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 bzrlib.lazy_import import lazy_import
23
lazy_import(globals(), """
28
config as _mod_config,
37
revision as _mod_revision,
38
serializer as _mod_serializer,
47
from bzrlib.recordcounter import RecordCounter
48
from bzrlib.revisiontree import InventoryRevisionTree
49
from bzrlib.testament import Testament
50
from bzrlib.i18n import gettext
56
from bzrlib.decorators import (
61
from bzrlib.inventory import (
68
from bzrlib.repository import (
72
RepositoryFormatMetaDir,
77
from bzrlib.trace import (
82
class VersionedFileRepositoryFormat(RepositoryFormat):
83
"""Base class for all repository formats that are VersionedFiles-based."""
85
supports_full_versioned_files = True
86
supports_versioned_directories = True
87
supports_unreferenced_revisions = True
89
# Should commit add an inventory, or an inventory delta to the repository.
90
_commit_inv_deltas = True
91
# What order should fetch operations request streams in?
92
# The default is unordered as that is the cheapest for an origin to
94
_fetch_order = 'unordered'
95
# Does this repository format use deltas that can be fetched as-deltas ?
96
# (E.g. knits, where the knit deltas can be transplanted intact.
97
# We default to False, which will ensure that enough data to get
98
# a full text out of any fetch stream will be grabbed.
99
_fetch_uses_deltas = False
102
class VersionedFileCommitBuilder(CommitBuilder):
103
"""Commit builder implementation for versioned files based repositories.
106
# this commit builder supports the record_entry_contents interface
107
supports_record_entry_contents = True
109
# the default CommitBuilder does not manage trees whose root is versioned.
110
_versioned_root = False
112
def __init__(self, repository, parents, config_stack, timestamp=None,
113
timezone=None, committer=None, revprops=None,
114
revision_id=None, lossy=False):
115
super(VersionedFileCommitBuilder, self).__init__(repository,
116
parents, config_stack, timestamp, timezone, committer, revprops,
119
basis_id = self.parents[0]
121
basis_id = _mod_revision.NULL_REVISION
122
self.basis_delta_revision = basis_id
123
self.new_inventory = Inventory(None)
124
self._basis_delta = []
125
self.__heads = graph.HeadsCache(repository.get_graph()).heads
126
# memo'd check for no-op commits.
127
self._any_changes = False
128
# API compatibility, older code that used CommitBuilder did not call
129
# .record_delete(), which means the delta that is computed would not be
130
# valid. Callers that will call record_delete() should call
131
# .will_record_deletes() to indicate that.
132
self._recording_deletes = False
134
def will_record_deletes(self):
135
"""Tell the commit builder that deletes are being notified.
137
This enables the accumulation of an inventory delta; for the resulting
138
commit to be valid, deletes against the basis MUST be recorded via
139
builder.record_delete().
141
self._recording_deletes = True
143
def any_changes(self):
144
"""Return True if any entries were changed.
146
This includes merge-only changes. It is the core for the --unchanged
149
:return: True if any changes have occured.
151
return self._any_changes
153
def _ensure_fallback_inventories(self):
154
"""Ensure that appropriate inventories are available.
156
This only applies to repositories that are stacked, and is about
157
enusring the stacking invariants. Namely, that for any revision that is
158
present, we either have all of the file content, or we have the parent
159
inventory and the delta file content.
161
if not self.repository._fallback_repositories:
163
if not self.repository._format.supports_chks:
164
raise errors.BzrError("Cannot commit directly to a stacked branch"
165
" in pre-2a formats. See "
166
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
167
# This is a stacked repo, we need to make sure we have the parent
168
# inventories for the parents.
169
parent_keys = [(p,) for p in self.parents]
170
parent_map = self.repository.inventories._index.get_parent_map(parent_keys)
171
missing_parent_keys = set([pk for pk in parent_keys
172
if pk not in parent_map])
173
fallback_repos = list(reversed(self.repository._fallback_repositories))
174
missing_keys = [('inventories', pk[0])
175
for pk in missing_parent_keys]
177
while missing_keys and fallback_repos:
178
fallback_repo = fallback_repos.pop()
179
source = fallback_repo._get_source(self.repository._format)
180
sink = self.repository._get_sink()
181
stream = source.get_stream_for_missing_keys(missing_keys)
182
missing_keys = sink.insert_stream_without_locking(stream,
183
self.repository._format)
185
raise errors.BzrError('Unable to fill in parent inventories for a'
188
def commit(self, message):
189
"""Make the actual commit.
191
:return: The revision id of the recorded revision.
193
self._validate_unicode_text(message, 'commit message')
194
rev = _mod_revision.Revision(
195
timestamp=self._timestamp,
196
timezone=self._timezone,
197
committer=self._committer,
199
inventory_sha1=self.inv_sha1,
200
revision_id=self._new_revision_id,
201
properties=self._revprops)
202
rev.parent_ids = self.parents
203
if self._config_stack.get('create_signatures') == _mod_config.SIGN_ALWAYS:
204
testament = Testament(rev, self.revision_tree())
205
plaintext = testament.as_short_text()
206
self.repository.store_revision_signature(
207
gpg.GPGStrategy(self._config_stack), plaintext,
208
self._new_revision_id)
209
self.repository._add_revision(rev)
210
self._ensure_fallback_inventories()
211
self.repository.commit_write_group()
212
return self._new_revision_id
215
"""Abort the commit that is being built.
217
self.repository.abort_write_group()
219
def revision_tree(self):
220
"""Return the tree that was just committed.
222
After calling commit() this can be called to get a
223
RevisionTree representing the newly committed tree. This is
224
preferred to calling Repository.revision_tree() because that may
225
require deserializing the inventory, while we already have a copy in
228
if self.new_inventory is None:
229
self.new_inventory = self.repository.get_inventory(
230
self._new_revision_id)
231
return InventoryRevisionTree(self.repository, self.new_inventory,
232
self._new_revision_id)
234
def finish_inventory(self):
235
"""Tell the builder that the inventory is finished.
237
:return: The inventory id in the repository, which can be used with
238
repository.get_inventory.
240
if self.new_inventory is None:
241
# an inventory delta was accumulated without creating a new
243
basis_id = self.basis_delta_revision
244
# We ignore the 'inventory' returned by add_inventory_by_delta
245
# because self.new_inventory is used to hint to the rest of the
246
# system what code path was taken
247
self.inv_sha1, _ = self.repository.add_inventory_by_delta(
248
basis_id, self._basis_delta, self._new_revision_id,
251
if self.new_inventory.root is None:
252
raise AssertionError('Root entry should be supplied to'
253
' record_entry_contents, as of bzr 0.10.')
254
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
255
self.new_inventory.revision_id = self._new_revision_id
256
self.inv_sha1 = self.repository.add_inventory(
257
self._new_revision_id,
261
return self._new_revision_id
263
def _check_root(self, ie, parent_invs, tree):
264
"""Helper for record_entry_contents.
266
:param ie: An entry being added.
267
:param parent_invs: The inventories of the parent revisions of the
269
:param tree: The tree that is being committed.
271
# In this revision format, root entries have no knit or weave When
272
# serializing out to disk and back in root.revision is always
274
ie.revision = self._new_revision_id
276
def _require_root_change(self, tree):
277
"""Enforce an appropriate root object change.
279
This is called once when record_iter_changes is called, if and only if
280
the root was not in the delta calculated by record_iter_changes.
282
:param tree: The tree which is being committed.
284
if len(self.parents) == 0:
285
raise errors.RootMissing()
286
entry = entry_factory['directory'](tree.path2id(''), '',
288
entry.revision = self._new_revision_id
289
self._basis_delta.append(('', '', entry.file_id, entry))
291
def _get_delta(self, ie, basis_inv, path):
292
"""Get a delta against the basis inventory for ie."""
293
if not basis_inv.has_id(ie.file_id):
295
result = (None, path, ie.file_id, ie)
296
self._basis_delta.append(result)
298
elif ie != basis_inv[ie.file_id]:
300
# TODO: avoid tis id2path call.
301
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
302
self._basis_delta.append(result)
308
def _heads(self, file_id, revision_ids):
309
"""Calculate the graph heads for revision_ids in the graph of file_id.
311
This can use either a per-file graph or a global revision graph as we
312
have an identity relationship between the two graphs.
314
return self.__heads(revision_ids)
316
def get_basis_delta(self):
317
"""Return the complete inventory delta versus the basis inventory.
319
This has been built up with the calls to record_delete and
320
record_entry_contents. The client must have already called
321
will_record_deletes() to indicate that they will be generating a
324
:return: An inventory delta, suitable for use with apply_delta, or
325
Repository.add_inventory_by_delta, etc.
327
if not self._recording_deletes:
328
raise AssertionError("recording deletes not activated.")
329
return self._basis_delta
331
def record_delete(self, path, file_id):
332
"""Record that a delete occured against a basis tree.
334
This is an optional API - when used it adds items to the basis_delta
335
being accumulated by the commit builder. It cannot be called unless the
336
method will_record_deletes() has been called to inform the builder that
337
a delta is being supplied.
339
:param path: The path of the thing deleted.
340
:param file_id: The file id that was deleted.
342
if not self._recording_deletes:
343
raise AssertionError("recording deletes not activated.")
344
delta = (path, None, file_id, None)
345
self._basis_delta.append(delta)
346
self._any_changes = True
349
def record_entry_contents(self, ie, parent_invs, path, tree,
351
"""Record the content of ie from tree into the commit if needed.
353
Side effect: sets ie.revision when unchanged
355
:param ie: An inventory entry present in the commit.
356
:param parent_invs: The inventories of the parent revisions of the
358
:param path: The path the entry is at in the tree.
359
:param tree: The tree which contains this entry and should be used to
361
:param content_summary: Summary data from the tree about the paths
362
content - stat, length, exec, sha/link target. This is only
363
accessed when the entry has a revision of None - that is when it is
364
a candidate to commit.
365
:return: A tuple (change_delta, version_recorded, fs_hash).
366
change_delta is an inventory_delta change for this entry against
367
the basis tree of the commit, or None if no change occured against
369
version_recorded is True if a new version of the entry has been
370
recorded. For instance, committing a merge where a file was only
371
changed on the other side will return (delta, False).
372
fs_hash is either None, or the hash details for the path (currently
373
a tuple of the contents sha1 and the statvalue returned by
374
tree.get_file_with_stat()).
376
if self.new_inventory.root is None:
377
if ie.parent_id is not None:
378
raise errors.RootMissing()
379
self._check_root(ie, parent_invs, tree)
380
if ie.revision is None:
381
kind = content_summary[0]
383
# ie is carried over from a prior commit
385
# XXX: repository specific check for nested tree support goes here - if
386
# the repo doesn't want nested trees we skip it ?
387
if (kind == 'tree-reference' and
388
not self.repository._format.supports_tree_reference):
389
# mismatch between commit builder logic and repository:
390
# this needs the entry creation pushed down into the builder.
391
raise NotImplementedError('Missing repository subtree support.')
392
self.new_inventory.add(ie)
394
# TODO: slow, take it out of the inner loop.
396
basis_inv = parent_invs[0]
398
basis_inv = Inventory(root_id=None)
400
# ie.revision is always None if the InventoryEntry is considered
401
# for committing. We may record the previous parents revision if the
402
# content is actually unchanged against a sole head.
403
if ie.revision is not None:
404
if not self._versioned_root and path == '':
405
# repositories that do not version the root set the root's
406
# revision to the new commit even when no change occurs (more
407
# specifically, they do not record a revision on the root; and
408
# the rev id is assigned to the root during deserialisation -
409
# this masks when a change may have occurred against the basis.
410
# To match this we always issue a delta, because the revision
411
# of the root will always be changing.
412
if basis_inv.has_id(ie.file_id):
413
delta = (basis_inv.id2path(ie.file_id), path,
417
delta = (None, path, ie.file_id, ie)
418
self._basis_delta.append(delta)
419
return delta, False, None
421
# we don't need to commit this, because the caller already
422
# determined that an existing revision of this file is
423
# appropriate. If it's not being considered for committing then
424
# it and all its parents to the root must be unaltered so
425
# no-change against the basis.
426
if ie.revision == self._new_revision_id:
427
raise AssertionError("Impossible situation, a skipped "
428
"inventory entry (%r) claims to be modified in this "
429
"commit (%r).", (ie, self._new_revision_id))
430
return None, False, None
431
# XXX: Friction: parent_candidates should return a list not a dict
432
# so that we don't have to walk the inventories again.
433
parent_candidate_entries = ie.parent_candidates(parent_invs)
434
head_set = self._heads(ie.file_id, parent_candidate_entries.keys())
436
for inv in parent_invs:
437
if inv.has_id(ie.file_id):
438
old_rev = inv[ie.file_id].revision
439
if old_rev in head_set:
440
heads.append(inv[ie.file_id].revision)
441
head_set.remove(inv[ie.file_id].revision)
