202
352
commit to be valid, deletes against the basis MUST be recorded via
203
353
builder.record_delete().
205
raise NotImplementedError(self.will_record_deletes)
207
def record_iter_changes(self, tree, basis_revision_id, iter_changes):
355
self._recording_deletes = True
357
basis_id = self.parents[0]
359
basis_id = _mod_revision.NULL_REVISION
360
self.basis_delta_revision = basis_id
362
def record_entry_contents(self, ie, parent_invs, path, tree,
364
"""Record the content of ie from tree into the commit if needed.
366
Side effect: sets ie.revision when unchanged
368
:param ie: An inventory entry present in the commit.
369
:param parent_invs: The inventories of the parent revisions of the
371
:param path: The path the entry is at in the tree.
372
:param tree: The tree which contains this entry and should be used to
374
:param content_summary: Summary data from the tree about the paths
375
content - stat, length, exec, sha/link target. This is only
376
accessed when the entry has a revision of None - that is when it is
377
a candidate to commit.
378
:return: A tuple (change_delta, version_recorded, fs_hash).
379
change_delta is an inventory_delta change for this entry against
380
the basis tree of the commit, or None if no change occured against
382
version_recorded is True if a new version of the entry has been
383
recorded. For instance, committing a merge where a file was only
384
changed on the other side will return (delta, False).
385
fs_hash is either None, or the hash details for the path (currently
386
a tuple of the contents sha1 and the statvalue returned by
387
tree.get_file_with_stat()).
389
if self.new_inventory.root is None:
390
if ie.parent_id is not None:
391
raise errors.RootMissing()
392
self._check_root(ie, parent_invs, tree)
393
if ie.revision is None:
394
kind = content_summary[0]
396
# ie is carried over from a prior commit
398
# XXX: repository specific check for nested tree support goes here - if
399
# the repo doesn't want nested trees we skip it ?
400
if (kind == 'tree-reference' and
401
not self.repository._format.supports_tree_reference):
402
# mismatch between commit builder logic and repository:
403
# this needs the entry creation pushed down into the builder.
404
raise NotImplementedError('Missing repository subtree support.')
405
self.new_inventory.add(ie)
407
# TODO: slow, take it out of the inner loop.
409
basis_inv = parent_invs[0]
411
basis_inv = Inventory(root_id=None)
413
# ie.revision is always None if the InventoryEntry is considered
414
# for committing. We may record the previous parents revision if the
415
# content is actually unchanged against a sole head.
416
if ie.revision is not None:
417
if not self._versioned_root and path == '':
418
# repositories that do not version the root set the root's
419
# revision to the new commit even when no change occurs (more
420
# specifically, they do not record a revision on the root; and
421
# the rev id is assigned to the root during deserialisation -
422
# this masks when a change may have occurred against the basis.
423
# To match this we always issue a delta, because the revision
424
# of the root will always be changing.
425
if ie.file_id in basis_inv:
426
delta = (basis_inv.id2path(ie.file_id), path,
430
delta = (None, path, ie.file_id, ie)
431
self._basis_delta.append(delta)
432
return delta, False, None
434
# we don't need to commit this, because the caller already
435
# determined that an existing revision of this file is
436
# appropriate. If its not being considered for committing then
437
# it and all its parents to the root must be unaltered so
438
# no-change against the basis.
439
if ie.revision == self._new_revision_id:
440
raise AssertionError("Impossible situation, a skipped "
441
"inventory entry (%r) claims to be modified in this "
442
"commit (%r).", (ie, self._new_revision_id))
443
return None, False, None
444
# XXX: Friction: parent_candidates should return a list not a dict
445
# so that we don't have to walk the inventories again.
446
parent_candiate_entries = ie.parent_candidates(parent_invs)
447
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
449
for inv in parent_invs:
450
if ie.file_id in inv:
451
old_rev = inv[ie.file_id].revision
452
if old_rev in head_set:
453
heads.append(inv[ie.file_id].revision)
454
head_set.remove(inv[ie.file_id].revision)
457
# now we check to see if we need to write a new record to the
459
# We write a new entry unless there is one head to the ancestors, and
460
# the kind-derived content is unchanged.
462
# Cheapest check first: no ancestors, or more the one head in the
463
# ancestors, we write a new node.
467
# There is a single head, look it up for comparison
468
parent_entry = parent_candiate_entries[heads[0]]
469
# if the non-content specific data has changed, we'll be writing a
471
if (parent_entry.parent_id != ie.parent_id or
472
parent_entry.name != ie.name):
474
# now we need to do content specific checks:
476
# if the kind changed the content obviously has
477
if kind != parent_entry.kind:
479
# Stat cache fingerprint feedback for the caller - None as we usually
480
# don't generate one.
483
if content_summary[2] is None:
484
raise ValueError("Files must not have executable = None")
486
# We can't trust a check of the file length because of content
488
if (# if the exec bit has changed we have to store:
489
parent_entry.executable != content_summary[2]):
491
elif parent_entry.text_sha1 == content_summary[3]:
492
# all meta and content is unchanged (using a hash cache
493
# hit to check the sha)
494
ie.revision = parent_entry.revision
495
ie.text_size = parent_entry.text_size
496
ie.text_sha1 = parent_entry.text_sha1
497
ie.executable = parent_entry.executable
498
return self._get_delta(ie, basis_inv, path), False, None
500
# Either there is only a hash change(no hash cache entry,
501
# or same size content change), or there is no change on
503
# Provide the parent's hash to the store layer, so that the
504
# content is unchanged we will not store a new node.
505
nostore_sha = parent_entry.text_sha1
507
# We want to record a new node regardless of the presence or
508
# absence of a content change in the file.
510
ie.executable = content_summary[2]
511
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
513
text = file_obj.read()
517
ie.text_sha1, ie.text_size = self._add_text_to_weave(
518
ie.file_id, text, heads, nostore_sha)
519
# Let the caller know we generated a stat fingerprint.
520
fingerprint = (ie.text_sha1, stat_value)
521
except errors.ExistingContent:
522
# Turns out that the file content was unchanged, and we were
523
# only going to store a new node if it was changed. Carry over
525
ie.revision = parent_entry.revision
526
ie.text_size = parent_entry.text_size
527
ie.text_sha1 = parent_entry.text_sha1
528
ie.executable = parent_entry.executable
529
return self._get_delta(ie, basis_inv, path), False, None
530
elif kind == 'directory':
532
# all data is meta here, nothing specific to directory, so
534
ie.revision = parent_entry.revision
535
return self._get_delta(ie, basis_inv, path), False, None
536
self._add_text_to_weave(ie.file_id, '', heads, None)
537
elif kind == 'symlink':
538
current_link_target = content_summary[3]
540
# symlink target is not generic metadata, check if it has
542
if current_link_target != parent_entry.symlink_target:
545
# unchanged, carry over.
546
ie.revision = parent_entry.revision
547
ie.symlink_target = parent_entry.symlink_target
548
return self._get_delta(ie, basis_inv, path), False, None
549
ie.symlink_target = current_link_target
550
self._add_text_to_weave(ie.file_id, '', heads, None)
551
elif kind == 'tree-reference':
553
if content_summary[3] != parent_entry.reference_revision:
556
# unchanged, carry over.
557
ie.reference_revision = parent_entry.reference_revision
558
ie.revision = parent_entry.revision
559
return self._get_delta(ie, basis_inv, path), False, None
560
ie.reference_revision = content_summary[3]
561
if ie.reference_revision is None:
562
raise AssertionError("invalid content_summary for nested tree: %r"
563
% (content_summary,))
564
self._add_text_to_weave(ie.file_id, '', heads, None)
566
raise NotImplementedError('unknown kind')
567
ie.revision = self._new_revision_id
568
self._any_changes = True
569
return self._get_delta(ie, basis_inv, path), True, fingerprint
571
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
572
_entry_factory=entry_factory):
208
573
"""Record a new tree via iter_changes.
210
575
:param tree: The tree to obtain text contents from for changed objects.
215
580
to basis_revision_id. The iterator must not include any items with
216
581
a current kind of None - missing items must be either filtered out
217
582
or errored-on beefore record_iter_changes sees the item.
583
:param _entry_factory: Private method to bind entry_factory locally for
218
585
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
219
586
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,
588
# Create an inventory delta based on deltas between all the parents and
589
# deltas between all the parent inventories. We use inventory delta's
590
# between the inventory objects because iter_changes masks
591
# last-changed-field only changes.
593
# file_id -> change map, change is fileid, paths, changed, versioneds,
594
# parents, names, kinds, executables
596
# {file_id -> revision_id -> inventory entry, for entries in parent
597
# trees that are not parents[0]
601
revtrees = list(self.repository.revision_trees(self.parents))
602
except errors.NoSuchRevision:
603
# one or more ghosts, slow path.
605
for revision_id in self.parents:
607
revtrees.append(self.repository.revision_tree(revision_id))
608
except errors.NoSuchRevision:
610
basis_revision_id = _mod_revision.NULL_REVISION
612
revtrees.append(self.repository.revision_tree(
613
_mod_revision.NULL_REVISION))
614
# The basis inventory from a repository
616
basis_inv = revtrees[0].inventory
618
basis_inv = self.repository.revision_tree(
619
_mod_revision.NULL_REVISION).inventory
620
if len(self.parents) > 0:
621
if basis_revision_id != self.parents[0] and not ghost_basis:
623
"arbitrary basis parents not yet supported with merges")
624
for revtree in revtrees[1:]:
625
for change in revtree.inventory._make_delta(basis_inv):
626
if change[1] is None:
627
# Not present in this parent.
629
if change[2] not in merged_ids:
630
if change[0] is not None:
631
basis_entry = basis_inv[change[2]]
632
merged_ids[change[2]] = [
634
basis_entry.revision,
637
parent_entries[change[2]] = {
639
basis_entry.revision:basis_entry,
641
change[3].revision:change[3],
644
merged_ids[change[2]] = [change[3].revision]
645
parent_entries[change[2]] = {change[3].revision:change[3]}
647
merged_ids[change[2]].append(change[3].revision)
648
parent_entries[change[2]][change[3].revision] = change[3]
651
# Setup the changes from the tree:
652
# changes maps file_id -> (change, [parent revision_ids])
654
for change in iter_changes:
655
# This probably looks up in basis_inv way to much.
656
if change[1][0] is not None:
657
head_candidate = [basis_inv[change[0]].revision]
660
changes[change[0]] = change, merged_ids.get(change[0],
662
unchanged_merged = set(merged_ids) - set(changes)
663
# Extend the changes dict with synthetic changes to record merges of
665
for file_id in unchanged_merged:
666
# Record a merged version of these items that did not change vs the
667
# basis. This can be either identical parallel changes, or a revert
668
# of a specific file after a merge. The recorded content will be
669
# that of the current tree (which is the same as the basis), but
670
# the per-file graph will reflect a merge.
671
# NB:XXX: We are reconstructing path information we had, this
672
# should be preserved instead.
673
# inv delta change: (file_id, (path_in_source, path_in_target),
674
# changed_content, versioned, parent, name, kind,
677
basis_entry = basis_inv[file_id]
678
except errors.NoSuchId:
679
# a change from basis->some_parents but file_id isn't in basis
680
# so was new in the merge, which means it must have changed
681
# from basis -> current, and as it hasn't the add was reverted
682
# by the user. So we discard this change.
686
(basis_inv.id2path(file_id), tree.id2path(file_id)),
688
(basis_entry.parent_id, basis_entry.parent_id),
689
(basis_entry.name, basis_entry.name),
690
(basis_entry.kind, basis_entry.kind),
691
(basis_entry.executable, basis_entry.executable))
692
changes[file_id] = (change, merged_ids[file_id])
693
# changes contains tuples with the change and a set of inventory
694
# candidates for the file.
696
# old_path, new_path, file_id, new_inventory_entry
697
seen_root = False # Is the root in the basis delta?
