341
208
commit to be valid, deletes against the basis MUST be recorded via
342
209
builder.record_delete().
344
self._recording_deletes = True
346
basis_id = self.parents[0]
348
basis_id = _mod_revision.NULL_REVISION
349
self.basis_delta_revision = basis_id
351
def record_entry_contents(self, ie, parent_invs, path, tree,
353
"""Record the content of ie from tree into the commit if needed.
355
Side effect: sets ie.revision when unchanged
357
:param ie: An inventory entry present in the commit.
358
:param parent_invs: The inventories of the parent revisions of the
360
:param path: The path the entry is at in the tree.
361
:param tree: The tree which contains this entry and should be used to
363
:param content_summary: Summary data from the tree about the paths
364
content - stat, length, exec, sha/link target. This is only
365
accessed when the entry has a revision of None - that is when it is
366
a candidate to commit.
367
:return: A tuple (change_delta, version_recorded, fs_hash).
368
change_delta is an inventory_delta change for this entry against
369
the basis tree of the commit, or None if no change occured against
371
version_recorded is True if a new version of the entry has been
372
recorded. For instance, committing a merge where a file was only
373
changed on the other side will return (delta, False).
374
fs_hash is either None, or the hash details for the path (currently
375
a tuple of the contents sha1 and the statvalue returned by
376
tree.get_file_with_stat()).
378
if self.new_inventory.root is None:
379
if ie.parent_id is not None:
380
raise errors.RootMissing()
381
self._check_root(ie, parent_invs, tree)
382
if ie.revision is None:
383
kind = content_summary[0]
385
# ie is carried over from a prior commit
387
# XXX: repository specific check for nested tree support goes here - if
388
# the repo doesn't want nested trees we skip it ?
389
if (kind == 'tree-reference' and
390
not self.repository._format.supports_tree_reference):
391
# mismatch between commit builder logic and repository:
392
# this needs the entry creation pushed down into the builder.
393
raise NotImplementedError('Missing repository subtree support.')
394
self.new_inventory.add(ie)
396
# TODO: slow, take it out of the inner loop.
398
basis_inv = parent_invs[0]
400
basis_inv = Inventory(root_id=None)
402
# ie.revision is always None if the InventoryEntry is considered
403
# for committing. We may record the previous parents revision if the
404
# content is actually unchanged against a sole head.
405
if ie.revision is not None:
406
if not self._versioned_root and path == '':
407
# repositories that do not version the root set the root's
408
# revision to the new commit even when no change occurs (more
409
# specifically, they do not record a revision on the root; and
410
# the rev id is assigned to the root during deserialisation -
411
# this masks when a change may have occurred against the basis.
412
# To match this we always issue a delta, because the revision
413
# of the root will always be changing.
414
if ie.file_id in basis_inv:
415
delta = (basis_inv.id2path(ie.file_id), path,
419
delta = (None, path, ie.file_id, ie)
420
self._basis_delta.append(delta)
421
return delta, False, None
423
# we don't need to commit this, because the caller already
424
# determined that an existing revision of this file is
425
# appropriate. If its not being considered for committing then
426
# it and all its parents to the root must be unaltered so
427
# no-change against the basis.
428
if ie.revision == self._new_revision_id:
429
raise AssertionError("Impossible situation, a skipped "
430
"inventory entry (%r) claims to be modified in this "
431
"commit (%r).", (ie, self._new_revision_id))
432
return None, False, None
433
# XXX: Friction: parent_candidates should return a list not a dict
434
# so that we don't have to walk the inventories again.
435
parent_candiate_entries = ie.parent_candidates(parent_invs)
436
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
438
for inv in parent_invs:
439
if ie.file_id in inv:
440
old_rev = inv[ie.file_id].revision
441
if old_rev in head_set:
442
heads.append(inv[ie.file_id].revision)
443
head_set.remove(inv[ie.file_id].revision)
446
# now we check to see if we need to write a new record to the
448
# We write a new entry unless there is one head to the ancestors, and
449
# the kind-derived content is unchanged.
451
# Cheapest check first: no ancestors, or more the one head in the
452
# ancestors, we write a new node.
456
# There is a single head, look it up for comparison
457
parent_entry = parent_candiate_entries[heads[0]]
458
# if the non-content specific data has changed, we'll be writing a
460
if (parent_entry.parent_id != ie.parent_id or
461
parent_entry.name != ie.name):
463
# now we need to do content specific checks:
465
# if the kind changed the content obviously has
466
if kind != parent_entry.kind:
468
# Stat cache fingerprint feedback for the caller - None as we usually
469
# don't generate one.
472
if content_summary[2] is None:
473
raise ValueError("Files must not have executable = None")
475
# We can't trust a check of the file length because of content
477
if (# if the exec bit has changed we have to store:
478
parent_entry.executable != content_summary[2]):
480
elif parent_entry.text_sha1 == content_summary[3]:
481
# all meta and content is unchanged (using a hash cache
482
# hit to check the sha)
483
ie.revision = parent_entry.revision
484
ie.text_size = parent_entry.text_size
485
ie.text_sha1 = parent_entry.text_sha1
486
ie.executable = parent_entry.executable
487
return self._get_delta(ie, basis_inv, path), False, None
489
# Either there is only a hash change(no hash cache entry,
490
# or same size content change), or there is no change on
492
# Provide the parent's hash to the store layer, so that the
493
# content is unchanged we will not store a new node.
494
nostore_sha = parent_entry.text_sha1
496
# We want to record a new node regardless of the presence or
497
# absence of a content change in the file.
499
ie.executable = content_summary[2]
500
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
502
text = file_obj.read()
506
ie.text_sha1, ie.text_size = self._add_text_to_weave(
507
ie.file_id, text, heads, nostore_sha)
508
# Let the caller know we generated a stat fingerprint.
509
fingerprint = (ie.text_sha1, stat_value)
510
except errors.ExistingContent:
511
# Turns out that the file content was unchanged, and we were
512
# only going to store a new node if it was changed. Carry over
514
ie.revision = parent_entry.revision
515
ie.text_size = parent_entry.text_size
516
ie.text_sha1 = parent_entry.text_sha1
517
ie.executable = parent_entry.executable
518
return self._get_delta(ie, basis_inv, path), False, None
519
elif kind == 'directory':
521
# all data is meta here, nothing specific to directory, so
523
ie.revision = parent_entry.revision
524
return self._get_delta(ie, basis_inv, path), False, None
525
self._add_text_to_weave(ie.file_id, '', heads, None)
526
elif kind == 'symlink':
527
current_link_target = content_summary[3]
529
# symlink target is not generic metadata, check if it has
531
if current_link_target != parent_entry.symlink_target:
534
# unchanged, carry over.
535
ie.revision = parent_entry.revision
536
ie.symlink_target = parent_entry.symlink_target
537
return self._get_delta(ie, basis_inv, path), False, None
538
ie.symlink_target = current_link_target
539
self._add_text_to_weave(ie.file_id, '', heads, None)
540
elif kind == 'tree-reference':
542
if content_summary[3] != parent_entry.reference_revision:
545
# unchanged, carry over.
546
ie.reference_revision = parent_entry.reference_revision
547
ie.revision = parent_entry.revision
548
return self._get_delta(ie, basis_inv, path), False, None
549
ie.reference_revision = content_summary[3]
550
if ie.reference_revision is None:
551
raise AssertionError("invalid content_summary for nested tree: %r"
552
% (content_summary,))
553
self._add_text_to_weave(ie.file_id, '', heads, None)
555
raise NotImplementedError('unknown kind')
556
ie.revision = self._new_revision_id
557
self._any_changes = True
558
return self._get_delta(ie, basis_inv, path), True, fingerprint
560
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
561
_entry_factory=entry_factory):
211
raise NotImplementedError(self.will_record_deletes)
213
def record_iter_changes(self, tree, basis_revision_id, iter_changes):
562
214
"""Record a new tree via iter_changes.
564
216
:param tree: The tree to obtain text contents from for changed objects.
569
221
to basis_revision_id. The iterator must not include any items with
570
222
a current kind of None - missing items must be either filtered out
571
223
or errored-on beefore record_iter_changes sees the item.
572
:param _entry_factory: Private method to bind entry_factory locally for
574
224
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
575
225
tree._observed_sha1.
577
# Create an inventory delta based on deltas between all the parents and
578
# deltas between all the parent inventories. We use inventory delta's
579
# between the inventory objects because iter_changes masks
580
# last-changed-field only changes.
582
# file_id -> change map, change is fileid, paths, changed, versioneds,
583
# parents, names, kinds, executables
585
# {file_id -> revision_id -> inventory entry, for entries in parent
586
# trees that are not parents[0]
590
revtrees = list(self.repository.revision_trees(self.parents))
591
except errors.NoSuchRevision:
592
# one or more ghosts, slow path.
594
for revision_id in self.parents:
596
revtrees.append(self.repository.revision_tree(revision_id))
597
except errors.NoSuchRevision:
599
basis_revision_id = _mod_revision.NULL_REVISION
601
revtrees.append(self.repository.revision_tree(
602
_mod_revision.NULL_REVISION))
603
# The basis inventory from a repository
605
basis_inv = revtrees[0].inventory
607
basis_inv = self.repository.revision_tree(
608
_mod_revision.NULL_REVISION).inventory
609
if len(self.parents) > 0:
610
if basis_revision_id != self.parents[0] and not ghost_basis:
612
"arbitrary basis parents not yet supported with merges")
613
for revtree in revtrees[1:]:
614
for change in revtree.inventory._make_delta(basis_inv):
615
if change[1] is None:
616
# Not present in this parent.
618
if change[2] not in merged_ids:
619
if change[0] is not None:
620
basis_entry = basis_inv[change[2]]
621
merged_ids[change[2]] = [
623
basis_entry.revision,
626
parent_entries[change[2]] = {
628
basis_entry.revision:basis_entry,
630
change[3].revision:change[3],
633
merged_ids[change[2]] = [change[3].revision]
634
parent_entries[change[2]] = {change[3].revision:change[3]}
636
merged_ids[change[2]].append(change[3].revision)
637
parent_entries[change[2]][change[3].revision] = change[3]
640
# Setup the changes from the tree:
641
# changes maps file_id -> (change, [parent revision_ids])
643
for change in iter_changes:
644
# This probably looks up in basis_inv way to much.
645
if change[1][0] is not None:
646
head_candidate = [basis_inv[change[0]].revision]
649
changes[change[0]] = change, merged_ids.get(change[0],
651
unchanged_merged = set(merged_ids) - set(changes)
652
# Extend the changes dict with synthetic changes to record merges of
654
for file_id in unchanged_merged:
655
# Record a merged version of these items that did not change vs the
656
# basis. This can be either identical parallel changes, or a revert
657
# of a specific file after a merge. The recorded content will be
658
# that of the current tree (which is the same as the basis), but
659
# the per-file graph will reflect a merge.
660
# NB:XXX: We are reconstructing path information we had, this
661
# should be preserved instead.
662
# inv delta change: (file_id, (path_in_source, path_in_target),
663
# changed_content, versioned, parent, name, kind,
666
basis_entry = basis_inv[file_id]
667
except errors.NoSuchId:
668
# a change from basis->some_parents but file_id isn't in basis
669
# so was new in the merge, which means it must have changed
670
# from basis -> current, and as it hasn't the add was reverted
671
# by the user. So we discard this change.
675
(basis_inv.id2path(file_id), tree.id2path(file_id)),
677
(basis_entry.parent_id, basis_entry.parent_id),
678
(basis_entry.name, basis_entry.name),
679
(basis_entry.kind, basis_entry.kind),
680
(basis_entry.executable, basis_entry.executable))
681
changes[file_id] = (change, merged_ids[file_id])
682
# changes contains tuples with the change and a set of inventory
683
# candidates for the file.
685
# old_path, new_path, file_id, new_inventory_entry
686
seen_root = False # Is the root in the basis delta?
687
inv_delta = self._basis_delta
688
modified_rev = self._new_revision_id
689
for change, head_candidates in changes.values():
690
if change[3][1]: # versioned in target.