444
# now we check to see if we need to write a new record to the
446
# We write a new entry unless there is one head to the ancestors, and
447
# the kind-derived content is unchanged.
449
# Cheapest check first: no ancestors, or more the one head in the
450
# ancestors, we write a new node.
454
# There is a single head, look it up for comparison
455
parent_entry = parent_candidate_entries[heads[0]]
456
# if the non-content specific data has changed, we'll be writing a
458
if (parent_entry.parent_id != ie.parent_id or
459
parent_entry.name != ie.name):
461
# now we need to do content specific checks:
463
# if the kind changed the content obviously has
464
if kind != parent_entry.kind:
466
# Stat cache fingerprint feedback for the caller - None as we usually
467
# don't generate one.
470
if content_summary[2] is None:
471
raise ValueError("Files must not have executable = None")
473
# We can't trust a check of the file length because of content
475
if (# if the exec bit has changed we have to store:
476
parent_entry.executable != content_summary[2]):
478
elif parent_entry.text_sha1 == content_summary[3]:
479
# all meta and content is unchanged (using a hash cache
480
# hit to check the sha)
481
ie.revision = parent_entry.revision
482
ie.text_size = parent_entry.text_size
483
ie.text_sha1 = parent_entry.text_sha1
484
ie.executable = parent_entry.executable
485
return self._get_delta(ie, basis_inv, path), False, None
487
# Either there is only a hash change(no hash cache entry,
488
# or same size content change), or there is no change on
490
# Provide the parent's hash to the store layer, so that the
491
# content is unchanged we will not store a new node.
492
nostore_sha = parent_entry.text_sha1
494
# We want to record a new node regardless of the presence or
495
# absence of a content change in the file.
497
ie.executable = content_summary[2]
498
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
500
text = file_obj.read()
504
ie.text_sha1, ie.text_size = self._add_text_to_weave(
505
ie.file_id, text, heads, nostore_sha)
506
# Let the caller know we generated a stat fingerprint.
507
fingerprint = (ie.text_sha1, stat_value)
508
except errors.ExistingContent:
509
# Turns out that the file content was unchanged, and we were
510
# only going to store a new node if it was changed. Carry over
512
ie.revision = parent_entry.revision
513
ie.text_size = parent_entry.text_size
514
ie.text_sha1 = parent_entry.text_sha1
515
ie.executable = parent_entry.executable
516
return self._get_delta(ie, basis_inv, path), False, None
517
elif kind == 'directory':
519
# all data is meta here, nothing specific to directory, so
521
ie.revision = parent_entry.revision
522
return self._get_delta(ie, basis_inv, path), False, None
523
self._add_text_to_weave(ie.file_id, '', heads, None)
524
elif kind == 'symlink':
525
current_link_target = content_summary[3]
527
# symlink target is not generic metadata, check if it has
529
if current_link_target != parent_entry.symlink_target:
532
# unchanged, carry over.
533
ie.revision = parent_entry.revision
534
ie.symlink_target = parent_entry.symlink_target
535
return self._get_delta(ie, basis_inv, path), False, None
536
ie.symlink_target = current_link_target
537
self._add_text_to_weave(ie.file_id, '', heads, None)
538
elif kind == 'tree-reference':
540
if content_summary[3] != parent_entry.reference_revision:
543
# unchanged, carry over.
544
ie.reference_revision = parent_entry.reference_revision
545
ie.revision = parent_entry.revision
546
return self._get_delta(ie, basis_inv, path), False, None
547
ie.reference_revision = content_summary[3]
548
if ie.reference_revision is None:
549
raise AssertionError("invalid content_summary for nested tree: %r"
550
% (content_summary,))
551
self._add_text_to_weave(ie.file_id, '', heads, None)
553
raise NotImplementedError('unknown kind')
554
ie.revision = self._new_revision_id
555
# The initial commit adds a root directory, but this in itself is not
556
# a worthwhile commit.
557
if (self.basis_delta_revision != _mod_revision.NULL_REVISION or
559
self._any_changes = True
560
return self._get_delta(ie, basis_inv, path), True, fingerprint
562
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
563
_entry_factory=entry_factory):
564
"""Record a new tree via iter_changes.
566
:param tree: The tree to obtain text contents from for changed objects.
567
:param basis_revision_id: The revision id of the tree the iter_changes
568
has been generated against. Currently assumed to be the same
569
as self.parents[0] - if it is not, errors may occur.
570
:param iter_changes: An iter_changes iterator with the changes to apply
571
to basis_revision_id. The iterator must not include any items with
572
a current kind of None - missing items must be either filtered out
573
or errored-on before record_iter_changes sees the item.
574
:param _entry_factory: Private method to bind entry_factory locally for
576
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
579
# Create an inventory delta based on deltas between all the parents and
580
# deltas between all the parent inventories. We use inventory delta's
581
# between the inventory objects because iter_changes masks
582
# last-changed-field only changes.
584
# file_id -> change map, change is fileid, paths, changed, versioneds,
585
# parents, names, kinds, executables
587
# {file_id -> revision_id -> inventory entry, for entries in parent
588
# trees that are not parents[0]
592
revtrees = list(self.repository.revision_trees(self.parents))
593
except errors.NoSuchRevision:
594
# one or more ghosts, slow path.
596
for revision_id in self.parents:
598
revtrees.append(self.repository.revision_tree(revision_id))
599
except errors.NoSuchRevision:
601
basis_revision_id = _mod_revision.NULL_REVISION
603
revtrees.append(self.repository.revision_tree(
604
_mod_revision.NULL_REVISION))
605
# The basis inventory from a repository
607
basis_tree = revtrees[0]
609
basis_tree = self.repository.revision_tree(
610
_mod_revision.NULL_REVISION)
611
basis_inv = basis_tree.root_inventory
612
if len(self.parents) > 0:
613
if basis_revision_id != self.parents[0] and not ghost_basis:
615
"arbitrary basis parents not yet supported with merges")
616
for revtree in revtrees[1:]:
617
for change in revtree.root_inventory._make_delta(basis_inv):
618
if change[1] is None:
619
# Not present in this parent.
621
if change[2] not in merged_ids:
622
if change[0] is not None:
623
basis_entry = basis_inv[change[2]]
624
merged_ids[change[2]] = [
626
basis_entry.revision,
629
parent_entries[change[2]] = {
631
basis_entry.revision:basis_entry,
633
change[3].revision:change[3],
636
merged_ids[change[2]] = [change[3].revision]
637
parent_entries[change[2]] = {change[3].revision:change[3]}
639
merged_ids[change[2]].append(change[3].revision)
640
parent_entries[change[2]][change[3].revision] = change[3]
643
# Setup the changes from the tree:
644
# changes maps file_id -> (change, [parent revision_ids])
646
for change in iter_changes:
647
# This probably looks up in basis_inv way to much.
648
if change[1][0] is not None:
649
head_candidate = [basis_inv[change[0]].revision]
652
changes[change[0]] = change, merged_ids.get(change[0],
654
unchanged_merged = set(merged_ids) - set(changes)
655
# Extend the changes dict with synthetic changes to record merges of
657
for file_id in unchanged_merged:
658
# Record a merged version of these items that did not change vs the
659
# basis. This can be either identical parallel changes, or a revert
660
# of a specific file after a merge. The recorded content will be
661
# that of the current tree (which is the same as the basis), but
662
# the per-file graph will reflect a merge.
663
# NB:XXX: We are reconstructing path information we had, this
664
# should be preserved instead.
665
# inv delta change: (file_id, (path_in_source, path_in_target),
666
# changed_content, versioned, parent, name, kind,
669
basis_entry = basis_inv[file_id]
670
except errors.NoSuchId:
671
# a change from basis->some_parents but file_id isn't in basis
672
# so was new in the merge, which means it must have changed
673
# from basis -> current, and as it hasn't the add was reverted
674
# by the user. So we discard this change.
678
(basis_inv.id2path(file_id), tree.id2path(file_id)),
680
(basis_entry.parent_id, basis_entry.parent_id),
681
(basis_entry.name, basis_entry.name),
682
(basis_entry.kind, basis_entry.kind),
683
(basis_entry.executable, basis_entry.executable))
684
changes[file_id] = (change, merged_ids[file_id])
685
# changes contains tuples with the change and a set of inventory
686
# candidates for the file.
688
# old_path, new_path, file_id, new_inventory_entry
689
seen_root = False # Is the root in the basis delta?
690
inv_delta = self._basis_delta
691
modified_rev = self._new_revision_id
692
for change, head_candidates in changes.values():
693
if change[3][1]: # versioned in target.
694
# Several things may be happening here:
695
# We may have a fork in the per-file graph
696
# - record a change with the content from tree
697
# We may have a change against < all trees
698
# - carry over the tree that hasn't changed
699
# We may have a change against all trees
700
# - record the change with the content from tree
703
entry = _entry_factory[kind](file_id, change[5][1],
705
head_set = self._heads(change[0], set(head_candidates))
708
for head_candidate in head_candidates:
709
if head_candidate in head_set:
710
heads.append(head_candidate)
711
head_set.remove(head_candidate)
714
# Could be a carry-over situation:
715
parent_entry_revs = parent_entries.get(file_id, None)
716
if parent_entry_revs:
717
parent_entry = parent_entry_revs.get(heads[0], None)
720
if parent_entry is None:
721
# The parent iter_changes was called against is the one
722
# that is the per-file head, so any change is relevant
723
# iter_changes is valid.
724
carry_over_possible = False
726
# could be a carry over situation
727
# A change against the basis may just indicate a merge,
728
# we need to check the content against the source of the
729
# merge to determine if it was changed after the merge
731
if (parent_entry.kind != entry.kind or
732
parent_entry.parent_id != entry.parent_id or
733
parent_entry.name != entry.name):
734
# Metadata common to all entries has changed
735
# against per-file parent
736
carry_over_possible = False
738
carry_over_possible = True
739
# per-type checks for changes against the parent_entry
742
# Cannot be a carry-over situation
743
carry_over_possible = False
744
# Populate the entry in the delta
746
# XXX: There is still a small race here: If someone reverts the content of a file
747
# after iter_changes examines and decides it has changed,
748
# we will unconditionally record a new version even if some
749
# other process reverts it while commit is running (with
750
# the revert happening after iter_changes did its
753
entry.executable = True
755
entry.executable = False
756
if (carry_over_possible and
757
parent_entry.executable == entry.executable):
758
# Check the file length, content hash after reading
760
nostore_sha = parent_entry.text_sha1
763
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
765
text = file_obj.read()
769
entry.text_sha1, entry.text_size = self._add_text_to_weave(
770
file_id, text, heads, nostore_sha)
771
yield file_id, change[1][1], (entry.text_sha1, stat_value)
772
except errors.ExistingContent:
773
# No content change against a carry_over parent
774
# Perhaps this should also yield a fs hash update?
776
entry.text_size = parent_entry.text_size
777
entry.text_sha1 = parent_entry.text_sha1
778
elif kind == 'symlink':
780
entry.symlink_target = tree.get_symlink_target(file_id)
781
if (carry_over_possible and
782
parent_entry.symlink_target == entry.symlink_target):
785
self._add_text_to_weave(change[0], '', heads, None)
786
elif kind == 'directory':
787
if carry_over_possible:
790
# Nothing to set on the entry.
791
# XXX: split into the Root and nonRoot versions.
792
if change[1][1] != '' or self.repository.supports_rich_root():
793
self._add_text_to_weave(change[0], '', heads, None)
794
elif kind == 'tree-reference':
795
if not self.repository._format.supports_tree_reference:
796
# This isn't quite sane as an error, but we shouldn't
797
# ever see this code path in practice: tree's don't
798
# permit references when the repo doesn't support tree
800
raise errors.UnsupportedOperation(tree.add_reference,
802
reference_revision = tree.get_reference_revision(change[0])
803
entry.reference_revision = reference_revision
804
if (carry_over_possible and
805
parent_entry.reference_revision == reference_revision):
808
self._add_text_to_weave(change[0], '', heads, None)
810
raise AssertionError('unknown kind %r' % kind)
812
entry.revision = modified_rev
814
entry.revision = parent_entry.revision
817
new_path = change[1][1]
818
inv_delta.append((change[1][0], new_path, change[0], entry))
821
self.new_inventory = None
822
# The initial commit adds a root directory, but this in itself is not
823
# a worthwhile commit.
824
if ((len(inv_delta) > 0 and basis_revision_id != _mod_revision.NULL_REVISION) or
825
(len(inv_delta) > 1 and basis_revision_id == _mod_revision.NULL_REVISION)):
826
# This should perhaps be guarded by a check that the basis we
827
# commit against is the basis for the commit and if not do a delta
829
self._any_changes = True
831
# housekeeping root entry changes do not affect no-change commits.
832
self._require_root_change(tree)
833
self.basis_delta_revision = basis_revision_id
835
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
836
parent_keys = tuple([(file_id, parent) for parent in parents])
837
return self.repository.texts._add_text(
838
(file_id, self._new_revision_id), parent_keys, new_text,
839
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
842
class VersionedFileRootCommitBuilder(VersionedFileCommitBuilder):
843
"""This commitbuilder actually records the root id"""
845
# the root entry gets versioned properly by this builder.
846
_versioned_root = True
848
def _check_root(self, ie, parent_invs, tree):
849
"""Helper for record_entry_contents.
851
:param ie: An entry being added.
852
:param parent_invs: The inventories of the parent revisions of the
854
:param tree: The tree that is being committed.
857
def _require_root_change(self, tree):
858
"""Enforce an appropriate root object change.
860
This is called once when record_iter_changes is called, if and only if
861
the root was not in the delta calculated by record_iter_changes.
863
:param tree: The tree which is being committed.
865
# versioned roots do not change unless the tree found a change.
868
class VersionedFileRepository(Repository):
869
"""Repository holding history for one or more branches.
