree.root_inventory._make_delta(basis_tree.root_inventory)
2788
for old_path, new_path, file_id, new_entry in delta:
2789
if new_path is None:
2790
# This file_id isn't present in the new rev, so we don't
2794
# Rich roots are handled elsewhere...
2796
kind = new_entry.kind
2797
if kind != 'directory' and kind != 'file':
2798
# No text record associated with this inventory entry.
2800
# This is a directory or file that has changed somehow.
2801
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
2802
deltas.append((len(delta), basis_id, delta))
2804
return deltas[0][1:]
2806
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
2807
"""Find all parent revisions that are absent, but for which the
2808
inventory is present, and copy those inventories.
2810
This is necessary to preserve correctness when the source is stacked
2811
without fallbacks configured. (Note that in cases like upgrade the
2812
source may be not have _fallback_repositories even though it is
2815
parent_revs = set(itertools.chain.from_iterable(viewvalues(
2817
present_parents = self.source.get_parent_map(parent_revs)
2818
absent_parents = parent_revs.difference(present_parents)
2819
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
2820
(rev_id,) for rev_id in absent_parents)
2821
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
2822
for parent_tree in self.source.revision_trees(parent_inv_ids):
2823
current_revision_id = parent_tree.get_revision_id()
2824
parents_parents_keys = parent_invs_keys_for_stacking[
2825
(current_revision_id,)]
2826
parents_parents = [key[-1] for key in parents_parents_keys]
2827
basis_id = _mod_revision.NULL_REVISION
2828
basis_tree = self.source.revision_tree(basis_id)
2829
delta = parent_tree.root_inventory._make_delta(
2830
basis_tree.root_inventory)
2831
self.target.add_inventory_by_delta(
2832
basis_id, delta, current_revision_id, parents_parents)
2833
cache[current_revision_id] = parent_tree
2835
def _fetch_batch(self, revision_ids, basis_id, cache):
2836
"""Fetch across a few revisions.
2838
:param revision_ids: The revisions to copy
2839
:param basis_id: The revision_id of a tree that must be in cache, used
2840
as a basis for delta when no other base is available
2841
:param cache: A cache of RevisionTrees that we can use.
2842
:return: The revision_id of the last converted tree. The RevisionTree
2843
for it will be in cache
2845
# Walk though all revisions; get inventory deltas, copy referenced
2846
# texts that delta references, insert the delta, revision and
2848
root_keys_to_create = set()
2851
pending_revisions = []
2852
parent_map = self.source.get_parent_map(revision_ids)
2853
self._fetch_parent_invs_for_stacking(parent_map, cache)
2854
self.source._safe_to_return_from_cache = True
2855
for tree in self.source.revision_trees(revision_ids):
2856
# Find a inventory delta for this revision.
2857
# Find text entries that need to be copied, too.
2858
current_revision_id = tree.get_revision_id()
2859
parent_ids = parent_map.get(current_revision_id, ())
2860
parent_trees = self._get_trees(parent_ids, cache)
2861
possible_trees = list(parent_trees)
2862
if len(possible_trees) == 0:
2863
# There either aren't any parents, or the parents are ghosts,
2864
# so just use the last converted tree.
2865
possible_trees.append((basis_id, cache[basis_id]))
2866
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
2868
revision = self.source.get_revision(current_revision_id)
2869
pending_deltas.append((basis_id, delta,
2870
current_revision_id, revision.parent_ids))
2871
if self._converting_to_rich_root:
2872
self._revision_id_to_root_id[current_revision_id] = \
2874
# Determine which texts are in present in this revision but not in
2875
# any of the available parents.
2876
texts_possibly_new_in_tree = set()
2877
for old_path, new_path, file_id, entry in delta:
2878
if new_path is None:
2879
# This file_id isn't present in the new rev
2883
if not self.target.supports_rich_root():
2884
# The target doesn't support rich root, so we don't
2887
if self._converting_to_rich_root:
2888
# This can't be copied normally, we have to insert
2890
root_keys_to_create.add((file_id, entry.revision))
2893
texts_possibly_new_in_tree.add((file_id, entry.revision))
2894
for basis_id, basis_tree in possible_trees:
2895
basis_inv = basis_tree.root_inventory
2896
for file_key in list(texts_possibly_new_in_tree):
2897
file_id, file_revision = file_key
2899
entry = basis_inv[file_id]
2900
except errors.NoSuchId:
2902
if entry.revision == file_revision:
2903
texts_possibly_new_in_tree.remove(file_key)
2904
text_keys.update(texts_possibly_new_in_tree)
2905
pending_revisions.append(revision)
2906
cache[current_revision_id] = tree
2907
basis_id = current_revision_id
2908
self.source._safe_to_return_from_cache = False
2910
from_texts = self.source.texts
2911
to_texts = self.target.texts
2912
if root_keys_to_create:
2913
root_stream = _mod_fetch._new_root_data_stream(
2914
root_keys_to_create, self._revision_id_to_root_id, parent_map,
2916
to_texts.insert_record_stream(root_stream)
2917
to_texts.insert_record_stream(from_texts.get_record_stream(
2918
text_keys, self.target._format._fetch_order,
2919
not self.target._format._fetch_uses_deltas))
2920
# insert inventory deltas
2921
for delta in pending_deltas:
2922
self.target.add_inventory_by_delta(*delta)
2923
if self.target._fallback_repositories:
