202
342
commit to be valid, deletes against the basis MUST be recorded via
203
343
builder.record_delete().
205
raise NotImplementedError(self.will_record_deletes)
207
def record_iter_changes(self, tree, basis_revision_id, iter_changes):
345
self._recording_deletes = True
347
basis_id = self.parents[0]
349
basis_id = _mod_revision.NULL_REVISION
350
self.basis_delta_revision = basis_id
352
def record_entry_contents(self, ie, parent_invs, path, tree,
354
"""Record the content of ie from tree into the commit if needed.
356
Side effect: sets ie.revision when unchanged
358
:param ie: An inventory entry present in the commit.
359
:param parent_invs: The inventories of the parent revisions of the
361
:param path: The path the entry is at in the tree.
362
:param tree: The tree which contains this entry and should be used to
364
:param content_summary: Summary data from the tree about the paths
365
content - stat, length, exec, sha/link target. This is only
366
accessed when the entry has a revision of None - that is when it is
367
a candidate to commit.
368
:return: A tuple (change_delta, version_recorded, fs_hash).
369
change_delta is an inventory_delta change for this entry against
370
the basis tree of the commit, or None if no change occured against
372
version_recorded is True if a new version of the entry has been
373
recorded. For instance, committing a merge where a file was only
374
changed on the other side will return (delta, False).
375
fs_hash is either None, or the hash details for the path (currently
376
a tuple of the contents sha1 and the statvalue returned by
377
tree.get_file_with_stat()).
379
if self.new_inventory.root is None:
380
if ie.parent_id is not None:
381
raise errors.RootMissing()
382
self._check_root(ie, parent_invs, tree)
383
if ie.revision is None:
384
kind = content_summary[0]
386
# ie is carried over from a prior commit
388
# XXX: repository specific check for nested tree support goes here - if
389
# the repo doesn't want nested trees we skip it ?
390
if (kind == 'tree-reference' and
391
not self.repository._format.supports_tree_reference):
392
# mismatch between commit builder logic and repository:
393
# this needs the entry creation pushed down into the builder.
394
raise NotImplementedError('Missing repository subtree support.')
395
self.new_inventory.add(ie)
397
# TODO: slow, take it out of the inner loop.
399
basis_inv = parent_invs[0]
401
basis_inv = Inventory(root_id=None)
403
# ie.revision is always None if the InventoryEntry is considered
404
# for committing. We may record the previous parents revision if the
405
# content is actually unchanged against a sole head.
406
if ie.revision is not None:
407
if not self._versioned_root and path == '':
408
# repositories that do not version the root set the root's
409
# revision to the new commit even when no change occurs (more
410
# specifically, they do not record a revision on the root; and
411
# the rev id is assigned to the root during deserialisation -
412
# this masks when a change may have occurred against the basis.
413
# To match this we always issue a delta, because the revision
414
# of the root will always be changing.
415
if ie.file_id in basis_inv:
416
delta = (basis_inv.id2path(ie.file_id), path,
420
delta = (None, path, ie.file_id, ie)
421
self._basis_delta.append(delta)
422
return delta, False, None
424
# we don't need to commit this, because the caller already
425
# determined that an existing revision of this file is
426
# appropriate. If its not being considered for committing then
427
# it and all its parents to the root must be unaltered so
428
# no-change against the basis.
429
if ie.revision == self._new_revision_id:
430
raise AssertionError("Impossible situation, a skipped "
431
"inventory entry (%r) claims to be modified in this "
432
"commit (%r).", (ie, self._new_revision_id))
433
return None, False, None
434
# XXX: Friction: parent_candidates should return a list not a dict
435
# so that we don't have to walk the inventories again.
436
parent_candiate_entries = ie.parent_candidates(parent_invs)
437
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
439
for inv in parent_invs:
440
if ie.file_id in inv:
441
old_rev = inv[ie.file_id].revision
442
if old_rev in head_set:
443
heads.append(inv[ie.file_id].revision)
444
head_set.remove(inv[ie.file_id].revision)
447
# now we check to see if we need to write a new record to the
449
# We write a new entry unless there is one head to the ancestors, and
450
# the kind-derived content is unchanged.
452
# Cheapest check first: no ancestors, or more the one head in the
453
# ancestors, we write a new node.
457
# There is a single head, look it up for comparison
458
parent_entry = parent_candiate_entries[heads[0]]
459
# if the non-content specific data has changed, we'll be writing a
461
if (parent_entry.parent_id != ie.parent_id or
462
parent_entry.name != ie.name):
464
# now we need to do content specific checks:
466
# if the kind changed the content obviously has
467
if kind != parent_entry.kind:
469
# Stat cache fingerprint feedback for the caller - None as we usually
470
# don't generate one.
473
if content_summary[2] is None:
474
raise ValueError("Files must not have executable = None")
476
# We can't trust a check of the file length because of content
478
if (# if the exec bit has changed we have to store:
479
parent_entry.executable != content_summary[2]):
481
elif parent_entry.text_sha1 == content_summary[3]:
482
# all meta and content is unchanged (using a hash cache
483
# hit to check the sha)
484
ie.revision = parent_entry.revision
485
ie.text_size = parent_entry.text_size
486
ie.text_sha1 = parent_entry.text_sha1
487
ie.executable = parent_entry.executable
488
return self._get_delta(ie, basis_inv, path), False, None
490
# Either there is only a hash change(no hash cache entry,
491
# or same size content change), or there is no change on
493
# Provide the parent's hash to the store layer, so that the
494
# content is unchanged we will not store a new node.
495
nostore_sha = parent_entry.text_sha1
497
# We want to record a new node regardless of the presence or
498
# absence of a content change in the file.
500
ie.executable = content_summary[2]
501
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
503
text = file_obj.read()
507
ie.text_sha1, ie.text_size = self._add_text_to_weave(
508
ie.file_id, text, heads, nostore_sha)
509
# Let the caller know we generated a stat fingerprint.
510
fingerprint = (ie.text_sha1, stat_value)
511
except errors.ExistingContent:
512
# Turns out that the file content was unchanged, and we were
513
# only going to store a new node if it was changed. Carry over
515
ie.revision = parent_entry.revision
516
ie.text_size = parent_entry.text_size
517
ie.text_sha1 = parent_entry.text_sha1
518
ie.executable = parent_entry.executable
519
return self._get_delta(ie, basis_inv, path), False, None
520
elif kind == 'directory':
522
# all data is meta here, nothing specific to directory, so
524
ie.revision = parent_entry.revision
525
return self._get_delta(ie, basis_inv, path), False, None
526
self._add_text_to_weave(ie.file_id, '', heads, None)
527
elif kind == 'symlink':
528
current_link_target = content_summary[3]
530
# symlink target is not generic metadata, check if it has
532
if current_link_target != parent_entry.symlink_target:
535
# unchanged, carry over.
536
ie.revision = parent_entry.revision
537
ie.symlink_target = parent_entry.symlink_target
538
return self._get_delta(ie, basis_inv, path), False, None
539
ie.symlink_target = current_link_target
540
self._add_text_to_weave(ie.file_id, '', heads, None)
541
elif kind == 'tree-reference':
543
if content_summary[3] != parent_entry.reference_revision:
546
# unchanged, carry over.
547
ie.reference_revision = parent_entry.reference_revision
548
ie.revision = parent_entry.revision
549
return self._get_delta(ie, basis_inv, path), False, None
550
ie.reference_revision = content_summary[3]
551
if ie.reference_revision is None:
552
raise AssertionError("invalid content_summary for nested tree: %r"
553
% (content_summary,))
554
self._add_text_to_weave(ie.file_id, '', heads, None)
556
raise NotImplementedError('unknown kind')
557
ie.revision = self._new_revision_id
558
self._any_changes = True
559
return self._get_delta(ie, basis_inv, path), True, fingerprint
561
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
562
_entry_factory=entry_factory):
208
563
"""Record a new tree via iter_changes.
210
565
:param tree: The tree to obtain text contents from for changed objects.
215
570
to basis_revision_id. The iterator must not include any items with
216
571
a current kind of None - missing items must be either filtered out
217
572
or errored-on beefore record_iter_changes sees the item.
573
:param _entry_factory: Private method to bind entry_factory locally for
218
575
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
219
576
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,
578
# Create an inventory delta based on deltas between all the parents and
579
# deltas between all the parent inventories. We use inventory delta's
580
# between the inventory objects because iter_changes masks
581
# last-changed-field only changes.
583
# file_id -> change map, change is fileid, paths, changed, versioneds,
584
# parents, names, kinds, executables
586
# {file_id -> revision_id -> inventory entry, for entries in parent
587
# trees that are not parents[0]
591
revtrees = list(self.repository.revision_trees(self.parents))
592
except errors.NoSuchRevision:
593
# one or more ghosts, slow path.
595
for revision_id in self.parents:
597
revtrees.append(self.repository.revision_tree(revision_id))
598
except errors.NoSuchRevision:
600
basis_revision_id = _mod_revision.NULL_REVISION
602
revtrees.append(self.repository.revision_tree(
603
_mod_revision.NULL_REVISION))
604
# The basis inventory from a repository
606
basis_inv = revtrees[0].inventory
608
basis_inv = self.repository.revision_tree(
609
_mod_revision.NULL_REVISION).inventory
610
if len(self.parents) > 0:
611
if basis_revision_id != self.parents[0] and not ghost_basis:
613
"arbitrary basis parents not yet supported with merges")
614
for revtree in revtrees[1:]:
615
for change in revtree.inventory._make_delta(basis_inv):
616
if change[1] is None:
617
# Not present in this parent.
619
if change[2] not in merged_ids:
620
if change[0] is not None:
621
basis_entry = basis_inv[change[2]]
622
merged_ids[change[2]] = [
624
basis_entry.revision,
627
parent_entries[change[2]] = {
629
basis_entry.revision:basis_entry,
631
change[3].revision:change[3],
634
merged_ids[change[2]] = [change[3].revision]
635
parent_entries[change[2]] = {change[3].revision:change[3]}
637
merged_ids[change[2]].append(change[3].revision)
638
parent_entries[change[2]][change[3].revision] = change[3]
641
# Setup the changes from the tree:
642
# changes maps file_id -> (change, [parent revision_ids])
644
for change in iter_changes:
645
# This probably looks up in basis_inv way to much.
646
if change[1][0] is not None:
647
head_candidate = [basis_inv[change[0]].revision]
650
changes[change[0]] = change, merged_ids.get(change[0],
652
unchanged_merged = set(merged_ids) - set(changes)
653
# Extend the changes dict with synthetic changes to record merges of
655
for file_id in unchanged_merged:
656
# Record a merged version of these items that did not change vs the
657
# basis. This can be either identical parallel changes, or a revert
658
# of a specific file after a merge. The recorded content will be
659
# that of the current tree (which is the same as the basis), but
660
# the per-file graph will reflect a merge.
661
# NB:XXX: We are reconstructing path information we had, this
662
# should be preserved instead.
663
# inv delta change: (file_id, (path_in_source, path_in_target),
664
# changed_content, versioned, parent, name, kind,
667
basis_entry = basis_inv[file_id]
668
except errors.NoSuchId:
669
# a change from basis->some_parents but file_id isn't in basis
670
# so was new in the merge, which means it must have changed
671
# from basis -> current, and as it hasn't the add was reverted
672
# by the user. So we discard this change.
676
(basis_inv.id2path(file_id), tree.id2path(file_id)),
678
(basis_entry.parent_id, basis_entry.parent_id),
679
(basis_entry.name, basis_entry.name),
680
(basis_entry.kind, basis_entry.kind),
681
(basis_entry.executable, basis_entry.executable))
682
changes[file_id] = (change, merged_ids[file_id])
683
# changes contains tuples with the change and a set of inventory
684
# candidates for the file.
686
# old_path, new_path, file_id, new_inventory_entry
687
seen_root = False # Is the root in the basis delta?
