341
202
commit to be valid, deletes against the basis MUST be recorded via
342
203
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):
205
raise NotImplementedError(self.will_record_deletes)
207
def record_iter_changes(self, tree, basis_revision_id, iter_changes):
562
208
"""Record a new tree via iter_changes.
564
210
:param tree: The tree to obtain text contents from for changed objects.
569
215
to basis_revision_id. The iterator must not include any items with
570
216
a current kind of None - missing items must be either filtered out
571
217
or errored-on beefore record_iter_changes sees the item.
572
:param _entry_factory: Private method to bind entry_factory locally for
574
218
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
575
219
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.
221
raise NotImplementedError(self.record_iter_changes)
224
class RepositoryWriteLockResult(LogicalLockResult):
225
"""The result of write locking a repository.
227
:ivar repository_token: The token obtained from the underlying lock, or
229
:ivar unlock: A callable which will unlock the lock.
232
def __init__(self, unlock, repository_token):
233
LogicalLockResult.__init__(self, unlock)
234
self.repository_token = repository_token
237
return "RepositoryWriteLockResult(%s, %s)" % (self.repository_token,
862
241
######################################################################
866
class Repository(_RelockDebugMixin):
245
class Repository(_RelockDebugMixin, controldir.ControlComponent):
867
246
"""Repository holding history for one or more branches.
869
248
The repository holds and retrieves historical information including
870
249
revisions and file history. It's normally accessed only by the Branch,
871
250
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.
252
See VersionedFileRepository in bzrlib.vf_repository for the
253
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
256
def abort_write_group(self, suppress_errors=False):
931
257
"""Commit the contents accrued within the current write group.
1000
313
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._serialise_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
315
def all_revision_ids(self):
1131
316
"""Returns a list of all the revision ids in the repository.
1156
341
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
344
def create(a_bzrdir):
1285
345
"""Construct the current default format repository in a_bzrdir."""
1971
909
signature = gpg_strategy.sign(plaintext)
1972
910
self.add_signature_text(revision_id, signature)
1975
912
def add_signature_text(self, revision_id, signature):
1976
self.signatures.add_lines((revision_id,), (),
1977
osutils.split_lines(signature))
1979
def find_text_key_references(self):
1980
"""Find the text key references within the repository.
1982
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1983
to whether they were referred to by the inventory of the
1984
revision_id that they contain. The inventory texts from all present
1985
revision ids are assessed to generate this report.
1987
revision_keys = self.revisions.keys()
1988
w = self.inventories
1989
pb = ui.ui_factory.nested_progress_bar()
1991
return self._find_text_key_references_from_xml_inventory_lines(
1992
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1996
def _find_text_key_references_from_xml_inventory_lines(self,
1998
"""Core routine for extracting references to texts from inventories.
2000
This performs the translation of xml lines to revision ids.
2002
:param line_iterator: An iterator of lines, origin_version_id
2003
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2004
to whether they were referred to by the inventory of the
2005
revision_id that they contain. Note that if that revision_id was
2006
not part of the line_iterator's output then False will be given -
2007
even though it may actually refer to that key.
2009
if not self._serializer.support_altered_by_hack:
2010
raise AssertionError(
2011
"_find_text_key_references_from_xml_inventory_lines only "
2012
"supported for branches which store inventory as unnested xml"
2013
", not on %r" % self)
2016
# this code needs to read every new line in every inventory for the
2017
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2018
# not present in one of those inventories is unnecessary but not
2019
# harmful because we are filtering by the revision id marker in the
2020
# inventory lines : we only select file ids altered in one of those
2021
# revisions. We don't need to see all lines in the inventory because
2022
# only those added in an inventory in rev X can contain a revision=X
2024
unescape_revid_cache = {}
2025
unescape_fileid_cache = {}
2027
# jam 20061218 In a big fetch, this handles hundreds of thousands
2028
# of lines, so it has had a lot of inlining and optimizing done.
2029
# Sorry that it is a little bit messy.
2030
# Move several functions to be local variables, since this is a long
2032
search = self._file_ids_altered_regex.search
2033
unescape = _unescape_xml
2034
setdefault = result.setdefault
2035
for line, line_key in line_iterator:
2036
match = search(line)
2039
# One call to match.group() returning multiple items is quite a
2040
# bit faster than 2 calls to match.group() each returning 1
2041
file_id, revision_id = match.group('file_id', 'revision_id')
2043
# Inlining the cache lookups helps a lot when you make 170,000
2044
# lines and 350k ids, versus 8.4 unique ids.
2045
# Using a cache helps in 2 ways:
2046
# 1) Avoids unnecessary decoding calls
2047
# 2) Re-uses cached strings, which helps in future set and
2049
# (2) is enough that removing encoding entirely along with
2050
# the cache (so we are using plain strings) results in no
2051
# performance improvement.
2053
revision_id = unescape_revid_cache[revision_id]
2055
unescaped = unescape(revision_id)
2056
unescape_revid_cache[revision_id] = unescaped
2057
revision_id = unescaped
2059
# Note that unconditionally unescaping means that we deserialise
2060
# every fileid, which for general 'pull' is not great, but we don't
2061
# really want to have some many fulltexts that this matters anyway.
2064
file_id = unescape_fileid_cache[file_id]
2066
unescaped = unescape(file_id)
2067
unescape_fileid_cache[file_id] = unescaped
2070
key = (file_id, revision_id)
2071
setdefault(key, False)
2072
if revision_id == line_key[-1]:
2076
def _inventory_xml_lines_for_keys(self, keys):
2077
"""Get a line iterator of the sort needed for findind references.
2079
Not relevant for non-xml inventory repositories.
2081
Ghosts in revision_keys are ignored.
2083
:param revision_keys: The revision keys for the inventories to inspect.
2084
:return: An iterator over (inventory line, revid) for the fulltexts of
2085
all of the xml inventories specified by revision_keys.
2087
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2088
for record in stream:
2089
if record.storage_kind != 'absent':
2090
chunks = record.get_bytes_as('chunked')
2091
revid = record.key[-1]
2092
lines = osutils.chunks_to_lines(chunks)
2096
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2098
"""Helper routine for fileids_altered_by_revision_ids.
2100
This performs the translation of xml lines to revision ids.
2102
:param line_iterator: An iterator of lines, origin_version_id
2103
:param revision_keys: The revision ids to filter for. This should be a
2104
set or other type which supports efficient __contains__ lookups, as
2105
the revision key from each parsed line will be looked up in the
2106
revision_keys filter.
2107
:return: a dictionary mapping altered file-ids to an iterable of
2108
revision_ids. Each altered file-ids has the exact revision_ids that
2109
altered it listed explicitly.
2111
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2112
line_iterator).iterkeys())
2113
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2114
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2115
self._inventory_xml_lines_for_keys(parent_keys)))
2116
new_keys = seen - parent_seen
2118
setdefault = result.setdefault
2119
for key in new_keys:
2120
setdefault(key[0], set()).add(key[-1])
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)
2123
920
def _find_parent_ids_of_revisions(self, revision_ids):
2124
921
"""Find all parent ids that are mentioned in the revision graph.
2175
951
uniquely identify the file version in the caller's context. (Examples:
2176
952
an index number or a TreeTransform trans_id.)
2178
bytes_iterator is an iterable of bytestrings for the file. The
2179
kind of iterable and length of the bytestrings are unspecified, but for
2180
this implementation, it is a list of bytes produced by
2181
VersionedFile.get_record_stream().
2183
954
:param desired_files: a list of (file_id, revision_id, identifier)
2187
for file_id, revision_id, callable_data in desired_files:
2188
text_keys[(file_id, revision_id)] = callable_data
2189
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2190
if record.storage_kind == 'absent':
2191
raise errors.RevisionNotPresent(record.key, self)
2192
yield text_keys[record.key], record.get_bytes_as('chunked')
2194
def _generate_text_key_index(self, text_key_references=None,
2196
"""Generate a new text key index for the repository.
2198
This is an expensive function that will take considerable time to run.
2200
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2201
list of parents, also text keys. When a given key has no parents,
2202
the parents list will be [NULL_REVISION].
