202
335
commit to be valid, deletes against the basis MUST be recorded via
203
336
builder.record_delete().
205
raise NotImplementedError(self.will_record_deletes)
207
def record_iter_changes(self, tree, basis_revision_id, iter_changes):
338
self._recording_deletes = True
340
basis_id = self.parents[0]
342
basis_id = _mod_revision.NULL_REVISION
343
self.basis_delta_revision = basis_id
345
def record_entry_contents(self, ie, parent_invs, path, tree,
347
"""Record the content of ie from tree into the commit if needed.
349
Side effect: sets ie.revision when unchanged
351
:param ie: An inventory entry present in the commit.
352
:param parent_invs: The inventories of the parent revisions of the
354
:param path: The path the entry is at in the tree.
355
:param tree: The tree which contains this entry and should be used to
357
:param content_summary: Summary data from the tree about the paths
358
content - stat, length, exec, sha/link target. This is only
359
accessed when the entry has a revision of None - that is when it is
360
a candidate to commit.
361
:return: A tuple (change_delta, version_recorded, fs_hash).
362
change_delta is an inventory_delta change for this entry against
363
the basis tree of the commit, or None if no change occured against
365
version_recorded is True if a new version of the entry has been
366
recorded. For instance, committing a merge where a file was only
367
changed on the other side will return (delta, False).
368
fs_hash is either None, or the hash details for the path (currently
369
a tuple of the contents sha1 and the statvalue returned by
370
tree.get_file_with_stat()).
372
if self.new_inventory.root is None:
373
if ie.parent_id is not None:
374
raise errors.RootMissing()
375
self._check_root(ie, parent_invs, tree)
376
if ie.revision is None:
377
kind = content_summary[0]
379
# ie is carried over from a prior commit
381
# XXX: repository specific check for nested tree support goes here - if
382
# the repo doesn't want nested trees we skip it ?
383
if (kind == 'tree-reference' and
384
not self.repository._format.supports_tree_reference):
385
# mismatch between commit builder logic and repository:
386
# this needs the entry creation pushed down into the builder.
387
raise NotImplementedError('Missing repository subtree support.')
388
self.new_inventory.add(ie)
390
# TODO: slow, take it out of the inner loop.
392
basis_inv = parent_invs[0]
394
basis_inv = Inventory(root_id=None)
396
# ie.revision is always None if the InventoryEntry is considered
397
# for committing. We may record the previous parents revision if the
398
# content is actually unchanged against a sole head.
399
if ie.revision is not None:
400
if not self._versioned_root and path == '':
401
# repositories that do not version the root set the root's
402
# revision to the new commit even when no change occurs (more
403
# specifically, they do not record a revision on the root; and
404
# the rev id is assigned to the root during deserialisation -
405
# this masks when a change may have occurred against the basis.
406
# To match this we always issue a delta, because the revision
407
# of the root will always be changing.
408
if ie.file_id in basis_inv:
409
delta = (basis_inv.id2path(ie.file_id), path,
413
delta = (None, path, ie.file_id, ie)
414
self._basis_delta.append(delta)
415
return delta, False, None
417
# we don't need to commit this, because the caller already
418
# determined that an existing revision of this file is
419
# appropriate. If its not being considered for committing then
420
# it and all its parents to the root must be unaltered so
421
# no-change against the basis.
422
if ie.revision == self._new_revision_id:
423
raise AssertionError("Impossible situation, a skipped "
424
"inventory entry (%r) claims to be modified in this "
425
"commit (%r).", (ie, self._new_revision_id))
426
return None, False, None
427
# XXX: Friction: parent_candidates should return a list not a dict
428
# so that we don't have to walk the inventories again.
429
parent_candiate_entries = ie.parent_candidates(parent_invs)
430
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
432
for inv in parent_invs:
433
if ie.file_id in inv:
434
old_rev = inv[ie.file_id].revision
435
if old_rev in head_set:
436
heads.append(inv[ie.file_id].revision)
437
head_set.remove(inv[ie.file_id].revision)
440
# now we check to see if we need to write a new record to the
442
# We write a new entry unless there is one head to the ancestors, and
443
# the kind-derived content is unchanged.
445
# Cheapest check first: no ancestors, or more the one head in the
446
# ancestors, we write a new node.
450
# There is a single head, look it up for comparison
451
parent_entry = parent_candiate_entries[heads[0]]
452
# if the non-content specific data has changed, we'll be writing a
454
if (parent_entry.parent_id != ie.parent_id or
455
parent_entry.name != ie.name):
457
# now we need to do content specific checks:
459
# if the kind changed the content obviously has
460
if kind != parent_entry.kind:
462
# Stat cache fingerprint feedback for the caller - None as we usually
463
# don't generate one.
466
if content_summary[2] is None:
467
raise ValueError("Files must not have executable = None")
469
# We can't trust a check of the file length because of content
471
if (# if the exec bit has changed we have to store:
472
parent_entry.executable != content_summary[2]):
474
elif parent_entry.text_sha1 == content_summary[3]:
475
# all meta and content is unchanged (using a hash cache
476
# hit to check the sha)
477
ie.revision = parent_entry.revision
478
ie.text_size = parent_entry.text_size
479
ie.text_sha1 = parent_entry.text_sha1
480
ie.executable = parent_entry.executable
481
return self._get_delta(ie, basis_inv, path), False, None
483
# Either there is only a hash change(no hash cache entry,
484
# or same size content change), or there is no change on
486
# Provide the parent's hash to the store layer, so that the
487
# content is unchanged we will not store a new node.
488
nostore_sha = parent_entry.text_sha1
490
# We want to record a new node regardless of the presence or
491
# absence of a content change in the file.
493
ie.executable = content_summary[2]
494
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
496
text = file_obj.read()
500
ie.text_sha1, ie.text_size = self._add_text_to_weave(
501
ie.file_id, text, heads, nostore_sha)
502
# Let the caller know we generated a stat fingerprint.
503
fingerprint = (ie.text_sha1, stat_value)
504
except errors.ExistingContent:
505
# Turns out that the file content was unchanged, and we were
506
# only going to store a new node if it was changed. Carry over
508
ie.revision = parent_entry.revision
509
ie.text_size = parent_entry.text_size
510
ie.text_sha1 = parent_entry.text_sha1
511
ie.executable = parent_entry.executable
512
return self._get_delta(ie, basis_inv, path), False, None
513
elif kind == 'directory':
515
# all data is meta here, nothing specific to directory, so
517
ie.revision = parent_entry.revision
518
return self._get_delta(ie, basis_inv, path), False, None
519
self._add_text_to_weave(ie.file_id, '', heads, None)
520
elif kind == 'symlink':
521
current_link_target = content_summary[3]
523
# symlink target is not generic metadata, check if it has
525
if current_link_target != parent_entry.symlink_target:
528
# unchanged, carry over.
529
ie.revision = parent_entry.revision
530
ie.symlink_target = parent_entry.symlink_target
531
return self._get_delta(ie, basis_inv, path), False, None
532
ie.symlink_target = current_link_target
533
self._add_text_to_weave(ie.file_id, '', heads, None)
534
elif kind == 'tree-reference':
536
if content_summary[3] != parent_entry.reference_revision:
539
# unchanged, carry over.
540
ie.reference_revision = parent_entry.reference_revision
541
ie.revision = parent_entry.revision
542
return self._get_delta(ie, basis_inv, path), False, None
543
ie.reference_revision = content_summary[3]
544
if ie.reference_revision is None:
545
raise AssertionError("invalid content_summary for nested tree: %r"
546
% (content_summary,))
547
self._add_text_to_weave(ie.file_id, '', heads, None)
549
raise NotImplementedError('unknown kind')
550
ie.revision = self._new_revision_id
551
self._any_changes = True
552
return self._get_delta(ie, basis_inv, path), True, fingerprint
554
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
555
_entry_factory=entry_factory):
208
556
"""Record a new tree via iter_changes.
210
558
:param tree: The tree to obtain text contents from for changed objects.
215
563
to basis_revision_id. The iterator must not include any items with
216
564
a current kind of None - missing items must be either filtered out
217
565
or errored-on beefore record_iter_changes sees the item.
566
:param _entry_factory: Private method to bind entry_factory locally for
218
568
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
219
569
tree._observed_sha1.
221
raise NotImplementedError(self.record_iter_changes)
224
class RepositoryWriteLockResult(LogicalLockResult):
225
"""The result of write locking a repository.
227
:ivar repository_token: The token obtained from the underlying lock, or
229
:ivar unlock: A callable which will unlock the lock.
232
def __init__(self, unlock, repository_token):
233
LogicalLockResult.__init__(self, unlock)
234
self.repository_token = repository_token
237
return "RepositoryWriteLockResult(%s, %s)" % (self.repository_token,
571
# Create an inventory delta based on deltas between all the parents and
572
# deltas between all the parent inventories. We use inventory delta's
573
# between the inventory objects because iter_changes masks
574
# last-changed-field only changes.
576
# file_id -> change map, change is fileid, paths, changed, versioneds,
577
# parents, names, kinds, executables
579
# {file_id -> revision_id -> inventory entry, for entries in parent
580
# trees that are not parents[0]
584
revtrees = list(self.repository.revision_trees(self.parents))
585
except errors.NoSuchRevision:
586
# one or more ghosts, slow path.
588
for revision_id in self.parents:
590
revtrees.append(self.repository.revision_tree(revision_id))
591
except errors.NoSuchRevision:
593
basis_revision_id = _mod_revision.NULL_REVISION
595
revtrees.append(self.repository.revision_tree(
596
_mod_revision.NULL_REVISION))
597
# The basis inventory from a repository
599
basis_inv = revtrees[0].inventory
601
basis_inv = self.repository.revision_tree(
602
_mod_revision.NULL_REVISION).inventory
603
if len(self.parents) > 0:
604
if basis_revision_id != self.parents[0] and not ghost_basis:
606
"arbitrary basis parents not yet supported with merges")
607
for revtree in revtrees[1:]:
608
for change in revtree.inventory._make_delta(basis_inv):
609
if change[1] is None:
610
# Not present in this parent.
612
if change[2] not in merged_ids:
613
if change[0] is not None:
614
basis_entry = basis_inv[change[2]]
615
merged_ids[change[2]] = [
617
basis_entry.revision,
620
parent_entries[change[2]] = {
622
basis_entry.revision:basis_entry,
624
change[3].revision:change[3],
627
merged_ids[change[2]] = [change[3].revision]
628
parent_entries[change[2]] = {change[3].revision:change[3]}
630
merged_ids[change[2]].append(change[3].revision)
631
parent_entries[change[2]][change[3].revision] = change[3]
634
# Setup the changes from the tree:
635
# changes maps file_id -> (change, [parent revision_ids])
637
for change in iter_changes:
638
# This probably looks up in basis_inv way to much.
639
if change[1][0] is not None:
640
head_candidate = [basis_inv[change[0]].revision]
643
changes[change[0]] = change, merged_ids.get(change[0],
645
unchanged_merged = set(merged_ids) - set(changes)
646
# Extend the changes dict with synthetic changes to record merges of
648
for file_id in unchanged_merged:
649
# Record a merged version of these items that did not change vs the
650
# basis. This can be either identical parallel changes, or a revert
651
# of a specific file after a merge. The recorded content will be
652
# that of the current tree (which is the same as the basis), but
653
# the per-file graph will reflect a merge.
654
# NB:XXX: We are reconstructing path information we had, this
655
# should be preserved instead.
656
# inv delta change: (file_id, (path_in_source, path_in_target),
657
# changed_content, versioned, parent, name, kind,
660
basis_entry = basis_inv[file_id]
661
except errors.NoSuchId:
662
# a change from basis->some_parents but file_id isn't in basis
663
# so was new in the merge, which means it must have changed
664
# from basis -> current, and as it hasn't the add was reverted
665
# by the user. So we discard this change.
669
(basis_inv.id2path(file_id), tree.id2path(file_id)),
671
(basis_entry.parent_id, basis_entry.parent_id),
672
(basis_entry.name, basis_entry.name),
673
(basis_entry.kind, basis_entry.kind),
674
(basis_entry.executable, basis_entry.executable))
675
changes[file_id] = (change, merged_ids[file_id])
676
# changes contains tuples with the change and a set of inventory
677
# candidates for the file.
679
# old_path, new_path, file_id, new_inventory_entry
680
seen_root = False # Is the root in the basis delta?
