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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
from cStringIO import StringIO
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19
17
from bzrlib.lazy_import import lazy_import
20
18
lazy_import(globals(), """
25
23
from bzrlib import (
37
40
revision as _mod_revision,
46
from bzrlib.bundle import serializer
42
47
from bzrlib.revisiontree import RevisionTree
43
48
from bzrlib.store.versioned import VersionedFileStore
44
from bzrlib.store.text import TextStore
45
49
from bzrlib.testament import Testament
49
52
from bzrlib.decorators import needs_read_lock, needs_write_lock
50
53
from bzrlib.inter import InterObject
51
from bzrlib.inventory import Inventory, InventoryDirectory, ROOT_ID
52
from bzrlib.symbol_versioning import (
56
from bzrlib.trace import mutter, note, warning
54
from bzrlib.inventory import (
60
from bzrlib import registry
61
from bzrlib.trace import (
62
log_exception_quietly, note, mutter, mutter_callsite, warning)
59
65
# Old formats display a warning, but only once
60
66
_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
# We can't trust a check of the file length because of content
470
if (# 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
if ie.reference_revision is None:
544
raise AssertionError("invalid content_summary for nested tree: %r"
545
% (content_summary,))
546
self._add_text_to_weave(ie.file_id, '', heads, None)
548
raise NotImplementedError('unknown kind')
549
ie.revision = self._new_revision_id
550
self._any_changes = True
551
return self._get_delta(ie, basis_inv, path), True, fingerprint
553
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
554
_entry_factory=entry_factory):
555
"""Record a new tree via iter_changes.
557
:param tree: The tree to obtain text contents from for changed objects.
558
:param basis_revision_id: The revision id of the tree the iter_changes
559
has been generated against. Currently assumed to be the same
560
as self.parents[0] - if it is not, errors may occur.
561
:param iter_changes: An iter_changes iterator with the changes to apply
562
to basis_revision_id. The iterator must not include any items with
563
a current kind of None - missing items must be either filtered out
564
or errored-on beefore record_iter_changes sees the item.
565
:param _entry_factory: Private method to bind entry_factory locally for
567
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
570
# Create an inventory delta based on deltas between all the parents and
571
# deltas between all the parent inventories. We use inventory delta's
572
# between the inventory objects because iter_changes masks
573
# last-changed-field only changes.
575
# file_id -> change map, change is fileid, paths, changed, versioneds,
576
# parents, names, kinds, executables
578
# {file_id -> revision_id -> inventory entry, for entries in parent
579
# trees that are not parents[0]
583
revtrees = list(self.repository.revision_trees(self.parents))
584
except errors.NoSuchRevision:
585
# one or more ghosts, slow path.
587
for revision_id in self.parents:
589
revtrees.append(self.repository.revision_tree(revision_id))
590
except errors.NoSuchRevision:
592
basis_revision_id = _mod_revision.NULL_REVISION
594
revtrees.append(self.repository.revision_tree(
595
_mod_revision.NULL_REVISION))
596
# The basis inventory from a repository
598
basis_inv = revtrees[0].inventory
600
basis_inv = self.repository.revision_tree(
601
_mod_revision.NULL_REVISION).inventory
602
if len(self.parents) > 0:
603
if basis_revision_id != self.parents[0] and not ghost_basis:
605
"arbitrary basis parents not yet supported with merges")
606
for revtree in revtrees[1:]:
607
for change in revtree.inventory._make_delta(basis_inv):
608
if change[1] is None:
609
# Not present in this parent.
611
if change[2] not in merged_ids:
612
if change[0] is not None:
613
basis_entry = basis_inv[change[2]]
614
merged_ids[change[2]] = [
616
basis_entry.revision,
619
parent_entries[change[2]] = {
621
basis_entry.revision:basis_entry,
623
change[3].revision:change[3],
626
merged_ids[change[2]] = [change[3].revision]
627
parent_entries[change[2]] = {change[3].revision:change[3]}
629
merged_ids[change[2]].append(change[3].revision)
630
parent_entries[change[2]][change[3].revision] = change[3]
633
# Setup the changes from the tree:
634
# changes maps file_id -> (change, [parent revision_ids])
636
for change in iter_changes:
637
# This probably looks up in basis_inv way to much.
638
if change[1][0] is not None:
639
head_candidate = [basis_inv[change[0]].revision]
642
changes[change[0]] = change, merged_ids.get(change[0],
644
unchanged_merged = set(merged_ids) - set(changes)
645
# Extend the changes dict with synthetic changes to record merges of
647
for file_id in unchanged_merged:
648
# Record a merged version of these items that did not change vs the
649
# basis. This can be either identical parallel changes, or a revert
650
# of a specific file after a merge. The recorded content will be
651
# that of the current tree (which is the same as the basis), but
652
# the per-file graph will reflect a merge.
653
# NB:XXX: We are reconstructing path information we had, this
654
# should be preserved instead.
655
# inv delta change: (file_id, (path_in_source, path_in_target),
656
# changed_content, versioned, parent, name, kind,
659
basis_entry = basis_inv[file_id]
660
except errors.NoSuchId:
661
# a change from basis->some_parents but file_id isn't in basis
662
# so was new in the merge, which means it must have changed
663
# from basis -> current, and as it hasn't the add was reverted
664
# by the user. So we discard this change.
668
(basis_inv.id2path(file_id), tree.id2path(file_id)),
670
(basis_entry.parent_id, basis_entry.parent_id),
671
(basis_entry.name, basis_entry.name),
672
(basis_entry.kind, basis_entry.kind),
673
(basis_entry.executable, basis_entry.executable))
674
changes[file_id] = (change, merged_ids[file_id])
675
# changes contains tuples with the change and a set of inventory
676
# candidates for the file.
678
# old_path, new_path, file_id, new_inventory_entry
679
seen_root = False # Is the root in the basis delta?
680
inv_delta = self._basis_delta
681
modified_rev = self._new_revision_id
682
for change, head_candidates in changes.values():
683
if change[3][1]: # versioned in target.
684
# Several things may be happening here:
685
# We may have a fork in the per-file graph
686
# - record a change with the content from tree
687
# We may have a change against < all trees
688
# - carry over the tree that hasn't changed
689
# We may have a change against all trees
690
# - record the change with the content from tree
693
entry = _entry_factory[kind](file_id, change[5][1],
695
head_set = self._heads(change[0], set(head_candidates))
698
for head_candidate in head_candidates:
699
if head_candidate in head_set:
700
heads.append(head_candidate)
701
head_set.remove(head_candidate)
704
# Could be a carry-over situation:
705
parent_entry_revs = parent_entries.get(file_id, None)
706
if parent_entry_revs:
707
parent_entry = parent_entry_revs.get(heads[0], None)
710
if parent_entry is None:
711
# The parent iter_changes was called against is the one
712
# that is the per-file head, so any change is relevant
713
# iter_changes is valid.
714
carry_over_possible = False
716
# could be a carry over situation
717
# A change against the basis may just indicate a merge,
718
# we need to check the content against the source of the
719
# merge to determine if it was changed after the merge
721
if (parent_entry.kind != entry.kind or
722
parent_entry.parent_id != entry.parent_id or
723
parent_entry.name != entry.name):
724
# Metadata common to all entries has changed
725
# against per-file parent
726
carry_over_possible = False
728
carry_over_possible = True
729
# per-type checks for changes against the parent_entry
732
# Cannot be a carry-over situation
733
carry_over_possible = False
734
# Populate the entry in the delta
736
# XXX: There is still a small race here: If someone reverts the content of a file
737
# after iter_changes examines and decides it has changed,
738
# we will unconditionally record a new version even if some
739
# other process reverts it while commit is running (with
740
# the revert happening after iter_changes did it's
743
entry.executable = True
745
entry.executable = False
746
if (carry_over_possible and
747
parent_entry.executable == entry.executable):
748
# Check the file length, content hash after reading
750
nostore_sha = parent_entry.text_sha1
753
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
755
text = file_obj.read()
759
entry.text_sha1, entry.text_size = self._add_text_to_weave(
760
file_id, text, heads, nostore_sha)
761
yield file_id, change[1][1], (entry.text_sha1, stat_value)
762
except errors.ExistingContent:
763
# No content change against a carry_over parent
764
# Perhaps this should also yield a fs hash update?
766
entry.text_size = parent_entry.text_size
767
entry.text_sha1 = parent_entry.text_sha1
768
elif kind == 'symlink':
770
entry.symlink_target = tree.get_symlink_target(file_id)
771
if (carry_over_possible and
772
parent_entry.symlink_target == entry.symlink_target):
775
self._add_text_to_weave(change[0], '', heads, None)
776
elif kind == 'directory':
777
if carry_over_possible:
780
# Nothing to set on the entry.
781
# XXX: split into the Root and nonRoot versions.
782
if change[1][1] != '' or self.repository.supports_rich_root():
783
self._add_text_to_weave(change[0], '', heads, None)
784
elif kind == 'tree-reference':
785
if not self.repository._format.supports_tree_reference:
786
# This isn't quite sane as an error, but we shouldn't
787
# ever see this code path in practice: tree's don't
788
# permit references when the repo doesn't support tree
790
raise errors.UnsupportedOperation(tree.add_reference,
792
reference_revision = tree.get_reference_revision(change[0])
793
entry.reference_revision = reference_revision
794
if (carry_over_possible and
795
parent_entry.reference_revision == reference_revision):
798
self._add_text_to_weave(change[0], '', heads, None)
800
raise AssertionError('unknown kind %r' % kind)
802
entry.revision = modified_rev
804
entry.revision = parent_entry.revision
807
new_path = change[1][1]
808
inv_delta.append((change[1][0], new_path, change[0], entry))
811
self.new_inventory = None
813
# This should perhaps be guarded by a check that the basis we
814
# commit against is the basis for the commit and if not do a delta
816
self._any_changes = True
818
# housekeeping root entry changes do not affect no-change commits.
819
self._require_root_change(tree)
820
self.basis_delta_revision = basis_revision_id
822
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
823
parent_keys = tuple([(file_id, parent) for parent in parents])
824
return self.repository.texts._add_text(
825
(file_id, self._new_revision_id), parent_keys, new_text,
826
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
829
class RootCommitBuilder(CommitBuilder):
830
"""This commitbuilder actually records the root id"""
832
# the root entry gets versioned properly by this builder.
833
_versioned_root = True
835
def _check_root(self, ie, parent_invs, tree):
836
"""Helper for record_entry_contents.
838
:param ie: An entry being added.
839
:param parent_invs: The inventories of the parent revisions of the
841
:param tree: The tree that is being committed.
844
def _require_root_change(self, tree):
845
"""Enforce an appropriate root object change.
847
This is called once when record_iter_changes is called, if and only if
848
the root was not in the delta calculated by record_iter_changes.
850
:param tree: The tree which is being committed.
852
# versioned roots do not change unless the tree found a change.
855
######################################################################
63
859
class Repository(object):
64
860
"""Repository holding history for one or more branches.
67
863
revisions and file history. It's normally accessed only by the Branch,
68
864
which views a particular line of development through that history.
70
The Repository builds on top of Stores and a Transport, which respectively
71
describe the disk data format and the way of accessing the (possibly
866
The Repository builds on top of some byte storage facilies (the revisions,
867
signatures, inventories, texts and chk_bytes attributes) and a Transport,
868
which respectively provide byte storage and a means to access the (possibly
871
The byte storage facilities are addressed via tuples, which we refer to
872
as 'keys' throughout the code base. Revision_keys, inventory_keys and
873
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
874
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
875
byte string made up of a hash identifier and a hash value.
876
We use this interface because it allows low friction with the underlying
877
code that implements disk indices, network encoding and other parts of
880
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
881
the serialised revisions for the repository. This can be used to obtain
882
revision graph information or to access raw serialised revisions.
883
The result of trying to insert data into the repository via this store
884
is undefined: it should be considered read-only except for implementors
886
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
887
the serialised signatures for the repository. This can be used to
888
obtain access to raw serialised signatures. The result of trying to
889
insert data into the repository via this store is undefined: it should
890
be considered read-only except for implementors of repositories.
891
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
892
the serialised inventories for the repository. This can be used to
893
obtain unserialised inventories. The result of trying to insert data
894
into the repository via this store is undefined: it should be
895
considered read-only except for implementors of repositories.
896
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
897
texts of files and directories for the repository. This can be used to
898
obtain file texts or file graphs. Note that Repository.iter_file_bytes
899
is usually a better interface for accessing file texts.
900
The result of trying to insert data into the repository via this store
901
is undefined: it should be considered read-only except for implementors
903
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
904
any data the repository chooses to store or have indexed by its hash.
905
The result of trying to insert data into the repository via this store
906
is undefined: it should be considered read-only except for implementors
908
:ivar _transport: Transport for file access to repository, typically
909
pointing to .bzr/repository.
912
# What class to use for a CommitBuilder. Often its simpler to change this
913
# in a Repository class subclass rather than to override
914
# get_commit_builder.
915
_commit_builder_class = CommitBuilder
916
# The search regex used by xml based repositories to determine what things
917
# where changed in a single commit.
75
918
_file_ids_altered_regex = lazy_regex.lazy_compile(
76
919
r'file_id="(?P<file_id>[^"]+)"'
77
r'.*revision="(?P<revision_id>[^"]+)"'
920
r'.* revision="(?P<revision_id>[^"]+)"'
81
def add_inventory(self, revid, inv, parents):
82
"""Add the inventory inv to the repository as revid.
84
:param parents: The revision ids of the parents that revid
923
def abort_write_group(self, suppress_errors=False):
924
"""Commit the contents accrued within the current write group.
926
:param suppress_errors: if true, abort_write_group will catch and log
927
unexpected errors that happen during the abort, rather than
928
allowing them to propagate. Defaults to False.
930
:seealso: start_write_group.
932
if self._write_group is not self.get_transaction():
933
# has an unlock or relock occured ?
936
'(suppressed) mismatched lock context and write group. %r, %r',
937
self._write_group, self.get_transaction())
939
raise errors.BzrError(
940
'mismatched lock context and write group. %r, %r' %
941
(self._write_group, self.get_transaction()))
943
self._abort_write_group()
944
except Exception, exc:
945
self._write_group = None
946
if not suppress_errors:
948
mutter('abort_write_group failed')
949
log_exception_quietly()
950
note('bzr: ERROR (ignored): %s', exc)
951
self._write_group = None
953
def _abort_write_group(self):
954
"""Template method for per-repository write group cleanup.