691
# Several things may be happening here:
692
# We may have a fork in the per-file graph
693
# - record a change with the content from tree
694
# We may have a change against < all trees
695
# - carry over the tree that hasn't changed
696
# We may have a change against all trees
697
# - record the change with the content from tree
700
entry = _entry_factory[kind](file_id, change[5][1],
702
head_set = self._heads(change[0], set(head_candidates))
705
for head_candidate in head_candidates:
706
if head_candidate in head_set:
707
heads.append(head_candidate)
708
head_set.remove(head_candidate)
711
# Could be a carry-over situation:
712
parent_entry_revs = parent_entries.get(file_id, None)
713
if parent_entry_revs:
714
parent_entry = parent_entry_revs.get(heads[0], None)
717
if parent_entry is None:
718
# The parent iter_changes was called against is the one
719
# that is the per-file head, so any change is relevant
720
# iter_changes is valid.
721
carry_over_possible = False
723
# could be a carry over situation
724
# A change against the basis may just indicate a merge,
725
# we need to check the content against the source of the
726
# merge to determine if it was changed after the merge
728
if (parent_entry.kind != entry.kind or
729
parent_entry.parent_id != entry.parent_id or
730
parent_entry.name != entry.name):
731
# Metadata common to all entries has changed
732
# against per-file parent
733
carry_over_possible = False
735
carry_over_possible = True
736
# per-type checks for changes against the parent_entry
739
# Cannot be a carry-over situation
740
carry_over_possible = False
741
# Populate the entry in the delta
743
# XXX: There is still a small race here: If someone reverts the content of a file
744
# after iter_changes examines and decides it has changed,
745
# we will unconditionally record a new version even if some
746
# other process reverts it while commit is running (with
747
# the revert happening after iter_changes did it's
750
entry.executable = True
752
entry.executable = False
753
if (carry_over_possible and
754
parent_entry.executable == entry.executable):
755
# Check the file length, content hash after reading
757
nostore_sha = parent_entry.text_sha1
760
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
762
text = file_obj.read()
766
entry.text_sha1, entry.text_size = self._add_text_to_weave(
767
file_id, text, heads, nostore_sha)
768
yield file_id, change[1][1], (entry.text_sha1, stat_value)
769
except errors.ExistingContent:
770
# No content change against a carry_over parent
771
# Perhaps this should also yield a fs hash update?
773
entry.text_size = parent_entry.text_size
774
entry.text_sha1 = parent_entry.text_sha1
775
elif kind == 'symlink':
777
entry.symlink_target = tree.get_symlink_target(file_id)
778
if (carry_over_possible and
779
parent_entry.symlink_target == entry.symlink_target):
782
self._add_text_to_weave(change[0], '', heads, None)
783
elif kind == 'directory':
784
if carry_over_possible:
787
# Nothing to set on the entry.
788
# XXX: split into the Root and nonRoot versions.
789
if change[1][1] != '' or self.repository.supports_rich_root():
790
self._add_text_to_weave(change[0], '', heads, None)
791
elif kind == 'tree-reference':
792
if not self.repository._format.supports_tree_reference:
793
# This isn't quite sane as an error, but we shouldn't
794
# ever see this code path in practice: tree's don't
795
# permit references when the repo doesn't support tree
797
raise errors.UnsupportedOperation(tree.add_reference,
799
reference_revision = tree.get_reference_revision(change[0])
800
entry.reference_revision = reference_revision
801
if (carry_over_possible and
802
parent_entry.reference_revision == reference_revision):
805
self._add_text_to_weave(change[0], '', heads, None)
807
raise AssertionError('unknown kind %r' % kind)
809
entry.revision = modified_rev
811
entry.revision = parent_entry.revision
814
new_path = change[1][1]
815
inv_delta.append((change[1][0], new_path, change[0], entry))
818
self.new_inventory = None
820
# This should perhaps be guarded by a check that the basis we
821
# commit against is the basis for the commit and if not do a delta
823
self._any_changes = True
825
# housekeeping root entry changes do not affect no-change commits.
826
self._require_root_change(tree)
827
self.basis_delta_revision = basis_revision_id
829
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
830
parent_keys = tuple([(file_id, parent) for parent in parents])
831
return self.repository.texts._add_text(
832
(file_id, self._new_revision_id), parent_keys, new_text,
833
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
836
class RootCommitBuilder(CommitBuilder):
837
"""This commitbuilder actually records the root id"""
839
# the root entry gets versioned properly by this builder.
840
_versioned_root = True
842
def _check_root(self, ie, parent_invs, tree):
843
"""Helper for record_entry_contents.
845
:param ie: An entry being added.
846
:param parent_invs: The inventories of the parent revisions of the
848
:param tree: The tree that is being committed.
851
def _require_root_change(self, tree):
852
"""Enforce an appropriate root object change.
854
This is called once when record_iter_changes is called, if and only if
855
the root was not in the delta calculated by record_iter_changes.
857
:param tree: The tree which is being committed.
859
# versioned roots do not change unless the tree found a change.
227
raise NotImplementedError(self.record_iter_changes)
230
class RepositoryWriteLockResult(LogicalLockResult):
231
"""The result of write locking a repository.
233
:ivar repository_token: The token obtained from the underlying lock, or
235
:ivar unlock: A callable which will unlock the lock.
238
def __init__(self, unlock, repository_token):
239
LogicalLockResult.__init__(self, unlock)
240
self.repository_token = repository_token
243
return "RepositoryWriteLockResult(%s, %s)" % (self.repository_token,
862
247
######################################################################
866
class Repository(_RelockDebugMixin):
251
class Repository(_RelockDebugMixin, controldir.ControlComponent):
867
252
"""Repository holding history for one or more branches.
869
254
The repository holds and retrieves historical information including
870
255
revisions and file history. It's normally accessed only by the Branch,
871
256
which views a particular line of development through that history.
873
The Repository builds on top of some byte storage facilies (the revisions,
874
signatures, inventories, texts and chk_bytes attributes) and a Transport,
875
which respectively provide byte storage and a means to access the (possibly
878
The byte storage facilities are addressed via tuples, which we refer to
879
as 'keys' throughout the code base. Revision_keys, inventory_keys and
880
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
881
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
882
byte string made up of a hash identifier and a hash value.
883
We use this interface because it allows low friction with the underlying
884
code that implements disk indices, network encoding and other parts of
887
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
888
the serialised revisions for the repository. This can be used to obtain
889
revision graph information or to access raw serialised revisions.
890
The result of trying to insert data into the repository via this store
891
is undefined: it should be considered read-only except for implementors
893
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
894
the serialised signatures for the repository. This can be used to
895
obtain access to raw serialised signatures. The result of trying to
896
insert data into the repository via this store is undefined: it should
897
be considered read-only except for implementors of repositories.
898
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
899
the serialised inventories for the repository. This can be used to
900
obtain unserialised inventories. The result of trying to insert data
901
into the repository via this store is undefined: it should be
902
considered read-only except for implementors of repositories.
903
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
904
texts of files and directories for the repository. This can be used to
905
obtain file texts or file graphs. Note that Repository.iter_file_bytes
906
is usually a better interface for accessing file texts.
907
The result of trying to insert data into the repository via this store
908
is undefined: it should be considered read-only except for implementors
910
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
911
any data the repository chooses to store or have indexed by its hash.
912
The result of trying to insert data into the repository via this store
913
is undefined: it should be considered read-only except for implementors
915
:ivar _transport: Transport for file access to repository, typically
916
pointing to .bzr/repository.
258
See VersionedFileRepository in bzrlib.vf_repository for the
259
base class for most Bazaar repositories.
919
# What class to use for a CommitBuilder. Often its simpler to change this
920
# in a Repository class subclass rather than to override
921
# get_commit_builder.
922
_commit_builder_class = CommitBuilder
923
# The search regex used by xml based repositories to determine what things
924
# where changed in a single commit.
925
_file_ids_altered_regex = lazy_regex.lazy_compile(
926
r'file_id="(?P<file_id>[^"]+)"'
927
r'.* revision="(?P<revision_id>[^"]+)"'
930
262
def abort_write_group(self, suppress_errors=False):
931
263
"""Commit the contents accrued within the current write group.
1000
319
return InterRepository._assert_same_model(self, repository)
1002
def add_inventory(self, revision_id, inv, parents):
1003
"""Add the inventory inv to the repository as revision_id.
1005
:param parents: The revision ids of the parents that revision_id
1006
is known to have and are in the repository already.
1008
:returns: The validator(which is a sha1 digest, though what is sha'd is
1009
repository format specific) of the serialized inventory.
1011
if not self.is_in_write_group():
1012
raise AssertionError("%r not in write group" % (self,))
1013
_mod_revision.check_not_reserved_id(revision_id)
1014
if not (inv.revision_id is None or inv.revision_id == revision_id):
1015
raise AssertionError(
1016
"Mismatch between inventory revision"
1017
" id and insertion revid (%r, %r)"
1018
% (inv.revision_id, revision_id))
1019
if inv.root is None:
1020
raise AssertionError()
1021
return self._add_inventory_checked(revision_id, inv, parents)
1023
def _add_inventory_checked(self, revision_id, inv, parents):
1024
"""Add inv to the repository after checking the inputs.
1026
This function can be overridden to allow different inventory styles.
1028
:seealso: add_inventory, for the contract.
1030
inv_lines = self._serializer.write_inventory_to_lines(inv)
1031
return self._inventory_add_lines(revision_id, parents,
1032
inv_lines, check_content=False)
1034
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1035
parents, basis_inv=None, propagate_caches=False):
1036
"""Add a new inventory expressed as a delta against another revision.
1038
See the inventory developers documentation for the theory behind
1041
:param basis_revision_id: The inventory id the delta was created
1042
against. (This does not have to be a direct parent.)
1043
:param delta: The inventory delta (see Inventory.apply_delta for
1045
:param new_revision_id: The revision id that the inventory is being
1047
:param parents: The revision ids of the parents that revision_id is
1048
known to have and are in the repository already. These are supplied
1049
for repositories that depend on the inventory graph for revision
1050
graph access, as well as for those that pun ancestry with delta
1052
:param basis_inv: The basis inventory if it is already known,
1054
:param propagate_caches: If True, the caches for this inventory are
1055
copied to and updated for the result if possible.
1057
:returns: (validator, new_inv)
1058
The validator(which is a sha1 digest, though what is sha'd is
1059
repository format specific) of the serialized inventory, and the
1060
resulting inventory.
1062
if not self.is_in_write_group():
1063
raise AssertionError("%r not in write group" % (self,))
1064
_mod_revision.check_not_reserved_id(new_revision_id)
1065
basis_tree = self.revision_tree(basis_revision_id)
1066
basis_tree.lock_read()
1068
# Note that this mutates the inventory of basis_tree, which not all
1069
# inventory implementations may support: A better idiom would be to
1070
# return a new inventory, but as there is no revision tree cache in
1071
# repository this is safe for now - RBC 20081013
1072
if basis_inv is None:
1073
basis_inv = basis_tree.inventory
1074
basis_inv.apply_delta(delta)
1075
basis_inv.revision_id = new_revision_id
1076
return (self.add_inventory(new_revision_id, basis_inv, parents),
1081
def _inventory_add_lines(self, revision_id, parents, lines,
1082
check_content=True):
1083
"""Store lines in inv_vf and return the sha1 of the inventory."""
1084
parents = [(parent,) for parent in parents]
1085
result = self.inventories.add_lines((revision_id,), parents, lines,
1086
check_content=check_content)[0]
1087
self.inventories._access.flush()
1090
def add_revision(self, revision_id, rev, inv=None, config=None):
1091
"""Add rev to the revision store as revision_id.
1093
:param revision_id: the revision id to use.
1094
:param rev: The revision object.
1095
:param inv: The inventory for the revision. if None, it will be looked
1096
up in the inventory storer
1097
:param config: If None no digital signature will be created.
1098
If supplied its signature_needed method will be used
1099
to determine if a signature should be made.
1101
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1103
_mod_revision.check_not_reserved_id(revision_id)
1104
if config is not None and config.signature_needed():
1106
inv = self.get_inventory(revision_id)
1107
plaintext = Testament(rev, inv).as_short_text()
1108
self.store_revision_signature(
1109
gpg.GPGStrategy(config), plaintext, revision_id)
1110
# check inventory present
1111
if not self.inventories.get_parent_map([(revision_id,)]):
1113
raise errors.WeaveRevisionNotPresent(revision_id,
1116
# yes, this is not suitable for adding with ghosts.
1117
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1120
key = (revision_id,)
1121
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1122
self._add_revision(rev)
1124
def _add_revision(self, revision):
1125
text = self._serializer.write_revision_to_string(revision)
1126
key = (revision.revision_id,)
1127
parents = tuple((parent,) for parent in revision.parent_ids)
1128
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1130
321
def all_revision_ids(self):
1131
322
"""Returns a list of all the revision ids in the repository.