681
inv_delta = self._basis_delta
682
modified_rev = self._new_revision_id
683
for change, head_candidates in changes.values():
684
if change[3][1]: # versioned in target.
685
# Several things may be happening here:
686
# We may have a fork in the per-file graph
687
# - record a change with the content from tree
688
# We may have a change against < all trees
689
# - carry over the tree that hasn't changed
690
# We may have a change against all trees
691
# - record the change with the content from tree
694
entry = _entry_factory[kind](file_id, change[5][1],
696
head_set = self._heads(change[0], set(head_candidates))
699
for head_candidate in head_candidates:
700
if head_candidate in head_set:
701
heads.append(head_candidate)
702
head_set.remove(head_candidate)
705
# Could be a carry-over situation:
706
parent_entry_revs = parent_entries.get(file_id, None)
707
if parent_entry_revs:
708
parent_entry = parent_entry_revs.get(heads[0], None)
711
if parent_entry is None:
712
# The parent iter_changes was called against is the one
713
# that is the per-file head, so any change is relevant
714
# iter_changes is valid.
715
carry_over_possible = False
717
# could be a carry over situation
718
# A change against the basis may just indicate a merge,
719
# we need to check the content against the source of the
720
# merge to determine if it was changed after the merge
722
if (parent_entry.kind != entry.kind or
723
parent_entry.parent_id != entry.parent_id or
724
parent_entry.name != entry.name):
725
# Metadata common to all entries has changed
726
# against per-file parent
727
carry_over_possible = False
729
carry_over_possible = True
730
# per-type checks for changes against the parent_entry
733
# Cannot be a carry-over situation
734
carry_over_possible = False
735
# Populate the entry in the delta
737
# XXX: There is still a small race here: If someone reverts the content of a file
738
# after iter_changes examines and decides it has changed,
739
# we will unconditionally record a new version even if some
740
# other process reverts it while commit is running (with
741
# the revert happening after iter_changes did it's
744
entry.executable = True
746
entry.executable = False
747
if (carry_over_possible and
748
parent_entry.executable == entry.executable):
749
# Check the file length, content hash after reading
751
nostore_sha = parent_entry.text_sha1
754
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
756
text = file_obj.read()
760
entry.text_sha1, entry.text_size = self._add_text_to_weave(
761
file_id, text, heads, nostore_sha)
762
yield file_id, change[1][1], (entry.text_sha1, stat_value)
763
except errors.ExistingContent:
764
# No content change against a carry_over parent
765
# Perhaps this should also yield a fs hash update?
767
entry.text_size = parent_entry.text_size
768
entry.text_sha1 = parent_entry.text_sha1
769
elif kind == 'symlink':
771
entry.symlink_target = tree.get_symlink_target(file_id)
772
if (carry_over_possible and
773
parent_entry.symlink_target == entry.symlink_target):
776
self._add_text_to_weave(change[0], '', heads, None)
777
elif kind == 'directory':
778
if carry_over_possible:
781
# Nothing to set on the entry.
782
# XXX: split into the Root and nonRoot versions.
783
if change[1][1] != '' or self.repository.supports_rich_root():
784
self._add_text_to_weave(change[0], '', heads, None)
785
elif kind == 'tree-reference':
786
if not self.repository._format.supports_tree_reference:
787
# This isn't quite sane as an error, but we shouldn't
788
# ever see this code path in practice: tree's don't
789
# permit references when the repo doesn't support tree
791
raise errors.UnsupportedOperation(tree.add_reference,
793
reference_revision = tree.get_reference_revision(change[0])
794
entry.reference_revision = reference_revision
795
if (carry_over_possible and
796
parent_entry.reference_revision == reference_revision):
799
self._add_text_to_weave(change[0], '', heads, None)
801
raise AssertionError('unknown kind %r' % kind)
803
entry.revision = modified_rev
805
entry.revision = parent_entry.revision
808
new_path = change[1][1]
809
inv_delta.append((change[1][0], new_path, change[0], entry))
812
self.new_inventory = None
814
# This should perhaps be guarded by a check that the basis we
815
# commit against is the basis for the commit and if not do a delta
817
self._any_changes = True
819
# housekeeping root entry changes do not affect no-change commits.
820
self._require_root_change(tree)
821
self.basis_delta_revision = basis_revision_id
823
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
824
parent_keys = tuple([(file_id, parent) for parent in parents])
825
return self.repository.texts._add_text(
826
(file_id, self._new_revision_id), parent_keys, new_text,
827
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
830
class RootCommitBuilder(CommitBuilder):
831
"""This commitbuilder actually records the root id"""
833
# the root entry gets versioned properly by this builder.
834
_versioned_root = True
836
def _check_root(self, ie, parent_invs, tree):
837
"""Helper for record_entry_contents.
839
:param ie: An entry being added.
840
:param parent_invs: The inventories of the parent revisions of the
842
:param tree: The tree that is being committed.
845
def _require_root_change(self, tree):
846
"""Enforce an appropriate root object change.
848
This is called once when record_iter_changes is called, if and only if
849
the root was not in the delta calculated by record_iter_changes.
851
:param tree: The tree which is being committed.
853
# versioned roots do not change unless the tree found a change.
241
856
######################################################################
245
class Repository(_RelockDebugMixin, controldir.ControlComponent):
860
class Repository(_RelockDebugMixin):
246
861
"""Repository holding history for one or more branches.
248
863
The repository holds and retrieves historical information including
249
864
revisions and file history. It's normally accessed only by the Branch,
250
865
which views a particular line of development through that history.
252
See VersionedFileRepository in bzrlib.vf_repository for the
253
base class for most Bazaar repositories.
867
The Repository builds on top of some byte storage facilies (the revisions,
868
signatures, inventories, texts and chk_bytes attributes) and a Transport,
869
which respectively provide byte storage and a means to access the (possibly
872
The byte storage facilities are addressed via tuples, which we refer to
873
as 'keys' throughout the code base. Revision_keys, inventory_keys and
874
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
875
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
876
byte string made up of a hash identifier and a hash value.
877
We use this interface because it allows low friction with the underlying
878
code that implements disk indices, network encoding and other parts of
881
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
882
the serialised revisions for the repository. This can be used to obtain
883
revision graph information or to access raw serialised revisions.
884
The result of trying to insert data into the repository via this store
885
is undefined: it should be considered read-only except for implementors
887
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
888
the serialised signatures for the repository. This can be used to
889
obtain access to raw serialised signatures. The result of trying to
890
insert data into the repository via this store is undefined: it should
891
be considered read-only except for implementors of repositories.
892
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
893
the serialised inventories for the repository. This can be used to
894
obtain unserialised inventories. The result of trying to insert data
895
into the repository via this store is undefined: it should be
896
considered read-only except for implementors of repositories.
897
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
898
texts of files and directories for the repository. This can be used to
899
obtain file texts or file graphs. Note that Repository.iter_file_bytes
900
is usually a better interface for accessing file texts.
901
The result of trying to insert data into the repository via this store
902
is undefined: it should be considered read-only except for implementors
904
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
905
any data the repository chooses to store or have indexed by its hash.
906
The result of trying to insert data into the repository via this store
907
is undefined: it should be considered read-only except for implementors
909
:ivar _transport: Transport for file access to repository, typically
910
pointing to .bzr/repository.
913
# What class to use for a CommitBuilder. Often its simpler to change this
914
# in a Repository class subclass rather than to override
915
# get_commit_builder.
916
_commit_builder_class = CommitBuilder
917
# The search regex used by xml based repositories to determine what things
918
# where changed in a single commit.
919
_file_ids_altered_regex = lazy_regex.lazy_compile(
920
r'file_id="(?P<file_id>[^"]+)"'
921
r'.* revision="(?P<revision_id>[^"]+)"'
256
924
def abort_write_group(self, suppress_errors=False):
257
925
"""Commit the contents accrued within the current write group.
313
994
return InterRepository._assert_same_model(self, repository)
996
def add_inventory(self, revision_id, inv, parents):
997
"""Add the inventory inv to the repository as revision_id.
999
:param parents: The revision ids of the parents that revision_id
1000
is known to have and are in the repository already.
1002
:returns: The validator(which is a sha1 digest, though what is sha'd is
1003
repository format specific) of the serialized inventory.
1005
if not self.is_in_write_group():
1006
raise AssertionError("%r not in write group" % (self,))
1007
_mod_revision.check_not_reserved_id(revision_id)
1008
if not (inv.revision_id is None or inv.revision_id == revision_id):
1009
raise AssertionError(
1010
"Mismatch between inventory revision"
1011
" id and insertion revid (%r, %r)"
1012
% (inv.revision_id, revision_id))
1013
if inv.root is None:
1014
raise AssertionError()
1015
return self._add_inventory_checked(revision_id, inv, parents)
1017
def _add_inventory_checked(self, revision_id, inv, parents):
1018
"""Add inv to the repository after checking the inputs.
1020
This function can be overridden to allow different inventory styles.
1022
:seealso: add_inventory, for the contract.
1024
inv_lines = self._serialise_inventory_to_lines(inv)
1025
return self._inventory_add_lines(revision_id, parents,
1026
inv_lines, check_content=False)
1028
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1029
parents, basis_inv=None, propagate_caches=False):
1030
"""Add a new inventory expressed as a delta against another revision.
1032
See the inventory developers documentation for the theory behind
1035
:param basis_revision_id: The inventory id the delta was created
1036
against. (This does not have to be a direct parent.)
1037
:param delta: The inventory delta (see Inventory.apply_delta for
1039
:param new_revision_id: The revision id that the inventory is being
1041
:param parents: The revision ids of the parents that revision_id is
1042
known to have and are in the repository already. These are supplied
1043
for repositories that depend on the inventory graph for revision
1044
graph access, as well as for those that pun ancestry with delta
1046
:param basis_inv: The basis inventory if it is already known,
1048
:param propagate_caches: If True, the caches for this inventory are
1049
copied to and updated for the result if possible.
1051
:returns: (validator, new_inv)
1052
The validator(which is a sha1 digest, though what is sha'd is
1053
repository format specific) of the serialized inventory, and the
1054
resulting inventory.
1056
if not self.is_in_write_group():
1057
raise AssertionError("%r not in write group" % (self,))
1058
_mod_revision.check_not_reserved_id(new_revision_id)
1059
basis_tree = self.revision_tree(basis_revision_id)
1060
basis_tree.lock_read()
1062
# Note that this mutates the inventory of basis_tree, which not all
1063
# inventory implementations may support: A better idiom would be to
1064
# return a new inventory, but as there is no revision tree cache in
1065
# repository this is safe for now - RBC 20081013
1066
if basis_inv is None:
1067
basis_inv = basis_tree.inventory
1068
basis_inv.apply_delta(delta)
1069
basis_inv.revision_id = new_revision_id
1070
return (self.add_inventory(new_revision_id, basis_inv, parents),
1075
def _inventory_add_lines(self, revision_id, parents, lines,
1076
check_content=True):
1077
"""Store lines in inv_vf and return the sha1 of the inventory."""
1078
parents = [(parent,) for parent in parents]
1079
result = self.inventories.add_lines((revision_id,), parents, lines,
1080
check_content=check_content)[0]
1081
self.inventories._access.flush()
1084
def add_revision(self, revision_id, rev, inv=None, config=None):
1085
"""Add rev to the revision store as revision_id.
1087
:param revision_id: the revision id to use.
1088
:param rev: The revision object.
1089
:param inv: The inventory for the revision. if None, it will be looked
1090
up in the inventory storer
1091
:param config: If None no digital signature will be created.