956
This is called during abort before the write group is considered to be
957
finished and should cleanup any internal state accrued during the write
958
group. There is no requirement that data handed to the repository be
959
*not* made available - this is not a rollback - but neither should any
960
attempt be made to ensure that data added is fully commited. Abort is
961
invoked when an error has occured so futher disk or network operations
962
may not be possible or may error and if possible should not be
966
def add_fallback_repository(self, repository):
967
"""Add a repository to use for looking up data not held locally.
969
:param repository: A repository.
971
if not self._format.supports_external_lookups:
972
raise errors.UnstackableRepositoryFormat(self._format, self.base)
974
# This repository will call fallback.unlock() when we transition to
975
# the unlocked state, so we make sure to increment the lock count
976
repository.lock_read()
977
self._check_fallback_repository(repository)
978
self._fallback_repositories.append(repository)
979
self.texts.add_fallback_versioned_files(repository.texts)
980
self.inventories.add_fallback_versioned_files(repository.inventories)
981
self.revisions.add_fallback_versioned_files(repository.revisions)
982
self.signatures.add_fallback_versioned_files(repository.signatures)
983
if self.chk_bytes is not None:
984
self.chk_bytes.add_fallback_versioned_files(repository.chk_bytes)
986
def _check_fallback_repository(self, repository):
987
"""Check that this repository can fallback to repository safely.
989
Raise an error if not.
991
:param repository: A repository to fallback to.
993
return InterRepository._assert_same_model(self, repository)
995
def add_inventory(self, revision_id, inv, parents):
996
"""Add the inventory inv to the repository as revision_id.
998
:param parents: The revision ids of the parents that revision_id
85
999
is known to have and are in the repository already.
87
returns the sha1 of the serialized inventory.
89
_mod_revision.check_not_reserved_id(revid)
90
assert inv.revision_id is None or inv.revision_id == revid, \
91
"Mismatch between inventory revision" \
92
" id and insertion revid (%r, %r)" % (inv.revision_id, revid)
93
assert inv.root is not None
94
inv_text = self.serialise_inventory(inv)
95
inv_sha1 = osutils.sha_string(inv_text)
96
inv_vf = self.control_weaves.get_weave('inventory',
97
self.get_transaction())
98
self._inventory_add_lines(inv_vf, revid, parents, osutils.split_lines(inv_text))
101
def _inventory_add_lines(self, inv_vf, revid, parents, lines):
103
for parent in parents:
105
final_parents.append(parent)
107
inv_vf.add_lines(revid, final_parents, lines)
110
def add_revision(self, rev_id, rev, inv=None, config=None):
111
"""Add rev to the revision store as rev_id.
113
:param rev_id: the revision id to use.
1001
:returns: The validator(which is a sha1 digest, though what is sha'd is
1002
repository format specific) of the serialized inventory.
1004
if not self.is_in_write_group():
1005
raise AssertionError("%r not in write group" % (self,))
1006
_mod_revision.check_not_reserved_id(revision_id)
1007
if not (inv.revision_id is None or inv.revision_id == revision_id):
1008
raise AssertionError(
1009
"Mismatch between inventory revision"
1010
" id and insertion revid (%r, %r)"
1011
% (inv.revision_id, revision_id))
1012
if inv.root is None:
1013
raise AssertionError()
1014
return self._add_inventory_checked(revision_id, inv, parents)
1016
def _add_inventory_checked(self, revision_id, inv, parents):
1017
"""Add inv to the repository after checking the inputs.
1019
This function can be overridden to allow different inventory styles.
1021
:seealso: add_inventory, for the contract.
1023
inv_lines = self._serialise_inventory_to_lines(inv)
1024
return self._inventory_add_lines(revision_id, parents,
1025
inv_lines, check_content=False)
1027
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1028
parents, basis_inv=None, propagate_caches=False):
1029
"""Add a new inventory expressed as a delta against another revision.
1031
See the inventory developers documentation for the theory behind
1034
:param basis_revision_id: The inventory id the delta was created
1035
against. (This does not have to be a direct parent.)
1036
:param delta: The inventory delta (see Inventory.apply_delta for
1038
:param new_revision_id: The revision id that the inventory is being
1040
:param parents: The revision ids of the parents that revision_id is
1041
known to have and are in the repository already. These are supplied
1042
for repositories that depend on the inventory graph for revision
1043
graph access, as well as for those that pun ancestry with delta
1045
:param basis_inv: The basis inventory if it is already known,
1047
:param propagate_caches: If True, the caches for this inventory are
1048
copied to and updated for the result if possible.
1050
:returns: (validator, new_inv)
1051
The validator(which is a sha1 digest, though what is sha'd is
1052
repository format specific) of the serialized inventory, and the
1053
resulting inventory.
1055
if not self.is_in_write_group():
1056
raise AssertionError("%r not in write group" % (self,))
1057
_mod_revision.check_not_reserved_id(new_revision_id)
1058
basis_tree = self.revision_tree(basis_revision_id)
1059
basis_tree.lock_read()
1061
# Note that this mutates the inventory of basis_tree, which not all
1062
# inventory implementations may support: A better idiom would be to
1063
# return a new inventory, but as there is no revision tree cache in
1064
# repository this is safe for now - RBC 20081013
1065
if basis_inv is None:
1066
basis_inv = basis_tree.inventory
1067
basis_inv.apply_delta(delta)
1068
basis_inv.revision_id = new_revision_id
1069
return (self.add_inventory(new_revision_id, basis_inv, parents),
1074
def _inventory_add_lines(self, revision_id, parents, lines,
1075
check_content=True):
1076
"""Store lines in inv_vf and return the sha1 of the inventory."""
1077
parents = [(parent,) for parent in parents]
1078
result = self.inventories.add_lines((revision_id,), parents, lines,
1079
check_content=check_content)[0]
1080
self.inventories._access.flush()
1083
def add_revision(self, revision_id, rev, inv=None, config=None):
1084
"""Add rev to the revision store as revision_id.
1086
:param revision_id: the revision id to use.
114
1087
:param rev: The revision object.
115
1088
:param inv: The inventory for the revision. if None, it will be looked
116
1089
up in the inventory storer
299
1531
control = bzrdir.BzrDir.open(base)
300
1532
return control.open_repository()
302
def copy_content_into(self, destination, revision_id=None, basis=None):
1534
def copy_content_into(self, destination, revision_id=None):
303
1535
"""Make a complete copy of the content in self into destination.
1537
This is a destructive operation! Do not use it on existing
1540
return InterRepository.get(self, destination).copy_content(revision_id)
1542
def commit_write_group(self):
1543
"""Commit the contents accrued within the current write group.
1545
:seealso: start_write_group.
305
This is a destructive operation! Do not use it on existing
308
return InterRepository.get(self, destination).copy_content(revision_id, basis)
310
def fetch(self, source, revision_id=None, pb=None):
1547
:return: it may return an opaque hint that can be passed to 'pack'.
1549
if self._write_group is not self.get_transaction():
1550
# has an unlock or relock occured ?
1551
raise errors.BzrError('mismatched lock context %r and '
1553
(self.get_transaction(), self._write_group))
1554
result = self._commit_write_group()
1555
self._write_group = None
1558
def _commit_write_group(self):
1559
"""Template method for per-repository write group cleanup.
1561
This is called before the write group is considered to be
1562
finished and should ensure that all data handed to the repository
1563
for writing during the write group is safely committed (to the
1564
extent possible considering file system caching etc).
1567
def suspend_write_group(self):
1568
raise errors.UnsuspendableWriteGroup(self)
1570
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1571
"""Return the keys of missing inventory parents for revisions added in
1574
A revision is not complete if the inventory delta for that revision
1575
cannot be calculated. Therefore if the parent inventories of a
1576
revision are not present, the revision is incomplete, and e.g. cannot
1577
be streamed by a smart server. This method finds missing inventory
1578
parents for revisions added in this write group.
1580
if not self._format.supports_external_lookups:
1581
# This is only an issue for stacked repositories
1583
if not self.is_in_write_group():
1584
raise AssertionError('not in a write group')
1586
# XXX: We assume that every added revision already has its
1587
# corresponding inventory, so we only check for parent inventories that
1588
# might be missing, rather than all inventories.
1589
parents = set(self.revisions._index.get_missing_parents())
1590
parents.discard(_mod_revision.NULL_REVISION)
1591
unstacked_inventories = self.inventories._index
1592
present_inventories = unstacked_inventories.get_parent_map(
1593
key[-1:] for key in parents)
1594
parents.difference_update(present_inventories)
1595
if len(parents) == 0:
1596
# No missing parent inventories.
1598
if not check_for_missing_texts:
1599
return set(('inventories', rev_id) for (rev_id,) in parents)
1600
# Ok, now we have a list of missing inventories. But these only matter
1601
# if the inventories that reference them are missing some texts they
1602
# appear to introduce.
1603
# XXX: Texts referenced by all added inventories need to be present,
1604
# but at the moment we're only checking for texts referenced by
1605
# inventories at the graph's edge.
1606
key_deps = self.revisions._index._key_dependencies
1607
key_deps.add_keys(present_inventories)
1608
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1609
file_ids = self.fileids_altered_by_revision_ids(referrers)
1610
missing_texts = set()
1611
for file_id, version_ids in file_ids.iteritems():
1612
missing_texts.update(
1613
(file_id, version_id) for version_id in version_ids)
1614
present_texts = self.texts.get_parent_map(missing_texts)
1615
missing_texts.difference_update(present_texts)
1616
if not missing_texts:
1617
# No texts are missing, so all revisions and their deltas are
1620
# Alternatively the text versions could be returned as the missing
1621
# keys, but this is likely to be less data.
1622
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1625
def refresh_data(self):
1626
"""Re-read any data needed to to synchronise with disk.
1628
This method is intended to be called after another repository instance
1629
(such as one used by a smart server) has inserted data into the
1630
repository. It may not be called during a write group, but may be
1631
called at any other time.
1633
if self.is_in_write_group():
1634
raise errors.InternalBzrError(
1635
"May not refresh_data while in a write group.")
1636
self._refresh_data()
1638
def resume_write_group(self, tokens):
1639
if not self.is_write_locked():
1640
raise errors.NotWriteLocked(self)
1641
if self._write_group:
1642
raise errors.BzrError('already in a write group')
1643
self._resume_write_group(tokens)
1644
# so we can detect unlock/relock - the write group is now entered.
1645
self._write_group = self.get_transaction()
1647
def _resume_write_group(self, tokens):
1648
raise errors.UnsuspendableWriteGroup(self)
1650
def fetch(self, source, revision_id=None, pb=None, find_ghosts=False,
311
1652
"""Fetch the content required to construct revision_id from source.
313
If revision_id is None all content is copied.
1654
If revision_id is None and fetch_spec is None, then all content is
1657
fetch() may not be used when the repository is in a write group -
1658
either finish the current write group before using fetch, or use
1659
fetch before starting the write group.
1661
:param find_ghosts: Find and copy revisions in the source that are
1662
ghosts in the target (and not reachable directly by walking out to
1663
the first-present revision in target from revision_id).
1664
:param revision_id: If specified, all the content needed for this
1665
revision ID will be copied to the target. Fetch will determine for
1666
itself which content needs to be copied.
1667
:param fetch_spec: If specified, a SearchResult or
1668
PendingAncestryResult that describes which revisions to copy. This
1669
allows copying multiple heads at once. Mutually exclusive with
315
return InterRepository.get(source, self).fetch(revision_id=revision_id,
318
def get_commit_builder(self, branch, parents, config, timestamp=None,
319
timezone=None, committer=None, revprops=None,
1672
if fetch_spec is not None and revision_id is not None:
1673
raise AssertionError(
1674
"fetch_spec and revision_id are mutually exclusive.")
1675
if self.is_in_write_group():
1676
raise errors.InternalBzrError(
1677
"May not fetch while in a write group.")
1678
# fast path same-url fetch operations
1679
# TODO: lift out to somewhere common with RemoteRepository
1680
# <https://bugs.edge.launchpad.net/bzr/+bug/401646>
1681
if (self.has_same_location(source)
1682
and fetch_spec is None
1683
and self._has_same_fallbacks(source)):
1684
# check that last_revision is in 'from' and then return a
1686
if (revision_id is not None and
1687
not _mod_revision.is_null(revision_id)):
1688
self.get_revision(revision_id)
1690
# if there is no specific appropriate InterRepository, this will get
1691
# the InterRepository base class, which raises an
1692
# IncompatibleRepositories when asked to fetch.
1693
inter = InterRepository.get(source, self)
1694
return inter.fetch(revision_id=revision_id, pb=pb,
1695
find_ghosts=find_ghosts, fetch_spec=fetch_spec)
1697
def create_bundle(self, target, base, fileobj, format=None):
1698
return serializer.write_bundle(self, target, base, fileobj, format)
1700
def get_commit_builder(self, branch, parents, config, timestamp=None,
1701
timezone=None, committer=None, revprops=None,
320
1702
revision_id=None):
321
1703
"""Obtain a CommitBuilder for this repository.
323
1705
:param branch: Branch to commit to.
324
1706
:param parents: Revision ids of the parents of the new revision.
325
1707
:param config: Configuration to use.
351
1796
# Most control formats need the repository to be specifically
352
1797
# created, but on some old all-in-one formats it's not needed
354
dest_repo = self._format.initialize(a_bzrdir, shared=self.is_shared())
1799
dest_repo = self._format.initialize(a_bzrdir, shared=shared)
355
1800
except errors.UninitializableFormat:
356
1801
dest_repo = a_bzrdir.open_repository()
357
self.copy_content_into(dest_repo, revision_id, basis)
358
1802
return dest_repo
1804
def _get_sink(self):
1805
"""Return a sink for streaming into this repository."""
1806
return StreamSink(self)
1808
def _get_source(self, to_format):
1809
"""Return a source for streaming from this repository."""
1810
return StreamSource(self, to_format)
360
1812
@needs_read_lock
361
1813
def has_revision(self, revision_id):
362
1814
"""True if this repository has a copy of the revision."""
363
return self._revision_store.has_revision_id(revision_id,
364
self.get_transaction())
1815
return revision_id in self.has_revisions((revision_id,))
1818
def has_revisions(self, revision_ids):
1819
"""Probe to find out the presence of multiple revisions.
1821
:param revision_ids: An iterable of revision_ids.
1822
:return: A set of the revision_ids that were present.
1824
parent_map = self.revisions.get_parent_map(
1825
[(rev_id,) for rev_id in revision_ids])
1827
if _mod_revision.NULL_REVISION in revision_ids:
1828
result.add(_mod_revision.NULL_REVISION)
1829
result.update([key[0] for key in parent_map])
1833
def get_revision(self, revision_id):
1834
"""Return the Revision object for a named revision."""