1156
347
self.control_files.break_lock()
1159
def _eliminate_revisions_not_present(self, revision_ids):
1160
"""Check every revision id in revision_ids to see if we have it.
1162
Returns a set of the present revisions.
1165
graph = self.get_graph()
1166
parent_map = graph.get_parent_map(revision_ids)
1167
# The old API returned a list, should this actually be a set?
1168
return parent_map.keys()
1170
def _check_inventories(self, checker):
1171
"""Check the inventories found from the revision scan.
1173
This is responsible for verifying the sha1 of inventories and
1174
creating a pending_keys set that covers data referenced by inventories.
1176
bar = ui.ui_factory.nested_progress_bar()
1178
self._do_check_inventories(checker, bar)
1182
def _do_check_inventories(self, checker, bar):
1183
"""Helper for _check_inventories."""
1185
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1186
kinds = ['chk_bytes', 'texts']
1187
count = len(checker.pending_keys)
1188
bar.update("inventories", 0, 2)
1189
current_keys = checker.pending_keys
1190
checker.pending_keys = {}
1191
# Accumulate current checks.
1192
for key in current_keys:
1193
if key[0] != 'inventories' and key[0] not in kinds:
1194
checker._report_items.append('unknown key type %r' % (key,))
1195
keys[key[0]].add(key[1:])
1196
if keys['inventories']:
1197
# NB: output order *should* be roughly sorted - topo or
1198
# inverse topo depending on repository - either way decent
1199
# to just delta against. However, pre-CHK formats didn't
1200
# try to optimise inventory layout on disk. As such the
1201
# pre-CHK code path does not use inventory deltas.
1203
for record in self.inventories.check(keys=keys['inventories']):
1204
if record.storage_kind == 'absent':
1205
checker._report_items.append(
1206
'Missing inventory {%s}' % (record.key,))
1208
last_object = self._check_record('inventories', record,
1209
checker, last_object,
1210
current_keys[('inventories',) + record.key])
1211
del keys['inventories']
1214
bar.update("texts", 1)
1215
while (checker.pending_keys or keys['chk_bytes']
1217
# Something to check.
1218
current_keys = checker.pending_keys
1219
checker.pending_keys = {}
1220
# Accumulate current checks.
1221
for key in current_keys:
1222
if key[0] not in kinds:
1223
checker._report_items.append('unknown key type %r' % (key,))
1224
keys[key[0]].add(key[1:])
1225
# Check the outermost kind only - inventories || chk_bytes || texts
1229
for record in getattr(self, kind).check(keys=keys[kind]):
1230
if record.storage_kind == 'absent':
1231
checker._report_items.append(
1232
'Missing %s {%s}' % (kind, record.key,))
1234
last_object = self._check_record(kind, record,
1235
checker, last_object, current_keys[(kind,) + record.key])
1239
def _check_record(self, kind, record, checker, last_object, item_data):
1240
"""Check a single text from this repository."""
1241
if kind == 'inventories':
1242
rev_id = record.key[0]
1243
inv = self._deserialise_inventory(rev_id,
1244
record.get_bytes_as('fulltext'))
1245
if last_object is not None:
1246
delta = inv._make_delta(last_object)
1247
for old_path, path, file_id, ie in delta:
1250
ie.check(checker, rev_id, inv)
1252
for path, ie in inv.iter_entries():
1253
ie.check(checker, rev_id, inv)
1254
if self._format.fast_deltas:
1256
elif kind == 'chk_bytes':
1257
# No code written to check chk_bytes for this repo format.
1258
checker._report_items.append(
1259
'unsupported key type chk_bytes for %s' % (record.key,))
1260
elif kind == 'texts':
1261
self._check_text(record, checker, item_data)
1263
checker._report_items.append(
1264
'unknown key type %s for %s' % (kind, record.key))
1266
def _check_text(self, record, checker, item_data):
1267
"""Check a single text."""
1268
# Check it is extractable.
1269
# TODO: check length.
1270
if record.storage_kind == 'chunked':
1271
chunks = record.get_bytes_as(record.storage_kind)
1272
sha1 = osutils.sha_strings(chunks)
1273
length = sum(map(len, chunks))
1275
content = record.get_bytes_as('fulltext')
1276
sha1 = osutils.sha_string(content)
1277
length = len(content)
1278
if item_data and sha1 != item_data[1]:
1279
checker._report_items.append(
1280
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1281
(record.key, sha1, item_data[1], item_data[2]))
1284
def create(a_bzrdir):
1285
"""Construct the current default format repository in a_bzrdir."""
1286
return RepositoryFormat.get_default_format().initialize(a_bzrdir)
350
def create(controldir):
351
"""Construct the current default format repository in controldir."""
352
return RepositoryFormat.get_default_format().initialize(controldir)
1288
def __init__(self, _format, a_bzrdir, control_files):
354
def __init__(self, _format, controldir, control_files):
1289
355
"""instantiate a Repository.
1291
357
:param _format: The format of the repository on disk.
1292
:param a_bzrdir: The BzrDir of the repository.
1294
In the future we will have a single api for all stores for
1295
getting file texts, inventories and revisions, then
1296
this construct will accept instances of those things.
358
:param controldir: The ControlDir of the repository.
359
:param control_files: Control files to use for locking, etc.
361
# In the future we will have a single api for all stores for
362
# getting file texts, inventories and revisions, then
363
# this construct will accept instances of those things.
1298
364
super(Repository, self).__init__()
1299
365
self._format = _format
1300
366
# the following are part of the public API for Repository:
1301
self.bzrdir = a_bzrdir
367
self.bzrdir = controldir
1302
368
self.control_files = control_files
1303
self._transport = control_files._transport
1304
self.base = self._transport.base
1306
self._reconcile_does_inventory_gc = True
1307
self._reconcile_fixes_text_parents = False
1308
self._reconcile_backsup_inventory = True
1309
370
self._write_group = None
1310
371
# Additional places to query for data.
1311
372
self._fallback_repositories = []
1312
# An InventoryEntry cache, used during deserialization
1313
self._inventory_entry_cache = fifo_cache.FIFOCache(10*1024)
1314
# Is it safe to return inventory entries directly from the entry cache,
1315
# rather copying them?
1316
self._safe_to_return_from_cache = False
375
def user_transport(self):
376
return self.bzrdir.user_transport
379
def control_transport(self):
380
return self._transport
1318
382
def __repr__(self):
1319
383
if self._fallback_repositories:
1954
909
signature = gpg_strategy.sign(plaintext)
1955
910
self.add_signature_text(revision_id, signature)
1958
912
def add_signature_text(self, revision_id, signature):
1959
self.signatures.add_lines((revision_id,), (),
1960
osutils.split_lines(signature))
1962
def find_text_key_references(self):
1963
"""Find the text key references within the repository.
1965
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1966
to whether they were referred to by the inventory of the
1967
revision_id that they contain. The inventory texts from all present
1968
revision ids are assessed to generate this report.
1970
revision_keys = self.revisions.keys()
1971
w = self.inventories
1972
pb = ui.ui_factory.nested_progress_bar()
1974
return self._find_text_key_references_from_xml_inventory_lines(
1975
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1979
def _find_text_key_references_from_xml_inventory_lines(self,
1981
"""Core routine for extracting references to texts from inventories.
1983
This performs the translation of xml lines to revision ids.
1985
:param line_iterator: An iterator of lines, origin_version_id
1986
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1987
to whether they were referred to by the inventory of the
1988
revision_id that they contain. Note that if that revision_id was
1989
not part of the line_iterator's output then False will be given -
1990
even though it may actually refer to that key.
1992
if not self._serializer.support_altered_by_hack:
1993
raise AssertionError(
1994
"_find_text_key_references_from_xml_inventory_lines only "
1995
"supported for branches which store inventory as unnested xml"
1996
", not on %r" % self)
1999
# this code needs to read every new line in every inventory for the
2000
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2001
# not present in one of those inventories is unnecessary but not
2002
# harmful because we are filtering by the revision id marker in the
2003
# inventory lines : we only select file ids altered in one of those
2004
# revisions. We don't need to see all lines in the inventory because
2005
# only those added in an inventory in rev X can contain a revision=X
2007
unescape_revid_cache = {}
2008
unescape_fileid_cache = {}
2010
# jam 20061218 In a big fetch, this handles hundreds of thousands
2011
# of lines, so it has had a lot of inlining and optimizing done.
2012
# Sorry that it is a little bit messy.
2013
# Move several functions to be local variables, since this is a long
2015
search = self._file_ids_altered_regex.search
2016
unescape = _unescape_xml
2017
setdefault = result.setdefault
2018
for line, line_key in line_iterator:
2019
match = search(line)
2022
# One call to match.group() returning multiple items is quite a
2023
# bit faster than 2 calls to match.group() each returning 1
2024
file_id, revision_id = match.group('file_id', 'revision_id')
2026
# Inlining the cache lookups helps a lot when you make 170,000
2027
# lines and 350k ids, versus 8.4 unique ids.
2028
# Using a cache helps in 2 ways:
2029
# 1) Avoids unnecessary decoding calls
2030
# 2) Re-uses cached strings, which helps in future set and
2032
# (2) is enough that removing encoding entirely along with
2033
# the cache (so we are using plain strings) results in no
2034
# performance improvement.
2036
revision_id = unescape_revid_cache[revision_id]
2038
unescaped = unescape(revision_id)
2039
unescape_revid_cache[revision_id] = unescaped
2040
revision_id = unescaped
2042
# Note that unconditionally unescaping means that we deserialise
2043
# every fileid, which for general 'pull' is not great, but we don't
2044
# really want to have some many fulltexts that this matters anyway.
2047
file_id = unescape_fileid_cache[file_id]
2049
unescaped = unescape(file_id)
2050
unescape_fileid_cache[file_id] = unescaped
2053
key = (file_id, revision_id)
2054
setdefault(key, False)
2055
if revision_id == line_key[-1]:
2059
def _inventory_xml_lines_for_keys(self, keys):
2060
"""Get a line iterator of the sort needed for findind references.
2062
Not relevant for non-xml inventory repositories.
2064
Ghosts in revision_keys are ignored.
2066
:param revision_keys: The revision keys for the inventories to inspect.
2067
:return: An iterator over (inventory line, revid) for the fulltexts of
2068
all of the xml inventories specified by revision_keys.
2070
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2071
for record in stream:
2072
if record.storage_kind != 'absent':
2073
chunks = record.get_bytes_as('chunked')
2074
revid = record.key[-1]
2075
lines = osutils.chunks_to_lines(chunks)
2079
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2081
"""Helper routine for fileids_altered_by_revision_ids.
2083
This performs the translation of xml lines to revision ids.
2085
:param line_iterator: An iterator of lines, origin_version_id
2086
:param revision_keys: The revision ids to filter for. This should be a
2087
set or other type which supports efficient __contains__ lookups, as
2088
the revision key from each parsed line will be looked up in the
2089
revision_keys filter.
2090
:return: a dictionary mapping altered file-ids to an iterable of
2091
revision_ids. Each altered file-ids has the exact revision_ids that
2092
altered it listed explicitly.
2094
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2095
line_iterator).iterkeys())
2096
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2097
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2098
self._inventory_xml_lines_for_keys(parent_keys)))
2099
new_keys = seen - parent_seen
2101
setdefault = result.setdefault
2102
for key in new_keys:
2103
setdefault(key[0], set()).add(key[-1])
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)
2106
920
def _find_parent_ids_of_revisions(self, revision_ids):
2107
921
"""Find all parent ids that are mentioned in the revision graph.
2158
941
uniquely identify the file version in the caller's context. (Examples:
2159
942
an index number or a TreeTransform trans_id.)
2161
bytes_iterator is an iterable of bytestrings for the file. The
2162
kind of iterable and length of the bytestrings are unspecified, but for
2163
this implementation, it is a list of bytes produced by
2164
VersionedFile.get_record_stream().