688
inv_delta = self._basis_delta
689
modified_rev = self._new_revision_id
690
for change, head_candidates in changes.values():
691
if change[3][1]: # versioned in target.
692
# Several things may be happening here:
693
# We may have a fork in the per-file graph
694
# - record a change with the content from tree
695
# We may have a change against < all trees
696
# - carry over the tree that hasn't changed
697
# We may have a change against all trees
698
# - record the change with the content from tree
701
entry = _entry_factory[kind](file_id, change[5][1],
703
head_set = self._heads(change[0], set(head_candidates))
706
for head_candidate in head_candidates:
707
if head_candidate in head_set:
708
heads.append(head_candidate)
709
head_set.remove(head_candidate)
712
# Could be a carry-over situation:
713
parent_entry_revs = parent_entries.get(file_id, None)
714
if parent_entry_revs:
715
parent_entry = parent_entry_revs.get(heads[0], None)
718
if parent_entry is None:
719
# The parent iter_changes was called against is the one
720
# that is the per-file head, so any change is relevant
721
# iter_changes is valid.
722
carry_over_possible = False
724
# could be a carry over situation
725
# A change against the basis may just indicate a merge,
726
# we need to check the content against the source of the
727
# merge to determine if it was changed after the merge
729
if (parent_entry.kind != entry.kind or
730
parent_entry.parent_id != entry.parent_id or
731
parent_entry.name != entry.name):
732
# Metadata common to all entries has changed
733
# against per-file parent
734
carry_over_possible = False
736
carry_over_possible = True
737
# per-type checks for changes against the parent_entry
740
# Cannot be a carry-over situation
741
carry_over_possible = False
742
# Populate the entry in the delta
744
# XXX: There is still a small race here: If someone reverts the content of a file
745
# after iter_changes examines and decides it has changed,
746
# we will unconditionally record a new version even if some
747
# other process reverts it while commit is running (with
748
# the revert happening after iter_changes did it's
751
entry.executable = True
753
entry.executable = False
754
if (carry_over_possible and
755
parent_entry.executable == entry.executable):
756
# Check the file length, content hash after reading
758
nostore_sha = parent_entry.text_sha1
761
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
763
text = file_obj.read()
767
entry.text_sha1, entry.text_size = self._add_text_to_weave(
768
file_id, text, heads, nostore_sha)
769
yield file_id, change[1][1], (entry.text_sha1, stat_value)
770
except errors.ExistingContent:
771
# No content change against a carry_over parent
772
# Perhaps this should also yield a fs hash update?
774
entry.text_size = parent_entry.text_size
775
entry.text_sha1 = parent_entry.text_sha1
776
elif kind == 'symlink':
778
entry.symlink_target = tree.get_symlink_target(file_id)
779
if (carry_over_possible and
780
parent_entry.symlink_target == entry.symlink_target):
783
self._add_text_to_weave(change[0], '', heads, None)
784
elif kind == 'directory':
785
if carry_over_possible:
788
# Nothing to set on the entry.
789
# XXX: split into the Root and nonRoot versions.
790
if change[1][1] != '' or self.repository.supports_rich_root():
791
self._add_text_to_weave(change[0], '', heads, None)
792
elif kind == 'tree-reference':
793
if not self.repository._format.supports_tree_reference:
794
# This isn't quite sane as an error, but we shouldn't
795
# ever see this code path in practice: tree's don't
796
# permit references when the repo doesn't support tree
798
raise errors.UnsupportedOperation(tree.add_reference,
800
reference_revision = tree.get_reference_revision(change[0])
801
entry.reference_revision = reference_revision
802
if (carry_over_possible and
803
parent_entry.reference_revision == reference_revision):
806
self._add_text_to_weave(change[0], '', heads, None)
808
raise AssertionError('unknown kind %r' % kind)
810
entry.revision = modified_rev
812
entry.revision = parent_entry.revision
815
new_path = change[1][1]
816
inv_delta.append((change[1][0], new_path, change[0], entry))
819
self.new_inventory = None
821
# This should perhaps be guarded by a check that the basis we
822
# commit against is the basis for the commit and if not do a delta
824
self._any_changes = True
826
# housekeeping root entry changes do not affect no-change commits.
827
self._require_root_change(tree)
828
self.basis_delta_revision = basis_revision_id
830
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
831
parent_keys = tuple([(file_id, parent) for parent in parents])
832
return self.repository.texts._add_text(
833
(file_id, self._new_revision_id), parent_keys, new_text,
834
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
837
class RootCommitBuilder(CommitBuilder):
838
"""This commitbuilder actually records the root id"""
840
# the root entry gets versioned properly by this builder.
841
_versioned_root = True
843
def _check_root(self, ie, parent_invs, tree):
844
"""Helper for record_entry_contents.
846
:param ie: An entry being added.
847
:param parent_invs: The inventories of the parent revisions of the
849
:param tree: The tree that is being committed.
852
def _require_root_change(self, tree):
853
"""Enforce an appropriate root object change.
855
This is called once when record_iter_changes is called, if and only if
856
the root was not in the delta calculated by record_iter_changes.
858
:param tree: The tree which is being committed.
860
# versioned roots do not change unless the tree found a change.
241
863
######################################################################
245
class Repository(_RelockDebugMixin, controldir.ControlComponent):
867
class Repository(_RelockDebugMixin, bzrdir.ControlComponent):
246
868
"""Repository holding history for one or more branches.
248
870
The repository holds and retrieves historical information including
249
871
revisions and file history. It's normally accessed only by the Branch,
250
872
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.
874
The Repository builds on top of some byte storage facilies (the revisions,
875
signatures, inventories, texts and chk_bytes attributes) and a Transport,
876
which respectively provide byte storage and a means to access the (possibly
879
The byte storage facilities are addressed via tuples, which we refer to
880
as 'keys' throughout the code base. Revision_keys, inventory_keys and
881
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
882
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
883
byte string made up of a hash identifier and a hash value.
884
We use this interface because it allows low friction with the underlying
885
code that implements disk indices, network encoding and other parts of
888
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
889
the serialised revisions for the repository. This can be used to obtain
890
revision graph information or to access raw serialised revisions.
891
The result of trying to insert data into the repository via this store
892
is undefined: it should be considered read-only except for implementors
894
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
895
the serialised signatures for the repository. This can be used to
896
obtain access to raw serialised signatures. The result of trying to
897
insert data into the repository via this store is undefined: it should
898
be considered read-only except for implementors of repositories.
899
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
900
the serialised inventories for the repository. This can be used to
901
obtain unserialised inventories. The result of trying to insert data
902
into the repository via this store is undefined: it should be
903
considered read-only except for implementors of repositories.
904
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
905
texts of files and directories for the repository. This can be used to
906
obtain file texts or file graphs. Note that Repository.iter_file_bytes
907
is usually a better interface for accessing file texts.
908
The result of trying to insert data into the repository via this store
909
is undefined: it should be considered read-only except for implementors
911
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
912
any data the repository chooses to store or have indexed by its hash.
913
The result of trying to insert data into the repository via this store
914
is undefined: it should be considered read-only except for implementors
916
:ivar _transport: Transport for file access to repository, typically
917
pointing to .bzr/repository.
920
# What class to use for a CommitBuilder. Often its simpler to change this
921
# in a Repository class subclass rather than to override
922
# get_commit_builder.
923
_commit_builder_class = CommitBuilder
924
# The search regex used by xml based repositories to determine what things
925
# where changed in a single commit.
926
_file_ids_altered_regex = lazy_regex.lazy_compile(
927
r'file_id="(?P<file_id>[^"]+)"'
928
r'.* revision="(?P<revision_id>[^"]+)"'
256
931
def abort_write_group(self, suppress_errors=False):
257
932
"""Commit the contents accrued within the current write group.
313
1001
return InterRepository._assert_same_model(self, repository)
1003
def add_inventory(self, revision_id, inv, parents):
1004
"""Add the inventory inv to the repository as revision_id.
1006
:param parents: The revision ids of the parents that revision_id
1007
is known to have and are in the repository already.
1009
:returns: The validator(which is a sha1 digest, though what is sha'd is
1010
repository format specific) of the serialized inventory.
1012
if not self.is_in_write_group():
1013
raise AssertionError("%r not in write group" % (self,))
1014
_mod_revision.check_not_reserved_id(revision_id)
1015
if not (inv.revision_id is None or inv.revision_id == revision_id):
1016
raise AssertionError(
1017
"Mismatch between inventory revision"
1018
" id and insertion revid (%r, %r)"
1019
% (inv.revision_id, revision_id))
1020
if inv.root is None:
1021
raise AssertionError()
1022
return self._add_inventory_checked(revision_id, inv, parents)
1024
def _add_inventory_checked(self, revision_id, inv, parents):
1025
"""Add inv to the repository after checking the inputs.
1027
This function can be overridden to allow different inventory styles.
1029
:seealso: add_inventory, for the contract.
1031
inv_lines = self._serializer.write_inventory_to_lines(inv)
1032
return self._inventory_add_lines(revision_id, parents,
1033
inv_lines, check_content=False)
1035
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1036
parents, basis_inv=None, propagate_caches=False):
1037
"""Add a new inventory expressed as a delta against another revision.
1039
See the inventory developers documentation for the theory behind
1042
:param basis_revision_id: The inventory id the delta was created
1043
against. (This does not have to be a direct parent.)
1044
:param delta: The inventory delta (see Inventory.apply_delta for
1046
:param new_revision_id: The revision id that the inventory is being
1048
:param parents: The revision ids of the parents that revision_id is
1049
known to have and are in the repository already. These are supplied
1050
for repositories that depend on the inventory graph for revision
1051
graph access, as well as for those that pun ancestry with delta
1053
:param basis_inv: The basis inventory if it is already known,
1055
:param propagate_caches: If True, the caches for this inventory are
1056
copied to and updated for the result if possible.