1835
return self.get_revisions([revision_id])[0]
366
1837
@needs_read_lock
367
1838
def get_revision_reconcile(self, revision_id):
368
1839
"""'reconcile' helper routine that allows access to a revision always.
370
1841
This variant of get_revision does not cross check the weave graph
371
1842
against the revision one as get_revision does: but it should only
372
1843
be used by reconcile, or reconcile-alike commands that are correcting
373
1844
or testing the revision graph.
375
if not revision_id or not isinstance(revision_id, basestring):
376
raise errors.InvalidRevisionId(revision_id=revision_id,
378
return self._revision_store.get_revisions([revision_id],
379
self.get_transaction())[0]
1846
return self._get_revisions([revision_id])[0]
380
1848
@needs_read_lock
381
1849
def get_revisions(self, revision_ids):
382
return self._revision_store.get_revisions(revision_ids,
383
self.get_transaction())
1850
"""Get many revisions at once.
1852
Repositories that need to check data on every revision read should
1853
subclass this method.
1855
return self._get_revisions(revision_ids)
1858
def _get_revisions(self, revision_ids):
1859
"""Core work logic to get many revisions without sanity checks."""
1861
for revid, rev in self._iter_revisions(revision_ids):
1863
raise errors.NoSuchRevision(self, revid)
1865
return [revs[revid] for revid in revision_ids]
1867
def _iter_revisions(self, revision_ids):
1868
"""Iterate over revision objects.
1870
:param revision_ids: An iterable of revisions to examine. None may be
1871
passed to request all revisions known to the repository. Note that
1872
not all repositories can find unreferenced revisions; for those
1873
repositories only referenced ones will be returned.
1874
:return: An iterator of (revid, revision) tuples. Absent revisions (
1875
those asked for but not available) are returned as (revid, None).
1877
if revision_ids is None:
1878
revision_ids = self.all_revision_ids()
1880
for rev_id in revision_ids:
1881
if not rev_id or not isinstance(rev_id, basestring):
1882
raise errors.InvalidRevisionId(revision_id=rev_id, branch=self)
1883
keys = [(key,) for key in revision_ids]
1884
stream = self.revisions.get_record_stream(keys, 'unordered', True)
1885
for record in stream:
1886
revid = record.key[0]
1887
if record.storage_kind == 'absent':
1890
text = record.get_bytes_as('fulltext')
1891
rev = self._serializer.read_revision_from_string(text)
385
1894
@needs_read_lock
386
1895
def get_revision_xml(self, revision_id):
387
rev = self.get_revision(revision_id)
1896
# TODO: jam 20070210 This shouldn't be necessary since get_revision
1897
# would have already do it.
1898
# TODO: jam 20070210 Just use _serializer.write_revision_to_string()
1899
# TODO: this can't just be replaced by:
1900
# return self._serializer.write_revision_to_string(
1901
# self.get_revision(revision_id))
1902
# as cStringIO preservers the encoding unlike write_revision_to_string
1903
# or some other call down the path.
1904
rev = self.get_revision(revision_id)
1905
rev_tmp = cStringIO.StringIO()
389
1906
# the current serializer..
390
self._revision_store._serializer.write_revision(rev, rev_tmp)
1907
self._serializer.write_revision(rev, rev_tmp)
392
1909
return rev_tmp.getvalue()
395
def get_revision(self, revision_id):
396
"""Return the Revision object for a named revision"""
397
r = self.get_revision_reconcile(revision_id)
398
# weave corruption can lead to absent revision markers that should be
400
# the following test is reasonably cheap (it needs a single weave read)
401
# and the weave is cached in read transactions. In write transactions
402
# it is not cached but typically we only read a small number of
403
# revisions. For knits when they are introduced we will probably want
404
# to ensure that caching write transactions are in use.
405
inv = self.get_inventory_weave()
406
self._check_revision_parents(r, inv)
410
def get_deltas_for_revisions(self, revisions):
1911
def get_deltas_for_revisions(self, revisions, specific_fileids=None):
411
1912
"""Produce a generator of revision deltas.
413
1914
Note that the input is a sequence of REVISIONS, not revision_ids.
414
1915
Trees will be held in memory until the generator exits.
415
1916
Each delta is relative to the revision's lefthand predecessor.
1918
:param specific_fileids: if not None, the result is filtered
1919
so that only those file-ids, their parents and their
1920
children are included.
1922
# Get the revision-ids of interest
417
1923
required_trees = set()
418
1924
for revision in revisions:
419
1925
required_trees.add(revision.revision_id)
420
1926
required_trees.update(revision.parent_ids[:1])
421
trees = dict((t.get_revision_id(), t) for
422
t in self.revision_trees(required_trees))
1928
# Get the matching filtered trees. Note that it's more
1929
# efficient to pass filtered trees to changes_from() rather
1930
# than doing the filtering afterwards. changes_from() could
1931
# arguably do the filtering itself but it's path-based, not
1932
# file-id based, so filtering before or afterwards is
1934
if specific_fileids is None:
1935
trees = dict((t.get_revision_id(), t) for
1936
t in self.revision_trees(required_trees))
1938
trees = dict((t.get_revision_id(), t) for
1939
t in self._filtered_revision_trees(required_trees,
1942
# Calculate the deltas
423
1943
for revision in revisions:
424
1944
if not revision.parent_ids:
425
old_tree = self.revision_tree(None)
1945
old_tree = self.revision_tree(_mod_revision.NULL_REVISION)
427
1947
old_tree = trees[revision.parent_ids[0]]
428
1948
yield trees[revision.revision_id].changes_from(old_tree)
430
1950
@needs_read_lock
431
def get_revision_delta(self, revision_id):
1951
def get_revision_delta(self, revision_id, specific_fileids=None):
432
1952
"""Return the delta for one revision.
434
1954
The delta is relative to the left-hand predecessor of the
1957
:param specific_fileids: if not None, the result is filtered
1958
so that only those file-ids, their parents and their
1959
children are included.
437
1961
r = self.get_revision(revision_id)
438
return list(self.get_deltas_for_revisions([r]))[0]
440
def _check_revision_parents(self, revision, inventory):
441
"""Private to Repository and Fetch.
443
This checks the parentage of revision in an inventory weave for
444
consistency and is only applicable to inventory-weave-for-ancestry
445
using repository formats & fetchers.
447
weave_parents = inventory.get_parents(revision.revision_id)
448
weave_names = inventory.versions()
449
for parent_id in revision.parent_ids:
450
if parent_id in weave_names:
451
# this parent must not be a ghost.
452
if not parent_id in weave_parents:
454
raise errors.CorruptRepository(self)
1962
return list(self.get_deltas_for_revisions([r],
1963
specific_fileids=specific_fileids))[0]
456
1965
@needs_write_lock
457
1966
def store_revision_signature(self, gpg_strategy, plaintext, revision_id):
458
1967
signature = gpg_strategy.sign(plaintext)
459
self._revision_store.add_revision_signature_text(revision_id,
461
self.get_transaction())
463
def fileids_altered_by_revision_ids(self, revision_ids):
464
"""Find the file ids and versions affected by revisions.
466
:param revisions: an iterable containing revision ids.
467
:return: a dictionary mapping altered file-ids to an iterable of
468
revision_ids. Each altered file-ids has the exact revision_ids that
469
altered it listed explicitly.
471
assert self._serializer.support_altered_by_hack, \
472
("fileids_altered_by_revision_ids only supported for branches "
473
"which store inventory as unnested xml, not on %r" % self)
474
selected_revision_ids = set(revision_ids)
475
w = self.get_inventory_weave()
1968
self.add_signature_text(revision_id, signature)
1971
def add_signature_text(self, revision_id, signature):
1972
self.signatures.add_lines((revision_id,), (),
1973
osutils.split_lines(signature))
1975
def find_text_key_references(self):
1976
"""Find the text key references within the repository.
1978
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1979
to whether they were referred to by the inventory of the
1980
revision_id that they contain. The inventory texts from all present
1981
revision ids are assessed to generate this report.
1983
revision_keys = self.revisions.keys()
1984
w = self.inventories
1985
pb = ui.ui_factory.nested_progress_bar()
1987
return self._find_text_key_references_from_xml_inventory_lines(
1988
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1992
def _find_text_key_references_from_xml_inventory_lines(self,
1994
"""Core routine for extracting references to texts from inventories.
1996
This performs the translation of xml lines to revision ids.
1998
:param line_iterator: An iterator of lines, origin_version_id
1999
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2000
to whether they were referred to by the inventory of the
2001
revision_id that they contain. Note that if that revision_id was
2002
not part of the line_iterator's output then False will be given -
2003
even though it may actually refer to that key.
2005
if not self._serializer.support_altered_by_hack:
2006
raise AssertionError(
2007
"_find_text_key_references_from_xml_inventory_lines only "
2008
"supported for branches which store inventory as unnested xml"
2009
", not on %r" % self)
478
2012
# this code needs to read every new line in every inventory for the
479
2013
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
480
# not present in one of those inventories is unnecessary but not
2014
# not present in one of those inventories is unnecessary but not
481
2015
# harmful because we are filtering by the revision id marker in the
482
# inventory lines : we only select file ids altered in one of those
2016
# inventory lines : we only select file ids altered in one of those
483
2017
# revisions. We don't need to see all lines in the inventory because
484
2018
# only those added in an inventory in rev X can contain a revision=X
494
2028
search = self._file_ids_altered_regex.search
495
2029
unescape = _unescape_xml
496
2030
setdefault = result.setdefault
497
pb = ui.ui_factory.nested_progress_bar()
499
for line in w.iter_lines_added_or_present_in_versions(
500
selected_revision_ids, pb=pb):
504
# One call to match.group() returning multiple items is quite a
505
# bit faster than 2 calls to match.group() each returning 1
506
file_id, revision_id = match.group('file_id', 'revision_id')
508
# Inlining the cache lookups helps a lot when you make 170,000
509
# lines and 350k ids, versus 8.4 unique ids.
510
# Using a cache helps in 2 ways:
511
# 1) Avoids unnecessary decoding calls
512
# 2) Re-uses cached strings, which helps in future set and
514
# (2) is enough that removing encoding entirely along with
515
# the cache (so we are using plain strings) results in no
516
# performance improvement.
518
revision_id = unescape_revid_cache[revision_id]
520
unescaped = unescape(revision_id)
521
unescape_revid_cache[revision_id] = unescaped
522
revision_id = unescaped
524
if revision_id in selected_revision_ids:
526
file_id = unescape_fileid_cache[file_id]
528
unescaped = unescape(file_id)
529
unescape_fileid_cache[file_id] = unescaped
531
setdefault(file_id, set()).add(revision_id)
537
def get_inventory_weave(self):
538
return self.control_weaves.get_weave('inventory',
539
self.get_transaction())
2031
for line, line_key in line_iterator:
2032
match = search(line)
2035
# One call to match.group() returning multiple items is quite a
2036
# bit faster than 2 calls to match.group() each returning 1
2037
file_id, revision_id = match.group('file_id', 'revision_id')
2039
# Inlining the cache lookups helps a lot when you make 170,000
2040
# lines and 350k ids, versus 8.4 unique ids.
2041
# Using a cache helps in 2 ways:
2042
# 1) Avoids unnecessary decoding calls
2043
# 2) Re-uses cached strings, which helps in future set and
2045
# (2) is enough that removing encoding entirely along with
2046
# the cache (so we are using plain strings) results in no
2047
# performance improvement.
2049
revision_id = unescape_revid_cache[revision_id]
2051
unescaped = unescape(revision_id)
2052
unescape_revid_cache[revision_id] = unescaped
2053
revision_id = unescaped
2055
# Note that unconditionally unescaping means that we deserialise
2056
# every fileid, which for general 'pull' is not great, but we don't
2057
# really want to have some many fulltexts that this matters anyway.
2060
file_id = unescape_fileid_cache[file_id]
2062
unescaped = unescape(file_id)
2063
unescape_fileid_cache[file_id] = unescaped
2066
key = (file_id, revision_id)
2067
setdefault(key, False)
2068
if revision_id == line_key[-1]:
2072
def _inventory_xml_lines_for_keys(self, keys):
2073
"""Get a line iterator of the sort needed for findind references.
2075
Not relevant for non-xml inventory repositories.
2077
Ghosts in revision_keys are ignored.
2079
:param revision_keys: The revision keys for the inventories to inspect.
2080
:return: An iterator over (inventory line, revid) for the fulltexts of
2081
all of the xml inventories specified by revision_keys.
2083
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2084
for record in stream:
2085
if record.storage_kind != 'absent':
2086
chunks = record.get_bytes_as('chunked')
2087
revid = record.key[-1]
2088
lines = osutils.chunks_to_lines(chunks)
2092
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2094
"""Helper routine for fileids_altered_by_revision_ids.
2096
This performs the translation of xml lines to revision ids.
2098
:param line_iterator: An iterator of lines, origin_version_id
2099
:param revision_keys: The revision ids to filter for. This should be a
2100
set or other type which supports efficient __contains__ lookups, as
2101
the revision key from each parsed line will be looked up in the
2102
revision_keys filter.
2103
:return: a dictionary mapping altered file-ids to an iterable of
2104
revision_ids. Each altered file-ids has the exact revision_ids that
2105
altered it listed explicitly.
2107
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2108
line_iterator).iterkeys())
2109
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2110
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2111
self._inventory_xml_lines_for_keys(parent_keys)))
2112
new_keys = seen - parent_seen
2114
setdefault = result.setdefault
2115
for key in new_keys:
2116
setdefault(key[0], set()).add(key[-1])
2119
def _find_parent_ids_of_revisions(self, revision_ids):
2120
"""Find all parent ids that are mentioned in the revision graph.
2122
:return: set of revisions that are parents of revision_ids which are
2123
not part of revision_ids themselves
2125
parent_map = self.get_parent_map(revision_ids)
2127
map(parent_ids.update, parent_map.itervalues())
2128
parent_ids.difference_update(revision_ids)
2129
parent_ids.discard(_mod_revision.NULL_REVISION)
2132
def _find_parent_keys_of_revisions(self, revision_keys):
2133
"""Similar to _find_parent_ids_of_revisions, but used with keys.
2135
:param revision_keys: An iterable of revision_keys.
2136
:return: The parents of all revision_keys that are not already in
2139
parent_map = self.revisions.get_parent_map(revision_keys)
2141
map(parent_keys.update, parent_map.itervalues())
2142
parent_keys.difference_update(revision_keys)
2143
parent_keys.discard(_mod_revision.NULL_REVISION)
2146
def fileids_altered_by_revision_ids(self, revision_ids, _inv_weave=None):
2147
"""Find the file ids and versions affected by revisions.