1092
If supplied its signature_needed method will be used
1093
to determine if a signature should be made.
1095
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1097
_mod_revision.check_not_reserved_id(revision_id)
1098
if config is not None and config.signature_needed():
1100
inv = self.get_inventory(revision_id)
1101
plaintext = Testament(rev, inv).as_short_text()
1102
self.store_revision_signature(
1103
gpg.GPGStrategy(config), plaintext, revision_id)
1104
# check inventory present
1105
if not self.inventories.get_parent_map([(revision_id,)]):
1107
raise errors.WeaveRevisionNotPresent(revision_id,
1110
# yes, this is not suitable for adding with ghosts.
1111
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1114
key = (revision_id,)
1115
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1116
self._add_revision(rev)
1118
def _add_revision(self, revision):
1119
text = self._serializer.write_revision_to_string(revision)
1120
key = (revision.revision_id,)
1121
parents = tuple((parent,) for parent in revision.parent_ids)
1122
self.revisions.add_lines(key, parents, osutils.split_lines(text))
315
1124
def all_revision_ids(self):
316
1125
"""Returns a list of all the revision ids in the repository.
352
1161
# The old API returned a list, should this actually be a set?
353
1162
return parent_map.keys()
1164
def _check_inventories(self, checker):
1165
"""Check the inventories found from the revision scan.
1167
This is responsible for verifying the sha1 of inventories and
1168
creating a pending_keys set that covers data referenced by inventories.
1170
bar = ui.ui_factory.nested_progress_bar()
1172
self._do_check_inventories(checker, bar)
1176
def _do_check_inventories(self, checker, bar):
1177
"""Helper for _check_inventories."""
1179
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1180
kinds = ['chk_bytes', 'texts']
1181
count = len(checker.pending_keys)
1182
bar.update("inventories", 0, 2)
1183
current_keys = checker.pending_keys
1184
checker.pending_keys = {}
1185
# Accumulate current checks.
1186
for key in current_keys:
1187
if key[0] != 'inventories' and key[0] not in kinds:
1188
checker._report_items.append('unknown key type %r' % (key,))
1189
keys[key[0]].add(key[1:])
1190
if keys['inventories']:
1191
# NB: output order *should* be roughly sorted - topo or
1192
# inverse topo depending on repository - either way decent
1193
# to just delta against. However, pre-CHK formats didn't
1194
# try to optimise inventory layout on disk. As such the
1195
# pre-CHK code path does not use inventory deltas.
1197
for record in self.inventories.check(keys=keys['inventories']):
1198
if record.storage_kind == 'absent':
1199
checker._report_items.append(
1200
'Missing inventory {%s}' % (record.key,))
1202
last_object = self._check_record('inventories', record,
1203
checker, last_object,
1204
current_keys[('inventories',) + record.key])
1205
del keys['inventories']
1208
bar.update("texts", 1)
1209
while (checker.pending_keys or keys['chk_bytes']
1211
# Something to check.
1212
current_keys = checker.pending_keys
1213
checker.pending_keys = {}
1214
# Accumulate current checks.
1215
for key in current_keys:
1216
if key[0] not in kinds:
1217
checker._report_items.append('unknown key type %r' % (key,))
1218
keys[key[0]].add(key[1:])
1219
# Check the outermost kind only - inventories || chk_bytes || texts
1223
for record in getattr(self, kind).check(keys=keys[kind]):
1224
if record.storage_kind == 'absent':
1225
checker._report_items.append(
1226
'Missing %s {%s}' % (kind, record.key,))
1228
last_object = self._check_record(kind, record,
1229
checker, last_object, current_keys[(kind,) + record.key])
1233
def _check_record(self, kind, record, checker, last_object, item_data):
1234
"""Check a single text from this repository."""
1235
if kind == 'inventories':
1236
rev_id = record.key[0]
1237
inv = self.deserialise_inventory(rev_id,
1238
record.get_bytes_as('fulltext'))
1239
if last_object is not None:
1240
delta = inv._make_delta(last_object)
1241
for old_path, path, file_id, ie in delta:
1244
ie.check(checker, rev_id, inv)
1246
for path, ie in inv.iter_entries():
1247
ie.check(checker, rev_id, inv)
1248
if self._format.fast_deltas:
1250
elif kind == 'chk_bytes':
1251
# No code written to check chk_bytes for this repo format.
1252
checker._report_items.append(
1253
'unsupported key type chk_bytes for %s' % (record.key,))
1254
elif kind == 'texts':
1255
self._check_text(record, checker, item_data)
1257
checker._report_items.append(
1258
'unknown key type %s for %s' % (kind, record.key))
1260
def _check_text(self, record, checker, item_data):
1261
"""Check a single text."""
1262
# Check it is extractable.
1263
# TODO: check length.
1264
if record.storage_kind == 'chunked':
1265
chunks = record.get_bytes_as(record.storage_kind)
1266
sha1 = osutils.sha_strings(chunks)
1267
length = sum(map(len, chunks))
1269
content = record.get_bytes_as('fulltext')
1270
sha1 = osutils.sha_string(content)
1271
length = len(content)
1272
if item_data and sha1 != item_data[1]:
1273
checker._report_items.append(
1274
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1275
(record.key, sha1, item_data[1], item_data[2]))
356
1278
def create(a_bzrdir):
357
1279
"""Construct the current default format repository in a_bzrdir."""
921
1971
signature = gpg_strategy.sign(plaintext)
922
1972
self.add_signature_text(revision_id, signature)
924
1975
def add_signature_text(self, revision_id, signature):
925
"""Store a signature text for a revision.
927
:param revision_id: Revision id of the revision
928
:param signature: Signature text.
930
raise NotImplementedError(self.add_signature_text)
1976
self.signatures.add_lines((revision_id,), (),
1977
osutils.split_lines(signature))
1979
def find_text_key_references(self):
1980
"""Find the text key references within the repository.
1982
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1983
to whether they were referred to by the inventory of the
1984
revision_id that they contain. The inventory texts from all present
1985
revision ids are assessed to generate this report.
1987
revision_keys = self.revisions.keys()
1988
w = self.inventories
1989
pb = ui.ui_factory.nested_progress_bar()
1991
return self._find_text_key_references_from_xml_inventory_lines(
1992
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1996
def _find_text_key_references_from_xml_inventory_lines(self,
1998
"""Core routine for extracting references to texts from inventories.
2000
This performs the translation of xml lines to revision ids.
2002
:param line_iterator: An iterator of lines, origin_version_id
2003
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2004
to whether they were referred to by the inventory of the
2005
revision_id that they contain. Note that if that revision_id was
2006
not part of the line_iterator's output then False will be given -
2007
even though it may actually refer to that key.
2009
if not self._serializer.support_altered_by_hack:
2010
raise AssertionError(
2011
"_find_text_key_references_from_xml_inventory_lines only "
2012
"supported for branches which store inventory as unnested xml"
2013
", not on %r" % self)
2016
# this code needs to read every new line in every inventory for the
2017
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2018
# not present in one of those inventories is unnecessary but not
2019
# harmful because we are filtering by the revision id marker in the
2020
# inventory lines : we only select file ids altered in one of those
2021
# revisions. We don't need to see all lines in the inventory because
2022
# only those added in an inventory in rev X can contain a revision=X
2024
unescape_revid_cache = {}
2025
unescape_fileid_cache = {}
2027
# jam 20061218 In a big fetch, this handles hundreds of thousands
2028
# of lines, so it has had a lot of inlining and optimizing done.
2029
# Sorry that it is a little bit messy.
2030
# Move several functions to be local variables, since this is a long
2032
search = self._file_ids_altered_regex.search
2033
unescape = _unescape_xml
2034
setdefault = result.setdefault
2035
for line, line_key in line_iterator:
2036
match = search(line)
2039
# One call to match.group() returning multiple items is quite a
2040
# bit faster than 2 calls to match.group() each returning 1
2041
file_id, revision_id = match.group('file_id', 'revision_id')
2043
# Inlining the cache lookups helps a lot when you make 170,000
2044
# lines and 350k ids, versus 8.4 unique ids.
2045
# Using a cache helps in 2 ways:
2046
# 1) Avoids unnecessary decoding calls
2047
# 2) Re-uses cached strings, which helps in future set and
2049
# (2) is enough that removing encoding entirely along with
2050
# the cache (so we are using plain strings) results in no
2051
# performance improvement.
2053
revision_id = unescape_revid_cache[revision_id]
2055
unescaped = unescape(revision_id)
2056
unescape_revid_cache[revision_id] = unescaped
2057
revision_id = unescaped
2059
# Note that unconditionally unescaping means that we deserialise
2060
# every fileid, which for general 'pull' is not great, but we don't
2061
# really want to have some many fulltexts that this matters anyway.
2064
file_id = unescape_fileid_cache[file_id]
2066
unescaped = unescape(file_id)
2067
unescape_fileid_cache[file_id] = unescaped
2070
key = (file_id, revision_id)
2071
setdefault(key, False)
2072
if revision_id == line_key[-1]:
2076
def _inventory_xml_lines_for_keys(self, keys):
2077
"""Get a line iterator of the sort needed for findind references.
2079
Not relevant for non-xml inventory repositories.
2081
Ghosts in revision_keys are ignored.
2083
:param revision_keys: The revision keys for the inventories to inspect.
2084
:return: An iterator over (inventory line, revid) for the fulltexts of
2085
all of the xml inventories specified by revision_keys.
2087
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2088
for record in stream:
2089
if record.storage_kind != 'absent':
2090
chunks = record.get_bytes_as('chunked')
2091
revid = record.key[-1]
2092
lines = osutils.chunks_to_lines(chunks)
2096
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2098
"""Helper routine for fileids_altered_by_revision_ids.
2100
This performs the translation of xml lines to revision ids.
2102
:param line_iterator: An iterator of lines, origin_version_id
2103
:param revision_keys: The revision ids to filter for. This should be a
2104
set or other type which supports efficient __contains__ lookups, as
2105
the revision key from each parsed line will be looked up in the
2106
revision_keys filter.
2107
:return: a dictionary mapping altered file-ids to an iterable of
2108
revision_ids. Each altered file-ids has the exact revision_ids that
2109
altered it listed explicitly.
2111
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2112
line_iterator).iterkeys())
2113
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2114
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2115
self._inventory_xml_lines_for_keys(parent_keys)))
2116
new_keys = seen - parent_seen
2118
setdefault = result.setdefault
2119
for key in new_keys:
2120
setdefault(key[0], set()).add(key[-1])
932
2123
def _find_parent_ids_of_revisions(self, revision_ids):
933
2124
"""Find all parent ids that are mentioned in the revision graph.
963
2179
uniquely identify the file version in the caller's context. (Examples:
964
2180
an index number or a TreeTransform trans_id.)
2182
bytes_iterator is an iterable of bytestrings for the file. The
2183
kind of iterable and length of the bytestrings are unspecified, but for
2184
this implementation, it is a list of bytes produced by
2185
VersionedFile.get_record_stream().
966
2187
:param desired_files: a list of (file_id, revision_id, identifier)
969
raise NotImplementedError(self.iter_files_bytes)
2191
for file_id, revision_id, callable_data in desired_files:
2192
text_keys[(file_id, revision_id)] = callable_data
2193
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2194
if record.storage_kind == 'absent':
2195
raise errors.RevisionNotPresent(record.key, self)
2196
yield text_keys[record.key], record.get_bytes_as('chunked')
2198
def _generate_text_key_index(self, text_key_references=None,
2200
"""Generate a new text key index for the repository.
2202
This is an expensive function that will take considerable time to run.
2204
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2205
list of parents, also text keys. When a given key has no parents,
2206
the parents list will be [NULL_REVISION].