2204
# All revisions, to find inventory parents.
2205
if ancestors is None:
2206
graph = self.get_graph()
2207
ancestors = graph.get_parent_map(self.all_revision_ids())
2208
if text_key_references is None:
2209
text_key_references = self.find_text_key_references()
2210
pb = ui.ui_factory.nested_progress_bar()
2212
return self._do_generate_text_key_index(ancestors,
2213
text_key_references, pb)
2217
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2218
"""Helper for _generate_text_key_index to avoid deep nesting."""
2219
revision_order = tsort.topo_sort(ancestors)
2220
invalid_keys = set()
2222
for revision_id in revision_order:
2223
revision_keys[revision_id] = set()
2224
text_count = len(text_key_references)
2225
# a cache of the text keys to allow reuse; costs a dict of all the
2226
# keys, but saves a 2-tuple for every child of a given key.
2228
for text_key, valid in text_key_references.iteritems():
2230
invalid_keys.add(text_key)
2232
revision_keys[text_key[1]].add(text_key)
2233
text_key_cache[text_key] = text_key
2234
del text_key_references
2236
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2237
NULL_REVISION = _mod_revision.NULL_REVISION
2238
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2239
# too small for large or very branchy trees. However, for 55K path
2240
# trees, it would be easy to use too much memory trivially. Ideally we
2241
# could gauge this by looking at available real memory etc, but this is
2242
# always a tricky proposition.
2243
inventory_cache = lru_cache.LRUCache(10)
2244
batch_size = 10 # should be ~150MB on a 55K path tree
2245
batch_count = len(revision_order) / batch_size + 1
2247
pb.update("Calculating text parents", processed_texts, text_count)
2248
for offset in xrange(batch_count):
2249
to_query = revision_order[offset * batch_size:(offset + 1) *
2253
for revision_id in to_query:
2254
parent_ids = ancestors[revision_id]
2255
for text_key in revision_keys[revision_id]:
2256
pb.update("Calculating text parents", processed_texts)
2257
processed_texts += 1
2258
candidate_parents = []
2259
for parent_id in parent_ids:
2260
parent_text_key = (text_key[0], parent_id)
2262
check_parent = parent_text_key not in \
2263
revision_keys[parent_id]
2265
# the parent parent_id is a ghost:
2266
check_parent = False
2267
# truncate the derived graph against this ghost.
2268
parent_text_key = None
2270
# look at the parent commit details inventories to
2271
# determine possible candidates in the per file graph.
2274
inv = inventory_cache[parent_id]
2276
inv = self.revision_tree(parent_id).inventory
2277
inventory_cache[parent_id] = inv
2279
parent_entry = inv[text_key[0]]
2280
except (KeyError, errors.NoSuchId):
2282
if parent_entry is not None:
2284
text_key[0], parent_entry.revision)
2286
parent_text_key = None
2287
if parent_text_key is not None:
2288
candidate_parents.append(
2289
text_key_cache[parent_text_key])
2290
parent_heads = text_graph.heads(candidate_parents)
2291
new_parents = list(parent_heads)
2292
new_parents.sort(key=lambda x:candidate_parents.index(x))
2293
if new_parents == []:
2294
new_parents = [NULL_REVISION]
2295
text_index[text_key] = new_parents
2297
for text_key in invalid_keys:
2298
text_index[text_key] = [NULL_REVISION]
2301
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2302
"""Get an iterable listing the keys of all the data introduced by a set
2305
The keys will be ordered so that the corresponding items can be safely
2306
fetched and inserted in that order.
2308
:returns: An iterable producing tuples of (knit-kind, file-id,
2309
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2310
'revisions'. file-id is None unless knit-kind is 'file'.
2312
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2315
for result in self._find_non_file_keys_to_fetch(revision_ids):
2318
def _find_file_keys_to_fetch(self, revision_ids, pb):
2319
# XXX: it's a bit weird to control the inventory weave caching in this
2320
# generator. Ideally the caching would be done in fetch.py I think. Or
2321
# maybe this generator should explicitly have the contract that it
2322
# should not be iterated until the previously yielded item has been
2324
inv_w = self.inventories
2326
# file ids that changed
2327
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2329
num_file_ids = len(file_ids)
2330
for file_id, altered_versions in file_ids.iteritems():
2332
pb.update("Fetch texts", count, num_file_ids)
2334
yield ("file", file_id, altered_versions)
2336
def _find_non_file_keys_to_fetch(self, revision_ids):
2338
yield ("inventory", None, revision_ids)
2341
# XXX: Note ATM no callers actually pay attention to this return
2342
# instead they just use the list of revision ids and ignore
2343
# missing sigs. Consider removing this work entirely
2344
revisions_with_signatures = set(self.signatures.get_parent_map(
2345
[(r,) for r in revision_ids]))
2346
revisions_with_signatures = set(
2347
[r for (r,) in revisions_with_signatures])
2348
revisions_with_signatures.intersection_update(revision_ids)
2349
yield ("signatures", None, revisions_with_signatures)
2352
yield ("revisions", None, revision_ids)
2355
def get_inventory(self, revision_id):
2356
"""Get Inventory object by revision id."""
2357
return self.iter_inventories([revision_id]).next()
2359
def iter_inventories(self, revision_ids, ordering=None):
2360
"""Get many inventories by revision_ids.
2362
This will buffer some or all of the texts used in constructing the
2363
inventories in memory, but will only parse a single inventory at a
2366
:param revision_ids: The expected revision ids of the inventories.
2367
:param ordering: optional ordering, e.g. 'topological'. If not
2368
specified, the order of revision_ids will be preserved (by
2369
buffering if necessary).
2370
:return: An iterator of inventories.
2372
if ((None in revision_ids)
2373
or (_mod_revision.NULL_REVISION in revision_ids)):
2374
raise ValueError('cannot get null revision inventory')
2375
return self._iter_inventories(revision_ids, ordering)
2377
def _iter_inventories(self, revision_ids, ordering):
2378
"""single-document based inventory iteration."""
2379
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2380
for text, revision_id in inv_xmls:
2381
yield self._deserialise_inventory(revision_id, text)
2383
def _iter_inventory_xmls(self, revision_ids, ordering):
2384
if ordering is None:
2385
order_as_requested = True
2386
ordering = 'unordered'
2388
order_as_requested = False
2389
keys = [(revision_id,) for revision_id in revision_ids]
2392
if order_as_requested:
2393
key_iter = iter(keys)
2394
next_key = key_iter.next()
2395
stream = self.inventories.get_record_stream(keys, ordering, True)
2397
for record in stream:
2398
if record.storage_kind != 'absent':
2399
chunks = record.get_bytes_as('chunked')
2400
if order_as_requested:
2401
text_chunks[record.key] = chunks
2403
yield ''.join(chunks), record.key[-1]
2405
raise errors.NoSuchRevision(self, record.key)
2406
if order_as_requested:
2407
# Yield as many results as we can while preserving order.
2408
while next_key in text_chunks:
2409
chunks = text_chunks.pop(next_key)
2410
yield ''.join(chunks), next_key[-1]
2412
next_key = key_iter.next()
2413
except StopIteration:
2414
# We still want to fully consume the get_record_stream,
2415
# just in case it is not actually finished at this point
2419
def _deserialise_inventory(self, revision_id, xml):
2420
"""Transform the xml into an inventory object.
2422
:param revision_id: The expected revision id of the inventory.
2423
:param xml: A serialised inventory.
2425
result = self._serializer.read_inventory_from_string(xml, revision_id,
2426
entry_cache=self._inventory_entry_cache,
2427
return_from_cache=self._safe_to_return_from_cache)
2428
if result.revision_id != revision_id:
2429
raise AssertionError('revision id mismatch %s != %s' % (
2430
result.revision_id, revision_id))
2433
def _serialise_inventory(self, inv):
2434
return self._serializer.write_inventory_to_string(inv)
2436
def _serialise_inventory_to_lines(self, inv):
2437
return self._serializer.write_inventory_to_lines(inv)
2439
def get_serializer_format(self):
2440
return self._serializer.format_num
2443
def _get_inventory_xml(self, revision_id):
2444
"""Get serialized inventory as a string."""
2445
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2447
text, revision_id = texts.next()
2448
except StopIteration:
2449
raise errors.HistoryMissing(self, 'inventory', revision_id)
957
raise NotImplementedError(self.iter_files_bytes)
2452
959
def get_rev_id_for_revno(self, revno, known_pair):
2453
960
"""Return the revision id of a revno, given a later (revno, revid)
2801
1277
except UnicodeDecodeError:
2802
1278
raise errors.NonAsciiRevisionId(method, self)
2804
def revision_graph_can_have_wrong_parents(self):
2805
"""Is it possible for this repository to have a revision graph with
2808
If True, then this repository must also implement
2809
_find_inconsistent_revision_parents so that check and reconcile can
2810
check for inconsistencies before proceeding with other checks that may
2811
depend on the revision index being consistent.
2813
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2816
# remove these delegates a while after bzr 0.15
2817
def __make_delegated(name, from_module):
2818
def _deprecated_repository_forwarder():
2819
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2820
% (name, from_module),
2823
m = __import__(from_module, globals(), locals(), [name])
2825
return getattr(m, name)
2826
except AttributeError:
2827
raise AttributeError('module %s has no name %s'
2829
globals()[name] = _deprecated_repository_forwarder
2832
'AllInOneRepository',
2833
'WeaveMetaDirRepository',
2834
'PreSplitOutRepositoryFormat',
2835
'RepositoryFormat4',
2836
'RepositoryFormat5',
2837
'RepositoryFormat6',
2838
'RepositoryFormat7',
2840
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2844
'RepositoryFormatKnit',
2845
'RepositoryFormatKnit1',
2847
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2850
def install_revision(repository, rev, revision_tree):
2851
"""Install all revision data into a repository."""
2852
install_revisions(repository, [(rev, revision_tree, None)])
2855
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2856
"""Install all revision data into a repository.
2858
Accepts an iterable of revision, tree, signature tuples. The signature
2861
repository.start_write_group()
2863
inventory_cache = lru_cache.LRUCache(10)
2864
for n, (revision, revision_tree, signature) in enumerate(iterable):
2865
_install_revision(repository, revision, revision_tree, signature,
2868
pb.update('Transferring revisions', n + 1, num_revisions)
2870
repository.abort_write_group()
2873
repository.commit_write_group()
2876
def _install_revision(repository, rev, revision_tree, signature,
2878
"""Install all revision data into a repository."""
2879
present_parents = []
2881
for p_id in rev.parent_ids:
2882
if repository.has_revision(p_id):
2883
present_parents.append(p_id)
2884
parent_trees[p_id] = repository.revision_tree(p_id)
2886
parent_trees[p_id] = repository.revision_tree(
2887
_mod_revision.NULL_REVISION)
2889
inv = revision_tree.inventory
2890
entries = inv.iter_entries()
2891
# backwards compatibility hack: skip the root id.
2892
if not repository.supports_rich_root():
2893
path, root = entries.next()
2894
if root.revision != rev.revision_id:
2895
raise errors.IncompatibleRevision(repr(repository))
2897
for path, ie in entries:
2898
text_keys[(ie.file_id, ie.revision)] = ie
2899
text_parent_map = repository.texts.get_parent_map(text_keys)
2900
missing_texts = set(text_keys) - set(text_parent_map)
2901
# Add the texts that are not already present
2902
for text_key in missing_texts:
2903
ie = text_keys[text_key]
2905
# FIXME: TODO: The following loop overlaps/duplicates that done by
2906
# commit to determine parents. There is a latent/real bug here where
2907
# the parents inserted are not those commit would do - in particular
2908
# they are not filtered by heads(). RBC, AB
2909
for revision, tree in parent_trees.iteritems():
2910
if ie.file_id not in tree:
2912
parent_id = tree.inventory[ie.file_id].revision
2913
if parent_id in text_parents:
2915
text_parents.append((ie.file_id, parent_id))
2916
lines = revision_tree.get_file(ie.file_id).readlines()
2917
repository.texts.add_lines(text_key, text_parents, lines)
2919
# install the inventory
2920
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2921
# Cache this inventory
2922
inventory_cache[rev.revision_id] = inv
2924
basis_inv = inventory_cache[rev.parent_ids[0]]
2926
repository.add_inventory(rev.revision_id, inv, present_parents)
2928
delta = inv._make_delta(basis_inv)
2929
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2930
rev.revision_id, present_parents)
2932
repository.add_inventory(rev.revision_id, inv, present_parents)
2933
except errors.RevisionAlreadyPresent:
2935
if signature is not None:
2936
repository.add_signature_text(rev.revision_id, signature)
2937
repository.add_revision(rev.revision_id, rev, inv)
2940
1281
class MetaDirRepository(Repository):
2941
1282
"""Repositories in the new meta-dir layout.
3297
1621
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
3298
1622
format_registry.register_lazy(
3299
1623
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
3300
'bzrlib.repofmt.pack_repo',
1624
'bzrlib.repofmt.knitpack_repo',
3301
1625
'RepositoryFormatKnitPack1',
3303
1627
format_registry.register_lazy(
3304
1628
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
3305
'bzrlib.repofmt.pack_repo',
1629
'bzrlib.repofmt.knitpack_repo',
3306
1630
'RepositoryFormatKnitPack3',
3308
1632
format_registry.register_lazy(
3309
1633
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
3310
'bzrlib.repofmt.pack_repo',
1634
'bzrlib.repofmt.knitpack_repo',
3311
1635
'RepositoryFormatKnitPack4',
3313
1637
format_registry.register_lazy(
3314
1638
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
3315
'bzrlib.repofmt.pack_repo',
1639
'bzrlib.repofmt.knitpack_repo',
3316
1640
'RepositoryFormatKnitPack5',
3318
1642
format_registry.register_lazy(
3319
1643
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
3320
'bzrlib.repofmt.pack_repo',
1644
'bzrlib.repofmt.knitpack_repo',
3321
1645
'RepositoryFormatKnitPack5RichRoot',
3323
1647
format_registry.register_lazy(
3324
1648
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
3325
'bzrlib.repofmt.pack_repo',
1649
'bzrlib.repofmt.knitpack_repo',
3326
1650
'RepositoryFormatKnitPack5RichRootBroken',
3328
1652
format_registry.register_lazy(
3329
1653
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
3330
'bzrlib.repofmt.pack_repo',
1654
'bzrlib.repofmt.knitpack_repo',
3331
1655
'RepositoryFormatKnitPack6',
3333
1657
format_registry.register_lazy(
3334
1658
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
3335
'bzrlib.repofmt.pack_repo',
1659
'bzrlib.repofmt.knitpack_repo',
3336
1660
'RepositoryFormatKnitPack6RichRoot',
1662
format_registry.register_lazy(
1663
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
1664
'bzrlib.repofmt.groupcompress_repo',
1665
'RepositoryFormat2a',
3339
1668
# Development formats.
3340
# Obsolete but kept pending a CHK based subtree format.
1669
# Check their docstrings to see if/when they are obsolete.
3341
1670
format_registry.register_lazy(
3342
1671
("Bazaar development format 2 with subtree support "
3343
1672
"(needs bzr.dev from before 1.8)\n"),
3344
'bzrlib.repofmt.pack_repo',
1673
'bzrlib.repofmt.knitpack_repo',
3345
1674
'RepositoryFormatPackDevelopment2Subtree',
3348
# 1.14->1.16 go below here
3349
format_registry.register_lazy(
3350
'Bazaar development format - group compression and chk inventory'
3351
' (needs bzr.dev from 1.14)\n',
3352
'bzrlib.repofmt.groupcompress_repo',
3353
'RepositoryFormatCHK1',
3356
format_registry.register_lazy(
3357
'Bazaar development format - chk repository with bencode revision '
3358
'serialization (needs bzr.dev from 1.16)\n',
3359
'bzrlib.repofmt.groupcompress_repo',
3360
'RepositoryFormatCHK2',
3362
format_registry.register_lazy(
3363
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3364
'bzrlib.repofmt.groupcompress_repo',
3365
'RepositoryFormat2a',
1676
format_registry.register_lazy(
1677
'Bazaar development format 8\n',
1678
'bzrlib.repofmt.groupcompress_repo',
1679
'RepositoryFormat2aSubtree',
3408
1721
:param revision_id: if None all content is copied, if NULL_REVISION no
3409
1722
content is copied.