2149
:param revisions: an iterable containing revision ids.
2150
:param _inv_weave: The inventory weave from this repository or None.
2151
If None, the inventory weave will be opened automatically.
2152
:return: a dictionary mapping altered file-ids to an iterable of
2153
revision_ids. Each altered file-ids has the exact revision_ids that
2154
altered it listed explicitly.
2156
selected_keys = set((revid,) for revid in revision_ids)
2157
w = _inv_weave or self.inventories
2158
pb = ui.ui_factory.nested_progress_bar()
2160
return self._find_file_ids_from_xml_inventory_lines(
2161
w.iter_lines_added_or_present_in_keys(
2162
selected_keys, pb=pb),
2167
def iter_files_bytes(self, desired_files):
2168
"""Iterate through file versions.
2170
Files will not necessarily be returned in the order they occur in
2171
desired_files. No specific order is guaranteed.
2173
Yields pairs of identifier, bytes_iterator. identifier is an opaque
2174
value supplied by the caller as part of desired_files. It should
2175
uniquely identify the file version in the caller's context. (Examples:
2176
an index number or a TreeTransform trans_id.)
2178
bytes_iterator is an iterable of bytestrings for the file. The
2179
kind of iterable and length of the bytestrings are unspecified, but for
2180
this implementation, it is a list of bytes produced by
2181
VersionedFile.get_record_stream().
2183
:param desired_files: a list of (file_id, revision_id, identifier)
2187
for file_id, revision_id, callable_data in desired_files:
2188
text_keys[(file_id, revision_id)] = callable_data
2189
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2190
if record.storage_kind == 'absent':
2191
raise errors.RevisionNotPresent(record.key, self)
2192
yield text_keys[record.key], record.get_bytes_as('chunked')
2194
def _generate_text_key_index(self, text_key_references=None,
2196
"""Generate a new text key index for the repository.
2198
This is an expensive function that will take considerable time to run.
2200
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2201
list of parents, also text keys. When a given key has no parents,
2202
the parents list will be [NULL_REVISION].
2204
# All revisions, to find inventory parents.
2205
if ancestors is None:
2206
graph = self.get_graph()
2207
ancestors = graph.get_parent_map(self.all_revision_ids())
2208
if text_key_references is None:
2209
text_key_references = self.find_text_key_references()
2210
pb = ui.ui_factory.nested_progress_bar()
2212
return self._do_generate_text_key_index(ancestors,
2213
text_key_references, pb)
2217
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2218
"""Helper for _generate_text_key_index to avoid deep nesting."""
2219
revision_order = tsort.topo_sort(ancestors)
2220
invalid_keys = set()
2222
for revision_id in revision_order:
2223
revision_keys[revision_id] = set()
2224
text_count = len(text_key_references)
2225
# a cache of the text keys to allow reuse; costs a dict of all the
2226
# keys, but saves a 2-tuple for every child of a given key.
2228
for text_key, valid in text_key_references.iteritems():
2230
invalid_keys.add(text_key)
2232
revision_keys[text_key[1]].add(text_key)
2233
text_key_cache[text_key] = text_key
2234
del text_key_references
2236
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2237
NULL_REVISION = _mod_revision.NULL_REVISION
2238
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2239
# too small for large or very branchy trees. However, for 55K path
2240
# trees, it would be easy to use too much memory trivially. Ideally we
2241
# could gauge this by looking at available real memory etc, but this is
2242
# always a tricky proposition.
2243
inventory_cache = lru_cache.LRUCache(10)
2244
batch_size = 10 # should be ~150MB on a 55K path tree
2245
batch_count = len(revision_order) / batch_size + 1
2247
pb.update("Calculating text parents", processed_texts, text_count)
2248
for offset in xrange(batch_count):
2249
to_query = revision_order[offset * batch_size:(offset + 1) *
2253
for revision_id in to_query:
2254
parent_ids = ancestors[revision_id]
2255
for text_key in revision_keys[revision_id]:
2256
pb.update("Calculating text parents", processed_texts)
2257
processed_texts += 1
2258
candidate_parents = []
2259
for parent_id in parent_ids:
2260
parent_text_key = (text_key[0], parent_id)
2262
check_parent = parent_text_key not in \
2263
revision_keys[parent_id]
2265
# the parent parent_id is a ghost:
2266
check_parent = False
2267
# truncate the derived graph against this ghost.
2268
parent_text_key = None
2270
# look at the parent commit details inventories to
2271
# determine possible candidates in the per file graph.
2274
inv = inventory_cache[parent_id]
2276
inv = self.revision_tree(parent_id).inventory
2277
inventory_cache[parent_id] = inv
2279
parent_entry = inv[text_key[0]]
2280
except (KeyError, errors.NoSuchId):
2282
if parent_entry is not None:
2284
text_key[0], parent_entry.revision)
2286
parent_text_key = None
2287
if parent_text_key is not None:
2288
candidate_parents.append(
2289
text_key_cache[parent_text_key])
2290
parent_heads = text_graph.heads(candidate_parents)
2291
new_parents = list(parent_heads)
2292
new_parents.sort(key=lambda x:candidate_parents.index(x))
2293
if new_parents == []:
2294
new_parents = [NULL_REVISION]
2295
text_index[text_key] = new_parents
2297
for text_key in invalid_keys:
2298
text_index[text_key] = [NULL_REVISION]
2301
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2302
"""Get an iterable listing the keys of all the data introduced by a set
2305
The keys will be ordered so that the corresponding items can be safely
2306
fetched and inserted in that order.
2308
:returns: An iterable producing tuples of (knit-kind, file-id,
2309
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2310
'revisions'. file-id is None unless knit-kind is 'file'.
2312
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2315
for result in self._find_non_file_keys_to_fetch(revision_ids):
2318
def _find_file_keys_to_fetch(self, revision_ids, pb):
2319
# XXX: it's a bit weird to control the inventory weave caching in this
2320
# generator. Ideally the caching would be done in fetch.py I think. Or
2321
# maybe this generator should explicitly have the contract that it
2322
# should not be iterated until the previously yielded item has been
2324
inv_w = self.inventories
2326
# file ids that changed
2327
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2329
num_file_ids = len(file_ids)
2330
for file_id, altered_versions in file_ids.iteritems():
2332
pb.update("fetch texts", count, num_file_ids)
2334
yield ("file", file_id, altered_versions)
2336
def _find_non_file_keys_to_fetch(self, revision_ids):
2338
yield ("inventory", None, revision_ids)
2341
# XXX: Note ATM no callers actually pay attention to this return
2342
# instead they just use the list of revision ids and ignore
2343
# missing sigs. Consider removing this work entirely
2344
revisions_with_signatures = set(self.signatures.get_parent_map(
2345
[(r,) for r in revision_ids]))
2346
revisions_with_signatures = set(
2347
[r for (r,) in revisions_with_signatures])
2348
revisions_with_signatures.intersection_update(revision_ids)
2349
yield ("signatures", None, revisions_with_signatures)
2352
yield ("revisions", None, revision_ids)
541
2354
@needs_read_lock
542
2355
def get_inventory(self, revision_id):
543
"""Get Inventory object by hash."""
544
return self.deserialise_inventory(
545
revision_id, self.get_inventory_xml(revision_id))
2356
"""Get Inventory object by revision id."""
2357
return self.iter_inventories([revision_id]).next()
2359
def iter_inventories(self, revision_ids, ordering=None):
2360
"""Get many inventories by revision_ids.
2362
This will buffer some or all of the texts used in constructing the
2363
inventories in memory, but will only parse a single inventory at a
2366
:param revision_ids: The expected revision ids of the inventories.
2367
:param ordering: optional ordering, e.g. 'topological'. If not
2368
specified, the order of revision_ids will be preserved (by
2369
buffering if necessary).
2370
:return: An iterator of inventories.
2372
if ((None in revision_ids)
2373
or (_mod_revision.NULL_REVISION in revision_ids)):
2374
raise ValueError('cannot get null revision inventory')
2375
return self._iter_inventories(revision_ids, ordering)
2377
def _iter_inventories(self, revision_ids, ordering):
2378
"""single-document based inventory iteration."""
2379
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2380
for text, revision_id in inv_xmls:
2381
yield self.deserialise_inventory(revision_id, text)
2383
def _iter_inventory_xmls(self, revision_ids, ordering):
2384
if ordering is None:
2385
order_as_requested = True
2386
ordering = 'unordered'
2388
order_as_requested = False
2389
keys = [(revision_id,) for revision_id in revision_ids]
2392
if order_as_requested:
2393
key_iter = iter(keys)
2394
next_key = key_iter.next()
2395
stream = self.inventories.get_record_stream(keys, ordering, True)
2397
for record in stream:
2398
if record.storage_kind != 'absent':
2399
chunks = record.get_bytes_as('chunked')
2400
if order_as_requested:
2401
text_chunks[record.key] = chunks
2403
yield ''.join(chunks), record.key[-1]
2405
raise errors.NoSuchRevision(self, record.key)
2406
if order_as_requested:
2407
# Yield as many results as we can while preserving order.
2408
while next_key in text_chunks:
2409
chunks = text_chunks.pop(next_key)
2410
yield ''.join(chunks), next_key[-1]
2412
next_key = key_iter.next()
2413
except StopIteration:
2414
# We still want to fully consume the get_record_stream,
2415
# just in case it is not actually finished at this point
547
2419
def deserialise_inventory(self, revision_id, xml):
548
"""Transform the xml into an inventory object.
2420
"""Transform the xml into an inventory object.
550
2422
:param revision_id: The expected revision id of the inventory.
551
2423
:param xml: A serialised inventory.
553
result = self._serializer.read_inventory_from_string(xml)
554
result.root.revision = revision_id
2425
result = self._serializer.read_inventory_from_string(xml, revision_id,
2426
entry_cache=self._inventory_entry_cache)
2427
if result.revision_id != revision_id:
2428
raise AssertionError('revision id mismatch %s != %s' % (
2429
result.revision_id, revision_id))
557
2432
def serialise_inventory(self, inv):
558
2433
return self._serializer.write_inventory_to_string(inv)
2435
def _serialise_inventory_to_lines(self, inv):
2436
return self._serializer.write_inventory_to_lines(inv)
2438
def get_serializer_format(self):
2439
return self._serializer.format_num
560
2441
@needs_read_lock
561
2442
def get_inventory_xml(self, revision_id):
562
2443
"""Get inventory XML as a file object."""
2444
texts = self._iter_inventory_xmls([revision_id], 'unordered')
564
assert isinstance(revision_id, basestring), type(revision_id)
565
iw = self.get_inventory_weave()
566
return iw.get_text(revision_id)
2446
text, revision_id = texts.next()
2447
except StopIteration:
568
2448
raise errors.HistoryMissing(self, 'inventory', revision_id)
570
2451
@needs_read_lock
571
2452
def get_inventory_sha1(self, revision_id):
697
2537
reconciler = RepoReconciler(self, thorough=thorough)
698
2538
reconciler.reconcile()
699
2539
return reconciler
2541
def _refresh_data(self):
2542
"""Helper called from lock_* to ensure coherency with disk.
2544
The default implementation does nothing; it is however possible
2545
for repositories to maintain loaded indices across multiple locks
2546
by checking inside their implementation of this method to see
2547
whether their indices are still valid. This depends of course on
2548
the disk format being validatable in this manner. This method is
2549
also called by the refresh_data() public interface to cause a refresh
2550
to occur while in a write lock so that data inserted by a smart server
2551
push operation is visible on the client's instance of the physical
701
2555
@needs_read_lock
702
2556
def revision_tree(self, revision_id):
703
2557
"""Return Tree for a revision on this branch.
705
`revision_id` may be None for the empty tree revision.
2559
`revision_id` may be NULL_REVISION for the empty tree revision.
2561
revision_id = _mod_revision.ensure_null(revision_id)
707
2562
# TODO: refactor this to use an existing revision object
708
2563
# so we don't need to read it in twice.
709
if revision_id is None or revision_id == _mod_revision.NULL_REVISION:
710
return RevisionTree(self, Inventory(root_id=None),
2564
if revision_id == _mod_revision.NULL_REVISION:
2565
return RevisionTree(self, Inventory(root_id=None),
711
2566
_mod_revision.NULL_REVISION)
713
2568
inv = self.get_revision_inventory(revision_id)
714
2569
return RevisionTree(self, inv, revision_id)
717
2571
def revision_trees(self, revision_ids):
718
"""Return Tree for a revision on this branch.
720
`revision_id` may not be None or 'null:'"""
721
assert None not in revision_ids
722
assert _mod_revision.NULL_REVISION not in revision_ids
723
texts = self.get_inventory_weave().get_texts(revision_ids)
724
for text, revision_id in zip(texts, revision_ids):
725
inv = self.deserialise_inventory(revision_id, text)
726
yield RevisionTree(self, inv, revision_id)
2572
"""Return Trees for revisions in this repository.
2574
:param revision_ids: a sequence of revision-ids;
2575
a revision-id may not be None or 'null:'
2577
inventories = self.iter_inventories(revision_ids)
2578
for inv in inventories:
2579
yield RevisionTree(self, inv, inv.revision_id)
2581
def _filtered_revision_trees(self, revision_ids, file_ids):
2582
"""Return Tree for a revision on this branch with only some files.
2584
:param revision_ids: a sequence of revision-ids;
2585
a revision-id may not be None or 'null:'
2586
:param file_ids: if not None, the result is filtered
2587
so that only those file-ids, their parents and their
2588
children are included.
2590
inventories = self.iter_inventories(revision_ids)
2591
for inv in inventories:
2592
# Should we introduce a FilteredRevisionTree class rather
2593
# than pre-filter the inventory here?
2594
filtered_inv = inv.filter(file_ids)
2595
yield RevisionTree(self, filtered_inv, filtered_inv.revision_id)
728
2597
@needs_read_lock
729
def get_ancestry(self, revision_id):
2598
def get_ancestry(self, revision_id, topo_sorted=True):
730
2599
"""Return a list of revision-ids integrated by a revision.
732
The first element of the list is always None, indicating the origin
733
revision. This might change when we have history horizons, or
2601
The first element of the list is always None, indicating the origin
2602
revision. This might change when we have history horizons, or
734
2603
perhaps we should have a new API.
736
2605
This is topologically sorted.
738
if revision_id is None:
2607
if _mod_revision.is_null(revision_id):
740
2609
if not self.has_revision(revision_id):
741
2610
raise errors.NoSuchRevision(self, revision_id)
742
w = self.get_inventory_weave()
743
candidates = w.get_ancestry(revision_id)
744
return [None] + candidates # self._eliminate_revisions_not_present(candidates)
747
def print_file(self, file, revision_id):
748
"""Print `file` to stdout.
750
FIXME RBC 20060125 as John Meinel points out this is a bad api
751
- it writes to stdout, it assumes that that is valid etc. Fix
752
by creating a new more flexible convenience function.
2611
graph = self.get_graph()
2613
search = graph._make_breadth_first_searcher([revision_id])
2616
found, ghosts = search.next_with_ghosts()
2617
except StopIteration:
2620
if _mod_revision.NULL_REVISION in keys:
2621
keys.remove(_mod_revision.NULL_REVISION)
2623
parent_map = graph.get_parent_map(keys)
2624
keys = tsort.topo_sort(parent_map)
2625
return [None] + list(keys)
2627
def pack(self, hint=None):
2628
"""Compress the data within the repository.
2630
This operation only makes sense for some repository types. For other
2631
types it should be a no-op that just returns.
2633
This stub method does not require a lock, but subclasses should use
2634
@needs_write_lock as this is a long running call its reasonable to
2635
implicitly lock for the user.
2637
:param hint: If not supplied, the whole repository is packed.
2638
If supplied, the repository may use the hint parameter as a
2639
hint for the parts of the repository to pack. A hint can be
2640
obtained from the result of commit_write_group(). Out of
2641
date hints are simply ignored, because concurrent operations
2642
can obsolete them rapidly.
754
tree = self.revision_tree(revision_id)
755
# use inventory as it was in that revision
756
file_id = tree.inventory.path2id(file)
758
# TODO: jam 20060427 Write a test for this code path
759
# it had a bug in it, and was raising the wrong
761
raise errors.BzrError("%r is not present in revision %s" % (file, revision_id))
762
tree.print_file(file_id)
764
2645
def get_transaction(self):
765
2646
return self.control_files.get_transaction()
767
def revision_parents(self, revid):
768
return self.get_inventory_weave().parent_names(revid)
2648
def get_parent_map(self, revision_ids):
2649
"""See graph.StackedParentsProvider.get_parent_map"""
2650
# revisions index works in keys; this just works in revisions
2651
# therefore wrap and unwrap
2654
for revision_id in revision_ids:
2655
if revision_id == _mod_revision.NULL_REVISION:
2656
result[revision_id] = ()
2657
elif revision_id is None:
2658
raise ValueError('get_parent_map(None) is not valid')
2660
query_keys.append((revision_id ,))
2661
for ((revision_id,), parent_keys) in \
2662
self.revisions.get_parent_map(query_keys).iteritems():
2664
result[revision_id] = tuple(parent_revid
2665
for (parent_revid,) in parent_keys)
2667
result[revision_id] = (_mod_revision.NULL_REVISION,)
2670
def _make_parents_provider(self):
2673
def get_graph(self, other_repository=None):
2674
"""Return the graph walker for this repository format"""
2675
parents_provider = self._make_parents_provider()
2676
if (other_repository is not None and
2677
not self.has_same_location(other_repository)):
2678
parents_provider = graph.StackedParentsProvider(
2679
[parents_provider, other_repository._make_parents_provider()])
2680
return graph.Graph(parents_provider)
2682
def _get_versioned_file_checker(self, text_key_references=None,
2684
"""Return an object suitable for checking versioned files.
2686
:param text_key_references: if non-None, an already built
2687
dictionary mapping text keys ((fileid, revision_id) tuples)
2688
to whether they were referred to by the inventory of the
2689
revision_id that they contain. If None, this will be
2691
:param ancestors: Optional result from
2692
self.get_graph().get_parent_map(self.all_revision_ids()) if already
2695
return _VersionedFileChecker(self,
2696
text_key_references=text_key_references, ancestors=ancestors)
2698
def revision_ids_to_search_result(self, result_set):
2699
"""Convert a set of revision ids to a graph SearchResult."""
2700
result_parents = set()
2701
for parents in self.get_graph().get_parent_map(
2702
result_set).itervalues():
2703
result_parents.update(parents)
2704
included_keys = result_set.intersection(result_parents)
2705
start_keys = result_set.difference(included_keys)
2706
exclude_keys = result_parents.difference(result_set)
2707
result = graph.SearchResult(start_keys, exclude_keys,
2708
len(result_set), result_set)
770
2711
@needs_write_lock
771
2712
def set_make_working_trees(self, new_value):
1235
3403
content is copied.
1236
3404
:param pb: optional progress bar to use for progress reports. If not
1237
3405
provided a default one will be created.
1239
Returns the copied revision count and the failed revisions in a tuple:
1242
raise NotImplementedError(self.fetch)
3408
from bzrlib.fetch import RepoFetcher
3409
f = RepoFetcher(to_repository=self.target,
3410
from_repository=self.source,
3411
last_revision=revision_id,
3412
fetch_spec=fetch_spec,
3413
pb=pb, find_ghosts=find_ghosts)
3415
def _walk_to_common_revisions(self, revision_ids):
3416
"""Walk out from revision_ids in source to revisions target has.
3418
:param revision_ids: The start point for the search.
3419
:return: A set of revision ids.
3421
target_graph = self.target.get_graph()
3422
revision_ids = frozenset(revision_ids)
3423
missing_revs = set()
3424
source_graph = self.source.get_graph()
3425
# ensure we don't pay silly lookup costs.
3426
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3427
null_set = frozenset([_mod_revision.NULL_REVISION])
3428
searcher_exhausted = False
3432
# Iterate the searcher until we have enough next_revs
3433
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3435
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3436
next_revs.update(next_revs_part)
3437
ghosts.update(ghosts_part)
3438
except StopIteration:
3439
searcher_exhausted = True
3441
# If there are ghosts in the source graph, and the caller asked for
3442
# them, make sure that they are present in the target.
3443
# We don't care about other ghosts as we can't fetch them and
3444
# haven't been asked to.
3445
ghosts_to_check = set(revision_ids.intersection(ghosts))
3446
revs_to_get = set(next_revs).union(ghosts_to_check)
3448
have_revs = set(target_graph.get_parent_map(revs_to_get))
3449
# we always have NULL_REVISION present.
3450
have_revs = have_revs.union(null_set)
3451
# Check if the target is missing any ghosts we need.
3452
ghosts_to_check.difference_update(have_revs)
3454
# One of the caller's revision_ids is a ghost in both the
3455
# source and the target.
3456
raise errors.NoSuchRevision(
3457
self.source, ghosts_to_check.pop())
3458
missing_revs.update(next_revs - have_revs)
3459
# Because we may have walked past the original stop point, make
3460
# sure everything is stopped
3461
stop_revs = searcher.find_seen_ancestors(have_revs)
3462
searcher.stop_searching_any(stop_revs)
3463
if searcher_exhausted:
3465
return searcher.get_result()
1244
3467
@needs_read_lock
1245
def missing_revision_ids(self, revision_id=None):
3468
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1246
3469
"""Return the revision ids that source has that target does not.
1248
These are returned in topological order.
1250
3471
:param revision_id: only return revision ids included by this
3473
:param find_ghosts: If True find missing revisions in deep history
3474
rather than just finding the surface difference.
3475
:return: A bzrlib.graph.SearchResult.
3477
# stop searching at found target revisions.
3478
if not find_ghosts and revision_id is not None:
3479
return self._walk_to_common_revisions([revision_id])
1253
3480
# generic, possibly worst case, slow code path.
1254
3481
target_ids = set(self.target.all_revision_ids())
1255
3482
if revision_id is not None:
1256
3483
source_ids = self.source.get_ancestry(revision_id)
1257
assert source_ids[0] is None
3484
if source_ids[0] is not None:
3485
raise AssertionError()
1258
3486
source_ids.pop(0)
1260
3488
source_ids = self.source.all_revision_ids()
1261
3489
result_set = set(source_ids).difference(target_ids)
1262
# this may look like a no-op: its not. It preserves the ordering
1263
# other_ids had while only returning the members from other_ids
1264
# that we've decided we need.
1265
return [rev_id for rev_id in source_ids if rev_id in result_set]
3490
return self.source.revision_ids_to_search_result(result_set)
3493
def _same_model(source, target):
3494
"""True if source and target have the same data representation.
3496
Note: this is always called on the base class; overriding it in a
3497
subclass will have no effect.
3500
InterRepository._assert_same_model(source, target)
3502
except errors.IncompatibleRepositories, e:
3506
def _assert_same_model(source, target):
3507
"""Raise an exception if two repositories do not use the same model.
3509
if source.supports_rich_root() != target.supports_rich_root():
3510
raise errors.IncompatibleRepositories(source, target,
3511
"different rich-root support")
3512
if source._serializer != target._serializer:
3513
raise errors.IncompatibleRepositories(source, target,
3514
"different serializers")
1268
3517
class InterSameDataRepository(InterRepository):
1269
3518
"""Code for converting between repositories that represent the same data.
1271
3520
Data format and model must match for this to work.
1275
3524
def _get_repo_format_to_test(self):
1276
"""Repository format for testing with."""
1277
return RepositoryFormat.get_default_format()
3525
"""Repository format for testing with.
3527
InterSameData can pull from subtree to subtree and from non-subtree to
3528
non-subtree, so we test this with the richest repository format.
3530
from bzrlib.repofmt import knitrepo
3531
return knitrepo.RepositoryFormatKnit3()
1280
3534
def is_compatible(source, target):
1281
if not isinstance(source, Repository):
1283
if not isinstance(target, Repository):
1285
if source._format.rich_root_data == target._format.rich_root_data:
1291
def copy_content(self, revision_id=None, basis=None):
1292
"""Make a complete copy of the content in self into destination.
1294
This is a destructive operation! Do not use it on existing
1297
:param revision_id: Only copy the content needed to construct
1298
revision_id and its parents.
1299
:param basis: Copy the needed data preferentially from basis.
1302
self.target.set_make_working_trees(self.source.make_working_trees())
1303
except NotImplementedError:
1305
# grab the basis available data
1306
if basis is not None:
1307
self.target.fetch(basis, revision_id=revision_id)
1308
# but don't bother fetching if we have the needed data now.
1309
if (revision_id not in (None, _mod_revision.NULL_REVISION) and
1310
self.target.has_revision(revision_id)):
1312
self.target.fetch(self.source, revision_id=revision_id)
1315
def fetch(self, revision_id=None, pb=None):
1316
"""See InterRepository.fetch()."""
1317
from bzrlib.fetch import GenericRepoFetcher
1318
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1319
self.source, self.source._format, self.target,
1320
self.target._format)
1321
f = GenericRepoFetcher(to_repository=self.target,
1322
from_repository=self.source,
1323
last_revision=revision_id,
1325
return f.count_copied, f.failed_revisions
3535
return InterRepository._same_model(source, target)
1328
3538
class InterWeaveRepo(InterSameDataRepository):
1329
"""Optimised code paths between Weave based repositories."""
3539
"""Optimised code paths between Weave based repositories.
3541
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3542
implemented lazy inter-object optimisation.
1332
3546
def _get_repo_format_to_test(self):
1526
3715
def is_compatible(source, target):
1527
if not isinstance(source, Repository):
1529
if not isinstance(target, Repository):
1531
if not source._format.rich_root_data and target._format.rich_root_data:
3716
"""Be compatible with Knit2 source and Knit3 target"""
3717
# This is redundant with format.check_conversion_target(), however that
3718
# raises an exception, and we just want to say "False" as in we won't
3719
# support converting between these formats.
3720
if 'IDS_never' in debug.debug_flags:
3722
if source.supports_rich_root() and not target.supports_rich_root():
3724
if (source._format.supports_tree_reference
3725
and not target._format.supports_tree_reference):
3727
if target._fallback_repositories and target._format.supports_chks:
3728
# IDS doesn't know how to copy CHKs for the parent inventories it
3729
# adds to stacked repos.
3731
if 'IDS_always' in debug.debug_flags:
1537
def fetch(self, revision_id=None, pb=None):
1538
"""See InterRepository.fetch()."""
1539
from bzrlib.fetch import Model1toKnit2Fetcher
1540
f = Model1toKnit2Fetcher(to_repository=self.target,
1541
from_repository=self.source,
1542
last_revision=revision_id,
1544
return f.count_copied, f.failed_revisions
1547
def copy_content(self, revision_id=None, basis=None):
1548
"""Make a complete copy of the content in self into destination.
1550
This is a destructive operation! Do not use it on existing
1553
:param revision_id: Only copy the content needed to construct
1554
revision_id and its parents.
1555
:param basis: Copy the needed data preferentially from basis.
1558
self.target.set_make_working_trees(self.source.make_working_trees())
1559
except NotImplementedError:
1561
# grab the basis available data
1562
if basis is not None:
1563
self.target.fetch(basis, revision_id=revision_id)
1564
# but don't bother fetching if we have the needed data now.
1565
if (revision_id not in (None, _mod_revision.NULL_REVISION) and
1566
self.target.has_revision(revision_id)):
1568
self.target.fetch(self.source, revision_id=revision_id)
1571
class InterKnit1and2(InterKnitRepo):
1574
def _get_repo_format_to_test(self):
1578
def is_compatible(source, target):
1579
"""Be compatible with Knit1 source and Knit2 target"""
1580
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit2
1582
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit1, \
1583
RepositoryFormatKnit2
1584
return (isinstance(source._format, (RepositoryFormatKnit1)) and
1585
isinstance(target._format, (RepositoryFormatKnit2)))
1586
except AttributeError:
1590
def fetch(self, revision_id=None, pb=None):
1591
"""See InterRepository.fetch()."""
1592
from bzrlib.fetch import Knit1to2Fetcher
1593
mutter("Using fetch logic to copy between %s(%s) and %s(%s)",
1594
self.source, self.source._format, self.target,
1595
self.target._format)
1596
f = Knit1to2Fetcher(to_repository=self.target,
1597
from_repository=self.source,
1598
last_revision=revision_id,
1600
return f.count_copied, f.failed_revisions
3733
# Only use this code path for local source and target. IDS does far
3734
# too much IO (both bandwidth and roundtrips) over a network.
3735
if not source.bzrdir.transport.base.startswith('file:///'):
3737
if not target.bzrdir.transport.base.startswith('file:///'):
3741
def _get_trees(self, revision_ids, cache):
3743
for rev_id in revision_ids:
3745
possible_trees.append((rev_id, cache[rev_id]))
3747
# Not cached, but inventory might be present anyway.
3749
tree = self.source.revision_tree(rev_id)
3750
except errors.NoSuchRevision:
3751
# Nope, parent is ghost.
3754
cache[rev_id] = tree
3755
possible_trees.append((rev_id, tree))
3756
return possible_trees
3758
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3759
"""Get the best delta and base for this revision.