1058
:returns: (validator, new_inv)
1059
The validator(which is a sha1 digest, though what is sha'd is
1060
repository format specific) of the serialized inventory, and the
1061
resulting inventory.
1063
if not self.is_in_write_group():
1064
raise AssertionError("%r not in write group" % (self,))
1065
_mod_revision.check_not_reserved_id(new_revision_id)
1066
basis_tree = self.revision_tree(basis_revision_id)
1067
basis_tree.lock_read()
1069
# Note that this mutates the inventory of basis_tree, which not all
1070
# inventory implementations may support: A better idiom would be to
1071
# return a new inventory, but as there is no revision tree cache in
1072
# repository this is safe for now - RBC 20081013
1073
if basis_inv is None:
1074
basis_inv = basis_tree.inventory
1075
basis_inv.apply_delta(delta)
1076
basis_inv.revision_id = new_revision_id
1077
return (self.add_inventory(new_revision_id, basis_inv, parents),
1082
def _inventory_add_lines(self, revision_id, parents, lines,
1083
check_content=True):
1084
"""Store lines in inv_vf and return the sha1 of the inventory."""
1085
parents = [(parent,) for parent in parents]
1086
result = self.inventories.add_lines((revision_id,), parents, lines,
1087
check_content=check_content)[0]
1088
self.inventories._access.flush()
1091
def add_revision(self, revision_id, rev, inv=None, config=None):
1092
"""Add rev to the revision store as revision_id.
1094
:param revision_id: the revision id to use.
1095
:param rev: The revision object.
1096
:param inv: The inventory for the revision. if None, it will be looked
1097
up in the inventory storer
1098
:param config: If None no digital signature will be created.
1099
If supplied its signature_needed method will be used
1100
to determine if a signature should be made.
1102
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1104
_mod_revision.check_not_reserved_id(revision_id)
1105
if config is not None and config.signature_needed():
1107
inv = self.get_inventory(revision_id)
1108
plaintext = Testament(rev, inv).as_short_text()
1109
self.store_revision_signature(
1110
gpg.GPGStrategy(config), plaintext, revision_id)
1111
# check inventory present
1112
if not self.inventories.get_parent_map([(revision_id,)]):
1114
raise errors.WeaveRevisionNotPresent(revision_id,
1117
# yes, this is not suitable for adding with ghosts.
1118
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1121
key = (revision_id,)
1122
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1123
self._add_revision(rev)
1125
def _add_revision(self, revision):
1126
text = self._serializer.write_revision_to_string(revision)
1127
key = (revision.revision_id,)
1128
parents = tuple((parent,) for parent in revision.parent_ids)
1129
self.revisions.add_lines(key, parents, osutils.split_lines(text))
315
1131
def all_revision_ids(self):
316
1132
"""Returns a list of all the revision ids in the repository.
352
1168
# The old API returned a list, should this actually be a set?
353
1169
return parent_map.keys()
1171
def _check_inventories(self, checker):
1172
"""Check the inventories found from the revision scan.
1174
This is responsible for verifying the sha1 of inventories and
1175
creating a pending_keys set that covers data referenced by inventories.
1177
bar = ui.ui_factory.nested_progress_bar()
1179
self._do_check_inventories(checker, bar)
1183
def _do_check_inventories(self, checker, bar):
1184
"""Helper for _check_inventories."""
1186
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1187
kinds = ['chk_bytes', 'texts']
1188
count = len(checker.pending_keys)
1189
bar.update("inventories", 0, 2)
1190
current_keys = checker.pending_keys
1191
checker.pending_keys = {}
1192
# Accumulate current checks.
1193
for key in current_keys:
1194
if key[0] != 'inventories' and key[0] not in kinds:
1195
checker._report_items.append('unknown key type %r' % (key,))
1196
keys[key[0]].add(key[1:])
1197
if keys['inventories']:
1198
# NB: output order *should* be roughly sorted - topo or
1199
# inverse topo depending on repository - either way decent
1200
# to just delta against. However, pre-CHK formats didn't
1201
# try to optimise inventory layout on disk. As such the
1202
# pre-CHK code path does not use inventory deltas.
1204
for record in self.inventories.check(keys=keys['inventories']):
1205
if record.storage_kind == 'absent':
1206
checker._report_items.append(
1207
'Missing inventory {%s}' % (record.key,))
1209
last_object = self._check_record('inventories', record,
1210
checker, last_object,
1211
current_keys[('inventories',) + record.key])
1212
del keys['inventories']
1215
bar.update("texts", 1)
1216
while (checker.pending_keys or keys['chk_bytes']
1218
# Something to check.
1219
current_keys = checker.pending_keys
1220
checker.pending_keys = {}
1221
# Accumulate current checks.
1222
for key in current_keys:
1223
if key[0] not in kinds:
1224
checker._report_items.append('unknown key type %r' % (key,))
1225
keys[key[0]].add(key[1:])
1226
# Check the outermost kind only - inventories || chk_bytes || texts
1230
for record in getattr(self, kind).check(keys=keys[kind]):
1231
if record.storage_kind == 'absent':
1232
checker._report_items.append(
1233
'Missing %s {%s}' % (kind, record.key,))
1235
last_object = self._check_record(kind, record,
1236
checker, last_object, current_keys[(kind,) + record.key])
1240
def _check_record(self, kind, record, checker, last_object, item_data):
1241
"""Check a single text from this repository."""
1242
if kind == 'inventories':
1243
rev_id = record.key[0]
1244
inv = self._deserialise_inventory(rev_id,
1245
record.get_bytes_as('fulltext'))
1246
if last_object is not None:
1247
delta = inv._make_delta(last_object)
1248
for old_path, path, file_id, ie in delta:
1251
ie.check(checker, rev_id, inv)
1253
for path, ie in inv.iter_entries():
1254
ie.check(checker, rev_id, inv)
1255
if self._format.fast_deltas:
1257
elif kind == 'chk_bytes':
1258
# No code written to check chk_bytes for this repo format.
1259
checker._report_items.append(
1260
'unsupported key type chk_bytes for %s' % (record.key,))
1261
elif kind == 'texts':
1262
self._check_text(record, checker, item_data)
1264
checker._report_items.append(
1265
'unknown key type %s for %s' % (kind, record.key))
1267
def _check_text(self, record, checker, item_data):
1268
"""Check a single text."""
1269
# Check it is extractable.
1270
# TODO: check length.
1271
if record.storage_kind == 'chunked':
1272
chunks = record.get_bytes_as(record.storage_kind)
1273
sha1 = osutils.sha_strings(chunks)
1274
length = sum(map(len, chunks))
1276
content = record.get_bytes_as('fulltext')
1277
sha1 = osutils.sha_string(content)
1278
length = len(content)
1279
if item_data and sha1 != item_data[1]:
1280
checker._report_items.append(
1281
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1282
(record.key, sha1, item_data[1], item_data[2]))
356
1285
def create(a_bzrdir):
357
1286
"""Construct the current default format repository in a_bzrdir."""
921
1962
signature = gpg_strategy.sign(plaintext)
922
1963
self.add_signature_text(revision_id, signature)
924
1966
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)
1967
self.signatures.add_lines((revision_id,), (),
1968
osutils.split_lines(signature))
1970
def find_text_key_references(self):
1971
"""Find the text key references within the repository.
1973
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1974
to whether they were referred to by the inventory of the
1975
revision_id that they contain. The inventory texts from all present
1976
revision ids are assessed to generate this report.
1978
revision_keys = self.revisions.keys()
1979
w = self.inventories
1980
pb = ui.ui_factory.nested_progress_bar()
1982
return self._find_text_key_references_from_xml_inventory_lines(
1983
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1987
def _find_text_key_references_from_xml_inventory_lines(self,
1989
"""Core routine for extracting references to texts from inventories.
1991
This performs the translation of xml lines to revision ids.
1993
:param line_iterator: An iterator of lines, origin_version_id
1994
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1995
to whether they were referred to by the inventory of the
1996
revision_id that they contain. Note that if that revision_id was
1997
not part of the line_iterator's output then False will be given -
1998
even though it may actually refer to that key.
2000
if not self._serializer.support_altered_by_hack:
2001
raise AssertionError(
2002
"_find_text_key_references_from_xml_inventory_lines only "
2003
"supported for branches which store inventory as unnested xml"
2004
", not on %r" % self)
2007
# this code needs to read every new line in every inventory for the
2008
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2009
# not present in one of those inventories is unnecessary but not
2010
# harmful because we are filtering by the revision id marker in the
2011
# inventory lines : we only select file ids altered in one of those
2012
# revisions. We don't need to see all lines in the inventory because
2013
# only those added in an inventory in rev X can contain a revision=X
2015
unescape_revid_cache = {}
2016
unescape_fileid_cache = {}
2018
# jam 20061218 In a big fetch, this handles hundreds of thousands
2019
# of lines, so it has had a lot of inlining and optimizing done.
2020
# Sorry that it is a little bit messy.
2021
# Move several functions to be local variables, since this is a long
2023
search = self._file_ids_altered_regex.search
2024
unescape = _unescape_xml
2025
setdefault = result.setdefault
2026
for line, line_key in line_iterator:
2027
match = search(line)
2030
# One call to match.group() returning multiple items is quite a
2031
# bit faster than 2 calls to match.group() each returning 1
2032
file_id, revision_id = match.group('file_id', 'revision_id')
2034
# Inlining the cache lookups helps a lot when you make 170,000
2035
# lines and 350k ids, versus 8.4 unique ids.
2036
# Using a cache helps in 2 ways:
2037
# 1) Avoids unnecessary decoding calls
2038
# 2) Re-uses cached strings, which helps in future set and
2040
# (2) is enough that removing encoding entirely along with
2041
# the cache (so we are using plain strings) results in no
2042
# performance improvement.
2044
revision_id = unescape_revid_cache[revision_id]
2046
unescaped = unescape(revision_id)
2047
unescape_revid_cache[revision_id] = unescaped
2048
revision_id = unescaped
2050
# Note that unconditionally unescaping means that we deserialise
2051
# every fileid, which for general 'pull' is not great, but we don't
2052
# really want to have some many fulltexts that this matters anyway.
2055
file_id = unescape_fileid_cache[file_id]
2057
unescaped = unescape(file_id)
2058
unescape_fileid_cache[file_id] = unescaped
2061
key = (file_id, revision_id)
2062
setdefault(key, False)
2063
if revision_id == line_key[-1]:
2067
def _inventory_xml_lines_for_keys(self, keys):
2068
"""Get a line iterator of the sort needed for findind references.