3413
ui.ui_factory.warn_experimental_format_fetch(self)
3414
f = _mod_fetch.RepoFetcher(to_repository=self.target,
3415
from_repository=self.source,
3416
last_revision=revision_id,
3417
fetch_spec=fetch_spec,
3418
find_ghosts=find_ghosts)
3420
def _walk_to_common_revisions(self, revision_ids):
3421
"""Walk out from revision_ids in source to revisions target has.
3423
:param revision_ids: The start point for the search.
3424
:return: A set of revision ids.
3426
target_graph = self.target.get_graph()
3427
revision_ids = frozenset(revision_ids)
3428
missing_revs = set()
3429
source_graph = self.source.get_graph()
3430
# ensure we don't pay silly lookup costs.
3431
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3432
null_set = frozenset([_mod_revision.NULL_REVISION])
3433
searcher_exhausted = False
3437
# Iterate the searcher until we have enough next_revs
3438
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3440
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3441
next_revs.update(next_revs_part)
3442
ghosts.update(ghosts_part)
3443
except StopIteration:
3444
searcher_exhausted = True
3446
# If there are ghosts in the source graph, and the caller asked for
3447
# them, make sure that they are present in the target.
3448
# We don't care about other ghosts as we can't fetch them and
3449
# haven't been asked to.
3450
ghosts_to_check = set(revision_ids.intersection(ghosts))
3451
revs_to_get = set(next_revs).union(ghosts_to_check)
3453
have_revs = set(target_graph.get_parent_map(revs_to_get))
3454
# we always have NULL_REVISION present.
3455
have_revs = have_revs.union(null_set)
3456
# Check if the target is missing any ghosts we need.
3457
ghosts_to_check.difference_update(have_revs)
3459
# One of the caller's revision_ids is a ghost in both the
3460
# source and the target.
3461
raise errors.NoSuchRevision(
3462
self.source, ghosts_to_check.pop())
3463
missing_revs.update(next_revs - have_revs)
3464
# Because we may have walked past the original stop point, make
3465
# sure everything is stopped
3466
stop_revs = searcher.find_seen_ancestors(have_revs)
3467
searcher.stop_searching_any(stop_revs)
3468
if searcher_exhausted:
3470
return searcher.get_result()
1725
raise NotImplementedError(self.fetch)
3472
1727
@needs_read_lock
3473
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1728
def search_missing_revision_ids(self,
1729
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
1730
find_ghosts=True, revision_ids=None, if_present_ids=None,
3474
1732
"""Return the revision ids that source has that target does not.
3476
1734
:param revision_id: only return revision ids included by this
1736
:param revision_ids: return revision ids included by these
1737
revision_ids. NoSuchRevision will be raised if any of these
1738
revisions are not present.
1739
:param if_present_ids: like revision_ids, but will not cause
1740
NoSuchRevision if any of these are absent, instead they will simply
1741
not be in the result. This is useful for e.g. finding revisions
1742
to fetch for tags, which may reference absent revisions.
3478
1743
:param find_ghosts: If True find missing revisions in deep history
3479
1744
rather than just finding the surface difference.
1745
:param limit: Maximum number of revisions to return, topologically
3480
1747
:return: A bzrlib.graph.SearchResult.
3482
# stop searching at found target revisions.
3483
if not find_ghosts and revision_id is not None:
3484
return self._walk_to_common_revisions([revision_id])
3485
# generic, possibly worst case, slow code path.
3486
target_ids = set(self.target.all_revision_ids())
3487
if revision_id is not None:
3488
source_ids = self.source.get_ancestry(revision_id)
3489
if source_ids[0] is not None:
3490
raise AssertionError()
3493
source_ids = self.source.all_revision_ids()
3494
result_set = set(source_ids).difference(target_ids)
3495
return self.source.revision_ids_to_search_result(result_set)
1749
raise NotImplementedError(self.search_missing_revision_ids)
3498
1752
def _same_model(source, target):
3519
1773
"different serializers")
3522
class InterSameDataRepository(InterRepository):
3523
"""Code for converting between repositories that represent the same data.
3525
Data format and model must match for this to work.
3529
def _get_repo_format_to_test(self):
3530
"""Repository format for testing with.
3532
InterSameData can pull from subtree to subtree and from non-subtree to
3533
non-subtree, so we test this with the richest repository format.
3535
from bzrlib.repofmt import knitrepo
3536
return knitrepo.RepositoryFormatKnit3()
3539
def is_compatible(source, target):
3540
return InterRepository._same_model(source, target)
3543
class InterWeaveRepo(InterSameDataRepository):
3544
"""Optimised code paths between Weave based repositories.
3546
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3547
implemented lazy inter-object optimisation.
3551
def _get_repo_format_to_test(self):
3552
from bzrlib.repofmt import weaverepo
3553
return weaverepo.RepositoryFormat7()
3556
def is_compatible(source, target):
3557
"""Be compatible with known Weave formats.
3559
We don't test for the stores being of specific types because that
3560
could lead to confusing results, and there is no need to be
3563
from bzrlib.repofmt.weaverepo import (
3569
return (isinstance(source._format, (RepositoryFormat5,
3571
RepositoryFormat7)) and
3572
isinstance(target._format, (RepositoryFormat5,
3574
RepositoryFormat7)))
3575
except AttributeError:
3579
def copy_content(self, revision_id=None):
3580
"""See InterRepository.copy_content()."""
3581
# weave specific optimised path:
3583
self.target.set_make_working_trees(self.source.make_working_trees())
3584
except (errors.RepositoryUpgradeRequired, NotImplemented):
3586
# FIXME do not peek!
3587
if self.source._transport.listable():
3588
pb = ui.ui_factory.nested_progress_bar()
3590
self.target.texts.insert_record_stream(
3591
self.source.texts.get_record_stream(
3592
self.source.texts.keys(), 'topological', False))
3593
pb.update('Copying inventory', 0, 1)
3594
self.target.inventories.insert_record_stream(
3595
self.source.inventories.get_record_stream(
3596
self.source.inventories.keys(), 'topological', False))
3597
self.target.signatures.insert_record_stream(
3598
self.source.signatures.get_record_stream(
3599
self.source.signatures.keys(),
3601
self.target.revisions.insert_record_stream(
3602
self.source.revisions.get_record_stream(
3603
self.source.revisions.keys(),
3604
'topological', True))
3608
self.target.fetch(self.source, revision_id=revision_id)
3611
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3612
"""See InterRepository.missing_revision_ids()."""
3613
# we want all revisions to satisfy revision_id in source.
3614
# but we don't want to stat every file here and there.
3615
# we want then, all revisions other needs to satisfy revision_id
3616
# checked, but not those that we have locally.
3617
# so the first thing is to get a subset of the revisions to
3618
# satisfy revision_id in source, and then eliminate those that
3619
# we do already have.
3620
# this is slow on high latency connection to self, but as this
3621
# disk format scales terribly for push anyway due to rewriting
3622
# inventory.weave, this is considered acceptable.
3624
if revision_id is not None:
3625
source_ids = self.source.get_ancestry(revision_id)
3626
if source_ids[0] is not None:
3627
raise AssertionError()
3630
source_ids = self.source._all_possible_ids()
3631
source_ids_set = set(source_ids)
3632
# source_ids is the worst possible case we may need to pull.
3633
# now we want to filter source_ids against what we actually
3634
# have in target, but don't try to check for existence where we know
3635
# we do not have a revision as that would be pointless.
3636
target_ids = set(self.target._all_possible_ids())
3637
possibly_present_revisions = target_ids.intersection(source_ids_set)
3638
actually_present_revisions = set(
3639
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3640
required_revisions = source_ids_set.difference(actually_present_revisions)
3641
if revision_id is not None:
3642
# we used get_ancestry to determine source_ids then we are assured all
3643
# revisions referenced are present as they are installed in topological order.
3644
# and the tip revision was validated by get_ancestry.
3645
result_set = required_revisions
3647
# if we just grabbed the possibly available ids, then
3648
# we only have an estimate of whats available and need to validate
3649
# that against the revision records.
3651
self.source._eliminate_revisions_not_present(required_revisions))
3652
return self.source.revision_ids_to_search_result(result_set)
3655
class InterKnitRepo(InterSameDataRepository):
3656
"""Optimised code paths between Knit based repositories."""