871
The repository holds and retrieves historical information including
872
revisions and file history. It's normally accessed only by the Branch,
873
which views a particular line of development through that history.
875
The Repository builds on top of some byte storage facilies (the revisions,
876
signatures, inventories, texts and chk_bytes attributes) and a Transport,
877
which respectively provide byte storage and a means to access the (possibly
880
The byte storage facilities are addressed via tuples, which we refer to
881
as 'keys' throughout the code base. Revision_keys, inventory_keys and
882
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
883
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
884
byte string made up of a hash identifier and a hash value.
885
We use this interface because it allows low friction with the underlying
886
code that implements disk indices, network encoding and other parts of
889
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
890
the serialised revisions for the repository. This can be used to obtain
891
revision graph information or to access raw serialised revisions.
892
The result of trying to insert data into the repository via this store
893
is undefined: it should be considered read-only except for implementors
895
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
896
the serialised signatures for the repository. This can be used to
897
obtain access to raw serialised signatures. The result of trying to
898
insert data into the repository via this store is undefined: it should
899
be considered read-only except for implementors of repositories.
900
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
901
the serialised inventories for the repository. This can be used to
902
obtain unserialised inventories. The result of trying to insert data
903
into the repository via this store is undefined: it should be
904
considered read-only except for implementors of repositories.
905
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
906
texts of files and directories for the repository. This can be used to
907
obtain file texts or file graphs. Note that Repository.iter_file_bytes
908
is usually a better interface for accessing file texts.
909
The result of trying to insert data into the repository via this store
910
is undefined: it should be considered read-only except for implementors
912
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
913
any data the repository chooses to store or have indexed by its hash.
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 _transport: Transport for file access to repository, typically
918
pointing to .bzr/repository.
921
# What class to use for a CommitBuilder. Often it's simpler to change this
922
# in a Repository class subclass rather than to override
923
# get_commit_builder.
924
_commit_builder_class = VersionedFileCommitBuilder
926
def add_fallback_repository(self, repository):
927
"""Add a repository to use for looking up data not held locally.
929
:param repository: A repository.
931
if not self._format.supports_external_lookups:
932
raise errors.UnstackableRepositoryFormat(self._format, self.base)
933
# This can raise an exception, so should be done before we lock the
934
# fallback repository.
935
self._check_fallback_repository(repository)
937
# This repository will call fallback.unlock() when we transition to
938
# the unlocked state, so we make sure to increment the lock count
939
repository.lock_read()
940
self._fallback_repositories.append(repository)
941
self.texts.add_fallback_versioned_files(repository.texts)
942
self.inventories.add_fallback_versioned_files(repository.inventories)
943
self.revisions.add_fallback_versioned_files(repository.revisions)
944
self.signatures.add_fallback_versioned_files(repository.signatures)
945
if self.chk_bytes is not None:
946
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
948
@only_raises(errors.LockNotHeld, errors.LockBroken)
950
super(VersionedFileRepository, self).unlock()
951
if self.control_files._lock_count == 0:
952
self._inventory_entry_cache.clear()
954
def add_inventory(self, revision_id, inv, parents):
955
"""Add the inventory inv to the repository as revision_id.
957
:param parents: The revision ids of the parents that revision_id
958
is known to have and are in the repository already.
960
:returns: The validator(which is a sha1 digest, though what is sha'd is
961
repository format specific) of the serialized inventory.
963
if not self.is_in_write_group():
964
raise AssertionError("%r not in write group" % (self,))
965
_mod_revision.check_not_reserved_id(revision_id)
966
if not (inv.revision_id is None or inv.revision_id == revision_id):
967
raise AssertionError(
968
"Mismatch between inventory revision"
969
" id and insertion revid (%r, %r)"
970
% (inv.revision_id, revision_id))
972
raise errors.RootMissing()
973
return self._add_inventory_checked(revision_id, inv, parents)
975
def _add_inventory_checked(self, revision_id, inv, parents):
976
"""Add inv to the repository after checking the inputs.
978
This function can be overridden to allow different inventory styles.
980
:seealso: add_inventory, for the contract.
982
inv_lines = self._serializer.write_inventory_to_lines(inv)
983
return self._inventory_add_lines(revision_id, parents,
984
inv_lines, check_content=False)
986
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
987
parents, basis_inv=None, propagate_caches=False):
988
"""Add a new inventory expressed as a delta against another revision.
990
See the inventory developers documentation for the theory behind
993
:param basis_revision_id: The inventory id the delta was created
994
against. (This does not have to be a direct parent.)
995
:param delta: The inventory delta (see Inventory.apply_delta for
997
:param new_revision_id: The revision id that the inventory is being
999
:param parents: The revision ids of the parents that revision_id is
1000
known to have and are in the repository already. These are supplied
1001
for repositories that depend on the inventory graph for revision
1002
graph access, as well as for those that pun ancestry with delta
1004
:param basis_inv: The basis inventory if it is already known,
1006
:param propagate_caches: If True, the caches for this inventory are
1007
copied to and updated for the result if possible.
1009
:returns: (validator, new_inv)
1010
The validator(which is a sha1 digest, though what is sha'd is
1011
repository format specific) of the serialized inventory, and the
1012
resulting inventory.
1014
if not self.is_in_write_group():
1015
raise AssertionError("%r not in write group" % (self,))
1016
_mod_revision.check_not_reserved_id(new_revision_id)
1017
basis_tree = self.revision_tree(basis_revision_id)
1018
basis_tree.lock_read()
1020
# Note that this mutates the inventory of basis_tree, which not all
1021
# inventory implementations may support: A better idiom would be to
1022
# return a new inventory, but as there is no revision tree cache in
1023
# repository this is safe for now - RBC 20081013
1024
if basis_inv is None:
1025
basis_inv = basis_tree.root_inventory
1026
basis_inv.apply_delta(delta)
1027
basis_inv.revision_id = new_revision_id
1028
return (self.add_inventory(new_revision_id, basis_inv, parents),
1033
def _inventory_add_lines(self, revision_id, parents, lines,
1034
check_content=True):
1035
"""Store lines in inv_vf and return the sha1 of the inventory."""
1036
parents = [(parent,) for parent in parents]
1037
result = self.inventories.add_lines((revision_id,), parents, lines,
1038
check_content=check_content)[0]
1039
self.inventories._access.flush()
1042
def add_revision(self, revision_id, rev, inv=None):
1043
"""Add rev to the revision store as revision_id.
1045
:param revision_id: the revision id to use.
1046
:param rev: The revision object.
1047
:param inv: The inventory for the revision. if None, it will be looked
1048
up in the inventory storer
1050
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1052
_mod_revision.check_not_reserved_id(revision_id)
1053
# check inventory present
1054
if not self.inventories.get_parent_map([(revision_id,)]):
1056
raise errors.WeaveRevisionNotPresent(revision_id,
1059
# yes, this is not suitable for adding with ghosts.
1060
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1063
key = (revision_id,)
1064
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1065
self._add_revision(rev)
1067
def _add_revision(self, revision):
1068
text = self._serializer.write_revision_to_string(revision)
1069
key = (revision.revision_id,)
1070
parents = tuple((parent,) for parent in revision.parent_ids)
1071
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1073
def _check_inventories(self, checker):
1074
"""Check the inventories found from the revision scan.
1076
This is responsible for verifying the sha1 of inventories and
1077
creating a pending_keys set that covers data referenced by inventories.
1079
bar = ui.ui_factory.nested_progress_bar()
1081
self._do_check_inventories(checker, bar)
1085
def _do_check_inventories(self, checker, bar):
1086
"""Helper for _check_inventories."""
1088
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1089
kinds = ['chk_bytes', 'texts']
1090
count = len(checker.pending_keys)
1091
bar.update(gettext("inventories"), 0, 2)
1092
current_keys = checker.pending_keys
1093
checker.pending_keys = {}
1094
# Accumulate current checks.
1095
for key in current_keys:
1096
if key[0] != 'inventories' and key[0] not in kinds:
1097
checker._report_items.append('unknown key type %r' % (key,))
1098
keys[key[0]].add(key[1:])
1099
if keys['inventories']:
1100
# NB: output order *should* be roughly sorted - topo or
1101
# inverse topo depending on repository - either way decent
1102
# to just delta against. However, pre-CHK formats didn't
1103
# try to optimise inventory layout on disk. As such the
1104
# pre-CHK code path does not use inventory deltas.
1106
for record in self.inventories.check(keys=keys['inventories']):
1107
if record.storage_kind == 'absent':
1108
checker._report_items.append(
1109
'Missing inventory {%s}' % (record.key,))
1111
last_object = self._check_record('inventories', record,
1112
checker, last_object,
1113
current_keys[('inventories',) + record.key])
1114
del keys['inventories']
1117
bar.update(gettext("texts"), 1)
1118
while (checker.pending_keys or keys['chk_bytes']
1120
# Something to check.
1121
current_keys = checker.pending_keys
1122
checker.pending_keys = {}
1123
# Accumulate current checks.
1124
for key in current_keys:
1125
if key[0] not in kinds:
1126
checker._report_items.append('unknown key type %r' % (key,))
1127
keys[key[0]].add(key[1:])
1128
# Check the outermost kind only - inventories || chk_bytes || texts
1132
for record in getattr(self, kind).check(keys=keys[kind]):
1133
if record.storage_kind == 'absent':
1134
checker._report_items.append(
1135
'Missing %s {%s}' % (kind, record.key,))
1137
last_object = self._check_record(kind, record,
1138
checker, last_object, current_keys[(kind,) + record.key])
1142
def _check_record(self, kind, record, checker, last_object, item_data):
1143
"""Check a single text from this repository."""
1144
if kind == 'inventories':
1145
rev_id = record.key[0]
1146
inv = self._deserialise_inventory(rev_id,
1147
record.get_bytes_as('fulltext'))
1148
if last_object is not None:
1149
delta = inv._make_delta(last_object)
1150
for old_path, path, file_id, ie in delta:
1153
ie.check(checker, rev_id, inv)
1155
for path, ie in inv.iter_entries():
1156
ie.check(checker, rev_id, inv)
1157
if self._format.fast_deltas:
1159
elif kind == 'chk_bytes':
1160
# No code written to check chk_bytes for this repo format.
1161
checker._report_items.append(
1162
'unsupported key type chk_bytes for %s' % (record.key,))
1163
elif kind == 'texts':
1164
self._check_text(record, checker, item_data)
1166
checker._report_items.append(
1167
'unknown key type %s for %s' % (kind, record.key))
1169
def _check_text(self, record, checker, item_data):
1170
"""Check a single text."""
1171
# Check it is extractable.
1172
# TODO: check length.
1173
if record.storage_kind == 'chunked':
1174
chunks = record.get_bytes_as(record.storage_kind)
1175
sha1 = osutils.sha_strings(chunks)
1176
length = sum(map(len, chunks))
1178
content = record.get_bytes_as('fulltext')
1179
sha1 = osutils.sha_string(content)
1180
length = len(content)
1181
if item_data and sha1 != item_data[1]:
1182
checker._report_items.append(
1183
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1184
(record.key, sha1, item_data[1], item_data[2]))
1187
def _eliminate_revisions_not_present(self, revision_ids):
1188
"""Check every revision id in revision_ids to see if we have it.
1190
Returns a set of the present revisions.
1193
graph = self.get_graph()
1194
parent_map = graph.get_parent_map(revision_ids)
1195
# The old API returned a list, should this actually be a set?
1196
return parent_map.keys()
1198
def __init__(self, _format, a_bzrdir, control_files):
1199
"""Instantiate a VersionedFileRepository.
1201
:param _format: The format of the repository on disk.
1202
:param controldir: The ControlDir of the repository.
1203
:param control_files: Control files to use for locking, etc.
1205
# In the future we will have a single api for all stores for
1206
# getting file texts, inventories and revisions, then
1207
# this construct will accept instances of those things.
1208
super(VersionedFileRepository, self).__init__(_format, a_bzrdir,
1210
self._transport = control_files._transport
1211
self.base = self._transport.base
1213
self._reconcile_does_inventory_gc = True
1214
self._reconcile_fixes_text_parents = False
1215
self._reconcile_backsup_inventory = True
1216
# An InventoryEntry cache, used during deserialization
1217
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1218
# Is it safe to return inventory entries directly from the entry cache,
1219
# rather copying them?
1220
self._safe_to_return_from_cache = False
1222
def fetch(self, source, revision_id=None, find_ghosts=False,
1224
"""Fetch the content required to construct revision_id from source.
1226
If revision_id is None and fetch_spec is None, then all content is
1229
fetch() may not be used when the repository is in a write group -
1230
either finish the current write group before using fetch, or use
1231
fetch before starting the write group.
1233
:param find_ghosts: Find and copy revisions in the source that are
1234
ghosts in the target (and not reachable directly by walking out to
1235
the first-present revision in target from revision_id).
1236
:param revision_id: If specified, all the content needed for this
1237
revision ID will be copied to the target. Fetch will determine for
1238
itself which content needs to be copied.
1239
:param fetch_spec: If specified, a SearchResult or
1240
PendingAncestryResult that describes which revisions to copy. This
1241
allows copying multiple heads at once. Mutually exclusive with
1244
if fetch_spec is not None and revision_id is not None:
1245
raise AssertionError(
1246
"fetch_spec and revision_id are mutually exclusive.")