3761
:return: (basis_id, delta)
3764
# Generate deltas against each tree, to find the shortest.
3765
texts_possibly_new_in_tree = set()
3766
for basis_id, basis_tree in possible_trees:
3767
delta = tree.inventory._make_delta(basis_tree.inventory)
3768
for old_path, new_path, file_id, new_entry in delta:
3769
if new_path is None:
3770
# This file_id isn't present in the new rev, so we don't
3774
# Rich roots are handled elsewhere...
3776
kind = new_entry.kind
3777
if kind != 'directory' and kind != 'file':
3778
# No text record associated with this inventory entry.
3780
# This is a directory or file that has changed somehow.
3781
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3782
deltas.append((len(delta), basis_id, delta))
3784
return deltas[0][1:]
3786
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3787
"""Find all parent revisions that are absent, but for which the
3788
inventory is present, and copy those inventories.
3790
This is necessary to preserve correctness when the source is stacked
3791
without fallbacks configured. (Note that in cases like upgrade the
3792
source may be not have _fallback_repositories even though it is
3796
for parents in parent_map.values():
3797
parent_revs.update(parents)
3798
present_parents = self.source.get_parent_map(parent_revs)
3799
absent_parents = set(parent_revs).difference(present_parents)
3800
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3801
(rev_id,) for rev_id in absent_parents)
3802
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3803
for parent_tree in self.source.revision_trees(parent_inv_ids):
3804
current_revision_id = parent_tree.get_revision_id()
3805
parents_parents_keys = parent_invs_keys_for_stacking[
3806
(current_revision_id,)]
3807
parents_parents = [key[-1] for key in parents_parents_keys]
3808
basis_id = _mod_revision.NULL_REVISION
3809
basis_tree = self.source.revision_tree(basis_id)
3810
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3811
self.target.add_inventory_by_delta(
3812
basis_id, delta, current_revision_id, parents_parents)
3813
cache[current_revision_id] = parent_tree
3815
def _fetch_batch(self, revision_ids, basis_id, cache):
3816
"""Fetch across a few revisions.
3818
:param revision_ids: The revisions to copy
3819
:param basis_id: The revision_id of a tree that must be in cache, used
3820
as a basis for delta when no other base is available
3821
:param cache: A cache of RevisionTrees that we can use.
3822
:return: The revision_id of the last converted tree. The RevisionTree
3823
for it will be in cache
3825
# Walk though all revisions; get inventory deltas, copy referenced
3826
# texts that delta references, insert the delta, revision and
3828
root_keys_to_create = set()
3831
pending_revisions = []
3832
parent_map = self.source.get_parent_map(revision_ids)
3833
self._fetch_parent_invs_for_stacking(parent_map, cache)
3834
for tree in self.source.revision_trees(revision_ids):
3835
# Find a inventory delta for this revision.
3836
# Find text entries that need to be copied, too.
3837
current_revision_id = tree.get_revision_id()
3838
parent_ids = parent_map.get(current_revision_id, ())
3839
parent_trees = self._get_trees(parent_ids, cache)
3840
possible_trees = list(parent_trees)
3841
if len(possible_trees) == 0:
3842
# There either aren't any parents, or the parents are ghosts,
3843
# so just use the last converted tree.
3844
possible_trees.append((basis_id, cache[basis_id]))
3845
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3847
revision = self.source.get_revision(current_revision_id)
3848
pending_deltas.append((basis_id, delta,
3849
current_revision_id, revision.parent_ids))
3850
if self._converting_to_rich_root:
3851
self._revision_id_to_root_id[current_revision_id] = \
3853
# Determine which texts are in present in this revision but not in
3854
# any of the available parents.
3855
texts_possibly_new_in_tree = set()
3856
for old_path, new_path, file_id, entry in delta:
3857
if new_path is None:
3858
# This file_id isn't present in the new rev
3862
if not self.target.supports_rich_root():
3863
# The target doesn't support rich root, so we don't
3866
if self._converting_to_rich_root:
3867
# This can't be copied normally, we have to insert
3869
root_keys_to_create.add((file_id, entry.revision))
3872
texts_possibly_new_in_tree.add((file_id, entry.revision))
3873
for basis_id, basis_tree in possible_trees:
3874
basis_inv = basis_tree.inventory
3875
for file_key in list(texts_possibly_new_in_tree):
3876
file_id, file_revision = file_key
3878
entry = basis_inv[file_id]
3879
except errors.NoSuchId:
3881
if entry.revision == file_revision:
3882
texts_possibly_new_in_tree.remove(file_key)
3883
text_keys.update(texts_possibly_new_in_tree)
3884
pending_revisions.append(revision)
3885
cache[current_revision_id] = tree
3886
basis_id = current_revision_id
3888
from_texts = self.source.texts
3889
to_texts = self.target.texts
3890
if root_keys_to_create:
3891
from bzrlib.fetch import _new_root_data_stream
3892
root_stream = _new_root_data_stream(
3893
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3895
to_texts.insert_record_stream(root_stream)
3896
to_texts.insert_record_stream(from_texts.get_record_stream(
3897
text_keys, self.target._format._fetch_order,
3898
not self.target._format._fetch_uses_deltas))
3899
# insert inventory deltas
3900
for delta in pending_deltas:
3901
self.target.add_inventory_by_delta(*delta)
3902
if self.target._fallback_repositories:
3903
# Make sure this stacked repository has all the parent inventories
3904
# for the new revisions that we are about to insert. We do this
3905
# before adding the revisions so that no revision is added until
3906
# all the inventories it may depend on are added.
3907
# Note that this is overzealous, as we may have fetched these in an
3910
revision_ids = set()
3911
for revision in pending_revisions:
3912
revision_ids.add(revision.revision_id)
3913
parent_ids.update(revision.parent_ids)
3914
parent_ids.difference_update(revision_ids)
3915
parent_ids.discard(_mod_revision.NULL_REVISION)
3916
parent_map = self.source.get_parent_map(parent_ids)
3917
# we iterate over parent_map and not parent_ids because we don't
3918
# want to try copying any revision which is a ghost
3919
for parent_tree in self.source.revision_trees(parent_map):
3920
current_revision_id = parent_tree.get_revision_id()
3921
parents_parents = parent_map[current_revision_id]
3922
possible_trees = self._get_trees(parents_parents, cache)
3923
if len(possible_trees) == 0:
3924
# There either aren't any parents, or the parents are
3925
# ghosts, so just use the last converted tree.
3926
possible_trees.append((basis_id, cache[basis_id]))
3927
basis_id, delta = self._get_delta_for_revision(parent_tree,
3928
parents_parents, possible_trees)
3929
self.target.add_inventory_by_delta(
3930
basis_id, delta, current_revision_id, parents_parents)
3931
# insert signatures and revisions
3932
for revision in pending_revisions:
3934
signature = self.source.get_signature_text(
3935
revision.revision_id)
3936
self.target.add_signature_text(revision.revision_id,
3938
except errors.NoSuchRevision:
3940
self.target.add_revision(revision.revision_id, revision)
3943
def _fetch_all_revisions(self, revision_ids, pb):
3944
"""Fetch everything for the list of revisions.
3946
:param revision_ids: The list of revisions to fetch. Must be in
3948
:param pb: A ProgressTask
3951
basis_id, basis_tree = self._get_basis(revision_ids[0])
3953
cache = lru_cache.LRUCache(100)
3954
cache[basis_id] = basis_tree
3955
del basis_tree # We don't want to hang on to it here
3957
for offset in range(0, len(revision_ids), batch_size):
3958
self.target.start_write_group()
3960
pb.update('Transferring revisions', offset,
3962
batch = revision_ids[offset:offset+batch_size]
3963
basis_id = self._fetch_batch(batch, basis_id, cache)
3965
self.target.abort_write_group()
3968
hint = self.target.commit_write_group()
3971
if hints and self.target._format.pack_compresses:
3972
self.target.pack(hint=hints)
3973
pb.update('Transferring revisions', len(revision_ids),
3977
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3979
"""See InterRepository.fetch()."""
3980
if fetch_spec is not None:
3981
raise AssertionError("Not implemented yet...")
3982
if (not self.source.supports_rich_root()
3983
and self.target.supports_rich_root()):
3984
self._converting_to_rich_root = True
3985
self._revision_id_to_root_id = {}
3987
self._converting_to_rich_root = False
3988
revision_ids = self.target.search_missing_revision_ids(self.source,
3989
revision_id, find_ghosts=find_ghosts).get_keys()
3990
if not revision_ids:
3992
revision_ids = tsort.topo_sort(
3993
self.source.get_graph().get_parent_map(revision_ids))
3994
if not revision_ids:
3996
# Walk though all revisions; get inventory deltas, copy referenced
3997
# texts that delta references, insert the delta, revision and
4000
my_pb = ui.ui_factory.nested_progress_bar()
4003
symbol_versioning.warn(
4004
symbol_versioning.deprecated_in((1, 14, 0))
4005
% "pb parameter to fetch()")
4008
self._fetch_all_revisions(revision_ids, pb)
4010
if my_pb is not None:
4012
return len(revision_ids), 0
4014
def _get_basis(self, first_revision_id):
4015
"""Get a revision and tree which exists in the target.
4017
This assumes that first_revision_id is selected for transmission
4018
because all other ancestors are already present. If we can't find an
4019
ancestor we fall back to NULL_REVISION since we know that is safe.
4021
:return: (basis_id, basis_tree)
4023
first_rev = self.source.get_revision(first_revision_id)
4025
basis_id = first_rev.parent_ids[0]
4026
# only valid as a basis if the target has it
4027
self.target.get_revision(basis_id)
4028
# Try to get a basis tree - if its a ghost it will hit the
4029
# NoSuchRevision case.
4030
basis_tree = self.source.revision_tree(basis_id)
4031
except (IndexError, errors.NoSuchRevision):
4032
basis_id = _mod_revision.NULL_REVISION
4033
basis_tree = self.source.revision_tree(basis_id)
4034
return basis_id, basis_tree
4037
InterRepository.register_optimiser(InterDifferingSerializer)
1603
4038
InterRepository.register_optimiser(InterSameDataRepository)
1604
4039
InterRepository.register_optimiser(InterWeaveRepo)
1605
4040
InterRepository.register_optimiser(InterKnitRepo)
1606
InterRepository.register_optimiser(InterModel1and2)
1607
InterRepository.register_optimiser(InterKnit1and2)
1610
class RepositoryTestProviderAdapter(object):
1611
"""A tool to generate a suite testing multiple repository formats at once.
1613
This is done by copying the test once for each transport and injecting
1614
the transport_server, transport_readonly_server, and bzrdir_format and
1615
repository_format classes into each copy. Each copy is also given a new id()
1616
to make it easy to identify.
1619
def __init__(self, transport_server, transport_readonly_server, formats):
1620
self._transport_server = transport_server
1621
self._transport_readonly_server = transport_readonly_server
1622
self._formats = formats
1624
def adapt(self, test):
1625
result = unittest.TestSuite()
1626
for repository_format, bzrdir_format in self._formats:
1627
from copy import deepcopy
1628
new_test = deepcopy(test)
1629
new_test.transport_server = self._transport_server
1630
new_test.transport_readonly_server = self._transport_readonly_server
1631
new_test.bzrdir_format = bzrdir_format
1632
new_test.repository_format = repository_format
1633
def make_new_test_id():
1634
new_id = "%s(%s)" % (new_test.id(), repository_format.__class__.__name__)
1635
return lambda: new_id
1636
new_test.id = make_new_test_id()
1637
result.addTest(new_test)
1641
class InterRepositoryTestProviderAdapter(object):
1642
"""A tool to generate a suite testing multiple inter repository formats.
1644
This is done by copying the test once for each interrepo provider and injecting
1645
the transport_server, transport_readonly_server, repository_format and
1646
repository_to_format classes into each copy.
1647
Each copy is also given a new id() to make it easy to identify.
1650
def __init__(self, transport_server, transport_readonly_server, formats):
1651
self._transport_server = transport_server
1652
self._transport_readonly_server = transport_readonly_server
1653
self._formats = formats
1655
def adapt(self, test):
1656
result = unittest.TestSuite()
1657
for interrepo_class, repository_format, repository_format_to in self._formats:
1658
from copy import deepcopy
1659
new_test = deepcopy(test)
1660
new_test.transport_server = self._transport_server
1661
new_test.transport_readonly_server = self._transport_readonly_server
1662
new_test.interrepo_class = interrepo_class
1663
new_test.repository_format = repository_format
1664
new_test.repository_format_to = repository_format_to
1665
def make_new_test_id():
1666
new_id = "%s(%s)" % (new_test.id(), interrepo_class.__name__)
1667
return lambda: new_id
1668
new_test.id = make_new_test_id()
1669
result.addTest(new_test)
1673
def default_test_list():
1674
"""Generate the default list of interrepo permutations to test."""
1675
from bzrlib.repofmt import knitrepo, weaverepo
1677
# test the default InterRepository between format 6 and the current
1679
# XXX: robertc 20060220 reinstate this when there are two supported
1680
# formats which do not have an optimal code path between them.
1681
#result.append((InterRepository,
1682
# RepositoryFormat6(),
1683
# RepositoryFormatKnit1()))
1684
for optimiser_class in InterRepository._optimisers:
1685
format_to_test = optimiser_class._get_repo_format_to_test()
1686
if format_to_test is not None:
1687
result.append((optimiser_class,
1688
format_to_test, format_to_test))
1689
# if there are specific combinations we want to use, we can add them
1691
result.append((InterModel1and2,
1692
weaverepo.RepositoryFormat5(),
1693
knitrepo.RepositoryFormatKnit2()))
1694
result.append((InterKnit1and2,
1695
knitrepo.RepositoryFormatKnit1(),
1696
knitrepo.RepositoryFormatKnit2()))
1700
4043
class CopyConverter(object):
1701
4044
"""A repository conversion tool which just performs a copy of the content.
1703
4046
This is slow but quite reliable.
1749
4092
self.pb.update(message, self.count, self.total)
1752
class CommitBuilder(object):
1753
"""Provides an interface to build up a commit.
1755
This allows describing a tree to be committed without needing to
1756
know the internals of the format of the repository.