2166
944
:param desired_files: a list of (file_id, revision_id, identifier)
2170
for file_id, revision_id, callable_data in desired_files:
2171
text_keys[(file_id, revision_id)] = callable_data
2172
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2173
if record.storage_kind == 'absent':
2174
raise errors.RevisionNotPresent(record.key, self)
2175
yield text_keys[record.key], record.get_bytes_as('chunked')
2177
def _generate_text_key_index(self, text_key_references=None,
2179
"""Generate a new text key index for the repository.
2181
This is an expensive function that will take considerable time to run.
2183
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2184
list of parents, also text keys. When a given key has no parents,
2185
the parents list will be [NULL_REVISION].
2187
# All revisions, to find inventory parents.
2188
if ancestors is None:
2189
graph = self.get_graph()
2190
ancestors = graph.get_parent_map(self.all_revision_ids())
2191
if text_key_references is None:
2192
text_key_references = self.find_text_key_references()
2193
pb = ui.ui_factory.nested_progress_bar()
2195
return self._do_generate_text_key_index(ancestors,
2196
text_key_references, pb)
2200
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2201
"""Helper for _generate_text_key_index to avoid deep nesting."""
2202
revision_order = tsort.topo_sort(ancestors)
2203
invalid_keys = set()
2205
for revision_id in revision_order:
2206
revision_keys[revision_id] = set()
2207
text_count = len(text_key_references)
2208
# a cache of the text keys to allow reuse; costs a dict of all the
2209
# keys, but saves a 2-tuple for every child of a given key.
2211
for text_key, valid in text_key_references.iteritems():
2213
invalid_keys.add(text_key)
2215
revision_keys[text_key[1]].add(text_key)
2216
text_key_cache[text_key] = text_key
2217
del text_key_references
2219
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2220
NULL_REVISION = _mod_revision.NULL_REVISION
2221
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2222
# too small for large or very branchy trees. However, for 55K path
2223
# trees, it would be easy to use too much memory trivially. Ideally we
2224
# could gauge this by looking at available real memory etc, but this is
2225
# always a tricky proposition.
2226
inventory_cache = lru_cache.LRUCache(10)
2227
batch_size = 10 # should be ~150MB on a 55K path tree
2228
batch_count = len(revision_order) / batch_size + 1
2230
pb.update("Calculating text parents", processed_texts, text_count)
2231
for offset in xrange(batch_count):
2232
to_query = revision_order[offset * batch_size:(offset + 1) *
2236
for revision_id in to_query:
2237
parent_ids = ancestors[revision_id]
2238
for text_key in revision_keys[revision_id]:
2239
pb.update("Calculating text parents", processed_texts)
2240
processed_texts += 1
2241
candidate_parents = []
2242
for parent_id in parent_ids:
2243
parent_text_key = (text_key[0], parent_id)
2245
check_parent = parent_text_key not in \
2246
revision_keys[parent_id]
2248
# the parent parent_id is a ghost:
2249
check_parent = False
2250
# truncate the derived graph against this ghost.
2251
parent_text_key = None
2253
# look at the parent commit details inventories to
2254
# determine possible candidates in the per file graph.
2257
inv = inventory_cache[parent_id]
2259
inv = self.revision_tree(parent_id).inventory
2260
inventory_cache[parent_id] = inv
2262
parent_entry = inv[text_key[0]]
2263
except (KeyError, errors.NoSuchId):
2265
if parent_entry is not None:
2267
text_key[0], parent_entry.revision)
2269
parent_text_key = None
2270
if parent_text_key is not None:
2271
candidate_parents.append(
2272
text_key_cache[parent_text_key])
2273
parent_heads = text_graph.heads(candidate_parents)
2274
new_parents = list(parent_heads)
2275
new_parents.sort(key=lambda x:candidate_parents.index(x))
2276
if new_parents == []:
2277
new_parents = [NULL_REVISION]
2278
text_index[text_key] = new_parents
2280
for text_key in invalid_keys:
2281
text_index[text_key] = [NULL_REVISION]
2284
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2285
"""Get an iterable listing the keys of all the data introduced by a set
2288
The keys will be ordered so that the corresponding items can be safely
2289
fetched and inserted in that order.
2291
:returns: An iterable producing tuples of (knit-kind, file-id,
2292
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2293
'revisions'. file-id is None unless knit-kind is 'file'.
2295
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2298
for result in self._find_non_file_keys_to_fetch(revision_ids):
2301
def _find_file_keys_to_fetch(self, revision_ids, pb):
2302
# XXX: it's a bit weird to control the inventory weave caching in this
2303
# generator. Ideally the caching would be done in fetch.py I think. Or
2304
# maybe this generator should explicitly have the contract that it
2305
# should not be iterated until the previously yielded item has been
2307
inv_w = self.inventories
2309
# file ids that changed
2310
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2312
num_file_ids = len(file_ids)
2313
for file_id, altered_versions in file_ids.iteritems():
2315
pb.update("Fetch texts", count, num_file_ids)
2317
yield ("file", file_id, altered_versions)
2319
def _find_non_file_keys_to_fetch(self, revision_ids):
2321
yield ("inventory", None, revision_ids)
2324
# XXX: Note ATM no callers actually pay attention to this return
2325
# instead they just use the list of revision ids and ignore
2326
# missing sigs. Consider removing this work entirely
2327
revisions_with_signatures = set(self.signatures.get_parent_map(
2328
[(r,) for r in revision_ids]))
2329
revisions_with_signatures = set(
2330
[r for (r,) in revisions_with_signatures])
2331
revisions_with_signatures.intersection_update(revision_ids)
2332
yield ("signatures", None, revisions_with_signatures)
2335
yield ("revisions", None, revision_ids)
2338
def get_inventory(self, revision_id):
2339
"""Get Inventory object by revision id."""
2340
return self.iter_inventories([revision_id]).next()
2342
def iter_inventories(self, revision_ids, ordering=None):
2343
"""Get many inventories by revision_ids.
2345
This will buffer some or all of the texts used in constructing the
2346
inventories in memory, but will only parse a single inventory at a
2349
:param revision_ids: The expected revision ids of the inventories.
2350
:param ordering: optional ordering, e.g. 'topological'. If not
2351
specified, the order of revision_ids will be preserved (by
2352
buffering if necessary).
2353
:return: An iterator of inventories.
2355
if ((None in revision_ids)
2356
or (_mod_revision.NULL_REVISION in revision_ids)):
2357
raise ValueError('cannot get null revision inventory')
2358
return self._iter_inventories(revision_ids, ordering)
2360
def _iter_inventories(self, revision_ids, ordering):
2361
"""single-document based inventory iteration."""
2362
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2363
for text, revision_id in inv_xmls:
2364
yield self._deserialise_inventory(revision_id, text)
2366
def _iter_inventory_xmls(self, revision_ids, ordering):
2367
if ordering is None:
2368
order_as_requested = True
2369
ordering = 'unordered'
2371
order_as_requested = False
2372
keys = [(revision_id,) for revision_id in revision_ids]
2375
if order_as_requested:
2376
key_iter = iter(keys)
2377
next_key = key_iter.next()
2378
stream = self.inventories.get_record_stream(keys, ordering, True)
2380
for record in stream:
2381
if record.storage_kind != 'absent':
2382
chunks = record.get_bytes_as('chunked')
2383
if order_as_requested:
2384
text_chunks[record.key] = chunks
2386
yield ''.join(chunks), record.key[-1]
2388
raise errors.NoSuchRevision(self, record.key)
2389
if order_as_requested:
2390
# Yield as many results as we can while preserving order.
2391
while next_key in text_chunks:
2392
chunks = text_chunks.pop(next_key)
2393
yield ''.join(chunks), next_key[-1]
2395
next_key = key_iter.next()
2396
except StopIteration:
2397
# We still want to fully consume the get_record_stream,
2398
# just in case it is not actually finished at this point
2402
def _deserialise_inventory(self, revision_id, xml):
2403
"""Transform the xml into an inventory object.
2405
:param revision_id: The expected revision id of the inventory.
2406
:param xml: A serialised inventory.
2408
result = self._serializer.read_inventory_from_string(xml, revision_id,
2409
entry_cache=self._inventory_entry_cache,
2410
return_from_cache=self._safe_to_return_from_cache)
2411
if result.revision_id != revision_id:
2412
raise AssertionError('revision id mismatch %s != %s' % (
2413
result.revision_id, revision_id))
2416
def get_serializer_format(self):
2417
return self._serializer.format_num
2420
def _get_inventory_xml(self, revision_id):
2421
"""Get serialized inventory as a string."""
2422
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2424
text, revision_id = texts.next()
2425
except StopIteration:
2426
raise errors.HistoryMissing(self, 'inventory', revision_id)
947
raise NotImplementedError(self.iter_files_bytes)
2429
949
def get_rev_id_for_revno(self, revno, known_pair):
2430
950
"""Return the revision id of a revno, given a later (revno, revid)
2762
1221
except UnicodeDecodeError:
2763
1222
raise errors.NonAsciiRevisionId(method, self)
2765
def revision_graph_can_have_wrong_parents(self):
2766
"""Is it possible for this repository to have a revision graph with
2769
If True, then this repository must also implement
2770
_find_inconsistent_revision_parents so that check and reconcile can
2771
check for inconsistencies before proceeding with other checks that may
2772
depend on the revision index being consistent.
2774
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2777
# remove these delegates a while after bzr 0.15
2778
def __make_delegated(name, from_module):
2779
def _deprecated_repository_forwarder():
2780
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2781
% (name, from_module),
2784
m = __import__(from_module, globals(), locals(), [name])
2786
return getattr(m, name)
2787
except AttributeError:
2788
raise AttributeError('module %s has no name %s'
2790
globals()[name] = _deprecated_repository_forwarder
2793
'AllInOneRepository',
2794
'WeaveMetaDirRepository',
2795
'PreSplitOutRepositoryFormat',
2796
'RepositoryFormat4',
2797
'RepositoryFormat5',
2798
'RepositoryFormat6',
2799
'RepositoryFormat7',
2801
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2805
'RepositoryFormatKnit',
2806
'RepositoryFormatKnit1',
2808
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2811
def install_revision(repository, rev, revision_tree):
2812
"""Install all revision data into a repository."""
2813
install_revisions(repository, [(rev, revision_tree, None)])
2816
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2817
"""Install all revision data into a repository.
2819
Accepts an iterable of revision, tree, signature tuples. The signature
2822
repository.start_write_group()
2824
inventory_cache = lru_cache.LRUCache(10)
2825
for n, (revision, revision_tree, signature) in enumerate(iterable):
2826
_install_revision(repository, revision, revision_tree, signature,
2829
pb.update('Transferring revisions', n + 1, num_revisions)
2831
repository.abort_write_group()
2834
repository.commit_write_group()
2837
def _install_revision(repository, rev, revision_tree, signature,
2839
"""Install all revision data into a repository."""
2840
present_parents = []
2842
for p_id in rev.parent_ids:
2843
if repository.has_revision(p_id):
2844
present_parents.append(p_id)
2845
parent_trees[p_id] = repository.revision_tree(p_id)
2847
parent_trees[p_id] = repository.revision_tree(
2848
_mod_revision.NULL_REVISION)
2850
inv = revision_tree.inventory
2851
entries = inv.iter_entries()
2852
# backwards compatibility hack: skip the root id.
2853
if not repository.supports_rich_root():
2854
path, root = entries.next()
2855
if root.revision != rev.revision_id:
2856
raise errors.IncompatibleRevision(repr(repository))
2858
for path, ie in entries:
2859
text_keys[(ie.file_id, ie.revision)] = ie
2860
text_parent_map = repository.texts.get_parent_map(text_keys)
2861
missing_texts = set(text_keys) - set(text_parent_map)
2862
# Add the texts that are not already present
2863
for text_key in missing_texts:
2864
ie = text_keys[text_key]
2866
# FIXME: TODO: The following loop overlaps/duplicates that done by
2867
# commit to determine parents. There is a latent/real bug here where
2868
# the parents inserted are not those commit would do - in particular
2869
# they are not filtered by heads(). RBC, AB
2870
for revision, tree in parent_trees.iteritems():
2871
if ie.file_id not in tree:
2873
parent_id = tree.inventory[ie.file_id].revision
2874
if parent_id in text_parents:
2876
text_parents.append((ie.file_id, parent_id))
2877
lines = revision_tree.get_file(ie.file_id).readlines()
2878
repository.texts.add_lines(text_key, text_parents, lines)
2880
# install the inventory
2881
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2882
# Cache this inventory
2883
inventory_cache[rev.revision_id] = inv
2885
basis_inv = inventory_cache[rev.parent_ids[0]]
2887
repository.add_inventory(rev.revision_id, inv, present_parents)
2889
delta = inv._make_delta(basis_inv)
2890
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2891
rev.revision_id, present_parents)
2893
repository.add_inventory(rev.revision_id, inv, present_parents)
2894
except errors.RevisionAlreadyPresent:
2896
if signature is not None:
2897
repository.add_signature_text(rev.revision_id, signature)
2898
repository.add_revision(rev.revision_id, rev, inv)
2901
1225
class MetaDirRepository(Repository):
2902
1226
"""Repositories in the new meta-dir layout.