1247
if self.is_in_write_group():
1248
raise errors.InternalBzrError(
1249
"May not fetch while in a write group.")
1250
# fast path same-url fetch operations
1251
# TODO: lift out to somewhere common with RemoteRepository
1252
# <https://bugs.launchpad.net/bzr/+bug/401646>
1253
if (self.has_same_location(source)
1254
and fetch_spec is None
1255
and self._has_same_fallbacks(source)):
1256
# check that last_revision is in 'from' and then return a
1258
if (revision_id is not None and
1259
not _mod_revision.is_null(revision_id)):
1260
self.get_revision(revision_id)
1262
inter = InterRepository.get(source, self)
1263
if (fetch_spec is not None and
1264
not getattr(inter, "supports_fetch_spec", False)):
1265
raise errors.UnsupportedOperation(
1266
"fetch_spec not supported for %r" % inter)
1267
return inter.fetch(revision_id=revision_id,
1268
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1271
def gather_stats(self, revid=None, committers=None):
1272
"""See Repository.gather_stats()."""
1273
result = super(VersionedFileRepository, self).gather_stats(revid, committers)
1274
# now gather global repository information
1275
# XXX: This is available for many repos regardless of listability.
1276
if self.user_transport.listable():
1277
# XXX: do we want to __define len__() ?
1278
# Maybe the versionedfiles object should provide a different
1279
# method to get the number of keys.
1280
result['revisions'] = len(self.revisions.keys())
1281
# result['size'] = t
1284
def get_commit_builder(self, branch, parents, config_stack, timestamp=None,
1285
timezone=None, committer=None, revprops=None,
1286
revision_id=None, lossy=False):
1287
"""Obtain a CommitBuilder for this repository.
1289
:param branch: Branch to commit to.
1290
:param parents: Revision ids of the parents of the new revision.
1291
:param config_stack: Configuration stack to use.
1292
:param timestamp: Optional timestamp recorded for commit.
1293
:param timezone: Optional timezone for timestamp.
1294
:param committer: Optional committer to set for commit.
1295
:param revprops: Optional dictionary of revision properties.
1296
:param revision_id: Optional revision id.
1297
:param lossy: Whether to discard data that can not be natively
1298
represented, when pushing to a foreign VCS
1300
if self._fallback_repositories and not self._format.supports_chks:
1301
raise errors.BzrError("Cannot commit directly to a stacked branch"
1302
" in pre-2a formats. See "
1303
"https://bugs.launchpad.net/bzr/+bug/375013 for details.")
1304
result = self._commit_builder_class(self, parents, config_stack,
1305
timestamp, timezone, committer, revprops, revision_id,
1307
self.start_write_group()
1310
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1311
"""Return the keys of missing inventory parents for revisions added in
1314
A revision is not complete if the inventory delta for that revision
1315
cannot be calculated. Therefore if the parent inventories of a
1316
revision are not present, the revision is incomplete, and e.g. cannot
1317
be streamed by a smart server. This method finds missing inventory
1318
parents for revisions added in this write group.
1320
if not self._format.supports_external_lookups:
1321
# This is only an issue for stacked repositories
1323
if not self.is_in_write_group():
1324
raise AssertionError('not in a write group')
1326
# XXX: We assume that every added revision already has its
1327
# corresponding inventory, so we only check for parent inventories that
1328
# might be missing, rather than all inventories.
1329
parents = set(self.revisions._index.get_missing_parents())
1330
parents.discard(_mod_revision.NULL_REVISION)
1331
unstacked_inventories = self.inventories._index
1332
present_inventories = unstacked_inventories.get_parent_map(
1333
key[-1:] for key in parents)
1334
parents.difference_update(present_inventories)
1335
if len(parents) == 0:
1336
# No missing parent inventories.
1338
if not check_for_missing_texts:
1339
return set(('inventories', rev_id) for (rev_id,) in parents)
1340
# Ok, now we have a list of missing inventories. But these only matter
1341
# if the inventories that reference them are missing some texts they
1342
# appear to introduce.
1343
# XXX: Texts referenced by all added inventories need to be present,
1344
# but at the moment we're only checking for texts referenced by
1345
# inventories at the graph's edge.
1346
key_deps = self.revisions._index._key_dependencies
1347
key_deps.satisfy_refs_for_keys(present_inventories)
1348
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1349
file_ids = self.fileids_altered_by_revision_ids(referrers)
1350
missing_texts = set()
1351
for file_id, version_ids in file_ids.iteritems():
1352
missing_texts.update(
1353
(file_id, version_id) for version_id in version_ids)
1354
present_texts = self.texts.get_parent_map(missing_texts)
1355
missing_texts.difference_update(present_texts)
1356
if not missing_texts:
1357
# No texts are missing, so all revisions and their deltas are
1360
# Alternatively the text versions could be returned as the missing
1361
# keys, but this is likely to be less data.
1362
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1366
def has_revisions(self, revision_ids):
1367
"""Probe to find out the presence of multiple revisions.
1369
:param revision_ids: An iterable of revision_ids.
1370
:return: A set of the revision_ids that were present.
1372
parent_map = self.revisions.get_parent_map(
1373
[(rev_id,) for rev_id in revision_ids])
1375
if _mod_revision.NULL_REVISION in revision_ids:
1376
result.add(_mod_revision.NULL_REVISION)
1377
result.update([key[0] for key in parent_map])
1381
def get_revision_reconcile(self, revision_id):
1382
"""'reconcile' helper routine that allows access to a revision always.
1384
This variant of get_revision does not cross check the weave graph
1385
against the revision one as get_revision does: but it should only
1386
be used by reconcile, or reconcile-alike commands that are correcting
1387
or testing the revision graph.
1389
return self._get_revisions([revision_id])[0]
1392
def get_revisions(self, revision_ids):
1393
"""Get many revisions at once.
1395
Repositories that need to check data on every revision read should
1396
subclass this method.
1398
return self._get_revisions(revision_ids)
1401
def _get_revisions(self, revision_ids):
1402
"""Core work logic to get many revisions without sanity checks."""
1404
for revid, rev in self._iter_revisions(revision_ids):
1406
raise errors.NoSuchRevision(self, revid)
1408
return [revs[revid] for revid in revision_ids]
1410
def _iter_revisions(self, revision_ids):
1411
"""Iterate over revision objects.
1413
:param revision_ids: An iterable of revisions to examine. None may be
1414
passed to request all revisions known to the repository. Note that
1415
not all repositories can find unreferenced revisions; for those
1416
repositories only referenced ones will be returned.
1417
:return: An iterator of (revid, revision) tuples. Absent revisions (
1418
those asked for but not available) are returned as (revid, None).
1420
if revision_ids is None:
1421
revision_ids = self.all_revision_ids()
1423
for rev_id in revision_ids:
1424
if not rev_id or not isinstance(rev_id, basestring):
1425
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1426
keys = [(key,) for key in revision_ids]
1427
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1428
for record in stream:
1429
revid = record.key[0]
1430
if record.storage_kind == 'absent':
1433
text = record.get_bytes_as('fulltext')
1434
rev = self._serializer.read_revision_from_string(text)
1438
def add_signature_text(self, revision_id, signature):
1439
"""Store a signature text for a revision.
1441
:param revision_id: Revision id of the revision
1442
:param signature: Signature text.
1444
self.signatures.add_lines((revision_id,), (),
1445
osutils.split_lines(signature))
1447
def find_text_key_references(self):
1448
"""Find the text key references within the repository.
1450
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1451
to whether they were referred to by the inventory of the
1452
revision_id that they contain. The inventory texts from all present
1453
revision ids are assessed to generate this report.
1455
revision_keys = self.revisions.keys()
1456
w = self.inventories
1457
pb = ui.ui_factory.nested_progress_bar()
1459
return self._serializer._find_text_key_references(
1460
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1464
def _inventory_xml_lines_for_keys(self, keys):
1465
"""Get a line iterator of the sort needed for findind references.
1467
Not relevant for non-xml inventory repositories.
1469
Ghosts in revision_keys are ignored.
1471
:param revision_keys: The revision keys for the inventories to inspect.
1472
:return: An iterator over (inventory line, revid) for the fulltexts of
1473
all of the xml inventories specified by revision_keys.
1475
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1476
for record in stream:
1477
if record.storage_kind != 'absent':
1478
chunks = record.get_bytes_as('chunked')
1479
revid = record.key[-1]
1480
lines = osutils.chunks_to_lines(chunks)
1484
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1486
"""Helper routine for fileids_altered_by_revision_ids.
1488
This performs the translation of xml lines to revision ids.
1490
:param line_iterator: An iterator of lines, origin_version_id
1491
:param revision_keys: The revision ids to filter for. This should be a
1492
set or other type which supports efficient __contains__ lookups, as
1493
the revision key from each parsed line will be looked up in the
1494
revision_keys filter.
1495
:return: a dictionary mapping altered file-ids to an iterable of
1496
revision_ids. Each altered file-ids has the exact revision_ids that
1497
altered it listed explicitly.
1499
seen = set(self._serializer._find_text_key_references(
1500
line_iterator).iterkeys())
1501
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
1502
parent_seen = set(self._serializer._find_text_key_references(
1503
self._inventory_xml_lines_for_keys(parent_keys)))
1504
new_keys = seen - parent_seen
1506
setdefault = result.setdefault
1507
for key in new_keys:
1508
setdefault(key[0], set()).add(key[-1])
1511
def _find_parent_keys_of_revisions(self, revision_keys):
1512
"""Similar to _find_parent_ids_of_revisions, but used with keys.
1514
:param revision_keys: An iterable of revision_keys.
1515
:return: The parents of all revision_keys that are not already in
1518
parent_map = self.revisions.get_parent_map(revision_keys)
1520
map(parent_keys.update, parent_map.itervalues())
1521
parent_keys.difference_update(revision_keys)
1522
parent_keys.discard(_mod_revision.NULL_REVISION)
1525
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
1526
"""Find the file ids and versions affected by revisions.
1528
:param revisions: an iterable containing revision ids.
1529
:param _inv_weave: The inventory weave from this repository or None.
1530
If None, the inventory weave will be opened automatically.
1531
:return: a dictionary mapping altered file-ids to an iterable of
1532
revision_ids. Each altered file-ids has the exact revision_ids that
1533
altered it listed explicitly.
1535
selected_keys = set((revid,) for revid in revision_ids)
1536
w = _inv_weave or self.inventories
1537
return self._find_file_ids_from_xml_inventory_lines(
1538
w.iter_lines_added_or_present_in_keys(
1539
selected_keys, pb=None),
1542
def iter_files_bytes(self, desired_files):
1543
"""Iterate through file versions.
1545
Files will not necessarily be returned in the order they occur in
1546
desired_files. No specific order is guaranteed.
1548
Yields pairs of identifier, bytes_iterator. identifier is an opaque
1549
value supplied by the caller as part of desired_files. It should
1550
uniquely identify the file version in the caller's context. (Examples:
1551
an index number or a TreeTransform trans_id.)
1553
bytes_iterator is an iterable of bytestrings for the file. The
1554
kind of iterable and length of the bytestrings are unspecified, but for
1555
this implementation, it is a list of bytes produced by
1556
VersionedFile.get_record_stream().
1558
:param desired_files: a list of (file_id, revision_id, identifier)
1562
for file_id, revision_id, callable_data in desired_files:
1563
text_keys[(file_id, revision_id)] = callable_data
1564
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
1565
if record.storage_kind == 'absent':
1566
raise errors.RevisionNotPresent(record.key[1], record.key[0])
1567
yield text_keys[record.key], record.get_bytes_as('chunked')
1569
def _generate_text_key_index(self, text_key_references=None,
1571
"""Generate a new text key index for the repository.
1573
This is an expensive function that will take considerable time to run.
1575
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
1576
list of parents, also text keys. When a given key has no parents,
1577
the parents list will be [NULL_REVISION].
1579
# All revisions, to find inventory parents.
1580
if ancestors is None:
1581
graph = self.get_graph()
1582
ancestors = graph.get_parent_map(self.all_revision_ids())
1583
if text_key_references is None:
1584
text_key_references = self.find_text_key_references()
1585
pb = ui.ui_factory.nested_progress_bar()
1587
return self._do_generate_text_key_index(ancestors,
1588
text_key_references, pb)
1592
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
1593
"""Helper for _generate_text_key_index to avoid deep nesting."""
1594
revision_order = tsort.topo_sort(ancestors)
1595
invalid_keys = set()
1597
for revision_id in revision_order:
1598
revision_keys[revision_id] = set()
1599
text_count = len(text_key_references)
1600
# a cache of the text keys to allow reuse; costs a dict of all the
1601
# keys, but saves a 2-tuple for every child of a given key.
1603
for text_key, valid in text_key_references.iteritems():
1605
invalid_keys.add(text_key)
1607
revision_keys[text_key[1]].add(text_key)
1608
text_key_cache[text_key] = text_key
1609
del text_key_references
1611
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
1612
NULL_REVISION = _mod_revision.NULL_REVISION
1613
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
1614
# too small for large or very branchy trees. However, for 55K path
1615
# trees, it would be easy to use too much memory trivially. Ideally we
1616
# could gauge this by looking at available real memory etc, but this is
1617
# always a tricky proposition.
1618
inventory_cache = lru_cache.LRUCache(10)
1619
batch_size = 10 # should be ~150MB on a 55K path tree
1620
batch_count = len(revision_order) / batch_size + 1
1622
pb.update(gettext("Calculating text parents"), processed_texts, text_count)
1623
for offset in xrange(batch_count):
1624
to_query = revision_order[offset * batch_size:(offset + 1) *
1628
for revision_id in to_query:
1629
parent_ids = ancestors[revision_id]
1630
for text_key in revision_keys[revision_id]:
1631
pb.update(gettext("Calculating text parents"), processed_texts)
1632
processed_texts += 1
1633
candidate_parents = []
1634
for parent_id in parent_ids:
1635
parent_text_key = (text_key[0], parent_id)
1637
check_parent = parent_text_key not in \
1638
revision_keys[parent_id]
1640
# the parent parent_id is a ghost:
1641
check_parent = False
1642
# truncate the derived graph against this ghost.
1643
parent_text_key = None
1645
# look at the parent commit details inventories to
1646
# determine possible candidates in the per file graph.
1649
inv = inventory_cache[parent_id]
1651
inv = self.revision_tree(parent_id).root_inventory
1652
inventory_cache[parent_id] = inv
1654
parent_entry = inv[text_key[0]]
1655
except (KeyError, errors.NoSuchId):
1657
if parent_entry is not None:
1659
text_key[0], parent_entry.revision)
1661
parent_text_key = None
1662
if parent_text_key is not None:
1663
candidate_parents.append(
1664
text_key_cache[parent_text_key])
1665
parent_heads = text_graph.heads(candidate_parents)
1666
new_parents = list(parent_heads)
1667
new_parents.sort(key=lambda x:candidate_parents.index(x))
1668
if new_parents == []:
1669
new_parents = [NULL_REVISION]
1670
text_index[text_key] = new_parents
1672
for text_key in invalid_keys:
1673
text_index[text_key] = [NULL_REVISION]
1676
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
1677
"""Get an iterable listing the keys of all the data introduced by a set
1680
The keys will be ordered so that the corresponding items can be safely
1681
fetched and inserted in that order.
1683
:returns: An iterable producing tuples of (knit-kind, file-id,
1684
versions). knit-kind is one of 'file', 'inventory', 'signatures',
1685
'revisions'. file-id is None unless knit-kind is 'file'.
1687
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
1690
for result in self._find_non_file_keys_to_fetch(revision_ids):
1693
def _find_file_keys_to_fetch(self, revision_ids, pb):
1694
# XXX: it's a bit weird to control the inventory weave caching in this
1695
# generator. Ideally the caching would be done in fetch.py I think. Or
1696
# maybe this generator should explicitly have the contract that it
1697
# should not be iterated until the previously yielded item has been
1699
inv_w = self.inventories
1701
# file ids that changed
1702
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
1704
num_file_ids = len(file_ids)
1705
for file_id, altered_versions in file_ids.iteritems():
1707
pb.update(gettext("Fetch texts"), count, num_file_ids)
1709
yield ("file", file_id, altered_versions)
1711
def _find_non_file_keys_to_fetch(self, revision_ids):
1713
yield ("inventory", None, revision_ids)
1716
# XXX: Note ATM no callers actually pay attention to this return
1717
# instead they just use the list of revision ids and ignore
1718
# missing sigs. Consider removing this work entirely
1719
revisions_with_signatures = set(self.signatures.get_parent_map(
1720
[(r,) for r in revision_ids]))
1721
revisions_with_signatures = set(
1722
[r for (r,) in revisions_with_signatures])
1723
revisions_with_signatures.intersection_update(revision_ids)
1724
yield ("signatures", None, revisions_with_signatures)
1727
yield ("revisions", None, revision_ids)
1730
def get_inventory(self, revision_id):
1731
"""Get Inventory object by revision id."""
1732
return self.iter_inventories([revision_id]).next()
1734
def iter_inventories(self, revision_ids, ordering=None):
1735
"""Get many inventories by revision_ids.
1737
This will buffer some or all of the texts used in constructing the
1738
inventories in memory, but will only parse a single inventory at a
1741
:param revision_ids: The expected revision ids of the inventories.
1742
:param ordering: optional ordering, e.g. 'topological'. If not
1743
specified, the order of revision_ids will be preserved (by
1744
buffering if necessary).
1745
:return: An iterator of inventories.
1747
if ((None in revision_ids)
1748
or (_mod_revision.NULL_REVISION in revision_ids)):
1749
raise ValueError('cannot get null revision inventory')
1750
for inv, revid in self._iter_inventories(revision_ids, ordering):
1752
raise errors.NoSuchRevision(self, revid)
1755
def _iter_inventories(self, revision_ids, ordering):
1756
"""single-document based inventory iteration."""
1757
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
1758
for text, revision_id in inv_xmls:
1760
yield None, revision_id
1762
yield self._deserialise_inventory(revision_id, text), revision_id
1764
def _iter_inventory_xmls(self, revision_ids, ordering):
1765
if ordering is None:
1766
order_as_requested = True
1767
ordering = 'unordered'
1769
order_as_requested = False
1770
keys = [(revision_id,) for revision_id in revision_ids]
1773
if order_as_requested:
1774
key_iter = iter(keys)
1775
next_key = key_iter.next()
1776
stream = self.inventories.get_record_stream(keys, ordering, True)
1778
for record in stream:
1779
if record.storage_kind != 'absent':
1780
chunks = record.get_bytes_as('chunked')
1781
if order_as_requested:
1782
text_chunks[record.key] = chunks
1784
yield ''.join(chunks), record.key[-1]
1786
yield None, record.key[-1]
1787
if order_as_requested:
1788
# Yield as many results as we can while preserving order.
1789
while next_key in text_chunks:
1790
chunks = text_chunks.pop(next_key)
1791
yield ''.join(chunks), next_key[-1]
1793
next_key = key_iter.next()
1794
except StopIteration:
1795
# We still want to fully consume the get_record_stream,
1796
# just in case it is not actually finished at this point
1800
def _deserialise_inventory(self, revision_id, xml):
1801
"""Transform the xml into an inventory object.
1803
:param revision_id: The expected revision id of the inventory.
1804
:param xml: A serialised inventory.
1806
result = self._serializer.read_inventory_from_string(xml, revision_id,
1807
entry_cache=self._inventory_entry_cache,
1808
return_from_cache=self._safe_to_return_from_cache)
1809
if result.revision_id != revision_id:
1810
raise AssertionError('revision id mismatch %s != %s' % (
1811
result.revision_id, revision_id))
1814
def get_serializer_format(self):
1815
return self._serializer.format_num
1818
def _get_inventory_xml(self, revision_id):
1819
"""Get serialized inventory as a string."""
1820
texts = self._iter_inventory_xmls([revision_id], 'unordered')
1821
text, revision_id = texts.next()
1823
raise errors.NoSuchRevision(self, revision_id)
1827
def revision_tree(self, revision_id):
1828
"""Return Tree for a revision on this branch.
1830
`revision_id` may be NULL_REVISION for the empty tree revision.
1832
revision_id = _mod_revision.ensure_null(revision_id)
1833
# TODO: refactor this to use an existing revision object
1834
# so we don't need to read it in twice.
1835
if revision_id == _mod_revision.NULL_REVISION:
1836
return InventoryRevisionTree(self,
1837
Inventory(root_id=None), _mod_revision.NULL_REVISION)
1839
inv = self.get_inventory(revision_id)
1840
return InventoryRevisionTree(self, inv, revision_id)
1842
def revision_trees(self, revision_ids):
1843
"""Return Trees for revisions in this repository.
1845
:param revision_ids: a sequence of revision-ids;
1846
a revision-id may not be None or 'null:'
1848
inventories = self.iter_inventories(revision_ids)
1849
for inv in inventories:
1850
yield InventoryRevisionTree(self, inv, inv.revision_id)
1852
def _filtered_revision_trees(self, revision_ids, file_ids):
1853
"""Return Tree for a revision on this branch with only some files.
1855
:param revision_ids: a sequence of revision-ids;
1856
a revision-id may not be None or 'null:'
1857
:param file_ids: if not None, the result is filtered
1858
so that only those file-ids, their parents and their
1859
children are included.
1861
inventories = self.iter_inventories(revision_ids)
1862
for inv in inventories:
1863
# Should we introduce a FilteredRevisionTree class rather
1864
# than pre-filter the inventory here?
1865
filtered_inv = inv.filter(file_ids)
1866
yield InventoryRevisionTree(self, filtered_inv, filtered_inv.revision_id)
1868
def get_parent_map(self, revision_ids):
1869
"""See graph.StackedParentsProvider.get_parent_map"""
1870
# revisions index works in keys; this just works in revisions
1871
# therefore wrap and unwrap
1874
for revision_id in revision_ids:
1875
if revision_id == _mod_revision.NULL_REVISION:
1876
result[revision_id] = ()
1877
elif revision_id is None:
1878
raise ValueError('get_parent_map(None) is not valid')
1880
query_keys.append((revision_id ,))
1881
for ((revision_id,), parent_keys) in \
1882
self.revisions.get_parent_map(query_keys).iteritems():
1884
result[revision_id] = tuple([parent_revid
1885
for (parent_revid,) in parent_keys])
1887
result[revision_id] = (_mod_revision.NULL_REVISION,)
1891
def get_known_graph_ancestry(self, revision_ids):
1892
"""Return the known graph for a set of revision ids and their ancestors.
1894
st = static_tuple.StaticTuple
1895
revision_keys = [st(r_id).intern() for r_id in revision_ids]
1896
known_graph = self.revisions.get_known_graph_ancestry(revision_keys)
1897
return graph.GraphThunkIdsToKeys(known_graph)
1900
def get_file_graph(self):
1901
"""Return the graph walker for text revisions."""
1902
return graph.Graph(self.texts)
1904
def revision_ids_to_search_result(self, result_set):
1905
"""Convert a set of revision ids to a graph SearchResult."""
1906
result_parents = set()
1907
for parents in self.get_graph().get_parent_map(
1908
result_set).itervalues():
1909
result_parents.update(parents)
1910
included_keys = result_set.intersection(result_parents)
1911
start_keys = result_set.difference(included_keys)
1912
exclude_keys = result_parents.difference(result_set)
1913
result = vf_search.SearchResult(start_keys, exclude_keys,
1914
len(result_set), result_set)
1917
def _get_versioned_file_checker(self, text_key_references=None,
1919
"""Return an object suitable for checking versioned files.
1921
:param text_key_references: if non-None, an already built
1922
dictionary mapping text keys ((fileid, revision_id) tuples)
1923
to whether they were referred to by the inventory of the
1924
revision_id that they contain. If None, this will be
1926
:param ancestors: Optional result from
1927
self.get_graph().get_parent_map(self.all_revision_ids()) if already
1930
return _VersionedFileChecker(self,
1931
text_key_references=text_key_references, ancestors=ancestors)
1934
def has_signature_for_revision_id(self, revision_id):
1935
"""Query for a revision signature for revision_id in the repository."""
1936
if not self.has_revision(revision_id):
1937
raise errors.NoSuchRevision(self, revision_id)
1938
sig_present = (1 == len(
1939
self.signatures.get_parent_map([(revision_id,)])))
1943
def get_signature_text(self, revision_id):
1944
"""Return the text for a signature."""
1945
stream = self.signatures.get_record_stream([(revision_id,)],
1947
record = stream.next()
1948
if record.storage_kind == 'absent':
1949
raise errors.NoSuchRevision(self, revision_id)
1950
return record.get_bytes_as('fulltext')
1953
def _check(self, revision_ids, callback_refs, check_repo):
1954
result = check.VersionedFileCheck(self, check_repo=check_repo)
1955
result.check(callback_refs)
1958
def _find_inconsistent_revision_parents(self, revisions_iterator=None):
1959
"""Find revisions with different parent lists in the revision object
1960
and in the index graph.
1962
:param revisions_iterator: None, or an iterator of (revid,
1963
Revision-or-None). This iterator controls the revisions checked.
1964
:returns: an iterator yielding tuples of (revison-id, parents-in-index,
1965
parents-in-revision).
1967
if not self.is_locked():
1968
raise AssertionError()
1970
if revisions_iterator is None:
1971
revisions_iterator = self._iter_revisions(None)
1972
for revid, revision in revisions_iterator:
1973
if revision is None:
1975
parent_map = vf.get_parent_map([(revid,)])
1976
parents_according_to_index = tuple(parent[-1] for parent in
1977
parent_map[(revid,)])
1978
parents_according_to_revision = tuple(revision.parent_ids)
1979
if parents_according_to_index != parents_according_to_revision:
1980
yield (revid, parents_according_to_index,
1981
parents_according_to_revision)
1983
def _check_for_inconsistent_revision_parents(self):
1984
inconsistencies = list(self._find_inconsistent_revision_parents())
1986
raise errors.BzrCheckError(
1987
"Revision knit has inconsistent parents.")
1989
def _get_sink(self):
1990
"""Return a sink for streaming into this repository."""
1991
return StreamSink(self)
1993
def _get_source(self, to_format):
1994
"""Return a source for streaming from this repository."""
1995
return StreamSource(self, to_format)
1998
class MetaDirVersionedFileRepository(MetaDirRepository,
1999
VersionedFileRepository):
2000
"""Repositories in a meta-dir, that work via versioned file objects."""