2924
# Make sure this stacked repository has all the parent inventories
2925
# for the new revisions that we are about to insert. We do this
2926
# before adding the revisions so that no revision is added until
2927
# all the inventories it may depend on are added.
2928
# Note that this is overzealous, as we may have fetched these in an
2931
revision_ids = set()
2932
for revision in pending_revisions:
2933
revision_ids.add(revision.revision_id)
2934
parent_ids.update(revision.parent_ids)
2935
parent_ids.difference_update(revision_ids)
2936
parent_ids.discard(_mod_revision.NULL_REVISION)
2937
parent_map = self.source.get_parent_map(parent_ids)
2938
# we iterate over parent_map and not parent_ids because we don't
2939
# want to try copying any revision which is a ghost
2940
for parent_tree in self.source.revision_trees(parent_map):
2941
current_revision_id = parent_tree.get_revision_id()
2942
parents_parents = parent_map[current_revision_id]
2943
possible_trees = self._get_trees(parents_parents, cache)
2944
if len(possible_trees) == 0:
2945
# There either aren't any parents, or the parents are
2946
# ghosts, so just use the last converted tree.
2947
possible_trees.append((basis_id, cache[basis_id]))
2948
basis_id, delta = self._get_delta_for_revision(parent_tree,
2949
parents_parents, possible_trees)
2950
self.target.add_inventory_by_delta(
2951
basis_id, delta, current_revision_id, parents_parents)
2952
# insert signatures and revisions
2953
for revision in pending_revisions:
2955
signature = self.source.get_signature_text(
2956
revision.revision_id)
2957
self.target.add_signature_text(revision.revision_id,
2959
except errors.NoSuchRevision:
2961
self.target.add_revision(revision.revision_id, revision)
2964
def _fetch_all_revisions(self, revision_ids, pb):
2965
"""Fetch everything for the list of revisions.
2967
:param revision_ids: The list of revisions to fetch. Must be in
2969
:param pb: A ProgressTask
2972
basis_id, basis_tree = self._get_basis(revision_ids[0])
2974
cache = lru_cache.LRUCache(100)
2975
cache[basis_id] = basis_tree
2976
del basis_tree # We don't want to hang on to it here
2980
for offset in range(0, len(revision_ids), batch_size):
2981
self.target.start_write_group()
2983
pb.update(gettext('Transferring revisions'), offset,
2985
batch = revision_ids[offset:offset+batch_size]
2986
basis_id = self._fetch_batch(batch, basis_id, cache)
2988
self.source._safe_to_return_from_cache = False
2989
self.target.abort_write_group()
2992
hint = self.target.commit_write_group()
2995
if hints and self.target._format.pack_compresses:
2996
self.target.pack(hint=hints)
2997
pb.update(gettext('Transferring revisions'), len(revision_ids),
3001
def fetch(self, revision_id=None, find_ghosts=False,
3003
"""See InterRepository.fetch()."""
3004
if fetch_spec is not None:
3005
revision_ids = fetch_spec.get_keys()
3008
if self.source._format.experimental:
3009
ui.ui_factory.show_user_warning('experimental_format_fetch',
3010
from_format=self.source._format,
3011
to_format=self.target._format)
3012
if (not self.source.supports_rich_root()
3013
and self.target.supports_rich_root()):
3014
self._converting_to_rich_root = True
3015
self._revision_id_to_root_id = {}
3017
self._converting_to_rich_root = False
3018
# See <https://launchpad.net/bugs/456077> asking for a warning here
3019
if self.source._format.network_name() != self.target._format.network_name():
3020
ui.ui_factory.show_user_warning('cross_format_fetch',
3021
from_format=self.source._format,
3022
to_format=self.target._format)
3023
if revision_ids is None:
3025
search_revision_ids = [revision_id]
3027
search_revision_ids = None
3028
revision_ids = self.target.search_missing_revision_ids(self.source,
3029
revision_ids=search_revision_ids,
3030
find_ghosts=find_ghosts).get_keys()
3031
if not revision_ids:
3033
revision_ids = tsort.topo_sort(
3034
self.source.get_graph().get_parent_map(revision_ids))
3035
if not revision_ids:
3037
# Walk though all revisions; get inventory deltas, copy referenced
3038
# texts that delta references, insert the delta, revision and
3040
pb = ui.ui_factory.nested_progress_bar()
3042
self._fetch_all_revisions(revision_ids, pb)
3045
return len(revision_ids), 0
3047
def _get_basis(self, first_revision_id):
3048
"""Get a revision and tree which exists in the target.