3258
1558
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
3259
1559
format_registry.register_lazy(
3260
1560
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
3261
'bzrlib.repofmt.pack_repo',
1561
'bzrlib.repofmt.knitpack_repo',
3262
1562
'RepositoryFormatKnitPack1',
3264
1564
format_registry.register_lazy(
3265
1565
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
3266
'bzrlib.repofmt.pack_repo',
1566
'bzrlib.repofmt.knitpack_repo',
3267
1567
'RepositoryFormatKnitPack3',
3269
1569
format_registry.register_lazy(
3270
1570
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
3271
'bzrlib.repofmt.pack_repo',
1571
'bzrlib.repofmt.knitpack_repo',
3272
1572
'RepositoryFormatKnitPack4',
3274
1574
format_registry.register_lazy(
3275
1575
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
3276
'bzrlib.repofmt.pack_repo',
1576
'bzrlib.repofmt.knitpack_repo',
3277
1577
'RepositoryFormatKnitPack5',
3279
1579
format_registry.register_lazy(
3280
1580
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
3281
'bzrlib.repofmt.pack_repo',
1581
'bzrlib.repofmt.knitpack_repo',
3282
1582
'RepositoryFormatKnitPack5RichRoot',
3284
1584
format_registry.register_lazy(
3285
1585
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3286
'bzrlib.repofmt.pack_repo',
1586
'bzrlib.repofmt.knitpack_repo',
3287
1587
'RepositoryFormatKnitPack5RichRootBroken',
3289
1589
format_registry.register_lazy(
3290
1590
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3291
'bzrlib.repofmt.pack_repo',
1591
'bzrlib.repofmt.knitpack_repo',
3292
1592
'RepositoryFormatKnitPack6',
3294
1594
format_registry.register_lazy(
3295
1595
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3296
'bzrlib.repofmt.pack_repo',
1596
'bzrlib.repofmt.knitpack_repo',
3297
1597
'RepositoryFormatKnitPack6RichRoot',
1599
format_registry.register_lazy(
1600
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
1601
'bzrlib.repofmt.groupcompress_repo',
1602
'RepositoryFormat2a',
3300
1605
# Development formats.
3301
# Obsolete but kept pending a CHK based subtree format.
1606
# Check their docstrings to see if/when they are obsolete.
3302
1607
format_registry.register_lazy(
3303
1608
("Bazaar development format 2 with subtree support "
3304
1609
"(needs bzr.dev from before 1.8)\n"),
3305
'bzrlib.repofmt.pack_repo',
1610
'bzrlib.repofmt.knitpack_repo',
3306
1611
'RepositoryFormatPackDevelopment2Subtree',
3309
# 1.14->1.16 go below here
3310
format_registry.register_lazy(
3311
'Bazaar development format - group compression and chk inventory'
3312
' (needs bzr.dev from 1.14)\n',
3313
'bzrlib.repofmt.groupcompress_repo',
3314
'RepositoryFormatCHK1',
3317
format_registry.register_lazy(
3318
'Bazaar development format - chk repository with bencode revision '
3319
'serialization (needs bzr.dev from 1.16)\n',
3320
'bzrlib.repofmt.groupcompress_repo',
3321
'RepositoryFormatCHK2',
3323
format_registry.register_lazy(
3324
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3325
'bzrlib.repofmt.groupcompress_repo',
3326
'RepositoryFormat2a',
1613
format_registry.register_lazy(
1614
'Bazaar development format 8\n',
1615
'bzrlib.repofmt.groupcompress_repo',
1616
'RepositoryFormat2aSubtree',
3360
1649
self.target.fetch(self.source, revision_id=revision_id)
3362
1651
@needs_write_lock
3363
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
1652
def fetch(self, revision_id=None, find_ghosts=False):
3365
1653
"""Fetch the content required to construct revision_id.
3367
1655
The content is copied from self.source to self.target.
3369
1657
:param revision_id: if None all content is copied, if NULL_REVISION no
3370
1658
content is copied.
3374
ui.ui_factory.warn_experimental_format_fetch(self)
3375
from bzrlib.fetch import RepoFetcher
3376
# See <https://launchpad.net/bugs/456077> asking for a warning here
3377
if self.source._format.network_name() != self.target._format.network_name():
3378
ui.ui_factory.show_user_warning('cross_format_fetch',
3379
from_format=self.source._format,
3380
to_format=self.target._format)
3381
f = RepoFetcher(to_repository=self.target,
3382
from_repository=self.source,
3383
last_revision=revision_id,
3384
fetch_spec=fetch_spec,
3385
find_ghosts=find_ghosts)
3387
def _walk_to_common_revisions(self, revision_ids):
3388
"""Walk out from revision_ids in source to revisions target has.
3390
:param revision_ids: The start point for the search.
3391
:return: A set of revision ids.
3393
target_graph = self.target.get_graph()
3394
revision_ids = frozenset(revision_ids)
3395
missing_revs = set()
3396
source_graph = self.source.get_graph()
3397
# ensure we don't pay silly lookup costs.
3398
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3399
null_set = frozenset([_mod_revision.NULL_REVISION])
3400
searcher_exhausted = False
3404
# Iterate the searcher until we have enough next_revs
3405
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3407
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3408
next_revs.update(next_revs_part)
3409
ghosts.update(ghosts_part)
3410
except StopIteration:
3411
searcher_exhausted = True
3413
# If there are ghosts in the source graph, and the caller asked for
3414
# them, make sure that they are present in the target.
3415
# We don't care about other ghosts as we can't fetch them and
3416
# haven't been asked to.
3417
ghosts_to_check = set(revision_ids.intersection(ghosts))
3418
revs_to_get = set(next_revs).union(ghosts_to_check)
3420
have_revs = set(target_graph.get_parent_map(revs_to_get))
3421
# we always have NULL_REVISION present.
3422
have_revs = have_revs.union(null_set)
3423
# Check if the target is missing any ghosts we need.
3424
ghosts_to_check.difference_update(have_revs)
3426
# One of the caller's revision_ids is a ghost in both the
3427
# source and the target.
3428
raise errors.NoSuchRevision(
3429
self.source, ghosts_to_check.pop())
3430
missing_revs.update(next_revs - have_revs)
3431
# Because we may have walked past the original stop point, make
3432
# sure everything is stopped
3433
stop_revs = searcher.find_seen_ancestors(have_revs)
3434
searcher.stop_searching_any(stop_revs)
3435
if searcher_exhausted:
3437
return searcher.get_result()
1661
raise NotImplementedError(self.fetch)
3439
1663
@needs_read_lock
3440
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1664
def search_missing_revision_ids(self,
1665
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
1666
find_ghosts=True, revision_ids=None, if_present_ids=None,
3441
1668
"""Return the revision ids that source has that target does not.
3443
1670
:param revision_id: only return revision ids included by this
1672
:param revision_ids: return revision ids included by these
1673
revision_ids. NoSuchRevision will be raised if any of these
1674
revisions are not present.
1675
:param if_present_ids: like revision_ids, but will not cause
1676
NoSuchRevision if any of these are absent, instead they will simply
1677
not be in the result. This is useful for e.g. finding revisions
1678
to fetch for tags, which may reference absent revisions.
3445
1679
:param find_ghosts: If True find missing revisions in deep history
3446
1680
rather than just finding the surface difference.
1681
:param limit: Maximum number of revisions to return, topologically
3447
1683
:return: A bzrlib.graph.SearchResult.
3449
# stop searching at found target revisions.
3450
if not find_ghosts and revision_id is not None:
3451
return self._walk_to_common_revisions([revision_id])
3452
# generic, possibly worst case, slow code path.
3453
target_ids = set(self.target.all_revision_ids())
3454
if revision_id is not None:
3455
source_ids = self.source.get_ancestry(revision_id)
3456
if source_ids[0] is not None:
3457
raise AssertionError()
3460
source_ids = self.source.all_revision_ids()
3461
result_set = set(source_ids).difference(target_ids)
3462
return self.source.revision_ids_to_search_result(result_set)
1685
raise NotImplementedError(self.search_missing_revision_ids)
3465
1688
def _same_model(source, target):
3486
1709
"different serializers")
3489
class InterSameDataRepository(InterRepository):
3490
"""Code for converting between repositories that represent the same data.
3492
Data format and model must match for this to work.
3496
def _get_repo_format_to_test(self):
3497
"""Repository format for testing with.
3499
InterSameData can pull from subtree to subtree and from non-subtree to
3500
non-subtree, so we test this with the richest repository format.
3502
from bzrlib.repofmt import knitrepo
3503
return knitrepo.RepositoryFormatKnit3()
3506
def is_compatible(source, target):
3507
return InterRepository._same_model(source, target)
3510
class InterWeaveRepo(InterSameDataRepository):
3511
"""Optimised code paths between Weave based repositories.
3513
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3514
implemented lazy inter-object optimisation.
3518
def _get_repo_format_to_test(self):
3519
from bzrlib.repofmt import weaverepo
3520
return weaverepo.RepositoryFormat7()
3523
def is_compatible(source, target):
3524
"""Be compatible with known Weave formats.
3526
We don't test for the stores being of specific types because that
3527
could lead to confusing results, and there is no need to be
3530
from bzrlib.repofmt.weaverepo import (
3536
return (isinstance(source._format, (RepositoryFormat5,
3538
RepositoryFormat7)) and
3539
isinstance(target._format, (RepositoryFormat5,
3541
RepositoryFormat7)))
3542
except AttributeError:
3546
def copy_content(self, revision_id=None):
3547
"""See InterRepository.copy_content()."""
3548
# weave specific optimised path:
3550
self.target.set_make_working_trees(self.source.make_working_trees())
3551
except (errors.RepositoryUpgradeRequired, NotImplemented):
3553
# FIXME do not peek!
3554
if self.source._transport.listable():
3555
pb = ui.ui_factory.nested_progress_bar()
3557
self.target.texts.insert_record_stream(
3558
self.source.texts.get_record_stream(
3559
self.source.texts.keys(), 'topological', False))
3560
pb.update('Copying inventory', 0, 1)
3561
self.target.inventories.insert_record_stream(
3562
self.source.inventories.get_record_stream(
3563
self.source.inventories.keys(), 'topological', False))
3564
self.target.signatures.insert_record_stream(
3565
self.source.signatures.get_record_stream(
3566
self.source.signatures.keys(),
3568
self.target.revisions.insert_record_stream(
3569
self.source.revisions.get_record_stream(
3570
self.source.revisions.keys(),
3571
'topological', True))
3575
self.target.fetch(self.source, revision_id=revision_id)
3578
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3579
"""See InterRepository.missing_revision_ids()."""
3580
# we want all revisions to satisfy revision_id in source.
3581
# but we don't want to stat every file here and there.
3582
# we want then, all revisions other needs to satisfy revision_id
3583
# checked, but not those that we have locally.
3584
# so the first thing is to get a subset of the revisions to
3585
# satisfy revision_id in source, and then eliminate those that
3586
# we do already have.
3587
# this is slow on high latency connection to self, but as this
3588
# disk format scales terribly for push anyway due to rewriting
3589
# inventory.weave, this is considered acceptable.
3591
if revision_id is not None:
3592
source_ids = self.source.get_ancestry(revision_id)
3593
if source_ids[0] is not None:
3594
raise AssertionError()
3597
source_ids = self.source._all_possible_ids()
3598
source_ids_set = set(source_ids)
3599
# source_ids is the worst possible case we may need to pull.
3600
# now we want to filter source_ids against what we actually
3601
# have in target, but don't try to check for existence where we know
3602
# we do not have a revision as that would be pointless.
3603
target_ids = set(self.target._all_possible_ids())
3604
possibly_present_revisions = target_ids.intersection(source_ids_set)
3605
actually_present_revisions = set(
3606
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3607
required_revisions = source_ids_set.difference(actually_present_revisions)
3608
if revision_id is not None:
3609
# we used get_ancestry to determine source_ids then we are assured all
3610
# revisions referenced are present as they are installed in topological order.
3611
# and the tip revision was validated by get_ancestry.
3612
result_set = required_revisions
3614
# if we just grabbed the possibly available ids, then
3615
# we only have an estimate of whats available and need to validate
3616
# that against the revision records.
3618
self.source._eliminate_revisions_not_present(required_revisions))
3619
return self.source.revision_ids_to_search_result(result_set)
3622
class InterKnitRepo(InterSameDataRepository):
3623
"""Optimised code paths between Knit based repositories."""
3626
def _get_repo_format_to_test(self):
3627
from bzrlib.repofmt import knitrepo
3628
return knitrepo.RepositoryFormatKnit1()
3631
def is_compatible(source, target):
3632
"""Be compatible with known Knit formats.
3634
We don't test for the stores being of specific types because that
3635
could lead to confusing results, and there is no need to be
3638
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3640
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3641
isinstance(target._format, RepositoryFormatKnit))
3642
except AttributeError:
3644
return are_knits and InterRepository._same_model(source, target)
3647
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3648
"""See InterRepository.missing_revision_ids()."""
3649
if revision_id is not None:
3650
source_ids = self.source.get_ancestry(revision_id)
3651
if source_ids[0] is not None:
3652
raise AssertionError()
3655
source_ids = self.source.all_revision_ids()
3656
source_ids_set = set(source_ids)
3657
# source_ids is the worst possible case we may need to pull.
3658
# now we want to filter source_ids against what we actually
3659
# have in target, but don't try to check for existence where we know
3660
# we do not have a revision as that would be pointless.
3661
target_ids = set(self.target.all_revision_ids())
3662
possibly_present_revisions = target_ids.intersection(source_ids_set)
3663
actually_present_revisions = set(
3664
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3665
required_revisions = source_ids_set.difference(actually_present_revisions)
3666
if revision_id is not None:
3667
# we used get_ancestry to determine source_ids then we are assured all
3668
# revisions referenced are present as they are installed in topological order.
3669
# and the tip revision was validated by get_ancestry.
3670
result_set = required_revisions
3672
# if we just grabbed the possibly available ids, then
3673
# we only have an estimate of whats available and need to validate
3674
# that against the revision records.
3676
self.source._eliminate_revisions_not_present(required_revisions))
3677
return self.source.revision_ids_to_search_result(result_set)
3680
class InterDifferingSerializer(InterRepository):
3683
def _get_repo_format_to_test(self):
3687
def is_compatible(source, target):
3688
"""Be compatible with Knit2 source and Knit3 target"""
3689
# This is redundant with format.check_conversion_target(), however that
3690
# raises an exception, and we just want to say "False" as in we won't
3691
# support converting between these formats.
3692
if 'IDS_never' in debug.debug_flags:
3694
if source.supports_rich_root() and not target.supports_rich_root():
3696
if (source._format.supports_tree_reference
3697
and not target._format.supports_tree_reference):
3699
if target._fallback_repositories and target._format.supports_chks:
3700
# IDS doesn't know how to copy CHKs for the parent inventories it
3701
# adds to stacked repos.
3703
if 'IDS_always' in debug.debug_flags:
3705
# Only use this code path for local source and target. IDS does far
3706
# too much IO (both bandwidth and roundtrips) over a network.
3707
if not source.bzrdir.transport.base.startswith('file:///'):
3709
if not target.bzrdir.transport.base.startswith('file:///'):
3713
def _get_trees(self, revision_ids, cache):
3715
for rev_id in revision_ids:
3717
possible_trees.append((rev_id, cache[rev_id]))
3719
# Not cached, but inventory might be present anyway.
3721
tree = self.source.revision_tree(rev_id)
3722
except errors.NoSuchRevision:
3723
# Nope, parent is ghost.
3726
cache[rev_id] = tree
3727
possible_trees.append((rev_id, tree))
3728
return possible_trees
3730
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3731
"""Get the best delta and base for this revision.
3733
:return: (basis_id, delta)
3736
# Generate deltas against each tree, to find the shortest.
3737
texts_possibly_new_in_tree = set()
3738
for basis_id, basis_tree in possible_trees:
3739
delta = tree.inventory._make_delta(basis_tree.inventory)
3740
for old_path, new_path, file_id, new_entry in delta:
3741
if new_path is None:
3742
# This file_id isn't present in the new rev, so we don't
3746
# Rich roots are handled elsewhere...
3748
kind = new_entry.kind
3749
if kind != 'directory' and kind != 'file':
3750
# No text record associated with this inventory entry.
3752
# This is a directory or file that has changed somehow.
3753
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3754
deltas.append((len(delta), basis_id, delta))
3756
return deltas[0][1:]
3758
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3759
"""Find all parent revisions that are absent, but for which the
3760
inventory is present, and copy those inventories.
3762
This is necessary to preserve correctness when the source is stacked
3763
without fallbacks configured. (Note that in cases like upgrade the
3764
source may be not have _fallback_repositories even though it is
3768
for parents in parent_map.values():
3769
parent_revs.update(parents)
3770
present_parents = self.source.get_parent_map(parent_revs)
3771
absent_parents = set(parent_revs).difference(present_parents)
3772
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3773
(rev_id,) for rev_id in absent_parents)
3774
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3775
for parent_tree in self.source.revision_trees(parent_inv_ids):
3776
current_revision_id = parent_tree.get_revision_id()
3777
parents_parents_keys = parent_invs_keys_for_stacking[
3778
(current_revision_id,)]
3779
parents_parents = [key[-1] for key in parents_parents_keys]
3780
basis_id = _mod_revision.NULL_REVISION
3781
basis_tree = self.source.revision_tree(basis_id)
3782
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3783
self.target.add_inventory_by_delta(
3784
basis_id, delta, current_revision_id, parents_parents)
3785
cache[current_revision_id] = parent_tree
3787
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3788
"""Fetch across a few revisions.
3790
:param revision_ids: The revisions to copy
3791
:param basis_id: The revision_id of a tree that must be in cache, used
3792
as a basis for delta when no other base is available
3793
:param cache: A cache of RevisionTrees that we can use.
3794
:param a_graph: A Graph object to determine the heads() of the
3795
rich-root data stream.
3796
:return: The revision_id of the last converted tree. The RevisionTree
3797
for it will be in cache
3799
# Walk though all revisions; get inventory deltas, copy referenced
3800
# texts that delta references, insert the delta, revision and
3802
root_keys_to_create = set()
3805
pending_revisions = []
3806
parent_map = self.source.get_parent_map(revision_ids)
3807
self._fetch_parent_invs_for_stacking(parent_map, cache)
3808
self.source._safe_to_return_from_cache = True
3809
for tree in self.source.revision_trees(revision_ids):
3810
# Find a inventory delta for this revision.
3811
# Find text entries that need to be copied, too.
3812
current_revision_id = tree.get_revision_id()
3813
parent_ids = parent_map.get(current_revision_id, ())
3814
parent_trees = self._get_trees(parent_ids, cache)
3815
possible_trees = list(parent_trees)
3816
if len(possible_trees) == 0:
3817
# There either aren't any parents, or the parents are ghosts,
3818
# so just use the last converted tree.
3819
possible_trees.append((basis_id, cache[basis_id]))
3820
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3822
revision = self.source.get_revision(current_revision_id)
3823
pending_deltas.append((basis_id, delta,
3824
current_revision_id, revision.parent_ids))
3825
if self._converting_to_rich_root:
3826
self._revision_id_to_root_id[current_revision_id] = \
3828
# Determine which texts are in present in this revision but not in
3829
# any of the available parents.
3830
texts_possibly_new_in_tree = set()
3831
for old_path, new_path, file_id, entry in delta:
3832
if new_path is None:
3833
# This file_id isn't present in the new rev
3837
if not self.target.supports_rich_root():
3838
# The target doesn't support rich root, so we don't
3841
if self._converting_to_rich_root:
3842
# This can't be copied normally, we have to insert
3844
root_keys_to_create.add((file_id, entry.revision))
3847
texts_possibly_new_in_tree.add((file_id, entry.revision))
3848
for basis_id, basis_tree in possible_trees:
3849
basis_inv = basis_tree.inventory
3850
for file_key in list(texts_possibly_new_in_tree):
3851
file_id, file_revision = file_key
3853
entry = basis_inv[file_id]
3854
except errors.NoSuchId:
3856
if entry.revision == file_revision:
3857
texts_possibly_new_in_tree.remove(file_key)
3858
text_keys.update(texts_possibly_new_in_tree)
3859
pending_revisions.append(revision)
3860
cache[current_revision_id] = tree
3861
basis_id = current_revision_id
3862
self.source._safe_to_return_from_cache = False
3864
from_texts = self.source.texts
3865
to_texts = self.target.texts
3866
if root_keys_to_create:
3867
root_stream = _mod_fetch._new_root_data_stream(
3868
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3869
self.source, graph=a_graph)
3870
to_texts.insert_record_stream(root_stream)
3871
to_texts.insert_record_stream(from_texts.get_record_stream(
3872
text_keys, self.target._format._fetch_order,
3873
not self.target._format._fetch_uses_deltas))
3874
# insert inventory deltas
3875
for delta in pending_deltas:
3876
self.target.add_inventory_by_delta(*delta)
3877
if self.target._fallback_repositories:
3878
# Make sure this stacked repository has all the parent inventories
3879
# for the new revisions that we are about to insert. We do this
3880
# before adding the revisions so that no revision is added until
3881
# all the inventories it may depend on are added.
3882
# Note that this is overzealous, as we may have fetched these in an
3885
revision_ids = set()
3886
for revision in pending_revisions:
3887
revision_ids.add(revision.revision_id)
3888
parent_ids.update(revision.parent_ids)
3889
parent_ids.difference_update(revision_ids)
3890
parent_ids.discard(_mod_revision.NULL_REVISION)
3891
parent_map = self.source.get_parent_map(parent_ids)
3892
# we iterate over parent_map and not parent_ids because we don't
3893
# want to try copying any revision which is a ghost
3894
for parent_tree in self.source.revision_trees(parent_map):
3895
current_revision_id = parent_tree.get_revision_id()
3896
parents_parents = parent_map[current_revision_id]
3897
possible_trees = self._get_trees(parents_parents, cache)
3898
if len(possible_trees) == 0:
3899
# There either aren't any parents, or the parents are
3900
# ghosts, so just use the last converted tree.
3901
possible_trees.append((basis_id, cache[basis_id]))
3902
basis_id, delta = self._get_delta_for_revision(parent_tree,
3903
parents_parents, possible_trees)
3904
self.target.add_inventory_by_delta(
3905
basis_id, delta, current_revision_id, parents_parents)
3906
# insert signatures and revisions
3907
for revision in pending_revisions:
3909
signature = self.source.get_signature_text(
3910
revision.revision_id)
3911
self.target.add_signature_text(revision.revision_id,
3913
except errors.NoSuchRevision:
3915
self.target.add_revision(revision.revision_id, revision)
3918
def _fetch_all_revisions(self, revision_ids, pb):
3919
"""Fetch everything for the list of revisions.
3921
:param revision_ids: The list of revisions to fetch. Must be in
3923
:param pb: A ProgressTask
3926
basis_id, basis_tree = self._get_basis(revision_ids[0])
3928
cache = lru_cache.LRUCache(100)
3929
cache[basis_id] = basis_tree
3930
del basis_tree # We don't want to hang on to it here
3932
if self._converting_to_rich_root and len(revision_ids) > 100:
3933
a_graph = _mod_fetch._get_rich_root_heads_graph(self.source,
3938
for offset in range(0, len(revision_ids), batch_size):
3939
self.target.start_write_group()
3941
pb.update('Transferring revisions', offset,
3943
batch = revision_ids[offset:offset+batch_size]
3944
basis_id = self._fetch_batch(batch, basis_id, cache,
3947
self.source._safe_to_return_from_cache = False
3948
self.target.abort_write_group()
3951
hint = self.target.commit_write_group()
3954
if hints and self.target._format.pack_compresses:
3955
self.target.pack(hint=hints)
3956
pb.update('Transferring revisions', len(revision_ids),
3960
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3962
"""See InterRepository.fetch()."""
3963
if fetch_spec is not None:
3964
raise AssertionError("Not implemented yet...")
3965
ui.ui_factory.warn_experimental_format_fetch(self)
3966
if (not self.source.supports_rich_root()
3967
and self.target.supports_rich_root()):
3968
self._converting_to_rich_root = True
3969
self._revision_id_to_root_id = {}
3971
self._converting_to_rich_root = False
3972
# See <https://launchpad.net/bugs/456077> asking for a warning here
3973
if self.source._format.network_name() != self.target._format.network_name():
3974
ui.ui_factory.show_user_warning('cross_format_fetch',
3975
from_format=self.source._format,
3976
to_format=self.target._format)
3977
revision_ids = self.target.search_missing_revision_ids(self.source,
3978
revision_id, find_ghosts=find_ghosts).get_keys()
3979
if not revision_ids:
3981
revision_ids = tsort.topo_sort(
3982
self.source.get_graph().get_parent_map(revision_ids))
3983
if not revision_ids:
3985
# Walk though all revisions; get inventory deltas, copy referenced
3986
# texts that delta references, insert the delta, revision and
3989
my_pb = ui.ui_factory.nested_progress_bar()
3992
symbol_versioning.warn(
3993
symbol_versioning.deprecated_in((1, 14, 0))
3994
% "pb parameter to fetch()")
3997
self._fetch_all_revisions(revision_ids, pb)
3999
if my_pb is not None:
4001
return len(revision_ids), 0
4003
def _get_basis(self, first_revision_id):
4004
"""Get a revision and tree which exists in the target.
4006
This assumes that first_revision_id is selected for transmission
4007
because all other ancestors are already present. If we can't find an
4008
ancestor we fall back to NULL_REVISION since we know that is safe.
4010
:return: (basis_id, basis_tree)
4012
first_rev = self.source.get_revision(first_revision_id)
4014
basis_id = first_rev.parent_ids[0]
4015
# only valid as a basis if the target has it
4016
self.target.get_revision(basis_id)
4017
# Try to get a basis tree - if its a ghost it will hit the
4018
# NoSuchRevision case.
4019
basis_tree = self.source.revision_tree(basis_id)
4020
except (IndexError, errors.NoSuchRevision):
4021
basis_id = _mod_revision.NULL_REVISION
4022
basis_tree = self.source.revision_tree(basis_id)
4023
return basis_id, basis_tree
4026
InterRepository.register_optimiser(InterDifferingSerializer)
4027
InterRepository.register_optimiser(InterSameDataRepository)
4028
InterRepository.register_optimiser(InterWeaveRepo)
4029
InterRepository.register_optimiser(InterKnitRepo)
4032
1712
class CopyConverter(object):
4033
1713
"""A repository conversion tool which just performs a copy of the content.
4055
1735
# trigger an assertion if not such
4056
1736
repo._format.get_format_string()
4057
1737
self.repo_dir = repo.bzrdir
4058
pb.update('Moving repository to repository.backup')
1738
pb.update(gettext('Moving repository to repository.backup'))
4059
1739
self.repo_dir.transport.move('repository', 'repository.backup')
4060
1740
backup_transport = self.repo_dir.transport.clone('repository.backup')
4061
1741
repo._format.check_conversion_target(self.target_format)
4062
1742
self.source_repo = repo._format.open(self.repo_dir,
4064
1744
_override_transport=backup_transport)
4065
pb.update('Creating new repository')
1745
pb.update(gettext('Creating new repository'))
4066
1746
converted = self.target_format.initialize(self.repo_dir,
4067
1747
self.source_repo.is_shared())
4068
1748
converted.lock_write()
4070
pb.update('Copying content')
1750
pb.update(gettext('Copying content'))
4071
1751
self.source_repo.copy_content_into(converted)
4073
1753
converted.unlock()
4074
pb.update('Deleting old repository content')
1754
pb.update(gettext('Deleting old repository content'))
4075
1755
self.repo_dir.transport.delete_tree('repository.backup')
4076
ui.ui_factory.note('repository converted')
1756
ui.ui_factory.note(gettext('repository converted'))
4089
def _unescaper(match, _map=_unescape_map):
4090
code = match.group(1)
4094
if not code.startswith('#'):
4096
return unichr(int(code[1:])).encode('utf8')
4102
def _unescape_xml(data):
4103
"""Unescape predefined XML entities in a string of data."""
4105
if _unescape_re is None:
4106
_unescape_re = re.compile('\&([^;]*);')
4107
return _unescape_re.sub(_unescaper, data)
4110
class _VersionedFileChecker(object):
4112
def __init__(self, repository, text_key_references=None, ancestors=None):
4113
self.repository = repository
4114
self.text_index = self.repository._generate_text_key_index(
4115
text_key_references=text_key_references, ancestors=ancestors)
4117
def calculate_file_version_parents(self, text_key):
4118
"""Calculate the correct parents for a file version according to
4121
parent_keys = self.text_index[text_key]
4122
if parent_keys == [_mod_revision.NULL_REVISION]:
4124
return tuple(parent_keys)
4126
def check_file_version_parents(self, texts, progress_bar=None):
4127
"""Check the parents stored in a versioned file are correct.
4129
It also detects file versions that are not referenced by their
4130
corresponding revision's inventory.
4132
:returns: A tuple of (wrong_parents, dangling_file_versions).
4133
wrong_parents is a dict mapping {revision_id: (stored_parents,
4134
correct_parents)} for each revision_id where the stored parents
4135
are not correct. dangling_file_versions is a set of (file_id,
4136
revision_id) tuples for versions that are present in this versioned
4137
file, but not used by the corresponding inventory.
4139
local_progress = None
4140
if progress_bar is None:
4141
local_progress = ui.ui_factory.nested_progress_bar()
4142
progress_bar = local_progress
4144
return self._check_file_version_parents(texts, progress_bar)
4147
local_progress.finished()
4149
def _check_file_version_parents(self, texts, progress_bar):
4150
"""See check_file_version_parents."""
4152
self.file_ids = set([file_id for file_id, _ in
4153
self.text_index.iterkeys()])
4154
# text keys is now grouped by file_id
4155
n_versions = len(self.text_index)
4156
progress_bar.update('loading text store', 0, n_versions)
4157
parent_map = self.repository.texts.get_parent_map(self.text_index)
4158
# On unlistable transports this could well be empty/error...
4159
text_keys = self.repository.texts.keys()
4160
unused_keys = frozenset(text_keys) - set(self.text_index)
4161
for num, key in enumerate(self.text_index.iterkeys()):
4162
progress_bar.update('checking text graph', num, n_versions)
4163
correct_parents = self.calculate_file_version_parents(key)
4165
knit_parents = parent_map[key]
4166
except errors.RevisionNotPresent:
4169
if correct_parents != knit_parents:
4170
wrong_parents[key] = (knit_parents, correct_parents)
4171
return wrong_parents, unused_keys
4174
def _old_get_graph(repository, revision_id):
4175
"""DO NOT USE. That is all. I'm serious."""
4176
graph = repository.get_graph()
4177
revision_graph = dict(((key, value) for key, value in
4178
graph.iter_ancestry([revision_id]) if value is not None))
4179
return _strip_NULL_ghosts(revision_graph)
4182
1760
def _strip_NULL_ghosts(revision_graph):
4183
1761
"""Also don't use this. more compatibility code for unmigrated clients."""
4184
1762
# Filter ghosts, and null:
4190
1768
return revision_graph
4193
class StreamSink(object):
4194
"""An object that can insert a stream into a repository.
4196
This interface handles the complexity of reserialising inventories and
4197
revisions from different formats, and allows unidirectional insertion into
4198
stacked repositories without looking for the missing basis parents
4202
def __init__(self, target_repo):
4203
self.target_repo = target_repo
4205
def insert_stream(self, stream, src_format, resume_tokens):
4206
"""Insert a stream's content into the target repository.
4208
:param src_format: a bzr repository format.
4210
:return: a list of resume tokens and an iterable of keys additional
4211
items required before the insertion can be completed.
4213
self.target_repo.lock_write()
4216
self.target_repo.resume_write_group(resume_tokens)
4219
self.target_repo.start_write_group()
4222
# locked_insert_stream performs a commit|suspend.
4223
return self._locked_insert_stream(stream, src_format, is_resume)
4225
self.target_repo.abort_write_group(suppress_errors=True)
4228
self.target_repo.unlock()
4230
def _locked_insert_stream(self, stream, src_format, is_resume):
4231
to_serializer = self.target_repo._format._serializer
4232
src_serializer = src_format._serializer
4234
if to_serializer == src_serializer:
4235
# If serializers match and the target is a pack repository, set the
4236
# write cache size on the new pack. This avoids poor performance
4237
# on transports where append is unbuffered (such as
4238
# RemoteTransport). This is safe to do because nothing should read
4239
# back from the target repository while a stream with matching
4240
# serialization is being inserted.
4241
# The exception is that a delta record from the source that should
4242
# be a fulltext may need to be expanded by the target (see
4243
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4244
# explicitly flush any buffered writes first in that rare case.
4246
new_pack = self.target_repo._pack_collection._new_pack
4247
except AttributeError:
4248
# Not a pack repository
4251
new_pack.set_write_cache_size(1024*1024)
4252
for substream_type, substream in stream:
4253
if 'stream' in debug.debug_flags:
4254
mutter('inserting substream: %s', substream_type)
4255
if substream_type == 'texts':
4256
self.target_repo.texts.insert_record_stream(substream)
4257
elif substream_type == 'inventories':
4258
if src_serializer == to_serializer:
4259
self.target_repo.inventories.insert_record_stream(
4262
self._extract_and_insert_inventories(
4263
substream, src_serializer)
4264
elif substream_type == 'inventory-deltas':
4265
self._extract_and_insert_inventory_deltas(
4266
substream, src_serializer)
4267
elif substream_type == 'chk_bytes':
4268
# XXX: This doesn't support conversions, as it assumes the
4269
# conversion was done in the fetch code.
4270
self.target_repo.chk_bytes.insert_record_stream(substream)
4271
elif substream_type == 'revisions':
4272
# This may fallback to extract-and-insert more often than
4273
# required if the serializers are different only in terms of
4275
if src_serializer == to_serializer:
4276
self.target_repo.revisions.insert_record_stream(
4279
self._extract_and_insert_revisions(substream,
4281
elif substream_type == 'signatures':
4282
self.target_repo.signatures.insert_record_stream(substream)
4284
raise AssertionError('kaboom! %s' % (substream_type,))
4285
# Done inserting data, and the missing_keys calculations will try to
4286
# read back from the inserted data, so flush the writes to the new pack
4287
# (if this is pack format).
4288
if new_pack is not None:
4289
new_pack._write_data('', flush=True)
4290
# Find all the new revisions (including ones from resume_tokens)
4291
missing_keys = self.target_repo.get_missing_parent_inventories(
4292
check_for_missing_texts=is_resume)
4294
for prefix, versioned_file in (
4295
('texts', self.target_repo.texts),
4296
('inventories', self.target_repo.inventories),
4297
('revisions', self.target_repo.revisions),
4298
('signatures', self.target_repo.signatures),
4299
('chk_bytes', self.target_repo.chk_bytes),
4301
if versioned_file is None:
4303
# TODO: key is often going to be a StaticTuple object
4304
# I don't believe we can define a method by which
4305
# (prefix,) + StaticTuple will work, though we could
4306
# define a StaticTuple.sq_concat that would allow you to
4307
# pass in either a tuple or a StaticTuple as the second
4308
# object, so instead we could have:
4309
# StaticTuple(prefix) + key here...
4310
missing_keys.update((prefix,) + key for key in
4311
versioned_file.get_missing_compression_parent_keys())
4312
except NotImplementedError:
4313
# cannot even attempt suspending, and missing would have failed
4314
# during stream insertion.
4315
missing_keys = set()
4318
# suspend the write group and tell the caller what we is
4319
# missing. We know we can suspend or else we would not have
4320
# entered this code path. (All repositories that can handle
4321
# missing keys can handle suspending a write group).
4322
write_group_tokens = self.target_repo.suspend_write_group()
4323
return write_group_tokens, missing_keys
4324
hint = self.target_repo.commit_write_group()
4325
if (to_serializer != src_serializer and
4326
self.target_repo._format.pack_compresses):
4327
self.target_repo.pack(hint=hint)
4330
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4331
target_rich_root = self.target_repo._format.rich_root_data
4332
target_tree_refs = self.target_repo._format.supports_tree_reference
4333
for record in substream:
4334
# Insert the delta directly
4335
inventory_delta_bytes = record.get_bytes_as('fulltext')
4336
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4338
parse_result = deserialiser.parse_text_bytes(
4339
inventory_delta_bytes)
4340
except inventory_delta.IncompatibleInventoryDelta, err:
4341
trace.mutter("Incompatible delta: %s", err.msg)
4342
raise errors.IncompatibleRevision(self.target_repo._format)
4343
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4344
revision_id = new_id
4345
parents = [key[0] for key in record.parents]
4346
self.target_repo.add_inventory_by_delta(
4347
basis_id, inv_delta, revision_id, parents)
4349
def _extract_and_insert_inventories(self, substream, serializer,
4351
"""Generate a new inventory versionedfile in target, converting data.
4353
The inventory is retrieved from the source, (deserializing it), and
4354
stored in the target (reserializing it in a different format).
4356
target_rich_root = self.target_repo._format.rich_root_data
4357
target_tree_refs = self.target_repo._format.supports_tree_reference
4358
for record in substream:
4359
# It's not a delta, so it must be a fulltext in the source
4360
# serializer's format.
4361
bytes = record.get_bytes_as('fulltext')
4362
revision_id = record.key[0]
4363
inv = serializer.read_inventory_from_string(bytes, revision_id)
4364
parents = [key[0] for key in record.parents]
4365
self.target_repo.add_inventory(revision_id, inv, parents)
4366
# No need to keep holding this full inv in memory when the rest of
4367
# the substream is likely to be all deltas.
4370
def _extract_and_insert_revisions(self, substream, serializer):
4371
for record in substream:
4372
bytes = record.get_bytes_as('fulltext')
4373
revision_id = record.key[0]
4374
rev = serializer.read_revision_from_string(bytes)
4375
if rev.revision_id != revision_id:
4376
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4377
self.target_repo.add_revision(revision_id, rev)
4380
if self.target_repo._format._fetch_reconcile:
4381
self.target_repo.reconcile()
4384
class StreamSource(object):
4385
"""A source of a stream for fetching between repositories."""
4387
def __init__(self, from_repository, to_format):
4388
"""Create a StreamSource streaming from from_repository."""
4389
self.from_repository = from_repository
4390
self.to_format = to_format
4392
def delta_on_metadata(self):
4393
"""Return True if delta's are permitted on metadata streams.
4395
That is on revisions and signatures.
4397
src_serializer = self.from_repository._format._serializer
4398
target_serializer = self.to_format._serializer
4399
return (self.to_format._fetch_uses_deltas and
4400
src_serializer == target_serializer)
4402
def _fetch_revision_texts(self, revs):
4403
# fetch signatures first and then the revision texts
4404
# may need to be a InterRevisionStore call here.
4405
from_sf = self.from_repository.signatures
4406
# A missing signature is just skipped.
4407
keys = [(rev_id,) for rev_id in revs]
4408
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4410
self.to_format._fetch_order,
4411
not self.to_format._fetch_uses_deltas))
4412
# If a revision has a delta, this is actually expanded inside the
4413
# insert_record_stream code now, which is an alternate fix for
4415
from_rf = self.from_repository.revisions
4416
revisions = from_rf.get_record_stream(
4418
self.to_format._fetch_order,
4419
not self.delta_on_metadata())
4420
return [('signatures', signatures), ('revisions', revisions)]
4422
def _generate_root_texts(self, revs):
4423
"""This will be called by get_stream between fetching weave texts and
4424
fetching the inventory weave.
4426
if self._rich_root_upgrade():
4427
return _mod_fetch.Inter1and2Helper(
4428
self.from_repository).generate_root_texts(revs)
4432
def get_stream(self, search):
4434
revs = search.get_keys()
4435
graph = self.from_repository.get_graph()
4436
revs = tsort.topo_sort(graph.get_parent_map(revs))
4437
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4439
for knit_kind, file_id, revisions in data_to_fetch:
4440
if knit_kind != phase:
4442
# Make a new progress bar for this phase
4443
if knit_kind == "file":
4444
# Accumulate file texts
4445
text_keys.extend([(file_id, revision) for revision in
4447
elif knit_kind == "inventory":
4448
# Now copy the file texts.
4449
from_texts = self.from_repository.texts
4450
yield ('texts', from_texts.get_record_stream(
4451
text_keys, self.to_format._fetch_order,
4452
not self.to_format._fetch_uses_deltas))
4453
# Cause an error if a text occurs after we have done the
4456
# Before we process the inventory we generate the root
4457
# texts (if necessary) so that the inventories references
4459
for _ in self._generate_root_texts(revs):
4461
# we fetch only the referenced inventories because we do not
4462
# know for unselected inventories whether all their required
4463
# texts are present in the other repository - it could be
4465
for info in self._get_inventory_stream(revs):
4467
elif knit_kind == "signatures":
4468
# Nothing to do here; this will be taken care of when
4469
# _fetch_revision_texts happens.
4471
elif knit_kind == "revisions":
4472
for record in self._fetch_revision_texts(revs):
4475
raise AssertionError("Unknown knit kind %r" % knit_kind)
4477
def get_stream_for_missing_keys(self, missing_keys):
4478
# missing keys can only occur when we are byte copying and not
4479
# translating (because translation means we don't send
4480
# unreconstructable deltas ever).
4482
keys['texts'] = set()
4483
keys['revisions'] = set()
4484
keys['inventories'] = set()
4485
keys['chk_bytes'] = set()
4486
keys['signatures'] = set()
4487
for key in missing_keys:
4488
keys[key[0]].add(key[1:])
4489
if len(keys['revisions']):
4490
# If we allowed copying revisions at this point, we could end up
4491
# copying a revision without copying its required texts: a
4492
# violation of the requirements for repository integrity.
4493
raise AssertionError(
4494
'cannot copy revisions to fill in missing deltas %s' % (
4495
keys['revisions'],))
4496
for substream_kind, keys in keys.iteritems():
4497
vf = getattr(self.from_repository, substream_kind)
4498
if vf is None and keys:
4499
raise AssertionError(
4500
"cannot fill in keys for a versioned file we don't"
4501
" have: %s needs %s" % (substream_kind, keys))
4503
# No need to stream something we don't have
4505
if substream_kind == 'inventories':
4506
# Some missing keys are genuinely ghosts, filter those out.
4507
present = self.from_repository.inventories.get_parent_map(keys)
4508
revs = [key[0] for key in present]
4509
# Get the inventory stream more-or-less as we do for the
4510
# original stream; there's no reason to assume that records
4511
# direct from the source will be suitable for the sink. (Think
4512
# e.g. 2a -> 1.9-rich-root).
4513
for info in self._get_inventory_stream(revs, missing=True):
4517
# Ask for full texts always so that we don't need more round trips
4518
# after this stream.
4519
# Some of the missing keys are genuinely ghosts, so filter absent
4520
# records. The Sink is responsible for doing another check to
4521
# ensure that ghosts don't introduce missing data for future
4523
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4524
self.to_format._fetch_order, True))
4525
yield substream_kind, stream
4527
def inventory_fetch_order(self):
4528
if self._rich_root_upgrade():
4529
return 'topological'
4531
return self.to_format._fetch_order
4533
def _rich_root_upgrade(self):
4534
return (not self.from_repository._format.rich_root_data and
4535
self.to_format.rich_root_data)
4537
def _get_inventory_stream(self, revision_ids, missing=False):
4538
from_format = self.from_repository._format
4539
if (from_format.supports_chks and self.to_format.supports_chks and
4540
from_format.network_name() == self.to_format.network_name()):
4541
raise AssertionError(
4542
"this case should be handled by GroupCHKStreamSource")
4543
elif 'forceinvdeltas' in debug.debug_flags:
4544
return self._get_convertable_inventory_stream(revision_ids,
4545
delta_versus_null=missing)
4546
elif from_format.network_name() == self.to_format.network_name():
4548
return self._get_simple_inventory_stream(revision_ids,
4550
elif (not from_format.supports_chks and not self.to_format.supports_chks
4551
and from_format._serializer == self.to_format._serializer):
4552
# Essentially the same format.
4553
return self._get_simple_inventory_stream(revision_ids,
4556
# Any time we switch serializations, we want to use an
4557
# inventory-delta based approach.
4558
return self._get_convertable_inventory_stream(revision_ids,
4559
delta_versus_null=missing)
4561
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4562
# NB: This currently reopens the inventory weave in source;
4563
# using a single stream interface instead would avoid this.
4564
from_weave = self.from_repository.inventories
4566
delta_closure = True
4568
delta_closure = not self.delta_on_metadata()
4569
yield ('inventories', from_weave.get_record_stream(
4570
[(rev_id,) for rev_id in revision_ids],
4571
self.inventory_fetch_order(), delta_closure))
4573
def _get_convertable_inventory_stream(self, revision_ids,
4574
delta_versus_null=False):
4575
# The two formats are sufficiently different that there is no fast
4576
# path, so we need to send just inventorydeltas, which any
4577
# sufficiently modern client can insert into any repository.
4578
# The StreamSink code expects to be able to
4579
# convert on the target, so we need to put bytes-on-the-wire that can
4580
# be converted. That means inventory deltas (if the remote is <1.19,
4581
# RemoteStreamSink will fallback to VFS to insert the deltas).
4582
yield ('inventory-deltas',
4583
self._stream_invs_as_deltas(revision_ids,
4584
delta_versus_null=delta_versus_null))
4586
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4587
"""Return a stream of inventory-deltas for the given rev ids.
4589
:param revision_ids: The list of inventories to transmit
4590
:param delta_versus_null: Don't try to find a minimal delta for this
4591
entry, instead compute the delta versus the NULL_REVISION. This
4592
effectively streams a complete inventory. Used for stuff like
4593
filling in missing parents, etc.
4595
from_repo = self.from_repository
4596
revision_keys = [(rev_id,) for rev_id in revision_ids]
4597
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4598
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4600
inventories = self.from_repository.iter_inventories(
4601
revision_ids, 'topological')
4602
format = from_repo._format
4603
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4604
inventory_cache = lru_cache.LRUCache(50)
4605
null_inventory = from_repo.revision_tree(
4606
_mod_revision.NULL_REVISION).inventory
4607
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4608
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4609
# repo back into a non-rich-root repo ought to be allowed)
4610
serializer = inventory_delta.InventoryDeltaSerializer(
4611
versioned_root=format.rich_root_data,
4612
tree_references=format.supports_tree_reference)
4613
for inv in inventories:
4614
key = (inv.revision_id,)
4615
parent_keys = parent_map.get(key, ())
4617
if not delta_versus_null and parent_keys:
4618
# The caller did not ask for complete inventories and we have
4619
# some parents that we can delta against. Make a delta against
4620
# each parent so that we can find the smallest.
4621
parent_ids = [parent_key[0] for parent_key in parent_keys]
4622
for parent_id in parent_ids:
4623
if parent_id not in invs_sent_so_far:
4624
# We don't know that the remote side has this basis, so
4627
if parent_id == _mod_revision.NULL_REVISION:
4628
parent_inv = null_inventory
4630
parent_inv = inventory_cache.get(parent_id, None)
4631
if parent_inv is None:
4632
parent_inv = from_repo.get_inventory(parent_id)
4633
candidate_delta = inv._make_delta(parent_inv)
4634
if (delta is None or
4635
len(delta) > len(candidate_delta)):
4636
delta = candidate_delta
4637
basis_id = parent_id
4639
# Either none of the parents ended up being suitable, or we
4640
# were asked to delta against NULL
4641
basis_id = _mod_revision.NULL_REVISION
4642
delta = inv._make_delta(null_inventory)
4643
invs_sent_so_far.add(inv.revision_id)
4644
inventory_cache[inv.revision_id] = inv
4645
delta_serialized = ''.join(
4646
serializer.delta_to_lines(basis_id, key[-1], delta))
4647
yield versionedfile.FulltextContentFactory(
4648
key, parent_keys, None, delta_serialized)
4651
1771
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4652
1772
stop_revision=None):
4653
1773
"""Extend the partial history to include a given index
added
removed
bzrlib/repository.py