2002
def __init__(self, _format, a_bzrdir, control_files):
2003
super(MetaDirVersionedFileRepository, self).__init__(_format, a_bzrdir,
2007
class MetaDirVersionedFileRepositoryFormat(RepositoryFormatMetaDir,
2008
VersionedFileRepositoryFormat):
2009
"""Base class for repository formats using versioned files in metadirs."""
2012
class StreamSink(object):
2013
"""An object that can insert a stream into a repository.
2015
This interface handles the complexity of reserialising inventories and
2016
revisions from different formats, and allows unidirectional insertion into
2017
stacked repositories without looking for the missing basis parents
2021
def __init__(self, target_repo):
2022
self.target_repo = target_repo
2024
def insert_stream(self, stream, src_format, resume_tokens):
2025
"""Insert a stream's content into the target repository.
2027
:param src_format: a bzr repository format.
2029
:return: a list of resume tokens and an iterable of keys additional
2030
items required before the insertion can be completed.
2032
self.target_repo.lock_write()
2035
self.target_repo.resume_write_group(resume_tokens)
2038
self.target_repo.start_write_group()
2041
# locked_insert_stream performs a commit|suspend.
2042
missing_keys = self.insert_stream_without_locking(stream,
2043
src_format, is_resume)
2045
# suspend the write group and tell the caller what we is
2046
# missing. We know we can suspend or else we would not have
2047
# entered this code path. (All repositories that can handle
2048
# missing keys can handle suspending a write group).
2049
write_group_tokens = self.target_repo.suspend_write_group()
2050
return write_group_tokens, missing_keys
2051
hint = self.target_repo.commit_write_group()
2052
to_serializer = self.target_repo._format._serializer
2053
src_serializer = src_format._serializer
2054
if (to_serializer != src_serializer and
2055
self.target_repo._format.pack_compresses):
2056
self.target_repo.pack(hint=hint)
2059
self.target_repo.abort_write_group(suppress_errors=True)
2062
self.target_repo.unlock()
2064
def insert_stream_without_locking(self, stream, src_format,
2066
"""Insert a stream's content into the target repository.
2068
This assumes that you already have a locked repository and an active
2071
:param src_format: a bzr repository format.
2072
:param is_resume: Passed down to get_missing_parent_inventories to
2073
indicate if we should be checking for missing texts at the same
2076
:return: A set of keys that are missing.
2078
if not self.target_repo.is_write_locked():
2079
raise errors.ObjectNotLocked(self)
2080
if not self.target_repo.is_in_write_group():
2081
raise errors.BzrError('you must already be in a write group')
2082
to_serializer = self.target_repo._format._serializer
2083
src_serializer = src_format._serializer
2085
if to_serializer == src_serializer:
2086
# If serializers match and the target is a pack repository, set the
2087
# write cache size on the new pack. This avoids poor performance
2088
# on transports where append is unbuffered (such as
2089
# RemoteTransport). This is safe to do because nothing should read
2090
# back from the target repository while a stream with matching
2091
# serialization is being inserted.
2092
# The exception is that a delta record from the source that should
2093
# be a fulltext may need to be expanded by the target (see
2094
# test_fetch_revisions_with_deltas_into_pack); but we take care to
2095
# explicitly flush any buffered writes first in that rare case.
2097
new_pack = self.target_repo._pack_collection._new_pack
2098
except AttributeError:
2099
# Not a pack repository
2102
new_pack.set_write_cache_size(1024*1024)
2103
for substream_type, substream in stream:
2104
if 'stream' in debug.debug_flags:
2105
mutter('inserting substream: %s', substream_type)
2106
if substream_type == 'texts':
2107
self.target_repo.texts.insert_record_stream(substream)
2108
elif substream_type == 'inventories':
2109
if src_serializer == to_serializer:
2110
self.target_repo.inventories.insert_record_stream(
2113
self._extract_and_insert_inventories(
2114
substream, src_serializer)
2115
elif substream_type == 'inventory-deltas':
2116
self._extract_and_insert_inventory_deltas(
2117
substream, src_serializer)
2118
elif substream_type == 'chk_bytes':
2119
# XXX: This doesn't support conversions, as it assumes the
2120
# conversion was done in the fetch code.
2121
self.target_repo.chk_bytes.insert_record_stream(substream)
2122
elif substream_type == 'revisions':
2123
# This may fallback to extract-and-insert more often than
2124
# required if the serializers are different only in terms of
2126
if src_serializer == to_serializer:
2127
self.target_repo.revisions.insert_record_stream(substream)
2129
self._extract_and_insert_revisions(substream,
2131
elif substream_type == 'signatures':
2132
self.target_repo.signatures.insert_record_stream(substream)
2134
raise AssertionError('kaboom! %s' % (substream_type,))
2135
# Done inserting data, and the missing_keys calculations will try to
2136
# read back from the inserted data, so flush the writes to the new pack
2137
# (if this is pack format).
2138
if new_pack is not None:
2139
new_pack._write_data('', flush=True)
2140
# Find all the new revisions (including ones from resume_tokens)
2141
missing_keys = self.target_repo.get_missing_parent_inventories(
2142
check_for_missing_texts=is_resume)
2144
for prefix, versioned_file in (
2145
('texts', self.target_repo.texts),
2146
('inventories', self.target_repo.inventories),
2147
('revisions', self.target_repo.revisions),
2148
('signatures', self.target_repo.signatures),
2149
('chk_bytes', self.target_repo.chk_bytes),
2151
if versioned_file is None:
2153
# TODO: key is often going to be a StaticTuple object
2154
# I don't believe we can define a method by which
2155
# (prefix,) + StaticTuple will work, though we could
2156
# define a StaticTuple.sq_concat that would allow you to
2157
# pass in either a tuple or a StaticTuple as the second
2158
# object, so instead we could have:
2159
# StaticTuple(prefix) + key here...
2160
missing_keys.update((prefix,) + key for key in
2161
versioned_file.get_missing_compression_parent_keys())
2162
except NotImplementedError:
2163
# cannot even attempt suspending, and missing would have failed
2164
# during stream insertion.
2165
missing_keys = set()
2168
def _extract_and_insert_inventory_deltas(self, substream, serializer):
2169
target_rich_root = self.target_repo._format.rich_root_data
2170
target_tree_refs = self.target_repo._format.supports_tree_reference
2171
for record in substream:
2172
# Insert the delta directly
2173
inventory_delta_bytes = record.get_bytes_as('fulltext')
2174
deserialiser = inventory_delta.InventoryDeltaDeserializer()
2176
parse_result = deserialiser.parse_text_bytes(
2177
inventory_delta_bytes)
2178
except inventory_delta.IncompatibleInventoryDelta, err:
2179
mutter("Incompatible delta: %s", err.msg)
2180
raise errors.IncompatibleRevision(self.target_repo._format)
2181
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
2182
revision_id = new_id
2183
parents = [key[0] for key in record.parents]
2184
self.target_repo.add_inventory_by_delta(
2185
basis_id, inv_delta, revision_id, parents)
2187
def _extract_and_insert_inventories(self, substream, serializer,
2189
"""Generate a new inventory versionedfile in target, converting data.
2191
The inventory is retrieved from the source, (deserializing it), and
2192
stored in the target (reserializing it in a different format).
2194
target_rich_root = self.target_repo._format.rich_root_data
2195
target_tree_refs = self.target_repo._format.supports_tree_reference
2196
for record in substream:
2197
# It's not a delta, so it must be a fulltext in the source
2198
# serializer's format.
2199
bytes = record.get_bytes_as('fulltext')
2200
revision_id = record.key[0]
2201
inv = serializer.read_inventory_from_string(bytes, revision_id)
2202
parents = [key[0] for key in record.parents]
2203
self.target_repo.add_inventory(revision_id, inv, parents)
2204
# No need to keep holding this full inv in memory when the rest of
2205
# the substream is likely to be all deltas.
2208
def _extract_and_insert_revisions(self, substream, serializer):
2209
for record in substream:
2210
bytes = record.get_bytes_as('fulltext')
2211
revision_id = record.key[0]
2212
rev = serializer.read_revision_from_string(bytes)
2213
if rev.revision_id != revision_id:
2214
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
2215
self.target_repo.add_revision(revision_id, rev)
2218
if self.target_repo._format._fetch_reconcile:
2219
self.target_repo.reconcile()
2222
class StreamSource(object):
2223
"""A source of a stream for fetching between repositories."""
2225
def __init__(self, from_repository, to_format):
2226
"""Create a StreamSource streaming from from_repository."""
2227
self.from_repository = from_repository
2228
self.to_format = to_format
2229
self._record_counter = RecordCounter()
2231
def delta_on_metadata(self):
2232
"""Return True if delta's are permitted on metadata streams.
2234
That is on revisions and signatures.
2236
src_serializer = self.from_repository._format._serializer
2237
target_serializer = self.to_format._serializer
2238
return (self.to_format._fetch_uses_deltas and
2239
src_serializer == target_serializer)
2241
def _fetch_revision_texts(self, revs):
2242
# fetch signatures first and then the revision texts
2243
# may need to be a InterRevisionStore call here.
2244
from_sf = self.from_repository.signatures
2245
# A missing signature is just skipped.
2246
keys = [(rev_id,) for rev_id in revs]
2247
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
2249
self.to_format._fetch_order,
2250
not self.to_format._fetch_uses_deltas))
2251
# If a revision has a delta, this is actually expanded inside the
2252
# insert_record_stream code now, which is an alternate fix for
2254
from_rf = self.from_repository.revisions
2255
revisions = from_rf.get_record_stream(
2257
self.to_format._fetch_order,
2258
not self.delta_on_metadata())
2259
return [('signatures', signatures), ('revisions', revisions)]
2261
def _generate_root_texts(self, revs):
2262
"""This will be called by get_stream between fetching weave texts and
2263
fetching the inventory weave.
2265
if self._rich_root_upgrade():
2266
return _mod_fetch.Inter1and2Helper(
2267
self.from_repository).generate_root_texts(revs)
2271
def get_stream(self, search):
2273
revs = search.get_keys()
2274
graph = self.from_repository.get_graph()
2275
revs = tsort.topo_sort(graph.get_parent_map(revs))
2276
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
2278
for knit_kind, file_id, revisions in data_to_fetch:
2279
if knit_kind != phase:
2281
# Make a new progress bar for this phase
2282
if knit_kind == "file":
2283
# Accumulate file texts
2284
text_keys.extend([(file_id, revision) for revision in
2286
elif knit_kind == "inventory":
2287
# Now copy the file texts.
2288
from_texts = self.from_repository.texts
2289
yield ('texts', from_texts.get_record_stream(
2290
text_keys, self.to_format._fetch_order,
2291
not self.to_format._fetch_uses_deltas))
2292
# Cause an error if a text occurs after we have done the
2295
# Before we process the inventory we generate the root
2296
# texts (if necessary) so that the inventories references
2298
for _ in self._generate_root_texts(revs):
2300
# we fetch only the referenced inventories because we do not
2301
# know for unselected inventories whether all their required
2302
# texts are present in the other repository - it could be
2304
for info in self._get_inventory_stream(revs):
2306
elif knit_kind == "signatures":
2307
# Nothing to do here; this will be taken care of when
2308
# _fetch_revision_texts happens.
2310
elif knit_kind == "revisions":
2311
for record in self._fetch_revision_texts(revs):
2314
raise AssertionError("Unknown knit kind %r" % knit_kind)
2316
def get_stream_for_missing_keys(self, missing_keys):
2317
# missing keys can only occur when we are byte copying and not
2318
# translating (because translation means we don't send
2319
# unreconstructable deltas ever).
2321
keys['texts'] = set()
2322
keys['revisions'] = set()
2323
keys['inventories'] = set()
2324
keys['chk_bytes'] = set()
2325
keys['signatures'] = set()
2326
for key in missing_keys:
2327
keys[key[0]].add(key[1:])
2328
if len(keys['revisions']):
2329
# If we allowed copying revisions at this point, we could end up
2330
# copying a revision without copying its required texts: a
2331
# violation of the requirements for repository integrity.
2332
raise AssertionError(
2333
'cannot copy revisions to fill in missing deltas %s' % (
2334
keys['revisions'],))
2335
for substream_kind, keys in keys.iteritems():
2336
vf = getattr(self.from_repository, substream_kind)
2337
if vf is None and keys:
2338
raise AssertionError(
2339
"cannot fill in keys for a versioned file we don't"
2340
" have: %s needs %s" % (substream_kind, keys))
2342
# No need to stream something we don't have
2344
if substream_kind == 'inventories':
2345
# Some missing keys are genuinely ghosts, filter those out.
2346
present = self.from_repository.inventories.get_parent_map(keys)
2347
revs = [key[0] for key in present]
2348
# Get the inventory stream more-or-less as we do for the
2349
# original stream; there's no reason to assume that records
2350
# direct from the source will be suitable for the sink. (Think
2351
# e.g. 2a -> 1.9-rich-root).
2352
for info in self._get_inventory_stream(revs, missing=True):
2356
# Ask for full texts always so that we don't need more round trips
2357
# after this stream.
2358
# Some of the missing keys are genuinely ghosts, so filter absent
2359
# records. The Sink is responsible for doing another check to
2360
# ensure that ghosts don't introduce missing data for future
2362
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
2363
self.to_format._fetch_order, True))
2364
yield substream_kind, stream
2366
def inventory_fetch_order(self):
2367
if self._rich_root_upgrade():
2368
return 'topological'
2370
return self.to_format._fetch_order
2372
def _rich_root_upgrade(self):
2373
return (not self.from_repository._format.rich_root_data and
2374
self.to_format.rich_root_data)
2376
def _get_inventory_stream(self, revision_ids, missing=False):
2377
from_format = self.from_repository._format
2378
if (from_format.supports_chks and self.to_format.supports_chks and
2379
from_format.network_name() == self.to_format.network_name()):
2380
raise AssertionError(
2381
"this case should be handled by GroupCHKStreamSource")
2382
elif 'forceinvdeltas' in debug.debug_flags:
2383
return self._get_convertable_inventory_stream(revision_ids,
2384
delta_versus_null=missing)
2385
elif from_format.network_name() == self.to_format.network_name():
2387
return self._get_simple_inventory_stream(revision_ids,
2389
elif (not from_format.supports_chks and not self.to_format.supports_chks
2390
and from_format._serializer == self.to_format._serializer):
2391
# Essentially the same format.
2392
return self._get_simple_inventory_stream(revision_ids,
2395
# Any time we switch serializations, we want to use an
2396
# inventory-delta based approach.
2397
return self._get_convertable_inventory_stream(revision_ids,
2398
delta_versus_null=missing)
2400
def _get_simple_inventory_stream(self, revision_ids, missing=False):
2401
# NB: This currently reopens the inventory weave in source;
2402
# using a single stream interface instead would avoid this.
2403
from_weave = self.from_repository.inventories
2405
delta_closure = True
2407
delta_closure = not self.delta_on_metadata()
2408
yield ('inventories', from_weave.get_record_stream(
2409
[(rev_id,) for rev_id in revision_ids],
2410
self.inventory_fetch_order(), delta_closure))
2412
def _get_convertable_inventory_stream(self, revision_ids,
2413
delta_versus_null=False):
2414
# The two formats are sufficiently different that there is no fast
2415
# path, so we need to send just inventorydeltas, which any
2416
# sufficiently modern client can insert into any repository.
2417
# The StreamSink code expects to be able to
2418
# convert on the target, so we need to put bytes-on-the-wire that can
2419
# be converted. That means inventory deltas (if the remote is <1.19,
2420
# RemoteStreamSink will fallback to VFS to insert the deltas).
2421
yield ('inventory-deltas',
2422
self._stream_invs_as_deltas(revision_ids,
2423
delta_versus_null=delta_versus_null))
2425
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
2426
"""Return a stream of inventory-deltas for the given rev ids.
2428
:param revision_ids: The list of inventories to transmit
2429
:param delta_versus_null: Don't try to find a minimal delta for this
2430
entry, instead compute the delta versus the NULL_REVISION. This
2431
effectively streams a complete inventory. Used for stuff like
2432
filling in missing parents, etc.
2434
from_repo = self.from_repository
2435
revision_keys = [(rev_id,) for rev_id in revision_ids]
2436
parent_map = from_repo.inventories.get_parent_map(revision_keys)
2437
# XXX: possibly repos could implement a more efficient iter_inv_deltas
2439
inventories = self.from_repository.iter_inventories(
2440
revision_ids, 'topological')
2441
format = from_repo._format
2442
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
2443
inventory_cache = lru_cache.LRUCache(50)
2444
null_inventory = from_repo.revision_tree(
2445
_mod_revision.NULL_REVISION).root_inventory
2446
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
2447
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
2448
# repo back into a non-rich-root repo ought to be allowed)
2449
serializer = inventory_delta.InventoryDeltaSerializer(
2450
versioned_root=format.rich_root_data,
2451
tree_references=format.supports_tree_reference)
2452
for inv in inventories:
2453
key = (inv.revision_id,)
2454
parent_keys = parent_map.get(key, ())
2456
if not delta_versus_null and parent_keys:
2457
# The caller did not ask for complete inventories and we have
2458
# some parents that we can delta against. Make a delta against
2459
# each parent so that we can find the smallest.
2460
parent_ids = [parent_key[0] for parent_key in parent_keys]
2461
for parent_id in parent_ids:
2462
if parent_id not in invs_sent_so_far:
2463
# We don't know that the remote side has this basis, so
2466
if parent_id == _mod_revision.NULL_REVISION:
2467
parent_inv = null_inventory
2469
parent_inv = inventory_cache.get(parent_id, None)
2470
if parent_inv is None:
2471
parent_inv = from_repo.get_inventory(parent_id)
2472
candidate_delta = inv._make_delta(parent_inv)
2473
if (delta is None or
2474
len(delta) > len(candidate_delta)):
2475
delta = candidate_delta
2476
basis_id = parent_id
2478
# Either none of the parents ended up being suitable, or we
2479
# were asked to delta against NULL
2480
basis_id = _mod_revision.NULL_REVISION
2481
delta = inv._make_delta(null_inventory)
2482
invs_sent_so_far.add(inv.revision_id)
2483
inventory_cache[inv.revision_id] = inv
2484
delta_serialized = ''.join(
2485
serializer.delta_to_lines(basis_id, key[-1], delta))
2486
yield versionedfile.FulltextContentFactory(
2487
key, parent_keys, None, delta_serialized)
2490
class _VersionedFileChecker(object):
2492
def __init__(self, repository, text_key_references=None, ancestors=None):
2493
self.repository = repository
2494
self.text_index = self.repository._generate_text_key_index(
2495
text_key_references=text_key_references, ancestors=ancestors)
2497
def calculate_file_version_parents(self, text_key):
2498
"""Calculate the correct parents for a file version according to
2501
parent_keys = self.text_index[text_key]
2502
if parent_keys == [_mod_revision.NULL_REVISION]:
2504
return tuple(parent_keys)
2506
def check_file_version_parents(self, texts, progress_bar=None):
2507
"""Check the parents stored in a versioned file are correct.
2509
It also detects file versions that are not referenced by their
2510
corresponding revision's inventory.
2512
:returns: A tuple of (wrong_parents, dangling_file_versions).
2513
wrong_parents is a dict mapping {revision_id: (stored_parents,
2514
correct_parents)} for each revision_id where the stored parents
2515
are not correct. dangling_file_versions is a set of (file_id,
2516
revision_id) tuples for versions that are present in this versioned
2517
file, but not used by the corresponding inventory.
2519
local_progress = None
2520
if progress_bar is None:
2521
local_progress = ui.ui_factory.nested_progress_bar()
2522
progress_bar = local_progress
2524
return self._check_file_version_parents(texts, progress_bar)
2527
local_progress.finished()
2529
def _check_file_version_parents(self, texts, progress_bar):
2530
"""See check_file_version_parents."""
2532
self.file_ids = set([file_id for file_id, _ in
2533
self.text_index.iterkeys()])
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.iterkeys()):
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 bzrlib.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
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
2648
find_ghosts=True, revision_ids=None, if_present_ids=None,
2650
"""Return the revision ids that source has that target does not.
2652
:param revision_id: only return revision ids included by this
2654
:param revision_ids: return revision ids included by these
2655
revision_ids. NoSuchRevision will be raised if any of these
2656
revisions are not present.
2657
:param if_present_ids: like revision_ids, but will not cause
2658
NoSuchRevision if any of these are absent, instead they will simply
2659
not be in the result. This is useful for e.g. finding revisions
2660
to fetch for tags, which may reference absent revisions.
2661
:param find_ghosts: If True find missing revisions in deep history
2662
rather than just finding the surface difference.
2663
:return: A bzrlib.graph.SearchResult.
2665
if symbol_versioning.deprecated_passed(revision_id):
2666
symbol_versioning.warn(
2667
'search_missing_revision_ids(revision_id=...) was '
2668
'deprecated in 2.4. Use revision_ids=[...] instead.',
2669
DeprecationWarning, stacklevel=2)
2670
if revision_ids is not None:
2671
raise AssertionError(
2672
'revision_ids is mutually exclusive with revision_id')
2673
if revision_id is not None:
2674
revision_ids = [revision_id]
2676
# stop searching at found target revisions.
2677
if not find_ghosts and (revision_ids is not None or if_present_ids is
2679
result = self._walk_to_common_revisions(revision_ids,
2680
if_present_ids=if_present_ids)
2683
result_set = result.get_keys()
2685
# generic, possibly worst case, slow code path.
2686
target_ids = set(self.target.all_revision_ids())
2687
source_ids = self._present_source_revisions_for(
2688
revision_ids, if_present_ids)
2689
result_set = set(source_ids).difference(target_ids)
2690
if limit is not None:
2691
topo_ordered = self.source.get_graph().iter_topo_order(result_set)
2692
result_set = set(itertools.islice(topo_ordered, limit))
2693
return self.source.revision_ids_to_search_result(result_set)
2695
def _present_source_revisions_for(self, revision_ids, if_present_ids=None):
2696
"""Returns set of all revisions in ancestry of revision_ids present in
2699
:param revision_ids: if None, all revisions in source are returned.
2700
:param if_present_ids: like revision_ids, but if any/all of these are
2701
absent no error is raised.
2703
if revision_ids is not None or if_present_ids is not None:
2704
# First, ensure all specified revisions exist. Callers expect
2705
# NoSuchRevision when they pass absent revision_ids here.
2706
if revision_ids is None:
2707
revision_ids = set()
2708
if if_present_ids is None:
2709
if_present_ids = set()
2710
revision_ids = set(revision_ids)
2711
if_present_ids = set(if_present_ids)
2712
all_wanted_ids = revision_ids.union(if_present_ids)
2713
graph = self.source.get_graph()
2714
present_revs = set(graph.get_parent_map(all_wanted_ids))
2715
missing = revision_ids.difference(present_revs)
2717
raise errors.NoSuchRevision(self.source, missing.pop())
2718
found_ids = all_wanted_ids.intersection(present_revs)
2719
source_ids = [rev_id for (rev_id, parents) in
2720
graph.iter_ancestry(found_ids)
2721
if rev_id != _mod_revision.NULL_REVISION
2722
and parents is not None]
2724
source_ids = self.source.all_revision_ids()
2725
return set(source_ids)
2728
def _get_repo_format_to_test(self):
2732
def is_compatible(cls, source, target):
2733
# The default implementation is compatible with everything
2734
return (source._format.supports_full_versioned_files and
2735
target._format.supports_full_versioned_files)
2738
class InterDifferingSerializer(InterVersionedFileRepository):
2741
def _get_repo_format_to_test(self):
2745
def is_compatible(source, target):
2746
if not source._format.supports_full_versioned_files:
2748
if not target._format.supports_full_versioned_files:
2750
# This is redundant with format.check_conversion_target(), however that
2751
# raises an exception, and we just want to say "False" as in we won't
2752
# support converting between these formats.
2753
if 'IDS_never' in debug.debug_flags:
2755
if source.supports_rich_root() and not target.supports_rich_root():
2757
if (source._format.supports_tree_reference
2758
and not target._format.supports_tree_reference):
2760
if target._fallback_repositories and target._format.supports_chks:
2761
# IDS doesn't know how to copy CHKs for the parent inventories it
2762
# adds to stacked repos.
2764
if 'IDS_always' in debug.debug_flags:
2766
# Only use this code path for local source and target. IDS does far
2767
# too much IO (both bandwidth and roundtrips) over a network.
2768
if not source.bzrdir.transport.base.startswith('file:///'):
2770
if not target.bzrdir.transport.base.startswith('file:///'):
2774
def _get_trees(self, revision_ids, cache):
2776
for rev_id in revision_ids:
2778
possible_trees.append((rev_id, cache[rev_id]))
2780
# Not cached, but inventory might be present anyway.
2782
tree = self.source.revision_tree(rev_id)
2783
except errors.NoSuchRevision:
2784
# Nope, parent is ghost.
2787
cache[rev_id] = tree
2788
possible_trees.append((rev_id, tree))
2789
return possible_trees
2791
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
2792
"""Get the best delta and base for this revision.
2794
:return: (basis_id, delta)
2797
# Generate deltas against each tree, to find the shortest.
2798
# FIXME: Support nested trees
2799
texts_possibly_new_in_tree = set()
2800
for basis_id, basis_tree in possible_trees:
2801
delta = tree.root_inventory._make_delta(basis_tree.root_inventory)
2802
for old_path, new_path, file_id, new_entry in delta:
2803
if new_path is None:
2804
# This file_id isn't present in the new rev, so we don't
2808
# Rich roots are handled elsewhere...
2810
kind = new_entry.kind
2811
if kind != 'directory' and kind != 'file':
2812
# No text record associated with this inventory entry.
2814
# This is a directory or file that has changed somehow.
2815
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
2816
deltas.append((len(delta), basis_id, delta))
2818
return deltas[0][1:]
2820
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
2821
"""Find all parent revisions that are absent, but for which the
2822
inventory is present, and copy those inventories.
2824
This is necessary to preserve correctness when the source is stacked
2825
without fallbacks configured. (Note that in cases like upgrade the
2826
source may be not have _fallback_repositories even though it is
2830
for parents in parent_map.values():
2831
parent_revs.update(parents)
2832
present_parents = self.source.get_parent_map(parent_revs)
2833
absent_parents = set(parent_revs).difference(present_parents)
2834
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
2835
(rev_id,) for rev_id in absent_parents)
2836
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
2837
for parent_tree in self.source.revision_trees(parent_inv_ids):
2838
current_revision_id = parent_tree.get_revision_id()
2839
parents_parents_keys = parent_invs_keys_for_stacking[
2840
(current_revision_id,)]
2841
parents_parents = [key[-1] for key in parents_parents_keys]
2842
basis_id = _mod_revision.NULL_REVISION
2843
basis_tree = self.source.revision_tree(basis_id)
2844
delta = parent_tree.root_inventory._make_delta(
2845
basis_tree.root_inventory)
2846
self.target.add_inventory_by_delta(
2847
basis_id, delta, current_revision_id, parents_parents)
2848
cache[current_revision_id] = parent_tree
2850
def _fetch_batch(self, revision_ids, basis_id, cache):
2851
"""Fetch across a few revisions.