2208
# All revisions, to find inventory parents.
2209
if ancestors is None:
2210
graph = self.get_graph()
2211
ancestors = graph.get_parent_map(self.all_revision_ids())
2212
if text_key_references is None:
2213
text_key_references = self.find_text_key_references()
2214
pb = ui.ui_factory.nested_progress_bar()
2216
return self._do_generate_text_key_index(ancestors,
2217
text_key_references, pb)
2221
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2222
"""Helper for _generate_text_key_index to avoid deep nesting."""
2223
revision_order = tsort.topo_sort(ancestors)
2224
invalid_keys = set()
2226
for revision_id in revision_order:
2227
revision_keys[revision_id] = set()
2228
text_count = len(text_key_references)
2229
# a cache of the text keys to allow reuse; costs a dict of all the
2230
# keys, but saves a 2-tuple for every child of a given key.
2232
for text_key, valid in text_key_references.iteritems():
2234
invalid_keys.add(text_key)
2236
revision_keys[text_key[1]].add(text_key)
2237
text_key_cache[text_key] = text_key
2238
del text_key_references
2240
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2241
NULL_REVISION = _mod_revision.NULL_REVISION
2242
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2243
# too small for large or very branchy trees. However, for 55K path
2244
# trees, it would be easy to use too much memory trivially. Ideally we
2245
# could gauge this by looking at available real memory etc, but this is
2246
# always a tricky proposition.
2247
inventory_cache = lru_cache.LRUCache(10)
2248
batch_size = 10 # should be ~150MB on a 55K path tree
2249
batch_count = len(revision_order) / batch_size + 1
2251
pb.update("Calculating text parents", processed_texts, text_count)
2252
for offset in xrange(batch_count):
2253
to_query = revision_order[offset * batch_size:(offset + 1) *
2257
for revision_id in to_query:
2258
parent_ids = ancestors[revision_id]
2259
for text_key in revision_keys[revision_id]:
2260
pb.update("Calculating text parents", processed_texts)
2261
processed_texts += 1
2262
candidate_parents = []
2263
for parent_id in parent_ids:
2264
parent_text_key = (text_key[0], parent_id)
2266
check_parent = parent_text_key not in \
2267
revision_keys[parent_id]
2269
# the parent parent_id is a ghost:
2270
check_parent = False
2271
# truncate the derived graph against this ghost.
2272
parent_text_key = None
2274
# look at the parent commit details inventories to
2275
# determine possible candidates in the per file graph.
2278
inv = inventory_cache[parent_id]
2280
inv = self.revision_tree(parent_id).inventory
2281
inventory_cache[parent_id] = inv
2283
parent_entry = inv[text_key[0]]
2284
except (KeyError, errors.NoSuchId):
2286
if parent_entry is not None:
2288
text_key[0], parent_entry.revision)
2290
parent_text_key = None
2291
if parent_text_key is not None:
2292
candidate_parents.append(
2293
text_key_cache[parent_text_key])
2294
parent_heads = text_graph.heads(candidate_parents)
2295
new_parents = list(parent_heads)
2296
new_parents.sort(key=lambda x:candidate_parents.index(x))
2297
if new_parents == []:
2298
new_parents = [NULL_REVISION]
2299
text_index[text_key] = new_parents
2301
for text_key in invalid_keys:
2302
text_index[text_key] = [NULL_REVISION]
2305
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2306
"""Get an iterable listing the keys of all the data introduced by a set
2309
The keys will be ordered so that the corresponding items can be safely
2310
fetched and inserted in that order.
2312
:returns: An iterable producing tuples of (knit-kind, file-id,
2313
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2314
'revisions'. file-id is None unless knit-kind is 'file'.
2316
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2319
for result in self._find_non_file_keys_to_fetch(revision_ids):
2322
def _find_file_keys_to_fetch(self, revision_ids, pb):
2323
# XXX: it's a bit weird to control the inventory weave caching in this
2324
# generator. Ideally the caching would be done in fetch.py I think. Or
2325
# maybe this generator should explicitly have the contract that it
2326
# should not be iterated until the previously yielded item has been
2328
inv_w = self.inventories
2330
# file ids that changed
2331
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2333
num_file_ids = len(file_ids)
2334
for file_id, altered_versions in file_ids.iteritems():
2336
pb.update("Fetch texts", count, num_file_ids)
2338
yield ("file", file_id, altered_versions)
2340
def _find_non_file_keys_to_fetch(self, revision_ids):
2342
yield ("inventory", None, revision_ids)
2345
# XXX: Note ATM no callers actually pay attention to this return
2346
# instead they just use the list of revision ids and ignore
2347
# missing sigs. Consider removing this work entirely
2348
revisions_with_signatures = set(self.signatures.get_parent_map(
2349
[(r,) for r in revision_ids]))
2350
revisions_with_signatures = set(
2351
[r for (r,) in revisions_with_signatures])
2352
revisions_with_signatures.intersection_update(revision_ids)
2353
yield ("signatures", None, revisions_with_signatures)
2356
yield ("revisions", None, revision_ids)
2359
def get_inventory(self, revision_id):
2360
"""Get Inventory object by revision id."""
2361
return self.iter_inventories([revision_id]).next()
2363
def iter_inventories(self, revision_ids, ordering=None):
2364
"""Get many inventories by revision_ids.
2366
This will buffer some or all of the texts used in constructing the
2367
inventories in memory, but will only parse a single inventory at a
2370
:param revision_ids: The expected revision ids of the inventories.
2371
:param ordering: optional ordering, e.g. 'topological'. If not
2372
specified, the order of revision_ids will be preserved (by
2373
buffering if necessary).
2374
:return: An iterator of inventories.
2376
if ((None in revision_ids)
2377
or (_mod_revision.NULL_REVISION in revision_ids)):
2378
raise ValueError('cannot get null revision inventory')
2379
return self._iter_inventories(revision_ids, ordering)
2381
def _iter_inventories(self, revision_ids, ordering):
2382
"""single-document based inventory iteration."""
2383
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2384
for text, revision_id in inv_xmls:
2385
yield self.deserialise_inventory(revision_id, text)
2387
def _iter_inventory_xmls(self, revision_ids, ordering):
2388
if ordering is None:
2389
order_as_requested = True
2390
ordering = 'unordered'
2392
order_as_requested = False
2393
keys = [(revision_id,) for revision_id in revision_ids]
2396
if order_as_requested:
2397
key_iter = iter(keys)
2398
next_key = key_iter.next()
2399
stream = self.inventories.get_record_stream(keys, ordering, True)
2401
for record in stream:
2402
if record.storage_kind != 'absent':
2403
chunks = record.get_bytes_as('chunked')
2404
if order_as_requested:
2405
text_chunks[record.key] = chunks
2407
yield ''.join(chunks), record.key[-1]
2409
raise errors.NoSuchRevision(self, record.key)
2410
if order_as_requested:
2411
# Yield as many results as we can while preserving order.
2412
while next_key in text_chunks:
2413
chunks = text_chunks.pop(next_key)
2414
yield ''.join(chunks), next_key[-1]
2416
next_key = key_iter.next()
2417
except StopIteration:
2418
# We still want to fully consume the get_record_stream,
2419
# just in case it is not actually finished at this point
2423
def deserialise_inventory(self, revision_id, xml):
2424
"""Transform the xml into an inventory object.
2426
:param revision_id: The expected revision id of the inventory.
2427
:param xml: A serialised inventory.
2429
result = self._serializer.read_inventory_from_string(xml, revision_id,
2430
entry_cache=self._inventory_entry_cache)
2431
if result.revision_id != revision_id:
2432
raise AssertionError('revision id mismatch %s != %s' % (
2433
result.revision_id, revision_id))
2436
def serialise_inventory(self, inv):
2437
return self._serializer.write_inventory_to_string(inv)
2439
def _serialise_inventory_to_lines(self, inv):
2440
return self._serializer.write_inventory_to_lines(inv)
2442
def get_serializer_format(self):
2443
return self._serializer.format_num
2446
def get_inventory_xml(self, revision_id):
2447
"""Get inventory XML as a file object."""
2448
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2450
text, revision_id = texts.next()
2451
except StopIteration:
2452
raise errors.HistoryMissing(self, 'inventory', revision_id)
2456
def get_inventory_sha1(self, revision_id):
2457
"""Return the sha1 hash of the inventory entry
2459
return self.get_revision(revision_id).inventory_sha1
971
2461
def get_rev_id_for_revno(self, revno, known_pair):
972
2462
"""Return the revision id of a revno, given a later (revno, revid)
1289
2801
except UnicodeDecodeError:
1290
2802
raise errors.NonAsciiRevisionId(method, self)
2804
def revision_graph_can_have_wrong_parents(self):
2805
"""Is it possible for this repository to have a revision graph with
2808
If True, then this repository must also implement
2809
_find_inconsistent_revision_parents so that check and reconcile can
2810
check for inconsistencies before proceeding with other checks that may
2811
depend on the revision index being consistent.
2813
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2816
# remove these delegates a while after bzr 0.15
2817
def __make_delegated(name, from_module):
2818
def _deprecated_repository_forwarder():
2819
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2820
% (name, from_module),
2823
m = __import__(from_module, globals(), locals(), [name])
2825
return getattr(m, name)
2826
except AttributeError:
2827
raise AttributeError('module %s has no name %s'
2829
globals()[name] = _deprecated_repository_forwarder
2832
'AllInOneRepository',
2833
'WeaveMetaDirRepository',
2834
'PreSplitOutRepositoryFormat',
2835
'RepositoryFormat4',
2836
'RepositoryFormat5',
2837
'RepositoryFormat6',
2838
'RepositoryFormat7',
2840
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2844
'RepositoryFormatKnit',
2845
'RepositoryFormatKnit1',
2847
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2850
def install_revision(repository, rev, revision_tree):
2851
"""Install all revision data into a repository."""
2852
install_revisions(repository, [(rev, revision_tree, None)])
2855
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2856
"""Install all revision data into a repository.
2858
Accepts an iterable of revision, tree, signature tuples. The signature
2861
repository.start_write_group()
2863
inventory_cache = lru_cache.LRUCache(10)
2864
for n, (revision, revision_tree, signature) in enumerate(iterable):
2865
_install_revision(repository, revision, revision_tree, signature,
2868
pb.update('Transferring revisions', n + 1, num_revisions)
2870
repository.abort_write_group()
2873
repository.commit_write_group()
2876
def _install_revision(repository, rev, revision_tree, signature,
2878
"""Install all revision data into a repository."""
2879
present_parents = []
2881
for p_id in rev.parent_ids:
2882
if repository.has_revision(p_id):
2883
present_parents.append(p_id)
2884
parent_trees[p_id] = repository.revision_tree(p_id)
2886
parent_trees[p_id] = repository.revision_tree(
2887
_mod_revision.NULL_REVISION)
2889
inv = revision_tree.inventory
2890
entries = inv.iter_entries()
2891
# backwards compatibility hack: skip the root id.
2892
if not repository.supports_rich_root():
2893
path, root = entries.next()
2894
if root.revision != rev.revision_id:
2895
raise errors.IncompatibleRevision(repr(repository))
2897
for path, ie in entries:
2898
text_keys[(ie.file_id, ie.revision)] = ie
2899
text_parent_map = repository.texts.get_parent_map(text_keys)
2900
missing_texts = set(text_keys) - set(text_parent_map)
2901
# Add the texts that are not already present
2902
for text_key in missing_texts:
2903
ie = text_keys[text_key]
2905
# FIXME: TODO: The following loop overlaps/duplicates that done by
2906
# commit to determine parents. There is a latent/real bug here where
2907
# the parents inserted are not those commit would do - in particular
2908
# they are not filtered by heads(). RBC, AB
2909
for revision, tree in parent_trees.iteritems():
2910
if ie.file_id not in tree:
2912
parent_id = tree.inventory[ie.file_id].revision
2913
if parent_id in text_parents:
2915
text_parents.append((ie.file_id, parent_id))
2916
lines = revision_tree.get_file(ie.file_id).readlines()
2917
repository.texts.add_lines(text_key, text_parents, lines)
2919
# install the inventory
2920
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2921
# Cache this inventory
2922
inventory_cache[rev.revision_id] = inv
2924
basis_inv = inventory_cache[rev.parent_ids[0]]
2926
repository.add_inventory(rev.revision_id, inv, present_parents)
2928
delta = inv._make_delta(basis_inv)
2929
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2930
rev.revision_id, present_parents)
2932
repository.add_inventory(rev.revision_id, inv, present_parents)
2933
except errors.RevisionAlreadyPresent:
2935
if signature is not None:
2936
repository.add_signature_text(rev.revision_id, signature)
2937
repository.add_revision(rev.revision_id, rev, inv)
1293
2940
class MetaDirRepository(Repository):
1294
2941
"""Repositories in the new meta-dir layout.