2070
Not relevant for non-xml inventory repositories.
2072
Ghosts in revision_keys are ignored.
2074
:param revision_keys: The revision keys for the inventories to inspect.
2075
:return: An iterator over (inventory line, revid) for the fulltexts of
2076
all of the xml inventories specified by revision_keys.
2078
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2079
for record in stream:
2080
if record.storage_kind != 'absent':
2081
chunks = record.get_bytes_as('chunked')
2082
revid = record.key[-1]
2083
lines = osutils.chunks_to_lines(chunks)
2087
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2089
"""Helper routine for fileids_altered_by_revision_ids.
2091
This performs the translation of xml lines to revision ids.
2093
:param line_iterator: An iterator of lines, origin_version_id
2094
:param revision_keys: The revision ids to filter for. This should be a
2095
set or other type which supports efficient __contains__ lookups, as
2096
the revision key from each parsed line will be looked up in the
2097
revision_keys filter.
2098
:return: a dictionary mapping altered file-ids to an iterable of
2099
revision_ids. Each altered file-ids has the exact revision_ids that
2100
altered it listed explicitly.
2102
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2103
line_iterator).iterkeys())
2104
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2105
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2106
self._inventory_xml_lines_for_keys(parent_keys)))
2107
new_keys = seen - parent_seen
2109
setdefault = result.setdefault
2110
for key in new_keys:
2111
setdefault(key[0], set()).add(key[-1])
932
2114
def _find_parent_ids_of_revisions(self, revision_ids):
933
2115
"""Find all parent ids that are mentioned in the revision graph.
963
2166
uniquely identify the file version in the caller's context. (Examples:
964
2167
an index number or a TreeTransform trans_id.)
2169
bytes_iterator is an iterable of bytestrings for the file. The
2170
kind of iterable and length of the bytestrings are unspecified, but for
2171
this implementation, it is a list of bytes produced by
2172
VersionedFile.get_record_stream().
966
2174
:param desired_files: a list of (file_id, revision_id, identifier)
969
raise NotImplementedError(self.iter_files_bytes)
2178
for file_id, revision_id, callable_data in desired_files:
2179
text_keys[(file_id, revision_id)] = callable_data
2180
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2181
if record.storage_kind == 'absent':
2182
raise errors.RevisionNotPresent(record.key, self)
2183
yield text_keys[record.key], record.get_bytes_as('chunked')
2185
def _generate_text_key_index(self, text_key_references=None,
2187
"""Generate a new text key index for the repository.
2189
This is an expensive function that will take considerable time to run.
2191
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2192
list of parents, also text keys. When a given key has no parents,
2193
the parents list will be [NULL_REVISION].
2195
# All revisions, to find inventory parents.
2196
if ancestors is None:
2197
graph = self.get_graph()
2198
ancestors = graph.get_parent_map(self.all_revision_ids())
2199
if text_key_references is None:
2200
text_key_references = self.find_text_key_references()
2201
pb = ui.ui_factory.nested_progress_bar()
2203
return self._do_generate_text_key_index(ancestors,
2204
text_key_references, pb)
2208
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2209
"""Helper for _generate_text_key_index to avoid deep nesting."""
2210
revision_order = tsort.topo_sort(ancestors)
2211
invalid_keys = set()
2213
for revision_id in revision_order:
2214
revision_keys[revision_id] = set()
2215
text_count = len(text_key_references)
2216
# a cache of the text keys to allow reuse; costs a dict of all the
2217
# keys, but saves a 2-tuple for every child of a given key.
2219
for text_key, valid in text_key_references.iteritems():
2221
invalid_keys.add(text_key)
2223
revision_keys[text_key[1]].add(text_key)
2224
text_key_cache[text_key] = text_key
2225
del text_key_references
2227
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2228
NULL_REVISION = _mod_revision.NULL_REVISION
2229
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2230
# too small for large or very branchy trees. However, for 55K path
2231
# trees, it would be easy to use too much memory trivially. Ideally we
2232
# could gauge this by looking at available real memory etc, but this is
2233
# always a tricky proposition.
2234
inventory_cache = lru_cache.LRUCache(10)
2235
batch_size = 10 # should be ~150MB on a 55K path tree
2236
batch_count = len(revision_order) / batch_size + 1
2238
pb.update("Calculating text parents", processed_texts, text_count)
2239
for offset in xrange(batch_count):
2240
to_query = revision_order[offset * batch_size:(offset + 1) *
2244
for revision_id in to_query:
2245
parent_ids = ancestors[revision_id]
2246
for text_key in revision_keys[revision_id]:
2247
pb.update("Calculating text parents", processed_texts)
2248
processed_texts += 1
2249
candidate_parents = []
2250
for parent_id in parent_ids:
2251
parent_text_key = (text_key[0], parent_id)
2253
check_parent = parent_text_key not in \
2254
revision_keys[parent_id]
2256
# the parent parent_id is a ghost:
2257
check_parent = False
2258
# truncate the derived graph against this ghost.
2259
parent_text_key = None
2261
# look at the parent commit details inventories to
2262
# determine possible candidates in the per file graph.
2265
inv = inventory_cache[parent_id]
2267
inv = self.revision_tree(parent_id).inventory
2268
inventory_cache[parent_id] = inv
2270
parent_entry = inv[text_key[0]]
2271
except (KeyError, errors.NoSuchId):
2273
if parent_entry is not None:
2275
text_key[0], parent_entry.revision)
2277
parent_text_key = None
2278
if parent_text_key is not None:
2279
candidate_parents.append(
2280
text_key_cache[parent_text_key])
2281
parent_heads = text_graph.heads(candidate_parents)
2282
new_parents = list(parent_heads)
2283
new_parents.sort(key=lambda x:candidate_parents.index(x))
2284
if new_parents == []:
2285
new_parents = [NULL_REVISION]
2286
text_index[text_key] = new_parents
2288
for text_key in invalid_keys:
2289
text_index[text_key] = [NULL_REVISION]
2292
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2293
"""Get an iterable listing the keys of all the data introduced by a set
2296
The keys will be ordered so that the corresponding items can be safely
2297
fetched and inserted in that order.
2299
:returns: An iterable producing tuples of (knit-kind, file-id,
2300
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2301
'revisions'. file-id is None unless knit-kind is 'file'.
2303
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2306
for result in self._find_non_file_keys_to_fetch(revision_ids):
2309
def _find_file_keys_to_fetch(self, revision_ids, pb):
2310
# XXX: it's a bit weird to control the inventory weave caching in this
2311
# generator. Ideally the caching would be done in fetch.py I think. Or
2312
# maybe this generator should explicitly have the contract that it
2313
# should not be iterated until the previously yielded item has been
2315
inv_w = self.inventories
2317
# file ids that changed
2318
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2320
num_file_ids = len(file_ids)
2321
for file_id, altered_versions in file_ids.iteritems():
2323
pb.update("Fetch texts", count, num_file_ids)
2325
yield ("file", file_id, altered_versions)
2327
def _find_non_file_keys_to_fetch(self, revision_ids):
2329
yield ("inventory", None, revision_ids)
2332
# XXX: Note ATM no callers actually pay attention to this return
2333
# instead they just use the list of revision ids and ignore
2334
# missing sigs. Consider removing this work entirely
2335
revisions_with_signatures = set(self.signatures.get_parent_map(
2336
[(r,) for r in revision_ids]))
2337
revisions_with_signatures = set(
2338
[r for (r,) in revisions_with_signatures])
2339
revisions_with_signatures.intersection_update(revision_ids)
2340
yield ("signatures", None, revisions_with_signatures)
2343
yield ("revisions", None, revision_ids)
2346
def get_inventory(self, revision_id):
2347
"""Get Inventory object by revision id."""
2348
return self.iter_inventories([revision_id]).next()
2350
def iter_inventories(self, revision_ids, ordering=None):
2351
"""Get many inventories by revision_ids.
2353
This will buffer some or all of the texts used in constructing the
2354
inventories in memory, but will only parse a single inventory at a
2357
:param revision_ids: The expected revision ids of the inventories.
2358
:param ordering: optional ordering, e.g. 'topological'. If not
2359
specified, the order of revision_ids will be preserved (by
2360
buffering if necessary).
2361
:return: An iterator of inventories.
2363
if ((None in revision_ids)
2364
or (_mod_revision.NULL_REVISION in revision_ids)):
2365
raise ValueError('cannot get null revision inventory')
2366
return self._iter_inventories(revision_ids, ordering)
2368
def _iter_inventories(self, revision_ids, ordering):
2369
"""single-document based inventory iteration."""
2370
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2371
for text, revision_id in inv_xmls:
2372
yield self._deserialise_inventory(revision_id, text)
2374
def _iter_inventory_xmls(self, revision_ids, ordering):
2375
if ordering is None:
2376
order_as_requested = True
2377
ordering = 'unordered'
2379
order_as_requested = False
2380
keys = [(revision_id,) for revision_id in revision_ids]
2383
if order_as_requested:
2384
key_iter = iter(keys)
2385
next_key = key_iter.next()
2386
stream = self.inventories.get_record_stream(keys, ordering, True)
2388
for record in stream:
2389
if record.storage_kind != 'absent':
2390
chunks = record.get_bytes_as('chunked')
2391
if order_as_requested:
2392
text_chunks[record.key] = chunks
2394
yield ''.join(chunks), record.key[-1]
2396
raise errors.NoSuchRevision(self, record.key)
2397
if order_as_requested:
2398
# Yield as many results as we can while preserving order.
2399
while next_key in text_chunks:
2400
chunks = text_chunks.pop(next_key)
2401
yield ''.join(chunks), next_key[-1]
2403
next_key = key_iter.next()
2404
except StopIteration:
2405
# We still want to fully consume the get_record_stream,
2406
# just in case it is not actually finished at this point
2410
def _deserialise_inventory(self, revision_id, xml):
2411
"""Transform the xml into an inventory object.
2413
:param revision_id: The expected revision id of the inventory.
2414
:param xml: A serialised inventory.
2416
result = self._serializer.read_inventory_from_string(xml, revision_id,
2417
entry_cache=self._inventory_entry_cache,
2418
return_from_cache=self._safe_to_return_from_cache)
2419
if result.revision_id != revision_id:
2420
raise AssertionError('revision id mismatch %s != %s' % (
2421
result.revision_id, revision_id))
2424
def get_serializer_format(self):
2425
return self._serializer.format_num
2428
def _get_inventory_xml(self, revision_id):
2429
"""Get serialized inventory as a string."""
2430
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2432
text, revision_id = texts.next()
2433
except StopIteration:
2434
raise errors.HistoryMissing(self, 'inventory', revision_id)
971
2437
def get_rev_id_for_revno(self, revno, known_pair):
972
2438
"""Return the revision id of a revno, given a later (revno, revid)
1289
2782
except UnicodeDecodeError:
1290
2783
raise errors.NonAsciiRevisionId(method, self)
2785
def revision_graph_can_have_wrong_parents(self):
2786
"""Is it possible for this repository to have a revision graph with
2789
If True, then this repository must also implement
2790
_find_inconsistent_revision_parents so that check and reconcile can
2791
check for inconsistencies before proceeding with other checks that may
2792
depend on the revision index being consistent.
2794
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2797
# remove these delegates a while after bzr 0.15
2798
def __make_delegated(name, from_module):
2799
def _deprecated_repository_forwarder():
2800
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2801
% (name, from_module),
2804
m = __import__(from_module, globals(), locals(), [name])
2806
return getattr(m, name)
2807
except AttributeError:
2808
raise AttributeError('module %s has no name %s'
2810
globals()[name] = _deprecated_repository_forwarder
2813
'AllInOneRepository',
2814
'WeaveMetaDirRepository',
2815
'PreSplitOutRepositoryFormat',
2816
'RepositoryFormat4',
2817
'RepositoryFormat5',
2818
'RepositoryFormat6',
2819
'RepositoryFormat7',
2821
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2825
'RepositoryFormatKnit',
2826
'RepositoryFormatKnit1',
2828
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2831
def install_revision(repository, rev, revision_tree):
2832
"""Install all revision data into a repository."""
2833
install_revisions(repository, [(rev, revision_tree, None)])
2836
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2837
"""Install all revision data into a repository.
2839
Accepts an iterable of revision, tree, signature tuples. The signature
2842
repository.start_write_group()
2844
inventory_cache = lru_cache.LRUCache(10)
2845
for n, (revision, revision_tree, signature) in enumerate(iterable):
2846
_install_revision(repository, revision, revision_tree, signature,
2849
pb.update('Transferring revisions', n + 1, num_revisions)
2851
repository.abort_write_group()
2854
repository.commit_write_group()
2857
def _install_revision(repository, rev, revision_tree, signature,
2859
"""Install all revision data into a repository."""
2860
present_parents = []
2862
for p_id in rev.parent_ids:
2863
if repository.has_revision(p_id):
2864
present_parents.append(p_id)
2865
parent_trees[p_id] = repository.revision_tree(p_id)
2867
parent_trees[p_id] = repository.revision_tree(
2868
_mod_revision.NULL_REVISION)
2870
inv = revision_tree.inventory
2871
entries = inv.iter_entries()
2872
# backwards compatibility hack: skip the root id.
2873
if not repository.supports_rich_root():
2874
path, root = entries.next()
2875
if root.revision != rev.revision_id:
2876
raise errors.IncompatibleRevision(repr(repository))
2878
for path, ie in entries:
2879
text_keys[(ie.file_id, ie.revision)] = ie
2880
text_parent_map = repository.texts.get_parent_map(text_keys)
2881
missing_texts = set(text_keys) - set(text_parent_map)
2882
# Add the texts that are not already present
2883
for text_key in missing_texts:
2884
ie = text_keys[text_key]
2886
# FIXME: TODO: The following loop overlaps/duplicates that done by
2887
# commit to determine parents. There is a latent/real bug here where
2888
# the parents inserted are not those commit would do - in particular
2889
# they are not filtered by heads(). RBC, AB
2890
for revision, tree in parent_trees.iteritems():
2891
if ie.file_id not in tree:
2893
parent_id = tree.inventory[ie.file_id].revision
2894
if parent_id in text_parents:
2896
text_parents.append((ie.file_id, parent_id))
2897
lines = revision_tree.get_file(ie.file_id).readlines()
2898
repository.texts.add_lines(text_key, text_parents, lines)
2900
# install the inventory
2901
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2902
# Cache this inventory
2903
inventory_cache[rev.revision_id] = inv
2905
basis_inv = inventory_cache[rev.parent_ids[0]]
2907
repository.add_inventory(rev.revision_id, inv, present_parents)
2909
delta = inv._make_delta(basis_inv)
2910
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2911
rev.revision_id, present_parents)
2913
repository.add_inventory(rev.revision_id, inv, present_parents)
2914
except errors.RevisionAlreadyPresent:
2916
if signature is not None:
2917
repository.add_signature_text(rev.revision_id, signature)
2918
repository.add_revision(rev.revision_id, rev, inv)
1293
2921
class MetaDirRepository(Repository):
1294
2922
"""Repositories in the new meta-dir layout.
1633
3287
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
1634
3288
format_registry.register_lazy(
1635
3289
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
1636
'bzrlib.repofmt.knitpack_repo',
3290
'bzrlib.repofmt.pack_repo',
1637
3291
'RepositoryFormatKnitPack1',
1639
3293
format_registry.register_lazy(
1640
3294
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
1641
'bzrlib.repofmt.knitpack_repo',
3295
'bzrlib.repofmt.pack_repo',
1642
3296
'RepositoryFormatKnitPack3',
1644
3298
format_registry.register_lazy(
1645
3299
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
1646
'bzrlib.repofmt.knitpack_repo',
3300
'bzrlib.repofmt.pack_repo',
1647
3301
'RepositoryFormatKnitPack4',
1649
3303
format_registry.register_lazy(
1650
3304
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
1651
'bzrlib.repofmt.knitpack_repo',
3305
'bzrlib.repofmt.pack_repo',
1652
3306
'RepositoryFormatKnitPack5',
1654
3308
format_registry.register_lazy(
1655
3309
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
1656
'bzrlib.repofmt.knitpack_repo',
3310
'bzrlib.repofmt.pack_repo',
1657
3311
'RepositoryFormatKnitPack5RichRoot',
1659
3313
format_registry.register_lazy(
1660
3314
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
1661
'bzrlib.repofmt.knitpack_repo',
3315
'bzrlib.repofmt.pack_repo',
1662
3316
'RepositoryFormatKnitPack5RichRootBroken',
1664
3318
format_registry.register_lazy(
1665
3319
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
1666
'bzrlib.repofmt.knitpack_repo',
3320
'bzrlib.repofmt.pack_repo',
1667
3321
'RepositoryFormatKnitPack6',
1669
3323
format_registry.register_lazy(
1670
3324
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
1671
'bzrlib.repofmt.knitpack_repo',
3325
'bzrlib.repofmt.pack_repo',
1672
3326
'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
3329
# Development formats.
1681
# Check their docstrings to see if/when they are obsolete.
3330
# Obsolete but kept pending a CHK based subtree format.
1682
3331
format_registry.register_lazy(
1683
3332
("Bazaar development format 2 with subtree support "
1684
3333
"(needs bzr.dev from before 1.8)\n"),
1685
'bzrlib.repofmt.knitpack_repo',
3334
'bzrlib.repofmt.pack_repo',
1686
3335
'RepositoryFormatPackDevelopment2Subtree',
1688
format_registry.register_lazy(
1689
'Bazaar development format 8\n',
1690
'bzrlib.repofmt.groupcompress_repo',
1691
'RepositoryFormat2aSubtree',
3338
# 1.14->1.16 go below here
3339
format_registry.register_lazy(
3340
'Bazaar development format - group compression and chk inventory'
3341
' (needs bzr.dev from 1.14)\n',
3342
'bzrlib.repofmt.groupcompress_repo',
3343
'RepositoryFormatCHK1',
3346
format_registry.register_lazy(
3347
'Bazaar development format - chk repository with bencode revision '
3348
'serialization (needs bzr.dev from 1.16)\n',
3349
'bzrlib.repofmt.groupcompress_repo',
3350
'RepositoryFormatCHK2',
3352
format_registry.register_lazy(
3353
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3354
'bzrlib.repofmt.groupcompress_repo',
3355
'RepositoryFormat2a',
1733
3398
:param revision_id: if None all content is copied, if NULL_REVISION no
1734
3399
content is copied.
1737
raise NotImplementedError(self.fetch)
3403
ui.ui_factory.warn_experimental_format_fetch(self)
3404
from bzrlib.fetch import RepoFetcher
3405
# See <https://launchpad.net/bugs/456077> asking for a warning here
3406
if self.source._format.network_name() != self.target._format.network_name():
3407
ui.ui_factory.show_user_warning('cross_format_fetch',
3408
from_format=self.source._format,
3409
to_format=self.target._format)
3410
f = RepoFetcher(to_repository=self.target,
3411
from_repository=self.source,
3412
last_revision=revision_id,
3413
fetch_spec=fetch_spec,
3414
find_ghosts=find_ghosts)
3416
def _walk_to_common_revisions(self, revision_ids):
3417
"""Walk out from revision_ids in source to revisions target has.
3419
:param revision_ids: The start point for the search.
3420
:return: A set of revision ids.
3422
target_graph = self.target.get_graph()
3423
revision_ids = frozenset(revision_ids)
3424
missing_revs = set()
3425
source_graph = self.source.get_graph()
3426
# ensure we don't pay silly lookup costs.
3427
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3428
null_set = frozenset([_mod_revision.NULL_REVISION])
3429
searcher_exhausted = False
3433
# Iterate the searcher until we have enough next_revs
3434
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3436
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3437
next_revs.update(next_revs_part)
3438
ghosts.update(ghosts_part)
3439
except StopIteration:
3440
searcher_exhausted = True
3442
# If there are ghosts in the source graph, and the caller asked for
3443
# them, make sure that they are present in the target.
3444
# We don't care about other ghosts as we can't fetch them and
3445
# haven't been asked to.
3446
ghosts_to_check = set(revision_ids.intersection(ghosts))
3447
revs_to_get = set(next_revs).union(ghosts_to_check)
3449
have_revs = set(target_graph.get_parent_map(revs_to_get))
3450
# we always have NULL_REVISION present.
3451
have_revs = have_revs.union(null_set)
3452
# Check if the target is missing any ghosts we need.
3453
ghosts_to_check.difference_update(have_revs)
3455
# One of the caller's revision_ids is a ghost in both the
3456
# source and the target.
3457
raise errors.NoSuchRevision(
3458
self.source, ghosts_to_check.pop())
3459
missing_revs.update(next_revs - have_revs)
3460
# Because we may have walked past the original stop point, make
3461
# sure everything is stopped
3462
stop_revs = searcher.find_seen_ancestors(have_revs)
3463
searcher.stop_searching_any(stop_revs)
3464
if searcher_exhausted:
3466
return searcher.get_result()
1739
3468
@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,
3469
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1744
3470
"""Return the revision ids that source has that target does not.
1746
3472
: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
3474
:param find_ghosts: If True find missing revisions in deep history
1756
3475
rather than just finding the surface difference.
1757
:param limit: Maximum number of revisions to return, topologically
1759
3476
:return: A bzrlib.graph.SearchResult.