1759
record_root_entry = False
1760
def __init__(self, repository, parents, config, timestamp=None,
1761
timezone=None, committer=None, revprops=None,
1763
"""Initiate a CommitBuilder.
1765
:param repository: Repository to commit to.
1766
:param parents: Revision ids of the parents of the new revision.
1767
:param config: Configuration to use.
1768
:param timestamp: Optional timestamp recorded for commit.
1769
:param timezone: Optional timezone for timestamp.
1770
:param committer: Optional committer to set for commit.
1771
:param revprops: Optional dictionary of revision properties.
1772
:param revision_id: Optional revision id.
1774
self._config = config
1776
if committer is None:
1777
self._committer = self._config.username()
1779
assert isinstance(committer, basestring), type(committer)
1780
self._committer = committer
1782
self.new_inventory = Inventory(None)
1783
self._new_revision_id = revision_id
1784
self.parents = parents
1785
self.repository = repository
1788
if revprops is not None:
1789
self._revprops.update(revprops)
1791
if timestamp is None:
1792
timestamp = time.time()
1793
# Restrict resolution to 1ms
1794
self._timestamp = round(timestamp, 3)
1796
if timezone is None:
1797
self._timezone = osutils.local_time_offset()
1799
self._timezone = int(timezone)
1801
self._generate_revision_if_needed()
1803
def commit(self, message):
1804
"""Make the actual commit.
1806
:return: The revision id of the recorded revision.
1808
rev = _mod_revision.Revision(
1809
timestamp=self._timestamp,
1810
timezone=self._timezone,
1811
committer=self._committer,
1813
inventory_sha1=self.inv_sha1,
1814
revision_id=self._new_revision_id,
1815
properties=self._revprops)
1816
rev.parent_ids = self.parents
1817
self.repository.add_revision(self._new_revision_id, rev,
1818
self.new_inventory, self._config)
1819
return self._new_revision_id
1821
def revision_tree(self):
1822
"""Return the tree that was just committed.
1824
After calling commit() this can be called to get a RevisionTree
1825
representing the newly committed tree. This is preferred to
1826
calling Repository.revision_tree() because that may require
1827
deserializing the inventory, while we already have a copy in
1830
return RevisionTree(self.repository, self.new_inventory,
1831
self._new_revision_id)
1833
def finish_inventory(self):
1834
"""Tell the builder that the inventory is finished."""
1835
if self.new_inventory.root is None:
1836
symbol_versioning.warn('Root entry should be supplied to'
1837
' record_entry_contents, as of bzr 0.10.',
1838
DeprecationWarning, stacklevel=2)
1839
self.new_inventory.add(InventoryDirectory(ROOT_ID, '', None))
1840
self.new_inventory.revision_id = self._new_revision_id
1841
self.inv_sha1 = self.repository.add_inventory(
1842
self._new_revision_id,
1847
def _gen_revision_id(self):
1848
"""Return new revision-id."""
1849
return generate_ids.gen_revision_id(self._config.username(),
1852
def _generate_revision_if_needed(self):
1853
"""Create a revision id if None was supplied.
1855
If the repository can not support user-specified revision ids
1856
they should override this function and raise CannotSetRevisionId
1857
if _new_revision_id is not None.
1859
:raises: CannotSetRevisionId
1861
if self._new_revision_id is None:
1862
self._new_revision_id = self._gen_revision_id()
1864
def record_entry_contents(self, ie, parent_invs, path, tree):
1865
"""Record the content of ie from tree into the commit if needed.
1867
Side effect: sets ie.revision when unchanged
1869
:param ie: An inventory entry present in the commit.
1870
:param parent_invs: The inventories of the parent revisions of the
1872
:param path: The path the entry is at in the tree.
1873
:param tree: The tree which contains this entry and should be used to
1876
if self.new_inventory.root is None and ie.parent_id is not None:
1877
symbol_versioning.warn('Root entry should be supplied to'
1878
' record_entry_contents, as of bzr 0.10.',
1879
DeprecationWarning, stacklevel=2)
1880
self.record_entry_contents(tree.inventory.root.copy(), parent_invs,
1882
self.new_inventory.add(ie)
1884
# ie.revision is always None if the InventoryEntry is considered
1885
# for committing. ie.snapshot will record the correct revision
1886
# which may be the sole parent if it is untouched.
1887
if ie.revision is not None:
1890
# In this revision format, root entries have no knit or weave
1891
if ie is self.new_inventory.root:
1892
# When serializing out to disk and back in
1893
# root.revision is always _new_revision_id
1894
ie.revision = self._new_revision_id
1896
previous_entries = ie.find_previous_heads(
1898
self.repository.weave_store,
1899
self.repository.get_transaction())
1900
# we are creating a new revision for ie in the history store
1902
ie.snapshot(self._new_revision_id, path, previous_entries, tree, self)
1904
def modified_directory(self, file_id, file_parents):
1905
"""Record the presence of a symbolic link.
1907
:param file_id: The file_id of the link to record.
1908
:param file_parents: The per-file parent revision ids.
1910
self._add_text_to_weave(file_id, [], file_parents.keys())
1912
def modified_file_text(self, file_id, file_parents,
1913
get_content_byte_lines, text_sha1=None,
1915
"""Record the text of file file_id
1917
:param file_id: The file_id of the file to record the text of.
1918
:param file_parents: The per-file parent revision ids.
1919
:param get_content_byte_lines: A callable which will return the byte
1921
:param text_sha1: Optional SHA1 of the file contents.
1922
:param text_size: Optional size of the file contents.
1924
# mutter('storing text of file {%s} in revision {%s} into %r',
1925
# file_id, self._new_revision_id, self.repository.weave_store)
1926
# special case to avoid diffing on renames or
1928
if (len(file_parents) == 1
1929
and text_sha1 == file_parents.values()[0].text_sha1
1930
and text_size == file_parents.values()[0].text_size):
1931
previous_ie = file_parents.values()[0]
1932
versionedfile = self.repository.weave_store.get_weave(file_id,
1933
self.repository.get_transaction())
1934
versionedfile.clone_text(self._new_revision_id,
1935
previous_ie.revision, file_parents.keys())
1936
return text_sha1, text_size
1938
new_lines = get_content_byte_lines()
1939
# TODO: Rather than invoking sha_strings here, _add_text_to_weave
1940
# should return the SHA1 and size
1941
self._add_text_to_weave(file_id, new_lines, file_parents.keys())
1942
return osutils.sha_strings(new_lines), \
1943
sum(map(len, new_lines))
1945
def modified_link(self, file_id, file_parents, link_target):
1946
"""Record the presence of a symbolic link.
1948
:param file_id: The file_id of the link to record.
1949
:param file_parents: The per-file parent revision ids.
1950
:param link_target: Target location of this link.
1952
self._add_text_to_weave(file_id, [], file_parents.keys())
1954
def _add_text_to_weave(self, file_id, new_lines, parents):
1955
versionedfile = self.repository.weave_store.get_weave_or_empty(
1956
file_id, self.repository.get_transaction())
1957
versionedfile.add_lines(self._new_revision_id, parents, new_lines)
1958
versionedfile.clear_cache()
1961
class _CommitBuilder(CommitBuilder):
1962
"""Temporary class so old CommitBuilders are detected properly
1964
Note: CommitBuilder works whether or not root entry is recorded.
1967
record_root_entry = True
1970
class RootCommitBuilder(CommitBuilder):
1971
"""This commitbuilder actually records the root id"""
1973
record_root_entry = True
1975
def record_entry_contents(self, ie, parent_invs, path, tree):
1976
"""Record the content of ie from tree into the commit if needed.
1978
Side effect: sets ie.revision when unchanged
1980
:param ie: An inventory entry present in the commit.
1981
:param parent_invs: The inventories of the parent revisions of the
1983
:param path: The path the entry is at in the tree.
1984
:param tree: The tree which contains this entry and should be used to
1987
assert self.new_inventory.root is not None or ie.parent_id is None
1988
self.new_inventory.add(ie)
1990
# ie.revision is always None if the InventoryEntry is considered
1991
# for committing. ie.snapshot will record the correct revision
1992
# which may be the sole parent if it is untouched.
1993
if ie.revision is not None:
1996
previous_entries = ie.find_previous_heads(
1998
self.repository.weave_store,
1999
self.repository.get_transaction())
2000
# we are creating a new revision for ie in the history store
2002
ie.snapshot(self._new_revision_id, path, previous_entries, tree, self)
2005
4095
_unescape_map = {
2024
4120
if _unescape_re is None:
2025
4121
_unescape_re = re.compile('\&([^;]*);')
2026
4122
return _unescape_re.sub(_unescaper, data)
4125
class _VersionedFileChecker(object):
4127
def __init__(self, repository, text_key_references=None, ancestors=None):
4128
self.repository = repository
4129
self.text_index = self.repository._generate_text_key_index(
4130
text_key_references=text_key_references, ancestors=ancestors)
4132
def calculate_file_version_parents(self, text_key):
4133
"""Calculate the correct parents for a file version according to
4136
parent_keys = self.text_index[text_key]
4137
if parent_keys == [_mod_revision.NULL_REVISION]:
4139
return tuple(parent_keys)
4141
def check_file_version_parents(self, texts, progress_bar=None):
4142
"""Check the parents stored in a versioned file are correct.
4144
It also detects file versions that are not referenced by their
4145
corresponding revision's inventory.
4147
:returns: A tuple of (wrong_parents, dangling_file_versions).
4148
wrong_parents is a dict mapping {revision_id: (stored_parents,
4149
correct_parents)} for each revision_id where the stored parents
4150
are not correct. dangling_file_versions is a set of (file_id,
4151
revision_id) tuples for versions that are present in this versioned
4152
file, but not used by the corresponding inventory.
4154
local_progress = None
4155
if progress_bar is None:
4156
local_progress = ui.ui_factory.nested_progress_bar()
4157
progress_bar = local_progress
4159
return self._check_file_version_parents(texts, progress_bar)
4162
local_progress.finished()
4164
def _check_file_version_parents(self, texts, progress_bar):
4165
"""See check_file_version_parents."""
4167
self.file_ids = set([file_id for file_id, _ in
4168
self.text_index.iterkeys()])
4169
# text keys is now grouped by file_id
4170
n_versions = len(self.text_index)
4171
progress_bar.update('loading text store', 0, n_versions)
4172
parent_map = self.repository.texts.get_parent_map(self.text_index)
4173
# On unlistable transports this could well be empty/error...
4174
text_keys = self.repository.texts.keys()
4175
unused_keys = frozenset(text_keys) - set(self.text_index)
4176
for num, key in enumerate(self.text_index.iterkeys()):
4177
progress_bar.update('checking text graph', num, n_versions)
4178
correct_parents = self.calculate_file_version_parents(key)
4180
knit_parents = parent_map[key]
4181
except errors.RevisionNotPresent:
4184
if correct_parents != knit_parents:
4185
wrong_parents[key] = (knit_parents, correct_parents)
4186
return wrong_parents, unused_keys
4189
def _old_get_graph(repository, revision_id):
4190
"""DO NOT USE. That is all. I'm serious."""
4191
graph = repository.get_graph()
4192
revision_graph = dict(((key, value) for key, value in
4193
graph.iter_ancestry([revision_id]) if value is not None))
4194
return _strip_NULL_ghosts(revision_graph)
4197
def _strip_NULL_ghosts(revision_graph):
4198
"""Also don't use this. more compatibility code for unmigrated clients."""
4199
# Filter ghosts, and null:
4200
if _mod_revision.NULL_REVISION in revision_graph:
4201
del revision_graph[_mod_revision.NULL_REVISION]
4202
for key, parents in revision_graph.items():
4203
revision_graph[key] = tuple(parent for parent in parents if parent
4205
return revision_graph
4208
class StreamSink(object):
4209
"""An object that can insert a stream into a repository.
4211
This interface handles the complexity of reserialising inventories and
4212
revisions from different formats, and allows unidirectional insertion into
4213
stacked repositories without looking for the missing basis parents
4217
def __init__(self, target_repo):
4218
self.target_repo = target_repo
4220
def insert_stream(self, stream, src_format, resume_tokens):
4221
"""Insert a stream's content into the target repository.
4223
:param src_format: a bzr repository format.
4225
:return: a list of resume tokens and an iterable of keys additional
4226
items required before the insertion can be completed.
4228
self.target_repo.lock_write()
4231
self.target_repo.resume_write_group(resume_tokens)
4234
self.target_repo.start_write_group()
4237
# locked_insert_stream performs a commit|suspend.
4238
return self._locked_insert_stream(stream, src_format, is_resume)
4240
self.target_repo.abort_write_group(suppress_errors=True)
4243
self.target_repo.unlock()
4245
def _locked_insert_stream(self, stream, src_format, is_resume):
4246
to_serializer = self.target_repo._format._serializer
4247
src_serializer = src_format._serializer
4249
if to_serializer == src_serializer:
4250
# If serializers match and the target is a pack repository, set the
4251
# write cache size on the new pack. This avoids poor performance
4252
# on transports where append is unbuffered (such as
4253
# RemoteTransport). This is safe to do because nothing should read
4254
# back from the target repository while a stream with matching
4255
# serialization is being inserted.
4256
# The exception is that a delta record from the source that should
4257
# be a fulltext may need to be expanded by the target (see
4258
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4259
# explicitly flush any buffered writes first in that rare case.
4261
new_pack = self.target_repo._pack_collection._new_pack
4262
except AttributeError:
4263
# Not a pack repository
4266
new_pack.set_write_cache_size(1024*1024)
4267
for substream_type, substream in stream:
4268
if 'stream' in debug.debug_flags:
4269
mutter('inserting substream: %s', substream_type)
4270
if substream_type == 'texts':
4271
self.target_repo.texts.insert_record_stream(substream)
4272
elif substream_type == 'inventories':
4273
if src_serializer == to_serializer:
4274
self.target_repo.inventories.insert_record_stream(
4277
self._extract_and_insert_inventories(
4278
substream, src_serializer)
4279
elif substream_type == 'inventory-deltas':
4280
self._extract_and_insert_inventory_deltas(
4281
substream, src_serializer)
4282
elif substream_type == 'chk_bytes':