1633
3295
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
1634
3296
format_registry.register_lazy(
1635
3297
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
1636
'bzrlib.repofmt.knitpack_repo',
3298
'bzrlib.repofmt.pack_repo',
1637
3299
'RepositoryFormatKnitPack1',
1639
3301
format_registry.register_lazy(
1640
3302
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
1641
'bzrlib.repofmt.knitpack_repo',
3303
'bzrlib.repofmt.pack_repo',
1642
3304
'RepositoryFormatKnitPack3',
1644
3306
format_registry.register_lazy(
1645
3307
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
1646
'bzrlib.repofmt.knitpack_repo',
3308
'bzrlib.repofmt.pack_repo',
1647
3309
'RepositoryFormatKnitPack4',
1649
3311
format_registry.register_lazy(
1650
3312
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
1651
'bzrlib.repofmt.knitpack_repo',
3313
'bzrlib.repofmt.pack_repo',
1652
3314
'RepositoryFormatKnitPack5',
1654
3316
format_registry.register_lazy(
1655
3317
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
1656
'bzrlib.repofmt.knitpack_repo',
3318
'bzrlib.repofmt.pack_repo',
1657
3319
'RepositoryFormatKnitPack5RichRoot',
1659
3321
format_registry.register_lazy(
1660
3322
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
1661
'bzrlib.repofmt.knitpack_repo',
3323
'bzrlib.repofmt.pack_repo',
1662
3324
'RepositoryFormatKnitPack5RichRootBroken',
1664
3326
format_registry.register_lazy(
1665
3327
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
1666
'bzrlib.repofmt.knitpack_repo',
3328
'bzrlib.repofmt.pack_repo',
1667
3329
'RepositoryFormatKnitPack6',
1669
3331
format_registry.register_lazy(
1670
3332
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
1671
'bzrlib.repofmt.knitpack_repo',
3333
'bzrlib.repofmt.pack_repo',
1672
3334
'RepositoryFormatKnitPack6RichRoot',
1674
format_registry.register_lazy(
1675
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
1676
'bzrlib.repofmt.groupcompress_repo',
1677
'RepositoryFormat2a',
1680
3337
# Development formats.
1681
# Check their docstrings to see if/when they are obsolete.
3338
# Obsolete but kept pending a CHK based subtree format.
1682
3339
format_registry.register_lazy(
1683
3340
("Bazaar development format 2 with subtree support "
1684
3341
"(needs bzr.dev from before 1.8)\n"),
1685
'bzrlib.repofmt.knitpack_repo',
3342
'bzrlib.repofmt.pack_repo',
1686
3343
'RepositoryFormatPackDevelopment2Subtree',
1688
format_registry.register_lazy(
1689
'Bazaar development format 8\n',
1690
'bzrlib.repofmt.groupcompress_repo',
1691
'RepositoryFormat2aSubtree',
3346
# 1.14->1.16 go below here
3347
format_registry.register_lazy(
3348
'Bazaar development format - group compression and chk inventory'
3349
' (needs bzr.dev from 1.14)\n',
3350
'bzrlib.repofmt.groupcompress_repo',
3351
'RepositoryFormatCHK1',
3354
format_registry.register_lazy(
3355
'Bazaar development format - chk repository with bencode revision '
3356
'serialization (needs bzr.dev from 1.16)\n',
3357
'bzrlib.repofmt.groupcompress_repo',
3358
'RepositoryFormatCHK2',
3360
format_registry.register_lazy(
3361
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3362
'bzrlib.repofmt.groupcompress_repo',
3363
'RepositoryFormat2a',
1733
3406
:param revision_id: if None all content is copied, if NULL_REVISION no
1734
3407
content is copied.
3408
:param pb: optional progress bar to use for progress reports. If not
3409
provided a default one will be created.
1737
raise NotImplementedError(self.fetch)
3412
from bzrlib.fetch import RepoFetcher
3413
f = RepoFetcher(to_repository=self.target,
3414
from_repository=self.source,
3415
last_revision=revision_id,
3416
fetch_spec=fetch_spec,
3417
pb=pb, find_ghosts=find_ghosts)
3419
def _walk_to_common_revisions(self, revision_ids):
3420
"""Walk out from revision_ids in source to revisions target has.
3422
:param revision_ids: The start point for the search.
3423
:return: A set of revision ids.
3425
target_graph = self.target.get_graph()
3426
revision_ids = frozenset(revision_ids)
3427
missing_revs = set()
3428
source_graph = self.source.get_graph()
3429
# ensure we don't pay silly lookup costs.
3430
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3431
null_set = frozenset([_mod_revision.NULL_REVISION])
3432
searcher_exhausted = False
3436
# Iterate the searcher until we have enough next_revs
3437
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3439
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3440
next_revs.update(next_revs_part)
3441
ghosts.update(ghosts_part)
3442
except StopIteration:
3443
searcher_exhausted = True
3445
# If there are ghosts in the source graph, and the caller asked for
3446
# them, make sure that they are present in the target.
3447
# We don't care about other ghosts as we can't fetch them and
3448
# haven't been asked to.
3449
ghosts_to_check = set(revision_ids.intersection(ghosts))
3450
revs_to_get = set(next_revs).union(ghosts_to_check)
3452
have_revs = set(target_graph.get_parent_map(revs_to_get))
3453
# we always have NULL_REVISION present.
3454
have_revs = have_revs.union(null_set)
3455
# Check if the target is missing any ghosts we need.
3456
ghosts_to_check.difference_update(have_revs)
3458
# One of the caller's revision_ids is a ghost in both the
3459
# source and the target.
3460
raise errors.NoSuchRevision(
3461
self.source, ghosts_to_check.pop())
3462
missing_revs.update(next_revs - have_revs)
3463
# Because we may have walked past the original stop point, make
3464
# sure everything is stopped
3465
stop_revs = searcher.find_seen_ancestors(have_revs)
3466
searcher.stop_searching_any(stop_revs)
3467
if searcher_exhausted:
3469
return searcher.get_result()
1739
3471
@needs_read_lock
1740
def search_missing_revision_ids(self,
1741
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
1742
find_ghosts=True, revision_ids=None, if_present_ids=None,
3472
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1744
3473
"""Return the revision ids that source has that target does not.
1746
3475
:param revision_id: only return revision ids included by this
1748
:param revision_ids: return revision ids included by these
1749
revision_ids. NoSuchRevision will be raised if any of these
1750
revisions are not present.
1751
:param if_present_ids: like revision_ids, but will not cause
1752
NoSuchRevision if any of these are absent, instead they will simply
1753
not be in the result. This is useful for e.g. finding revisions
1754
to fetch for tags, which may reference absent revisions.
1755
3477
:param find_ghosts: If True find missing revisions in deep history
1756
3478
rather than just finding the surface difference.
1757
:param limit: Maximum number of revisions to return, topologically
1759
3479
:return: A bzrlib.graph.SearchResult.
1761
raise NotImplementedError(self.search_missing_revision_ids)
3481
# stop searching at found target revisions.
3482
if not find_ghosts and revision_id is not None:
3483
return self._walk_to_common_revisions([revision_id])
3484
# generic, possibly worst case, slow code path.
3485
target_ids = set(self.target.all_revision_ids())
3486
if revision_id is not None:
3487
source_ids = self.source.get_ancestry(revision_id)
3488
if source_ids[0] is not None:
3489
raise AssertionError()
3492
source_ids = self.source.all_revision_ids()
3493
result_set = set(source_ids).difference(target_ids)
3494
return self.source.revision_ids_to_search_result(result_set)
1764
3497
def _same_model(source, target):
1785
3518
"different serializers")
3521
class InterSameDataRepository(InterRepository):
3522
"""Code for converting between repositories that represent the same data.
3524
Data format and model must match for this to work.
3528
def _get_repo_format_to_test(self):
3529
"""Repository format for testing with.
3531
InterSameData can pull from subtree to subtree and from non-subtree to
3532
non-subtree, so we test this with the richest repository format.
3534
from bzrlib.repofmt import knitrepo
3535
return knitrepo.RepositoryFormatKnit3()
3538
def is_compatible(source, target):
3539
return InterRepository._same_model(source, target)
3542
class InterWeaveRepo(InterSameDataRepository):
3543
"""Optimised code paths between Weave based repositories.
3545
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3546
implemented lazy inter-object optimisation.
3550
def _get_repo_format_to_test(self):
3551
from bzrlib.repofmt import weaverepo
3552
return weaverepo.RepositoryFormat7()
3555
def is_compatible(source, target):
3556
"""Be compatible with known Weave formats.
3558
We don't test for the stores being of specific types because that
3559
could lead to confusing results, and there is no need to be
3562
from bzrlib.repofmt.weaverepo import (
3568
return (isinstance(source._format, (RepositoryFormat5,
3570
RepositoryFormat7)) and
3571
isinstance(target._format, (RepositoryFormat5,
3573
RepositoryFormat7)))
3574
except AttributeError:
3578
def copy_content(self, revision_id=None):
3579
"""See InterRepository.copy_content()."""
3580
# weave specific optimised path:
3582
self.target.set_make_working_trees(self.source.make_working_trees())
3583
except (errors.RepositoryUpgradeRequired, NotImplemented):
3585
# FIXME do not peek!
3586
if self.source._transport.listable():
3587
pb = ui.ui_factory.nested_progress_bar()
3589
self.target.texts.insert_record_stream(
3590
self.source.texts.get_record_stream(
3591
self.source.texts.keys(), 'topological', False))
3592
pb.update('Copying inventory', 0, 1)
3593
self.target.inventories.insert_record_stream(
3594
self.source.inventories.get_record_stream(
3595
self.source.inventories.keys(), 'topological', False))
3596
self.target.signatures.insert_record_stream(
3597
self.source.signatures.get_record_stream(
3598
self.source.signatures.keys(),
3600
self.target.revisions.insert_record_stream(
3601
self.source.revisions.get_record_stream(
3602
self.source.revisions.keys(),
3603
'topological', True))
3607
self.target.fetch(self.source, revision_id=revision_id)
3610
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3611
"""See InterRepository.missing_revision_ids()."""
3612
# we want all revisions to satisfy revision_id in source.
3613
# but we don't want to stat every file here and there.
3614
# we want then, all revisions other needs to satisfy revision_id
3615
# checked, but not those that we have locally.
3616
# so the first thing is to get a subset of the revisions to
3617
# satisfy revision_id in source, and then eliminate those that
3618
# we do already have.
3619
# this is slow on high latency connection to self, but as this
3620
# disk format scales terribly for push anyway due to rewriting
3621
# inventory.weave, this is considered acceptable.
3623
if revision_id is not None:
3624
source_ids = self.source.get_ancestry(revision_id)
3625
if source_ids[0] is not None:
3626
raise AssertionError()
3629
source_ids = self.source._all_possible_ids()
3630
source_ids_set = set(source_ids)
3631
# source_ids is the worst possible case we may need to pull.
3632
# now we want to filter source_ids against what we actually
3633
# have in target, but don't try to check for existence where we know
3634
# we do not have a revision as that would be pointless.
3635
target_ids = set(self.target._all_possible_ids())
3636
possibly_present_revisions = target_ids.intersection(source_ids_set)
3637
actually_present_revisions = set(
3638
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3639
required_revisions = source_ids_set.difference(actually_present_revisions)
3640
if revision_id is not None:
3641
# we used get_ancestry to determine source_ids then we are assured all
3642
# revisions referenced are present as they are installed in topological order.
3643
# and the tip revision was validated by get_ancestry.
3644
result_set = required_revisions
3646
# if we just grabbed the possibly available ids, then
3647
# we only have an estimate of whats available and need to validate
3648
# that against the revision records.
3650
self.source._eliminate_revisions_not_present(required_revisions))
3651
return self.source.revision_ids_to_search_result(result_set)
3654
class InterKnitRepo(InterSameDataRepository):
3655
"""Optimised code paths between Knit based repositories."""