3659
def _get_repo_format_to_test(self):
3660
from bzrlib.repofmt import knitrepo
3661
return knitrepo.RepositoryFormatKnit1()
3664
def is_compatible(source, target):
3665
"""Be compatible with known Knit formats.
3667
We don't test for the stores being of specific types because that
3668
could lead to confusing results, and there is no need to be
3671
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3673
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3674
isinstance(target._format, RepositoryFormatKnit))
3675
except AttributeError:
3677
return are_knits and InterRepository._same_model(source, target)
3680
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3681
"""See InterRepository.missing_revision_ids()."""
3682
if revision_id is not None:
3683
source_ids = self.source.get_ancestry(revision_id)
3684
if source_ids[0] is not None:
3685
raise AssertionError()
3688
source_ids = self.source.all_revision_ids()
3689
source_ids_set = set(source_ids)
3690
# source_ids is the worst possible case we may need to pull.
3691
# now we want to filter source_ids against what we actually
3692
# have in target, but don't try to check for existence where we know
3693
# we do not have a revision as that would be pointless.
3694
target_ids = set(self.target.all_revision_ids())
3695
possibly_present_revisions = target_ids.intersection(source_ids_set)
3696
actually_present_revisions = set(
3697
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3698
required_revisions = source_ids_set.difference(actually_present_revisions)
3699
if revision_id is not None:
3700
# we used get_ancestry to determine source_ids then we are assured all
3701
# revisions referenced are present as they are installed in topological order.
3702
# and the tip revision was validated by get_ancestry.
3703
result_set = required_revisions
3705
# if we just grabbed the possibly available ids, then
3706
# we only have an estimate of whats available and need to validate
3707
# that against the revision records.
3709
self.source._eliminate_revisions_not_present(required_revisions))
3710
return self.source.revision_ids_to_search_result(result_set)
3713
class InterDifferingSerializer(InterRepository):
3716
def _get_repo_format_to_test(self):
3720
def is_compatible(source, target):
3721
"""Be compatible with Knit2 source and Knit3 target"""
3722
# This is redundant with format.check_conversion_target(), however that
3723
# raises an exception, and we just want to say "False" as in we won't
3724
# support converting between these formats.
3725
if 'IDS_never' in debug.debug_flags:
3727
if source.supports_rich_root() and not target.supports_rich_root():
3729
if (source._format.supports_tree_reference
3730
and not target._format.supports_tree_reference):
3732
if target._fallback_repositories and target._format.supports_chks:
3733
# IDS doesn't know how to copy CHKs for the parent inventories it
3734
# adds to stacked repos.
3736
if 'IDS_always' in debug.debug_flags:
3738
# Only use this code path for local source and target. IDS does far
3739
# too much IO (both bandwidth and roundtrips) over a network.
3740
if not source.bzrdir.transport.base.startswith('file:///'):
3742
if not target.bzrdir.transport.base.startswith('file:///'):
3746
def _get_trees(self, revision_ids, cache):
3748
for rev_id in revision_ids:
3750
possible_trees.append((rev_id, cache[rev_id]))
3752
# Not cached, but inventory might be present anyway.
3754
tree = self.source.revision_tree(rev_id)
3755
except errors.NoSuchRevision:
3756
# Nope, parent is ghost.
3759
cache[rev_id] = tree
3760
possible_trees.append((rev_id, tree))
3761
return possible_trees
3763
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3764
"""Get the best delta and base for this revision.
3766
:return: (basis_id, delta)
3769
# Generate deltas against each tree, to find the shortest.
3770
texts_possibly_new_in_tree = set()
3771
for basis_id, basis_tree in possible_trees:
3772
delta = tree.inventory._make_delta(basis_tree.inventory)
3773
for old_path, new_path, file_id, new_entry in delta:
3774
if new_path is None:
3775
# This file_id isn't present in the new rev, so we don't
3779
# Rich roots are handled elsewhere...
3781
kind = new_entry.kind
3782
if kind != 'directory' and kind != 'file':
3783
# No text record associated with this inventory entry.
3785
# This is a directory or file that has changed somehow.
3786
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3787
deltas.append((len(delta), basis_id, delta))
3789
return deltas[0][1:]
3791
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3792
"""Find all parent revisions that are absent, but for which the
3793
inventory is present, and copy those inventories.
3795
This is necessary to preserve correctness when the source is stacked
3796
without fallbacks configured. (Note that in cases like upgrade the
3797
source may be not have _fallback_repositories even though it is
3801
for parents in parent_map.values():
3802
parent_revs.update(parents)
3803
present_parents = self.source.get_parent_map(parent_revs)
3804
absent_parents = set(parent_revs).difference(present_parents)
3805
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3806
(rev_id,) for rev_id in absent_parents)
3807
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3808
for parent_tree in self.source.revision_trees(parent_inv_ids):
3809
current_revision_id = parent_tree.get_revision_id()
3810
parents_parents_keys = parent_invs_keys_for_stacking[
3811
(current_revision_id,)]
3812
parents_parents = [key[-1] for key in parents_parents_keys]
3813
basis_id = _mod_revision.NULL_REVISION
3814
basis_tree = self.source.revision_tree(basis_id)
3815
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3816
self.target.add_inventory_by_delta(
3817
basis_id, delta, current_revision_id, parents_parents)
3818
cache[current_revision_id] = parent_tree
3820
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3821
"""Fetch across a few revisions.
3823
:param revision_ids: The revisions to copy
3824
:param basis_id: The revision_id of a tree that must be in cache, used
3825
as a basis for delta when no other base is available
3826
:param cache: A cache of RevisionTrees that we can use.
3827
:param a_graph: A Graph object to determine the heads() of the
3828
rich-root data stream.
3829
:return: The revision_id of the last converted tree. The RevisionTree
3830
for it will be in cache
3832
# Walk though all revisions; get inventory deltas, copy referenced
3833
# texts that delta references, insert the delta, revision and
3835
root_keys_to_create = set()
3838
pending_revisions = []
3839
parent_map = self.source.get_parent_map(revision_ids)
3840
self._fetch_parent_invs_for_stacking(parent_map, cache)
3841
self.source._safe_to_return_from_cache = True
3842
for tree in self.source.revision_trees(revision_ids):
3843
# Find a inventory delta for this revision.
3844
# Find text entries that need to be copied, too.
3845
current_revision_id = tree.get_revision_id()
3846
parent_ids = parent_map.get(current_revision_id, ())
3847
parent_trees = self._get_trees(parent_ids, cache)
3848
possible_trees = list(parent_trees)
3849
if len(possible_trees) == 0:
3850
# There either aren't any parents, or the parents are ghosts,
3851
# so just use the last converted tree.
3852
possible_trees.append((basis_id, cache[basis_id]))
3853
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3855
revision = self.source.get_revision(current_revision_id)
3856
pending_deltas.append((basis_id, delta,
3857
current_revision_id, revision.parent_ids))
3858
if self._converting_to_rich_root:
3859
self._revision_id_to_root_id[current_revision_id] = \
3861
# Determine which texts are in present in this revision but not in
3862
# any of the available parents.
3863
texts_possibly_new_in_tree = set()
3864
for old_path, new_path, file_id, entry in delta:
3865
if new_path is None:
3866
# This file_id isn't present in the new rev
3870
if not self.target.supports_rich_root():
3871
# The target doesn't support rich root, so we don't
3874
if self._converting_to_rich_root:
3875
# This can't be copied normally, we have to insert
3877
root_keys_to_create.add((file_id, entry.revision))
3880
texts_possibly_new_in_tree.add((file_id, entry.revision))
3881
for basis_id, basis_tree in possible_trees:
3882
basis_inv = basis_tree.inventory
3883
for file_key in list(texts_possibly_new_in_tree):
3884
file_id, file_revision = file_key
3886
entry = basis_inv[file_id]
3887
except errors.NoSuchId:
3889
if entry.revision == file_revision:
3890
texts_possibly_new_in_tree.remove(file_key)
3891
text_keys.update(texts_possibly_new_in_tree)
3892
pending_revisions.append(revision)
3893
cache[current_revision_id] = tree
3894
basis_id = current_revision_id
3895
self.source._safe_to_return_from_cache = False
3897
from_texts = self.source.texts
3898
to_texts = self.target.texts
3899
if root_keys_to_create:
3900
root_stream = _mod_fetch._new_root_data_stream(
3901
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3902
self.source, graph=a_graph)
3903
to_texts.insert_record_stream(root_stream)
3904
to_texts.insert_record_stream(from_texts.get_record_stream(
3905
text_keys, self.target._format._fetch_order,
3906
not self.target._format._fetch_uses_deltas))
3907
# insert inventory deltas
3908
for delta in pending_deltas:
3909
self.target.add_inventory_by_delta(*delta)
3910
if self.target._fallback_repositories:
3911
# Make sure this stacked repository has all the parent inventories
3912
# for the new revisions that we are about to insert. We do this
3913
# before adding the revisions so that no revision is added until
3914
# all the inventories it may depend on are added.
3915
# Note that this is overzealous, as we may have fetched these in an
3918
revision_ids = set()
3919
for revision in pending_revisions:
3920
revision_ids.add(revision.revision_id)
3921
parent_ids.update(revision.parent_ids)
3922
parent_ids.difference_update(revision_ids)
3923
parent_ids.discard(_mod_revision.NULL_REVISION)
3924
parent_map = self.source.get_parent_map(parent_ids)
3925
# we iterate over parent_map and not parent_ids because we don't
3926
# want to try copying any revision which is a ghost
3927
for parent_tree in self.source.revision_trees(parent_map):
3928
current_revision_id = parent_tree.get_revision_id()
3929
parents_parents = parent_map[current_revision_id]
3930
possible_trees = self._get_trees(parents_parents, cache)
3931
if len(possible_trees) == 0:
3932
# There either aren't any parents, or the parents are
3933
# ghosts, so just use the last converted tree.
3934
possible_trees.append((basis_id, cache[basis_id]))
3935
basis_id, delta = self._get_delta_for_revision(parent_tree,
3936
parents_parents, possible_trees)
3937
self.target.add_inventory_by_delta(
3938
basis_id, delta, current_revision_id, parents_parents)
3939
# insert signatures and revisions
3940
for revision in pending_revisions:
3942
signature = self.source.get_signature_text(
3943
revision.revision_id)
3944
self.target.add_signature_text(revision.revision_id,
3946
except errors.NoSuchRevision:
3948
self.target.add_revision(revision.revision_id, revision)
3951
def _fetch_all_revisions(self, revision_ids, pb):
3952
"""Fetch everything for the list of revisions.
3954
:param revision_ids: The list of revisions to fetch. Must be in
3956
:param pb: A ProgressTask
3959
basis_id, basis_tree = self._get_basis(revision_ids[0])
3961
cache = lru_cache.LRUCache(100)
3962
cache[basis_id] = basis_tree
3963
del basis_tree # We don't want to hang on to it here
3965
if self._converting_to_rich_root and len(revision_ids) > 100:
3966
a_graph = _mod_fetch._get_rich_root_heads_graph(self.source,
3971
for offset in range(0, len(revision_ids), batch_size):
3972
self.target.start_write_group()
3974
pb.update('Transferring revisions', offset,
3976
batch = revision_ids[offset:offset+batch_size]
3977
basis_id = self._fetch_batch(batch, basis_id, cache,
3980
self.source._safe_to_return_from_cache = False
3981
self.target.abort_write_group()
3984
hint = self.target.commit_write_group()
3987
if hints and self.target._format.pack_compresses:
3988
self.target.pack(hint=hints)
3989
pb.update('Transferring revisions', len(revision_ids),
3993
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3995
"""See InterRepository.fetch()."""
3996
if fetch_spec is not None:
3997
raise AssertionError("Not implemented yet...")
3998
# See <https://launchpad.net/bugs/456077> asking for a warning here
4000
# nb this is only active for local-local fetches; other things using
4002
ui.ui_factory.warn_cross_format_fetch(self.source._format,
4003
self.target._format)
4004
ui.ui_factory.warn_experimental_format_fetch(self)
4005
if (not self.source.supports_rich_root()
4006
and self.target.supports_rich_root()):
4007
self._converting_to_rich_root = True
4008
self._revision_id_to_root_id = {}
4010
self._converting_to_rich_root = False
4011
revision_ids = self.target.search_missing_revision_ids(self.source,
4012
revision_id, find_ghosts=find_ghosts).get_keys()
4013
if not revision_ids:
4015
revision_ids = tsort.topo_sort(
4016
self.source.get_graph().get_parent_map(revision_ids))
4017
if not revision_ids:
4019
# Walk though all revisions; get inventory deltas, copy referenced
4020
# texts that delta references, insert the delta, revision and
4023
my_pb = ui.ui_factory.nested_progress_bar()
4026
symbol_versioning.warn(
4027
symbol_versioning.deprecated_in((1, 14, 0))
4028
% "pb parameter to fetch()")
4031
self._fetch_all_revisions(revision_ids, pb)
4033
if my_pb is not None:
4035
return len(revision_ids), 0
4037
def _get_basis(self, first_revision_id):
4038
"""Get a revision and tree which exists in the target.
4040
This assumes that first_revision_id is selected for transmission
4041
because all other ancestors are already present. If we can't find an
4042
ancestor we fall back to NULL_REVISION since we know that is safe.
4044
:return: (basis_id, basis_tree)
4046
first_rev = self.source.get_revision(first_revision_id)
4048
basis_id = first_rev.parent_ids[0]
4049
# only valid as a basis if the target has it
4050
self.target.get_revision(basis_id)
4051
# Try to get a basis tree - if its a ghost it will hit the
4052
# NoSuchRevision case.
4053
basis_tree = self.source.revision_tree(basis_id)
4054
except (IndexError, errors.NoSuchRevision):
4055
basis_id = _mod_revision.NULL_REVISION
4056
basis_tree = self.source.revision_tree(basis_id)
4057
return basis_id, basis_tree
4060
InterRepository.register_optimiser(InterDifferingSerializer)
4061
InterRepository.register_optimiser(InterSameDataRepository)
4062
InterRepository.register_optimiser(InterWeaveRepo)
4063
InterRepository.register_optimiser(InterKnitRepo)
4066
1776
class CopyConverter(object):
4067
1777
"""A repository conversion tool which just performs a copy of the content.
4224
1832
return revision_graph
4227
class StreamSink(object):
4228
"""An object that can insert a stream into a repository.
4230
This interface handles the complexity of reserialising inventories and
4231
revisions from different formats, and allows unidirectional insertion into
4232
stacked repositories without looking for the missing basis parents
4236
def __init__(self, target_repo):
4237
self.target_repo = target_repo
4239
def insert_stream(self, stream, src_format, resume_tokens):
4240
"""Insert a stream's content into the target repository.
4242
:param src_format: a bzr repository format.
4244
:return: a list of resume tokens and an iterable of keys additional
4245
items required before the insertion can be completed.
4247
self.target_repo.lock_write()
4250
self.target_repo.resume_write_group(resume_tokens)
4253
self.target_repo.start_write_group()
4256
# locked_insert_stream performs a commit|suspend.
4257
return self._locked_insert_stream(stream, src_format, is_resume)
4259
self.target_repo.abort_write_group(suppress_errors=True)
4262
self.target_repo.unlock()
4264
def _locked_insert_stream(self, stream, src_format, is_resume):
4265
to_serializer = self.target_repo._format._serializer
4266
src_serializer = src_format._serializer
4268
if to_serializer == src_serializer:
4269
# If serializers match and the target is a pack repository, set the
4270
# write cache size on the new pack. This avoids poor performance
4271
# on transports where append is unbuffered (such as
4272
# RemoteTransport). This is safe to do because nothing should read
4273
# back from the target repository while a stream with matching
4274
# serialization is being inserted.
4275
# The exception is that a delta record from the source that should
4276
# be a fulltext may need to be expanded by the target (see
4277
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4278
# explicitly flush any buffered writes first in that rare case.
4280
new_pack = self.target_repo._pack_collection._new_pack
4281
except AttributeError:
4282
# Not a pack repository
4285
new_pack.set_write_cache_size(1024*1024)
4286
for substream_type, substream in stream:
4287
if 'stream' in debug.debug_flags:
4288
mutter('inserting substream: %s', substream_type)
4289
if substream_type == 'texts':
4290
self.target_repo.texts.insert_record_stream(substream)
4291
elif substream_type == 'inventories':
4292
if src_serializer == to_serializer:
4293
self.target_repo.inventories.insert_record_stream(
4296
self._extract_and_insert_inventories(
4297
substream, src_serializer)
4298
elif substream_type == 'inventory-deltas':
4299
ui.ui_factory.warn_cross_format_fetch(src_format,
4300
self.target_repo._format)
4301
self._extract_and_insert_inventory_deltas(
4302
substream, src_serializer)
4303
elif substream_type == 'chk_bytes':
4304
# XXX: This doesn't support conversions, as it assumes the
4305
# conversion was done in the fetch code.
4306
self.target_repo.chk_bytes.insert_record_stream(substream)
4307
elif substream_type == 'revisions':
4308
# This may fallback to extract-and-insert more often than
4309
# required if the serializers are different only in terms of
4311
if src_serializer == to_serializer:
4312
self.target_repo.revisions.insert_record_stream(
4315
self._extract_and_insert_revisions(substream,
4317
elif substream_type == 'signatures':
4318
self.target_repo.signatures.insert_record_stream(substream)
4320
raise AssertionError('kaboom! %s' % (substream_type,))
4321
# Done inserting data, and the missing_keys calculations will try to
4322
# read back from the inserted data, so flush the writes to the new pack
4323
# (if this is pack format).
4324
if new_pack is not None:
4325
new_pack._write_data('', flush=True)
4326
# Find all the new revisions (including ones from resume_tokens)
4327
missing_keys = self.target_repo.get_missing_parent_inventories(
4328
check_for_missing_texts=is_resume)
4330
for prefix, versioned_file in (
4331
('texts', self.target_repo.texts),
4332
('inventories', self.target_repo.inventories),
4333
('revisions', self.target_repo.revisions),
4334
('signatures', self.target_repo.signatures),
4335
('chk_bytes', self.target_repo.chk_bytes),
4337
if versioned_file is None:
4339
# TODO: key is often going to be a StaticTuple object
4340
# I don't believe we can define a method by which
4341
# (prefix,) + StaticTuple will work, though we could
4342
# define a StaticTuple.sq_concat that would allow you to
4343
# pass in either a tuple or a StaticTuple as the second
4344
# object, so instead we could have:
4345
# StaticTuple(prefix) + key here...
4346
missing_keys.update((prefix,) + key for key in
4347
versioned_file.get_missing_compression_parent_keys())
4348
except NotImplementedError:
4349
# cannot even attempt suspending, and missing would have failed
4350
# during stream insertion.
4351
missing_keys = set()
4354
# suspend the write group and tell the caller what we is
4355
# missing. We know we can suspend or else we would not have
4356
# entered this code path. (All repositories that can handle
4357
# missing keys can handle suspending a write group).
4358
write_group_tokens = self.target_repo.suspend_write_group()
4359
return write_group_tokens, missing_keys
4360
hint = self.target_repo.commit_write_group()
4361
if (to_serializer != src_serializer and
4362
self.target_repo._format.pack_compresses):
4363
self.target_repo.pack(hint=hint)
4366
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4367
target_rich_root = self.target_repo._format.rich_root_data
4368
target_tree_refs = self.target_repo._format.supports_tree_reference
4369
for record in substream:
4370
# Insert the delta directly
4371
inventory_delta_bytes = record.get_bytes_as('fulltext')
4372
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4374
parse_result = deserialiser.parse_text_bytes(
4375
inventory_delta_bytes)
4376
except inventory_delta.IncompatibleInventoryDelta, err:
4377
trace.mutter("Incompatible delta: %s", err.msg)
4378
raise errors.IncompatibleRevision(self.target_repo._format)
4379
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4380
revision_id = new_id
4381
parents = [key[0] for key in record.parents]
4382
self.target_repo.add_inventory_by_delta(
4383
basis_id, inv_delta, revision_id, parents)
4385
def _extract_and_insert_inventories(self, substream, serializer,
4387
"""Generate a new inventory versionedfile in target, converting data.
4389
The inventory is retrieved from the source, (deserializing it), and
4390
stored in the target (reserializing it in a different format).
4392
target_rich_root = self.target_repo._format.rich_root_data
4393
target_tree_refs = self.target_repo._format.supports_tree_reference
4394
for record in substream:
4395
# It's not a delta, so it must be a fulltext in the source
4396
# serializer's format.
4397
bytes = record.get_bytes_as('fulltext')
4398
revision_id = record.key[0]
4399
inv = serializer.read_inventory_from_string(bytes, revision_id)
4400
parents = [key[0] for key in record.parents]
4401
self.target_repo.add_inventory(revision_id, inv, parents)
4402
# No need to keep holding this full inv in memory when the rest of
4403
# the substream is likely to be all deltas.
4406
def _extract_and_insert_revisions(self, substream, serializer):
4407
for record in substream:
4408
bytes = record.get_bytes_as('fulltext')
4409
revision_id = record.key[0]
4410
rev = serializer.read_revision_from_string(bytes)
4411
if rev.revision_id != revision_id:
4412
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4413
self.target_repo.add_revision(revision_id, rev)
4416
if self.target_repo._format._fetch_reconcile:
4417
self.target_repo.reconcile()
4420
class StreamSource(object):
4421
"""A source of a stream for fetching between repositories."""
4423
def __init__(self, from_repository, to_format):
4424
"""Create a StreamSource streaming from from_repository."""
4425
self.from_repository = from_repository
4426
self.to_format = to_format
4428
def delta_on_metadata(self):
4429
"""Return True if delta's are permitted on metadata streams.
4431
That is on revisions and signatures.
4433
src_serializer = self.from_repository._format._serializer
4434
target_serializer = self.to_format._serializer
4435
return (self.to_format._fetch_uses_deltas and
4436
src_serializer == target_serializer)
4438
def _fetch_revision_texts(self, revs):
4439
# fetch signatures first and then the revision texts
4440
# may need to be a InterRevisionStore call here.
4441
from_sf = self.from_repository.signatures
4442
# A missing signature is just skipped.
4443
keys = [(rev_id,) for rev_id in revs]
4444
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4446
self.to_format._fetch_order,
4447
not self.to_format._fetch_uses_deltas))
4448
# If a revision has a delta, this is actually expanded inside the
4449
# insert_record_stream code now, which is an alternate fix for
4451
from_rf = self.from_repository.revisions
4452
revisions = from_rf.get_record_stream(
4454
self.to_format._fetch_order,
4455
not self.delta_on_metadata())
4456
return [('signatures', signatures), ('revisions', revisions)]
4458
def _generate_root_texts(self, revs):
4459
"""This will be called by get_stream between fetching weave texts and
4460
fetching the inventory weave.
4462
if self._rich_root_upgrade():
4463
return _mod_fetch.Inter1and2Helper(
4464
self.from_repository).generate_root_texts(revs)
4468
def get_stream(self, search):
4470
revs = search.get_keys()
4471
graph = self.from_repository.get_graph()
4472
revs = tsort.topo_sort(graph.get_parent_map(revs))
4473
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4475
for knit_kind, file_id, revisions in data_to_fetch:
4476
if knit_kind != phase:
4478
# Make a new progress bar for this phase
4479
if knit_kind == "file":
4480
# Accumulate file texts
4481
text_keys.extend([(file_id, revision) for revision in
4483
elif knit_kind == "inventory":
4484
# Now copy the file texts.
4485
from_texts = self.from_repository.texts
4486
yield ('texts', from_texts.get_record_stream(
4487
text_keys, self.to_format._fetch_order,
4488
not self.to_format._fetch_uses_deltas))
4489
# Cause an error if a text occurs after we have done the
4492
# Before we process the inventory we generate the root
4493
# texts (if necessary) so that the inventories references
4495
for _ in self._generate_root_texts(revs):
4497
# we fetch only the referenced inventories because we do not
4498
# know for unselected inventories whether all their required
4499
# texts are present in the other repository - it could be
4501
for info in self._get_inventory_stream(revs):
4503
elif knit_kind == "signatures":
4504
# Nothing to do here; this will be taken care of when
4505
# _fetch_revision_texts happens.
4507
elif knit_kind == "revisions":
4508
for record in self._fetch_revision_texts(revs):
4511
raise AssertionError("Unknown knit kind %r" % knit_kind)
4513
def get_stream_for_missing_keys(self, missing_keys):
4514
# missing keys can only occur when we are byte copying and not
4515
# translating (because translation means we don't send
4516
# unreconstructable deltas ever).
4518
keys['texts'] = set()
4519
keys['revisions'] = set()
4520
keys['inventories'] = set()
4521
keys['chk_bytes'] = set()
4522
keys['signatures'] = set()
4523
for key in missing_keys:
4524
keys[key[0]].add(key[1:])
4525
if len(keys['revisions']):
4526
# If we allowed copying revisions at this point, we could end up
4527
# copying a revision without copying its required texts: a
4528
# violation of the requirements for repository integrity.
4529
raise AssertionError(
4530
'cannot copy revisions to fill in missing deltas %s' % (
4531
keys['revisions'],))
4532
for substream_kind, keys in keys.iteritems():
4533
vf = getattr(self.from_repository, substream_kind)
4534
if vf is None and keys:
4535
raise AssertionError(
4536
"cannot fill in keys for a versioned file we don't"
4537
" have: %s needs %s" % (substream_kind, keys))
4539
# No need to stream something we don't have
4541
if substream_kind == 'inventories':
4542
# Some missing keys are genuinely ghosts, filter those out.
4543
present = self.from_repository.inventories.get_parent_map(keys)
4544
revs = [key[0] for key in present]
4545
# Get the inventory stream more-or-less as we do for the
4546
# original stream; there's no reason to assume that records
4547
# direct from the source will be suitable for the sink. (Think
4548
# e.g. 2a -> 1.9-rich-root).
4549
for info in self._get_inventory_stream(revs, missing=True):
4553
# Ask for full texts always so that we don't need more round trips
4554
# after this stream.
4555
# Some of the missing keys are genuinely ghosts, so filter absent
4556
# records. The Sink is responsible for doing another check to
4557
# ensure that ghosts don't introduce missing data for future
4559
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4560
self.to_format._fetch_order, True))
4561
yield substream_kind, stream
4563
def inventory_fetch_order(self):
4564
if self._rich_root_upgrade():
4565
return 'topological'
4567
return self.to_format._fetch_order
4569
def _rich_root_upgrade(self):
4570
return (not self.from_repository._format.rich_root_data and
4571
self.to_format.rich_root_data)
4573
def _get_inventory_stream(self, revision_ids, missing=False):
4574
from_format = self.from_repository._format
4575
if (from_format.supports_chks and self.to_format.supports_chks and
4576
from_format.network_name() == self.to_format.network_name()):
4577
raise AssertionError(
4578
"this case should be handled by GroupCHKStreamSource")
4579
elif 'forceinvdeltas' in debug.debug_flags:
4580
return self._get_convertable_inventory_stream(revision_ids,
4581
delta_versus_null=missing)
4582
elif from_format.network_name() == self.to_format.network_name():
4584
return self._get_simple_inventory_stream(revision_ids,
4586
elif (not from_format.supports_chks and not self.to_format.supports_chks
4587
and from_format._serializer == self.to_format._serializer):
4588
# Essentially the same format.
4589
return self._get_simple_inventory_stream(revision_ids,
4592
# Any time we switch serializations, we want to use an
4593
# inventory-delta based approach.
4594
return self._get_convertable_inventory_stream(revision_ids,
4595
delta_versus_null=missing)
4597
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4598
# NB: This currently reopens the inventory weave in source;
4599
# using a single stream interface instead would avoid this.
4600
from_weave = self.from_repository.inventories
4602
delta_closure = True
4604
delta_closure = not self.delta_on_metadata()
4605
yield ('inventories', from_weave.get_record_stream(
4606
[(rev_id,) for rev_id in revision_ids],
4607
self.inventory_fetch_order(), delta_closure))
4609
def _get_convertable_inventory_stream(self, revision_ids,
4610
delta_versus_null=False):
4611
# The two formats are sufficiently different that there is no fast
4612
# path, so we need to send just inventorydeltas, which any
4613
# sufficiently modern client can insert into any repository.
4614
# The StreamSink code expects to be able to
4615
# convert on the target, so we need to put bytes-on-the-wire that can
4616
# be converted. That means inventory deltas (if the remote is <1.19,
4617
# RemoteStreamSink will fallback to VFS to insert the deltas).
4618
yield ('inventory-deltas',
4619
self._stream_invs_as_deltas(revision_ids,
4620
delta_versus_null=delta_versus_null))
4622
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4623
"""Return a stream of inventory-deltas for the given rev ids.
4625
:param revision_ids: The list of inventories to transmit
4626
:param delta_versus_null: Don't try to find a minimal delta for this
4627
entry, instead compute the delta versus the NULL_REVISION. This
4628
effectively streams a complete inventory. Used for stuff like
4629
filling in missing parents, etc.
4631
from_repo = self.from_repository
4632
revision_keys = [(rev_id,) for rev_id in revision_ids]
4633
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4634
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4636
inventories = self.from_repository.iter_inventories(
4637
revision_ids, 'topological')
4638
format = from_repo._format
4639
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4640
inventory_cache = lru_cache.LRUCache(50)
4641
null_inventory = from_repo.revision_tree(
4642
_mod_revision.NULL_REVISION).inventory
4643
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4644
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4645
# repo back into a non-rich-root repo ought to be allowed)
4646
serializer = inventory_delta.InventoryDeltaSerializer(
4647
versioned_root=format.rich_root_data,
4648
tree_references=format.supports_tree_reference)
4649
for inv in inventories:
4650
key = (inv.revision_id,)
4651
parent_keys = parent_map.get(key, ())
4653
if not delta_versus_null and parent_keys:
4654
# The caller did not ask for complete inventories and we have
4655
# some parents that we can delta against. Make a delta against
4656
# each parent so that we can find the smallest.
4657
parent_ids = [parent_key[0] for parent_key in parent_keys]
4658
for parent_id in parent_ids:
4659
if parent_id not in invs_sent_so_far:
4660
# We don't know that the remote side has this basis, so
4663
if parent_id == _mod_revision.NULL_REVISION:
4664
parent_inv = null_inventory
4666
parent_inv = inventory_cache.get(parent_id, None)
4667
if parent_inv is None:
4668
parent_inv = from_repo.get_inventory(parent_id)
4669
candidate_delta = inv._make_delta(parent_inv)
4670
if (delta is None or
4671
len(delta) > len(candidate_delta)):
4672
delta = candidate_delta
4673
basis_id = parent_id
4675
# Either none of the parents ended up being suitable, or we
4676
# were asked to delta against NULL
4677
basis_id = _mod_revision.NULL_REVISION
4678
delta = inv._make_delta(null_inventory)
4679
invs_sent_so_far.add(inv.revision_id)
4680
inventory_cache[inv.revision_id] = inv
4681
delta_serialized = ''.join(
4682
serializer.delta_to_lines(basis_id, key[-1], delta))
4683
yield versionedfile.FulltextContentFactory(
4684
key, parent_keys, None, delta_serialized)
4687
1835
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4688
1836
stop_revision=None):
4689
1837
"""Extend the partial history to include a given index
added
removed
bzrlib/repository.py