698
inv_delta = self._basis_delta
699
modified_rev = self._new_revision_id
700
for change, head_candidates in changes.values():
701
if change[3][1]: # versioned in target.
702
# Several things may be happening here:
703
# We may have a fork in the per-file graph
704
# - record a change with the content from tree
705
# We may have a change against < all trees
706
# - carry over the tree that hasn't changed
707
# We may have a change against all trees
708
# - record the change with the content from tree
711
entry = _entry_factory[kind](file_id, change[5][1],
713
head_set = self._heads(change[0], set(head_candidates))
716
for head_candidate in head_candidates:
717
if head_candidate in head_set:
718
heads.append(head_candidate)
719
head_set.remove(head_candidate)
722
# Could be a carry-over situation:
723
parent_entry_revs = parent_entries.get(file_id, None)
724
if parent_entry_revs:
725
parent_entry = parent_entry_revs.get(heads[0], None)
728
if parent_entry is None:
729
# The parent iter_changes was called against is the one
730
# that is the per-file head, so any change is relevant
731
# iter_changes is valid.
732
carry_over_possible = False
734
# could be a carry over situation
735
# A change against the basis may just indicate a merge,
736
# we need to check the content against the source of the
737
# merge to determine if it was changed after the merge
739
if (parent_entry.kind != entry.kind or
740
parent_entry.parent_id != entry.parent_id or
741
parent_entry.name != entry.name):
742
# Metadata common to all entries has changed
743
# against per-file parent
744
carry_over_possible = False
746
carry_over_possible = True
747
# per-type checks for changes against the parent_entry
750
# Cannot be a carry-over situation
751
carry_over_possible = False
752
# Populate the entry in the delta
754
# XXX: There is still a small race here: If someone reverts the content of a file
755
# after iter_changes examines and decides it has changed,
756
# we will unconditionally record a new version even if some
757
# other process reverts it while commit is running (with
758
# the revert happening after iter_changes did it's
761
entry.executable = True
763
entry.executable = False
764
if (carry_over_possible and
765
parent_entry.executable == entry.executable):
766
# Check the file length, content hash after reading
768
nostore_sha = parent_entry.text_sha1
771
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
773
text = file_obj.read()
777
entry.text_sha1, entry.text_size = self._add_text_to_weave(
778
file_id, text, heads, nostore_sha)
779
yield file_id, change[1][1], (entry.text_sha1, stat_value)
780
except errors.ExistingContent:
781
# No content change against a carry_over parent
782
# Perhaps this should also yield a fs hash update?
784
entry.text_size = parent_entry.text_size
785
entry.text_sha1 = parent_entry.text_sha1
786
elif kind == 'symlink':
788
entry.symlink_target = tree.get_symlink_target(file_id)
789
if (carry_over_possible and
790
parent_entry.symlink_target == entry.symlink_target):
793
self._add_text_to_weave(change[0], '', heads, None)
794
elif kind == 'directory':
795
if carry_over_possible:
798
# Nothing to set on the entry.
799
# XXX: split into the Root and nonRoot versions.
800
if change[1][1] != '' or self.repository.supports_rich_root():
801
self._add_text_to_weave(change[0], '', heads, None)
802
elif kind == 'tree-reference':
803
if not self.repository._format.supports_tree_reference:
804
# This isn't quite sane as an error, but we shouldn't
805
# ever see this code path in practice: tree's don't
806
# permit references when the repo doesn't support tree
808
raise errors.UnsupportedOperation(tree.add_reference,
810
reference_revision = tree.get_reference_revision(change[0])
811
entry.reference_revision = reference_revision
812
if (carry_over_possible and
813
parent_entry.reference_revision == reference_revision):
816
self._add_text_to_weave(change[0], '', heads, None)
818
raise AssertionError('unknown kind %r' % kind)
820
entry.revision = modified_rev
822
entry.revision = parent_entry.revision
825
new_path = change[1][1]
826
inv_delta.append((change[1][0], new_path, change[0], entry))
829
self.new_inventory = None
831
# This should perhaps be guarded by a check that the basis we
832
# commit against is the basis for the commit and if not do a delta
834
self._any_changes = True
836
# housekeeping root entry changes do not affect no-change commits.
837
self._require_root_change(tree)
838
self.basis_delta_revision = basis_revision_id
840
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
841
parent_keys = tuple([(file_id, parent) for parent in parents])
842
return self.repository.texts._add_text(
843
(file_id, self._new_revision_id), parent_keys, new_text,
844
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
847
class RootCommitBuilder(CommitBuilder):
848
"""This commitbuilder actually records the root id"""
850
# the root entry gets versioned properly by this builder.
851
_versioned_root = True
853
def _check_root(self, ie, parent_invs, tree):
854
"""Helper for record_entry_contents.
856
:param ie: An entry being added.
857
:param parent_invs: The inventories of the parent revisions of the
859
:param tree: The tree that is being committed.
862
def _require_root_change(self, tree):
863
"""Enforce an appropriate root object change.
865
This is called once when record_iter_changes is called, if and only if
866
the root was not in the delta calculated by record_iter_changes.
868
:param tree: The tree which is being committed.
870
# versioned roots do not change unless the tree found a change.
241
873
######################################################################
245
class Repository(_RelockDebugMixin, controldir.ControlComponent):
877
class Repository(_RelockDebugMixin, bzrdir.ControlComponent):
246
878
"""Repository holding history for one or more branches.
248
880
The repository holds and retrieves historical information including
249
881
revisions and file history. It's normally accessed only by the Branch,
250
882
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.
884
The Repository builds on top of some byte storage facilies (the revisions,
885
signatures, inventories, texts and chk_bytes attributes) and a Transport,
886
which respectively provide byte storage and a means to access the (possibly
889
The byte storage facilities are addressed via tuples, which we refer to
890
as 'keys' throughout the code base. Revision_keys, inventory_keys and
891
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
892
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
893
byte string made up of a hash identifier and a hash value.
894
We use this interface because it allows low friction with the underlying
895
code that implements disk indices, network encoding and other parts of
898
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
899
the serialised revisions for the repository. This can be used to obtain
900
revision graph information or to access raw serialised revisions.
901
The result of trying to insert data into the repository via this store
902
is undefined: it should be considered read-only except for implementors
904
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
905
the serialised signatures for the repository. This can be used to
906
obtain access to raw serialised signatures. The result of trying to
907
insert data into the repository via this store is undefined: it should
908
be considered read-only except for implementors of repositories.
909
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
910
the serialised inventories for the repository. This can be used to
911
obtain unserialised inventories. The result of trying to insert data
912
into the repository via this store is undefined: it should be
913
considered read-only except for implementors of repositories.
914
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
915
texts of files and directories for the repository. This can be used to
916
obtain file texts or file graphs. Note that Repository.iter_file_bytes
917
is usually a better interface for accessing file texts.
918
The result of trying to insert data into the repository via this store
919
is undefined: it should be considered read-only except for implementors
921
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
922
any data the repository chooses to store or have indexed by its hash.
923
The result of trying to insert data into the repository via this store
924
is undefined: it should be considered read-only except for implementors
926
:ivar _transport: Transport for file access to repository, typically
927
pointing to .bzr/repository.
930
# What class to use for a CommitBuilder. Often its simpler to change this
931
# in a Repository class subclass rather than to override
932
# get_commit_builder.
933
_commit_builder_class = CommitBuilder
934
# The search regex used by xml based repositories to determine what things
935
# where changed in a single commit.
936
_file_ids_altered_regex = lazy_regex.lazy_compile(
937
r'file_id="(?P<file_id>[^"]+)"'
938
r'.* revision="(?P<revision_id>[^"]+)"'
256
941
def abort_write_group(self, suppress_errors=False):
257
942
"""Commit the contents accrued within the current write group.
313
1011
return InterRepository._assert_same_model(self, repository)
1013
def add_inventory(self, revision_id, inv, parents):
1014
"""Add the inventory inv to the repository as revision_id.
1016
:param parents: The revision ids of the parents that revision_id
1017
is known to have and are in the repository already.
1019
:returns: The validator(which is a sha1 digest, though what is sha'd is
1020
repository format specific) of the serialized inventory.
1022
if not self.is_in_write_group():
1023
raise AssertionError("%r not in write group" % (self,))
1024
_mod_revision.check_not_reserved_id(revision_id)
1025
if not (inv.revision_id is None or inv.revision_id == revision_id):
1026
raise AssertionError(
1027
"Mismatch between inventory revision"
1028
" id and insertion revid (%r, %r)"
1029
% (inv.revision_id, revision_id))
1030
if inv.root is None:
1031
raise AssertionError()
1032
return self._add_inventory_checked(revision_id, inv, parents)
1034
def _add_inventory_checked(self, revision_id, inv, parents):
1035
"""Add inv to the repository after checking the inputs.
1037
This function can be overridden to allow different inventory styles.
1039
:seealso: add_inventory, for the contract.
1041
inv_lines = self._serializer.write_inventory_to_lines(inv)
1042
return self._inventory_add_lines(revision_id, parents,
1043
inv_lines, check_content=False)
1045
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1046
parents, basis_inv=None, propagate_caches=False):
1047
"""Add a new inventory expressed as a delta against another revision.
1049
See the inventory developers documentation for the theory behind
1052
:param basis_revision_id: The inventory id the delta was created
1053
against. (This does not have to be a direct parent.)
1054
:param delta: The inventory delta (see Inventory.apply_delta for
1056
:param new_revision_id: The revision id that the inventory is being
1058
:param parents: The revision ids of the parents that revision_id is
1059
known to have and are in the repository already. These are supplied
1060
for repositories that depend on the inventory graph for revision
1061
graph access, as well as for those that pun ancestry with delta
1063
:param basis_inv: The basis inventory if it is already known,
1065
:param propagate_caches: If True, the caches for this inventory are
1066
copied to and updated for the result if possible.
1068
:returns: (validator, new_inv)
1069
The validator(which is a sha1 digest, though what is sha'd is
1070
repository format specific) of the serialized inventory, and the
1071
resulting inventory.
1073
if not self.is_in_write_group():
1074
raise AssertionError("%r not in write group" % (self,))
1075
_mod_revision.check_not_reserved_id(new_revision_id)
1076
basis_tree = self.revision_tree(basis_revision_id)
1077
basis_tree.lock_read()
1079
# Note that this mutates the inventory of basis_tree, which not all
1080
# inventory implementations may support: A better idiom would be to
1081
# return a new inventory, but as there is no revision tree cache in
1082
# repository this is safe for now - RBC 20081013
1083
if basis_inv is None:
1084
basis_inv = basis_tree.inventory
1085
basis_inv.apply_delta(delta)
1086
basis_inv.revision_id = new_revision_id
1087
return (self.add_inventory(new_revision_id, basis_inv, parents),
1092
def _inventory_add_lines(self, revision_id, parents, lines,
1093
check_content=True):
1094
"""Store lines in inv_vf and return the sha1 of the inventory."""