3050
This assumes that first_revision_id is selected for transmission
3051
because all other ancestors are already present. If we can't find an
3052
ancestor we fall back to NULL_REVISION since we know that is safe.
3054
:return: (basis_id, basis_tree)
3056
first_rev = self.source.get_revision(first_revision_id)
3058
basis_id = first_rev.parent_ids[0]
3059
# only valid as a basis if the target has it
3060
self.target.get_revision(basis_id)
3061
# Try to get a basis tree - if it's a ghost it will hit the
3062
# NoSuchRevision case.
3063
basis_tree = self.source.revision_tree(basis_id)
3064
except (IndexError, errors.NoSuchRevision):
3065
basis_id = _mod_revision.NULL_REVISION
3066
basis_tree = self.source.revision_tree(basis_id)
3067
return basis_id, basis_tree
3070
class InterSameDataRepository(InterVersionedFileRepository):
3071
"""Code for converting between repositories that represent the same data.
3073
Data format and model must match for this to work.
3077
def _get_repo_format_to_test(self):
3078
"""Repository format for testing with.
3080
InterSameData can pull from subtree to subtree and from non-subtree to
3081
non-subtree, so we test this with the richest repository format.
3083
from breezy.repofmt import knitrepo
3084
return knitrepo.RepositoryFormatKnit3()
3087
def is_compatible(source, target):
3089
InterRepository._same_model(source, target) and
3090
source._format.supports_full_versioned_files and
3091
target._format.supports_full_versioned_files)
3094
InterRepository.register_optimiser(InterVersionedFileRepository)
3095
InterRepository.register_optimiser(InterDifferingSerializer)
3096
InterRepository.register_optimiser(InterSameDataRepository)
3099
def install_revisions(repository, iterable, num_revisions=None, pb=None):
3100
"""Install all revision data into a repository.
3102
Accepts an iterable of revision, tree, signature tuples. The signature
3105
repository.start_write_group()
3107
inventory_cache = lru_cache.LRUCache(10)
3108
for n, (revision, revision_tree, signature) in enumerate(iterable):
3109
_install_revision(repository, revision, revision_tree, signature,
3112
pb.update(gettext('Transferring revisions'), n + 1, num_revisions)
3114
repository.abort_write_group()
3117
repository.commit_write_group()
3120
def _install_revision(repository, rev, revision_tree, signature,
3122
"""Install all revision data into a repository."""
3123
present_parents = []
3125
for p_id in rev.parent_ids:
3126
if repository.has_revision(p_id):
3127
present_parents.append(p_id)
3128
parent_trees[p_id] = repository.revision_tree(p_id)
3130
parent_trees[p_id] = repository.revision_tree(
3131
_mod_revision.NULL_REVISION)
3133
# FIXME: Support nested trees
3134
inv = revision_tree.root_inventory
3135
entries = inv.iter_entries()
3136
# backwards compatibility hack: skip the root id.
3137
if not repository.supports_rich_root():
3138
path, root = next(entries)
3139
if root.revision != rev.revision_id:
3140
raise errors.IncompatibleRevision(repr(repository))
3142
for path, ie in entries:
3143
text_keys[(ie.file_id, ie.revision)] = ie
3144
text_parent_map = repository.texts.get_parent_map(text_keys)
3145
missing_texts = set(text_keys) - set(text_parent_map)
3146
# Add the texts that are not already present
3147
for text_key in missing_texts:
3148
ie = text_keys[text_key]
3150
# FIXME: TODO: The following loop overlaps/duplicates that done by
3151
# commit to determine parents. There is a latent/real bug here where
3152
# the parents inserted are not those commit would do - in particular
3153
# they are not filtered by heads(). RBC, AB
3154
for revision, tree in viewitems(parent_trees):
3155
if not tree.has_id(ie.file_id):
3157
parent_id = tree.get_file_revision(ie.file_id)
3158
if parent_id in text_parents:
3160
text_parents.append((ie.file_id, parent_id))
3161
lines = revision_tree.get_file(ie.file_id).readlines()
3162
repository.texts.add_lines(text_key, text_parents, lines)
3164
# install the inventory
3165
if repository._format._commit_inv_deltas and len(rev.parent_ids):
3166
# Cache this inventory
3167
inventory_cache[rev.revision_id] = inv
3169
basis_inv = inventory_cache[rev.parent_ids[0]]
3171
repository.add_inventory(rev.revision_id, inv, present_parents)
3173
delta = inv._make_delta(basis_inv)
3174
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
3175
rev.revision_id, present_parents)
3177
repository.add_inventory(rev.revision_id, inv, present_parents)
3178
except errors.RevisionAlreadyPresent:
3180
if signature is not None:
3181
repository.add_signature_text(rev.revision_id, signature)
3182
repository.add_revision(rev.revision_id, rev, inv)
3185
def install_revision(repository, rev, revision_tree):
3186
"""Install all revision data into a repository."""
3187
install_revisions(repository, [(rev, revision_tree, None)])
added
removed
breezy/vf_repository.py