2853
:param revision_ids: The revisions to copy
2854
:param basis_id: The revision_id of a tree that must be in cache, used
2855
as a basis for delta when no other base is available
2856
:param cache: A cache of RevisionTrees that we can use.
2857
:return: The revision_id of the last converted tree. The RevisionTree
2858
for it will be in cache
2860
# Walk though all revisions; get inventory deltas, copy referenced
2861
# texts that delta references, insert the delta, revision and
2863
root_keys_to_create = set()
2866
pending_revisions = []
2867
parent_map = self.source.get_parent_map(revision_ids)
2868
self._fetch_parent_invs_for_stacking(parent_map, cache)
2869
self.source._safe_to_return_from_cache = True
2870
for tree in self.source.revision_trees(revision_ids):
2871
# Find a inventory delta for this revision.
2872
# Find text entries that need to be copied, too.
2873
current_revision_id = tree.get_revision_id()
2874
parent_ids = parent_map.get(current_revision_id, ())
2875
parent_trees = self._get_trees(parent_ids, cache)
2876
possible_trees = list(parent_trees)
2877
if len(possible_trees) == 0:
2878
# There either aren't any parents, or the parents are ghosts,
2879
# so just use the last converted tree.
2880
possible_trees.append((basis_id, cache[basis_id]))
2881
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
2883
revision = self.source.get_revision(current_revision_id)
2884
pending_deltas.append((basis_id, delta,
2885
current_revision_id, revision.parent_ids))
2886
if self._converting_to_rich_root:
2887
self._revision_id_to_root_id[current_revision_id] = \
2889
# Determine which texts are in present in this revision but not in
2890
# any of the available parents.
2891
texts_possibly_new_in_tree = set()
2892
for old_path, new_path, file_id, entry in delta:
2893
if new_path is None:
2894
# This file_id isn't present in the new rev
2898
if not self.target.supports_rich_root():
2899
# The target doesn't support rich root, so we don't
2902
if self._converting_to_rich_root:
2903
# This can't be copied normally, we have to insert
2905
root_keys_to_create.add((file_id, entry.revision))
2908
texts_possibly_new_in_tree.add((file_id, entry.revision))
2909
for basis_id, basis_tree in possible_trees:
2910
basis_inv = basis_tree.root_inventory
2911
for file_key in list(texts_possibly_new_in_tree):
2912
file_id, file_revision = file_key
2914
entry = basis_inv[file_id]
2915
except errors.NoSuchId:
2917
if entry.revision == file_revision:
2918
texts_possibly_new_in_tree.remove(file_key)
2919
text_keys.update(texts_possibly_new_in_tree)
2920
pending_revisions.append(revision)
2921
cache[current_revision_id] = tree
2922
basis_id = current_revision_id
2923
self.source._safe_to_return_from_cache = False
2925
from_texts = self.source.texts
2926
to_texts = self.target.texts
2927
if root_keys_to_create:
2928
root_stream = _mod_fetch._new_root_data_stream(
2929
root_keys_to_create, self._revision_id_to_root_id, parent_map,
2931
to_texts.insert_record_stream(root_stream)
2932
to_texts.insert_record_stream(from_texts.get_record_stream(
2933
text_keys, self.target._format._fetch_order,
2934
not self.target._format._fetch_uses_deltas))
2935
# insert inventory deltas
2936
for delta in pending_deltas:
2937
self.target.add_inventory_by_delta(*delta)
2938
if self.target._fallback_repositories:
2939
# Make sure this stacked repository has all the parent inventories
2940
# for the new revisions that we are about to insert. We do this
2941
# before adding the revisions so that no revision is added until
2942
# all the inventories it may depend on are added.
2943
# Note that this is overzealous, as we may have fetched these in an
2946
revision_ids = set()
2947
for revision in pending_revisions:
2948
revision_ids.add(revision.revision_id)
2949
parent_ids.update(revision.parent_ids)
2950
parent_ids.difference_update(revision_ids)
2951
parent_ids.discard(_mod_revision.NULL_REVISION)
2952
parent_map = self.source.get_parent_map(parent_ids)
2953
# we iterate over parent_map and not parent_ids because we don't
2954
# want to try copying any revision which is a ghost
2955
for parent_tree in self.source.revision_trees(parent_map):
2956
current_revision_id = parent_tree.get_revision_id()
2957
parents_parents = parent_map[current_revision_id]
2958
possible_trees = self._get_trees(parents_parents, cache)
2959
if len(possible_trees) == 0:
2960
# There either aren't any parents, or the parents are
2961
# ghosts, so just use the last converted tree.
2962
possible_trees.append((basis_id, cache[basis_id]))
2963
basis_id, delta = self._get_delta_for_revision(parent_tree,
2964
parents_parents, possible_trees)
2965
self.target.add_inventory_by_delta(
2966
basis_id, delta, current_revision_id, parents_parents)
2967
# insert signatures and revisions
2968
for revision in pending_revisions:
2970
signature = self.source.get_signature_text(
2971
revision.revision_id)
2972
self.target.add_signature_text(revision.revision_id,
2974
except errors.NoSuchRevision:
2976
self.target.add_revision(revision.revision_id, revision)
2979
def _fetch_all_revisions(self, revision_ids, pb):
2980
"""Fetch everything for the list of revisions.
2982
:param revision_ids: The list of revisions to fetch. Must be in
2984
:param pb: A ProgressTask
2987
basis_id, basis_tree = self._get_basis(revision_ids[0])
2989
cache = lru_cache.LRUCache(100)
2990
cache[basis_id] = basis_tree
2991
del basis_tree # We don't want to hang on to it here
2995
for offset in range(0, len(revision_ids), batch_size):
2996
self.target.start_write_group()
2998
pb.update(gettext('Transferring revisions'), offset,
3000
batch = revision_ids[offset:offset+batch_size]
3001
basis_id = self._fetch_batch(batch, basis_id, cache)
3003
self.source._safe_to_return_from_cache = False
3004
self.target.abort_write_group()
3007
hint = self.target.commit_write_group()
3010
if hints and self.target._format.pack_compresses:
3011
self.target.pack(hint=hints)
3012
pb.update(gettext('Transferring revisions'), len(revision_ids),
3016
def fetch(self, revision_id=None, find_ghosts=False,
3018
"""See InterRepository.fetch()."""
3019
if fetch_spec is not None:
3020
revision_ids = fetch_spec.get_keys()
3023
if self.source._format.experimental:
3024
ui.ui_factory.show_user_warning('experimental_format_fetch',
3025
from_format=self.source._format,
3026
to_format=self.target._format)
3027
if (not self.source.supports_rich_root()
3028
and self.target.supports_rich_root()):
3029
self._converting_to_rich_root = True
3030
self._revision_id_to_root_id = {}
3032
self._converting_to_rich_root = False
3033
# See <https://launchpad.net/bugs/456077> asking for a warning here
3034
if self.source._format.network_name() != self.target._format.network_name():
3035
ui.ui_factory.show_user_warning('cross_format_fetch',
3036
from_format=self.source._format,
3037
to_format=self.target._format)
3038
if revision_ids is None:
3040
search_revision_ids = [revision_id]
3042
search_revision_ids = None
3043
revision_ids = self.target.search_missing_revision_ids(self.source,
3044
revision_ids=search_revision_ids,
3045
find_ghosts=find_ghosts).get_keys()
3046
if not revision_ids:
3048
revision_ids = tsort.topo_sort(
3049
self.source.get_graph().get_parent_map(revision_ids))
3050
if not revision_ids:
3052
# Walk though all revisions; get inventory deltas, copy referenced
3053
# texts that delta references, insert the delta, revision and
3055
pb = ui.ui_factory.nested_progress_bar()
3057
self._fetch_all_revisions(revision_ids, pb)
3060
return len(revision_ids), 0
3062
def _get_basis(self, first_revision_id):
3063
"""Get a revision and tree which exists in the target.
3065
This assumes that first_revision_id is selected for transmission
3066
because all other ancestors are already present. If we can't find an
3067
ancestor we fall back to NULL_REVISION since we know that is safe.
3069
:return: (basis_id, basis_tree)
3071
first_rev = self.source.get_revision(first_revision_id)
3073
basis_id = first_rev.parent_ids[0]
3074
# only valid as a basis if the target has it
3075
self.target.get_revision(basis_id)
3076
# Try to get a basis tree - if it's a ghost it will hit the
3077
# NoSuchRevision case.
3078
basis_tree = self.source.revision_tree(basis_id)
3079
except (IndexError, errors.NoSuchRevision):
3080
basis_id = _mod_revision.NULL_REVISION
3081
basis_tree = self.source.revision_tree(basis_id)
3082
return basis_id, basis_tree
3085
class InterSameDataRepository(InterVersionedFileRepository):
3086
"""Code for converting between repositories that represent the same data.
3088
Data format and model must match for this to work.
3092
def _get_repo_format_to_test(self):
3093
"""Repository format for testing with.
3095
InterSameData can pull from subtree to subtree and from non-subtree to
3096
non-subtree, so we test this with the richest repository format.
3098
from bzrlib.repofmt import knitrepo
3099
return knitrepo.RepositoryFormatKnit3()
3102
def is_compatible(source, target):
3104
InterRepository._same_model(source, target) and
3105
source._format.supports_full_versioned_files and
3106
target._format.supports_full_versioned_files)
3109
InterRepository.register_optimiser(InterVersionedFileRepository)
3110
InterRepository.register_optimiser(InterDifferingSerializer)
3111
InterRepository.register_optimiser(InterSameDataRepository)
3114
def install_revisions(repository, iterable, num_revisions=None, pb=None):
3115
"""Install all revision data into a repository.
3117
Accepts an iterable of revision, tree, signature tuples. The signature
3120
repository.start_write_group()
3122
inventory_cache = lru_cache.LRUCache(10)
3123
for n, (revision, revision_tree, signature) in enumerate(iterable):
3124
_install_revision(repository, revision, revision_tree, signature,
3127
pb.update(gettext('Transferring revisions'), n + 1, num_revisions)
3129
repository.abort_write_group()
3132
repository.commit_write_group()
3135
def _install_revision(repository, rev, revision_tree, signature,
3137
"""Install all revision data into a repository."""
3138
present_parents = []
3140
for p_id in rev.parent_ids:
3141
if repository.has_revision(p_id):
3142
present_parents.append(p_id)
3143
parent_trees[p_id] = repository.revision_tree(p_id)
3145
parent_trees[p_id] = repository.revision_tree(
3146
_mod_revision.NULL_REVISION)
3148
# FIXME: Support nested trees
3149
inv = revision_tree.root_inventory
3150
entries = inv.iter_entries()
3151
# backwards compatibility hack: skip the root id.
3152
if not repository.supports_rich_root():
3153
path, root = entries.next()
3154
if root.revision != rev.revision_id:
3155
raise errors.IncompatibleRevision(repr(repository))
3157
for path, ie in entries:
3158
text_keys[(ie.file_id, ie.revision)] = ie
3159
text_parent_map = repository.texts.get_parent_map(text_keys)
3160
missing_texts = set(text_keys) - set(text_parent_map)
3161
# Add the texts that are not already present
3162
for text_key in missing_texts:
3163
ie = text_keys[text_key]
3165
# FIXME: TODO: The following loop overlaps/duplicates that done by
3166
# commit to determine parents. There is a latent/real bug here where
3167
# the parents inserted are not those commit would do - in particular
3168
# they are not filtered by heads(). RBC, AB
3169
for revision, tree in parent_trees.iteritems():
3170
if not tree.has_id(ie.file_id):
3172
parent_id = tree.get_file_revision(ie.file_id)
3173
if parent_id in text_parents:
3175
text_parents.append((ie.file_id, parent_id))
3176
lines = revision_tree.get_file(ie.file_id).readlines()
3177
repository.texts.add_lines(text_key, text_parents, lines)
3179
# install the inventory
3180
if repository._format._commit_inv_deltas and len(rev.parent_ids):
3181
# Cache this inventory
3182
inventory_cache[rev.revision_id] = inv
3184
basis_inv = inventory_cache[rev.parent_ids[0]]
3186
repository.add_inventory(rev.revision_id, inv, present_parents)
3188
delta = inv._make_delta(basis_inv)
3189
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
3190
rev.revision_id, present_parents)
3192
repository.add_inventory(rev.revision_id, inv, present_parents)
3193
except errors.RevisionAlreadyPresent:
3195
if signature is not None:
3196
repository.add_signature_text(rev.revision_id, signature)
3197
repository.add_revision(rev.revision_id, rev, inv)
3200
def install_revision(repository, rev, revision_tree):
3201
"""Install all revision data into a repository."""
3202
install_revisions(repository, [(rev, revision_tree, None)])
added
removed
bzrlib/vf_repository.py