1095
parents = [(parent,) for parent in parents]
1096
result = self.inventories.add_lines((revision_id,), parents, lines,
1097
check_content=check_content)[0]
1098
self.inventories._access.flush()
1101
def add_revision(self, revision_id, rev, inv=None, config=None):
1102
"""Add rev to the revision store as revision_id.
1104
:param revision_id: the revision id to use.
1105
:param rev: The revision object.
1106
:param inv: The inventory for the revision. if None, it will be looked
1107
up in the inventory storer
1108
:param config: If None no digital signature will be created.
1109
If supplied its signature_needed method will be used
1110
to determine if a signature should be made.
1112
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1114
_mod_revision.check_not_reserved_id(revision_id)
1115
if config is not None and config.signature_needed():
1117
inv = self.get_inventory(revision_id)
1118
plaintext = Testament(rev, inv).as_short_text()
1119
self.store_revision_signature(
1120
gpg.GPGStrategy(config), plaintext, revision_id)
1121
# check inventory present
1122
if not self.inventories.get_parent_map([(revision_id,)]):
1124
raise errors.WeaveRevisionNotPresent(revision_id,
1127
# yes, this is not suitable for adding with ghosts.
1128
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1131
key = (revision_id,)
1132
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1133
self._add_revision(rev)
1135
def _add_revision(self, revision):
1136
text = self._serializer.write_revision_to_string(revision)
1137
key = (revision.revision_id,)
1138
parents = tuple((parent,) for parent in revision.parent_ids)
1139
self.revisions.add_lines(key, parents, osutils.split_lines(text))
315
1141
def all_revision_ids(self):
316
1142
"""Returns a list of all the revision ids in the repository.
341
1167
self.control_files.break_lock()
1170
def _eliminate_revisions_not_present(self, revision_ids):
1171
"""Check every revision id in revision_ids to see if we have it.
1173
Returns a set of the present revisions.
1176
graph = self.get_graph()
1177
parent_map = graph.get_parent_map(revision_ids)
1178
# The old API returned a list, should this actually be a set?
1179
return parent_map.keys()
1181
def _check_inventories(self, checker):
1182
"""Check the inventories found from the revision scan.
1184
This is responsible for verifying the sha1 of inventories and
1185
creating a pending_keys set that covers data referenced by inventories.
1187
bar = ui.ui_factory.nested_progress_bar()
1189
self._do_check_inventories(checker, bar)
1193
def _do_check_inventories(self, checker, bar):
1194
"""Helper for _check_inventories."""
1196
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1197
kinds = ['chk_bytes', 'texts']
1198
count = len(checker.pending_keys)
1199
bar.update("inventories", 0, 2)
1200
current_keys = checker.pending_keys
1201
checker.pending_keys = {}
1202
# Accumulate current checks.
1203
for key in current_keys:
1204
if key[0] != 'inventories' and key[0] not in kinds:
1205
checker._report_items.append('unknown key type %r' % (key,))
1206
keys[key[0]].add(key[1:])
1207
if keys['inventories']:
1208
# NB: output order *should* be roughly sorted - topo or
1209
# inverse topo depending on repository - either way decent
1210
# to just delta against. However, pre-CHK formats didn't
1211
# try to optimise inventory layout on disk. As such the
1212
# pre-CHK code path does not use inventory deltas.
1214
for record in self.inventories.check(keys=keys['inventories']):
1215
if record.storage_kind == 'absent':
1216
checker._report_items.append(
1217
'Missing inventory {%s}' % (record.key,))
1219
last_object = self._check_record('inventories', record,
1220
checker, last_object,
1221
current_keys[('inventories',) + record.key])
1222
del keys['inventories']
1225
bar.update("texts", 1)
1226
while (checker.pending_keys or keys['chk_bytes']
1228
# Something to check.
1229
current_keys = checker.pending_keys
1230
checker.pending_keys = {}
1231
# Accumulate current checks.
1232
for key in current_keys:
1233
if key[0] not in kinds:
1234
checker._report_items.append('unknown key type %r' % (key,))
1235
keys[key[0]].add(key[1:])
1236
# Check the outermost kind only - inventories || chk_bytes || texts
1240
for record in getattr(self, kind).check(keys=keys[kind]):
1241
if record.storage_kind == 'absent':
1242
checker._report_items.append(
1243
'Missing %s {%s}' % (kind, record.key,))
1245
last_object = self._check_record(kind, record,
1246
checker, last_object, current_keys[(kind,) + record.key])
1250
def _check_record(self, kind, record, checker, last_object, item_data):
1251
"""Check a single text from this repository."""
1252
if kind == 'inventories':
1253
rev_id = record.key[0]
1254
inv = self._deserialise_inventory(rev_id,
1255
record.get_bytes_as('fulltext'))
1256
if last_object is not None:
1257
delta = inv._make_delta(last_object)
1258
for old_path, path, file_id, ie in delta:
1261
ie.check(checker, rev_id, inv)
1263
for path, ie in inv.iter_entries():
1264
ie.check(checker, rev_id, inv)
1265
if self._format.fast_deltas:
1267
elif kind == 'chk_bytes':
1268
# No code written to check chk_bytes for this repo format.
1269
checker._report_items.append(
1270
'unsupported key type chk_bytes for %s' % (record.key,))
1271
elif kind == 'texts':
1272
self._check_text(record, checker, item_data)
1274
checker._report_items.append(
1275
'unknown key type %s for %s' % (kind, record.key))
1277
def _check_text(self, record, checker, item_data):
1278
"""Check a single text."""
1279
# Check it is extractable.
1280
# TODO: check length.
1281
if record.storage_kind == 'chunked':
1282
chunks = record.get_bytes_as(record.storage_kind)
1283
sha1 = osutils.sha_strings(chunks)
1284
length = sum(map(len, chunks))
1286
content = record.get_bytes_as('fulltext')
1287
sha1 = osutils.sha_string(content)
1288
length = len(content)
1289
if item_data and sha1 != item_data[1]:
1290
checker._report_items.append(
1291
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1292
(record.key, sha1, item_data[1], item_data[2]))
344
1295
def create(a_bzrdir):
345
1296
"""Construct the current default format repository in a_bzrdir."""
909
1972
signature = gpg_strategy.sign(plaintext)
910
1973
self.add_signature_text(revision_id, signature)
912
1976
def add_signature_text(self, revision_id, signature):
913
"""Store a signature text for a revision.
915
:param revision_id: Revision id of the revision
916
:param signature: Signature text.
918
raise NotImplementedError(self.add_signature_text)
1977
self.signatures.add_lines((revision_id,), (),
1978
osutils.split_lines(signature))
1980
def find_text_key_references(self):
1981
"""Find the text key references within the repository.
1983
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1984
to whether they were referred to by the inventory of the
1985
revision_id that they contain. The inventory texts from all present
1986
revision ids are assessed to generate this report.
1988
revision_keys = self.revisions.keys()
1989
w = self.inventories
1990
pb = ui.ui_factory.nested_progress_bar()
1992
return self._find_text_key_references_from_xml_inventory_lines(
1993
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1997
def _find_text_key_references_from_xml_inventory_lines(self,
1999
"""Core routine for extracting references to texts from inventories.
2001
This performs the translation of xml lines to revision ids.
2003
:param line_iterator: An iterator of lines, origin_version_id
2004
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2005
to whether they were referred to by the inventory of the
2006
revision_id that they contain. Note that if that revision_id was
2007
not part of the line_iterator's output then False will be given -
2008
even though it may actually refer to that key.
2010
if not self._serializer.support_altered_by_hack:
2011
raise AssertionError(
2012
"_find_text_key_references_from_xml_inventory_lines only "
2013
"supported for branches which store inventory as unnested xml"
2014
", not on %r" % self)
2017
# this code needs to read every new line in every inventory for the
2018
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2019
# not present in one of those inventories is unnecessary but not
2020
# harmful because we are filtering by the revision id marker in the
2021
# inventory lines : we only select file ids altered in one of those
2022
# revisions. We don't need to see all lines in the inventory because
2023
# only those added in an inventory in rev X can contain a revision=X
2025
unescape_revid_cache = {}
2026
unescape_fileid_cache = {}
2028
# jam 20061218 In a big fetch, this handles hundreds of thousands
2029
# of lines, so it has had a lot of inlining and optimizing done.
2030
# Sorry that it is a little bit messy.
2031
# Move several functions to be local variables, since this is a long
2033
search = self._file_ids_altered_regex.search
2034
unescape = _unescape_xml
2035
setdefault = result.setdefault
2036
for line, line_key in line_iterator:
2037
match = search(line)
2040
# One call to match.group() returning multiple items is quite a
2041
# bit faster than 2 calls to match.group() each returning 1
2042
file_id, revision_id = match.group('file_id', 'revision_id')
2044
# Inlining the cache lookups helps a lot when you make 170,000
2045
# lines and 350k ids, versus 8.4 unique ids.
2046
# Using a cache helps in 2 ways:
2047
# 1) Avoids unnecessary decoding calls
2048
# 2) Re-uses cached strings, which helps in future set and
2050
# (2) is enough that removing encoding entirely along with
2051
# the cache (so we are using plain strings) results in no
2052
# performance improvement.
2054
revision_id = unescape_revid_cache[revision_id]
2056
unescaped = unescape(revision_id)
2057
unescape_revid_cache[revision_id] = unescaped
2058
revision_id = unescaped
2060
# Note that unconditionally unescaping means that we deserialise
2061
# every fileid, which for general 'pull' is not great, but we don't
2062
# really want to have some many fulltexts that this matters anyway.
2065
file_id = unescape_fileid_cache[file_id]
2067
unescaped = unescape(file_id)
2068
unescape_fileid_cache[file_id] = unescaped
2071
key = (file_id, revision_id)
2072
setdefault(key, False)
2073
if revision_id == line_key[-1]:
2077
def _inventory_xml_lines_for_keys(self, keys):
2078
"""Get a line iterator of the sort needed for findind references.
2080
Not relevant for non-xml inventory repositories.
2082
Ghosts in revision_keys are ignored.
2084
:param revision_keys: The revision keys for the inventories to inspect.
2085
:return: An iterator over (inventory line, revid) for the fulltexts of
2086
all of the xml inventories specified by revision_keys.
2088
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2089
for record in stream:
2090
if record.storage_kind != 'absent':
2091
chunks = record.get_bytes_as('chunked')
2092
revid = record.key[-1]
2093
lines = osutils.chunks_to_lines(chunks)
2097
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2099
"""Helper routine for fileids_altered_by_revision_ids.
2101
This performs the translation of xml lines to revision ids.
2103
:param line_iterator: An iterator of lines, origin_version_id
2104
:param revision_keys: The revision ids to filter for. This should be a
2105
set or other type which supports efficient __contains__ lookups, as
2106
the revision key from each parsed line will be looked up in the
2107
revision_keys filter.
2108
:return: a dictionary mapping altered file-ids to an iterable of
2109
revision_ids. Each altered file-ids has the exact revision_ids that
2110
altered it listed explicitly.
2112
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2113
line_iterator).iterkeys())
2114
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2115
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2116
self._inventory_xml_lines_for_keys(parent_keys)))
2117
new_keys = seen - parent_seen
2119
setdefault = result.setdefault
2120
for key in new_keys:
2121
setdefault(key[0], set()).add(key[-1])
920
2124
def _find_parent_ids_of_revisions(self, revision_ids):
921
2125
"""Find all parent ids that are mentioned in the revision graph.
951
2176
uniquely identify the file version in the caller's context. (Examples:
952
2177
an index number or a TreeTransform trans_id.)