4283
# XXX: This doesn't support conversions, as it assumes the
4284
# conversion was done in the fetch code.
4285
self.target_repo.chk_bytes.insert_record_stream(substream)
4286
elif substream_type == 'revisions':
4287
# This may fallback to extract-and-insert more often than
4288
# required if the serializers are different only in terms of
4290
if src_serializer == to_serializer:
4291
self.target_repo.revisions.insert_record_stream(
4294
self._extract_and_insert_revisions(substream,
4296
elif substream_type == 'signatures':
4297
self.target_repo.signatures.insert_record_stream(substream)
4299
raise AssertionError('kaboom! %s' % (substream_type,))
4300
# Done inserting data, and the missing_keys calculations will try to
4301
# read back from the inserted data, so flush the writes to the new pack
4302
# (if this is pack format).
4303
if new_pack is not None:
4304
new_pack._write_data('', flush=True)
4305
# Find all the new revisions (including ones from resume_tokens)
4306
missing_keys = self.target_repo.get_missing_parent_inventories(
4307
check_for_missing_texts=is_resume)
4309
for prefix, versioned_file in (
4310
('texts', self.target_repo.texts),
4311
('inventories', self.target_repo.inventories),
4312
('revisions', self.target_repo.revisions),
4313
('signatures', self.target_repo.signatures),
4314
('chk_bytes', self.target_repo.chk_bytes),
4316
if versioned_file is None:
4318
missing_keys.update((prefix,) + key for key in
4319
versioned_file.get_missing_compression_parent_keys())
4320
except NotImplementedError:
4321
# cannot even attempt suspending, and missing would have failed
4322
# during stream insertion.
4323
missing_keys = set()
4326
# suspend the write group and tell the caller what we is
4327
# missing. We know we can suspend or else we would not have
4328
# entered this code path. (All repositories that can handle
4329
# missing keys can handle suspending a write group).
4330
write_group_tokens = self.target_repo.suspend_write_group()
4331
return write_group_tokens, missing_keys
4332
hint = self.target_repo.commit_write_group()
4333
if (to_serializer != src_serializer and
4334
self.target_repo._format.pack_compresses):
4335
self.target_repo.pack(hint=hint)
4338
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4339
target_rich_root = self.target_repo._format.rich_root_data
4340
target_tree_refs = self.target_repo._format.supports_tree_reference
4341
for record in substream:
4342
# Insert the delta directly
4343
inventory_delta_bytes = record.get_bytes_as('fulltext')
4344
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4346
parse_result = deserialiser.parse_text_bytes(
4347
inventory_delta_bytes)
4348
except inventory_delta.IncompatibleInventoryDelta, err:
4349
trace.mutter("Incompatible delta: %s", err.msg)
4350
raise errors.IncompatibleRevision(self.target_repo._format)
4351
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4352
revision_id = new_id
4353
parents = [key[0] for key in record.parents]
4354
self.target_repo.add_inventory_by_delta(
4355
basis_id, inv_delta, revision_id, parents)
4357
def _extract_and_insert_inventories(self, substream, serializer,
4359
"""Generate a new inventory versionedfile in target, converting data.
4361
The inventory is retrieved from the source, (deserializing it), and
4362
stored in the target (reserializing it in a different format).
4364
target_rich_root = self.target_repo._format.rich_root_data
4365
target_tree_refs = self.target_repo._format.supports_tree_reference
4366
for record in substream:
4367
# It's not a delta, so it must be a fulltext in the source
4368
# serializer's format.
4369
bytes = record.get_bytes_as('fulltext')
4370
revision_id = record.key[0]
4371
inv = serializer.read_inventory_from_string(bytes, revision_id)
4372
parents = [key[0] for key in record.parents]
4373
self.target_repo.add_inventory(revision_id, inv, parents)
4374
# No need to keep holding this full inv in memory when the rest of
4375
# the substream is likely to be all deltas.
4378
def _extract_and_insert_revisions(self, substream, serializer):
4379
for record in substream:
4380
bytes = record.get_bytes_as('fulltext')
4381
revision_id = record.key[0]
4382
rev = serializer.read_revision_from_string(bytes)
4383
if rev.revision_id != revision_id:
4384
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4385
self.target_repo.add_revision(revision_id, rev)
4388
if self.target_repo._format._fetch_reconcile:
4389
self.target_repo.reconcile()
4392
class StreamSource(object):
4393
"""A source of a stream for fetching between repositories."""
4395
def __init__(self, from_repository, to_format):
4396
"""Create a StreamSource streaming from from_repository."""
4397
self.from_repository = from_repository
4398
self.to_format = to_format
4400
def delta_on_metadata(self):
4401
"""Return True if delta's are permitted on metadata streams.
4403
That is on revisions and signatures.
4405
src_serializer = self.from_repository._format._serializer
4406
target_serializer = self.to_format._serializer
4407
return (self.to_format._fetch_uses_deltas and
4408
src_serializer == target_serializer)
4410
def _fetch_revision_texts(self, revs):
4411
# fetch signatures first and then the revision texts
4412
# may need to be a InterRevisionStore call here.
4413
from_sf = self.from_repository.signatures
4414
# A missing signature is just skipped.
4415
keys = [(rev_id,) for rev_id in revs]
4416
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4418
self.to_format._fetch_order,
4419
not self.to_format._fetch_uses_deltas))
4420
# If a revision has a delta, this is actually expanded inside the
4421
# insert_record_stream code now, which is an alternate fix for
4423
from_rf = self.from_repository.revisions
4424
revisions = from_rf.get_record_stream(
4426
self.to_format._fetch_order,
4427
not self.delta_on_metadata())
4428
return [('signatures', signatures), ('revisions', revisions)]
4430
def _generate_root_texts(self, revs):
4431
"""This will be called by get_stream between fetching weave texts and
4432
fetching the inventory weave.
4434
if self._rich_root_upgrade():
4436
return bzrlib.fetch.Inter1and2Helper(
4437
self.from_repository).generate_root_texts(revs)
4441
def get_stream(self, search):
4443
revs = search.get_keys()
4444
graph = self.from_repository.get_graph()
4445
revs = tsort.topo_sort(graph.get_parent_map(revs))
4446
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4448
for knit_kind, file_id, revisions in data_to_fetch:
4449
if knit_kind != phase:
4451
# Make a new progress bar for this phase
4452
if knit_kind == "file":
4453
# Accumulate file texts
4454
text_keys.extend([(file_id, revision) for revision in
4456
elif knit_kind == "inventory":
4457
# Now copy the file texts.
4458
from_texts = self.from_repository.texts
4459
yield ('texts', from_texts.get_record_stream(
4460
text_keys, self.to_format._fetch_order,
4461
not self.to_format._fetch_uses_deltas))
4462
# Cause an error if a text occurs after we have done the
4465
# Before we process the inventory we generate the root
4466
# texts (if necessary) so that the inventories references
4468
for _ in self._generate_root_texts(revs):
4470
# we fetch only the referenced inventories because we do not
4471
# know for unselected inventories whether all their required
4472
# texts are present in the other repository - it could be
4474
for info in self._get_inventory_stream(revs):
4476
elif knit_kind == "signatures":
4477
# Nothing to do here; this will be taken care of when
4478
# _fetch_revision_texts happens.
4480
elif knit_kind == "revisions":
4481
for record in self._fetch_revision_texts(revs):
4484
raise AssertionError("Unknown knit kind %r" % knit_kind)
4486
def get_stream_for_missing_keys(self, missing_keys):
4487
# missing keys can only occur when we are byte copying and not
4488
# translating (because translation means we don't send
4489
# unreconstructable deltas ever).
4491
keys['texts'] = set()
4492
keys['revisions'] = set()
4493
keys['inventories'] = set()
4494
keys['chk_bytes'] = set()
4495
keys['signatures'] = set()
4496
for key in missing_keys:
4497
keys[key[0]].add(key[1:])
4498
if len(keys['revisions']):
4499
# If we allowed copying revisions at this point, we could end up
4500
# copying a revision without copying its required texts: a
4501
# violation of the requirements for repository integrity.
4502
raise AssertionError(
4503
'cannot copy revisions to fill in missing deltas %s' % (
4504
keys['revisions'],))
4505
for substream_kind, keys in keys.iteritems():
4506
vf = getattr(self.from_repository, substream_kind)
4507
if vf is None and keys:
4508
raise AssertionError(
4509
"cannot fill in keys for a versioned file we don't"
4510
" have: %s needs %s" % (substream_kind, keys))
4512
# No need to stream something we don't have
4514
if substream_kind == 'inventories':
4515
# Some missing keys are genuinely ghosts, filter those out.
4516
present = self.from_repository.inventories.get_parent_map(keys)
4517
revs = [key[0] for key in present]
4518
# Get the inventory stream more-or-less as we do for the
4519
# original stream; there's no reason to assume that records
4520
# direct from the source will be suitable for the sink. (Think
4521
# e.g. 2a -> 1.9-rich-root).
4522
for info in self._get_inventory_stream(revs, missing=True):
4526
# Ask for full texts always so that we don't need more round trips
4527
# after this stream.
4528
# Some of the missing keys are genuinely ghosts, so filter absent
4529
# records. The Sink is responsible for doing another check to
4530
# ensure that ghosts don't introduce missing data for future
4532
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4533
self.to_format._fetch_order, True))
4534
yield substream_kind, stream
4536
def inventory_fetch_order(self):
4537
if self._rich_root_upgrade():
4538
return 'topological'
4540
return self.to_format._fetch_order
4542
def _rich_root_upgrade(self):
4543
return (not self.from_repository._format.rich_root_data and
4544
self.to_format.rich_root_data)
4546
def _get_inventory_stream(self, revision_ids, missing=False):
4547
from_format = self.from_repository._format
4548
if (from_format.supports_chks and self.to_format.supports_chks and
4549
from_format.network_name() == self.to_format.network_name()):
4550
raise AssertionError(
4551
"this case should be handled by GroupCHKStreamSource")
4552
elif 'forceinvdeltas' in debug.debug_flags:
4553
return self._get_convertable_inventory_stream(revision_ids,
4554
delta_versus_null=missing)
4555
elif from_format.network_name() == self.to_format.network_name():
4557
return self._get_simple_inventory_stream(revision_ids,
4559
elif (not from_format.supports_chks and not self.to_format.supports_chks
4560
and from_format._serializer == self.to_format._serializer):
4561
# Essentially the same format.
4562
return self._get_simple_inventory_stream(revision_ids,
4565
# Any time we switch serializations, we want to use an
4566
# inventory-delta based approach.
4567
return self._get_convertable_inventory_stream(revision_ids,
4568
delta_versus_null=missing)
4570
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4571
# NB: This currently reopens the inventory weave in source;
4572
# using a single stream interface instead would avoid this.
4573
from_weave = self.from_repository.inventories
4575
delta_closure = True
4577
delta_closure = not self.delta_on_metadata()
4578
yield ('inventories', from_weave.get_record_stream(
4579
[(rev_id,) for rev_id in revision_ids],
4580
self.inventory_fetch_order(), delta_closure))
4582
def _get_convertable_inventory_stream(self, revision_ids,
4583
delta_versus_null=False):
4584
# The source is using CHKs, but the target either doesn't or it has a
4585
# different serializer. The StreamSink code expects to be able to
4586
# convert on the target, so we need to put bytes-on-the-wire that can
4587
# be converted. That means inventory deltas (if the remote is <1.19,
4588
# RemoteStreamSink will fallback to VFS to insert the deltas).
4589
yield ('inventory-deltas',
4590
self._stream_invs_as_deltas(revision_ids,
4591
delta_versus_null=delta_versus_null))
4593
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4594
"""Return a stream of inventory-deltas for the given rev ids.
4596
:param revision_ids: The list of inventories to transmit
4597
:param delta_versus_null: Don't try to find a minimal delta for this
4598
entry, instead compute the delta versus the NULL_REVISION. This
4599
effectively streams a complete inventory. Used for stuff like
4600
filling in missing parents, etc.
4602
from_repo = self.from_repository
4603
revision_keys = [(rev_id,) for rev_id in revision_ids]
4604
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4605
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4607
inventories = self.from_repository.iter_inventories(
4608
revision_ids, 'topological')
4609
format = from_repo._format
4610
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4611
inventory_cache = lru_cache.LRUCache(50)
4612
null_inventory = from_repo.revision_tree(
4613
_mod_revision.NULL_REVISION).inventory
4614
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4615
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4616
# repo back into a non-rich-root repo ought to be allowed)
4617
serializer = inventory_delta.InventoryDeltaSerializer(
4618
versioned_root=format.rich_root_data,
4619
tree_references=format.supports_tree_reference)
4620
for inv in inventories:
4621
key = (inv.revision_id,)
4622
parent_keys = parent_map.get(key, ())
4624
if not delta_versus_null and parent_keys:
4625
# The caller did not ask for complete inventories and we have
4626
# some parents that we can delta against. Make a delta against
4627
# each parent so that we can find the smallest.
4628
parent_ids = [parent_key[0] for parent_key in parent_keys]
4629
for parent_id in parent_ids:
4630
if parent_id not in invs_sent_so_far:
4631
# We don't know that the remote side has this basis, so
4634
if parent_id == _mod_revision.NULL_REVISION:
4635
parent_inv = null_inventory
4637
parent_inv = inventory_cache.get(parent_id, None)
4638
if parent_inv is None:
4639
parent_inv = from_repo.get_inventory(parent_id)
4640
candidate_delta = inv._make_delta(parent_inv)
4641
if (delta is None or
4642
len(delta) > len(candidate_delta)):
4643
delta = candidate_delta
4644
basis_id = parent_id
4646
# Either none of the parents ended up being suitable, or we
4647
# were asked to delta against NULL
4648
basis_id = _mod_revision.NULL_REVISION
4649
delta = inv._make_delta(null_inventory)
4650
invs_sent_so_far.add(inv.revision_id)
4651
inventory_cache[inv.revision_id] = inv
4652
delta_serialized = ''.join(
4653
serializer.delta_to_lines(basis_id, key[-1], delta))
4654
yield versionedfile.FulltextContentFactory(
4655
key, parent_keys, None, delta_serialized)
4658
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4659
stop_revision=None):
4660
"""Extend the partial history to include a given index
4662
If a stop_index is supplied, stop when that index has been reached.
4663
If a stop_revision is supplied, stop when that revision is
4664
encountered. Otherwise, stop when the beginning of history is
4667
:param stop_index: The index which should be present. When it is
4668
present, history extension will stop.
4669
:param stop_revision: The revision id which should be present. When
4670
it is encountered, history extension will stop.
4672
start_revision = partial_history_cache[-1]
4673
iterator = repo.iter_reverse_revision_history(start_revision)
4675
#skip the last revision in the list
4678
if (stop_index is not None and
4679
len(partial_history_cache) > stop_index):
4681
if partial_history_cache[-1] == stop_revision:
4683
revision_id = iterator.next()
4684
partial_history_cache.append(revision_id)
4685
except StopIteration: