13
13
# You should have received a copy of the GNU General Public License
14
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
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
from cStringIO import StringIO
17
19
from bzrlib.lazy_import import lazy_import
18
20
lazy_import(globals(), """
23
25
from bzrlib import (
40
37
revision as _mod_revision,
46
from bzrlib.bundle import serializer
47
42
from bzrlib.revisiontree import RevisionTree
48
43
from bzrlib.store.versioned import VersionedFileStore
44
from bzrlib.store.text import TextStore
49
45
from bzrlib.testament import Testament
52
49
from bzrlib.decorators import needs_read_lock, needs_write_lock
53
50
from bzrlib.inter import InterObject
54
from bzrlib.inventory import (
60
from bzrlib import registry
61
from bzrlib.trace import (
62
log_exception_quietly, note, mutter, mutter_callsite, warning)
51
from bzrlib.inventory import Inventory, InventoryDirectory, ROOT_ID
52
from bzrlib.symbol_versioning import (
56
from bzrlib.trace import mutter, note, warning
65
59
# Old formats display a warning, but only once
66
60
_deprecation_warning_done = False
69
class CommitBuilder(object):
70
"""Provides an interface to build up a commit.
72
This allows describing a tree to be committed without needing to
73
know the internals of the format of the repository.
76
# all clients should supply tree roots.
77
record_root_entry = True
78
# the default CommitBuilder does not manage trees whose root is versioned.
79
_versioned_root = False
81
def __init__(self, repository, parents, config, timestamp=None,
82
timezone=None, committer=None, revprops=None,
84
"""Initiate a CommitBuilder.
86
:param repository: Repository to commit to.
87
:param parents: Revision ids of the parents of the new revision.
88
:param config: Configuration to use.
89
:param timestamp: Optional timestamp recorded for commit.
90
:param timezone: Optional timezone for timestamp.
91
:param committer: Optional committer to set for commit.
92
:param revprops: Optional dictionary of revision properties.
93
:param revision_id: Optional revision id.
98
self._committer = self._config.username()
100
self._committer = committer
102
self.new_inventory = Inventory(None)
103
self._new_revision_id = revision_id
104
self.parents = parents
105
self.repository = repository
108
if revprops is not None:
109
self._validate_revprops(revprops)
110
self._revprops.update(revprops)
112
if timestamp is None:
113
timestamp = time.time()
114
# Restrict resolution to 1ms
115
self._timestamp = round(timestamp, 3)
118
self._timezone = osutils.local_time_offset()
120
self._timezone = int(timezone)
122
self._generate_revision_if_needed()
123
self.__heads = graph.HeadsCache(repository.get_graph()).heads
124
self._basis_delta = []
125
# API compatibility, older code that used CommitBuilder did not call
126
# .record_delete(), which means the delta that is computed would not be
127
# valid. Callers that will call record_delete() should call
128
# .will_record_deletes() to indicate that.
129
self._recording_deletes = False
130
# memo'd check for no-op commits.
131
self._any_changes = False
133
def any_changes(self):
134
"""Return True if any entries were changed.
136
This includes merge-only changes. It is the core for the --unchanged
139
:return: True if any changes have occured.
141
return self._any_changes
143
def _validate_unicode_text(self, text, context):
144
"""Verify things like commit messages don't have bogus characters."""
146
raise ValueError('Invalid value for %s: %r' % (context, text))
148
def _validate_revprops(self, revprops):
149
for key, value in revprops.iteritems():
150
# We know that the XML serializers do not round trip '\r'
151
# correctly, so refuse to accept them
152
if not isinstance(value, basestring):
153
raise ValueError('revision property (%s) is not a valid'
154
' (unicode) string: %r' % (key, value))
155
self._validate_unicode_text(value,
156
'revision property (%s)' % (key,))
158
def commit(self, message):
159
"""Make the actual commit.
161
:return: The revision id of the recorded revision.
163
self._validate_unicode_text(message, 'commit message')
164
rev = _mod_revision.Revision(
165
timestamp=self._timestamp,
166
timezone=self._timezone,
167
committer=self._committer,
169
inventory_sha1=self.inv_sha1,
170
revision_id=self._new_revision_id,
171
properties=self._revprops)
172
rev.parent_ids = self.parents
173
self.repository.add_revision(self._new_revision_id, rev,
174
self.new_inventory, self._config)
175
self.repository.commit_write_group()
176
return self._new_revision_id
179
"""Abort the commit that is being built.
181
self.repository.abort_write_group()
183
def revision_tree(self):
184
"""Return the tree that was just committed.
186
After calling commit() this can be called to get a RevisionTree
187
representing the newly committed tree. This is preferred to
188
calling Repository.revision_tree() because that may require
189
deserializing the inventory, while we already have a copy in
192
if self.new_inventory is None:
193
self.new_inventory = self.repository.get_inventory(
194
self._new_revision_id)
195
return RevisionTree(self.repository, self.new_inventory,
196
self._new_revision_id)
198
def finish_inventory(self):
199
"""Tell the builder that the inventory is finished.
201
:return: The inventory id in the repository, which can be used with
202
repository.get_inventory.
204
if self.new_inventory is None:
205
# an inventory delta was accumulated without creating a new
207
basis_id = self.basis_delta_revision
208
self.inv_sha1 = self.repository.add_inventory_by_delta(
209
basis_id, self._basis_delta, self._new_revision_id,
212
if self.new_inventory.root is None:
213
raise AssertionError('Root entry should be supplied to'
214
' record_entry_contents, as of bzr 0.10.')
215
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
216
self.new_inventory.revision_id = self._new_revision_id
217
self.inv_sha1 = self.repository.add_inventory(
218
self._new_revision_id,
222
return self._new_revision_id
224
def _gen_revision_id(self):
225
"""Return new revision-id."""
226
return generate_ids.gen_revision_id(self._config.username(),
229
def _generate_revision_if_needed(self):
230
"""Create a revision id if None was supplied.
232
If the repository can not support user-specified revision ids
233
they should override this function and raise CannotSetRevisionId
234
if _new_revision_id is not None.
236
:raises: CannotSetRevisionId
238
if self._new_revision_id is None:
239
self._new_revision_id = self._gen_revision_id()
240
self.random_revid = True
242
self.random_revid = False
244
def _heads(self, file_id, revision_ids):
245
"""Calculate the graph heads for revision_ids in the graph of file_id.
247
This can use either a per-file graph or a global revision graph as we
248
have an identity relationship between the two graphs.
250
return self.__heads(revision_ids)
252
def _check_root(self, ie, parent_invs, tree):
253
"""Helper for record_entry_contents.
255
:param ie: An entry being added.
256
:param parent_invs: The inventories of the parent revisions of the
258
:param tree: The tree that is being committed.
260
# In this revision format, root entries have no knit or weave When
261
# serializing out to disk and back in root.revision is always
263
ie.revision = self._new_revision_id
265
def _require_root_change(self, tree):
266
"""Enforce an appropriate root object change.
268
This is called once when record_iter_changes is called, if and only if
269
the root was not in the delta calculated by record_iter_changes.
271
:param tree: The tree which is being committed.
273
# NB: if there are no parents then this method is not called, so no
274
# need to guard on parents having length.
275
entry = entry_factory['directory'](tree.path2id(''), '',
277
entry.revision = self._new_revision_id
278
self._basis_delta.append(('', '', entry.file_id, entry))
280
def _get_delta(self, ie, basis_inv, path):
281
"""Get a delta against the basis inventory for ie."""
282
if ie.file_id not in basis_inv:
284
result = (None, path, ie.file_id, ie)
285
self._basis_delta.append(result)
287
elif ie != basis_inv[ie.file_id]:
289
# TODO: avoid tis id2path call.
290
result = (basis_inv.id2path(ie.file_id), path, ie.file_id, ie)
291
self._basis_delta.append(result)
297
def get_basis_delta(self):
298
"""Return the complete inventory delta versus the basis inventory.
300
This has been built up with the calls to record_delete and
301
record_entry_contents. The client must have already called
302
will_record_deletes() to indicate that they will be generating a
305
:return: An inventory delta, suitable for use with apply_delta, or
306
Repository.add_inventory_by_delta, etc.
308
if not self._recording_deletes:
309
raise AssertionError("recording deletes not activated.")
310
return self._basis_delta
312
def record_delete(self, path, file_id):
313
"""Record that a delete occured against a basis tree.
315
This is an optional API - when used it adds items to the basis_delta
316
being accumulated by the commit builder. It cannot be called unless the
317
method will_record_deletes() has been called to inform the builder that
318
a delta is being supplied.
320
:param path: The path of the thing deleted.
321
:param file_id: The file id that was deleted.
323
if not self._recording_deletes:
324
raise AssertionError("recording deletes not activated.")
325
delta = (path, None, file_id, None)
326
self._basis_delta.append(delta)
327
self._any_changes = True
330
def will_record_deletes(self):
331
"""Tell the commit builder that deletes are being notified.
333
This enables the accumulation of an inventory delta; for the resulting
334
commit to be valid, deletes against the basis MUST be recorded via
335
builder.record_delete().
337
self._recording_deletes = True
339
basis_id = self.parents[0]
341
basis_id = _mod_revision.NULL_REVISION
342
self.basis_delta_revision = basis_id
344
def record_entry_contents(self, ie, parent_invs, path, tree,
346
"""Record the content of ie from tree into the commit if needed.
348
Side effect: sets ie.revision when unchanged
350
:param ie: An inventory entry present in the commit.
351
:param parent_invs: The inventories of the parent revisions of the
353
:param path: The path the entry is at in the tree.
354
:param tree: The tree which contains this entry and should be used to
356
:param content_summary: Summary data from the tree about the paths
357
content - stat, length, exec, sha/link target. This is only
358
accessed when the entry has a revision of None - that is when it is
359
a candidate to commit.
360
:return: A tuple (change_delta, version_recorded, fs_hash).
361
change_delta is an inventory_delta change for this entry against
362
the basis tree of the commit, or None if no change occured against
364
version_recorded is True if a new version of the entry has been
365
recorded. For instance, committing a merge where a file was only
366
changed on the other side will return (delta, False).
367
fs_hash is either None, or the hash details for the path (currently
368
a tuple of the contents sha1 and the statvalue returned by
369
tree.get_file_with_stat()).
371
if self.new_inventory.root is None:
372
if ie.parent_id is not None:
373
raise errors.RootMissing()
374
self._check_root(ie, parent_invs, tree)
375
if ie.revision is None:
376
kind = content_summary[0]
378
# ie is carried over from a prior commit
380
# XXX: repository specific check for nested tree support goes here - if
381
# the repo doesn't want nested trees we skip it ?
382
if (kind == 'tree-reference' and
383
not self.repository._format.supports_tree_reference):
384
# mismatch between commit builder logic and repository:
385
# this needs the entry creation pushed down into the builder.
386
raise NotImplementedError('Missing repository subtree support.')
387
self.new_inventory.add(ie)
389
# TODO: slow, take it out of the inner loop.
391
basis_inv = parent_invs[0]
393
basis_inv = Inventory(root_id=None)
395
# ie.revision is always None if the InventoryEntry is considered
396
# for committing. We may record the previous parents revision if the
397
# content is actually unchanged against a sole head.
398
if ie.revision is not None:
399
if not self._versioned_root and path == '':
400
# repositories that do not version the root set the root's
401
# revision to the new commit even when no change occurs (more
402
# specifically, they do not record a revision on the root; and
403
# the rev id is assigned to the root during deserialisation -
404
# this masks when a change may have occurred against the basis.
405
# To match this we always issue a delta, because the revision
406
# of the root will always be changing.
407
if ie.file_id in basis_inv:
408
delta = (basis_inv.id2path(ie.file_id), path,
412
delta = (None, path, ie.file_id, ie)
413
self._basis_delta.append(delta)
414
return delta, False, None
416
# we don't need to commit this, because the caller already
417
# determined that an existing revision of this file is
418
# appropriate. If its not being considered for committing then
419
# it and all its parents to the root must be unaltered so
420
# no-change against the basis.
421
if ie.revision == self._new_revision_id:
422
raise AssertionError("Impossible situation, a skipped "
423
"inventory entry (%r) claims to be modified in this "
424
"commit (%r).", (ie, self._new_revision_id))
425
return None, False, None
426
# XXX: Friction: parent_candidates should return a list not a dict
427
# so that we don't have to walk the inventories again.
428
parent_candiate_entries = ie.parent_candidates(parent_invs)
429
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
431
for inv in parent_invs:
432
if ie.file_id in inv:
433
old_rev = inv[ie.file_id].revision
434
if old_rev in head_set:
435
heads.append(inv[ie.file_id].revision)
436
head_set.remove(inv[ie.file_id].revision)
439
# now we check to see if we need to write a new record to the
441
# We write a new entry unless there is one head to the ancestors, and
442
# the kind-derived content is unchanged.
444
# Cheapest check first: no ancestors, or more the one head in the
445
# ancestors, we write a new node.
449
# There is a single head, look it up for comparison
450
parent_entry = parent_candiate_entries[heads[0]]
451
# if the non-content specific data has changed, we'll be writing a
453
if (parent_entry.parent_id != ie.parent_id or
454
parent_entry.name != ie.name):
456
# now we need to do content specific checks:
458
# if the kind changed the content obviously has
459
if kind != parent_entry.kind:
461
# Stat cache fingerprint feedback for the caller - None as we usually
462
# don't generate one.
465
if content_summary[2] is None:
466
raise ValueError("Files must not have executable = None")
468
if (# if the file length changed we have to store:
469
parent_entry.text_size != content_summary[1] or
470
# if the exec bit has changed we have to store:
471
parent_entry.executable != content_summary[2]):
473
elif parent_entry.text_sha1 == content_summary[3]:
474
# all meta and content is unchanged (using a hash cache
475
# hit to check the sha)
476
ie.revision = parent_entry.revision
477
ie.text_size = parent_entry.text_size
478
ie.text_sha1 = parent_entry.text_sha1
479
ie.executable = parent_entry.executable
480
return self._get_delta(ie, basis_inv, path), False, None
482
# Either there is only a hash change(no hash cache entry,
483
# or same size content change), or there is no change on
485
# Provide the parent's hash to the store layer, so that the
486
# content is unchanged we will not store a new node.
487
nostore_sha = parent_entry.text_sha1
489
# We want to record a new node regardless of the presence or
490
# absence of a content change in the file.
492
ie.executable = content_summary[2]
493
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
495
text = file_obj.read()
499
ie.text_sha1, ie.text_size = self._add_text_to_weave(
500
ie.file_id, text, heads, nostore_sha)
501
# Let the caller know we generated a stat fingerprint.
502
fingerprint = (ie.text_sha1, stat_value)
503
except errors.ExistingContent:
504
# Turns out that the file content was unchanged, and we were
505
# only going to store a new node if it was changed. Carry over
507
ie.revision = parent_entry.revision
508
ie.text_size = parent_entry.text_size
509
ie.text_sha1 = parent_entry.text_sha1
510
ie.executable = parent_entry.executable
511
return self._get_delta(ie, basis_inv, path), False, None
512
elif kind == 'directory':
514
# all data is meta here, nothing specific to directory, so
516
ie.revision = parent_entry.revision
517
return self._get_delta(ie, basis_inv, path), False, None
518
self._add_text_to_weave(ie.file_id, '', heads, None)
519
elif kind == 'symlink':
520
current_link_target = content_summary[3]
522
# symlink target is not generic metadata, check if it has
524
if current_link_target != parent_entry.symlink_target:
527
# unchanged, carry over.
528
ie.revision = parent_entry.revision
529
ie.symlink_target = parent_entry.symlink_target
530
return self._get_delta(ie, basis_inv, path), False, None
531
ie.symlink_target = current_link_target
532
self._add_text_to_weave(ie.file_id, '', heads, None)
533
elif kind == 'tree-reference':
535
if content_summary[3] != parent_entry.reference_revision:
538
# unchanged, carry over.
539
ie.reference_revision = parent_entry.reference_revision
540
ie.revision = parent_entry.revision
541
return self._get_delta(ie, basis_inv, path), False, None
542
ie.reference_revision = content_summary[3]
543
self._add_text_to_weave(ie.file_id, '', heads, None)
545
raise NotImplementedError('unknown kind')
546
ie.revision = self._new_revision_id
547
self._any_changes = True
548
return self._get_delta(ie, basis_inv, path), True, fingerprint
550
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
551
_entry_factory=entry_factory):
552
"""Record a new tree via iter_changes.
554
:param tree: The tree to obtain text contents from for changed objects.
555
:param basis_revision_id: The revision id of the tree the iter_changes
556
has been generated against. Currently assumed to be the same
557
as self.parents[0] - if it is not, errors may occur.
558
:param iter_changes: An iter_changes iterator with the changes to apply
559
to basis_revision_id. The iterator must not include any items with
560
a current kind of None - missing items must be either filtered out
561
or errored-on beefore record_iter_changes sees the item.
562
:param _entry_factory: Private method to bind entry_factory locally for
564
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
567
# Create an inventory delta based on deltas between all the parents and
568
# deltas between all the parent inventories. We use inventory delta's
569
# between the inventory objects because iter_changes masks
570
# last-changed-field only changes.
572
# file_id -> change map, change is fileid, paths, changed, versioneds,
573
# parents, names, kinds, executables
575
# {file_id -> revision_id -> inventory entry, for entries in parent
576
# trees that are not parents[0]
580
revtrees = list(self.repository.revision_trees(self.parents))
581
except errors.NoSuchRevision:
582
# one or more ghosts, slow path.
584
for revision_id in self.parents:
586
revtrees.append(self.repository.revision_tree(revision_id))
587
except errors.NoSuchRevision:
589
basis_revision_id = _mod_revision.NULL_REVISION
591
revtrees.append(self.repository.revision_tree(
592
_mod_revision.NULL_REVISION))
593
# The basis inventory from a repository
595
basis_inv = revtrees[0].inventory
597
basis_inv = self.repository.revision_tree(
598
_mod_revision.NULL_REVISION).inventory
599
if len(self.parents) > 0:
600
if basis_revision_id != self.parents[0] and not ghost_basis:
602
"arbitrary basis parents not yet supported with merges")
603
for revtree in revtrees[1:]:
604
for change in revtree.inventory._make_delta(basis_inv):
605
if change[1] is None:
606
# Not present in this parent.
608
if change[2] not in merged_ids:
609
if change[0] is not None:
610
basis_entry = basis_inv[change[2]]
611
merged_ids[change[2]] = [
613
basis_entry.revision,
616
parent_entries[change[2]] = {
618
basis_entry.revision:basis_entry,
620
change[3].revision:change[3],
623
merged_ids[change[2]] = [change[3].revision]
624
parent_entries[change[2]] = {change[3].revision:change[3]}
626
merged_ids[change[2]].append(change[3].revision)
627
parent_entries[change[2]][change[3].revision] = change[3]
630
# Setup the changes from the tree:
631
# changes maps file_id -> (change, [parent revision_ids])
633
for change in iter_changes:
634
# This probably looks up in basis_inv way to much.
635
if change[1][0] is not None:
636
head_candidate = [basis_inv[change[0]].revision]
639
changes[change[0]] = change, merged_ids.get(change[0],
641
unchanged_merged = set(merged_ids) - set(changes)
642
# Extend the changes dict with synthetic changes to record merges of
644
for file_id in unchanged_merged:
645
# Record a merged version of these items that did not change vs the
646
# basis. This can be either identical parallel changes, or a revert
647
# of a specific file after a merge. The recorded content will be
648
# that of the current tree (which is the same as the basis), but
649
# the per-file graph will reflect a merge.
650
# NB:XXX: We are reconstructing path information we had, this
651
# should be preserved instead.
652
# inv delta change: (file_id, (path_in_source, path_in_target),
653
# changed_content, versioned, parent, name, kind,
656
basis_entry = basis_inv[file_id]
657
except errors.NoSuchId:
658
# a change from basis->some_parents but file_id isn't in basis
659
# so was new in the merge, which means it must have changed
660
# from basis -> current, and as it hasn't the add was reverted
661
# by the user. So we discard this change.
665
(basis_inv.id2path(file_id), tree.id2path(file_id)),
667
(basis_entry.parent_id, basis_entry.parent_id),
668
(basis_entry.name, basis_entry.name),
669
(basis_entry.kind, basis_entry.kind),
670
(basis_entry.executable, basis_entry.executable))
671
changes[file_id] = (change, merged_ids[file_id])
672
# changes contains tuples with the change and a set of inventory
673
# candidates for the file.
675
# old_path, new_path, file_id, new_inventory_entry
676
seen_root = False # Is the root in the basis delta?
677
inv_delta = self._basis_delta
678
modified_rev = self._new_revision_id
679
for change, head_candidates in changes.values():
680
if change[3][1]: # versioned in target.
681
# Several things may be happening here:
682
# We may have a fork in the per-file graph
683
# - record a change with the content from tree
684
# We may have a change against < all trees
685
# - carry over the tree that hasn't changed
686
# We may have a change against all trees
687
# - record the change with the content from tree
690
entry = _entry_factory[kind](file_id, change[5][1],
692
head_set = self._heads(change[0], set(head_candidates))
695
for head_candidate in head_candidates:
696
if head_candidate in head_set:
697
heads.append(head_candidate)
698
head_set.remove(head_candidate)
701
# Could be a carry-over situation:
702
parent_entry_revs = parent_entries.get(file_id, None)
703
if parent_entry_revs:
704
parent_entry = parent_entry_revs.get(heads[0], None)
707
if parent_entry is None:
708
# The parent iter_changes was called against is the one
709
# that is the per-file head, so any change is relevant
710
# iter_changes is valid.
711
carry_over_possible = False
713
# could be a carry over situation
714
# A change against the basis may just indicate a merge,
715
# we need to check the content against the source of the
716
# merge to determine if it was changed after the merge
718
if (parent_entry.kind != entry.kind or
719
parent_entry.parent_id != entry.parent_id or
720
parent_entry.name != entry.name):
721
# Metadata common to all entries has changed
722
# against per-file parent
723
carry_over_possible = False
725
carry_over_possible = True
726
# per-type checks for changes against the parent_entry
729
# Cannot be a carry-over situation
730
carry_over_possible = False
731
# Populate the entry in the delta
733
# XXX: There is still a small race here: If someone reverts the content of a file
734
# after iter_changes examines and decides it has changed,
735
# we will unconditionally record a new version even if some
736
# other process reverts it while commit is running (with
737
# the revert happening after iter_changes did it's
740
entry.executable = True
742
entry.executable = False
743
if (carry_over_possible and
744
parent_entry.executable == entry.executable):
745
# Check the file length, content hash after reading
747
nostore_sha = parent_entry.text_sha1
750
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
752
text = file_obj.read()
756
entry.text_sha1, entry.text_size = self._add_text_to_weave(
757
file_id, text, heads, nostore_sha)
758
yield file_id, change[1][1], (entry.text_sha1, stat_value)
759
except errors.ExistingContent:
760
# No content change against a carry_over parent
761
# Perhaps this should also yield a fs hash update?
763
entry.text_size = parent_entry.text_size
764
entry.text_sha1 = parent_entry.text_sha1
765
elif kind == 'symlink':
767
entry.symlink_target = tree.get_symlink_target(file_id)
768
if (carry_over_possible and
769
parent_entry.symlink_target == entry.symlink_target):
772
self._add_text_to_weave(change[0], '', heads, None)
773
elif kind == 'directory':
774
if carry_over_possible:
777
# Nothing to set on the entry.
778
# XXX: split into the Root and nonRoot versions.
779
if change[1][1] != '' or self.repository.supports_rich_root():
780
self._add_text_to_weave(change[0], '', heads, None)
781
elif kind == 'tree-reference':
782
if not self.repository._format.supports_tree_reference:
783
# This isn't quite sane as an error, but we shouldn't
784
# ever see this code path in practice: tree's don't
785
# permit references when the repo doesn't support tree
787
raise errors.UnsupportedOperation(tree.add_reference,
789
reference_revision = tree.get_reference_revision(change[0])
790
entry.reference_revision = reference_revision
791
if (carry_over_possible and
792
parent_entry.reference_revision == reference_revision):
795
self._add_text_to_weave(change[0], '', heads, None)
797
raise AssertionError('unknown kind %r' % kind)
799
entry.revision = modified_rev
801
entry.revision = parent_entry.revision
804
new_path = change[1][1]
805
inv_delta.append((change[1][0], new_path, change[0], entry))
808
self.new_inventory = None
810
self._any_changes = True
812
# housekeeping root entry changes do not affect no-change commits.
813
self._require_root_change(tree)
814
self.basis_delta_revision = basis_revision_id
816
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
817
parent_keys = tuple([(file_id, parent) for parent in parents])
818
return self.repository.texts._add_text(
819
(file_id, self._new_revision_id), parent_keys, new_text,
820
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
823
class RootCommitBuilder(CommitBuilder):
824
"""This commitbuilder actually records the root id"""
826
# the root entry gets versioned properly by this builder.
827
_versioned_root = True
829
def _check_root(self, ie, parent_invs, tree):
830
"""Helper for record_entry_contents.
832
:param ie: An entry being added.
833
:param parent_invs: The inventories of the parent revisions of the
835
:param tree: The tree that is being committed.
838
def _require_root_change(self, tree):
839
"""Enforce an appropriate root object change.
841
This is called once when record_iter_changes is called, if and only if
842
the root was not in the delta calculated by record_iter_changes.
844
:param tree: The tree which is being committed.
846
# versioned roots do not change unless the tree found a change.
849
63
######################################################################
853
66
class Repository(object):
854
67
"""Repository holding history for one or more branches.
857
70
revisions and file history. It's normally accessed only by the Branch,
858
71
which views a particular line of development through that history.
860
The Repository builds on top of some byte storage facilies (the revisions,
861
signatures, inventories, texts and chk_bytes attributes) and a Transport,
862
which respectively provide byte storage and a means to access the (possibly
73
The Repository builds on top of Stores and a Transport, which respectively
74
describe the disk data format and the way of accessing the (possibly
865
The byte storage facilities are addressed via tuples, which we refer to
866
as 'keys' throughout the code base. Revision_keys, inventory_keys and
867
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
868
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
869
byte string made up of a hash identifier and a hash value.
870
We use this interface because it allows low friction with the underlying
871
code that implements disk indices, network encoding and other parts of
874
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
875
the serialised revisions for the repository. This can be used to obtain
876
revision graph information or to access raw serialised revisions.
877
The result of trying to insert data into the repository via this store
878
is undefined: it should be considered read-only except for implementors
880
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
881
the serialised signatures for the repository. This can be used to
882
obtain access to raw serialised signatures. The result of trying to
883
insert data into the repository via this store is undefined: it should
884
be considered read-only except for implementors of repositories.
885
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
886
the serialised inventories for the repository. This can be used to
887
obtain unserialised inventories. The result of trying to insert data
888
into the repository via this store is undefined: it should be
889
considered read-only except for implementors of repositories.
890
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
891
texts of files and directories for the repository. This can be used to
892
obtain file texts or file graphs. Note that Repository.iter_file_bytes
893
is usually a better interface for accessing file texts.
894
The result of trying to insert data into the repository via this store
895
is undefined: it should be considered read-only except for implementors
897
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
898
any data the repository chooses to store or have indexed by its hash.
899
The result of trying to insert data into the repository via this store
900
is undefined: it should be considered read-only except for implementors
902
:ivar _transport: Transport for file access to repository, typically
903
pointing to .bzr/repository.
906
# What class to use for a CommitBuilder. Often its simpler to change this
907
# in a Repository class subclass rather than to override
908
# get_commit_builder.
909
_commit_builder_class = CommitBuilder
910
# The search regex used by xml based repositories to determine what things
911
# where changed in a single commit.
912
78
_file_ids_altered_regex = lazy_regex.lazy_compile(
913
79
r'file_id="(?P<file_id>[^"]+)"'
914
r'.* revision="(?P<revision_id>[^"]+)"'
80
r'.*revision="(?P<revision_id>[^"]+)"'
917
def abort_write_group(self, suppress_errors=False):
918
"""Commit the contents accrued within the current write group.
920
:param suppress_errors: if true, abort_write_group will catch and log
921
unexpected errors that happen during the abort, rather than
922
allowing them to propagate. Defaults to False.
924
:seealso: start_write_group.
926
if self._write_group is not self.get_transaction():
927
# has an unlock or relock occured ?
930
'(suppressed) mismatched lock context and write group. %r, %r',
931
self._write_group, self.get_transaction())
933
raise errors.BzrError(
934
'mismatched lock context and write group. %r, %r' %
935
(self._write_group, self.get_transaction()))
937
self._abort_write_group()
938
except Exception, exc:
939
self._write_group = None
940
if not suppress_errors:
942
mutter('abort_write_group failed')
943
log_exception_quietly()
944
note('bzr: ERROR (ignored): %s', exc)
945
self._write_group = None
947
def _abort_write_group(self):
948
"""Template method for per-repository write group cleanup.
950
This is called during abort before the write group is considered to be
951
finished and should cleanup any internal state accrued during the write
952
group. There is no requirement that data handed to the repository be
953
*not* made available - this is not a rollback - but neither should any
954
attempt be made to ensure that data added is fully commited. Abort is
955
invoked when an error has occured so futher disk or network operations
956
may not be possible or may error and if possible should not be
960
def add_fallback_repository(self, repository):
961
"""Add a repository to use for looking up data not held locally.
963
:param repository: A repository.
965
if not self._format.supports_external_lookups:
966
raise errors.UnstackableRepositoryFormat(self._format, self.base)
968
# This repository will call fallback.unlock() when we transition to
969
# the unlocked state, so we make sure to increment the lock count
970
repository.lock_read()
971
self._check_fallback_repository(repository)
972
self._fallback_repositories.append(repository)
973
self.texts.add_fallback_versioned_files(repository.texts)
974
self.inventories.add_fallback_versioned_files(repository.inventories)
975
self.revisions.add_fallback_versioned_files(repository.revisions)
976
self.signatures.add_fallback_versioned_files(repository.signatures)
977
if self.chk_bytes is not None:
978
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
980
def _check_fallback_repository(self, repository):
981
"""Check that this repository can fallback to repository safely.
983
Raise an error if not.
985
:param repository: A repository to fallback to.
987
return InterRepository._assert_same_model(self, repository)
989
84
def add_inventory(self, revision_id, inv, parents):
990
85
"""Add the inventory inv to the repository as revision_id.
992
87
:param parents: The revision ids of the parents that revision_id
993
88
is known to have and are in the repository already.
995
:returns: The validator(which is a sha1 digest, though what is sha'd is
996
repository format specific) of the serialized inventory.
90
returns the sha1 of the serialized inventory.
998
if not self.is_in_write_group():
999
raise AssertionError("%r not in write group" % (self,))
92
revision_id = osutils.safe_revision_id(revision_id)
1000
93
_mod_revision.check_not_reserved_id(revision_id)
1001
if not (inv.revision_id is None or inv.revision_id == revision_id):
1002
raise AssertionError(
1003
"Mismatch between inventory revision"
1004
" id and insertion revid (%r, %r)"
1005
% (inv.revision_id, revision_id))
1006
if inv.root is None:
1007
raise AssertionError()
1008
return self._add_inventory_checked(revision_id, inv, parents)
1010
def _add_inventory_checked(self, revision_id, inv, parents):
1011
"""Add inv to the repository after checking the inputs.
1013
This function can be overridden to allow different inventory styles.
1015
:seealso: add_inventory, for the contract.
1017
inv_lines = self._serialise_inventory_to_lines(inv)
1018
return self._inventory_add_lines(revision_id, parents,
1019
inv_lines, check_content=False)
1021
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1022
parents, basis_inv=None, propagate_caches=False):
1023
"""Add a new inventory expressed as a delta against another revision.
1025
See the inventory developers documentation for the theory behind
1028
:param basis_revision_id: The inventory id the delta was created
1029
against. (This does not have to be a direct parent.)
1030
:param delta: The inventory delta (see Inventory.apply_delta for
1032
:param new_revision_id: The revision id that the inventory is being
1034
:param parents: The revision ids of the parents that revision_id is
1035
known to have and are in the repository already. These are supplied
1036
for repositories that depend on the inventory graph for revision
1037
graph access, as well as for those that pun ancestry with delta
1039
:param basis_inv: The basis inventory if it is already known,
1041
:param propagate_caches: If True, the caches for this inventory are
1042
copied to and updated for the result if possible.
1044
:returns: (validator, new_inv)
1045
The validator(which is a sha1 digest, though what is sha'd is
1046
repository format specific) of the serialized inventory, and the
1047
resulting inventory.