2179
bytes_iterator is an iterable of bytestrings for the file. The
2180
kind of iterable and length of the bytestrings are unspecified, but for
2181
this implementation, it is a list of bytes produced by
2182
VersionedFile.get_record_stream().
954
2184
:param desired_files: a list of (file_id, revision_id, identifier)
957
raise NotImplementedError(self.iter_files_bytes)
2188
for file_id, revision_id, callable_data in desired_files:
2189
text_keys[(file_id, revision_id)] = callable_data
2190
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2191
if record.storage_kind == 'absent':
2192
raise errors.RevisionNotPresent(record.key, self)
2193
yield text_keys[record.key], record.get_bytes_as('chunked')
2195
def _generate_text_key_index(self, text_key_references=None,
2197
"""Generate a new text key index for the repository.
2199
This is an expensive function that will take considerable time to run.
2201
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2202
list of parents, also text keys. When a given key has no parents,
2203
the parents list will be [NULL_REVISION].
2205
# All revisions, to find inventory parents.
2206
if ancestors is None:
2207
graph = self.get_graph()
2208
ancestors = graph.get_parent_map(self.all_revision_ids())
2209
if text_key_references is None:
2210
text_key_references = self.find_text_key_references()
2211
pb = ui.ui_factory.nested_progress_bar()
2213
return self._do_generate_text_key_index(ancestors,
2214
text_key_references, pb)
2218
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2219
"""Helper for _generate_text_key_index to avoid deep nesting."""
2220
revision_order = tsort.topo_sort(ancestors)
2221
invalid_keys = set()
2223
for revision_id in revision_order:
2224
revision_keys[revision_id] = set()
2225
text_count = len(text_key_references)
2226
# a cache of the text keys to allow reuse; costs a dict of all the
2227
# keys, but saves a 2-tuple for every child of a given key.
2229
for text_key, valid in text_key_references.iteritems():
2231
invalid_keys.add(text_key)
2233
revision_keys[text_key[1]].add(text_key)
2234
text_key_cache[text_key] = text_key
2235
del text_key_references
2237
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2238
NULL_REVISION = _mod_revision.NULL_REVISION
2239
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2240
# too small for large or very branchy trees. However, for 55K path
2241
# trees, it would be easy to use too much memory trivially. Ideally we
2242
# could gauge this by looking at available real memory etc, but this is
2243
# always a tricky proposition.
2244
inventory_cache = lru_cache.LRUCache(10)
2245
batch_size = 10 # should be ~150MB on a 55K path tree
2246
batch_count = len(revision_order) / batch_size + 1
2248
pb.update("Calculating text parents", processed_texts, text_count)
2249
for offset in xrange(batch_count):
2250
to_query = revision_order[offset * batch_size:(offset + 1) *
2254
for revision_id in to_query:
2255
parent_ids = ancestors[revision_id]
2256
for text_key in revision_keys[revision_id]:
2257
pb.update("Calculating text parents", processed_texts)
2258
processed_texts += 1
2259
candidate_parents = []
2260
for parent_id in parent_ids:
2261
parent_text_key = (text_key[0], parent_id)
2263
check_parent = parent_text_key not in \
2264
revision_keys[parent_id]
2266
# the parent parent_id is a ghost:
2267
check_parent = False
2268
# truncate the derived graph against this ghost.
2269
parent_text_key = None
2271
# look at the parent commit details inventories to
2272
# determine possible candidates in the per file graph.
2275
inv = inventory_cache[parent_id]
2277
inv = self.revision_tree(parent_id).inventory
2278
inventory_cache[parent_id] = inv
2280
parent_entry = inv[text_key[0]]
2281
except (KeyError, errors.NoSuchId):
2283
if parent_entry is not None:
2285
text_key[0], parent_entry.revision)
2287
parent_text_key = None
2288
if parent_text_key is not None:
2289
candidate_parents.append(
2290
text_key_cache[parent_text_key])
2291
parent_heads = text_graph.heads(candidate_parents)
2292
new_parents = list(parent_heads)
2293
new_parents.sort(key=lambda x:candidate_parents.index(x))
2294
if new_parents == []:
2295
new_parents = [NULL_REVISION]
2296
text_index[text_key] = new_parents
2298
for text_key in invalid_keys:
2299
text_index[text_key] = [NULL_REVISION]
2302
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2303
"""Get an iterable listing the keys of all the data introduced by a set
2306
The keys will be ordered so that the corresponding items can be safely
2307
fetched and inserted in that order.
2309
:returns: An iterable producing tuples of (knit-kind, file-id,
2310
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2311
'revisions'. file-id is None unless knit-kind is 'file'.
2313
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2316
for result in self._find_non_file_keys_to_fetch(revision_ids):
2319
def _find_file_keys_to_fetch(self, revision_ids, pb):
2320
# XXX: it's a bit weird to control the inventory weave caching in this
2321
# generator. Ideally the caching would be done in fetch.py I think. Or
2322
# maybe this generator should explicitly have the contract that it
2323
# should not be iterated until the previously yielded item has been
2325
inv_w = self.inventories
2327
# file ids that changed
2328
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2330
num_file_ids = len(file_ids)
2331
for file_id, altered_versions in file_ids.iteritems():
2333
pb.update("Fetch texts", count, num_file_ids)
2335
yield ("file", file_id, altered_versions)
2337
def _find_non_file_keys_to_fetch(self, revision_ids):
2339
yield ("inventory", None, revision_ids)
2342
# XXX: Note ATM no callers actually pay attention to this return
2343
# instead they just use the list of revision ids and ignore
2344
# missing sigs. Consider removing this work entirely
2345
revisions_with_signatures = set(self.signatures.get_parent_map(
2346
[(r,) for r in revision_ids]))
2347
revisions_with_signatures = set(
2348
[r for (r,) in revisions_with_signatures])
2349
revisions_with_signatures.intersection_update(revision_ids)
2350
yield ("signatures", None, revisions_with_signatures)
2353
yield ("revisions", None, revision_ids)
2356
def get_inventory(self, revision_id):
2357
"""Get Inventory object by revision id."""
2358
return self.iter_inventories([revision_id]).next()
2360
def iter_inventories(self, revision_ids, ordering=None):
2361
"""Get many inventories by revision_ids.
2363
This will buffer some or all of the texts used in constructing the
2364
inventories in memory, but will only parse a single inventory at a
2367
:param revision_ids: The expected revision ids of the inventories.
2368
:param ordering: optional ordering, e.g. 'topological'. If not
2369
specified, the order of revision_ids will be preserved (by
2370
buffering if necessary).
2371
:return: An iterator of inventories.
2373
if ((None in revision_ids)
2374
or (_mod_revision.NULL_REVISION in revision_ids)):
2375
raise ValueError('cannot get null revision inventory')
2376
return self._iter_inventories(revision_ids, ordering)
2378
def _iter_inventories(self, revision_ids, ordering):
2379
"""single-document based inventory iteration."""
2380
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2381
for text, revision_id in inv_xmls:
2382
yield self._deserialise_inventory(revision_id, text)
2384
def _iter_inventory_xmls(self, revision_ids, ordering):
2385
if ordering is None:
2386
order_as_requested = True
2387
ordering = 'unordered'
2389
order_as_requested = False
2390
keys = [(revision_id,) for revision_id in revision_ids]
2393
if order_as_requested:
2394
key_iter = iter(keys)
2395
next_key = key_iter.next()
2396
stream = self.inventories.get_record_stream(keys, ordering, True)
2398
for record in stream:
2399
if record.storage_kind != 'absent':
2400
chunks = record.get_bytes_as('chunked')
2401
if order_as_requested:
2402
text_chunks[record.key] = chunks
2404
yield ''.join(chunks), record.key[-1]
2406
raise errors.NoSuchRevision(self, record.key)
2407
if order_as_requested:
2408
# Yield as many results as we can while preserving order.
2409
while next_key in text_chunks:
2410
chunks = text_chunks.pop(next_key)
2411
yield ''.join(chunks), next_key[-1]
2413
next_key = key_iter.next()
2414
except StopIteration:
2415
# We still want to fully consume the get_record_stream,
2416
# just in case it is not actually finished at this point
2420
def _deserialise_inventory(self, revision_id, xml):
2421
"""Transform the xml into an inventory object.
2423
:param revision_id: The expected revision id of the inventory.
2424
:param xml: A serialised inventory.
2426
result = self._serializer.read_inventory_from_string(xml, revision_id,
2427
entry_cache=self._inventory_entry_cache,
2428
return_from_cache=self._safe_to_return_from_cache)
2429
if result.revision_id != revision_id:
2430
raise AssertionError('revision id mismatch %s != %s' % (
2431
result.revision_id, revision_id))
2434
def get_serializer_format(self):
2435
return self._serializer.format_num
2438
def _get_inventory_xml(self, revision_id):
2439
"""Get serialized inventory as a string."""
2440
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2442
text, revision_id = texts.next()
2443
except StopIteration:
2444
raise errors.HistoryMissing(self, 'inventory', revision_id)
959
2447
def get_rev_id_for_revno(self, revno, known_pair):
960
2448
"""Return the revision id of a revno, given a later (revno, revid)
1277
2792
except UnicodeDecodeError:
1278
2793
raise errors.NonAsciiRevisionId(method, self)
2795
def revision_graph_can_have_wrong_parents(self):
2796
"""Is it possible for this repository to have a revision graph with
2799
If True, then this repository must also implement
2800
_find_inconsistent_revision_parents so that check and reconcile can
2801
check for inconsistencies before proceeding with other checks that may
2802
depend on the revision index being consistent.
2804
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2807
# remove these delegates a while after bzr 0.15
2808
def __make_delegated(name, from_module):
2809
def _deprecated_repository_forwarder():
2810
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2811
% (name, from_module),
2814
m = __import__(from_module, globals(), locals(), [name])
2816
return getattr(m, name)
2817
except AttributeError:
2818
raise AttributeError('module %s has no name %s'
2820
globals()[name] = _deprecated_repository_forwarder
2823
'AllInOneRepository',
2824
'WeaveMetaDirRepository',
2825
'PreSplitOutRepositoryFormat',
2826
'RepositoryFormat4',
2827
'RepositoryFormat5',
2828
'RepositoryFormat6',
2829
'RepositoryFormat7',
2831
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2835
'RepositoryFormatKnit',
2836
'RepositoryFormatKnit1',
2838
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2841
def install_revision(repository, rev, revision_tree):
2842
"""Install all revision data into a repository."""
2843
install_revisions(repository, [(rev, revision_tree, None)])
2846
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2847
"""Install all revision data into a repository.
2849
Accepts an iterable of revision, tree, signature tuples. The signature
2852
repository.start_write_group()
2854
inventory_cache = lru_cache.LRUCache(10)
2855
for n, (revision, revision_tree, signature) in enumerate(iterable):
2856
_install_revision(repository, revision, revision_tree, signature,
2859
pb.update('Transferring revisions', n + 1, num_revisions)
2861
repository.abort_write_group()
2864
repository.commit_write_group()
2867
def _install_revision(repository, rev, revision_tree, signature,
2869
"""Install all revision data into a repository."""
2870
present_parents = []
2872
for p_id in rev.parent_ids:
2873
if repository.has_revision(p_id):
2874
present_parents.append(p_id)
2875
parent_trees[p_id] = repository.revision_tree(p_id)
2877
parent_trees[p_id] = repository.revision_tree(
2878
_mod_revision.NULL_REVISION)
2880
inv = revision_tree.inventory
2881
entries = inv.iter_entries()
2882
# backwards compatibility hack: skip the root id.
2883
if not repository.supports_rich_root():
2884
path, root = entries.next()
2885
if root.revision != rev.revision_id:
2886
raise errors.IncompatibleRevision(repr(repository))
2888
for path, ie in entries:
2889
text_keys[(ie.file_id, ie.revision)] = ie
2890
text_parent_map = repository.texts.get_parent_map(text_keys)
2891
missing_texts = set(text_keys) - set(text_parent_map)
2892
# Add the texts that are not already present
2893
for text_key in missing_texts:
2894
ie = text_keys[text_key]
2896
# FIXME: TODO: The following loop overlaps/duplicates that done by
2897
# commit to determine parents. There is a latent/real bug here where
2898
# the parents inserted are not those commit would do - in particular
2899
# they are not filtered by heads(). RBC, AB
2900
for revision, tree in parent_trees.iteritems():
2901
if ie.file_id not in tree:
2903
parent_id = tree.inventory[ie.file_id].revision
2904
if parent_id in text_parents:
2906
text_parents.append((ie.file_id, parent_id))
2907
lines = revision_tree.get_file(ie.file_id).readlines()
2908
repository.texts.add_lines(text_key, text_parents, lines)
2910
# install the inventory
2911
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2912
# Cache this inventory
2913
inventory_cache[rev.revision_id] = inv
2915
basis_inv = inventory_cache[rev.parent_ids[0]]
2917
repository.add_inventory(rev.revision_id, inv, present_parents)
2919
delta = inv._make_delta(basis_inv)
2920
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2921
rev.revision_id, present_parents)
2923
repository.add_inventory(rev.revision_id, inv, present_parents)
2924
except errors.RevisionAlreadyPresent:
2926
if signature is not None:
2927
repository.add_signature_text(rev.revision_id, signature)
2928
repository.add_revision(rev.revision_id, rev, inv)
1281
2931
class MetaDirRepository(Repository):
1282
2932
"""Repositories in the new meta-dir layout.
1621
3297
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
1622
3298
format_registry.register_lazy(
1623
3299
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
1624
'bzrlib.repofmt.knitpack_repo',
3300
'bzrlib.repofmt.pack_repo',
1625
3301
'RepositoryFormatKnitPack1',
1627
3303
format_registry.register_lazy(
1628
3304
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
1629
'bzrlib.repofmt.knitpack_repo',
3305
'bzrlib.repofmt.pack_repo',
1630
3306
'RepositoryFormatKnitPack3',
1632
3308
format_registry.register_lazy(
1633
3309
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
1634
'bzrlib.repofmt.knitpack_repo',
3310
'bzrlib.repofmt.pack_repo',
1635
3311
'RepositoryFormatKnitPack4',
1637
3313
format_registry.register_lazy(
1638
3314
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
1639
'bzrlib.repofmt.knitpack_repo',
3315
'bzrlib.repofmt.pack_repo',
1640
3316
'RepositoryFormatKnitPack5',
1642
3318
format_registry.register_lazy(
1643
3319
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
1644
'bzrlib.repofmt.knitpack_repo',
3320
'bzrlib.repofmt.pack_repo',
1645
3321
'RepositoryFormatKnitPack5RichRoot',
1647
3323
format_registry.register_lazy(
1648
3324
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
1649
'bzrlib.repofmt.knitpack_repo',
3325
'bzrlib.repofmt.pack_repo',
1650
3326
'RepositoryFormatKnitPack5RichRootBroken',
1652
3328
format_registry.register_lazy(
1653
3329
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
1654
'bzrlib.repofmt.knitpack_repo',
3330
'bzrlib.repofmt.pack_repo',
1655
3331
'RepositoryFormatKnitPack6',
1657
3333
format_registry.register_lazy(
1658
3334
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
1659
'bzrlib.repofmt.knitpack_repo',
3335
'bzrlib.repofmt.pack_repo',
1660
3336
'RepositoryFormatKnitPack6RichRoot',
1662
format_registry.register_lazy(
1663
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
1664
'bzrlib.repofmt.groupcompress_repo',
1665
'RepositoryFormat2a',
1668
3339
# Development formats.
1669
# Check their docstrings to see if/when they are obsolete.
3340
# Obsolete but kept pending a CHK based subtree format.
1670
3341
format_registry.register_lazy(
1671
3342
("Bazaar development format 2 with subtree support "
1672
3343
"(needs bzr.dev from before 1.8)\n"),
1673
'bzrlib.repofmt.knitpack_repo',
3344
'bzrlib.repofmt.pack_repo',
1674
3345
'RepositoryFormatPackDevelopment2Subtree',
1676
format_registry.register_lazy(
1677
'Bazaar development format 8\n',
1678
'bzrlib.repofmt.groupcompress_repo',
1679
'RepositoryFormat2aSubtree',
3348
# 1.14->1.16 go below here
3349
format_registry.register_lazy(
3350
'Bazaar development format - group compression and chk inventory'
3351
' (needs bzr.dev from 1.14)\n',
3352
'bzrlib.repofmt.groupcompress_repo',
3353
'RepositoryFormatCHK1',
3356
format_registry.register_lazy(
3357
'Bazaar development format - chk repository with bencode revision '
3358
'serialization (needs bzr.dev from 1.16)\n',
3359
'bzrlib.repofmt.groupcompress_repo',
3360
'RepositoryFormatCHK2',
3362
format_registry.register_lazy(
3363
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3364
'bzrlib.repofmt.groupcompress_repo',
3365
'RepositoryFormat2a',
1721
3408
:param revision_id: if None all content is copied, if NULL_REVISION no
1722
3409
content is copied.
1725
raise NotImplementedError(self.fetch)
3413
ui.ui_factory.warn_experimental_format_fetch(self)
3414
from bzrlib.fetch import RepoFetcher
3415
# See <https://launchpad.net/bugs/456077> asking for a warning here
3416
if self.source._format.network_name() != self.target._format.network_name():
3417
ui.ui_factory.show_user_warning('cross_format_fetch',
3418
from_format=self.source._format,
3419
to_format=self.target._format)
3420
f = RepoFetcher(to_repository=self.target,
3421
from_repository=self.source,
3422
last_revision=revision_id,
3423
fetch_spec=fetch_spec,
3424
find_ghosts=find_ghosts)
3426
def _walk_to_common_revisions(self, revision_ids):
3427
"""Walk out from revision_ids in source to revisions target has.
3429
:param revision_ids: The start point for the search.
3430
:return: A set of revision ids.
3432
target_graph = self.target.get_graph()
3433
revision_ids = frozenset(revision_ids)
3434
missing_revs = set()
3435
source_graph = self.source.get_graph()
3436
# ensure we don't pay silly lookup costs.
3437
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3438
null_set = frozenset([_mod_revision.NULL_REVISION])
3439
searcher_exhausted = False
3443
# Iterate the searcher until we have enough next_revs
3444
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3446
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3447
next_revs.update(next_revs_part)
3448
ghosts.update(ghosts_part)
3449
except StopIteration:
3450
searcher_exhausted = True
3452
# If there are ghosts in the source graph, and the caller asked for
3453
# them, make sure that they are present in the target.
3454
# We don't care about other ghosts as we can't fetch them and
3455
# haven't been asked to.
3456
ghosts_to_check = set(revision_ids.intersection(ghosts))
3457
revs_to_get = set(next_revs).union(ghosts_to_check)
3459
have_revs = set(target_graph.get_parent_map(revs_to_get))
3460
# we always have NULL_REVISION present.
3461
have_revs = have_revs.union(null_set)
3462
# Check if the target is missing any ghosts we need.
3463
ghosts_to_check.difference_update(have_revs)
3465
# One of the caller's revision_ids is a ghost in both the
3466
# source and the target.
3467
raise errors.NoSuchRevision(
3468
self.source, ghosts_to_check.pop())
3469
missing_revs.update(next_revs - have_revs)
3470
# Because we may have walked past the original stop point, make
3471
# sure everything is stopped
3472
stop_revs = searcher.find_seen_ancestors(have_revs)
3473
searcher.stop_searching_any(stop_revs)
3474
if searcher_exhausted:
3476
return searcher.get_result()
1727
3478
@needs_read_lock
1728
def search_missing_revision_ids(self,
1729
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
1730
find_ghosts=True, revision_ids=None, if_present_ids=None,
3479
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1732
3480
"""Return the revision ids that source has that target does not.
1734
3482
:param revision_id: only return revision ids included by this
1736
:param revision_ids: return revision ids included by these
1737
revision_ids. NoSuchRevision will be raised if any of these
1738
revisions are not present.
1739
:param if_present_ids: like revision_ids, but will not cause
1740
NoSuchRevision if any of these are absent, instead they will simply
1741
not be in the result. This is useful for e.g. finding revisions
1742
to fetch for tags, which may reference absent revisions.
1743
3484
:param find_ghosts: If True find missing revisions in deep history
1744
3485
rather than just finding the surface difference.
1745
:param limit: Maximum number of revisions to return, topologically
1747
3486
:return: A bzrlib.graph.SearchResult.
1749
raise NotImplementedError(self.search_missing_revision_ids)
3488
# stop searching at found target revisions.
3489
if not find_ghosts and revision_id is not None:
3490
return self._walk_to_common_revisions([revision_id])
3491
# generic, possibly worst case, slow code path.
3492
target_ids = set(self.target.all_revision_ids())
3493
if revision_id is not None:
3494
source_ids = self.source.get_ancestry(revision_id)
3495
if source_ids[0] is not None:
3496
raise AssertionError()
3499
source_ids = self.source.all_revision_ids()
3500
result_set = set(source_ids).difference(target_ids)
3501
return self.source.revision_ids_to_search_result(result_set)
1752
3504
def _same_model(source, target):
1773
3525
"different serializers")
3528
class InterSameDataRepository(InterRepository):
3529
"""Code for converting between repositories that represent the same data.
3531
Data format and model must match for this to work.
3535
def _get_repo_format_to_test(self):
3536
"""Repository format for testing with.
3538
InterSameData can pull from subtree to subtree and from non-subtree to
3539
non-subtree, so we test this with the richest repository format.
3541
from bzrlib.repofmt import knitrepo
3542
return knitrepo.RepositoryFormatKnit3()
3545
def is_compatible(source, target):
3546
return InterRepository._same_model(source, target)
3549
class InterWeaveRepo(InterSameDataRepository):
3550
"""Optimised code paths between Weave based repositories.
3552
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3553
implemented lazy inter-object optimisation.
3557
def _get_repo_format_to_test(self):
3558
from bzrlib.repofmt import weaverepo
3559
return weaverepo.RepositoryFormat7()
3562
def is_compatible(source, target):
3563
"""Be compatible with known Weave formats.
3565
We don't test for the stores being of specific types because that
3566
could lead to confusing results, and there is no need to be
3569
from bzrlib.repofmt.weaverepo import (
3575
return (isinstance(source._format, (RepositoryFormat5,
3577
RepositoryFormat7)) and
3578
isinstance(target._format, (RepositoryFormat5,
3580
RepositoryFormat7)))
3581
except AttributeError:
3585
def copy_content(self, revision_id=None):
3586
"""See InterRepository.copy_content()."""