3658
def _get_repo_format_to_test(self):
3659
from bzrlib.repofmt import knitrepo
3660
return knitrepo.RepositoryFormatKnit1()
3663
def is_compatible(source, target):
3664
"""Be compatible with known Knit formats.
3666
We don't test for the stores being of specific types because that
3667
could lead to confusing results, and there is no need to be
3670
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3672
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3673
isinstance(target._format, RepositoryFormatKnit))
3674
except AttributeError:
3676
return are_knits and InterRepository._same_model(source, target)
3679
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3680
"""See InterRepository.missing_revision_ids()."""
3681
if revision_id is not None:
3682
source_ids = self.source.get_ancestry(revision_id)
3683
if source_ids[0] is not None:
3684
raise AssertionError()
3687
source_ids = self.source.all_revision_ids()
3688
source_ids_set = set(source_ids)
3689
# source_ids is the worst possible case we may need to pull.
3690
# now we want to filter source_ids against what we actually
3691
# have in target, but don't try to check for existence where we know
3692
# we do not have a revision as that would be pointless.
3693
target_ids = set(self.target.all_revision_ids())
3694
possibly_present_revisions = target_ids.intersection(source_ids_set)
3695
actually_present_revisions = set(
3696
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3697
required_revisions = source_ids_set.difference(actually_present_revisions)
3698
if revision_id is not None:
3699
# we used get_ancestry to determine source_ids then we are assured all
3700
# revisions referenced are present as they are installed in topological order.
3701
# and the tip revision was validated by get_ancestry.
3702
result_set = required_revisions
3704
# if we just grabbed the possibly available ids, then
3705
# we only have an estimate of whats available and need to validate
3706
# that against the revision records.
3708
self.source._eliminate_revisions_not_present(required_revisions))
3709
return self.source.revision_ids_to_search_result(result_set)
3712
class InterDifferingSerializer(InterRepository):
3715
def _get_repo_format_to_test(self):
3719
def is_compatible(source, target):
3720
"""Be compatible with Knit2 source and Knit3 target"""
3721
# This is redundant with format.check_conversion_target(), however that
3722
# raises an exception, and we just want to say "False" as in we won't
3723
# support converting between these formats.
3724
if 'IDS_never' in debug.debug_flags:
3726
if source.supports_rich_root() and not target.supports_rich_root():
3728
if (source._format.supports_tree_reference
3729
and not target._format.supports_tree_reference):
3731
if target._fallback_repositories and target._format.supports_chks:
3732
# IDS doesn't know how to copy CHKs for the parent inventories it
3733
# adds to stacked repos.
3735
if 'IDS_always' in debug.debug_flags:
3737
# Only use this code path for local source and target. IDS does far
3738
# too much IO (both bandwidth and roundtrips) over a network.
3739
if not source.bzrdir.transport.base.startswith('file:///'):
3741
if not target.bzrdir.transport.base.startswith('file:///'):
3745
def _get_trees(self, revision_ids, cache):
3747
for rev_id in revision_ids:
3749
possible_trees.append((rev_id, cache[rev_id]))
3751
# Not cached, but inventory might be present anyway.
3753
tree = self.source.revision_tree(rev_id)
3754
except errors.NoSuchRevision:
3755
# Nope, parent is ghost.
3758
cache[rev_id] = tree
3759
possible_trees.append((rev_id, tree))
3760
return possible_trees
3762
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3763
"""Get the best delta and base for this revision.
3765
:return: (basis_id, delta)
3768
# Generate deltas against each tree, to find the shortest.
3769
texts_possibly_new_in_tree = set()
3770
for basis_id, basis_tree in possible_trees:
3771
delta = tree.inventory._make_delta(basis_tree.inventory)
3772
for old_path, new_path, file_id, new_entry in delta:
3773
if new_path is None:
3774
# This file_id isn't present in the new rev, so we don't
3778
# Rich roots are handled elsewhere...
3780
kind = new_entry.kind
3781
if kind != 'directory' and kind != 'file':
3782
# No text record associated with this inventory entry.
3784
# This is a directory or file that has changed somehow.
3785
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3786
deltas.append((len(delta), basis_id, delta))
3788
return deltas[0][1:]
3790
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3791
"""Find all parent revisions that are absent, but for which the
3792
inventory is present, and copy those inventories.
3794
This is necessary to preserve correctness when the source is stacked
3795
without fallbacks configured. (Note that in cases like upgrade the
3796
source may be not have _fallback_repositories even though it is
3800
for parents in parent_map.values():
3801
parent_revs.update(parents)
3802
present_parents = self.source.get_parent_map(parent_revs)
3803
absent_parents = set(parent_revs).difference(present_parents)
3804
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3805
(rev_id,) for rev_id in absent_parents)
3806
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3807
for parent_tree in self.source.revision_trees(parent_inv_ids):
3808
current_revision_id = parent_tree.get_revision_id()
3809
parents_parents_keys = parent_invs_keys_for_stacking[
3810
(current_revision_id,)]
3811
parents_parents = [key[-1] for key in parents_parents_keys]
3812
basis_id = _mod_revision.NULL_REVISION
3813
basis_tree = self.source.revision_tree(basis_id)
3814
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3815
self.target.add_inventory_by_delta(
3816
basis_id, delta, current_revision_id, parents_parents)
3817
cache[current_revision_id] = parent_tree
3819
def _fetch_batch(self, revision_ids, basis_id, cache):
3820
"""Fetch across a few revisions.
3822
:param revision_ids: The revisions to copy
3823
:param basis_id: The revision_id of a tree that must be in cache, used
3824
as a basis for delta when no other base is available
3825
:param cache: A cache of RevisionTrees that we can use.
3826
:return: The revision_id of the last converted tree. The RevisionTree
3827
for it will be in cache
3829
# Walk though all revisions; get inventory deltas, copy referenced
3830
# texts that delta references, insert the delta, revision and
3832
root_keys_to_create = set()
3835
pending_revisions = []
3836
parent_map = self.source.get_parent_map(revision_ids)
3837
self._fetch_parent_invs_for_stacking(parent_map, cache)
3838
for tree in self.source.revision_trees(revision_ids):
3839
# Find a inventory delta for this revision.
3840
# Find text entries that need to be copied, too.
3841
current_revision_id = tree.get_revision_id()
3842
parent_ids = parent_map.get(current_revision_id, ())
3843
parent_trees = self._get_trees(parent_ids, cache)
3844
possible_trees = list(parent_trees)
3845
if len(possible_trees) == 0:
3846
# There either aren't any parents, or the parents are ghosts,
3847
# so just use the last converted tree.
3848
possible_trees.append((basis_id, cache[basis_id]))
3849
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3851
revision = self.source.get_revision(current_revision_id)
3852
pending_deltas.append((basis_id, delta,
3853
current_revision_id, revision.parent_ids))
3854
if self._converting_to_rich_root:
3855
self._revision_id_to_root_id[current_revision_id] = \
3857
# Determine which texts are in present in this revision but not in
3858
# any of the available parents.
3859
texts_possibly_new_in_tree = set()
3860
for old_path, new_path, file_id, entry in delta:
3861
if new_path is None:
3862
# This file_id isn't present in the new rev
3866
if not self.target.supports_rich_root():
3867
# The target doesn't support rich root, so we don't
3870
if self._converting_to_rich_root:
3871
# This can't be copied normally, we have to insert
3873
root_keys_to_create.add((file_id, entry.revision))
3876
texts_possibly_new_in_tree.add((file_id, entry.revision))
3877
for basis_id, basis_tree in possible_trees:
3878
basis_inv = basis_tree.inventory
3879
for file_key in list(texts_possibly_new_in_tree):
3880
file_id, file_revision = file_key
3882
entry = basis_inv[file_id]
3883
except errors.NoSuchId:
3885
if entry.revision == file_revision:
3886
texts_possibly_new_in_tree.remove(file_key)
3887
text_keys.update(texts_possibly_new_in_tree)
3888
pending_revisions.append(revision)
3889
cache[current_revision_id] = tree
3890
basis_id = current_revision_id
3892
from_texts = self.source.texts
3893
to_texts = self.target.texts
3894
if root_keys_to_create:
3895
from bzrlib.fetch import _new_root_data_stream
3896
root_stream = _new_root_data_stream(
3897
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3899
to_texts.insert_record_stream(root_stream)
3900
to_texts.insert_record_stream(from_texts.get_record_stream(
3901
text_keys, self.target._format._fetch_order,
3902
not self.target._format._fetch_uses_deltas))
3903
# insert inventory deltas
3904
for delta in pending_deltas:
3905
self.target.add_inventory_by_delta(*delta)
3906
if self.target._fallback_repositories:
3907
# Make sure this stacked repository has all the parent inventories
3908
# for the new revisions that we are about to insert. We do this
3909
# before adding the revisions so that no revision is added until
3910
# all the inventories it may depend on are added.
3911
# Note that this is overzealous, as we may have fetched these in an
3914
revision_ids = set()
3915
for revision in pending_revisions:
3916
revision_ids.add(revision.revision_id)
3917
parent_ids.update(revision.parent_ids)
3918
parent_ids.difference_update(revision_ids)
3919
parent_ids.discard(_mod_revision.NULL_REVISION)
3920
parent_map = self.source.get_parent_map(parent_ids)
3921
# we iterate over parent_map and not parent_ids because we don't
3922
# want to try copying any revision which is a ghost
3923
for parent_tree in self.source.revision_trees(parent_map):
3924
current_revision_id = parent_tree.get_revision_id()
3925
parents_parents = parent_map[current_revision_id]
3926
possible_trees = self._get_trees(parents_parents, cache)
3927
if len(possible_trees) == 0:
3928
# There either aren't any parents, or the parents are
3929
# ghosts, so just use the last converted tree.
3930
possible_trees.append((basis_id, cache[basis_id]))
3931
basis_id, delta = self._get_delta_for_revision(parent_tree,
3932
parents_parents, possible_trees)
3933
self.target.add_inventory_by_delta(
3934
basis_id, delta, current_revision_id, parents_parents)
3935
# insert signatures and revisions
3936
for revision in pending_revisions:
3938
signature = self.source.get_signature_text(
3939
revision.revision_id)
3940
self.target.add_signature_text(revision.revision_id,
3942
except errors.NoSuchRevision:
3944
self.target.add_revision(revision.revision_id, revision)
3947
def _fetch_all_revisions(self, revision_ids, pb):
3948
"""Fetch everything for the list of revisions.
3950
:param revision_ids: The list of revisions to fetch. Must be in
3952
:param pb: A ProgressTask
3955
basis_id, basis_tree = self._get_basis(revision_ids[0])
3957
cache = lru_cache.LRUCache(100)
3958
cache[basis_id] = basis_tree
3959
del basis_tree # We don't want to hang on to it here
3961
for offset in range(0, len(revision_ids), batch_size):
3962
self.target.start_write_group()
3964
pb.update('Transferring revisions', offset,
3966
batch = revision_ids[offset:offset+batch_size]
3967
basis_id = self._fetch_batch(batch, basis_id, cache)
3969
self.target.abort_write_group()
3972
hint = self.target.commit_write_group()
3975
if hints and self.target._format.pack_compresses:
3976
self.target.pack(hint=hints)
3977
pb.update('Transferring revisions', len(revision_ids),
3981
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3983
"""See InterRepository.fetch()."""