1761
raise NotImplementedError(self.search_missing_revision_ids)
3478
# stop searching at found target revisions.
3479
if not find_ghosts and revision_id is not None:
3480
return self._walk_to_common_revisions([revision_id])
3481
# generic, possibly worst case, slow code path.
3482
target_ids = set(self.target.all_revision_ids())
3483
if revision_id is not None:
3484
source_ids = self.source.get_ancestry(revision_id)
3485
if source_ids[0] is not None:
3486
raise AssertionError()
3489
source_ids = self.source.all_revision_ids()
3490
result_set = set(source_ids).difference(target_ids)
3491
return self.source.revision_ids_to_search_result(result_set)
1764
3494
def _same_model(source, target):
1785
3515
"different serializers")
3518
class InterSameDataRepository(InterRepository):
3519
"""Code for converting between repositories that represent the same data.
3521
Data format and model must match for this to work.
3525
def _get_repo_format_to_test(self):
3526
"""Repository format for testing with.
3528
InterSameData can pull from subtree to subtree and from non-subtree to
3529
non-subtree, so we test this with the richest repository format.
3531
from bzrlib.repofmt import knitrepo
3532
return knitrepo.RepositoryFormatKnit3()
3535
def is_compatible(source, target):
3536
return InterRepository._same_model(source, target)
3539
class InterWeaveRepo(InterSameDataRepository):
3540
"""Optimised code paths between Weave based repositories.
3542
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3543
implemented lazy inter-object optimisation.
3547
def _get_repo_format_to_test(self):
3548
from bzrlib.repofmt import weaverepo
3549
return weaverepo.RepositoryFormat7()
3552
def is_compatible(source, target):
3553
"""Be compatible with known Weave formats.
3555
We don't test for the stores being of specific types because that
3556
could lead to confusing results, and there is no need to be
3559
from bzrlib.repofmt.weaverepo import (
3565
return (isinstance(source._format, (RepositoryFormat5,
3567
RepositoryFormat7)) and
3568
isinstance(target._format, (RepositoryFormat5,
3570
RepositoryFormat7)))
3571
except AttributeError:
3575
def copy_content(self, revision_id=None):
3576
"""See InterRepository.copy_content()."""
3577
# weave specific optimised path:
3579
self.target.set_make_working_trees(self.source.make_working_trees())
3580
except (errors.RepositoryUpgradeRequired, NotImplemented):
3582
# FIXME do not peek!
3583
if self.source._transport.listable():
3584
pb = ui.ui_factory.nested_progress_bar()
3586
self.target.texts.insert_record_stream(
3587
self.source.texts.get_record_stream(
3588
self.source.texts.keys(), 'topological', False))
3589
pb.update('Copying inventory', 0, 1)
3590
self.target.inventories.insert_record_stream(
3591
self.source.inventories.get_record_stream(
3592
self.source.inventories.keys(), 'topological', False))
3593
self.target.signatures.insert_record_stream(
3594
self.source.signatures.get_record_stream(
3595
self.source.signatures.keys(),
3597
self.target.revisions.insert_record_stream(
3598
self.source.revisions.get_record_stream(
3599
self.source.revisions.keys(),
3600
'topological', True))
3604
self.target.fetch(self.source, revision_id=revision_id)
3607
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3608
"""See InterRepository.missing_revision_ids()."""
3609
# we want all revisions to satisfy revision_id in source.
3610
# but we don't want to stat every file here and there.
3611
# we want then, all revisions other needs to satisfy revision_id
3612
# checked, but not those that we have locally.
3613
# so the first thing is to get a subset of the revisions to
3614
# satisfy revision_id in source, and then eliminate those that
3615
# we do already have.
3616
# this is slow on high latency connection to self, but as this
3617
# disk format scales terribly for push anyway due to rewriting
3618
# inventory.weave, this is considered acceptable.
3620
if revision_id is not None:
3621
source_ids = self.source.get_ancestry(revision_id)
3622
if source_ids[0] is not None:
3623
raise AssertionError()
3626
source_ids = self.source._all_possible_ids()
3627
source_ids_set = set(source_ids)
3628
# source_ids is the worst possible case we may need to pull.
3629
# now we want to filter source_ids against what we actually
3630
# have in target, but don't try to check for existence where we know
3631
# we do not have a revision as that would be pointless.
3632
target_ids = set(self.target._all_possible_ids())
3633
possibly_present_revisions = target_ids.intersection(source_ids_set)
3634
actually_present_revisions = set(
3635
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3636
required_revisions = source_ids_set.difference(actually_present_revisions)
3637
if revision_id is not None:
3638
# we used get_ancestry to determine source_ids then we are assured all
3639
# revisions referenced are present as they are installed in topological order.
3640
# and the tip revision was validated by get_ancestry.
3641
result_set = required_revisions
3643
# if we just grabbed the possibly available ids, then
3644
# we only have an estimate of whats available and need to validate
3645
# that against the revision records.
3647
self.source._eliminate_revisions_not_present(required_revisions))
3648
return self.source.revision_ids_to_search_result(result_set)
3651
class InterKnitRepo(InterSameDataRepository):
3652
"""Optimised code paths between Knit based repositories."""
3655
def _get_repo_format_to_test(self):
3656
from bzrlib.repofmt import knitrepo
3657
return knitrepo.RepositoryFormatKnit1()
3660
def is_compatible(source, target):
3661
"""Be compatible with known Knit formats.
3663
We don't test for the stores being of specific types because that
3664
could lead to confusing results, and there is no need to be
3667
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3669
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3670
isinstance(target._format, RepositoryFormatKnit))
3671
except AttributeError:
3673
return are_knits and InterRepository._same_model(source, target)
3676
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3677
"""See InterRepository.missing_revision_ids()."""
3678
if revision_id is not None:
3679
source_ids = self.source.get_ancestry(revision_id)
3680
if source_ids[0] is not None:
3681
raise AssertionError()
3684
source_ids = self.source.all_revision_ids()
3685
source_ids_set = set(source_ids)
3686
# source_ids is the worst possible case we may need to pull.
3687
# now we want to filter source_ids against what we actually
3688
# have in target, but don't try to check for existence where we know
3689
# we do not have a revision as that would be pointless.
3690
target_ids = set(self.target.all_revision_ids())
3691
possibly_present_revisions = target_ids.intersection(source_ids_set)
3692
actually_present_revisions = set(
3693
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3694
required_revisions = source_ids_set.difference(actually_present_revisions)
3695
if revision_id is not None:
3696
# we used get_ancestry to determine source_ids then we are assured all
3697
# revisions referenced are present as they are installed in topological order.
3698
# and the tip revision was validated by get_ancestry.
3699
result_set = required_revisions
3701
# if we just grabbed the possibly available ids, then
3702
# we only have an estimate of whats available and need to validate
3703
# that against the revision records.
3705
self.source._eliminate_revisions_not_present(required_revisions))
3706
return self.source.revision_ids_to_search_result(result_set)
3709
class InterDifferingSerializer(InterRepository):
3712
def _get_repo_format_to_test(self):
3716
def is_compatible(source, target):
3717
"""Be compatible with Knit2 source and Knit3 target"""
3718
# This is redundant with format.check_conversion_target(), however that
3719
# raises an exception, and we just want to say "False" as in we won't
3720
# support converting between these formats.
3721
if 'IDS_never' in debug.debug_flags:
3723
if source.supports_rich_root() and not target.supports_rich_root():
3725
if (source._format.supports_tree_reference
3726
and not target._format.supports_tree_reference):
3728
if target._fallback_repositories and target._format.supports_chks:
3729
# IDS doesn't know how to copy CHKs for the parent inventories it
3730
# adds to stacked repos.
3732
if 'IDS_always' in debug.debug_flags:
3734
# Only use this code path for local source and target. IDS does far
3735
# too much IO (both bandwidth and roundtrips) over a network.
3736
if not source.bzrdir.transport.base.startswith('file:///'):
3738
if not target.bzrdir.transport.base.startswith('file:///'):
3742
def _get_trees(self, revision_ids, cache):
3744
for rev_id in revision_ids:
3746
possible_trees.append((rev_id, cache[rev_id]))
3748
# Not cached, but inventory might be present anyway.
3750
tree = self.source.revision_tree(rev_id)
3751
except errors.NoSuchRevision:
3752
# Nope, parent is ghost.
3755
cache[rev_id] = tree
3756
possible_trees.append((rev_id, tree))
3757
return possible_trees
3759
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3760
"""Get the best delta and base for this revision.
3762
:return: (basis_id, delta)
3765
# Generate deltas against each tree, to find the shortest.
3766
texts_possibly_new_in_tree = set()
3767
for basis_id, basis_tree in possible_trees:
3768
delta = tree.inventory._make_delta(basis_tree.inventory)
3769
for old_path, new_path, file_id, new_entry in delta:
3770
if new_path is None:
3771
# This file_id isn't present in the new rev, so we don't
3775
# Rich roots are handled elsewhere...
3777
kind = new_entry.kind
3778
if kind != 'directory' and kind != 'file':
3779
# No text record associated with this inventory entry.
3781
# This is a directory or file that has changed somehow.
3782
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3783
deltas.append((len(delta), basis_id, delta))
3785
return deltas[0][1:]
3787
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3788
"""Find all parent revisions that are absent, but for which the
3789
inventory is present, and copy those inventories.
3791
This is necessary to preserve correctness when the source is stacked
3792
without fallbacks configured. (Note that in cases like upgrade the
3793
source may be not have _fallback_repositories even though it is
3797
for parents in parent_map.values():
3798
parent_revs.update(parents)
3799
present_parents = self.source.get_parent_map(parent_revs)
3800
absent_parents = set(parent_revs).difference(present_parents)
3801
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3802
(rev_id,) for rev_id in absent_parents)
3803
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3804
for parent_tree in self.source.revision_trees(parent_inv_ids):
3805
current_revision_id = parent_tree.get_revision_id()
3806
parents_parents_keys = parent_invs_keys_for_stacking[
3807
(current_revision_id,)]
3808
parents_parents = [key[-1] for key in parents_parents_keys]
3809
basis_id = _mod_revision.NULL_REVISION
3810
basis_tree = self.source.revision_tree(basis_id)
3811
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3812
self.target.add_inventory_by_delta(
3813
basis_id, delta, current_revision_id, parents_parents)
3814
cache[current_revision_id] = parent_tree
3816
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3817
"""Fetch across a few revisions.
3819
:param revision_ids: The revisions to copy
3820
:param basis_id: The revision_id of a tree that must be in cache, used
3821
as a basis for delta when no other base is available
3822
:param cache: A cache of RevisionTrees that we can use.
3823
:param a_graph: A Graph object to determine the heads() of the
3824
rich-root data stream.
3825
:return: The revision_id of the last converted tree. The RevisionTree
3826
for it will be in cache
3828
# Walk though all revisions; get inventory deltas, copy referenced
3829
# texts that delta references, insert the delta, revision and
3831
root_keys_to_create = set()
3834
pending_revisions = []
3835
parent_map = self.source.get_parent_map(revision_ids)
3836
self._fetch_parent_invs_for_stacking(parent_map, cache)
3837
self.source._safe_to_return_from_cache = True
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
3891
self.source._safe_to_return_from_cache = False
3893
from_texts = self.source.texts
3894
to_texts = self.target.texts
3895
if root_keys_to_create:
3896
root_stream = _mod_fetch._new_root_data_stream(
3897
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3898
self.source, graph=a_graph)
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
if self._converting_to_rich_root and len(revision_ids) > 100:
3962
a_graph = _mod_fetch._get_rich_root_heads_graph(self.source,
3967
for offset in range(0, len(revision_ids), batch_size):
3968
self.target.start_write_group()
3970
pb.update('Transferring revisions', offset,
3972
batch = revision_ids[offset:offset+batch_size]
3973
basis_id = self._fetch_batch(batch, basis_id, cache,
3976
self.source._safe_to_return_from_cache = False
3977
self.target.abort_write_group()
3980
hint = self.target.commit_write_group()
3983
if hints and self.target._format.pack_compresses:
3984
self.target.pack(hint=hints)
3985
pb.update('Transferring revisions', len(revision_ids),
3989
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3991
"""See InterRepository.fetch()."""
3992
if fetch_spec is not None:
3993
raise AssertionError("Not implemented yet...")
3994
ui.ui_factory.warn_experimental_format_fetch(self)
3995
if (not self.source.supports_rich_root()
3996
and self.target.supports_rich_root()):
3997
self._converting_to_rich_root = True
3998
self._revision_id_to_root_id = {}
4000
self._converting_to_rich_root = False
4001
# See <https://launchpad.net/bugs/456077> asking for a warning here
4002
if self.source._format.network_name() != self.target._format.network_name():
4003
ui.ui_factory.show_user_warning('cross_format_fetch',
4004
from_format=self.source._format,
4005
to_format=self.target._format)
4006
revision_ids = self.target.search_missing_revision_ids(self.source,
4007
revision_id, find_ghosts=find_ghosts).get_keys()
4008
if not revision_ids:
4010
revision_ids = tsort.topo_sort(
4011
self.source.get_graph().get_parent_map(revision_ids))
4012
if not revision_ids:
4014
# Walk though all revisions; get inventory deltas, copy referenced
4015
# texts that delta references, insert the delta, revision and
4018
my_pb = ui.ui_factory.nested_progress_bar()
4021
symbol_versioning.warn(
4022
symbol_versioning.deprecated_in((1, 14, 0))
4023
% "pb parameter to fetch()")
4026
self._fetch_all_revisions(revision_ids, pb)
4028
if my_pb is not None:
4030
return len(revision_ids), 0
4032
def _get_basis(self, first_revision_id):
4033
"""Get a revision and tree which exists in the target.
4035
This assumes that first_revision_id is selected for transmission
4036
because all other ancestors are already present. If we can't find an
4037
ancestor we fall back to NULL_REVISION since we know that is safe.
4039
:return: (basis_id, basis_tree)
4041
first_rev = self.source.get_revision(first_revision_id)
4043
basis_id = first_rev.parent_ids[0]
4044
# only valid as a basis if the target has it
4045
self.target.get_revision(basis_id)
4046
# Try to get a basis tree - if its a ghost it will hit the
4047
# NoSuchRevision case.
4048
basis_tree = self.source.revision_tree(basis_id)
4049
except (IndexError, errors.NoSuchRevision):
4050
basis_id = _mod_revision.NULL_REVISION
4051
basis_tree = self.source.revision_tree(basis_id)
4052
return basis_id, basis_tree
4055
InterRepository.register_optimiser(InterDifferingSerializer)
4056
InterRepository.register_optimiser(InterSameDataRepository)
4057
InterRepository.register_optimiser(InterWeaveRepo)
4058
InterRepository.register_optimiser(InterKnitRepo)
1788
4061
class CopyConverter(object):
1789
4062
"""A repository conversion tool which just performs a copy of the content.
1844
4219
return revision_graph
4222
class StreamSink(object):
4223
"""An object that can insert a stream into a repository.
4225
This interface handles the complexity of reserialising inventories and
4226
revisions from different formats, and allows unidirectional insertion into
4227
stacked repositories without looking for the missing basis parents
4231
def __init__(self, target_repo):
4232
self.target_repo = target_repo
4234
def insert_stream(self, stream, src_format, resume_tokens):
4235
"""Insert a stream's content into the target repository.
4237
:param src_format: a bzr repository format.
4239
:return: a list of resume tokens and an iterable of keys additional
4240
items required before the insertion can be completed.
4242
self.target_repo.lock_write()
4245
self.target_repo.resume_write_group(resume_tokens)
4248
self.target_repo.start_write_group()
4251
# locked_insert_stream performs a commit|suspend.
4252
return self._locked_insert_stream(stream, src_format, is_resume)
4254
self.target_repo.abort_write_group(suppress_errors=True)
4257
self.target_repo.unlock()
4259
def _locked_insert_stream(self, stream, src_format, is_resume):
4260
to_serializer = self.target_repo._format._serializer
4261
src_serializer = src_format._serializer
4263
if to_serializer == src_serializer:
4264
# If serializers match and the target is a pack repository, set the
4265
# write cache size on the new pack. This avoids poor performance
4266
# on transports where append is unbuffered (such as
4267
# RemoteTransport). This is safe to do because nothing should read
4268
# back from the target repository while a stream with matching
4269
# serialization is being inserted.
4270
# The exception is that a delta record from the source that should
4271
# be a fulltext may need to be expanded by the target (see
4272
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4273
# explicitly flush any buffered writes first in that rare case.
4275
new_pack = self.target_repo._pack_collection._new_pack
4276
except AttributeError:
4277
# Not a pack repository
4280
new_pack.set_write_cache_size(1024*1024)
4281
for substream_type, substream in stream:
4282
if 'stream' in debug.debug_flags:
4283
mutter('inserting substream: %s', substream_type)
4284
if substream_type == 'texts':
4285
self.target_repo.texts.insert_record_stream(substream)
4286
elif substream_type == 'inventories':
4287
if src_serializer == to_serializer:
4288
self.target_repo.inventories.insert_record_stream(
4291
self._extract_and_insert_inventories(
4292
substream, src_serializer)
4293
elif substream_type == 'inventory-deltas':
4294
self._extract_and_insert_inventory_deltas(
4295
substream, src_serializer)
4296
elif substream_type == 'chk_bytes':
4297
# XXX: This doesn't support conversions, as it assumes the
4298
# conversion was done in the fetch code.
4299
self.target_repo.chk_bytes.insert_record_stream(substream)
4300
elif substream_type == 'revisions':
4301
# This may fallback to extract-and-insert more often than
4302
# required if the serializers are different only in terms of
4304
if src_serializer == to_serializer:
4305
self.target_repo.revisions.insert_record_stream(
4308
self._extract_and_insert_revisions(substream,
4310
elif substream_type == 'signatures':
4311
self.target_repo.signatures.insert_record_stream(substream)
4313
raise AssertionError('kaboom! %s' % (substream_type,))
4314
# Done inserting data, and the missing_keys calculations will try to
4315
# read back from the inserted data, so flush the writes to the new pack
4316
# (if this is pack format).
4317
if new_pack is not None:
4318
new_pack._write_data('', flush=True)
4319
# Find all the new revisions (including ones from resume_tokens)
4320
missing_keys = self.target_repo.get_missing_parent_inventories(
4321
check_for_missing_texts=is_resume)
4323
for prefix, versioned_file in (
4324
('texts', self.target_repo.texts),
4325
('inventories', self.target_repo.inventories),
4326
('revisions', self.target_repo.revisions),
4327
('signatures', self.target_repo.signatures),
4328
('chk_bytes', self.target_repo.chk_bytes),
4330
if versioned_file is None:
4332
# TODO: key is often going to be a StaticTuple object
4333
# I don't believe we can define a method by which
4334
# (prefix,) + StaticTuple will work, though we could
4335
# define a StaticTuple.sq_concat that would allow you to
4336
# pass in either a tuple or a StaticTuple as the second
4337
# object, so instead we could have:
4338
# StaticTuple(prefix) + key here...
4339
missing_keys.update((prefix,) + key for key in
4340
versioned_file.get_missing_compression_parent_keys())
4341
except NotImplementedError:
4342
# cannot even attempt suspending, and missing would have failed
4343
# during stream insertion.
4344
missing_keys = set()
4347
# suspend the write group and tell the caller what we is
4348
# missing. We know we can suspend or else we would not have
4349
# entered this code path. (All repositories that can handle
4350
# missing keys can handle suspending a write group).
4351
write_group_tokens = self.target_repo.suspend_write_group()
4352
return write_group_tokens, missing_keys
4353
hint = self.target_repo.commit_write_group()
4354
if (to_serializer != src_serializer and
4355
self.target_repo._format.pack_compresses):
4356
self.target_repo.pack(hint=hint)
4359
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4360
target_rich_root = self.target_repo._format.rich_root_data
4361
target_tree_refs = self.target_repo._format.supports_tree_reference
4362
for record in substream:
4363
# Insert the delta directly
4364
inventory_delta_bytes = record.get_bytes_as('fulltext')
4365
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4367
parse_result = deserialiser.parse_text_bytes(
4368
inventory_delta_bytes)
4369
except inventory_delta.IncompatibleInventoryDelta, err:
4370
trace.mutter("Incompatible delta: %s", err.msg)
4371
raise errors.IncompatibleRevision(self.target_repo._format)
4372
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4373
revision_id = new_id
4374
parents = [key[0] for key in record.parents]
4375
self.target_repo.add_inventory_by_delta(
4376
basis_id, inv_delta, revision_id, parents)
4378
def _extract_and_insert_inventories(self, substream, serializer,
4380
"""Generate a new inventory versionedfile in target, converting data.
4382
The inventory is retrieved from the source, (deserializing it), and
4383
stored in the target (reserializing it in a different format).