3587
# weave specific optimised path:
3589
self.target.set_make_working_trees(self.source.make_working_trees())
3590
except (errors.RepositoryUpgradeRequired, NotImplemented):
3592
# FIXME do not peek!
3593
if self.source._transport.listable():
3594
pb = ui.ui_factory.nested_progress_bar()
3596
self.target.texts.insert_record_stream(
3597
self.source.texts.get_record_stream(
3598
self.source.texts.keys(), 'topological', False))
3599
pb.update('Copying inventory', 0, 1)
3600
self.target.inventories.insert_record_stream(
3601
self.source.inventories.get_record_stream(
3602
self.source.inventories.keys(), 'topological', False))
3603
self.target.signatures.insert_record_stream(
3604
self.source.signatures.get_record_stream(
3605
self.source.signatures.keys(),
3607
self.target.revisions.insert_record_stream(
3608
self.source.revisions.get_record_stream(
3609
self.source.revisions.keys(),
3610
'topological', True))
3614
self.target.fetch(self.source, revision_id=revision_id)
3617
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3618
"""See InterRepository.missing_revision_ids()."""
3619
# we want all revisions to satisfy revision_id in source.
3620
# but we don't want to stat every file here and there.
3621
# we want then, all revisions other needs to satisfy revision_id
3622
# checked, but not those that we have locally.
3623
# so the first thing is to get a subset of the revisions to
3624
# satisfy revision_id in source, and then eliminate those that
3625
# we do already have.
3626
# this is slow on high latency connection to self, but as this
3627
# disk format scales terribly for push anyway due to rewriting
3628
# inventory.weave, this is considered acceptable.
3630
if revision_id is not None:
3631
source_ids = self.source.get_ancestry(revision_id)
3632
if source_ids[0] is not None:
3633
raise AssertionError()
3636
source_ids = self.source._all_possible_ids()
3637
source_ids_set = set(source_ids)
3638
# source_ids is the worst possible case we may need to pull.
3639
# now we want to filter source_ids against what we actually
3640
# have in target, but don't try to check for existence where we know
3641
# we do not have a revision as that would be pointless.
3642
target_ids = set(self.target._all_possible_ids())
3643
possibly_present_revisions = target_ids.intersection(source_ids_set)
3644
actually_present_revisions = set(
3645
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3646
required_revisions = source_ids_set.difference(actually_present_revisions)
3647
if revision_id is not None:
3648
# we used get_ancestry to determine source_ids then we are assured all
3649
# revisions referenced are present as they are installed in topological order.
3650
# and the tip revision was validated by get_ancestry.
3651
result_set = required_revisions
3653
# if we just grabbed the possibly available ids, then
3654
# we only have an estimate of whats available and need to validate
3655
# that against the revision records.
3657
self.source._eliminate_revisions_not_present(required_revisions))
3658
return self.source.revision_ids_to_search_result(result_set)
3661
class InterKnitRepo(InterSameDataRepository):
3662
"""Optimised code paths between Knit based repositories."""
3665
def _get_repo_format_to_test(self):
3666
from bzrlib.repofmt import knitrepo
3667
return knitrepo.RepositoryFormatKnit1()
3670
def is_compatible(source, target):
3671
"""Be compatible with known Knit formats.
3673
We don't test for the stores being of specific types because that
3674
could lead to confusing results, and there is no need to be
3677
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3679
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3680
isinstance(target._format, RepositoryFormatKnit))
3681
except AttributeError:
3683
return are_knits and InterRepository._same_model(source, target)
3686
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3687
"""See InterRepository.missing_revision_ids()."""
3688
if revision_id is not None:
3689
source_ids = self.source.get_ancestry(revision_id)
3690
if source_ids[0] is not None:
3691
raise AssertionError()
3694
source_ids = self.source.all_revision_ids()
3695
source_ids_set = set(source_ids)
3696
# source_ids is the worst possible case we may need to pull.
3697
# now we want to filter source_ids against what we actually
3698
# have in target, but don't try to check for existence where we know
3699
# we do not have a revision as that would be pointless.
3700
target_ids = set(self.target.all_revision_ids())
3701
possibly_present_revisions = target_ids.intersection(source_ids_set)
3702
actually_present_revisions = set(
3703
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3704
required_revisions = source_ids_set.difference(actually_present_revisions)
3705
if revision_id is not None:
3706
# we used get_ancestry to determine source_ids then we are assured all
3707
# revisions referenced are present as they are installed in topological order.
3708
# and the tip revision was validated by get_ancestry.
3709
result_set = required_revisions
3711
# if we just grabbed the possibly available ids, then
3712
# we only have an estimate of whats available and need to validate
3713
# that against the revision records.
3715
self.source._eliminate_revisions_not_present(required_revisions))
3716
return self.source.revision_ids_to_search_result(result_set)
3719
class InterDifferingSerializer(InterRepository):
3722
def _get_repo_format_to_test(self):
3726
def is_compatible(source, target):
3727
"""Be compatible with Knit2 source and Knit3 target"""
3728
# This is redundant with format.check_conversion_target(), however that
3729
# raises an exception, and we just want to say "False" as in we won't
3730
# support converting between these formats.
3731
if 'IDS_never' in debug.debug_flags:
3733
if source.supports_rich_root() and not target.supports_rich_root():
3735
if (source._format.supports_tree_reference
3736
and not target._format.supports_tree_reference):
3738
if target._fallback_repositories and target._format.supports_chks:
3739
# IDS doesn't know how to copy CHKs for the parent inventories it
3740
# adds to stacked repos.
3742
if 'IDS_always' in debug.debug_flags:
3744
# Only use this code path for local source and target. IDS does far
3745
# too much IO (both bandwidth and roundtrips) over a network.
3746
if not source.bzrdir.transport.base.startswith('file:///'):
3748
if not target.bzrdir.transport.base.startswith('file:///'):
3752
def _get_trees(self, revision_ids, cache):
3754
for rev_id in revision_ids:
3756
possible_trees.append((rev_id, cache[rev_id]))
3758
# Not cached, but inventory might be present anyway.
3760
tree = self.source.revision_tree(rev_id)
3761
except errors.NoSuchRevision:
3762
# Nope, parent is ghost.
3765
cache[rev_id] = tree
3766
possible_trees.append((rev_id, tree))
3767
return possible_trees
3769
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3770
"""Get the best delta and base for this revision.
3772
:return: (basis_id, delta)
3775
# Generate deltas against each tree, to find the shortest.
3776
texts_possibly_new_in_tree = set()
3777
for basis_id, basis_tree in possible_trees:
3778
delta = tree.inventory._make_delta(basis_tree.inventory)
3779
for old_path, new_path, file_id, new_entry in delta:
3780
if new_path is None:
3781
# This file_id isn't present in the new rev, so we don't
3785
# Rich roots are handled elsewhere...
3787
kind = new_entry.kind
3788
if kind != 'directory' and kind != 'file':
3789
# No text record associated with this inventory entry.
3791
# This is a directory or file that has changed somehow.
3792
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3793
deltas.append((len(delta), basis_id, delta))
3795
return deltas[0][1:]
3797
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3798
"""Find all parent revisions that are absent, but for which the
3799
inventory is present, and copy those inventories.
3801
This is necessary to preserve correctness when the source is stacked
3802
without fallbacks configured. (Note that in cases like upgrade the
3803
source may be not have _fallback_repositories even though it is
3807
for parents in parent_map.values():
3808
parent_revs.update(parents)
3809
present_parents = self.source.get_parent_map(parent_revs)
3810
absent_parents = set(parent_revs).difference(present_parents)
3811
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3812
(rev_id,) for rev_id in absent_parents)
3813
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3814
for parent_tree in self.source.revision_trees(parent_inv_ids):
3815
current_revision_id = parent_tree.get_revision_id()
3816
parents_parents_keys = parent_invs_keys_for_stacking[
3817
(current_revision_id,)]
3818
parents_parents = [key[-1] for key in parents_parents_keys]
3819
basis_id = _mod_revision.NULL_REVISION
3820
basis_tree = self.source.revision_tree(basis_id)
3821
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3822
self.target.add_inventory_by_delta(
3823
basis_id, delta, current_revision_id, parents_parents)
3824
cache[current_revision_id] = parent_tree
3826
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3827
"""Fetch across a few revisions.
3829
:param revision_ids: The revisions to copy
3830
:param basis_id: The revision_id of a tree that must be in cache, used
3831
as a basis for delta when no other base is available
3832
:param cache: A cache of RevisionTrees that we can use.
3833
:param a_graph: A Graph object to determine the heads() of the
3834
rich-root data stream.
3835
:return: The revision_id of the last converted tree. The RevisionTree
3836
for it will be in cache
3838
# Walk though all revisions; get inventory deltas, copy referenced
3839
# texts that delta references, insert the delta, revision and
3841
root_keys_to_create = set()
3844
pending_revisions = []
3845
parent_map = self.source.get_parent_map(revision_ids)
3846
self._fetch_parent_invs_for_stacking(parent_map, cache)
3847
self.source._safe_to_return_from_cache = True
3848
for tree in self.source.revision_trees(revision_ids):
3849
# Find a inventory delta for this revision.
3850
# Find text entries that need to be copied, too.
3851
current_revision_id = tree.get_revision_id()
3852
parent_ids = parent_map.get(current_revision_id, ())
3853
parent_trees = self._get_trees(parent_ids, cache)
3854
possible_trees = list(parent_trees)
3855
if len(possible_trees) == 0:
3856
# There either aren't any parents, or the parents are ghosts,
3857
# so just use the last converted tree.
3858
possible_trees.append((basis_id, cache[basis_id]))
3859
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3861
revision = self.source.get_revision(current_revision_id)
3862
pending_deltas.append((basis_id, delta,
3863
current_revision_id, revision.parent_ids))
3864
if self._converting_to_rich_root:
3865
self._revision_id_to_root_id[current_revision_id] = \
3867
# Determine which texts are in present in this revision but not in
3868
# any of the available parents.
3869
texts_possibly_new_in_tree = set()
3870
for old_path, new_path, file_id, entry in delta:
3871
if new_path is None:
3872
# This file_id isn't present in the new rev
3876
if not self.target.supports_rich_root():
3877
# The target doesn't support rich root, so we don't
3880
if self._converting_to_rich_root:
3881
# This can't be copied normally, we have to insert
3883
root_keys_to_create.add((file_id, entry.revision))
3886
texts_possibly_new_in_tree.add((file_id, entry.revision))
3887
for basis_id, basis_tree in possible_trees:
3888
basis_inv = basis_tree.inventory
3889
for file_key in list(texts_possibly_new_in_tree):
3890
file_id, file_revision = file_key
3892
entry = basis_inv[file_id]
3893
except errors.NoSuchId:
3895
if entry.revision == file_revision:
3896
texts_possibly_new_in_tree.remove(file_key)
3897
text_keys.update(texts_possibly_new_in_tree)
3898
pending_revisions.append(revision)
3899
cache[current_revision_id] = tree
3900
basis_id = current_revision_id
3901
self.source._safe_to_return_from_cache = False
3903
from_texts = self.source.texts
3904
to_texts = self.target.texts
3905
if root_keys_to_create:
3906
root_stream = _mod_fetch._new_root_data_stream(
3907
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3908
self.source, graph=a_graph)
3909
to_texts.insert_record_stream(root_stream)
3910
to_texts.insert_record_stream(from_texts.get_record_stream(
3911
text_keys, self.target._format._fetch_order,
3912
not self.target._format._fetch_uses_deltas))
3913
# insert inventory deltas
3914
for delta in pending_deltas:
3915
self.target.add_inventory_by_delta(*delta)
3916
if self.target._fallback_repositories:
3917
# Make sure this stacked repository has all the parent inventories
3918
# for the new revisions that we are about to insert. We do this
3919
# before adding the revisions so that no revision is added until
3920
# all the inventories it may depend on are added.
3921
# Note that this is overzealous, as we may have fetched these in an
3924
revision_ids = set()
3925
for revision in pending_revisions:
3926
revision_ids.add(revision.revision_id)
3927
parent_ids.update(revision.parent_ids)
3928
parent_ids.difference_update(revision_ids)
3929
parent_ids.discard(_mod_revision.NULL_REVISION)
3930
parent_map = self.source.get_parent_map(parent_ids)
3931
# we iterate over parent_map and not parent_ids because we don't
3932
# want to try copying any revision which is a ghost
3933
for parent_tree in self.source.revision_trees(parent_map):
3934
current_revision_id = parent_tree.get_revision_id()
3935
parents_parents = parent_map[current_revision_id]
3936
possible_trees = self._get_trees(parents_parents, cache)
3937
if len(possible_trees) == 0:
3938
# There either aren't any parents, or the parents are
3939
# ghosts, so just use the last converted tree.
3940
possible_trees.append((basis_id, cache[basis_id]))
3941
basis_id, delta = self._get_delta_for_revision(parent_tree,
3942
parents_parents, possible_trees)
3943
self.target.add_inventory_by_delta(
3944
basis_id, delta, current_revision_id, parents_parents)
3945
# insert signatures and revisions
3946
for revision in pending_revisions:
3948
signature = self.source.get_signature_text(
3949
revision.revision_id)
3950
self.target.add_signature_text(revision.revision_id,
3952
except errors.NoSuchRevision:
3954
self.target.add_revision(revision.revision_id, revision)
3957
def _fetch_all_revisions(self, revision_ids, pb):
3958
"""Fetch everything for the list of revisions.
3960
:param revision_ids: The list of revisions to fetch. Must be in
3962
:param pb: A ProgressTask
3965
basis_id, basis_tree = self._get_basis(revision_ids[0])
3967
cache = lru_cache.LRUCache(100)
3968
cache[basis_id] = basis_tree
3969
del basis_tree # We don't want to hang on to it here
3971
if self._converting_to_rich_root and len(revision_ids) > 100:
3972
a_graph = _mod_fetch._get_rich_root_heads_graph(self.source,
3977
for offset in range(0, len(revision_ids), batch_size):
3978
self.target.start_write_group()
3980
pb.update('Transferring revisions', offset,
3982
batch = revision_ids[offset:offset+batch_size]
3983
basis_id = self._fetch_batch(batch, basis_id, cache,
3986
self.source._safe_to_return_from_cache = False
3987
self.target.abort_write_group()
3990
hint = self.target.commit_write_group()
3993
if hints and self.target._format.pack_compresses:
3994
self.target.pack(hint=hints)
3995
pb.update('Transferring revisions', len(revision_ids),
3999
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
4001
"""See InterRepository.fetch()."""
4002
if fetch_spec is not None:
4003
raise AssertionError("Not implemented yet...")
4004
ui.ui_factory.warn_experimental_format_fetch(self)
4005
if (not self.source.supports_rich_root()
4006
and self.target.supports_rich_root()):
4007
self._converting_to_rich_root = True
4008
self._revision_id_to_root_id = {}
4010
self._converting_to_rich_root = False
4011
# See <https://launchpad.net/bugs/456077> asking for a warning here
4012
if self.source._format.network_name() != self.target._format.network_name():
4013
ui.ui_factory.show_user_warning('cross_format_fetch',
4014
from_format=self.source._format,
4015
to_format=self.target._format)
4016
revision_ids = self.target.search_missing_revision_ids(self.source,
4017
revision_id, find_ghosts=find_ghosts).get_keys()
4018
if not revision_ids:
4020
revision_ids = tsort.topo_sort(
4021
self.source.get_graph().get_parent_map(revision_ids))
4022
if not revision_ids:
4024
# Walk though all revisions; get inventory deltas, copy referenced
4025
# texts that delta references, insert the delta, revision and
4028
my_pb = ui.ui_factory.nested_progress_bar()
4031
symbol_versioning.warn(
4032
symbol_versioning.deprecated_in((1, 14, 0))
4033
% "pb parameter to fetch()")
4036
self._fetch_all_revisions(revision_ids, pb)
4038
if my_pb is not None:
4040
return len(revision_ids), 0
4042
def _get_basis(self, first_revision_id):
4043
"""Get a revision and tree which exists in the target.
4045
This assumes that first_revision_id is selected for transmission
4046
because all other ancestors are already present. If we can't find an
4047
ancestor we fall back to NULL_REVISION since we know that is safe.
4049
:return: (basis_id, basis_tree)
4051
first_rev = self.source.get_revision(first_revision_id)
4053
basis_id = first_rev.parent_ids[0]
4054
# only valid as a basis if the target has it
4055
self.target.get_revision(basis_id)
4056
# Try to get a basis tree - if its a ghost it will hit the
4057
# NoSuchRevision case.
4058
basis_tree = self.source.revision_tree(basis_id)
4059
except (IndexError, errors.NoSuchRevision):
4060
basis_id = _mod_revision.NULL_REVISION
4061
basis_tree = self.source.revision_tree(basis_id)
4062
return basis_id, basis_tree
4065
InterRepository.register_optimiser(InterDifferingSerializer)
4066
InterRepository.register_optimiser(InterSameDataRepository)
4067
InterRepository.register_optimiser(InterWeaveRepo)
4068
InterRepository.register_optimiser(InterKnitRepo)
1776
4071
class CopyConverter(object):
1777
4072
"""A repository conversion tool which just performs a copy of the content.
1832
4229
return revision_graph
4232
class StreamSink(object):
4233
"""An object that can insert a stream into a repository.
4235
This interface handles the complexity of reserialising inventories and
4236
revisions from different formats, and allows unidirectional insertion into
4237
stacked repositories without looking for the missing basis parents
4241
def __init__(self, target_repo):
4242
self.target_repo = target_repo
4244
def insert_stream(self, stream, src_format, resume_tokens):
4245
"""Insert a stream's content into the target repository.
4247
:param src_format: a bzr repository format.
4249
:return: a list of resume tokens and an iterable of keys additional
4250
items required before the insertion can be completed.
4252
self.target_repo.lock_write()
4255
self.target_repo.resume_write_group(resume_tokens)
4258
self.target_repo.start_write_group()
4261
# locked_insert_stream performs a commit|suspend.
4262
return self._locked_insert_stream(stream, src_format, is_resume)
4264
self.target_repo.abort_write_group(suppress_errors=True)
4267
self.target_repo.unlock()
4269
def _locked_insert_stream(self, stream, src_format, is_resume):
4270
to_serializer = self.target_repo._format._serializer
4271
src_serializer = src_format._serializer
4273
if to_serializer == src_serializer:
4274
# If serializers match and the target is a pack repository, set the
4275
# write cache size on the new pack. This avoids poor performance
4276
# on transports where append is unbuffered (such as
4277
# RemoteTransport). This is safe to do because nothing should read
4278
# back from the target repository while a stream with matching
4279
# serialization is being inserted.
4280
# The exception is that a delta record from the source that should
4281
# be a fulltext may need to be expanded by the target (see
4282
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4283
# explicitly flush any buffered writes first in that rare case.
4285
new_pack = self.target_repo._pack_collection._new_pack
4286
except AttributeError:
4287
# Not a pack repository
4290
new_pack.set_write_cache_size(1024*1024)
4291
for substream_type, substream in stream:
4292
if 'stream' in debug.debug_flags:
4293
mutter('inserting substream: %s', substream_type)
4294
if substream_type == 'texts':
4295
self.target_repo.texts.insert_record_stream(substream)
4296
elif substream_type == 'inventories':
4297
if src_serializer == to_serializer:
4298
self.target_repo.inventories.insert_record_stream(
4301
self._extract_and_insert_inventories(
4302
substream, src_serializer)
4303
elif substream_type == 'inventory-deltas':
4304
self._extract_and_insert_inventory_deltas(
4305
substream, src_serializer)
4306
elif substream_type == 'chk_bytes':
4307
# XXX: This doesn't support conversions, as it assumes the
4308
# conversion was done in the fetch code.
4309
self.target_repo.chk_bytes.insert_record_stream(substream)
4310
elif substream_type == 'revisions':
4311
# This may fallback to extract-and-insert more often than
4312
# required if the serializers are different only in terms of
4314
if src_serializer == to_serializer:
4315
self.target_repo.revisions.insert_record_stream(
4318
self._extract_and_insert_revisions(substream,
4320
elif substream_type == 'signatures':
4321
self.target_repo.signatures.insert_record_stream(substream)
4323
raise AssertionError('kaboom! %s' % (substream_type,))
4324
# Done inserting data, and the missing_keys calculations will try to
4325
# read back from the inserted data, so flush the writes to the new pack
4326
# (if this is pack format).
4327
if new_pack is not None:
4328
new_pack._write_data('', flush=True)
4329
# Find all the new revisions (including ones from resume_tokens)
4330
missing_keys = self.target_repo.get_missing_parent_inventories(
4331
check_for_missing_texts=is_resume)
4333
for prefix, versioned_file in (
4334
('texts', self.target_repo.texts),
4335
('inventories', self.target_repo.inventories),
4336
('revisions', self.target_repo.revisions),
4337
('signatures', self.target_repo.signatures),
4338
('chk_bytes', self.target_repo.chk_bytes),
4340
if versioned_file is None:
4342
# TODO: key is often going to be a StaticTuple object
4343
# I don't believe we can define a method by which
4344
# (prefix,) + StaticTuple will work, though we could
4345
# define a StaticTuple.sq_concat that would allow you to
4346
# pass in either a tuple or a StaticTuple as the second
4347
# object, so instead we could have:
4348
# StaticTuple(prefix) + key here...
4349
missing_keys.update((prefix,) + key for key in
4350
versioned_file.get_missing_compression_parent_keys())
4351
except NotImplementedError:
4352
# cannot even attempt suspending, and missing would have failed
4353
# during stream insertion.
4354
missing_keys = set()
4357
# suspend the write group and tell the caller what we is
4358
# missing. We know we can suspend or else we would not have
4359
# entered this code path. (All repositories that can handle
4360
# missing keys can handle suspending a write group).
4361
write_group_tokens = self.target_repo.suspend_write_group()
4362
return write_group_tokens, missing_keys
4363
hint = self.target_repo.commit_write_group()
4364
if (to_serializer != src_serializer and
4365
self.target_repo._format.pack_compresses):
4366
self.target_repo.pack(hint=hint)
4369
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4370
target_rich_root = self.target_repo._format.rich_root_data
4371
target_tree_refs = self.target_repo._format.supports_tree_reference
4372
for record in substream:
4373
# Insert the delta directly
4374
inventory_delta_bytes = record.get_bytes_as('fulltext')
4375
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4377
parse_result = deserialiser.parse_text_bytes(
4378
inventory_delta_bytes)
4379
except inventory_delta.IncompatibleInventoryDelta, err:
4380
trace.mutter("Incompatible delta: %s", err.msg)
4381
raise errors.IncompatibleRevision(self.target_repo._format)
4382
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4383
revision_id = new_id
4384
parents = [key[0] for key in record.parents]
4385
self.target_repo.add_inventory_by_delta(
4386
basis_id, inv_delta, revision_id, parents)
4388
def _extract_and_insert_inventories(self, substream, serializer,
4390
"""Generate a new inventory versionedfile in target, converting data.