3984
if fetch_spec is not None:
3985
raise AssertionError("Not implemented yet...")
3986
if (not self.source.supports_rich_root()
3987
and self.target.supports_rich_root()):
3988
self._converting_to_rich_root = True
3989
self._revision_id_to_root_id = {}
3991
self._converting_to_rich_root = False
3992
revision_ids = self.target.search_missing_revision_ids(self.source,
3993
revision_id, find_ghosts=find_ghosts).get_keys()
3994
if not revision_ids:
3996
revision_ids = tsort.topo_sort(
3997
self.source.get_graph().get_parent_map(revision_ids))
3998
if not revision_ids:
4000
# Walk though all revisions; get inventory deltas, copy referenced
4001
# texts that delta references, insert the delta, revision and
4004
my_pb = ui.ui_factory.nested_progress_bar()
4007
symbol_versioning.warn(
4008
symbol_versioning.deprecated_in((1, 14, 0))
4009
% "pb parameter to fetch()")
4012
self._fetch_all_revisions(revision_ids, pb)
4014
if my_pb is not None:
4016
return len(revision_ids), 0
4018
def _get_basis(self, first_revision_id):
4019
"""Get a revision and tree which exists in the target.
4021
This assumes that first_revision_id is selected for transmission
4022
because all other ancestors are already present. If we can't find an
4023
ancestor we fall back to NULL_REVISION since we know that is safe.
4025
:return: (basis_id, basis_tree)
4027
first_rev = self.source.get_revision(first_revision_id)
4029
basis_id = first_rev.parent_ids[0]
4030
# only valid as a basis if the target has it
4031
self.target.get_revision(basis_id)
4032
# Try to get a basis tree - if its a ghost it will hit the
4033
# NoSuchRevision case.
4034
basis_tree = self.source.revision_tree(basis_id)
4035
except (IndexError, errors.NoSuchRevision):
4036
basis_id = _mod_revision.NULL_REVISION
4037
basis_tree = self.source.revision_tree(basis_id)
4038
return basis_id, basis_tree
4041
InterRepository.register_optimiser(InterDifferingSerializer)
4042
InterRepository.register_optimiser(InterSameDataRepository)
4043
InterRepository.register_optimiser(InterWeaveRepo)
4044
InterRepository.register_optimiser(InterKnitRepo)
1788
4047
class CopyConverter(object):
1789
4048
"""A repository conversion tool which just performs a copy of the content.
1804
4063
:param to_convert: The disk object to convert.
1805
4064
:param pb: a progress bar to use for progress information.
1807
pb = ui.ui_factory.nested_progress_bar()
1810
4069
# this is only useful with metadir layouts - separated repo content.
1811
4070
# trigger an assertion if not such
1812
4071
repo._format.get_format_string()
1813
4072
self.repo_dir = repo.bzrdir
1814
pb.update('Moving repository to repository.backup')
4073
self.step('Moving repository to repository.backup')
1815
4074
self.repo_dir.transport.move('repository', 'repository.backup')
1816
4075
backup_transport = self.repo_dir.transport.clone('repository.backup')
1817
4076
repo._format.check_conversion_target(self.target_format)
1818
4077
self.source_repo = repo._format.open(self.repo_dir,
1820
4079
_override_transport=backup_transport)
1821
pb.update('Creating new repository')
4080
self.step('Creating new repository')
1822
4081
converted = self.target_format.initialize(self.repo_dir,
1823
4082
self.source_repo.is_shared())
1824
4083
converted.lock_write()
1826
pb.update('Copying content')
4085
self.step('Copying content')
1827
4086
self.source_repo.copy_content_into(converted)
1829
4088
converted.unlock()
1830
pb.update('Deleting old repository content')
4089
self.step('Deleting old repository content')
1831
4090
self.repo_dir.transport.delete_tree('repository.backup')
1832
4091
ui.ui_factory.note('repository converted')
4093
def step(self, message):
4094
"""Update the pb by a step."""
4096
self.pb.update(message, self.count, self.total)
4108
def _unescaper(match, _map=_unescape_map):
4109
code = match.group(1)
4113
if not code.startswith('#'):
4115
return unichr(int(code[1:])).encode('utf8')
4121
def _unescape_xml(data):
4122
"""Unescape predefined XML entities in a string of data."""
4124
if _unescape_re is None:
4125
_unescape_re = re.compile('\&([^;]*);')
4126
return _unescape_re.sub(_unescaper, data)
4129
class _VersionedFileChecker(object):
4131
def __init__(self, repository, text_key_references=None, ancestors=None):
4132
self.repository = repository
4133
self.text_index = self.repository._generate_text_key_index(
4134
text_key_references=text_key_references, ancestors=ancestors)
4136
def calculate_file_version_parents(self, text_key):
4137
"""Calculate the correct parents for a file version according to
4140
parent_keys = self.text_index[text_key]
4141
if parent_keys == [_mod_revision.NULL_REVISION]:
4143
return tuple(parent_keys)
4145
def check_file_version_parents(self, texts, progress_bar=None):
4146
"""Check the parents stored in a versioned file are correct.
4148
It also detects file versions that are not referenced by their
4149
corresponding revision's inventory.
4151
:returns: A tuple of (wrong_parents, dangling_file_versions).
4152
wrong_parents is a dict mapping {revision_id: (stored_parents,
4153
correct_parents)} for each revision_id where the stored parents
4154
are not correct. dangling_file_versions is a set of (file_id,
4155
revision_id) tuples for versions that are present in this versioned
4156
file, but not used by the corresponding inventory.
4158
local_progress = None
4159
if progress_bar is None:
4160
local_progress = ui.ui_factory.nested_progress_bar()
4161
progress_bar = local_progress
4163
return self._check_file_version_parents(texts, progress_bar)
4166
local_progress.finished()
4168
def _check_file_version_parents(self, texts, progress_bar):
4169
"""See check_file_version_parents."""
4171
self.file_ids = set([file_id for file_id, _ in
4172
self.text_index.iterkeys()])
4173
# text keys is now grouped by file_id
4174
n_versions = len(self.text_index)
4175
progress_bar.update('loading text store', 0, n_versions)
4176
parent_map = self.repository.texts.get_parent_map(self.text_index)
4177
# On unlistable transports this could well be empty/error...
4178
text_keys = self.repository.texts.keys()
4179
unused_keys = frozenset(text_keys) - set(self.text_index)
4180
for num, key in enumerate(self.text_index.iterkeys()):
4181
progress_bar.update('checking text graph', num, n_versions)
4182
correct_parents = self.calculate_file_version_parents(key)
4184
knit_parents = parent_map[key]
4185
except errors.RevisionNotPresent:
4188
if correct_parents != knit_parents:
4189
wrong_parents[key] = (knit_parents, correct_parents)
4190
return wrong_parents, unused_keys
4193
def _old_get_graph(repository, revision_id):
4194
"""DO NOT USE. That is all. I'm serious."""
4195
graph = repository.get_graph()
4196
revision_graph = dict(((key, value) for key, value in
4197
graph.iter_ancestry([revision_id]) if value is not None))
4198
return _strip_NULL_ghosts(revision_graph)
1836
4201
def _strip_NULL_ghosts(revision_graph):
1844
4209
return revision_graph
4212
class StreamSink(object):
4213
"""An object that can insert a stream into a repository.
4215
This interface handles the complexity of reserialising inventories and
4216
revisions from different formats, and allows unidirectional insertion into
4217
stacked repositories without looking for the missing basis parents
4221
def __init__(self, target_repo):
4222
self.target_repo = target_repo
4224
def insert_stream(self, stream, src_format, resume_tokens):
4225
"""Insert a stream's content into the target repository.
4227
:param src_format: a bzr repository format.
4229
:return: a list of resume tokens and an iterable of keys additional
4230
items required before the insertion can be completed.
4232
self.target_repo.lock_write()
4235
self.target_repo.resume_write_group(resume_tokens)
4238
self.target_repo.start_write_group()
4241
# locked_insert_stream performs a commit|suspend.
4242
return self._locked_insert_stream(stream, src_format, is_resume)
4244
self.target_repo.abort_write_group(suppress_errors=True)
4247
self.target_repo.unlock()
4249
def _locked_insert_stream(self, stream, src_format, is_resume):
4250
to_serializer = self.target_repo._format._serializer
4251
src_serializer = src_format._serializer
4253
if to_serializer == src_serializer:
4254
# If serializers match and the target is a pack repository, set the
4255
# write cache size on the new pack. This avoids poor performance
4256
# on transports where append is unbuffered (such as
4257
# RemoteTransport). This is safe to do because nothing should read
4258
# back from the target repository while a stream with matching
4259
# serialization is being inserted.
4260
# The exception is that a delta record from the source that should
4261
# be a fulltext may need to be expanded by the target (see
4262
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4263
# explicitly flush any buffered writes first in that rare case.
4265
new_pack = self.target_repo._pack_collection._new_pack
4266
except AttributeError:
4267
# Not a pack repository
4270
new_pack.set_write_cache_size(1024*1024)
4271
for substream_type, substream in stream:
4272
if 'stream' in debug.debug_flags:
4273
mutter('inserting substream: %s', substream_type)
4274
if substream_type == 'texts':
4275
self.target_repo.texts.insert_record_stream(substream)
4276
elif substream_type == 'inventories':
4277
if src_serializer == to_serializer:
4278
self.target_repo.inventories.insert_record_stream(
4281
self._extract_and_insert_inventories(
4282
substream, src_serializer)
4283
elif substream_type == 'inventory-deltas':
4284
self._extract_and_insert_inventory_deltas(
4285
substream, src_serializer)
4286
elif substream_type == 'chk_bytes':
4287
# XXX: This doesn't support conversions, as it assumes the
4288
# conversion was done in the fetch code.
4289
self.target_repo.chk_bytes.insert_record_stream(substream)
4290
elif substream_type == 'revisions':
4291
# This may fallback to extract-and-insert more often than
4292
# required if the serializers are different only in terms of
4294
if src_serializer == to_serializer:
4295
self.target_repo.revisions.insert_record_stream(
4298
self._extract_and_insert_revisions(substream,
4300
elif substream_type == 'signatures':
4301
self.target_repo.signatures.insert_record_stream(substream)
4303
raise AssertionError('kaboom! %s' % (substream_type,))
4304
# Done inserting data, and the missing_keys calculations will try to
4305
# read back from the inserted data, so flush the writes to the new pack
4306
# (if this is pack format).
4307
if new_pack is not None:
4308
new_pack._write_data('', flush=True)
4309
# Find all the new revisions (including ones from resume_tokens)
4310
missing_keys = self.target_repo.get_missing_parent_inventories(
4311
check_for_missing_texts=is_resume)
4313
for prefix, versioned_file in (
4314
('texts', self.target_repo.texts),
4315
('inventories', self.target_repo.inventories),
4316
('revisions', self.target_repo.revisions),
4317
('signatures', self.target_repo.signatures),
4318
('chk_bytes', self.target_repo.chk_bytes),
4320
if versioned_file is None:
4322
missing_keys.update((prefix,) + key for key in
4323
versioned_file.get_missing_compression_parent_keys())
4324
except NotImplementedError:
4325
# cannot even attempt suspending, and missing would have failed
4326
# during stream insertion.
4327
missing_keys = set()
4330
# suspend the write group and tell the caller what we is
4331
# missing. We know we can suspend or else we would not have
4332
# entered this code path. (All repositories that can handle
4333
# missing keys can handle suspending a write group).
4334
write_group_tokens = self.target_repo.suspend_write_group()
4335
return write_group_tokens, missing_keys
4336
hint = self.target_repo.commit_write_group()
4337
if (to_serializer != src_serializer and
4338
self.target_repo._format.pack_compresses):
4339
self.target_repo.pack(hint=hint)
4342
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4343
target_rich_root = self.target_repo._format.rich_root_data
4344
target_tree_refs = self.target_repo._format.supports_tree_reference
4345
for record in substream:
4346
# Insert the delta directly
4347
inventory_delta_bytes = record.get_bytes_as('fulltext')
4348
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4350
parse_result = deserialiser.parse_text_bytes(
4351
inventory_delta_bytes)
4352
except inventory_delta.IncompatibleInventoryDelta, err:
4353
trace.mutter("Incompatible delta: %s", err.msg)
4354
raise errors.IncompatibleRevision(self.target_repo._format)
4355
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4356
revision_id = new_id
4357
parents = [key[0] for key in record.parents]
4358
self.target_repo.add_inventory_by_delta(
4359
basis_id, inv_delta, revision_id, parents)
4361
def _extract_and_insert_inventories(self, substream, serializer,
4363
"""Generate a new inventory versionedfile in target, converting data.
4365
The inventory is retrieved from the source, (deserializing it), and
4366
stored in the target (reserializing it in a different format).
4368
target_rich_root = self.target_repo._format.rich_root_data
4369
target_tree_refs = self.target_repo._format.supports_tree_reference
4370
for record in substream:
4371
# It's not a delta, so it must be a fulltext in the source
4372
# serializer's format.
4373
bytes = record.get_bytes_as('fulltext')
4374
revision_id = record.key[0]
4375
inv = serializer.read_inventory_from_string(bytes, revision_id)
4376
parents = [key[0] for key in record.parents]
4377
self.target_repo.add_inventory(revision_id, inv, parents)
4378
# No need to keep holding this full inv in memory when the rest of
4379
# the substream is likely to be all deltas.
4382
def _extract_and_insert_revisions(self, substream, serializer):
4383
for record in substream:
4384
bytes = record.get_bytes_as('fulltext')
4385
revision_id = record.key[0]
4386
rev = serializer.read_revision_from_string(bytes)
4387
if rev.revision_id != revision_id:
4388
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4389
self.target_repo.add_revision(revision_id, rev)
4392
if self.target_repo._format._fetch_reconcile:
4393
self.target_repo.reconcile()
4396
class StreamSource(object):
4397
"""A source of a stream for fetching between repositories."""
4399
def __init__(self, from_repository, to_format):
4400
"""Create a StreamSource streaming from from_repository."""
4401
self.from_repository = from_repository
4402
self.to_format = to_format
4404
def delta_on_metadata(self):
4405
"""Return True if delta's are permitted on metadata streams.
4407
That is on revisions and signatures.
4409
src_serializer = self.from_repository._format._serializer
4410
target_serializer = self.to_format._serializer
4411
return (self.to_format._fetch_uses_deltas and
4412
src_serializer == target_serializer)
4414
def _fetch_revision_texts(self, revs):
4415
# fetch signatures first and then the revision texts
4416
# may need to be a InterRevisionStore call here.
4417
from_sf = self.from_repository.signatures
4418
# A missing signature is just skipped.
4419
keys = [(rev_id,) for rev_id in revs]
4420
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4422
self.to_format._fetch_order,
4423
not self.to_format._fetch_uses_deltas))
4424
# If a revision has a delta, this is actually expanded inside the
4425
# insert_record_stream code now, which is an alternate fix for
4427
from_rf = self.from_repository.revisions
4428
revisions = from_rf.get_record_stream(
4430
self.to_format._fetch_order,
4431
not self.delta_on_metadata())
4432
return [('signatures', signatures), ('revisions', revisions)]
4434
def _generate_root_texts(self, revs):
4435
"""This will be called by get_stream between fetching weave texts and
4436
fetching the inventory weave.
4438
if self._rich_root_upgrade():
4440
return bzrlib.fetch.Inter1and2Helper(
4441
self.from_repository).generate_root_texts(revs)
4445
def get_stream(self, search):
4447
revs = search.get_keys()
4448
graph = self.from_repository.get_graph()
4449
revs = tsort.topo_sort(graph.get_parent_map(revs))
4450
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4452
for knit_kind, file_id, revisions in data_to_fetch:
4453
if knit_kind != phase:
4455
# Make a new progress bar for this phase
4456
if knit_kind == "file":
4457
# Accumulate file texts
4458
text_keys.extend([(file_id, revision) for revision in
4460
elif knit_kind == "inventory":
4461
# Now copy the file texts.
4462
from_texts = self.from_repository.texts
4463
yield ('texts', from_texts.get_record_stream(
4464
text_keys, self.to_format._fetch_order,
4465
not self.to_format._fetch_uses_deltas))
4466
# Cause an error if a text occurs after we have done the
4469
# Before we process the inventory we generate the root
4470
# texts (if necessary) so that the inventories references
4472
for _ in self._generate_root_texts(revs):
4474
# we fetch only the referenced inventories because we do not
4475
# know for unselected inventories whether all their required
4476
# texts are present in the other repository - it could be
4478
for info in self._get_inventory_stream(revs):
4480
elif knit_kind == "signatures":
4481
# Nothing to do here; this will be taken care of when
4482
# _fetch_revision_texts happens.
4484
elif knit_kind == "revisions":
4485
for record in self._fetch_revision_texts(revs):
4488
raise AssertionError("Unknown knit kind %r" % knit_kind)
4490
def get_stream_for_missing_keys(self, missing_keys):
4491
# missing keys can only occur when we are byte copying and not
4492
# translating (because translation means we don't send
4493
# unreconstructable deltas ever).
4495
keys['texts'] = set()
4496
keys['revisions'] = set()
4497
keys['inventories'] = set()
4498
keys['chk_bytes'] = set()
4499
keys['signatures'] = set()
4500
for key in missing_keys:
4501
keys[key[0]].add(key[1:])
4502
if len(keys['revisions']):
4503
# If we allowed copying revisions at this point, we could end up
4504
# copying a revision without copying its required texts: a
4505
# violation of the requirements for repository integrity.
4506
raise AssertionError(
4507
'cannot copy revisions to fill in missing deltas %s' % (
4508
keys['revisions'],))
4509
for substream_kind, keys in keys.iteritems():
4510
vf = getattr(self.from_repository, substream_kind)
4511
if vf is None and keys:
4512
raise AssertionError(
4513
"cannot fill in keys for a versioned file we don't"
4514
" have: %s needs %s" % (substream_kind, keys))
4516
# No need to stream something we don't have
4518
if substream_kind == 'inventories':
4519
# Some missing keys are genuinely ghosts, filter those out.
4520
present = self.from_repository.inventories.get_parent_map(keys)
4521
revs = [key[0] for key in present]
4522
# Get the inventory stream more-or-less as we do for the
4523
# original stream; there's no reason to assume that records
4524
# direct from the source will be suitable for the sink. (Think
4525
# e.g. 2a -> 1.9-rich-root).
4526
for info in self._get_inventory_stream(revs, missing=True):
4530
# Ask for full texts always so that we don't need more round trips
4531
# after this stream.
4532
# Some of the missing keys are genuinely ghosts, so filter absent
4533
# records. The Sink is responsible for doing another check to
4534
# ensure that ghosts don't introduce missing data for future
4536
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4537
self.to_format._fetch_order, True))
4538
yield substream_kind, stream
4540
def inventory_fetch_order(self):
4541
if self._rich_root_upgrade():
4542
return 'topological'
4544
return self.to_format._fetch_order
4546
def _rich_root_upgrade(self):
4547
return (not self.from_repository._format.rich_root_data and
4548
self.to_format.rich_root_data)
4550
def _get_inventory_stream(self, revision_ids, missing=False):
4551
from_format = self.from_repository._format
4552
if (from_format.supports_chks and self.to_format.supports_chks and
4553
from_format.network_name() == self.to_format.network_name()):
4554
raise AssertionError(
4555
"this case should be handled by GroupCHKStreamSource")
4556
elif 'forceinvdeltas' in debug.debug_flags:
4557
return self._get_convertable_inventory_stream(revision_ids,
4558
delta_versus_null=missing)
4559
elif from_format.network_name() == self.to_format.network_name():
4561
return self._get_simple_inventory_stream(revision_ids,
4563
elif (not from_format.supports_chks and not self.to_format.supports_chks
4564
and from_format._serializer == self.to_format._serializer):
4565
# Essentially the same format.
4566
return self._get_simple_inventory_stream(revision_ids,
4569
# Any time we switch serializations, we want to use an
4570
# inventory-delta based approach.
4571
return self._get_convertable_inventory_stream(revision_ids,
4572
delta_versus_null=missing)
4574
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4575
# NB: This currently reopens the inventory weave in source;
4576
# using a single stream interface instead would avoid this.
4577
from_weave = self.from_repository.inventories
4579
delta_closure = True
4581
delta_closure = not self.delta_on_metadata()
4582
yield ('inventories', from_weave.get_record_stream(
4583
[(rev_id,) for rev_id in revision_ids],
4584
self.inventory_fetch_order(), delta_closure))
4586
def _get_convertable_inventory_stream(self, revision_ids,
4587
delta_versus_null=False):
4588
# The source is using CHKs, but the target either doesn't or it has a
4589
# different serializer. The StreamSink code expects to be able to
4590
# convert on the target, so we need to put bytes-on-the-wire that can
4591
# be converted. That means inventory deltas (if the remote is <1.19,
4592
# RemoteStreamSink will fallback to VFS to insert the deltas).
4593
yield ('inventory-deltas',
4594
self._stream_invs_as_deltas(revision_ids,
4595
delta_versus_null=delta_versus_null))
4597
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4598
"""Return a stream of inventory-deltas for the given rev ids.
4600
:param revision_ids: The list of inventories to transmit
4601
:param delta_versus_null: Don't try to find a minimal delta for this
4602
entry, instead compute the delta versus the NULL_REVISION. This
4603
effectively streams a complete inventory. Used for stuff like
4604
filling in missing parents, etc.
4606
from_repo = self.from_repository
4607
revision_keys = [(rev_id,) for rev_id in revision_ids]
4608
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4609
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4611
inventories = self.from_repository.iter_inventories(
4612
revision_ids, 'topological')
4613
format = from_repo._format
4614
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4615
inventory_cache = lru_cache.LRUCache(50)
4616
null_inventory = from_repo.revision_tree(
4617
_mod_revision.NULL_REVISION).inventory
4618
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4619
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4620
# repo back into a non-rich-root repo ought to be allowed)
4621
serializer = inventory_delta.InventoryDeltaSerializer(
4622
versioned_root=format.rich_root_data,
4623
tree_references=format.supports_tree_reference)
4624
for inv in inventories:
4625
key = (inv.revision_id,)
4626
parent_keys = parent_map.get(key, ())
4628
if not delta_versus_null and parent_keys:
4629
# The caller did not ask for complete inventories and we have
4630
# some parents that we can delta against. Make a delta against
4631
# each parent so that we can find the smallest.
4632
parent_ids = [parent_key[0] for parent_key in parent_keys]
4633
for parent_id in parent_ids:
4634
if parent_id not in invs_sent_so_far:
4635
# We don't know that the remote side has this basis, so
4638
if parent_id == _mod_revision.NULL_REVISION:
4639
parent_inv = null_inventory
4641
parent_inv = inventory_cache.get(parent_id, None)
4642
if parent_inv is None:
4643
parent_inv = from_repo.get_inventory(parent_id)
4644
candidate_delta = inv._make_delta(parent_inv)
4645
if (delta is None or
4646
len(delta) > len(candidate_delta)):
4647
delta = candidate_delta
4648
basis_id = parent_id
4650
# Either none of the parents ended up being suitable, or we
4651
# were asked to delta against NULL
4652
basis_id = _mod_revision.NULL_REVISION
4653
delta = inv._make_delta(null_inventory)
4654
invs_sent_so_far.add(inv.revision_id)
4655
inventory_cache[inv.revision_id] = inv
4656
delta_serialized = ''.join(
4657
serializer.delta_to_lines(basis_id, key[-1], delta))
4658
yield versionedfile.FulltextContentFactory(
4659
key, parent_keys, None, delta_serialized)
1847
4662
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
1848
4663
stop_revision=None):
1849
4664
"""Extend the partial history to include a given index