1049
if not self.is_in_write_group():
1050
raise AssertionError("%r not in write group" % (self,))
1051
_mod_revision.check_not_reserved_id(new_revision_id)
1052
basis_tree = self.revision_tree(basis_revision_id)
1053
basis_tree.lock_read()
1055
# Note that this mutates the inventory of basis_tree, which not all
1056
# inventory implementations may support: A better idiom would be to
1057
# return a new inventory, but as there is no revision tree cache in
1058
# repository this is safe for now - RBC 20081013
1059
if basis_inv is None:
1060
basis_inv = basis_tree.inventory
1061
basis_inv.apply_delta(delta)
1062
basis_inv.revision_id = new_revision_id
1063
return (self.add_inventory(new_revision_id, basis_inv, parents),
1068
def _inventory_add_lines(self, revision_id, parents, lines,
1069
check_content=True):
1070
"""Store lines in inv_vf and return the sha1 of the inventory."""
1071
parents = [(parent,) for parent in parents]
1072
result = self.inventories.add_lines((revision_id,), parents, lines,
1073
check_content=check_content)[0]
1074
self.inventories._access.flush()
94
assert inv.revision_id is None or inv.revision_id == revision_id, \
95
"Mismatch between inventory revision" \
96
" id and insertion revid (%r, %r)" % (inv.revision_id, revision_id)
97
assert inv.root is not None
98
inv_text = self.serialise_inventory(inv)
99
inv_sha1 = osutils.sha_string(inv_text)
100
inv_vf = self.control_weaves.get_weave('inventory',
101
self.get_transaction())
102
self._inventory_add_lines(inv_vf, revision_id, parents,
103
osutils.split_lines(inv_text))
106
def _inventory_add_lines(self, inv_vf, revision_id, parents, lines):
108
for parent in parents:
110
final_parents.append(parent)
112
inv_vf.add_lines(revision_id, final_parents, lines)
1077
115
def add_revision(self, revision_id, rev, inv=None, config=None):
1078
116
"""Add rev to the revision store as revision_id.
1525
311
control = bzrdir.BzrDir.open(base)
1526
312
return control.open_repository()
1528
def copy_content_into(self, destination, revision_id=None):
314
def copy_content_into(self, destination, revision_id=None, basis=None):
1529
315
"""Make a complete copy of the content in self into destination.
1531
This is a destructive operation! Do not use it on existing
317
This is a destructive operation! Do not use it on existing
1534
return InterRepository.get(self, destination).copy_content(revision_id)
1536
def commit_write_group(self):
1537
"""Commit the contents accrued within the current write group.
1539
:seealso: start_write_group.
1541
:return: it may return an opaque hint that can be passed to 'pack'.
1543
if self._write_group is not self.get_transaction():
1544
# has an unlock or relock occured ?
1545
raise errors.BzrError('mismatched lock context %r and '
1547
(self.get_transaction(), self._write_group))
1548
result = self._commit_write_group()
1549
self._write_group = None
1552
def _commit_write_group(self):
1553
"""Template method for per-repository write group cleanup.
1555
This is called before the write group is considered to be
1556
finished and should ensure that all data handed to the repository
1557
for writing during the write group is safely committed (to the
1558
extent possible considering file system caching etc).
1561
def suspend_write_group(self):
1562
raise errors.UnsuspendableWriteGroup(self)
1564
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1565
"""Return the keys of missing inventory parents for revisions added in
1568
A revision is not complete if the inventory delta for that revision
1569
cannot be calculated. Therefore if the parent inventories of a
1570
revision are not present, the revision is incomplete, and e.g. cannot
1571
be streamed by a smart server. This method finds missing inventory
1572
parents for revisions added in this write group.
1574
if not self._format.supports_external_lookups:
1575
# This is only an issue for stacked repositories
1577
if not self.is_in_write_group():
1578
raise AssertionError('not in a write group')
1580
# XXX: We assume that every added revision already has its
1581
# corresponding inventory, so we only check for parent inventories that
1582
# might be missing, rather than all inventories.
1583
parents = set(self.revisions._index.get_missing_parents())
1584
parents.discard(_mod_revision.NULL_REVISION)
1585
unstacked_inventories = self.inventories._index
1586
present_inventories = unstacked_inventories.get_parent_map(
1587
key[-1:] for key in parents)
1588
parents.difference_update(present_inventories)
1589
if len(parents) == 0:
1590
# No missing parent inventories.
1592
if not check_for_missing_texts:
1593
return set(('inventories', rev_id) for (rev_id,) in parents)
1594
# Ok, now we have a list of missing inventories. But these only matter
1595
# if the inventories that reference them are missing some texts they
1596
# appear to introduce.
1597
# XXX: Texts referenced by all added inventories need to be present,
1598
# but at the moment we're only checking for texts referenced by
1599
# inventories at the graph's edge.
1600
key_deps = self.revisions._index._key_dependencies
1601
key_deps.add_keys(present_inventories)
1602
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1603
file_ids = self.fileids_altered_by_revision_ids(referrers)
1604
missing_texts = set()
1605
for file_id, version_ids in file_ids.iteritems():
1606
missing_texts.update(
1607
(file_id, version_id) for version_id in version_ids)
1608
present_texts = self.texts.get_parent_map(missing_texts)
1609
missing_texts.difference_update(present_texts)
1610
if not missing_texts:
1611
# No texts are missing, so all revisions and their deltas are
1614
# Alternatively the text versions could be returned as the missing
1615
# keys, but this is likely to be less data.
1616
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1619
def refresh_data(self):
1620
"""Re-read any data needed to to synchronise with disk.
1622
This method is intended to be called after another repository instance
1623
(such as one used by a smart server) has inserted data into the
1624
repository. It may not be called during a write group, but may be
1625
called at any other time.
1627
if self.is_in_write_group():
1628
raise errors.InternalBzrError(
1629
"May not refresh_data while in a write group.")
1630
self._refresh_data()
1632
def resume_write_group(self, tokens):
1633
if not self.is_write_locked():
1634
raise errors.NotWriteLocked(self)
1635
if self._write_group:
1636
raise errors.BzrError('already in a write group')
1637
self._resume_write_group(tokens)
1638
# so we can detect unlock/relock - the write group is now entered.
1639
self._write_group = self.get_transaction()
1641
def _resume_write_group(self, tokens):
1642
raise errors.UnsuspendableWriteGroup(self)
1644
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
320
revision_id = osutils.safe_revision_id(revision_id)
321
return InterRepository.get(self, destination).copy_content(revision_id, basis)
323
def fetch(self, source, revision_id=None, pb=None):
1646
324
"""Fetch the content required to construct revision_id from source.
1648
If revision_id is None and fetch_spec is None, then all content is
1651
fetch() may not be used when the repository is in a write group -
1652
either finish the current write group before using fetch, or use
1653
fetch before starting the write group.
1655
:param find_ghosts: Find and copy revisions in the source that are
1656
ghosts in the target (and not reachable directly by walking out to
1657
the first-present revision in target from revision_id).
1658
:param revision_id: If specified, all the content needed for this
1659
revision ID will be copied to the target. Fetch will determine for
1660
itself which content needs to be copied.
1661
:param fetch_spec: If specified, a SearchResult or
1662
PendingAncestryResult that describes which revisions to copy. This
1663
allows copying multiple heads at once. Mutually exclusive with
326
If revision_id is None all content is copied.
1666
if fetch_spec is not None and revision_id is not None:
1667
raise AssertionError(
1668
"fetch_spec and revision_id are mutually exclusive.")
1669
if self.is_in_write_group():
1670
raise errors.InternalBzrError(
1671
"May not fetch while in a write group.")
1672
# fast path same-url fetch operations
1673
# TODO: lift out to somewhere common with RemoteRepository
1674
# <https://bugs.edge.launchpad.net/bzr/+bug/401646>
1675
if (self.has_same_location(source)
1676
and fetch_spec is None
1677
and self._has_same_fallbacks(source)):
1678
# check that last_revision is in 'from' and then return a
1680
if (revision_id is not None and
1681
not _mod_revision.is_null(revision_id)):
1682
self.get_revision(revision_id)
1684
# if there is no specific appropriate InterRepository, this will get
1685
# the InterRepository base class, which raises an
1686
# IncompatibleRepositories when asked to fetch.
1687
inter = InterRepository.get(source, self)
1688
return inter.fetch(revision_id=revision_id, pb=pb,
1689
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1691
def create_bundle(self, target, base, fileobj, format=None):
1692
return serializer.write_bundle(self, target, base, fileobj, format)
1694
def get_commit_builder(self, branch, parents, config, timestamp=None,
1695
timezone=None, committer=None, revprops=None,
328
revision_id = osutils.safe_revision_id(revision_id)
329
return InterRepository.get(source, self).fetch(revision_id=revision_id,
332
def get_commit_builder(self, branch, parents, config, timestamp=None,
333
timezone=None, committer=None, revprops=None,
1696
334
revision_id=None):
1697
335
"""Obtain a CommitBuilder for this repository.
1699
337
:param branch: Branch to commit to.
1700
338
:param parents: Revision ids of the parents of the new revision.
1701
339
:param config: Configuration to use.
1790
366
# Most control formats need the repository to be specifically
1791
367
# created, but on some old all-in-one formats it's not needed
1793
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
369
dest_repo = self._format.initialize(a_bzrdir, shared=self.is_shared())
1794
370
except errors.UninitializableFormat:
1795
371
dest_repo = a_bzrdir.open_repository()
372
self.copy_content_into(dest_repo, revision_id, basis)
1796
373
return dest_repo
1798
def _get_sink(self):
1799
"""Return a sink for streaming into this repository."""
1800
return StreamSink(self)
1802
def _get_source(self, to_format):
1803
"""Return a source for streaming from this repository."""
1804
return StreamSource(self, to_format)
1806
375
@needs_read_lock
1807
376
def has_revision(self, revision_id):
1808
377
"""True if this repository has a copy of the revision."""
1809
return revision_id in self.has_revisions((revision_id,))
1812
def has_revisions(self, revision_ids):
1813
"""Probe to find out the presence of multiple revisions.
1815
:param revision_ids: An iterable of revision_ids.
1816
:return: A set of the revision_ids that were present.
1818
parent_map = self.revisions.get_parent_map(
1819
[(rev_id,) for rev_id in revision_ids])
1821
if _mod_revision.NULL_REVISION in revision_ids:
1822
result.add(_mod_revision.NULL_REVISION)
1823
result.update([key[0] for key in parent_map])
1827
def get_revision(self, revision_id):
1828
"""Return the Revision object for a named revision."""
1829
return self.get_revisions([revision_id])[0]
378
revision_id = osutils.safe_revision_id(revision_id)
379
return self._revision_store.has_revision_id(revision_id,
380
self.get_transaction())
1831
382
@needs_read_lock
1832
383
def get_revision_reconcile(self, revision_id):
1833
384
"""'reconcile' helper routine that allows access to a revision always.
1835
386
This variant of get_revision does not cross check the weave graph
1836
387
against the revision one as get_revision does: but it should only
1837
388
be used by reconcile, or reconcile-alike commands that are correcting
1838
389
or testing the revision graph.
1840
return self._get_revisions([revision_id])[0]
391
if not revision_id or not isinstance(revision_id, basestring):
392
raise errors.InvalidRevisionId(revision_id=revision_id,
394
return self.get_revisions([revision_id])[0]
1842
396
@needs_read_lock
1843
397
def get_revisions(self, revision_ids):
1844
"""Get many revisions at once.
1846
Repositories that need to check data on every revision read should
1847
subclass this method.
1849
return self._get_revisions(revision_ids)
1852
def _get_revisions(self, revision_ids):
1853
"""Core work logic to get many revisions without sanity checks."""
1855
for revid, rev in self._iter_revisions(revision_ids):
1857
raise errors.NoSuchRevision(self, revid)
1859
return [revs[revid] for revid in revision_ids]
1861
def _iter_revisions(self, revision_ids):
1862
"""Iterate over revision objects.
1864
:param revision_ids: An iterable of revisions to examine. None may be
1865
passed to request all revisions known to the repository. Note that
1866
not all repositories can find unreferenced revisions; for those
1867
repositories only referenced ones will be returned.
1868
:return: An iterator of (revid, revision) tuples. Absent revisions (
1869
those asked for but not available) are returned as (revid, None).
1871
if revision_ids is None:
1872
revision_ids = self.all_revision_ids()
1874
for rev_id in revision_ids:
1875
if not rev_id or not isinstance(rev_id, basestring):
1876
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1877
keys = [(key,) for key in revision_ids]
1878
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1879
for record in stream:
1880
revid = record.key[0]
1881
if record.storage_kind == 'absent':
1884
text = record.get_bytes_as('fulltext')
1885
rev = self._serializer.read_revision_from_string(text)
398
revision_ids = [osutils.safe_revision_id(r) for r in revision_ids]
399
revs = self._revision_store.get_revisions(revision_ids,
400
self.get_transaction())
402
assert not isinstance(rev.revision_id, unicode)
403
for parent_id in rev.parent_ids:
404
assert not isinstance(parent_id, unicode)
1888
407
@needs_read_lock
1889
408
def get_revision_xml(self, revision_id):
1890
409
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1891
410
# would have already do it.
1892
411
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1893
# TODO: this can't just be replaced by:
1894
# return self._serializer.write_revision_to_string(
1895
# self.get_revision(revision_id))
1896
# as cStringIO preservers the encoding unlike write_revision_to_string
1897
# or some other call down the path.
412
revision_id = osutils.safe_revision_id(revision_id)
1898
413
rev = self.get_revision(revision_id)
1899
rev_tmp = cStringIO.StringIO()
1900
415
# the current serializer..
1901
self._serializer.write_revision(rev, rev_tmp)
416
self._revision_store._serializer.write_revision(rev, rev_tmp)
1903
418
return rev_tmp.getvalue()
1905
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
421
def get_revision(self, revision_id):
422
"""Return the Revision object for a named revision"""
423
# TODO: jam 20070210 get_revision_reconcile should do this for us
424
revision_id = osutils.safe_revision_id(revision_id)
425
r = self.get_revision_reconcile(revision_id)
426
# weave corruption can lead to absent revision markers that should be
428
# the following test is reasonably cheap (it needs a single weave read)
429
# and the weave is cached in read transactions. In write transactions
430
# it is not cached but typically we only read a small number of
431
# revisions. For knits when they are introduced we will probably want
432
# to ensure that caching write transactions are in use.
433
inv = self.get_inventory_weave()
434
self._check_revision_parents(r, inv)
438
def get_deltas_for_revisions(self, revisions):
1906
439
"""Produce a generator of revision deltas.
1908
441
Note that the input is a sequence of REVISIONS, not revision_ids.
1909
442
Trees will be held in memory until the generator exits.
1910
443
Each delta is relative to the revision's lefthand predecessor.
1912
:param specific_fileids: if not None, the result is filtered
1913
so that only those file-ids, their parents and their
1914
children are included.
1916
# Get the revision-ids of interest
1917
445
required_trees = set()
1918
446
for revision in revisions:
1919
447
required_trees.add(revision.revision_id)
1920
448
required_trees.update(revision.parent_ids[:1])
1922
# Get the matching filtered trees. Note that it's more
1923
# efficient to pass filtered trees to changes_from() rather
1924
# than doing the filtering afterwards. changes_from() could
1925
# arguably do the filtering itself but it's path-based, not
1926
# file-id based, so filtering before or afterwards is
1928
if specific_fileids is None:
1929
trees = dict((t.get_revision_id(), t) for
1930
t in self.revision_trees(required_trees))
1932
trees = dict((t.get_revision_id(), t) for
1933
t in self._filtered_revision_trees(required_trees,
1936
# Calculate the deltas
449
trees = dict((t.get_revision_id(), t) for
450
t in self.revision_trees(required_trees))
1937
451
for revision in revisions:
1938
452
if not revision.parent_ids:
1939
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
453
old_tree = self.revision_tree(None)
1941
455
old_tree = trees[revision.parent_ids[0]]
1942
456
yield trees[revision.revision_id].changes_from(old_tree)
1944
458
@needs_read_lock
1945
def get_revision_delta(self, revision_id, specific_fileids=None):
459
def get_revision_delta(self, revision_id):
1946
460
"""Return the delta for one revision.
1948
462
The delta is relative to the left-hand predecessor of the
1951
:param specific_fileids: if not None, the result is filtered
1952
so that only those file-ids, their parents and their
1953
children are included.
1955
465
r = self.get_revision(revision_id)
1956
return list(self.get_deltas_for_revisions([r],
1957
specific_fileids=specific_fileids))[0]
466
return list(self.get_deltas_for_revisions([r]))[0]
468
def _check_revision_parents(self, revision, inventory):
469
"""Private to Repository and Fetch.
471
This checks the parentage of revision in an inventory weave for
472
consistency and is only applicable to inventory-weave-for-ancestry
473
using repository formats & fetchers.
475
weave_parents = inventory.get_parents(revision.revision_id)
476
weave_names = inventory.versions()
477
for parent_id in revision.parent_ids:
478
if parent_id in weave_names:
479
# this parent must not be a ghost.
480
if not parent_id in weave_parents:
482
raise errors.CorruptRepository(self)
1959
484
@needs_write_lock
1960
485
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
486
revision_id = osutils.safe_revision_id(revision_id)
1961
487
signature = gpg_strategy.sign(plaintext)
1962
self.add_signature_text(revision_id, signature)
1965
def add_signature_text(self, revision_id, signature):
1966
self.signatures.add_lines((revision_id,), (),
1967
osutils.split_lines(signature))
1969
def find_text_key_references(self):
1970
"""Find the text key references within the repository.
1972
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1973
to whether they were referred to by the inventory of the
1974
revision_id that they contain. The inventory texts from all present
1975
revision ids are assessed to generate this report.
1977
revision_keys = self.revisions.keys()
1978
w = self.inventories
1979
pb = ui.ui_factory.nested_progress_bar()
1981
return self._find_text_key_references_from_xml_inventory_lines(
1982
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1986
def _find_text_key_references_from_xml_inventory_lines(self,
1988
"""Core routine for extracting references to texts from inventories.
1990
This performs the translation of xml lines to revision ids.
1992
:param line_iterator: An iterator of lines, origin_version_id
1993
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1994
to whether they were referred to by the inventory of the
1995
revision_id that they contain. Note that if that revision_id was
1996
not part of the line_iterator's output then False will be given -
1997
even though it may actually refer to that key.
1999
if not self._serializer.support_altered_by_hack:
2000
raise AssertionError(
2001
"_find_text_key_references_from_xml_inventory_lines only "
2002
"supported for branches which store inventory as unnested xml"
2003
", not on %r" % self)
488
self._revision_store.add_revision_signature_text(revision_id,
490
self.get_transaction())
492
def fileids_altered_by_revision_ids(self, revision_ids):
493
"""Find the file ids and versions affected by revisions.
495
:param revisions: an iterable containing revision ids.
496
:return: a dictionary mapping altered file-ids to an iterable of
497
revision_ids. Each altered file-ids has the exact revision_ids that
498
altered it listed explicitly.
500
assert self._serializer.support_altered_by_hack, \
501
("fileids_altered_by_revision_ids only supported for branches "
502
"which store inventory as unnested xml, not on %r" % self)
503
selected_revision_ids = set(osutils.safe_revision_id(r)
504
for r in revision_ids)
505
w = self.get_inventory_weave()
2006
508
# this code needs to read every new line in every inventory for the
2007
509
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2008
# not present in one of those inventories is unnecessary but not
510
# not present in one of those inventories is unnecessary but not
2009
511
# harmful because we are filtering by the revision id marker in the
2010
# inventory lines : we only select file ids altered in one of those
512
# inventory lines : we only select file ids altered in one of those
2011
513
# revisions. We don't need to see all lines in the inventory because
2012
514
# only those added in an inventory in rev X can contain a revision=X
2022
524
search = self._file_ids_altered_regex.search
2023
525
unescape = _unescape_xml
2024
526
setdefault = result.setdefault
2025
for line, line_key in line_iterator:
2026
match = search(line)
2029
# One call to match.group() returning multiple items is quite a
2030
# bit faster than 2 calls to match.group() each returning 1
2031
file_id, revision_id = match.group('file_id', 'revision_id')
2033
# Inlining the cache lookups helps a lot when you make 170,000
2034
# lines and 350k ids, versus 8.4 unique ids.
2035
# Using a cache helps in 2 ways:
2036
# 1) Avoids unnecessary decoding calls
2037
# 2) Re-uses cached strings, which helps in future set and
2039
# (2) is enough that removing encoding entirely along with
2040
# the cache (so we are using plain strings) results in no
2041
# performance improvement.
2043
revision_id = unescape_revid_cache[revision_id]
2045
unescaped = unescape(revision_id)
2046
unescape_revid_cache[revision_id] = unescaped
2047
revision_id = unescaped
2049
# Note that unconditionally unescaping means that we deserialise
2050
# every fileid, which for general 'pull' is not great, but we don't
2051
# really want to have some many fulltexts that this matters anyway.
2054
file_id = unescape_fileid_cache[file_id]
2056
unescaped = unescape(file_id)
2057
unescape_fileid_cache[file_id] = unescaped
2060
key = (file_id, revision_id)
2061
setdefault(key, False)
2062
if revision_id == line_key[-1]:
2066
def _inventory_xml_lines_for_keys(self, keys):
2067
"""Get a line iterator of the sort needed for findind references.
2069
Not relevant for non-xml inventory repositories.
2071
Ghosts in revision_keys are ignored.
2073
:param revision_keys: The revision keys for the inventories to inspect.
2074
:return: An iterator over (inventory line, revid) for the fulltexts of
2075
all of the xml inventories specified by revision_keys.
2077
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2078
for record in stream:
2079
if record.storage_kind != 'absent':
2080
chunks = record.get_bytes_as('chunked')
2081
revid = record.key[-1]
2082
lines = osutils.chunks_to_lines(chunks)
2086
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2088
"""Helper routine for fileids_altered_by_revision_ids.
2090
This performs the translation of xml lines to revision ids.
2092
:param line_iterator: An iterator of lines, origin_version_id
2093
:param revision_keys: The revision ids to filter for. This should be a
2094
set or other type which supports efficient __contains__ lookups, as
2095
the revision key from each parsed line will be looked up in the
2096
revision_keys filter.
2097
:return: a dictionary mapping altered file-ids to an iterable of
2098
revision_ids. Each altered file-ids has the exact revision_ids that
2099
altered it listed explicitly.
2101
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2102
line_iterator).iterkeys())
2103
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2104
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2105
self._inventory_xml_lines_for_keys(parent_keys)))
2106
new_keys = seen - parent_seen
2108
setdefault = result.setdefault
2109
for key in new_keys:
2110
setdefault(key[0], set()).add(key[-1])
2113
def _find_parent_ids_of_revisions(self, revision_ids):
2114
"""Find all parent ids that are mentioned in the revision graph.
2116
:return: set of revisions that are parents of revision_ids which are
2117
not part of revision_ids themselves
2119
parent_map = self.get_parent_map(revision_ids)
2121
map(parent_ids.update, parent_map.itervalues())
2122
parent_ids.difference_update(revision_ids)
2123
parent_ids.discard(_mod_revision.NULL_REVISION)
2126
def _find_parent_keys_of_revisions(self, revision_keys):
2127
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2129
:param revision_keys: An iterable of revision_keys.
2130
:return: The parents of all revision_keys that are not already in
2133
parent_map = self.revisions.get_parent_map(revision_keys)
2135
map(parent_keys.update, parent_map.itervalues())
2136
parent_keys.difference_update(revision_keys)
2137
parent_keys.discard(_mod_revision.NULL_REVISION)
2140
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2141
"""Find the file ids and versions affected by revisions.
2143
:param revisions: an iterable containing revision ids.
2144
:param _inv_weave: The inventory weave from this repository or None.
2145
If None, the inventory weave will be opened automatically.
2146
:return: a dictionary mapping altered file-ids to an iterable of
2147
revision_ids. Each altered file-ids has the exact revision_ids that
2148
altered it listed explicitly.
2150
selected_keys = set((revid,) for revid in revision_ids)
2151
w = _inv_weave or self.inventories
2152
pb = ui.ui_factory.nested_progress_bar()
2154
return self._find_file_ids_from_xml_inventory_lines(
2155
w.iter_lines_added_or_present_in_keys(
2156
selected_keys, pb=pb),
2161
def iter_files_bytes(self, desired_files):
2162
"""Iterate through file versions.
2164
Files will not necessarily be returned in the order they occur in
2165
desired_files. No specific order is guaranteed.
2167
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2168
value supplied by the caller as part of desired_files. It should
2169
uniquely identify the file version in the caller's context. (Examples:
2170
an index number or a TreeTransform trans_id.)
2172
bytes_iterator is an iterable of bytestrings for the file. The
2173
kind of iterable and length of the bytestrings are unspecified, but for
2174
this implementation, it is a list of bytes produced by
2175
VersionedFile.get_record_stream().
2177
:param desired_files: a list of (file_id, revision_id, identifier)
2181
for file_id, revision_id, callable_data in desired_files:
2182
text_keys[(file_id, revision_id)] = callable_data
2183
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2184
if record.storage_kind == 'absent':
2185
raise errors.RevisionNotPresent(record.key, self)
2186
yield text_keys[record.key], record.get_bytes_as('chunked')
2188
def _generate_text_key_index(self, text_key_references=None,
2190
"""Generate a new text key index for the repository.
2192
This is an expensive function that will take considerable time to run.
2194
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2195
list of parents, also text keys. When a given key has no parents,
2196
the parents list will be [NULL_REVISION].
2198
# All revisions, to find inventory parents.
2199
if ancestors is None:
2200
graph = self.get_graph()
2201
ancestors = graph.get_parent_map(self.all_revision_ids())
2202
if text_key_references is None:
2203
text_key_references = self.find_text_key_references()
2204
pb = ui.ui_factory.nested_progress_bar()
2206
return self._do_generate_text_key_index(ancestors,
2207
text_key_references, pb)
2211
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2212
"""Helper for _generate_text_key_index to avoid deep nesting."""
2213
revision_order = tsort.topo_sort(ancestors)
2214
invalid_keys = set()
2216
for revision_id in revision_order:
2217
revision_keys[revision_id] = set()
2218
text_count = len(text_key_references)
2219
# a cache of the text keys to allow reuse; costs a dict of all the
2220
# keys, but saves a 2-tuple for every child of a given key.
2222
for text_key, valid in text_key_references.iteritems():
2224
invalid_keys.add(text_key)
2226
revision_keys[text_key[1]].add(text_key)
2227
text_key_cache[text_key] = text_key
2228
del text_key_references
2230
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2231
NULL_REVISION = _mod_revision.NULL_REVISION
2232
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2233
# too small for large or very branchy trees. However, for 55K path
2234
# trees, it would be easy to use too much memory trivially. Ideally we
2235
# could gauge this by looking at available real memory etc, but this is
2236
# always a tricky proposition.
2237
inventory_cache = lru_cache.LRUCache(10)
2238
batch_size = 10 # should be ~150MB on a 55K path tree
2239
batch_count = len(revision_order) / batch_size + 1
2241
pb.update("Calculating text parents", processed_texts, text_count)
2242
for offset in xrange(batch_count):
2243
to_query = revision_order[offset * batch_size:(offset + 1) *
2247
for revision_id in to_query:
2248
parent_ids = ancestors[revision_id]
2249
for text_key in revision_keys[revision_id]:
2250
pb.update("Calculating text parents", processed_texts)
2251
processed_texts += 1
2252
candidate_parents = []
2253
for parent_id in parent_ids:
2254
parent_text_key = (text_key[0], parent_id)
2256
check_parent = parent_text_key not in \
2257
revision_keys[parent_id]
2259
# the parent parent_id is a ghost:
2260
check_parent = False
2261
# truncate the derived graph against this ghost.
2262
parent_text_key = None
2264
# look at the parent commit details inventories to
2265
# determine possible candidates in the per file graph.
2268
inv = inventory_cache[parent_id]
2270
inv = self.revision_tree(parent_id).inventory
2271
inventory_cache[parent_id] = inv
2273
parent_entry = inv[text_key[0]]
2274
except (KeyError, errors.NoSuchId):
2276
if parent_entry is not None:
2278
text_key[0], parent_entry.revision)
2280
parent_text_key = None
2281
if parent_text_key is not None:
2282
candidate_parents.append(
2283
text_key_cache[parent_text_key])
2284
parent_heads = text_graph.heads(candidate_parents)
2285
new_parents = list(parent_heads)
2286
new_parents.sort(key=lambda x:candidate_parents.index(x))
2287
if new_parents == []:
2288
new_parents = [NULL_REVISION]
2289
text_index[text_key] = new_parents
2291
for text_key in invalid_keys:
2292
text_index[text_key] = [NULL_REVISION]
2295
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2296
"""Get an iterable listing the keys of all the data introduced by a set
2299
The keys will be ordered so that the corresponding items can be safely
2300
fetched and inserted in that order.
2302
:returns: An iterable producing tuples of (knit-kind, file-id,
2303
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2304
'revisions'. file-id is None unless knit-kind is 'file'.
2306
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2309
for result in self._find_non_file_keys_to_fetch(revision_ids):
2312
def _find_file_keys_to_fetch(self, revision_ids, pb):
2313
# XXX: it's a bit weird to control the inventory weave caching in this
2314
# generator. Ideally the caching would be done in fetch.py I think. Or
2315
# maybe this generator should explicitly have the contract that it
2316
# should not be iterated until the previously yielded item has been
2318
inv_w = self.inventories
2320
# file ids that changed
2321
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2323
num_file_ids = len(file_ids)
2324
for file_id, altered_versions in file_ids.iteritems():
2326
pb.update("fetch texts", count, num_file_ids)
2328
yield ("file", file_id, altered_versions)
2330
def _find_non_file_keys_to_fetch(self, revision_ids):
2332
yield ("inventory", None, revision_ids)
2335
# XXX: Note ATM no callers actually pay attention to this return
2336
# instead they just use the list of revision ids and ignore
2337
# missing sigs. Consider removing this work entirely
2338
revisions_with_signatures = set(self.signatures.get_parent_map(
2339
[(r,) for r in revision_ids]))
2340
revisions_with_signatures = set(
2341
[r for (r,) in revisions_with_signatures])
2342
revisions_with_signatures.intersection_update(revision_ids)
2343
yield ("signatures", None, revisions_with_signatures)
2346
yield ("revisions", None, revision_ids)
527
pb = ui.ui_factory.nested_progress_bar()
529
for line in w.iter_lines_added_or_present_in_versions(
530
selected_revision_ids, pb=pb):
534
# One call to match.group() returning multiple items is quite a
535
# bit faster than 2 calls to match.group() each returning 1
536
file_id, revision_id = match.group('file_id', 'revision_id')
538
# Inlining the cache lookups helps a lot when you make 170,000
539
# lines and 350k ids, versus 8.4 unique ids.
540
# Using a cache helps in 2 ways:
541
# 1) Avoids unnecessary decoding calls
542
# 2) Re-uses cached strings, which helps in future set and
544
# (2) is enough that removing encoding entirely along with
545
# the cache (so we are using plain strings) results in no
546
# performance improvement.
548
revision_id = unescape_revid_cache[revision_id]
550
unescaped = unescape(revision_id)
551
unescape_revid_cache[revision_id] = unescaped
552
revision_id = unescaped
554
if revision_id in selected_revision_ids:
556
file_id = unescape_fileid_cache[file_id]
558
unescaped = unescape(file_id)
559
unescape_fileid_cache[file_id] = unescaped
561
setdefault(file_id, set()).add(revision_id)
567
def get_inventory_weave(self):
568
return self.control_weaves.get_weave('inventory',
569
self.get_transaction())
2348
571
@needs_read_lock
2349
572
def get_inventory(self, revision_id):
2350
"""Get Inventory object by revision id."""
2351
return self.iter_inventories([revision_id]).next()
2353
def iter_inventories(self, revision_ids, ordering=None):
2354
"""Get many inventories by revision_ids.
2356
This will buffer some or all of the texts used in constructing the
2357
inventories in memory, but will only parse a single inventory at a
2360
:param revision_ids: The expected revision ids of the inventories.
2361
:param ordering: optional ordering, e.g. 'topological'. If not
2362
specified, the order of revision_ids will be preserved (by
2363
buffering if necessary).
2364
:return: An iterator of inventories.
2366
if ((None in revision_ids)
2367
or (_mod_revision.NULL_REVISION in revision_ids)):
2368
raise ValueError('cannot get null revision inventory')
2369
return self._iter_inventories(revision_ids, ordering)
2371
def _iter_inventories(self, revision_ids, ordering):
2372
"""single-document based inventory iteration."""
2373
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2374
for text, revision_id in inv_xmls:
2375
yield self.deserialise_inventory(revision_id, text)
2377
def _iter_inventory_xmls(self, revision_ids, ordering):
2378
if ordering is None:
2379
order_as_requested = True
2380
ordering = 'unordered'
2382
order_as_requested = False
2383
keys = [(revision_id,) for revision_id in revision_ids]
2386
if order_as_requested:
2387
key_iter = iter(keys)
2388
next_key = key_iter.next()
2389
stream = self.inventories.get_record_stream(keys, ordering, True)
2391
for record in stream:
2392
if record.storage_kind != 'absent':
2393
chunks = record.get_bytes_as('chunked')
2394
if order_as_requested:
2395
text_chunks[record.key] = chunks
2397
yield ''.join(chunks), record.key[-1]
2399
raise errors.NoSuchRevision(self, record.key)
2400
if order_as_requested:
2401
# Yield as many results as we can while preserving order.
2402
while next_key in text_chunks:
2403
chunks = text_chunks.pop(next_key)
2404
yield ''.join(chunks), next_key[-1]
2406
next_key = key_iter.next()
2407
except StopIteration:
2408
# We still want to fully consume the get_record_stream,
2409
# just in case it is not actually finished at this point
573
"""Get Inventory object by hash."""
574
# TODO: jam 20070210 Technically we don't need to sanitize, since all
575
# called functions must sanitize.
576
revision_id = osutils.safe_revision_id(revision_id)
577
return self.deserialise_inventory(
578
revision_id, self.get_inventory_xml(revision_id))
2413
580
def deserialise_inventory(self, revision_id, xml):
2414
"""Transform the xml into an inventory object.
581
"""Transform the xml into an inventory object.
2416
583
:param revision_id: The expected revision id of the inventory.
2417
584
:param xml: A serialised inventory.
2419
result = self._serializer.read_inventory_from_string(xml, revision_id,
2420
entry_cache=self._inventory_entry_cache)
2421
if result.revision_id != revision_id:
2422
raise AssertionError('revision id mismatch %s != %s' % (
2423
result.revision_id, revision_id))
586
revision_id = osutils.safe_revision_id(revision_id)
587
result = self._serializer.read_inventory_from_string(xml)
588
result.root.revision = revision_id
2426
591
def serialise_inventory(self, inv):
2427
592
return self._serializer.write_inventory_to_string(inv)
2429
def _serialise_inventory_to_lines(self, inv):
2430
return self._serializer.write_inventory_to_lines(inv)
2432
def get_serializer_format(self):
2433
return self._serializer.format_num
2435
594
@needs_read_lock
2436
595
def get_inventory_xml(self, revision_id):
2437
596
"""Get inventory XML as a file object."""
2438
texts = self._iter_inventory_xmls([revision_id], 'unordered')
597
revision_id = osutils.safe_revision_id(revision_id)
2440
text, revision_id = texts.next()
2441
except StopIteration:
599
assert isinstance(revision_id, str), type(revision_id)
600
iw = self.get_inventory_weave()
601
return iw.get_text(revision_id)
2442
603
raise errors.HistoryMissing(self, 'inventory', revision_id)
2445
605
@needs_read_lock
2446
606
def get_inventory_sha1(self, revision_id):
2447
607
"""Return the sha1 hash of the inventory entry
609
# TODO: jam 20070210 Shouldn't this be deprecated / removed?
610
revision_id = osutils.safe_revision_id(revision_id)
2449
611
return self.get_revision(revision_id).inventory_sha1
2451
def get_rev_id_for_revno(self, revno, known_pair):
2452
"""Return the revision id of a revno, given a later (revno, revid)
2453
pair in the same history.
2455
:return: if found (True, revid). If the available history ran out
2456
before reaching the revno, then this returns
2457
(False, (closest_revno, closest_revid)).
2459
known_revno, known_revid = known_pair
2460
partial_history = [known_revid]
2461
distance_from_known = known_revno - revno
2462
if distance_from_known < 0:
2464
'requested revno (%d) is later than given known revno (%d)'
2465
% (revno, known_revno))
2468
self, partial_history, stop_index=distance_from_known)
2469
except errors.RevisionNotPresent, err:
2470
if err.revision_id == known_revid:
2471
# The start revision (known_revid) wasn't found.
2473
# This is a stacked repository with no fallbacks, or a there's a
2474
# left-hand ghost. Either way, even though the revision named in
2475
# the error isn't in this repo, we know it's the next step in this
2476
# left-hand history.
2477
partial_history.append(err.revision_id)
2478
if len(partial_history) <= distance_from_known:
2479
# Didn't find enough history to get a revid for the revno.
2480
earliest_revno = known_revno - len(partial_history) + 1
2481
return (False, (earliest_revno, partial_history[-1]))
2482
if len(partial_history) - 1 > distance_from_known:
2483
raise AssertionError('_iter_for_revno returned too much history')
2484
return (True, partial_history[-1])
614
def get_revision_graph(self, revision_id=None):
615
"""Return a dictionary containing the revision graph.
617
:param revision_id: The revision_id to get a graph from. If None, then
618
the entire revision graph is returned. This is a deprecated mode of
619
operation and will be removed in the future.
620
:return: a dictionary of revision_id->revision_parents_list.
622
# special case NULL_REVISION
623
if revision_id == _mod_revision.NULL_REVISION:
625
revision_id = osutils.safe_revision_id(revision_id)
626
a_weave = self.get_inventory_weave()
627
all_revisions = self._eliminate_revisions_not_present(
629
entire_graph = dict([(node, a_weave.get_parents(node)) for
630
node in all_revisions])
631
if revision_id is None:
633
elif revision_id not in entire_graph:
634
raise errors.NoSuchRevision(self, revision_id)
636
# add what can be reached from revision_id
638
pending = set([revision_id])
639
while len(pending) > 0:
641
result[node] = entire_graph[node]
642
for revision_id in result[node]:
643
if revision_id not in result:
644
pending.add(revision_id)
648
def get_revision_graph_with_ghosts(self, revision_ids=None):
649
"""Return a graph of the revisions with ghosts marked as applicable.
651
:param revision_ids: an iterable of revisions to graph or None for all.
652
:return: a Graph object with the graph reachable from revision_ids.
654
result = graph.Graph()
656
pending = set(self.all_revision_ids())
659
pending = set(osutils.safe_revision_id(r) for r in revision_ids)
660
# special case NULL_REVISION
661
if _mod_revision.NULL_REVISION in pending:
662
pending.remove(_mod_revision.NULL_REVISION)
663
required = set(pending)
666
revision_id = pending.pop()
668
rev = self.get_revision(revision_id)
669
except errors.NoSuchRevision:
670
if revision_id in required:
673
result.add_ghost(revision_id)
675
for parent_id in rev.parent_ids:
676
# is this queued or done ?
677
if (parent_id not in pending and
678
parent_id not in done):
680
pending.add(parent_id)
681
result.add_node(revision_id, rev.parent_ids)
682
done.add(revision_id)
685
def _get_history_vf(self):
686
"""Get a versionedfile whose history graph reflects all revisions.
688
For weave repositories, this is the inventory weave.
690
return self.get_inventory_weave()
2486
692
def iter_reverse_revision_history(self, revision_id):
2487
693
"""Iterate backwards through revision ids in the lefthand history
2531
736
reconciler = RepoReconciler(self, thorough=thorough)
2532
737
reconciler.reconcile()
2533
738
return reconciler
2535
def _refresh_data(self):
2536
"""Helper called from lock_* to ensure coherency with disk.
2538
The default implementation does nothing; it is however possible
2539
for repositories to maintain loaded indices across multiple locks
2540
by checking inside their implementation of this method to see
2541
whether their indices are still valid. This depends of course on
2542
the disk format being validatable in this manner. This method is
2543
also called by the refresh_data() public interface to cause a refresh
2544
to occur while in a write lock so that data inserted by a smart server
2545
push operation is visible on the client's instance of the physical
2549
740
@needs_read_lock
2550
741
def revision_tree(self, revision_id):
2551
742
"""Return Tree for a revision on this branch.
2553
`revision_id` may be NULL_REVISION for the empty tree revision.
744
`revision_id` may be None for the empty tree revision.
2555
revision_id = _mod_revision.ensure_null(revision_id)
2556
746
# TODO: refactor this to use an existing revision object
2557
747
# so we don't need to read it in twice.
2558
if revision_id == _mod_revision.NULL_REVISION:
2559
return RevisionTree(self, Inventory(root_id=None),
748
if revision_id is None or revision_id == _mod_revision.NULL_REVISION:
749
return RevisionTree(self, Inventory(root_id=None),
2560
750
_mod_revision.NULL_REVISION)
752
revision_id = osutils.safe_revision_id(revision_id)
2562
753
inv = self.get_revision_inventory(revision_id)
2563
754
return RevisionTree(self, inv, revision_id)
2565
757
def revision_trees(self, revision_ids):
2566
"""Return Trees for revisions in this repository.
2568
:param revision_ids: a sequence of revision-ids;
2569
a revision-id may not be None or 'null:'
2571
inventories = self.iter_inventories(revision_ids)
2572
for inv in inventories:
2573
yield RevisionTree(self, inv, inv.revision_id)
2575
def _filtered_revision_trees(self, revision_ids, file_ids):
2576
"""Return Tree for a revision on this branch with only some files.
2578
:param revision_ids: a sequence of revision-ids;
2579
a revision-id may not be None or 'null:'
2580
:param file_ids: if not None, the result is filtered
2581
so that only those file-ids, their parents and their
2582
children are included.
2584
inventories = self.iter_inventories(revision_ids)
2585
for inv in inventories:
2586
# Should we introduce a FilteredRevisionTree class rather
2587
# than pre-filter the inventory here?
2588
filtered_inv = inv.filter(file_ids)
2589
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
758
"""Return Tree for a revision on this branch.
760
`revision_id` may not be None or 'null:'"""
761
assert None not in revision_ids
762
assert _mod_revision.NULL_REVISION not in revision_ids
763
texts = self.get_inventory_weave().get_texts(revision_ids)
764
for text, revision_id in zip(texts, revision_ids):
765
inv = self.deserialise_inventory(revision_id, text)
766
yield RevisionTree(self, inv, revision_id)
2591
768
@needs_read_lock
2592
def get_ancestry(self, revision_id, topo_sorted=True):
769
def get_ancestry(self, revision_id):
2593
770
"""Return a list of revision-ids integrated by a revision.
2595
The first element of the list is always None, indicating the origin
2596
revision. This might change when we have history horizons, or
772
The first element of the list is always None, indicating the origin
773
revision. This might change when we have history horizons, or
2597
774
perhaps we should have a new API.
2599
776
This is topologically sorted.
2601
if _mod_revision.is_null(revision_id):
778
if revision_id is None:
780
revision_id = osutils.safe_revision_id(revision_id)
2603
781
if not self.has_revision(revision_id):
2604
782
raise errors.NoSuchRevision(self, revision_id)
2605
graph = self.get_graph()
2607
search = graph._make_breadth_first_searcher([revision_id])
2610
found, ghosts = search.next_with_ghosts()
2611
except StopIteration:
2614
if _mod_revision.NULL_REVISION in keys:
2615
keys.remove(_mod_revision.NULL_REVISION)
2617
parent_map = graph.get_parent_map(keys)
2618
keys = tsort.topo_sort(parent_map)
2619
return [None] + list(keys)
2621
def pack(self, hint=None):
2622
"""Compress the data within the repository.
2624
This operation only makes sense for some repository types. For other
2625
types it should be a no-op that just returns.
2627
This stub method does not require a lock, but subclasses should use
2628
@needs_write_lock as this is a long running call its reasonable to
2629
implicitly lock for the user.
2631
:param hint: If not supplied, the whole repository is packed.
2632
If supplied, the repository may use the hint parameter as a
2633
hint for the parts of the repository to pack. A hint can be
2634
obtained from the result of commit_write_group(). Out of
2635
date hints are simply ignored, because concurrent operations
2636
can obsolete them rapidly.
783
w = self.get_inventory_weave()
784
candidates = w.get_ancestry(revision_id)
785
return [None] + candidates # self._eliminate_revisions_not_present(candidates)
788
def print_file(self, file, revision_id):
789
"""Print `file` to stdout.
791
FIXME RBC 20060125 as John Meinel points out this is a bad api
792
- it writes to stdout, it assumes that that is valid etc. Fix
793
by creating a new more flexible convenience function.
795
revision_id = osutils.safe_revision_id(revision_id)
796
tree = self.revision_tree(revision_id)
797
# use inventory as it was in that revision
798
file_id = tree.inventory.path2id(file)
800
# TODO: jam 20060427 Write a test for this code path
801
# it had a bug in it, and was raising the wrong
803
raise errors.BzrError("%r is not present in revision %s" % (file, revision_id))
804
tree.print_file(file_id)
2639
806
def get_transaction(self):
2640
807
return self.control_files.get_transaction()
2642
def get_parent_map(self, revision_ids):
2643
"""See graph.StackedParentsProvider.get_parent_map"""
2644
# revisions index works in keys; this just works in revisions
2645
# therefore wrap and unwrap
2648
for revision_id in revision_ids:
2649
if revision_id == _mod_revision.NULL_REVISION:
2650
result[revision_id] = ()
2651
elif revision_id is None:
2652
raise ValueError('get_parent_map(None) is not valid')
2654
query_keys.append((revision_id ,))
2655
for ((revision_id,), parent_keys) in \
2656
self.revisions.get_parent_map(query_keys).iteritems():
2658
result[revision_id] = tuple(parent_revid
2659
for (parent_revid,) in parent_keys)
2661
result[revision_id] = (_mod_revision.NULL_REVISION,)
2664
def _make_parents_provider(self):
2667
def get_graph(self, other_repository=None):
2668
"""Return the graph walker for this repository format"""
2669
parents_provider = self._make_parents_provider()
2670
if (other_repository is not None and
2671
not self.has_same_location(other_repository)):
2672
parents_provider = graph.StackedParentsProvider(
2673
[parents_provider, other_repository._make_parents_provider()])
2674
return graph.Graph(parents_provider)
2676
def _get_versioned_file_checker(self, text_key_references=None,
2678
"""Return an object suitable for checking versioned files.
2680
:param text_key_references: if non-None, an already built
2681
dictionary mapping text keys ((fileid, revision_id) tuples)
2682
to whether they were referred to by the inventory of the
2683
revision_id that they contain. If None, this will be
2685
:param ancestors: Optional result from
2686
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2689
return _VersionedFileChecker(self,
2690
text_key_references=text_key_references, ancestors=ancestors)
2692
def revision_ids_to_search_result(self, result_set):
2693
"""Convert a set of revision ids to a graph SearchResult."""
2694
result_parents = set()
2695
for parents in self.get_graph().get_parent_map(
2696
result_set).itervalues():
2697
result_parents.update(parents)
2698
included_keys = result_set.intersection(result_parents)
2699
start_keys = result_set.difference(included_keys)
2700
exclude_keys = result_parents.difference(result_set)
2701
result = graph.SearchResult(start_keys, exclude_keys,
2702
len(result_set), result_set)
809
def revision_parents(self, revision_id):
810
revision_id = osutils.safe_revision_id(revision_id)
811
return self.get_inventory_weave().parent_names(revision_id)
2705
813
@needs_write_lock
2706
814
def set_make_working_trees(self, new_value):
3397
1291
content is copied.
3398
1292
:param pb: optional progress bar to use for progress reports. If not
3399
1293
provided a default one will be created.
3402
from bzrlib.fetch import RepoFetcher
3403
f = RepoFetcher(to_repository=self.target,
3404
from_repository=self.source,
3405
last_revision=revision_id,
3406
fetch_spec=fetch_spec,
3407
pb=pb, find_ghosts=find_ghosts)
3409
def _walk_to_common_revisions(self, revision_ids):
3410
"""Walk out from revision_ids in source to revisions target has.
3412
:param revision_ids: The start point for the search.
3413
:return: A set of revision ids.
3415
target_graph = self.target.get_graph()
3416
revision_ids = frozenset(revision_ids)
3417
missing_revs = set()
3418
source_graph = self.source.get_graph()
3419
# ensure we don't pay silly lookup costs.
3420
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3421
null_set = frozenset([_mod_revision.NULL_REVISION])
3422
searcher_exhausted = False
3426
# Iterate the searcher until we have enough next_revs
3427
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3429
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3430
next_revs.update(next_revs_part)
3431
ghosts.update(ghosts_part)
3432
except StopIteration:
3433
searcher_exhausted = True
3435
# If there are ghosts in the source graph, and the caller asked for
3436
# them, make sure that they are present in the target.
3437
# We don't care about other ghosts as we can't fetch them and
3438
# haven't been asked to.
3439
ghosts_to_check = set(revision_ids.intersection(ghosts))
3440
revs_to_get = set(next_revs).union(ghosts_to_check)
3442
have_revs = set(target_graph.get_parent_map(revs_to_get))
3443
# we always have NULL_REVISION present.
3444
have_revs = have_revs.union(null_set)
3445
# Check if the target is missing any ghosts we need.
3446
ghosts_to_check.difference_update(have_revs)
3448
# One of the caller's revision_ids is a ghost in both the
3449
# source and the target.
3450
raise errors.NoSuchRevision(
3451
self.source, ghosts_to_check.pop())
3452
missing_revs.update(next_revs - have_revs)
3453
# Because we may have walked past the original stop point, make
3454
# sure everything is stopped
3455
stop_revs = searcher.find_seen_ancestors(have_revs)
3456
searcher.stop_searching_any(stop_revs)
3457
if searcher_exhausted:
3459
return searcher.get_result()
1295
Returns the copied revision count and the failed revisions in a tuple:
1298
raise NotImplementedError(self.fetch)
3461
1300
@needs_read_lock
3462
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1301
def missing_revision_ids(self, revision_id=None):
3463
1302
"""Return the revision ids that source has that target does not.