4392
The inventory is retrieved from the source, (deserializing it), and
4393
stored in the target (reserializing it in a different format).
4395
target_rich_root = self.target_repo._format.rich_root_data
4396
target_tree_refs = self.target_repo._format.supports_tree_reference
4397
for record in substream:
4398
# It's not a delta, so it must be a fulltext in the source
4399
# serializer's format.
4400
bytes = record.get_bytes_as('fulltext')
4401
revision_id = record.key[0]
4402
inv = serializer.read_inventory_from_string(bytes, revision_id)
4403
parents = [key[0] for key in record.parents]
4404
self.target_repo.add_inventory(revision_id, inv, parents)
4405
# No need to keep holding this full inv in memory when the rest of
4406
# the substream is likely to be all deltas.
4409
def _extract_and_insert_revisions(self, substream, serializer):
4410
for record in substream:
4411
bytes = record.get_bytes_as('fulltext')
4412
revision_id = record.key[0]
4413
rev = serializer.read_revision_from_string(bytes)
4414
if rev.revision_id != revision_id:
4415
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4416
self.target_repo.add_revision(revision_id, rev)
4419
if self.target_repo._format._fetch_reconcile:
4420
self.target_repo.reconcile()
4423
class StreamSource(object):
4424
"""A source of a stream for fetching between repositories."""
4426
def __init__(self, from_repository, to_format):
4427
"""Create a StreamSource streaming from from_repository."""
4428
self.from_repository = from_repository
4429
self.to_format = to_format
4431
def delta_on_metadata(self):
4432
"""Return True if delta's are permitted on metadata streams.
4434
That is on revisions and signatures.
4436
src_serializer = self.from_repository._format._serializer
4437
target_serializer = self.to_format._serializer
4438
return (self.to_format._fetch_uses_deltas and
4439
src_serializer == target_serializer)
4441
def _fetch_revision_texts(self, revs):
4442
# fetch signatures first and then the revision texts
4443
# may need to be a InterRevisionStore call here.
4444
from_sf = self.from_repository.signatures
4445
# A missing signature is just skipped.
4446
keys = [(rev_id,) for rev_id in revs]
4447
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4449
self.to_format._fetch_order,
4450
not self.to_format._fetch_uses_deltas))
4451
# If a revision has a delta, this is actually expanded inside the
4452
# insert_record_stream code now, which is an alternate fix for
4454
from_rf = self.from_repository.revisions
4455
revisions = from_rf.get_record_stream(
4457
self.to_format._fetch_order,
4458
not self.delta_on_metadata())
4459
return [('signatures', signatures), ('revisions', revisions)]
4461
def _generate_root_texts(self, revs):
4462
"""This will be called by get_stream between fetching weave texts and
4463
fetching the inventory weave.
4465
if self._rich_root_upgrade():
4466
return _mod_fetch.Inter1and2Helper(
4467
self.from_repository).generate_root_texts(revs)
4471
def get_stream(self, search):
4473
revs = search.get_keys()
4474
graph = self.from_repository.get_graph()
4475
revs = tsort.topo_sort(graph.get_parent_map(revs))
4476
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4478
for knit_kind, file_id, revisions in data_to_fetch:
4479
if knit_kind != phase:
4481
# Make a new progress bar for this phase
4482
if knit_kind == "file":
4483
# Accumulate file texts
4484
text_keys.extend([(file_id, revision) for revision in
4486
elif knit_kind == "inventory":
4487
# Now copy the file texts.
4488
from_texts = self.from_repository.texts
4489
yield ('texts', from_texts.get_record_stream(
4490
text_keys, self.to_format._fetch_order,
4491
not self.to_format._fetch_uses_deltas))
4492
# Cause an error if a text occurs after we have done the
4495
# Before we process the inventory we generate the root
4496
# texts (if necessary) so that the inventories references
4498
for _ in self._generate_root_texts(revs):
4500
# we fetch only the referenced inventories because we do not
4501
# know for unselected inventories whether all their required
4502
# texts are present in the other repository - it could be
4504
for info in self._get_inventory_stream(revs):
4506
elif knit_kind == "signatures":
4507
# Nothing to do here; this will be taken care of when
4508
# _fetch_revision_texts happens.
4510
elif knit_kind == "revisions":
4511
for record in self._fetch_revision_texts(revs):
4514
raise AssertionError("Unknown knit kind %r" % knit_kind)
4516
def get_stream_for_missing_keys(self, missing_keys):
4517
# missing keys can only occur when we are byte copying and not
4518
# translating (because translation means we don't send
4519
# unreconstructable deltas ever).
4521
keys['texts'] = set()
4522
keys['revisions'] = set()
4523
keys['inventories'] = set()
4524
keys['chk_bytes'] = set()
4525
keys['signatures'] = set()
4526
for key in missing_keys:
4527
keys[key[0]].add(key[1:])
4528
if len(keys['revisions']):
4529
# If we allowed copying revisions at this point, we could end up
4530
# copying a revision without copying its required texts: a
4531
# violation of the requirements for repository integrity.
4532
raise AssertionError(
4533
'cannot copy revisions to fill in missing deltas %s' % (
4534
keys['revisions'],))
4535
for substream_kind, keys in keys.iteritems():
4536
vf = getattr(self.from_repository, substream_kind)
4537
if vf is None and keys:
4538
raise AssertionError(
4539
"cannot fill in keys for a versioned file we don't"
4540
" have: %s needs %s" % (substream_kind, keys))
4542
# No need to stream something we don't have
4544
if substream_kind == 'inventories':
4545
# Some missing keys are genuinely ghosts, filter those out.
4546
present = self.from_repository.inventories.get_parent_map(keys)
4547
revs = [key[0] for key in present]
4548
# Get the inventory stream more-or-less as we do for the
4549
# original stream; there's no reason to assume that records
4550
# direct from the source will be suitable for the sink. (Think
4551
# e.g. 2a -> 1.9-rich-root).
4552
for info in self._get_inventory_stream(revs, missing=True):
4556
# Ask for full texts always so that we don't need more round trips
4557
# after this stream.
4558
# Some of the missing keys are genuinely ghosts, so filter absent
4559
# records. The Sink is responsible for doing another check to
4560
# ensure that ghosts don't introduce missing data for future
4562
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4563
self.to_format._fetch_order, True))
4564
yield substream_kind, stream
4566
def inventory_fetch_order(self):
4567
if self._rich_root_upgrade():
4568
return 'topological'
4570
return self.to_format._fetch_order
4572
def _rich_root_upgrade(self):
4573
return (not self.from_repository._format.rich_root_data and
4574
self.to_format.rich_root_data)
4576
def _get_inventory_stream(self, revision_ids, missing=False):
4577
from_format = self.from_repository._format
4578
if (from_format.supports_chks and self.to_format.supports_chks and
4579
from_format.network_name() == self.to_format.network_name()):
4580
raise AssertionError(
4581
"this case should be handled by GroupCHKStreamSource")
4582
elif 'forceinvdeltas' in debug.debug_flags:
4583
return self._get_convertable_inventory_stream(revision_ids,
4584
delta_versus_null=missing)
4585
elif from_format.network_name() == self.to_format.network_name():
4587
return self._get_simple_inventory_stream(revision_ids,
4589
elif (not from_format.supports_chks and not self.to_format.supports_chks
4590
and from_format._serializer == self.to_format._serializer):
4591
# Essentially the same format.
4592
return self._get_simple_inventory_stream(revision_ids,
4595
# Any time we switch serializations, we want to use an
4596
# inventory-delta based approach.
4597
return self._get_convertable_inventory_stream(revision_ids,
4598
delta_versus_null=missing)
4600
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4601
# NB: This currently reopens the inventory weave in source;
4602
# using a single stream interface instead would avoid this.
4603
from_weave = self.from_repository.inventories
4605
delta_closure = True
4607
delta_closure = not self.delta_on_metadata()
4608
yield ('inventories', from_weave.get_record_stream(
4609
[(rev_id,) for rev_id in revision_ids],
4610
self.inventory_fetch_order(), delta_closure))
4612
def _get_convertable_inventory_stream(self, revision_ids,
4613
delta_versus_null=False):
4614
# The two formats are sufficiently different that there is no fast
4615
# path, so we need to send just inventorydeltas, which any
4616
# sufficiently modern client can insert into any repository.
4617
# The StreamSink code expects to be able to
4618
# convert on the target, so we need to put bytes-on-the-wire that can
4619
# be converted. That means inventory deltas (if the remote is <1.19,
4620
# RemoteStreamSink will fallback to VFS to insert the deltas).
4621
yield ('inventory-deltas',
4622
self._stream_invs_as_deltas(revision_ids,
4623
delta_versus_null=delta_versus_null))
4625
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4626
"""Return a stream of inventory-deltas for the given rev ids.
4628
:param revision_ids: The list of inventories to transmit
4629
:param delta_versus_null: Don't try to find a minimal delta for this
4630
entry, instead compute the delta versus the NULL_REVISION. This
4631
effectively streams a complete inventory. Used for stuff like
4632
filling in missing parents, etc.
4634
from_repo = self.from_repository
4635
revision_keys = [(rev_id,) for rev_id in revision_ids]
4636
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4637
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4639
inventories = self.from_repository.iter_inventories(
4640
revision_ids, 'topological')
4641
format = from_repo._format
4642
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4643
inventory_cache = lru_cache.LRUCache(50)
4644
null_inventory = from_repo.revision_tree(
4645
_mod_revision.NULL_REVISION).inventory
4646
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4647
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4648
# repo back into a non-rich-root repo ought to be allowed)
4649
serializer = inventory_delta.InventoryDeltaSerializer(
4650
versioned_root=format.rich_root_data,
4651
tree_references=format.supports_tree_reference)
4652
for inv in inventories:
4653
key = (inv.revision_id,)
4654
parent_keys = parent_map.get(key, ())
4656
if not delta_versus_null and parent_keys:
4657
# The caller did not ask for complete inventories and we have
4658
# some parents that we can delta against. Make a delta against
4659
# each parent so that we can find the smallest.
4660
parent_ids = [parent_key[0] for parent_key in parent_keys]
4661
for parent_id in parent_ids:
4662
if parent_id not in invs_sent_so_far:
4663
# We don't know that the remote side has this basis, so
4666
if parent_id == _mod_revision.NULL_REVISION:
4667
parent_inv = null_inventory
4669
parent_inv = inventory_cache.get(parent_id, None)
4670
if parent_inv is None:
4671
parent_inv = from_repo.get_inventory(parent_id)
4672
candidate_delta = inv._make_delta(parent_inv)
4673
if (delta is None or
4674
len(delta) > len(candidate_delta)):
4675
delta = candidate_delta
4676
basis_id = parent_id
4678
# Either none of the parents ended up being suitable, or we
4679
# were asked to delta against NULL
4680
basis_id = _mod_revision.NULL_REVISION
4681
delta = inv._make_delta(null_inventory)
4682
invs_sent_so_far.add(inv.revision_id)
4683
inventory_cache[inv.revision_id] = inv
4684
delta_serialized = ''.join(
4685
serializer.delta_to_lines(basis_id, key[-1], delta))
4686
yield versionedfile.FulltextContentFactory(
4687
key, parent_keys, None, delta_serialized)
1835
4690
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
1836
4691
stop_revision=None):
1837
4692
"""Extend the partial history to include a given index
added
removed
bzrlib/repository.py