4385
target_rich_root = self.target_repo._format.rich_root_data
4386
target_tree_refs = self.target_repo._format.supports_tree_reference
4387
for record in substream:
4388
# It's not a delta, so it must be a fulltext in the source
4389
# serializer's format.
4390
bytes = record.get_bytes_as('fulltext')
4391
revision_id = record.key[0]
4392
inv = serializer.read_inventory_from_string(bytes, revision_id)
4393
parents = [key[0] for key in record.parents]
4394
self.target_repo.add_inventory(revision_id, inv, parents)
4395
# No need to keep holding this full inv in memory when the rest of
4396
# the substream is likely to be all deltas.
4399
def _extract_and_insert_revisions(self, substream, serializer):
4400
for record in substream:
4401
bytes = record.get_bytes_as('fulltext')
4402
revision_id = record.key[0]
4403
rev = serializer.read_revision_from_string(bytes)
4404
if rev.revision_id != revision_id:
4405
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4406
self.target_repo.add_revision(revision_id, rev)
4409
if self.target_repo._format._fetch_reconcile:
4410
self.target_repo.reconcile()
4413
class StreamSource(object):
4414
"""A source of a stream for fetching between repositories."""
4416
def __init__(self, from_repository, to_format):
4417
"""Create a StreamSource streaming from from_repository."""
4418
self.from_repository = from_repository
4419
self.to_format = to_format
4421
def delta_on_metadata(self):
4422
"""Return True if delta's are permitted on metadata streams.
4424
That is on revisions and signatures.
4426
src_serializer = self.from_repository._format._serializer
4427
target_serializer = self.to_format._serializer
4428
return (self.to_format._fetch_uses_deltas and
4429
src_serializer == target_serializer)
4431
def _fetch_revision_texts(self, revs):
4432
# fetch signatures first and then the revision texts
4433
# may need to be a InterRevisionStore call here.
4434
from_sf = self.from_repository.signatures
4435
# A missing signature is just skipped.
4436
keys = [(rev_id,) for rev_id in revs]
4437
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4439
self.to_format._fetch_order,
4440
not self.to_format._fetch_uses_deltas))
4441
# If a revision has a delta, this is actually expanded inside the
4442
# insert_record_stream code now, which is an alternate fix for
4444
from_rf = self.from_repository.revisions
4445
revisions = from_rf.get_record_stream(
4447
self.to_format._fetch_order,
4448
not self.delta_on_metadata())
4449
return [('signatures', signatures), ('revisions', revisions)]
4451
def _generate_root_texts(self, revs):
4452
"""This will be called by get_stream between fetching weave texts and
4453
fetching the inventory weave.
4455
if self._rich_root_upgrade():
4456
return _mod_fetch.Inter1and2Helper(
4457
self.from_repository).generate_root_texts(revs)
4461
def get_stream(self, search):
4463
revs = search.get_keys()
4464
graph = self.from_repository.get_graph()
4465
revs = tsort.topo_sort(graph.get_parent_map(revs))
4466
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4468
for knit_kind, file_id, revisions in data_to_fetch:
4469
if knit_kind != phase:
4471
# Make a new progress bar for this phase
4472
if knit_kind == "file":
4473
# Accumulate file texts
4474
text_keys.extend([(file_id, revision) for revision in
4476
elif knit_kind == "inventory":
4477
# Now copy the file texts.
4478
from_texts = self.from_repository.texts
4479
yield ('texts', from_texts.get_record_stream(
4480
text_keys, self.to_format._fetch_order,
4481
not self.to_format._fetch_uses_deltas))
4482
# Cause an error if a text occurs after we have done the
4485
# Before we process the inventory we generate the root
4486
# texts (if necessary) so that the inventories references
4488
for _ in self._generate_root_texts(revs):
4490
# we fetch only the referenced inventories because we do not
4491
# know for unselected inventories whether all their required
4492
# texts are present in the other repository - it could be
4494
for info in self._get_inventory_stream(revs):
4496
elif knit_kind == "signatures":
4497
# Nothing to do here; this will be taken care of when
4498
# _fetch_revision_texts happens.
4500
elif knit_kind == "revisions":
4501
for record in self._fetch_revision_texts(revs):
4504
raise AssertionError("Unknown knit kind %r" % knit_kind)
4506
def get_stream_for_missing_keys(self, missing_keys):
4507
# missing keys can only occur when we are byte copying and not
4508
# translating (because translation means we don't send
4509
# unreconstructable deltas ever).
4511
keys['texts'] = set()
4512
keys['revisions'] = set()
4513
keys['inventories'] = set()
4514
keys['chk_bytes'] = set()
4515
keys['signatures'] = set()
4516
for key in missing_keys:
4517
keys[key[0]].add(key[1:])
4518
if len(keys['revisions']):
4519
# If we allowed copying revisions at this point, we could end up
4520
# copying a revision without copying its required texts: a
4521
# violation of the requirements for repository integrity.
4522
raise AssertionError(
4523
'cannot copy revisions to fill in missing deltas %s' % (
4524
keys['revisions'],))
4525
for substream_kind, keys in keys.iteritems():
4526
vf = getattr(self.from_repository, substream_kind)
4527
if vf is None and keys:
4528
raise AssertionError(
4529
"cannot fill in keys for a versioned file we don't"
4530
" have: %s needs %s" % (substream_kind, keys))
4532
# No need to stream something we don't have
4534
if substream_kind == 'inventories':
4535
# Some missing keys are genuinely ghosts, filter those out.
4536
present = self.from_repository.inventories.get_parent_map(keys)
4537
revs = [key[0] for key in present]
4538
# Get the inventory stream more-or-less as we do for the
4539
# original stream; there's no reason to assume that records
4540
# direct from the source will be suitable for the sink. (Think
4541
# e.g. 2a -> 1.9-rich-root).
4542
for info in self._get_inventory_stream(revs, missing=True):
4546
# Ask for full texts always so that we don't need more round trips
4547
# after this stream.
4548
# Some of the missing keys are genuinely ghosts, so filter absent
4549
# records. The Sink is responsible for doing another check to
4550
# ensure that ghosts don't introduce missing data for future
4552
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4553
self.to_format._fetch_order, True))
4554
yield substream_kind, stream
4556
def inventory_fetch_order(self):
4557
if self._rich_root_upgrade():
4558
return 'topological'
4560
return self.to_format._fetch_order
4562
def _rich_root_upgrade(self):
4563
return (not self.from_repository._format.rich_root_data and
4564
self.to_format.rich_root_data)
4566
def _get_inventory_stream(self, revision_ids, missing=False):
4567
from_format = self.from_repository._format
4568
if (from_format.supports_chks and self.to_format.supports_chks and
4569
from_format.network_name() == self.to_format.network_name()):
4570
raise AssertionError(
4571
"this case should be handled by GroupCHKStreamSource")
4572
elif 'forceinvdeltas' in debug.debug_flags:
4573
return self._get_convertable_inventory_stream(revision_ids,
4574
delta_versus_null=missing)
4575
elif from_format.network_name() == self.to_format.network_name():
4577
return self._get_simple_inventory_stream(revision_ids,
4579
elif (not from_format.supports_chks and not self.to_format.supports_chks
4580
and from_format._serializer == self.to_format._serializer):
4581
# Essentially the same format.
4582
return self._get_simple_inventory_stream(revision_ids,
4585
# Any time we switch serializations, we want to use an
4586
# inventory-delta based approach.
4587
return self._get_convertable_inventory_stream(revision_ids,
4588
delta_versus_null=missing)
4590
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4591
# NB: This currently reopens the inventory weave in source;
4592
# using a single stream interface instead would avoid this.
4593
from_weave = self.from_repository.inventories
4595
delta_closure = True
4597
delta_closure = not self.delta_on_metadata()
4598
yield ('inventories', from_weave.get_record_stream(
4599
[(rev_id,) for rev_id in revision_ids],
4600
self.inventory_fetch_order(), delta_closure))
4602
def _get_convertable_inventory_stream(self, revision_ids,
4603
delta_versus_null=False):
4604
# The two formats are sufficiently different that there is no fast
4605
# path, so we need to send just inventorydeltas, which any
4606
# sufficiently modern client can insert into any repository.
4607
# The StreamSink code expects to be able to
4608
# convert on the target, so we need to put bytes-on-the-wire that can
4609
# be converted. That means inventory deltas (if the remote is <1.19,
4610
# RemoteStreamSink will fallback to VFS to insert the deltas).
4611
yield ('inventory-deltas',
4612
self._stream_invs_as_deltas(revision_ids,
4613
delta_versus_null=delta_versus_null))
4615
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4616
"""Return a stream of inventory-deltas for the given rev ids.
4618
:param revision_ids: The list of inventories to transmit
4619
:param delta_versus_null: Don't try to find a minimal delta for this
4620
entry, instead compute the delta versus the NULL_REVISION. This
4621
effectively streams a complete inventory. Used for stuff like
4622
filling in missing parents, etc.
4624
from_repo = self.from_repository
4625
revision_keys = [(rev_id,) for rev_id in revision_ids]
4626
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4627
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4629
inventories = self.from_repository.iter_inventories(
4630
revision_ids, 'topological')
4631
format = from_repo._format
4632
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4633
inventory_cache = lru_cache.LRUCache(50)
4634
null_inventory = from_repo.revision_tree(
4635
_mod_revision.NULL_REVISION).inventory
4636
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4637
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4638
# repo back into a non-rich-root repo ought to be allowed)
4639
serializer = inventory_delta.InventoryDeltaSerializer(
4640
versioned_root=format.rich_root_data,
4641
tree_references=format.supports_tree_reference)
4642
for inv in inventories:
4643
key = (inv.revision_id,)
4644
parent_keys = parent_map.get(key, ())
4646
if not delta_versus_null and parent_keys:
4647
# The caller did not ask for complete inventories and we have
4648
# some parents that we can delta against. Make a delta against
4649
# each parent so that we can find the smallest.
4650
parent_ids = [parent_key[0] for parent_key in parent_keys]
4651
for parent_id in parent_ids:
4652
if parent_id not in invs_sent_so_far:
4653
# We don't know that the remote side has this basis, so
4656
if parent_id == _mod_revision.NULL_REVISION:
4657
parent_inv = null_inventory
4659
parent_inv = inventory_cache.get(parent_id, None)
4660
if parent_inv is None:
4661
parent_inv = from_repo.get_inventory(parent_id)
4662
candidate_delta = inv._make_delta(parent_inv)
4663
if (delta is None or
4664
len(delta) > len(candidate_delta)):
4665
delta = candidate_delta
4666
basis_id = parent_id
4668
# Either none of the parents ended up being suitable, or we
4669
# were asked to delta against NULL
4670
basis_id = _mod_revision.NULL_REVISION
4671
delta = inv._make_delta(null_inventory)
4672
invs_sent_so_far.add(inv.revision_id)
4673
inventory_cache[inv.revision_id] = inv
4674
delta_serialized = ''.join(
4675
serializer.delta_to_lines(basis_id, key[-1], delta))
4676
yield versionedfile.FulltextContentFactory(
4677
key, parent_keys, None, delta_serialized)
1847
4680
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
1848
4681
stop_revision=None):
1849
4682
"""Extend the partial history to include a given index
added
removed
bzrlib/repository.py