1304
These are returned in topological order.
3465
1306
:param revision_id: only return revision ids included by this
3467
:param find_ghosts: If True find missing revisions in deep history
3468
rather than just finding the surface difference.
3469
:return: A bzrlib.graph.SearchResult.
3471
# stop searching at found target revisions.
3472
if not find_ghosts and revision_id is not None:
3473
return self._walk_to_common_revisions([revision_id])
3474
1309
# generic, possibly worst case, slow code path.
3475
1310
target_ids = set(self.target.all_revision_ids())
3476
1311
if revision_id is not None:
1312
# TODO: jam 20070210 InterRepository is internal enough that it
1313
# should assume revision_ids are already utf-8
1314
revision_id = osutils.safe_revision_id(revision_id)
3477
1315
source_ids = self.source.get_ancestry(revision_id)
3478
if source_ids[0] is not None:
3479
raise AssertionError()
1316
assert source_ids[0] is None
3480
1317
source_ids.pop(0)
3482
1319
source_ids = self.source.all_revision_ids()
3483
1320
result_set = set(source_ids).difference(target_ids)
3484
return self.source.revision_ids_to_search_result(result_set)
3487
def _same_model(source, target):
3488
"""True if source and target have the same data representation.
3490
Note: this is always called on the base class; overriding it in a
3491
subclass will have no effect.
3494
InterRepository._assert_same_model(source, target)
3496
except errors.IncompatibleRepositories, e:
3500
def _assert_same_model(source, target):
3501
"""Raise an exception if two repositories do not use the same model.
3503
if source.supports_rich_root() != target.supports_rich_root():
3504
raise errors.IncompatibleRepositories(source, target,
3505
"different rich-root support")
3506
if source._serializer != target._serializer:
3507
raise errors.IncompatibleRepositories(source, target,
3508
"different serializers")
1321
# this may look like a no-op: its not. It preserves the ordering
1322
# other_ids had while only returning the members from other_ids
1323
# that we've decided we need.
1324
return [rev_id for rev_id in source_ids if rev_id in result_set]
3511
1327
class InterSameDataRepository(InterRepository):
3512
1328
"""Code for converting between repositories that represent the same data.
3514
1330
Data format and model must match for this to work.
3518
1334
def _get_repo_format_to_test(self):
3519
"""Repository format for testing with.
3521
InterSameData can pull from subtree to subtree and from non-subtree to
3522
non-subtree, so we test this with the richest repository format.
3524
from bzrlib.repofmt import knitrepo
3525
return knitrepo.RepositoryFormatKnit3()
1335
"""Repository format for testing with."""
1336
return RepositoryFormat.get_default_format()
3528
1339
def is_compatible(source, target):
3529
return InterRepository._same_model(source, target)
1340
if not isinstance(source, Repository):
1342
if not isinstance(target, Repository):
1344
if source._format.rich_root_data != target._format.rich_root_data:
1346
if source._serializer != target._serializer:
1352
def copy_content(self, revision_id=None, basis=None):
1353
"""Make a complete copy of the content in self into destination.
1355
This is a destructive operation! Do not use it on existing
1358
:param revision_id: Only copy the content needed to construct
1359
revision_id and its parents.
1360
:param basis: Copy the needed data preferentially from basis.
1363
self.target.set_make_working_trees(self.source.make_working_trees())
1364
except NotImplementedError:
1366
# TODO: jam 20070210 This is fairly internal, so we should probably
1367
# just assert that revision_id is not unicode.
1368
revision_id = osutils.safe_revision_id(revision_id)
1369
# grab the basis available data
1370
if basis is not None:
1371
self.target.fetch(basis, revision_id=revision_id)
1372
# but don't bother fetching if we have the needed data now.
1373
if (revision_id not in (None, _mod_revision.NULL_REVISION) and
1374
self.target.has_revision(revision_id)):
1376
self.target.fetch(self.source, revision_id=revision_id)
1379
def fetch(self, revision_id=None, pb=None):
1380
"""See InterRepository.fetch()."""
1381
from bzrlib.fetch import GenericRepoFetcher
1382
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1383
self.source, self.source._format, self.target,
1384
self.target._format)
1385
# TODO: jam 20070210 This should be an assert, not a translate
1386
revision_id = osutils.safe_revision_id(revision_id)
1387
f = GenericRepoFetcher(to_repository=self.target,
1388
from_repository=self.source,
1389
last_revision=revision_id,
1391
return f.count_copied, f.failed_revisions
3532
1394
class InterWeaveRepo(InterSameDataRepository):
3533
"""Optimised code paths between Weave based repositories.
3535
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3536
implemented lazy inter-object optimisation.
1395
"""Optimised code paths between Weave based repositories."""
3540
1398
def _get_repo_format_to_test(self):
3709
1598
def is_compatible(source, target):
3710
"""Be compatible with Knit2 source and Knit3 target"""
3711
# This is redundant with format.check_conversion_target(), however that
3712
# raises an exception, and we just want to say "False" as in we won't
3713
# support converting between these formats.
3714
if 'IDS_never' in debug.debug_flags:
3716
if source.supports_rich_root() and not target.supports_rich_root():
3718
if (source._format.supports_tree_reference
3719
and not target._format.supports_tree_reference):
3721
if target._fallback_repositories and target._format.supports_chks:
3722
# IDS doesn't know how to copy CHKs for the parent inventories it
3723
# adds to stacked repos.
3725
if 'IDS_always' in debug.debug_flags:
1599
if not isinstance(source, Repository):
1601
if not isinstance(target, Repository):
1603
if not source._format.rich_root_data and target._format.rich_root_data:
3727
# Only use this code path for local source and target. IDS does far
3728
# too much IO (both bandwidth and roundtrips) over a network.
3729
if not source.bzrdir.transport.base.startswith('file:///'):
3731
if not target.bzrdir.transport.base.startswith('file:///'):
3735
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3736
"""Get the best delta and base for this revision.
3738
:return: (basis_id, delta)
3740
possible_trees = [(parent_id, cache[parent_id])
3741
for parent_id in parent_ids
3742
if parent_id in cache]
3743
if len(possible_trees) == 0:
3744
# There either aren't any parents, or the parents aren't in the
3745
# cache, so just use the last converted tree
3746
possible_trees.append((basis_id, cache[basis_id]))
3748
for basis_id, basis_tree in possible_trees:
3749
delta = tree.inventory._make_delta(basis_tree.inventory)
3750
deltas.append((len(delta), basis_id, delta))
3752
return deltas[0][1:]
3754
def _fetch_batch(self, revision_ids, basis_id, cache):
3755
"""Fetch across a few revisions.
3757
:param revision_ids: The revisions to copy
3758
:param basis_id: The revision_id of a tree that must be in cache, used
3759
as a basis for delta when no other base is available
3760
:param cache: A cache of RevisionTrees that we can use.
3761
:return: The revision_id of the last converted tree. The RevisionTree
3762
for it will be in cache
3764
# Walk though all revisions; get inventory deltas, copy referenced
3765
# texts that delta references, insert the delta, revision and
3767
root_keys_to_create = set()
3770
pending_revisions = []
3771
parent_map = self.source.get_parent_map(revision_ids)
3772
for tree in self.source.revision_trees(revision_ids):
3773
current_revision_id = tree.get_revision_id()
3774
parent_ids = parent_map.get(current_revision_id, ())
3775
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3777
if self._converting_to_rich_root:
3778
self._revision_id_to_root_id[current_revision_id] = \
3780
# Find text entries that need to be copied
3781
for old_path, new_path, file_id, entry in delta:
3782
if new_path is not None:
3785
if not self.target.supports_rich_root():
3786
# The target doesn't support rich root, so we don't
3789
if self._converting_to_rich_root:
3790
# This can't be copied normally, we have to insert
3792
root_keys_to_create.add((file_id, entry.revision))
3794
text_keys.add((file_id, entry.revision))
3795
revision = self.source.get_revision(current_revision_id)
3796
pending_deltas.append((basis_id, delta,
3797
current_revision_id, revision.parent_ids))
3798
pending_revisions.append(revision)
3799
cache[current_revision_id] = tree
3800
basis_id = current_revision_id
3802
from_texts = self.source.texts
3803
to_texts = self.target.texts
3804
if root_keys_to_create:
3805
from bzrlib.fetch import _new_root_data_stream
3806
root_stream = _new_root_data_stream(
3807
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3809
to_texts.insert_record_stream(root_stream)
3810
to_texts.insert_record_stream(from_texts.get_record_stream(
3811
text_keys, self.target._format._fetch_order,
3812
not self.target._format._fetch_uses_deltas))
3813
# insert inventory deltas
3814
for delta in pending_deltas:
3815
self.target.add_inventory_by_delta(*delta)
3816
if self.target._fallback_repositories:
3817
# Make sure this stacked repository has all the parent inventories
3818
# for the new revisions that we are about to insert. We do this
3819
# before adding the revisions so that no revision is added until
3820
# all the inventories it may depend on are added.
3821
# Note that this is overzealous, as we may have fetched these in an
3824
revision_ids = set()
3825
for revision in pending_revisions:
3826
revision_ids.add(revision.revision_id)
3827
parent_ids.update(revision.parent_ids)
3828
parent_ids.difference_update(revision_ids)
3829
parent_ids.discard(_mod_revision.NULL_REVISION)
3830
parent_map = self.source.get_parent_map(parent_ids)
3831
# we iterate over parent_map and not parent_ids because we don't
3832
# want to try copying any revision which is a ghost
3833
for parent_tree in self.source.revision_trees(parent_map):
3834
current_revision_id = parent_tree.get_revision_id()
3835
parents_parents = parent_map[current_revision_id]
3836
basis_id, delta = self._get_delta_for_revision(parent_tree,
3837
parents_parents, basis_id, cache)
3838
self.target.add_inventory_by_delta(
3839
basis_id, delta, current_revision_id, parents_parents)
3840
# insert signatures and revisions
3841
for revision in pending_revisions:
3843
signature = self.source.get_signature_text(
3844
revision.revision_id)
3845
self.target.add_signature_text(revision.revision_id,
3847
except errors.NoSuchRevision:
3849
self.target.add_revision(revision.revision_id, revision)
3852
def _fetch_all_revisions(self, revision_ids, pb):
3853
"""Fetch everything for the list of revisions.
3855
:param revision_ids: The list of revisions to fetch. Must be in
3857
:param pb: A ProgressTask
3860
basis_id, basis_tree = self._get_basis(revision_ids[0])
3862
cache = lru_cache.LRUCache(100)
3863
cache[basis_id] = basis_tree
3864
del basis_tree # We don't want to hang on to it here
3866
for offset in range(0, len(revision_ids), batch_size):
3867
self.target.start_write_group()
3869
pb.update('Transferring revisions', offset,
3871
batch = revision_ids[offset:offset+batch_size]
3872
basis_id = self._fetch_batch(batch, basis_id, cache)
3874
self.target.abort_write_group()
3877
hint = self.target.commit_write_group()
3880
if hints and self.target._format.pack_compresses:
3881
self.target.pack(hint=hints)
3882
pb.update('Transferring revisions', len(revision_ids),
3886
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3888
"""See InterRepository.fetch()."""
3889
if fetch_spec is not None:
3890
raise AssertionError("Not implemented yet...")
3891
if (not self.source.supports_rich_root()
3892
and self.target.supports_rich_root()):
3893
self._converting_to_rich_root = True
3894
self._revision_id_to_root_id = {}
3896
self._converting_to_rich_root = False
3897
revision_ids = self.target.search_missing_revision_ids(self.source,
3898
revision_id, find_ghosts=find_ghosts).get_keys()
3899
if not revision_ids:
3901
revision_ids = tsort.topo_sort(
3902
self.source.get_graph().get_parent_map(revision_ids))
3903
if not revision_ids:
3905
# Walk though all revisions; get inventory deltas, copy referenced
3906
# texts that delta references, insert the delta, revision and
3909
my_pb = ui.ui_factory.nested_progress_bar()
3912
symbol_versioning.warn(
3913
symbol_versioning.deprecated_in((1, 14, 0))
3914
% "pb parameter to fetch()")
3917
self._fetch_all_revisions(revision_ids, pb)
3919
if my_pb is not None:
3921
return len(revision_ids), 0
3923
def _get_basis(self, first_revision_id):
3924
"""Get a revision and tree which exists in the target.
3926
This assumes that first_revision_id is selected for transmission
3927
because all other ancestors are already present. If we can't find an
3928
ancestor we fall back to NULL_REVISION since we know that is safe.
3930
:return: (basis_id, basis_tree)
3932
first_rev = self.source.get_revision(first_revision_id)
3934
basis_id = first_rev.parent_ids[0]
3935
# only valid as a basis if the target has it
3936
self.target.get_revision(basis_id)
3937
# Try to get a basis tree - if its a ghost it will hit the
3938
# NoSuchRevision case.
3939
basis_tree = self.source.revision_tree(basis_id)
3940
except (IndexError, errors.NoSuchRevision):
3941
basis_id = _mod_revision.NULL_REVISION
3942
basis_tree = self.source.revision_tree(basis_id)
3943
return basis_id, basis_tree
3946
InterRepository.register_optimiser(InterDifferingSerializer)
1609
def fetch(self, revision_id=None, pb=None):
1610
"""See InterRepository.fetch()."""
1611
from bzrlib.fetch import Model1toKnit2Fetcher
1612
# TODO: jam 20070210 This should be an assert, not a translate
1613
revision_id = osutils.safe_revision_id(revision_id)
1614
f = Model1toKnit2Fetcher(to_repository=self.target,
1615
from_repository=self.source,
1616
last_revision=revision_id,
1618
return f.count_copied, f.failed_revisions
1621
def copy_content(self, revision_id=None, basis=None):
1622
"""Make a complete copy of the content in self into destination.
1624
This is a destructive operation! Do not use it on existing
1627
:param revision_id: Only copy the content needed to construct
1628
revision_id and its parents.
1629
:param basis: Copy the needed data preferentially from basis.
1632
self.target.set_make_working_trees(self.source.make_working_trees())
1633
except NotImplementedError:
1635
# TODO: jam 20070210 Internal, assert, don't translate
1636
revision_id = osutils.safe_revision_id(revision_id)
1637
# grab the basis available data
1638
if basis is not None:
1639
self.target.fetch(basis, revision_id=revision_id)
1640
# but don't bother fetching if we have the needed data now.
1641
if (revision_id not in (None, _mod_revision.NULL_REVISION) and
1642
self.target.has_revision(revision_id)):
1644
self.target.fetch(self.source, revision_id=revision_id)
1647
class InterKnit1and2(InterKnitRepo):
1650
def _get_repo_format_to_test(self):
1654
def is_compatible(source, target):
1655
"""Be compatible with Knit1 source and Knit3 target"""
1656
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit3
1658
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit1, \
1659
RepositoryFormatKnit3
1660
return (isinstance(source._format, (RepositoryFormatKnit1)) and
1661
isinstance(target._format, (RepositoryFormatKnit3)))
1662
except AttributeError:
1666
def fetch(self, revision_id=None, pb=None):
1667
"""See InterRepository.fetch()."""
1668
from bzrlib.fetch import Knit1to2Fetcher
1669
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1670
self.source, self.source._format, self.target,
1671
self.target._format)
1672
# TODO: jam 20070210 This should be an assert, not a translate
1673
revision_id = osutils.safe_revision_id(revision_id)
1674
f = Knit1to2Fetcher(to_repository=self.target,
1675
from_repository=self.source,
1676
last_revision=revision_id,
1678
return f.count_copied, f.failed_revisions
3947
1681
InterRepository.register_optimiser(InterSameDataRepository)
3948
1682
InterRepository.register_optimiser(InterWeaveRepo)
3949
1683
InterRepository.register_optimiser(InterKnitRepo)
1684
InterRepository.register_optimiser(InterModel1and2)
1685
InterRepository.register_optimiser(InterKnit1and2)
1688
class RepositoryTestProviderAdapter(object):
1689
"""A tool to generate a suite testing multiple repository formats at once.
1691
This is done by copying the test once for each transport and injecting
1692
the transport_server, transport_readonly_server, and bzrdir_format and
1693
repository_format classes into each copy. Each copy is also given a new id()
1694
to make it easy to identify.
1697
def __init__(self, transport_server, transport_readonly_server, formats):
1698
self._transport_server = transport_server
1699
self._transport_readonly_server = transport_readonly_server
1700
self._formats = formats
1702
def adapt(self, test):
1703
result = unittest.TestSuite()
1704
for repository_format, bzrdir_format in self._formats:
1705
from copy import deepcopy
1706
new_test = deepcopy(test)
1707
new_test.transport_server = self._transport_server
1708
new_test.transport_readonly_server = self._transport_readonly_server
1709
new_test.bzrdir_format = bzrdir_format
1710
new_test.repository_format = repository_format
1711
def make_new_test_id():
1712
new_id = "%s(%s)" % (new_test.id(), repository_format.__class__.__name__)
1713
return lambda: new_id
1714
new_test.id = make_new_test_id()
1715
result.addTest(new_test)
1719
class InterRepositoryTestProviderAdapter(object):
1720
"""A tool to generate a suite testing multiple inter repository formats.
1722
This is done by copying the test once for each interrepo provider and injecting
1723
the transport_server, transport_readonly_server, repository_format and
1724
repository_to_format classes into each copy.
1725
Each copy is also given a new id() to make it easy to identify.
1728
def __init__(self, transport_server, transport_readonly_server, formats):
1729
self._transport_server = transport_server
1730
self._transport_readonly_server = transport_readonly_server
1731
self._formats = formats
1733
def adapt(self, test):
1734
result = unittest.TestSuite()
1735
for interrepo_class, repository_format, repository_format_to in self._formats:
1736
from copy import deepcopy
1737
new_test = deepcopy(test)
1738
new_test.transport_server = self._transport_server
1739
new_test.transport_readonly_server = self._transport_readonly_server
1740
new_test.interrepo_class = interrepo_class
1741
new_test.repository_format = repository_format
1742
new_test.repository_format_to = repository_format_to
1743
def make_new_test_id():
1744
new_id = "%s(%s)" % (new_test.id(), interrepo_class.__name__)
1745
return lambda: new_id
1746
new_test.id = make_new_test_id()
1747
result.addTest(new_test)
1751
def default_test_list():
1752
"""Generate the default list of interrepo permutations to test."""
1753
from bzrlib.repofmt import knitrepo, weaverepo
1755
# test the default InterRepository between format 6 and the current
1757
# XXX: robertc 20060220 reinstate this when there are two supported
1758
# formats which do not have an optimal code path between them.
1759
#result.append((InterRepository,
1760
# RepositoryFormat6(),
1761
# RepositoryFormatKnit1()))
1762
for optimiser_class in InterRepository._optimisers:
1763
format_to_test = optimiser_class._get_repo_format_to_test()
1764
if format_to_test is not None:
1765
result.append((optimiser_class,
1766
format_to_test, format_to_test))
1767
# if there are specific combinations we want to use, we can add them
1769
result.append((InterModel1and2,
1770
weaverepo.RepositoryFormat5(),
1771
knitrepo.RepositoryFormatKnit3()))
1772
result.append((InterKnit1and2,
1773
knitrepo.RepositoryFormatKnit1(),
1774
knitrepo.RepositoryFormatKnit3()))
3952
1778
class CopyConverter(object):
3953
1779
"""A repository conversion tool which just performs a copy of the content.
3955
1781
This is slow but quite reliable.
4001
1827
self.pb.update(message, self.count, self.total)
1830
class CommitBuilder(object):
1831
"""Provides an interface to build up a commit.
1833
This allows describing a tree to be committed without needing to
1834
know the internals of the format of the repository.
1837
record_root_entry = False
1838
def __init__(self, repository, parents, config, timestamp=None,
1839
timezone=None, committer=None, revprops=None,
1841
"""Initiate a CommitBuilder.
1843
:param repository: Repository to commit to.
1844
:param parents: Revision ids of the parents of the new revision.
1845
:param config: Configuration to use.
1846
:param timestamp: Optional timestamp recorded for commit.
1847
:param timezone: Optional timezone for timestamp.
1848
:param committer: Optional committer to set for commit.
1849
:param revprops: Optional dictionary of revision properties.
1850
:param revision_id: Optional revision id.
1852
self._config = config
1854
if committer is None:
1855
self._committer = self._config.username()
1857
assert isinstance(committer, basestring), type(committer)
1858
self._committer = committer
1860
self.new_inventory = Inventory(None)
1861
self._new_revision_id = osutils.safe_revision_id(revision_id)
1862
self.parents = parents
1863
self.repository = repository
1866
if revprops is not None:
1867
self._revprops.update(revprops)
1869
if timestamp is None:
1870
timestamp = time.time()
1871
# Restrict resolution to 1ms
1872
self._timestamp = round(timestamp, 3)
1874
if timezone is None:
1875
self._timezone = osutils.local_time_offset()
1877
self._timezone = int(timezone)
1879
self._generate_revision_if_needed()
1881
def commit(self, message):
1882
"""Make the actual commit.
1884
:return: The revision id of the recorded revision.
1886
rev = _mod_revision.Revision(
1887
timestamp=self._timestamp,
1888
timezone=self._timezone,
1889
committer=self._committer,
1891
inventory_sha1=self.inv_sha1,
1892
revision_id=self._new_revision_id,
1893
properties=self._revprops)
1894
rev.parent_ids = self.parents
1895
self.repository.add_revision(self._new_revision_id, rev,
1896
self.new_inventory, self._config)
1897
return self._new_revision_id
1899
def revision_tree(self):
1900
"""Return the tree that was just committed.
1902
After calling commit() this can be called to get a RevisionTree
1903
representing the newly committed tree. This is preferred to
1904
calling Repository.revision_tree() because that may require
1905
deserializing the inventory, while we already have a copy in
1908
return RevisionTree(self.repository, self.new_inventory,
1909
self._new_revision_id)
1911
def finish_inventory(self):
1912
"""Tell the builder that the inventory is finished."""
1913
if self.new_inventory.root is None:
1914
symbol_versioning.warn('Root entry should be supplied to'
1915
' record_entry_contents, as of bzr 0.10.',
1916
DeprecationWarning, stacklevel=2)
1917
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
1918
self.new_inventory.revision_id = self._new_revision_id
1919
self.inv_sha1 = self.repository.add_inventory(
1920
self._new_revision_id,
1925
def _gen_revision_id(self):
1926
"""Return new revision-id."""
1927
return generate_ids.gen_revision_id(self._config.username(),
1930
def _generate_revision_if_needed(self):
1931
"""Create a revision id if None was supplied.
1933
If the repository can not support user-specified revision ids
1934
they should override this function and raise CannotSetRevisionId
1935
if _new_revision_id is not None.
1937
:raises: CannotSetRevisionId
1939
if self._new_revision_id is None:
1940
self._new_revision_id = self._gen_revision_id()
1942
def record_entry_contents(self, ie, parent_invs, path, tree):
1943
"""Record the content of ie from tree into the commit if needed.
1945
Side effect: sets ie.revision when unchanged
1947
:param ie: An inventory entry present in the commit.
1948
:param parent_invs: The inventories of the parent revisions of the
1950
:param path: The path the entry is at in the tree.
1951
:param tree: The tree which contains this entry and should be used to
1954
if self.new_inventory.root is None and ie.parent_id is not None:
1955
symbol_versioning.warn('Root entry should be supplied to'
1956
' record_entry_contents, as of bzr 0.10.',
1957
DeprecationWarning, stacklevel=2)
1958
self.record_entry_contents(tree.inventory.root.copy(), parent_invs,
1960
self.new_inventory.add(ie)
1962
# ie.revision is always None if the InventoryEntry is considered
1963
# for committing. ie.snapshot will record the correct revision
1964
# which may be the sole parent if it is untouched.
1965
if ie.revision is not None:
1968
# In this revision format, root entries have no knit or weave
1969
if ie is self.new_inventory.root:
1970
# When serializing out to disk and back in
1971
# root.revision is always _new_revision_id
1972
ie.revision = self._new_revision_id
1974
previous_entries = ie.find_previous_heads(
1976
self.repository.weave_store,
1977
self.repository.get_transaction())
1978
# we are creating a new revision for ie in the history store
1980
ie.snapshot(self._new_revision_id, path, previous_entries, tree, self)
1982
def modified_directory(self, file_id, file_parents):
1983
"""Record the presence of a symbolic link.
1985
:param file_id: The file_id of the link to record.
1986
:param file_parents: The per-file parent revision ids.
1988
self._add_text_to_weave(file_id, [], file_parents.keys())
1990
def modified_reference(self, file_id, file_parents):
1991
"""Record the modification of a reference.
1993
:param file_id: The file_id of the link to record.
1994
:param file_parents: The per-file parent revision ids.
1996
self._add_text_to_weave(file_id, [], file_parents.keys())
1998
def modified_file_text(self, file_id, file_parents,
1999
get_content_byte_lines, text_sha1=None,
2001
"""Record the text of file file_id
2003
:param file_id: The file_id of the file to record the text of.
2004
:param file_parents: The per-file parent revision ids.
2005
:param get_content_byte_lines: A callable which will return the byte
2007
:param text_sha1: Optional SHA1 of the file contents.
2008
:param text_size: Optional size of the file contents.
2010
# mutter('storing text of file {%s} in revision {%s} into %r',
2011
# file_id, self._new_revision_id, self.repository.weave_store)
2012
# special case to avoid diffing on renames or
2014
if (len(file_parents) == 1
2015
and text_sha1 == file_parents.values()[0].text_sha1
2016
and text_size == file_parents.values()[0].text_size):
2017
previous_ie = file_parents.values()[0]
2018
versionedfile = self.repository.weave_store.get_weave(file_id,
2019
self.repository.get_transaction())
2020
versionedfile.clone_text(self._new_revision_id,
2021
previous_ie.revision, file_parents.keys())
2022
return text_sha1, text_size
2024
new_lines = get_content_byte_lines()
2025
# TODO: Rather than invoking sha_strings here, _add_text_to_weave
2026
# should return the SHA1 and size
2027
self._add_text_to_weave(file_id, new_lines, file_parents.keys())
2028
return osutils.sha_strings(new_lines), \
2029
sum(map(len, new_lines))
2031
def modified_link(self, file_id, file_parents, link_target):
2032
"""Record the presence of a symbolic link.
2034
:param file_id: The file_id of the link to record.
2035
:param file_parents: The per-file parent revision ids.
2036
:param link_target: Target location of this link.
2038
self._add_text_to_weave(file_id, [], file_parents.keys())
2040
def _add_text_to_weave(self, file_id, new_lines, parents):
2041
versionedfile = self.repository.weave_store.get_weave_or_empty(
2042
file_id, self.repository.get_transaction())
2043
versionedfile.add_lines(self._new_revision_id, parents, new_lines)
2044
versionedfile.clear_cache()
2047
class _CommitBuilder(CommitBuilder):
2048
"""Temporary class so old CommitBuilders are detected properly
2050
Note: CommitBuilder works whether or not root entry is recorded.
2053
record_root_entry = True
2056
class RootCommitBuilder(CommitBuilder):
2057
"""This commitbuilder actually records the root id"""
2059
record_root_entry = True
2061
def record_entry_contents(self, ie, parent_invs, path, tree):
2062
"""Record the content of ie from tree into the commit if needed.
2064
Side effect: sets ie.revision when unchanged
2066
:param ie: An inventory entry present in the commit.
2067
:param parent_invs: The inventories of the parent revisions of the
2069
:param path: The path the entry is at in the tree.
2070
:param tree: The tree which contains this entry and should be used to
2073
assert self.new_inventory.root is not None or ie.parent_id is None
2074
self.new_inventory.add(ie)
2076
# ie.revision is always None if the InventoryEntry is considered
2077
# for committing. ie.snapshot will record the correct revision
2078
# which may be the sole parent if it is untouched.
2079
if ie.revision is not None:
2082
previous_entries = ie.find_previous_heads(
2084
self.repository.weave_store,
2085
self.repository.get_transaction())
2086
# we are creating a new revision for ie in the history store
2088
ie.snapshot(self._new_revision_id, path, previous_entries, tree, self)
4004
2091
_unescape_map = {
4029
2116
if _unescape_re is None:
4030
2117
_unescape_re = re.compile('\&([^;]*);')
4031
2118
return _unescape_re.sub(_unescaper, data)
4034
class _VersionedFileChecker(object):
4036
def __init__(self, repository, text_key_references=None, ancestors=None):
4037
self.repository = repository
4038
self.text_index = self.repository._generate_text_key_index(
4039
text_key_references=text_key_references, ancestors=ancestors)
4041
def calculate_file_version_parents(self, text_key):
4042
"""Calculate the correct parents for a file version according to
4045
parent_keys = self.text_index[text_key]
4046
if parent_keys == [_mod_revision.NULL_REVISION]:
4048
return tuple(parent_keys)
4050
def check_file_version_parents(self, texts, progress_bar=None):
4051
"""Check the parents stored in a versioned file are correct.
4053
It also detects file versions that are not referenced by their
4054
corresponding revision's inventory.
4056
:returns: A tuple of (wrong_parents, dangling_file_versions).
4057
wrong_parents is a dict mapping {revision_id: (stored_parents,
4058
correct_parents)} for each revision_id where the stored parents
4059
are not correct. dangling_file_versions is a set of (file_id,
4060
revision_id) tuples for versions that are present in this versioned
4061
file, but not used by the corresponding inventory.
4063
local_progress = None
4064
if progress_bar is None:
4065
local_progress = ui.ui_factory.nested_progress_bar()
4066
progress_bar = local_progress
4068
return self._check_file_version_parents(texts, progress_bar)
4071
local_progress.finished()
4073
def _check_file_version_parents(self, texts, progress_bar):
4074
"""See check_file_version_parents."""
4076
self.file_ids = set([file_id for file_id, _ in
4077
self.text_index.iterkeys()])
4078
# text keys is now grouped by file_id
4079
n_versions = len(self.text_index)
4080
progress_bar.update('loading text store', 0, n_versions)
4081
parent_map = self.repository.texts.get_parent_map(self.text_index)
4082
# On unlistable transports this could well be empty/error...
4083
text_keys = self.repository.texts.keys()
4084
unused_keys = frozenset(text_keys) - set(self.text_index)
4085
for num, key in enumerate(self.text_index.iterkeys()):
4086
progress_bar.update('checking text graph', num, n_versions)
4087
correct_parents = self.calculate_file_version_parents(key)
4089
knit_parents = parent_map[key]
4090
except errors.RevisionNotPresent:
4093
if correct_parents != knit_parents:
4094
wrong_parents[key] = (knit_parents, correct_parents)
4095
return wrong_parents, unused_keys
4098
def _old_get_graph(repository, revision_id):
4099
"""DO NOT USE. That is all. I'm serious."""
4100
graph = repository.get_graph()
4101
revision_graph = dict(((key, value) for key, value in
4102
graph.iter_ancestry([revision_id]) if value is not None))
4103
return _strip_NULL_ghosts(revision_graph)
4106
def _strip_NULL_ghosts(revision_graph):
4107
"""Also don't use this. more compatibility code for unmigrated clients."""
4108
# Filter ghosts, and null:
4109
if _mod_revision.NULL_REVISION in revision_graph:
4110
del revision_graph[_mod_revision.NULL_REVISION]
4111
for key, parents in revision_graph.items():
4112
revision_graph[key] = tuple(parent for parent in parents if parent
4114
return revision_graph
4117
class StreamSink(object):
4118
"""An object that can insert a stream into a repository.
4120
This interface handles the complexity of reserialising inventories and
4121
revisions from different formats, and allows unidirectional insertion into
4122
stacked repositories without looking for the missing basis parents
4126
def __init__(self, target_repo):
4127
self.target_repo = target_repo
4129
def insert_stream(self, stream, src_format, resume_tokens):
4130
"""Insert a stream's content into the target repository.
4132
:param src_format: a bzr repository format.
4134
:return: a list of resume tokens and an iterable of keys additional
4135
items required before the insertion can be completed.
4137
self.target_repo.lock_write()
4140
self.target_repo.resume_write_group(resume_tokens)
4143
self.target_repo.start_write_group()
4146
# locked_insert_stream performs a commit|suspend.
4147
return self._locked_insert_stream(stream, src_format, is_resume)
4149
self.target_repo.abort_write_group(suppress_errors=True)
4152
self.target_repo.unlock()
4154
def _locked_insert_stream(self, stream, src_format, is_resume):
4155
to_serializer = self.target_repo._format._serializer
4156
src_serializer = src_format._serializer
4158
if to_serializer == src_serializer:
4159
# If serializers match and the target is a pack repository, set the
4160
# write cache size on the new pack. This avoids poor performance
4161
# on transports where append is unbuffered (such as
4162
# RemoteTransport). This is safe to do because nothing should read
4163
# back from the target repository while a stream with matching
4164
# serialization is being inserted.
4165
# The exception is that a delta record from the source that should
4166
# be a fulltext may need to be expanded by the target (see
4167
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4168
# explicitly flush any buffered writes first in that rare case.
4170
new_pack = self.target_repo._pack_collection._new_pack
4171
except AttributeError:
4172
# Not a pack repository
4175
new_pack.set_write_cache_size(1024*1024)
4176
for substream_type, substream in stream:
4177
if 'stream' in debug.debug_flags:
4178
mutter('inserting substream: %s', substream_type)
4179
if substream_type == 'texts':
4180
self.target_repo.texts.insert_record_stream(substream)
4181
elif substream_type == 'inventories':
4182
if src_serializer == to_serializer:
4183
self.target_repo.inventories.insert_record_stream(
4186
self._extract_and_insert_inventories(
4187
substream, src_serializer)
4188
elif substream_type == 'inventory-deltas':
4189
self._extract_and_insert_inventory_deltas(
4190
substream, src_serializer)
4191
elif substream_type == 'chk_bytes':
4192
# XXX: This doesn't support conversions, as it assumes the
4193
# conversion was done in the fetch code.
4194
self.target_repo.chk_bytes.insert_record_stream(substream)
4195
elif substream_type == 'revisions':
4196
# This may fallback to extract-and-insert more often than
4197
# required if the serializers are different only in terms of
4199
if src_serializer == to_serializer:
4200
self.target_repo.revisions.insert_record_stream(
4203
self._extract_and_insert_revisions(substream,
4205
elif substream_type == 'signatures':
4206
self.target_repo.signatures.insert_record_stream(substream)
4208
raise AssertionError('kaboom! %s' % (substream_type,))
4209
# Done inserting data, and the missing_keys calculations will try to
4210
# read back from the inserted data, so flush the writes to the new pack
4211
# (if this is pack format).
4212
if new_pack is not None:
4213
new_pack._write_data('', flush=True)
4214
# Find all the new revisions (including ones from resume_tokens)
4215
missing_keys = self.target_repo.get_missing_parent_inventories(
4216
check_for_missing_texts=is_resume)
4218
for prefix, versioned_file in (
4219
('texts', self.target_repo.texts),
4220
('inventories', self.target_repo.inventories),
4221
('revisions', self.target_repo.revisions),
4222
('signatures', self.target_repo.signatures),
4223
('chk_bytes', self.target_repo.chk_bytes),
4225
if versioned_file is None:
4227
missing_keys.update((prefix,) + key for key in
4228
versioned_file.get_missing_compression_parent_keys())
4229
except NotImplementedError:
4230
# cannot even attempt suspending, and missing would have failed
4231
# during stream insertion.
4232
missing_keys = set()
4235
# suspend the write group and tell the caller what we is
4236
# missing. We know we can suspend or else we would not have
4237
# entered this code path. (All repositories that can handle
4238
# missing keys can handle suspending a write group).
4239
write_group_tokens = self.target_repo.suspend_write_group()
4240
return write_group_tokens, missing_keys
4241
hint = self.target_repo.commit_write_group()
4242
if (to_serializer != src_serializer and
4243
self.target_repo._format.pack_compresses):
4244
self.target_repo.pack(hint=hint)
4247
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4248
target_rich_root = self.target_repo._format.rich_root_data
4249
target_tree_refs = self.target_repo._format.supports_tree_reference
4250
for record in substream:
4251
# Insert the delta directly
4252
inventory_delta_bytes = record.get_bytes_as('fulltext')
4253
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4255
parse_result = deserialiser.parse_text_bytes(
4256
inventory_delta_bytes)
4257
except inventory_delta.IncompatibleInventoryDelta, err:
4258
trace.mutter("Incompatible delta: %s", err.msg)
4259
raise errors.IncompatibleRevision(self.target_repo._format)
4260
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4261
revision_id = new_id
4262
parents = [key[0] for key in record.parents]
4263
self.target_repo.add_inventory_by_delta(
4264
basis_id, inv_delta, revision_id, parents)
4266
def _extract_and_insert_inventories(self, substream, serializer,
4268
"""Generate a new inventory versionedfile in target, converting data.
4270
The inventory is retrieved from the source, (deserializing it), and
4271
stored in the target (reserializing it in a different format).
4273
target_rich_root = self.target_repo._format.rich_root_data
4274
target_tree_refs = self.target_repo._format.supports_tree_reference
4275
for record in substream:
4276
# It's not a delta, so it must be a fulltext in the source
4277
# serializer's format.
4278
bytes = record.get_bytes_as('fulltext')
4279
revision_id = record.key[0]
4280
inv = serializer.read_inventory_from_string(bytes, revision_id)
4281
parents = [key[0] for key in record.parents]
4282
self.target_repo.add_inventory(revision_id, inv, parents)
4283
# No need to keep holding this full inv in memory when the rest of
4284
# the substream is likely to be all deltas.
4287
def _extract_and_insert_revisions(self, substream, serializer):
4288
for record in substream:
4289
bytes = record.get_bytes_as('fulltext')
4290
revision_id = record.key[0]
4291
rev = serializer.read_revision_from_string(bytes)
4292
if rev.revision_id != revision_id:
4293
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4294
self.target_repo.add_revision(revision_id, rev)
4297
if self.target_repo._format._fetch_reconcile:
4298
self.target_repo.reconcile()
4301
class StreamSource(object):
4302
"""A source of a stream for fetching between repositories."""
4304
def __init__(self, from_repository, to_format):
4305
"""Create a StreamSource streaming from from_repository."""
4306
self.from_repository = from_repository
4307
self.to_format = to_format
4309
def delta_on_metadata(self):
4310
"""Return True if delta's are permitted on metadata streams.
4312
That is on revisions and signatures.
4314
src_serializer = self.from_repository._format._serializer
4315
target_serializer = self.to_format._serializer
4316
return (self.to_format._fetch_uses_deltas and
4317
src_serializer == target_serializer)
4319
def _fetch_revision_texts(self, revs):
4320
# fetch signatures first and then the revision texts
4321
# may need to be a InterRevisionStore call here.
4322
from_sf = self.from_repository.signatures
4323
# A missing signature is just skipped.
4324
keys = [(rev_id,) for rev_id in revs]
4325
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4327
self.to_format._fetch_order,
4328
not self.to_format._fetch_uses_deltas))
4329
# If a revision has a delta, this is actually expanded inside the
4330
# insert_record_stream code now, which is an alternate fix for
4332
from_rf = self.from_repository.revisions
4333
revisions = from_rf.get_record_stream(
4335
self.to_format._fetch_order,
4336
not self.delta_on_metadata())
4337
return [('signatures', signatures), ('revisions', revisions)]
4339
def _generate_root_texts(self, revs):
4340
"""This will be called by get_stream between fetching weave texts and
4341
fetching the inventory weave.
4343
if self._rich_root_upgrade():
4345
return bzrlib.fetch.Inter1and2Helper(
4346
self.from_repository).generate_root_texts(revs)
4350
def get_stream(self, search):
4352
revs = search.get_keys()
4353
graph = self.from_repository.get_graph()
4354
revs = tsort.topo_sort(graph.get_parent_map(revs))
4355
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4357
for knit_kind, file_id, revisions in data_to_fetch:
4358
if knit_kind != phase:
4360
# Make a new progress bar for this phase
4361
if knit_kind == "file":
4362
# Accumulate file texts
4363
text_keys.extend([(file_id, revision) for revision in
4365
elif knit_kind == "inventory":
4366
# Now copy the file texts.
4367
from_texts = self.from_repository.texts
4368
yield ('texts', from_texts.get_record_stream(
4369
text_keys, self.to_format._fetch_order,
4370
not self.to_format._fetch_uses_deltas))
4371
# Cause an error if a text occurs after we have done the
4374
# Before we process the inventory we generate the root
4375
# texts (if necessary) so that the inventories references
4377
for _ in self._generate_root_texts(revs):
4379
# we fetch only the referenced inventories because we do not
4380
# know for unselected inventories whether all their required
4381
# texts are present in the other repository - it could be
4383
for info in self._get_inventory_stream(revs):
4385
elif knit_kind == "signatures":
4386
# Nothing to do here; this will be taken care of when
4387
# _fetch_revision_texts happens.
4389
elif knit_kind == "revisions":
4390
for record in self._fetch_revision_texts(revs):
4393
raise AssertionError("Unknown knit kind %r" % knit_kind)
4395
def get_stream_for_missing_keys(self, missing_keys):
4396
# missing keys can only occur when we are byte copying and not
4397
# translating (because translation means we don't send
4398
# unreconstructable deltas ever).
4400
keys['texts'] = set()
4401
keys['revisions'] = set()
4402
keys['inventories'] = set()
4403
keys['chk_bytes'] = set()
4404
keys['signatures'] = set()
4405
for key in missing_keys:
4406
keys[key[0]].add(key[1:])
4407
if len(keys['revisions']):
4408
# If we allowed copying revisions at this point, we could end up
4409
# copying a revision without copying its required texts: a
4410
# violation of the requirements for repository integrity.
4411
raise AssertionError(
4412
'cannot copy revisions to fill in missing deltas %s' % (
4413
keys['revisions'],))
4414
for substream_kind, keys in keys.iteritems():
4415
vf = getattr(self.from_repository, substream_kind)
4416
if vf is None and keys:
4417
raise AssertionError(
4418
"cannot fill in keys for a versioned file we don't"
4419
" have: %s needs %s" % (substream_kind, keys))
4421
# No need to stream something we don't have
4423
if substream_kind == 'inventories':
4424
# Some missing keys are genuinely ghosts, filter those out.
4425
present = self.from_repository.inventories.get_parent_map(keys)
4426
revs = [key[0] for key in present]
4427
# Get the inventory stream more-or-less as we do for the
4428
# original stream; there's no reason to assume that records
4429
# direct from the source will be suitable for the sink. (Think
4430
# e.g. 2a -> 1.9-rich-root).
4431
for info in self._get_inventory_stream(revs, missing=True):
4435
# Ask for full texts always so that we don't need more round trips
4436
# after this stream.
4437
# Some of the missing keys are genuinely ghosts, so filter absent
4438
# records. The Sink is responsible for doing another check to
4439
# ensure that ghosts don't introduce missing data for future
4441
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4442
self.to_format._fetch_order, True))
4443
yield substream_kind, stream
4445
def inventory_fetch_order(self):
4446
if self._rich_root_upgrade():
4447
return 'topological'
4449
return self.to_format._fetch_order
4451
def _rich_root_upgrade(self):
4452
return (not self.from_repository._format.rich_root_data and
4453
self.to_format.rich_root_data)
4455
def _get_inventory_stream(self, revision_ids, missing=False):
4456
from_format = self.from_repository._format
4457
if (from_format.supports_chks and self.to_format.supports_chks and
4458
from_format.network_name() == self.to_format.network_name()):
4459
raise AssertionError(
4460
"this case should be handled by GroupCHKStreamSource")
4461
elif 'forceinvdeltas' in debug.debug_flags:
4462
return self._get_convertable_inventory_stream(revision_ids,
4463
delta_versus_null=missing)
4464
elif from_format.network_name() == self.to_format.network_name():
4466
return self._get_simple_inventory_stream(revision_ids,
4468
elif (not from_format.supports_chks and not self.to_format.supports_chks
4469
and from_format._serializer == self.to_format._serializer):
4470
# Essentially the same format.
4471
return self._get_simple_inventory_stream(revision_ids,
4474
# Any time we switch serializations, we want to use an
4475
# inventory-delta based approach.
4476
return self._get_convertable_inventory_stream(revision_ids,
4477
delta_versus_null=missing)
4479
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4480
# NB: This currently reopens the inventory weave in source;
4481
# using a single stream interface instead would avoid this.
4482
from_weave = self.from_repository.inventories
4484
delta_closure = True
4486
delta_closure = not self.delta_on_metadata()
4487
yield ('inventories', from_weave.get_record_stream(
4488
[(rev_id,) for rev_id in revision_ids],
4489
self.inventory_fetch_order(), delta_closure))
4491
def _get_convertable_inventory_stream(self, revision_ids,
4492
delta_versus_null=False):
4493
# The source is using CHKs, but the target either doesn't or it has a
4494
# different serializer. The StreamSink code expects to be able to
4495
# convert on the target, so we need to put bytes-on-the-wire that can
4496
# be converted. That means inventory deltas (if the remote is <1.19,
4497
# RemoteStreamSink will fallback to VFS to insert the deltas).
4498
yield ('inventory-deltas',
4499
self._stream_invs_as_deltas(revision_ids,
4500
delta_versus_null=delta_versus_null))
4502
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4503
"""Return a stream of inventory-deltas for the given rev ids.
4505
:param revision_ids: The list of inventories to transmit
4506
:param delta_versus_null: Don't try to find a minimal delta for this
4507
entry, instead compute the delta versus the NULL_REVISION. This
4508
effectively streams a complete inventory. Used for stuff like
4509
filling in missing parents, etc.
4511
from_repo = self.from_repository
4512
revision_keys = [(rev_id,) for rev_id in revision_ids]
4513
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4514
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4516
inventories = self.from_repository.iter_inventories(
4517
revision_ids, 'topological')
4518
format = from_repo._format
4519
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4520
inventory_cache = lru_cache.LRUCache(50)
4521
null_inventory = from_repo.revision_tree(
4522
_mod_revision.NULL_REVISION).inventory
4523
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4524
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4525
# repo back into a non-rich-root repo ought to be allowed)
4526
serializer = inventory_delta.InventoryDeltaSerializer(
4527
versioned_root=format.rich_root_data,
4528
tree_references=format.supports_tree_reference)
4529
for inv in inventories:
4530
key = (inv.revision_id,)
4531
parent_keys = parent_map.get(key, ())
4533
if not delta_versus_null and parent_keys:
4534
# The caller did not ask for complete inventories and we have
4535
# some parents that we can delta against. Make a delta against
4536
# each parent so that we can find the smallest.
4537
parent_ids = [parent_key[0] for parent_key in parent_keys]
4538
for parent_id in parent_ids:
4539
if parent_id not in invs_sent_so_far:
4540
# We don't know that the remote side has this basis, so
4543
if parent_id == _mod_revision.NULL_REVISION:
4544
parent_inv = null_inventory
4546
parent_inv = inventory_cache.get(parent_id, None)
4547
if parent_inv is None:
4548
parent_inv = from_repo.get_inventory(parent_id)
4549
candidate_delta = inv._make_delta(parent_inv)
4550
if (delta is None or
4551
len(delta) > len(candidate_delta)):
4552
delta = candidate_delta
4553
basis_id = parent_id
4555
# Either none of the parents ended up being suitable, or we
4556
# were asked to delta against NULL
4557
basis_id = _mod_revision.NULL_REVISION
4558
delta = inv._make_delta(null_inventory)
4559
invs_sent_so_far.add(inv.revision_id)
4560
inventory_cache[inv.revision_id] = inv
4561
delta_serialized = ''.join(
4562
serializer.delta_to_lines(basis_id, key[-1], delta))
4563
yield versionedfile.FulltextContentFactory(
4564
key, parent_keys, None, delta_serialized)
4567
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4568
stop_revision=None):
4569
"""Extend the partial history to include a given index
4571
If a stop_index is supplied, stop when that index has been reached.
4572
If a stop_revision is supplied, stop when that revision is
4573
encountered. Otherwise, stop when the beginning of history is
4576
:param stop_index: The index which should be present. When it is
4577
present, history extension will stop.
4578
:param stop_revision: The revision id which should be present. When
4579
it is encountered, history extension will stop.
4581
start_revision = partial_history_cache[-1]
4582
iterator = repo.iter_reverse_revision_history(start_revision)
4584
#skip the last revision in the list
4587
if (stop_index is not None and
4588
len(partial_history_cache) > stop_index):
4590
if partial_history_cache[-1] == stop_revision:
4592
revision_id = iterator.next()
4593
partial_history_cache.append(revision_id)
4594
except StopIteration: