202
342
commit to be valid, deletes against the basis MUST be recorded via
203
343
builder.record_delete().
205
raise NotImplementedError(self.will_record_deletes)
207
def record_iter_changes(self, tree, basis_revision_id, iter_changes):
345
self._recording_deletes = True
347
basis_id = self.parents[0]
349
basis_id = _mod_revision.NULL_REVISION
350
self.basis_delta_revision = basis_id
352
def record_entry_contents(self, ie, parent_invs, path, tree,
354
"""Record the content of ie from tree into the commit if needed.
356
Side effect: sets ie.revision when unchanged
358
:param ie: An inventory entry present in the commit.
359
:param parent_invs: The inventories of the parent revisions of the
361
:param path: The path the entry is at in the tree.
362
:param tree: The tree which contains this entry and should be used to
364
:param content_summary: Summary data from the tree about the paths
365
content - stat, length, exec, sha/link target. This is only
366
accessed when the entry has a revision of None - that is when it is
367
a candidate to commit.
368
:return: A tuple (change_delta, version_recorded, fs_hash).
369
change_delta is an inventory_delta change for this entry against
370
the basis tree of the commit, or None if no change occured against
372
version_recorded is True if a new version of the entry has been
373
recorded. For instance, committing a merge where a file was only
374
changed on the other side will return (delta, False).
375
fs_hash is either None, or the hash details for the path (currently
376
a tuple of the contents sha1 and the statvalue returned by
377
tree.get_file_with_stat()).
379
if self.new_inventory.root is None:
380
if ie.parent_id is not None:
381
raise errors.RootMissing()
382
self._check_root(ie, parent_invs, tree)
383
if ie.revision is None:
384
kind = content_summary[0]
386
# ie is carried over from a prior commit
388
# XXX: repository specific check for nested tree support goes here - if
389
# the repo doesn't want nested trees we skip it ?
390
if (kind == 'tree-reference' and
391
not self.repository._format.supports_tree_reference):
392
# mismatch between commit builder logic and repository:
393
# this needs the entry creation pushed down into the builder.
394
raise NotImplementedError('Missing repository subtree support.')
395
self.new_inventory.add(ie)
397
# TODO: slow, take it out of the inner loop.
399
basis_inv = parent_invs[0]
401
basis_inv = Inventory(root_id=None)
403
# ie.revision is always None if the InventoryEntry is considered
404
# for committing. We may record the previous parents revision if the
405
# content is actually unchanged against a sole head.
406
if ie.revision is not None:
407
if not self._versioned_root and path == '':
408
# repositories that do not version the root set the root's
409
# revision to the new commit even when no change occurs (more
410
# specifically, they do not record a revision on the root; and
411
# the rev id is assigned to the root during deserialisation -
412
# this masks when a change may have occurred against the basis.
413
# To match this we always issue a delta, because the revision
414
# of the root will always be changing.
415
if ie.file_id in basis_inv:
416
delta = (basis_inv.id2path(ie.file_id), path,
420
delta = (None, path, ie.file_id, ie)
421
self._basis_delta.append(delta)
422
return delta, False, None
424
# we don't need to commit this, because the caller already
425
# determined that an existing revision of this file is
426
# appropriate. If its not being considered for committing then
427
# it and all its parents to the root must be unaltered so
428
# no-change against the basis.
429
if ie.revision == self._new_revision_id:
430
raise AssertionError("Impossible situation, a skipped "
431
"inventory entry (%r) claims to be modified in this "
432
"commit (%r).", (ie, self._new_revision_id))
433
return None, False, None
434
# XXX: Friction: parent_candidates should return a list not a dict
435
# so that we don't have to walk the inventories again.
436
parent_candiate_entries = ie.parent_candidates(parent_invs)
437
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
439
for inv in parent_invs:
440
if ie.file_id in inv:
441
old_rev = inv[ie.file_id].revision
442
if old_rev in head_set:
443
heads.append(inv[ie.file_id].revision)
444
head_set.remove(inv[ie.file_id].revision)
447
# now we check to see if we need to write a new record to the
449
# We write a new entry unless there is one head to the ancestors, and
450
# the kind-derived content is unchanged.
452
# Cheapest check first: no ancestors, or more the one head in the
453
# ancestors, we write a new node.
457
# There is a single head, look it up for comparison
458
parent_entry = parent_candiate_entries[heads[0]]
459
# if the non-content specific data has changed, we'll be writing a
461
if (parent_entry.parent_id != ie.parent_id or
462
parent_entry.name != ie.name):
464
# now we need to do content specific checks:
466
# if the kind changed the content obviously has
467
if kind != parent_entry.kind:
469
# Stat cache fingerprint feedback for the caller - None as we usually
470
# don't generate one.
473
if content_summary[2] is None:
474
raise ValueError("Files must not have executable = None")
476
# We can't trust a check of the file length because of content
478
if (# if the exec bit has changed we have to store:
479
parent_entry.executable != content_summary[2]):
481
elif parent_entry.text_sha1 == content_summary[3]:
482
# all meta and content is unchanged (using a hash cache
483
# hit to check the sha)
484
ie.revision = parent_entry.revision
485
ie.text_size = parent_entry.text_size
486
ie.text_sha1 = parent_entry.text_sha1
487
ie.executable = parent_entry.executable
488
return self._get_delta(ie, basis_inv, path), False, None
490
# Either there is only a hash change(no hash cache entry,
491
# or same size content change), or there is no change on
493
# Provide the parent's hash to the store layer, so that the
494
# content is unchanged we will not store a new node.
495
nostore_sha = parent_entry.text_sha1
497
# We want to record a new node regardless of the presence or
498
# absence of a content change in the file.
500
ie.executable = content_summary[2]
501
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
503
text = file_obj.read()
507
ie.text_sha1, ie.text_size = self._add_text_to_weave(
508
ie.file_id, text, heads, nostore_sha)
509
# Let the caller know we generated a stat fingerprint.
510
fingerprint = (ie.text_sha1, stat_value)
511
except errors.ExistingContent:
512
# Turns out that the file content was unchanged, and we were
513
# only going to store a new node if it was changed. Carry over
515
ie.revision = parent_entry.revision
516
ie.text_size = parent_entry.text_size
517
ie.text_sha1 = parent_entry.text_sha1
518
ie.executable = parent_entry.executable
519
return self._get_delta(ie, basis_inv, path), False, None
520
elif kind == 'directory':
522
# all data is meta here, nothing specific to directory, so
524
ie.revision = parent_entry.revision
525
return self._get_delta(ie, basis_inv, path), False, None
526
self._add_text_to_weave(ie.file_id, '', heads, None)
527
elif kind == 'symlink':
528
current_link_target = content_summary[3]
530
# symlink target is not generic metadata, check if it has
532
if current_link_target != parent_entry.symlink_target:
535
# unchanged, carry over.
536
ie.revision = parent_entry.revision
537
ie.symlink_target = parent_entry.symlink_target
538
return self._get_delta(ie, basis_inv, path), False, None
539
ie.symlink_target = current_link_target
540
self._add_text_to_weave(ie.file_id, '', heads, None)
541
elif kind == 'tree-reference':
543
if content_summary[3] != parent_entry.reference_revision:
546
# unchanged, carry over.
547
ie.reference_revision = parent_entry.reference_revision
548
ie.revision = parent_entry.revision
549
return self._get_delta(ie, basis_inv, path), False, None
550
ie.reference_revision = content_summary[3]
551
if ie.reference_revision is None:
552
raise AssertionError("invalid content_summary for nested tree: %r"
553
% (content_summary,))
554
self._add_text_to_weave(ie.file_id, '', heads, None)
556
raise NotImplementedError('unknown kind')
557
ie.revision = self._new_revision_id
558
self._any_changes = True
559
return self._get_delta(ie, basis_inv, path), True, fingerprint
561
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
562
_entry_factory=entry_factory):
208
563
"""Record a new tree via iter_changes.
210
565
:param tree: The tree to obtain text contents from for changed objects.
215
570
to basis_revision_id. The iterator must not include any items with
216
571
a current kind of None - missing items must be either filtered out
217
572
or errored-on beefore record_iter_changes sees the item.
573
:param _entry_factory: Private method to bind entry_factory locally for
218
575
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
219
576
tree._observed_sha1.
221
raise NotImplementedError(self.record_iter_changes)
224
class RepositoryWriteLockResult(LogicalLockResult):
578
# Create an inventory delta based on deltas between all the parents and
579
# deltas between all the parent inventories. We use inventory delta's
580
# between the inventory objects because iter_changes masks
581
# last-changed-field only changes.
583
# file_id -> change map, change is fileid, paths, changed, versioneds,
584
# parents, names, kinds, executables
586
# {file_id -> revision_id -> inventory entry, for entries in parent
587
# trees that are not parents[0]
591
revtrees = list(self.repository.revision_trees(self.parents))
592
except errors.NoSuchRevision:
593
# one or more ghosts, slow path.
595
for revision_id in self.parents:
597
revtrees.append(self.repository.revision_tree(revision_id))
598
except errors.NoSuchRevision:
600
basis_revision_id = _mod_revision.NULL_REVISION
602
revtrees.append(self.repository.revision_tree(
603
_mod_revision.NULL_REVISION))
604
# The basis inventory from a repository
606
basis_inv = revtrees[0].inventory
608
basis_inv = self.repository.revision_tree(
609
_mod_revision.NULL_REVISION).inventory
610
if len(self.parents) > 0:
611
if basis_revision_id != self.parents[0] and not ghost_basis:
613
"arbitrary basis parents not yet supported with merges")
614
for revtree in revtrees[1:]:
615
for change in revtree.inventory._make_delta(basis_inv):
616
if change[1] is None:
617
# Not present in this parent.
619
if change[2] not in merged_ids:
620
if change[0] is not None:
621
basis_entry = basis_inv[change[2]]
622
merged_ids[change[2]] = [
624
basis_entry.revision,
627
parent_entries[change[2]] = {
629
basis_entry.revision:basis_entry,
631
change[3].revision:change[3],
634
merged_ids[change[2]] = [change[3].revision]
635
parent_entries[change[2]] = {change[3].revision:change[3]}
637
merged_ids[change[2]].append(change[3].revision)
638
parent_entries[change[2]][change[3].revision] = change[3]
641
# Setup the changes from the tree:
642
# changes maps file_id -> (change, [parent revision_ids])
644
for change in iter_changes:
645
# This probably looks up in basis_inv way to much.
646
if change[1][0] is not None:
647
head_candidate = [basis_inv[change[0]].revision]
650
changes[change[0]] = change, merged_ids.get(change[0],
652
unchanged_merged = set(merged_ids) - set(changes)
653
# Extend the changes dict with synthetic changes to record merges of
655
for file_id in unchanged_merged:
656
# Record a merged version of these items that did not change vs the
657
# basis. This can be either identical parallel changes, or a revert
658
# of a specific file after a merge. The recorded content will be
659
# that of the current tree (which is the same as the basis), but
660
# the per-file graph will reflect a merge.
661
# NB:XXX: We are reconstructing path information we had, this
662
# should be preserved instead.
663
# inv delta change: (file_id, (path_in_source, path_in_target),
664
# changed_content, versioned, parent, name, kind,
667
basis_entry = basis_inv[file_id]
668
except errors.NoSuchId:
669
# a change from basis->some_parents but file_id isn't in basis
670
# so was new in the merge, which means it must have changed
671
# from basis -> current, and as it hasn't the add was reverted
672
# by the user. So we discard this change.
676
(basis_inv.id2path(file_id), tree.id2path(file_id)),
678
(basis_entry.parent_id, basis_entry.parent_id),
679
(basis_entry.name, basis_entry.name),
680
(basis_entry.kind, basis_entry.kind),
681
(basis_entry.executable, basis_entry.executable))
682
changes[file_id] = (change, merged_ids[file_id])
683
# changes contains tuples with the change and a set of inventory
684
# candidates for the file.
686
# old_path, new_path, file_id, new_inventory_entry
687
seen_root = False # Is the root in the basis delta?
688
inv_delta = self._basis_delta
689
modified_rev = self._new_revision_id
690
for change, head_candidates in changes.values():
691
if change[3][1]: # versioned in target.
692
# Several things may be happening here:
693
# We may have a fork in the per-file graph
694
# - record a change with the content from tree
695
# We may have a change against < all trees
696
# - carry over the tree that hasn't changed
697
# We may have a change against all trees
698
# - record the change with the content from tree
701
entry = _entry_factory[kind](file_id, change[5][1],
703
head_set = self._heads(change[0], set(head_candidates))
706
for head_candidate in head_candidates:
707
if head_candidate in head_set:
708
heads.append(head_candidate)
709
head_set.remove(head_candidate)
712
# Could be a carry-over situation:
713
parent_entry_revs = parent_entries.get(file_id, None)
714
if parent_entry_revs:
715
parent_entry = parent_entry_revs.get(heads[0], None)
718
if parent_entry is None:
719
# The parent iter_changes was called against is the one
720
# that is the per-file head, so any change is relevant
721
# iter_changes is valid.
722
carry_over_possible = False
724
# could be a carry over situation
725
# A change against the basis may just indicate a merge,
726
# we need to check the content against the source of the
727
# merge to determine if it was changed after the merge
729
if (parent_entry.kind != entry.kind or
730
parent_entry.parent_id != entry.parent_id or
731
parent_entry.name != entry.name):
732
# Metadata common to all entries has changed
733
# against per-file parent
734
carry_over_possible = False
736
carry_over_possible = True
737
# per-type checks for changes against the parent_entry
740
# Cannot be a carry-over situation
741
carry_over_possible = False
742
# Populate the entry in the delta
744
# XXX: There is still a small race here: If someone reverts the content of a file
745
# after iter_changes examines and decides it has changed,
746
# we will unconditionally record a new version even if some
747
# other process reverts it while commit is running (with
748
# the revert happening after iter_changes did it's
751
entry.executable = True
753
entry.executable = False
754
if (carry_over_possible and
755
parent_entry.executable == entry.executable):
756
# Check the file length, content hash after reading
758
nostore_sha = parent_entry.text_sha1
761
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
763
text = file_obj.read()
767
entry.text_sha1, entry.text_size = self._add_text_to_weave(
768
file_id, text, heads, nostore_sha)
769
yield file_id, change[1][1], (entry.text_sha1, stat_value)
770
except errors.ExistingContent:
771
# No content change against a carry_over parent
772
# Perhaps this should also yield a fs hash update?
774
entry.text_size = parent_entry.text_size
775
entry.text_sha1 = parent_entry.text_sha1
776
elif kind == 'symlink':
778
entry.symlink_target = tree.get_symlink_target(file_id)
779
if (carry_over_possible and
780
parent_entry.symlink_target == entry.symlink_target):
783
self._add_text_to_weave(change[0], '', heads, None)
784
elif kind == 'directory':
785
if carry_over_possible:
788
# Nothing to set on the entry.
789
# XXX: split into the Root and nonRoot versions.
790
if change[1][1] != '' or self.repository.supports_rich_root():
791
self._add_text_to_weave(change[0], '', heads, None)
792
elif kind == 'tree-reference':
793
if not self.repository._format.supports_tree_reference:
794
# This isn't quite sane as an error, but we shouldn't
795
# ever see this code path in practice: tree's don't
796
# permit references when the repo doesn't support tree
798
raise errors.UnsupportedOperation(tree.add_reference,
800
reference_revision = tree.get_reference_revision(change[0])
801
entry.reference_revision = reference_revision
802
if (carry_over_possible and
803
parent_entry.reference_revision == reference_revision):
806
self._add_text_to_weave(change[0], '', heads, None)
808
raise AssertionError('unknown kind %r' % kind)
810
entry.revision = modified_rev
812
entry.revision = parent_entry.revision
815
new_path = change[1][1]
816
inv_delta.append((change[1][0], new_path, change[0], entry))
819
self.new_inventory = None
821
# This should perhaps be guarded by a check that the basis we
822
# commit against is the basis for the commit and if not do a delta
824
self._any_changes = True
826
# housekeeping root entry changes do not affect no-change commits.
827
self._require_root_change(tree)
828
self.basis_delta_revision = basis_revision_id
830
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
831
parent_keys = tuple([(file_id, parent) for parent in parents])
832
return self.repository.texts._add_text(
833
(file_id, self._new_revision_id), parent_keys, new_text,
834
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
837
class RootCommitBuilder(CommitBuilder):
838
"""This commitbuilder actually records the root id"""
840
# the root entry gets versioned properly by this builder.
841
_versioned_root = True
843
def _check_root(self, ie, parent_invs, tree):
844
"""Helper for record_entry_contents.
846
:param ie: An entry being added.
847
:param parent_invs: The inventories of the parent revisions of the
849
:param tree: The tree that is being committed.
852
def _require_root_change(self, tree):
853
"""Enforce an appropriate root object change.
855
This is called once when record_iter_changes is called, if and only if
856
the root was not in the delta calculated by record_iter_changes.
858
:param tree: The tree which is being committed.
860
# versioned roots do not change unless the tree found a change.
863
class RepositoryWriteLockResult(object):
225
864
"""The result of write locking a repository.
227
866
:ivar repository_token: The token obtained from the underlying lock, or
313
1014
return InterRepository._assert_same_model(self, repository)
1016
def add_inventory(self, revision_id, inv, parents):
1017
"""Add the inventory inv to the repository as revision_id.
1019
:param parents: The revision ids of the parents that revision_id
1020
is known to have and are in the repository already.
1022
:returns: The validator(which is a sha1 digest, though what is sha'd is
1023
repository format specific) of the serialized inventory.
1025
if not self.is_in_write_group():
1026
raise AssertionError("%r not in write group" % (self,))
1027
_mod_revision.check_not_reserved_id(revision_id)
1028
if not (inv.revision_id is None or inv.revision_id == revision_id):
1029
raise AssertionError(
1030
"Mismatch between inventory revision"
1031
" id and insertion revid (%r, %r)"
1032
% (inv.revision_id, revision_id))
1033
if inv.root is None:
1034
raise AssertionError()
1035
return self._add_inventory_checked(revision_id, inv, parents)
1037
def _add_inventory_checked(self, revision_id, inv, parents):
1038
"""Add inv to the repository after checking the inputs.
1040
This function can be overridden to allow different inventory styles.
1042
:seealso: add_inventory, for the contract.
1044
inv_lines = self._serializer.write_inventory_to_lines(inv)
1045
return self._inventory_add_lines(revision_id, parents,
1046
inv_lines, check_content=False)
1048
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1049
parents, basis_inv=None, propagate_caches=False):
1050
"""Add a new inventory expressed as a delta against another revision.
1052
See the inventory developers documentation for the theory behind
1055
:param basis_revision_id: The inventory id the delta was created
1056
against. (This does not have to be a direct parent.)
1057
:param delta: The inventory delta (see Inventory.apply_delta for
1059
:param new_revision_id: The revision id that the inventory is being
1061
:param parents: The revision ids of the parents that revision_id is
1062
known to have and are in the repository already. These are supplied
1063
for repositories that depend on the inventory graph for revision
1064
graph access, as well as for those that pun ancestry with delta
1066
:param basis_inv: The basis inventory if it is already known,
1068
:param propagate_caches: If True, the caches for this inventory are
1069
copied to and updated for the result if possible.
1071
:returns: (validator, new_inv)
1072
The validator(which is a sha1 digest, though what is sha'd is
1073
repository format specific) of the serialized inventory, and the
1074
resulting inventory.
1076
if not self.is_in_write_group():
1077
raise AssertionError("%r not in write group" % (self,))
1078
_mod_revision.check_not_reserved_id(new_revision_id)
1079
basis_tree = self.revision_tree(basis_revision_id)
1080
basis_tree.lock_read()
1082
# Note that this mutates the inventory of basis_tree, which not all
1083
# inventory implementations may support: A better idiom would be to
1084
# return a new inventory, but as there is no revision tree cache in
1085
# repository this is safe for now - RBC 20081013
1086
if basis_inv is None:
1087
basis_inv = basis_tree.inventory
1088
basis_inv.apply_delta(delta)
1089
basis_inv.revision_id = new_revision_id
1090
return (self.add_inventory(new_revision_id, basis_inv, parents),
1095
def _inventory_add_lines(self, revision_id, parents, lines,
1096
check_content=True):
1097
"""Store lines in inv_vf and return the sha1 of the inventory."""
1098
parents = [(parent,) for parent in parents]
1099
result = self.inventories.add_lines((revision_id,), parents, lines,
1100
check_content=check_content)[0]
1101
self.inventories._access.flush()
1104
def add_revision(self, revision_id, rev, inv=None, config=None):
1105
"""Add rev to the revision store as revision_id.
1107
:param revision_id: the revision id to use.
1108
:param rev: The revision object.
1109
:param inv: The inventory for the revision. if None, it will be looked
1110
up in the inventory storer
1111
:param config: If None no digital signature will be created.
1112
If supplied its signature_needed method will be used
1113
to determine if a signature should be made.
1115
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1117
_mod_revision.check_not_reserved_id(revision_id)
1118
if config is not None and config.signature_needed():
1120
inv = self.get_inventory(revision_id)
1121
plaintext = Testament(rev, inv).as_short_text()
1122
self.store_revision_signature(
1123
gpg.GPGStrategy(config), plaintext, revision_id)
1124
# check inventory present
1125
if not self.inventories.get_parent_map([(revision_id,)]):
1127
raise errors.WeaveRevisionNotPresent(revision_id,
1130
# yes, this is not suitable for adding with ghosts.
1131
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1134
key = (revision_id,)
1135
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1136
self._add_revision(rev)
1138
def _add_revision(self, revision):
1139
text = self._serializer.write_revision_to_string(revision)
1140
key = (revision.revision_id,)
1141
parents = tuple((parent,) for parent in revision.parent_ids)
1142
self.revisions.add_lines(key, parents, osutils.split_lines(text))
315
1144
def all_revision_ids(self):
316
1145
"""Returns a list of all the revision ids in the repository.
352
1181
# The old API returned a list, should this actually be a set?
353
1182
return parent_map.keys()
1184
def _check_inventories(self, checker):
1185
"""Check the inventories found from the revision scan.
1187
This is responsible for verifying the sha1 of inventories and
1188
creating a pending_keys set that covers data referenced by inventories.
1190
bar = ui.ui_factory.nested_progress_bar()
1192
self._do_check_inventories(checker, bar)
1196
def _do_check_inventories(self, checker, bar):
1197
"""Helper for _check_inventories."""
1199
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1200
kinds = ['chk_bytes', 'texts']
1201
count = len(checker.pending_keys)
1202
bar.update("inventories", 0, 2)
1203
current_keys = checker.pending_keys
1204
checker.pending_keys = {}
1205
# Accumulate current checks.
1206
for key in current_keys:
1207
if key[0] != 'inventories' and key[0] not in kinds:
1208
checker._report_items.append('unknown key type %r' % (key,))
1209
keys[key[0]].add(key[1:])
1210
if keys['inventories']:
1211
# NB: output order *should* be roughly sorted - topo or
1212
# inverse topo depending on repository - either way decent
1213
# to just delta against. However, pre-CHK formats didn't
1214
# try to optimise inventory layout on disk. As such the
1215
# pre-CHK code path does not use inventory deltas.
1217
for record in self.inventories.check(keys=keys['inventories']):
1218
if record.storage_kind == 'absent':
1219
checker._report_items.append(
1220
'Missing inventory {%s}' % (record.key,))
1222
last_object = self._check_record('inventories', record,
1223
checker, last_object,
1224
current_keys[('inventories',) + record.key])
1225
del keys['inventories']
1228
bar.update("texts", 1)
1229
while (checker.pending_keys or keys['chk_bytes']
1231
# Something to check.
1232
current_keys = checker.pending_keys
1233
checker.pending_keys = {}
1234
# Accumulate current checks.
1235
for key in current_keys:
1236
if key[0] not in kinds:
1237
checker._report_items.append('unknown key type %r' % (key,))
1238
keys[key[0]].add(key[1:])
1239
# Check the outermost kind only - inventories || chk_bytes || texts
1243
for record in getattr(self, kind).check(keys=keys[kind]):
1244
if record.storage_kind == 'absent':
1245
checker._report_items.append(
1246
'Missing %s {%s}' % (kind, record.key,))
1248
last_object = self._check_record(kind, record,
1249
checker, last_object, current_keys[(kind,) + record.key])
1253
def _check_record(self, kind, record, checker, last_object, item_data):
1254
"""Check a single text from this repository."""
1255
if kind == 'inventories':
1256
rev_id = record.key[0]
1257
inv = self._deserialise_inventory(rev_id,
1258
record.get_bytes_as('fulltext'))
1259
if last_object is not None:
1260
delta = inv._make_delta(last_object)
1261
for old_path, path, file_id, ie in delta:
1264
ie.check(checker, rev_id, inv)
1266
for path, ie in inv.iter_entries():
1267
ie.check(checker, rev_id, inv)
1268
if self._format.fast_deltas:
1270
elif kind == 'chk_bytes':
1271
# No code written to check chk_bytes for this repo format.
1272
checker._report_items.append(
1273
'unsupported key type chk_bytes for %s' % (record.key,))
1274
elif kind == 'texts':
1275
self._check_text(record, checker, item_data)
1277
checker._report_items.append(
1278
'unknown key type %s for %s' % (kind, record.key))
1280
def _check_text(self, record, checker, item_data):
1281
"""Check a single text."""
1282
# Check it is extractable.
1283
# TODO: check length.
1284
if record.storage_kind == 'chunked':
1285
chunks = record.get_bytes_as(record.storage_kind)
1286
sha1 = osutils.sha_strings(chunks)
1287
length = sum(map(len, chunks))
1289
content = record.get_bytes_as('fulltext')
1290
sha1 = osutils.sha_string(content)
1291
length = len(content)
1292
if item_data and sha1 != item_data[1]:
1293
checker._report_items.append(
1294
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1295
(record.key, sha1, item_data[1], item_data[2]))
356
1298
def create(a_bzrdir):
357
1299
"""Construct the current default format repository in a_bzrdir."""
921
1982
signature = gpg_strategy.sign(plaintext)
922
1983
self.add_signature_text(revision_id, signature)
924
1986
def add_signature_text(self, revision_id, signature):
925
"""Store a signature text for a revision.
927
:param revision_id: Revision id of the revision
928
:param signature: Signature text.
930
raise NotImplementedError(self.add_signature_text)
1987
self.signatures.add_lines((revision_id,), (),
1988
osutils.split_lines(signature))
1990
def find_text_key_references(self):
1991
"""Find the text key references within the repository.
1993
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1994
to whether they were referred to by the inventory of the
1995
revision_id that they contain. The inventory texts from all present
1996
revision ids are assessed to generate this report.
1998
revision_keys = self.revisions.keys()
1999
w = self.inventories
2000
pb = ui.ui_factory.nested_progress_bar()
2002
return self._find_text_key_references_from_xml_inventory_lines(
2003
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
2007
def _find_text_key_references_from_xml_inventory_lines(self,
2009
"""Core routine for extracting references to texts from inventories.
2011
This performs the translation of xml lines to revision ids.
2013
:param line_iterator: An iterator of lines, origin_version_id
2014
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2015
to whether they were referred to by the inventory of the
2016
revision_id that they contain. Note that if that revision_id was
2017
not part of the line_iterator's output then False will be given -
2018
even though it may actually refer to that key.
2020
if not self._serializer.support_altered_by_hack:
2021
raise AssertionError(
2022
"_find_text_key_references_from_xml_inventory_lines only "
2023
"supported for branches which store inventory as unnested xml"
2024
", not on %r" % self)
2027
# this code needs to read every new line in every inventory for the
2028
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2029
# not present in one of those inventories is unnecessary but not
2030
# harmful because we are filtering by the revision id marker in the
2031
# inventory lines : we only select file ids altered in one of those
2032
# revisions. We don't need to see all lines in the inventory because
2033
# only those added in an inventory in rev X can contain a revision=X
2035
unescape_revid_cache = {}
2036
unescape_fileid_cache = {}
2038
# jam 20061218 In a big fetch, this handles hundreds of thousands
2039
# of lines, so it has had a lot of inlining and optimizing done.
2040
# Sorry that it is a little bit messy.
2041
# Move several functions to be local variables, since this is a long
2043
search = self._file_ids_altered_regex.search
2044
unescape = _unescape_xml
2045
setdefault = result.setdefault
2046
for line, line_key in line_iterator:
2047
match = search(line)
2050
# One call to match.group() returning multiple items is quite a
2051
# bit faster than 2 calls to match.group() each returning 1
2052
file_id, revision_id = match.group('file_id', 'revision_id')
2054
# Inlining the cache lookups helps a lot when you make 170,000
2055
# lines and 350k ids, versus 8.4 unique ids.
2056
# Using a cache helps in 2 ways:
2057
# 1) Avoids unnecessary decoding calls
2058
# 2) Re-uses cached strings, which helps in future set and
2060
# (2) is enough that removing encoding entirely along with
2061
# the cache (so we are using plain strings) results in no
2062
# performance improvement.
2064
revision_id = unescape_revid_cache[revision_id]
2066
unescaped = unescape(revision_id)
2067
unescape_revid_cache[revision_id] = unescaped
2068
revision_id = unescaped
2070
# Note that unconditionally unescaping means that we deserialise
2071
# every fileid, which for general 'pull' is not great, but we don't
2072
# really want to have some many fulltexts that this matters anyway.
2075
file_id = unescape_fileid_cache[file_id]
2077
unescaped = unescape(file_id)
2078
unescape_fileid_cache[file_id] = unescaped
2081
key = (file_id, revision_id)
2082
setdefault(key, False)
2083
if revision_id == line_key[-1]:
2087
def _inventory_xml_lines_for_keys(self, keys):
2088
"""Get a line iterator of the sort needed for findind references.
2090
Not relevant for non-xml inventory repositories.
2092
Ghosts in revision_keys are ignored.
2094
:param revision_keys: The revision keys for the inventories to inspect.
2095
:return: An iterator over (inventory line, revid) for the fulltexts of
2096
all of the xml inventories specified by revision_keys.
2098
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2099
for record in stream:
2100
if record.storage_kind != 'absent':
2101
chunks = record.get_bytes_as('chunked')
2102
revid = record.key[-1]
2103
lines = osutils.chunks_to_lines(chunks)
2107
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2109
"""Helper routine for fileids_altered_by_revision_ids.
2111
This performs the translation of xml lines to revision ids.
2113
:param line_iterator: An iterator of lines, origin_version_id
2114
:param revision_keys: The revision ids to filter for. This should be a
2115
set or other type which supports efficient __contains__ lookups, as
2116
the revision key from each parsed line will be looked up in the
2117
revision_keys filter.
2118
:return: a dictionary mapping altered file-ids to an iterable of
2119
revision_ids. Each altered file-ids has the exact revision_ids that
2120
altered it listed explicitly.
2122
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2123
line_iterator).iterkeys())
2124
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2125
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2126
self._inventory_xml_lines_for_keys(parent_keys)))
2127
new_keys = seen - parent_seen
2129
setdefault = result.setdefault
2130
for key in new_keys:
2131
setdefault(key[0], set()).add(key[-1])
932
2134
def _find_parent_ids_of_revisions(self, revision_ids):
933
2135
"""Find all parent ids that are mentioned in the revision graph.
963
2186
uniquely identify the file version in the caller's context. (Examples:
964
2187
an index number or a TreeTransform trans_id.)
2189
bytes_iterator is an iterable of bytestrings for the file. The
2190
kind of iterable and length of the bytestrings are unspecified, but for
2191
this implementation, it is a list of bytes produced by
2192
VersionedFile.get_record_stream().
966
2194
:param desired_files: a list of (file_id, revision_id, identifier)
969
raise NotImplementedError(self.iter_files_bytes)
2198
for file_id, revision_id, callable_data in desired_files:
2199
text_keys[(file_id, revision_id)] = callable_data
2200
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2201
if record.storage_kind == 'absent':
2202
raise errors.RevisionNotPresent(record.key, self)
2203
yield text_keys[record.key], record.get_bytes_as('chunked')
2205
def _generate_text_key_index(self, text_key_references=None,
2207
"""Generate a new text key index for the repository.
2209
This is an expensive function that will take considerable time to run.
2211
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2212
list of parents, also text keys. When a given key has no parents,
2213
the parents list will be [NULL_REVISION].
2215
# All revisions, to find inventory parents.
2216
if ancestors is None:
2217
graph = self.get_graph()
2218
ancestors = graph.get_parent_map(self.all_revision_ids())
2219
if text_key_references is None:
2220
text_key_references = self.find_text_key_references()
2221
pb = ui.ui_factory.nested_progress_bar()
2223
return self._do_generate_text_key_index(ancestors,
2224
text_key_references, pb)
2228
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2229
"""Helper for _generate_text_key_index to avoid deep nesting."""
2230
revision_order = tsort.topo_sort(ancestors)
2231
invalid_keys = set()
2233
for revision_id in revision_order:
2234
revision_keys[revision_id] = set()
2235
text_count = len(text_key_references)
2236
# a cache of the text keys to allow reuse; costs a dict of all the
2237
# keys, but saves a 2-tuple for every child of a given key.
2239
for text_key, valid in text_key_references.iteritems():
2241
invalid_keys.add(text_key)
2243
revision_keys[text_key[1]].add(text_key)
2244
text_key_cache[text_key] = text_key
2245
del text_key_references
2247
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2248
NULL_REVISION = _mod_revision.NULL_REVISION
2249
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2250
# too small for large or very branchy trees. However, for 55K path
2251
# trees, it would be easy to use too much memory trivially. Ideally we
2252
# could gauge this by looking at available real memory etc, but this is
2253
# always a tricky proposition.
2254
inventory_cache = lru_cache.LRUCache(10)
2255
batch_size = 10 # should be ~150MB on a 55K path tree
2256
batch_count = len(revision_order) / batch_size + 1
2258
pb.update("Calculating text parents", processed_texts, text_count)
2259
for offset in xrange(batch_count):
2260
to_query = revision_order[offset * batch_size:(offset + 1) *
2264
for revision_id in to_query:
2265
parent_ids = ancestors[revision_id]
2266
for text_key in revision_keys[revision_id]:
2267
pb.update("Calculating text parents", processed_texts)
2268
processed_texts += 1
2269
candidate_parents = []
2270
for parent_id in parent_ids:
2271
parent_text_key = (text_key[0], parent_id)
2273
check_parent = parent_text_key not in \
2274
revision_keys[parent_id]
2276
# the parent parent_id is a ghost:
2277
check_parent = False
2278
# truncate the derived graph against this ghost.
2279
parent_text_key = None
2281
# look at the parent commit details inventories to
2282
# determine possible candidates in the per file graph.
2285
inv = inventory_cache[parent_id]
2287
inv = self.revision_tree(parent_id).inventory
2288
inventory_cache[parent_id] = inv
2290
parent_entry = inv[text_key[0]]
2291
except (KeyError, errors.NoSuchId):
2293
if parent_entry is not None:
2295
text_key[0], parent_entry.revision)
2297
parent_text_key = None
2298
if parent_text_key is not None:
2299
candidate_parents.append(
2300
text_key_cache[parent_text_key])
2301
parent_heads = text_graph.heads(candidate_parents)
2302
new_parents = list(parent_heads)
2303
new_parents.sort(key=lambda x:candidate_parents.index(x))
2304
if new_parents == []:
2305
new_parents = [NULL_REVISION]
2306
text_index[text_key] = new_parents
2308
for text_key in invalid_keys:
2309
text_index[text_key] = [NULL_REVISION]
2312
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2313
"""Get an iterable listing the keys of all the data introduced by a set
2316
The keys will be ordered so that the corresponding items can be safely
2317
fetched and inserted in that order.
2319
:returns: An iterable producing tuples of (knit-kind, file-id,
2320
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2321
'revisions'. file-id is None unless knit-kind is 'file'.
2323
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2326
for result in self._find_non_file_keys_to_fetch(revision_ids):
2329
def _find_file_keys_to_fetch(self, revision_ids, pb):
2330
# XXX: it's a bit weird to control the inventory weave caching in this
2331
# generator. Ideally the caching would be done in fetch.py I think. Or
2332
# maybe this generator should explicitly have the contract that it
2333
# should not be iterated until the previously yielded item has been
2335
inv_w = self.inventories
2337
# file ids that changed
2338
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2340
num_file_ids = len(file_ids)
2341
for file_id, altered_versions in file_ids.iteritems():
2343
pb.update("Fetch texts", count, num_file_ids)
2345
yield ("file", file_id, altered_versions)
2347
def _find_non_file_keys_to_fetch(self, revision_ids):
2349
yield ("inventory", None, revision_ids)
2352
# XXX: Note ATM no callers actually pay attention to this return
2353
# instead they just use the list of revision ids and ignore
2354
# missing sigs. Consider removing this work entirely
2355
revisions_with_signatures = set(self.signatures.get_parent_map(
2356
[(r,) for r in revision_ids]))
2357
revisions_with_signatures = set(
2358
[r for (r,) in revisions_with_signatures])
2359
revisions_with_signatures.intersection_update(revision_ids)
2360
yield ("signatures", None, revisions_with_signatures)
2363
yield ("revisions", None, revision_ids)
2366
def get_inventory(self, revision_id):
2367
"""Get Inventory object by revision id."""
2368
return self.iter_inventories([revision_id]).next()
2370
def iter_inventories(self, revision_ids, ordering=None):
2371
"""Get many inventories by revision_ids.
2373
This will buffer some or all of the texts used in constructing the
2374
inventories in memory, but will only parse a single inventory at a
2377
:param revision_ids: The expected revision ids of the inventories.
2378
:param ordering: optional ordering, e.g. 'topological'. If not
2379
specified, the order of revision_ids will be preserved (by
2380
buffering if necessary).
2381
:return: An iterator of inventories.
2383
if ((None in revision_ids)
2384
or (_mod_revision.NULL_REVISION in revision_ids)):
2385
raise ValueError('cannot get null revision inventory')
2386
return self._iter_inventories(revision_ids, ordering)
2388
def _iter_inventories(self, revision_ids, ordering):
2389
"""single-document based inventory iteration."""
2390
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2391
for text, revision_id in inv_xmls:
2392
yield self._deserialise_inventory(revision_id, text)
2394
def _iter_inventory_xmls(self, revision_ids, ordering):
2395
if ordering is None:
2396
order_as_requested = True
2397
ordering = 'unordered'
2399
order_as_requested = False
2400
keys = [(revision_id,) for revision_id in revision_ids]
2403
if order_as_requested:
2404
key_iter = iter(keys)
2405
next_key = key_iter.next()
2406
stream = self.inventories.get_record_stream(keys, ordering, True)
2408
for record in stream:
2409
if record.storage_kind != 'absent':
2410
chunks = record.get_bytes_as('chunked')
2411
if order_as_requested:
2412
text_chunks[record.key] = chunks
2414
yield ''.join(chunks), record.key[-1]
2416
raise errors.NoSuchRevision(self, record.key)
2417
if order_as_requested:
2418
# Yield as many results as we can while preserving order.
2419
while next_key in text_chunks:
2420
chunks = text_chunks.pop(next_key)
2421
yield ''.join(chunks), next_key[-1]
2423
next_key = key_iter.next()
2424
except StopIteration:
2425
# We still want to fully consume the get_record_stream,
2426
# just in case it is not actually finished at this point
2430
def _deserialise_inventory(self, revision_id, xml):
2431
"""Transform the xml into an inventory object.
2433
:param revision_id: The expected revision id of the inventory.
2434
:param xml: A serialised inventory.
2436
result = self._serializer.read_inventory_from_string(xml, revision_id,
2437
entry_cache=self._inventory_entry_cache,
2438
return_from_cache=self._safe_to_return_from_cache)
2439
if result.revision_id != revision_id:
2440
raise AssertionError('revision id mismatch %s != %s' % (
2441
result.revision_id, revision_id))
2444
def get_serializer_format(self):
2445
return self._serializer.format_num
2448
def _get_inventory_xml(self, revision_id):
2449
"""Get serialized inventory as a string."""
2450
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2452
text, revision_id = texts.next()
2453
except StopIteration:
2454
raise errors.HistoryMissing(self, 'inventory', revision_id)
971
2457
def get_rev_id_for_revno(self, revno, known_pair):
972
2458
"""Return the revision id of a revno, given a later (revno, revid)
1289
2802
except UnicodeDecodeError:
1290
2803
raise errors.NonAsciiRevisionId(method, self)
2805
def revision_graph_can_have_wrong_parents(self):
2806
"""Is it possible for this repository to have a revision graph with
2809
If True, then this repository must also implement
2810
_find_inconsistent_revision_parents so that check and reconcile can
2811
check for inconsistencies before proceeding with other checks that may
2812
depend on the revision index being consistent.
2814
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2817
# remove these delegates a while after bzr 0.15
2818
def __make_delegated(name, from_module):
2819
def _deprecated_repository_forwarder():
2820
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2821
% (name, from_module),
2824
m = __import__(from_module, globals(), locals(), [name])
2826
return getattr(m, name)
2827
except AttributeError:
2828
raise AttributeError('module %s has no name %s'
2830
globals()[name] = _deprecated_repository_forwarder
2833
'AllInOneRepository',
2834
'WeaveMetaDirRepository',
2835
'PreSplitOutRepositoryFormat',
2836
'RepositoryFormat4',
2837
'RepositoryFormat5',
2838
'RepositoryFormat6',
2839
'RepositoryFormat7',
2841
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2845
'RepositoryFormatKnit',
2846
'RepositoryFormatKnit1',
2848
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2851
def install_revision(repository, rev, revision_tree):
2852
"""Install all revision data into a repository."""
2853
install_revisions(repository, [(rev, revision_tree, None)])
2856
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2857
"""Install all revision data into a repository.
2859
Accepts an iterable of revision, tree, signature tuples. The signature
2862
repository.start_write_group()
2864
inventory_cache = lru_cache.LRUCache(10)
2865
for n, (revision, revision_tree, signature) in enumerate(iterable):
2866
_install_revision(repository, revision, revision_tree, signature,
2869
pb.update('Transferring revisions', n + 1, num_revisions)
2871
repository.abort_write_group()
2874
repository.commit_write_group()
2877
def _install_revision(repository, rev, revision_tree, signature,
2879
"""Install all revision data into a repository."""
2880
present_parents = []
2882
for p_id in rev.parent_ids:
2883
if repository.has_revision(p_id):
2884
present_parents.append(p_id)
2885
parent_trees[p_id] = repository.revision_tree(p_id)
2887
parent_trees[p_id] = repository.revision_tree(
2888
_mod_revision.NULL_REVISION)
2890
inv = revision_tree.inventory
2891
entries = inv.iter_entries()
2892
# backwards compatibility hack: skip the root id.
2893
if not repository.supports_rich_root():
2894
path, root = entries.next()
2895
if root.revision != rev.revision_id:
2896
raise errors.IncompatibleRevision(repr(repository))
2898
for path, ie in entries:
2899
text_keys[(ie.file_id, ie.revision)] = ie
2900
text_parent_map = repository.texts.get_parent_map(text_keys)
2901
missing_texts = set(text_keys) - set(text_parent_map)
2902
# Add the texts that are not already present
2903
for text_key in missing_texts:
2904
ie = text_keys[text_key]
2906
# FIXME: TODO: The following loop overlaps/duplicates that done by
2907
# commit to determine parents. There is a latent/real bug here where
2908
# the parents inserted are not those commit would do - in particular
2909
# they are not filtered by heads(). RBC, AB
2910
for revision, tree in parent_trees.iteritems():
2911
if ie.file_id not in tree:
2913
parent_id = tree.inventory[ie.file_id].revision
2914
if parent_id in text_parents:
2916
text_parents.append((ie.file_id, parent_id))
2917
lines = revision_tree.get_file(ie.file_id).readlines()
2918
repository.texts.add_lines(text_key, text_parents, lines)
2920
# install the inventory
2921
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2922
# Cache this inventory
2923
inventory_cache[rev.revision_id] = inv
2925
basis_inv = inventory_cache[rev.parent_ids[0]]
2927
repository.add_inventory(rev.revision_id, inv, present_parents)
2929
delta = inv._make_delta(basis_inv)
2930
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2931
rev.revision_id, present_parents)
2933
repository.add_inventory(rev.revision_id, inv, present_parents)
2934
except errors.RevisionAlreadyPresent:
2936
if signature is not None:
2937
repository.add_signature_text(rev.revision_id, signature)
2938
repository.add_revision(rev.revision_id, rev, inv)
1293
2941
class MetaDirRepository(Repository):
1294
2942
"""Repositories in the new meta-dir layout.
1633
3307
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
1634
3308
format_registry.register_lazy(
1635
3309
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
1636
'bzrlib.repofmt.knitpack_repo',
3310
'bzrlib.repofmt.pack_repo',
1637
3311
'RepositoryFormatKnitPack1',
1639
3313
format_registry.register_lazy(
1640
3314
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
1641
'bzrlib.repofmt.knitpack_repo',
3315
'bzrlib.repofmt.pack_repo',
1642
3316
'RepositoryFormatKnitPack3',
1644
3318
format_registry.register_lazy(
1645
3319
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
1646
'bzrlib.repofmt.knitpack_repo',
3320
'bzrlib.repofmt.pack_repo',
1647
3321
'RepositoryFormatKnitPack4',
1649
3323
format_registry.register_lazy(
1650
3324
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
1651
'bzrlib.repofmt.knitpack_repo',
3325
'bzrlib.repofmt.pack_repo',
1652
3326
'RepositoryFormatKnitPack5',
1654
3328
format_registry.register_lazy(
1655
3329
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
1656
'bzrlib.repofmt.knitpack_repo',
3330
'bzrlib.repofmt.pack_repo',
1657
3331
'RepositoryFormatKnitPack5RichRoot',
1659
3333
format_registry.register_lazy(
1660
3334
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
1661
'bzrlib.repofmt.knitpack_repo',
3335
'bzrlib.repofmt.pack_repo',
1662
3336
'RepositoryFormatKnitPack5RichRootBroken',
1664
3338
format_registry.register_lazy(
1665
3339
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
1666
'bzrlib.repofmt.knitpack_repo',
3340
'bzrlib.repofmt.pack_repo',
1667
3341
'RepositoryFormatKnitPack6',
1669
3343
format_registry.register_lazy(
1670
3344
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
1671
'bzrlib.repofmt.knitpack_repo',
3345
'bzrlib.repofmt.pack_repo',
1672
3346
'RepositoryFormatKnitPack6RichRoot',
1674
format_registry.register_lazy(
1675
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
1676
'bzrlib.repofmt.groupcompress_repo',
1677
'RepositoryFormat2a',
1680
3349
# Development formats.
1681
# Check their docstrings to see if/when they are obsolete.
3350
# Obsolete but kept pending a CHK based subtree format.
1682
3351
format_registry.register_lazy(
1683
3352
("Bazaar development format 2 with subtree support "
1684
3353
"(needs bzr.dev from before 1.8)\n"),
1685
'bzrlib.repofmt.knitpack_repo',
3354
'bzrlib.repofmt.pack_repo',
1686
3355
'RepositoryFormatPackDevelopment2Subtree',
1688
format_registry.register_lazy(
1689
'Bazaar development format 8\n',
1690
'bzrlib.repofmt.groupcompress_repo',
1691
'RepositoryFormat2aSubtree',
3358
# 1.14->1.16 go below here
3359
format_registry.register_lazy(
3360
'Bazaar development format - group compression and chk inventory'
3361
' (needs bzr.dev from 1.14)\n',
3362
'bzrlib.repofmt.groupcompress_repo',
3363
'RepositoryFormatCHK1',
3366
format_registry.register_lazy(
3367
'Bazaar development format - chk repository with bencode revision '
3368
'serialization (needs bzr.dev from 1.16)\n',
3369
'bzrlib.repofmt.groupcompress_repo',
3370
'RepositoryFormatCHK2',
3372
format_registry.register_lazy(
3373
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3374
'bzrlib.repofmt.groupcompress_repo',
3375
'RepositoryFormat2a',
1733
3418
:param revision_id: if None all content is copied, if NULL_REVISION no
1734
3419
content is copied.
1737
raise NotImplementedError(self.fetch)
3423
ui.ui_factory.warn_experimental_format_fetch(self)
3424
from bzrlib.fetch import RepoFetcher
3425
# See <https://launchpad.net/bugs/456077> asking for a warning here
3426
if self.source._format.network_name() != self.target._format.network_name():
3427
ui.ui_factory.show_user_warning('cross_format_fetch',
3428
from_format=self.source._format,
3429
to_format=self.target._format)
3430
f = RepoFetcher(to_repository=self.target,
3431
from_repository=self.source,
3432
last_revision=revision_id,
3433
fetch_spec=fetch_spec,
3434
find_ghosts=find_ghosts)
3436
def _walk_to_common_revisions(self, revision_ids):
3437
"""Walk out from revision_ids in source to revisions target has.
3439
:param revision_ids: The start point for the search.
3440
:return: A set of revision ids.
3442
target_graph = self.target.get_graph()
3443
revision_ids = frozenset(revision_ids)
3444
missing_revs = set()
3445
source_graph = self.source.get_graph()
3446
# ensure we don't pay silly lookup costs.
3447
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3448
null_set = frozenset([_mod_revision.NULL_REVISION])
3449
searcher_exhausted = False
3453
# Iterate the searcher until we have enough next_revs
3454
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3456
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3457
next_revs.update(next_revs_part)
3458
ghosts.update(ghosts_part)
3459
except StopIteration:
3460
searcher_exhausted = True
3462
# If there are ghosts in the source graph, and the caller asked for
3463
# them, make sure that they are present in the target.
3464
# We don't care about other ghosts as we can't fetch them and
3465
# haven't been asked to.
3466
ghosts_to_check = set(revision_ids.intersection(ghosts))
3467
revs_to_get = set(next_revs).union(ghosts_to_check)
3469
have_revs = set(target_graph.get_parent_map(revs_to_get))
3470
# we always have NULL_REVISION present.
3471
have_revs = have_revs.union(null_set)
3472
# Check if the target is missing any ghosts we need.
3473
ghosts_to_check.difference_update(have_revs)
3475
# One of the caller's revision_ids is a ghost in both the
3476
# source and the target.
3477
raise errors.NoSuchRevision(
3478
self.source, ghosts_to_check.pop())
3479
missing_revs.update(next_revs - have_revs)
3480
# Because we may have walked past the original stop point, make
3481
# sure everything is stopped
3482
stop_revs = searcher.find_seen_ancestors(have_revs)
3483
searcher.stop_searching_any(stop_revs)
3484
if searcher_exhausted:
3486
return searcher.get_result()
1739
3488
@needs_read_lock
1740
def search_missing_revision_ids(self,
1741
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
1742
find_ghosts=True, revision_ids=None, if_present_ids=None,
3489
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1744
3490
"""Return the revision ids that source has that target does not.
1746
3492
:param revision_id: only return revision ids included by this
1748
:param revision_ids: return revision ids included by these
1749
revision_ids. NoSuchRevision will be raised if any of these
1750
revisions are not present.
1751
:param if_present_ids: like revision_ids, but will not cause
1752
NoSuchRevision if any of these are absent, instead they will simply
1753
not be in the result. This is useful for e.g. finding revisions
1754
to fetch for tags, which may reference absent revisions.
1755
3494
:param find_ghosts: If True find missing revisions in deep history
1756
3495
rather than just finding the surface difference.
1757
:param limit: Maximum number of revisions to return, topologically
1759
3496
:return: A bzrlib.graph.SearchResult.
1761
raise NotImplementedError(self.search_missing_revision_ids)
3498
# stop searching at found target revisions.
3499
if not find_ghosts and revision_id is not None:
3500
return self._walk_to_common_revisions([revision_id])
3501
# generic, possibly worst case, slow code path.
3502
target_ids = set(self.target.all_revision_ids())
3503
if revision_id is not None:
3504
source_ids = self.source.get_ancestry(revision_id)
3505
if source_ids[0] is not None:
3506
raise AssertionError()
3509
source_ids = self.source.all_revision_ids()
3510
result_set = set(source_ids).difference(target_ids)
3511
return self.source.revision_ids_to_search_result(result_set)
1764
3514
def _same_model(source, target):
1785
3535
"different serializers")
3538
class InterSameDataRepository(InterRepository):
3539
"""Code for converting between repositories that represent the same data.
3541
Data format and model must match for this to work.
3545
def _get_repo_format_to_test(self):
3546
"""Repository format for testing with.
3548
InterSameData can pull from subtree to subtree and from non-subtree to
3549
non-subtree, so we test this with the richest repository format.
3551
from bzrlib.repofmt import knitrepo
3552
return knitrepo.RepositoryFormatKnit3()
3555
def is_compatible(source, target):
3556
return InterRepository._same_model(source, target)
3559
class InterWeaveRepo(InterSameDataRepository):
3560
"""Optimised code paths between Weave based repositories.
3562
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3563
implemented lazy inter-object optimisation.
3567
def _get_repo_format_to_test(self):
3568
from bzrlib.repofmt import weaverepo
3569
return weaverepo.RepositoryFormat7()
3572
def is_compatible(source, target):
3573
"""Be compatible with known Weave formats.
3575
We don't test for the stores being of specific types because that
3576
could lead to confusing results, and there is no need to be
3579
from bzrlib.repofmt.weaverepo import (
3585
return (isinstance(source._format, (RepositoryFormat5,
3587
RepositoryFormat7)) and
3588
isinstance(target._format, (RepositoryFormat5,
3590
RepositoryFormat7)))
3591
except AttributeError:
3595
def copy_content(self, revision_id=None):
3596
"""See InterRepository.copy_content()."""
3597
# weave specific optimised path:
3599
self.target.set_make_working_trees(self.source.make_working_trees())
3600
except (errors.RepositoryUpgradeRequired, NotImplemented):
3602
# FIXME do not peek!
3603
if self.source._transport.listable():
3604
pb = ui.ui_factory.nested_progress_bar()
3606
self.target.texts.insert_record_stream(
3607
self.source.texts.get_record_stream(
3608
self.source.texts.keys(), 'topological', False))
3609
pb.update('Copying inventory', 0, 1)
3610
self.target.inventories.insert_record_stream(
3611
self.source.inventories.get_record_stream(
3612
self.source.inventories.keys(), 'topological', False))
3613
self.target.signatures.insert_record_stream(
3614
self.source.signatures.get_record_stream(
3615
self.source.signatures.keys(),
3617
self.target.revisions.insert_record_stream(
3618
self.source.revisions.get_record_stream(
3619
self.source.revisions.keys(),
3620
'topological', True))
3624
self.target.fetch(self.source, revision_id=revision_id)
3627
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3628
"""See InterRepository.missing_revision_ids()."""
3629
# we want all revisions to satisfy revision_id in source.
3630
# but we don't want to stat every file here and there.
3631
# we want then, all revisions other needs to satisfy revision_id
3632
# checked, but not those that we have locally.
3633
# so the first thing is to get a subset of the revisions to
3634
# satisfy revision_id in source, and then eliminate those that
3635
# we do already have.
3636
# this is slow on high latency connection to self, but as this
3637
# disk format scales terribly for push anyway due to rewriting
3638
# inventory.weave, this is considered acceptable.
3640
if revision_id is not None:
3641
source_ids = self.source.get_ancestry(revision_id)
3642
if source_ids[0] is not None:
3643
raise AssertionError()
3646
source_ids = self.source._all_possible_ids()
3647
source_ids_set = set(source_ids)
3648
# source_ids is the worst possible case we may need to pull.
3649
# now we want to filter source_ids against what we actually
3650
# have in target, but don't try to check for existence where we know
3651
# we do not have a revision as that would be pointless.
3652
target_ids = set(self.target._all_possible_ids())
3653
possibly_present_revisions = target_ids.intersection(source_ids_set)
3654
actually_present_revisions = set(
3655
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3656
required_revisions = source_ids_set.difference(actually_present_revisions)
3657
if revision_id is not None:
3658
# we used get_ancestry to determine source_ids then we are assured all
3659
# revisions referenced are present as they are installed in topological order.
3660
# and the tip revision was validated by get_ancestry.
3661
result_set = required_revisions
3663
# if we just grabbed the possibly available ids, then
3664
# we only have an estimate of whats available and need to validate
3665
# that against the revision records.
3667
self.source._eliminate_revisions_not_present(required_revisions))
3668
return self.source.revision_ids_to_search_result(result_set)
3671
class InterKnitRepo(InterSameDataRepository):
3672
"""Optimised code paths between Knit based repositories."""
3675
def _get_repo_format_to_test(self):
3676
from bzrlib.repofmt import knitrepo
3677
return knitrepo.RepositoryFormatKnit1()
3680
def is_compatible(source, target):
3681
"""Be compatible with known Knit formats.
3683
We don't test for the stores being of specific types because that
3684
could lead to confusing results, and there is no need to be
3687
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3689
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3690
isinstance(target._format, RepositoryFormatKnit))
3691
except AttributeError:
3693
return are_knits and InterRepository._same_model(source, target)
3696
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3697
"""See InterRepository.missing_revision_ids()."""
3698
if revision_id is not None:
3699
source_ids = self.source.get_ancestry(revision_id)
3700
if source_ids[0] is not None:
3701
raise AssertionError()
3704
source_ids = self.source.all_revision_ids()
3705
source_ids_set = set(source_ids)
3706
# source_ids is the worst possible case we may need to pull.
3707
# now we want to filter source_ids against what we actually
3708
# have in target, but don't try to check for existence where we know
3709
# we do not have a revision as that would be pointless.
3710
target_ids = set(self.target.all_revision_ids())
3711
possibly_present_revisions = target_ids.intersection(source_ids_set)
3712
actually_present_revisions = set(
3713
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3714
required_revisions = source_ids_set.difference(actually_present_revisions)
3715
if revision_id is not None:
3716
# we used get_ancestry to determine source_ids then we are assured all
3717
# revisions referenced are present as they are installed in topological order.
3718
# and the tip revision was validated by get_ancestry.
3719
result_set = required_revisions
3721
# if we just grabbed the possibly available ids, then
3722
# we only have an estimate of whats available and need to validate
3723
# that against the revision records.
3725
self.source._eliminate_revisions_not_present(required_revisions))
3726
return self.source.revision_ids_to_search_result(result_set)
3729
class InterDifferingSerializer(InterRepository):
3732
def _get_repo_format_to_test(self):
3736
def is_compatible(source, target):
3737
"""Be compatible with Knit2 source and Knit3 target"""
3738
# This is redundant with format.check_conversion_target(), however that
3739
# raises an exception, and we just want to say "False" as in we won't
3740
# support converting between these formats.
3741
if 'IDS_never' in debug.debug_flags:
3743
if source.supports_rich_root() and not target.supports_rich_root():
3745
if (source._format.supports_tree_reference
3746
and not target._format.supports_tree_reference):
3748
if target._fallback_repositories and target._format.supports_chks:
3749
# IDS doesn't know how to copy CHKs for the parent inventories it
3750
# adds to stacked repos.
3752
if 'IDS_always' in debug.debug_flags:
3754
# Only use this code path for local source and target. IDS does far
3755
# too much IO (both bandwidth and roundtrips) over a network.
3756
if not source.bzrdir.transport.base.startswith('file:///'):
3758
if not target.bzrdir.transport.base.startswith('file:///'):
3762
def _get_trees(self, revision_ids, cache):
3764
for rev_id in revision_ids:
3766
possible_trees.append((rev_id, cache[rev_id]))
3768
# Not cached, but inventory might be present anyway.
3770
tree = self.source.revision_tree(rev_id)
3771
except errors.NoSuchRevision:
3772
# Nope, parent is ghost.
3775
cache[rev_id] = tree
3776
possible_trees.append((rev_id, tree))
3777
return possible_trees
3779
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3780
"""Get the best delta and base for this revision.
3782
:return: (basis_id, delta)
3785
# Generate deltas against each tree, to find the shortest.
3786
texts_possibly_new_in_tree = set()
3787
for basis_id, basis_tree in possible_trees:
3788
delta = tree.inventory._make_delta(basis_tree.inventory)
3789
for old_path, new_path, file_id, new_entry in delta:
3790
if new_path is None:
3791
# This file_id isn't present in the new rev, so we don't
3795
# Rich roots are handled elsewhere...
3797
kind = new_entry.kind
3798
if kind != 'directory' and kind != 'file':
3799
# No text record associated with this inventory entry.
3801
# This is a directory or file that has changed somehow.
3802
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3803
deltas.append((len(delta), basis_id, delta))
3805
return deltas[0][1:]
3807
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3808
"""Find all parent revisions that are absent, but for which the
3809
inventory is present, and copy those inventories.
3811
This is necessary to preserve correctness when the source is stacked
3812
without fallbacks configured. (Note that in cases like upgrade the
3813
source may be not have _fallback_repositories even though it is
3817
for parents in parent_map.values():
3818
parent_revs.update(parents)
3819
present_parents = self.source.get_parent_map(parent_revs)
3820
absent_parents = set(parent_revs).difference(present_parents)
3821
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3822
(rev_id,) for rev_id in absent_parents)
3823
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3824
for parent_tree in self.source.revision_trees(parent_inv_ids):
3825
current_revision_id = parent_tree.get_revision_id()
3826
parents_parents_keys = parent_invs_keys_for_stacking[
3827
(current_revision_id,)]
3828
parents_parents = [key[-1] for key in parents_parents_keys]
3829
basis_id = _mod_revision.NULL_REVISION
3830
basis_tree = self.source.revision_tree(basis_id)
3831
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3832
self.target.add_inventory_by_delta(
3833
basis_id, delta, current_revision_id, parents_parents)
3834
cache[current_revision_id] = parent_tree
3836
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3837
"""Fetch across a few revisions.
3839
:param revision_ids: The revisions to copy
3840
:param basis_id: The revision_id of a tree that must be in cache, used
3841
as a basis for delta when no other base is available
3842
:param cache: A cache of RevisionTrees that we can use.
3843
:param a_graph: A Graph object to determine the heads() of the
3844
rich-root data stream.
3845
:return: The revision_id of the last converted tree. The RevisionTree
3846
for it will be in cache
3848
# Walk though all revisions; get inventory deltas, copy referenced
3849
# texts that delta references, insert the delta, revision and
3851
root_keys_to_create = set()
3854
pending_revisions = []
3855
parent_map = self.source.get_parent_map(revision_ids)
3856
self._fetch_parent_invs_for_stacking(parent_map, cache)
3857
self.source._safe_to_return_from_cache = True
3858
for tree in self.source.revision_trees(revision_ids):
3859
# Find a inventory delta for this revision.
3860
# Find text entries that need to be copied, too.
3861
current_revision_id = tree.get_revision_id()
3862
parent_ids = parent_map.get(current_revision_id, ())
3863
parent_trees = self._get_trees(parent_ids, cache)
3864
possible_trees = list(parent_trees)
3865
if len(possible_trees) == 0:
3866
# There either aren't any parents, or the parents are ghosts,
3867
# so just use the last converted tree.
3868
possible_trees.append((basis_id, cache[basis_id]))
3869
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3871
revision = self.source.get_revision(current_revision_id)
3872
pending_deltas.append((basis_id, delta,
3873
current_revision_id, revision.parent_ids))
3874
if self._converting_to_rich_root:
3875
self._revision_id_to_root_id[current_revision_id] = \
3877
# Determine which texts are in present in this revision but not in
3878
# any of the available parents.
3879
texts_possibly_new_in_tree = set()
3880
for old_path, new_path, file_id, entry in delta:
3881
if new_path is None:
3882
# This file_id isn't present in the new rev
3886
if not self.target.supports_rich_root():
3887
# The target doesn't support rich root, so we don't
3890
if self._converting_to_rich_root:
3891
# This can't be copied normally, we have to insert
3893
root_keys_to_create.add((file_id, entry.revision))
3896
texts_possibly_new_in_tree.add((file_id, entry.revision))
3897
for basis_id, basis_tree in possible_trees:
3898
basis_inv = basis_tree.inventory
3899
for file_key in list(texts_possibly_new_in_tree):
3900
file_id, file_revision = file_key
3902
entry = basis_inv[file_id]
3903
except errors.NoSuchId:
3905
if entry.revision == file_revision:
3906
texts_possibly_new_in_tree.remove(file_key)
3907
text_keys.update(texts_possibly_new_in_tree)
3908
pending_revisions.append(revision)
3909
cache[current_revision_id] = tree
3910
basis_id = current_revision_id
3911
self.source._safe_to_return_from_cache = False
3913
from_texts = self.source.texts
3914
to_texts = self.target.texts
3915
if root_keys_to_create:
3916
root_stream = _mod_fetch._new_root_data_stream(
3917
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3918
self.source, graph=a_graph)
3919
to_texts.insert_record_stream(root_stream)
3920
to_texts.insert_record_stream(from_texts.get_record_stream(
3921
text_keys, self.target._format._fetch_order,
3922
not self.target._format._fetch_uses_deltas))
3923
# insert inventory deltas
3924
for delta in pending_deltas:
3925
self.target.add_inventory_by_delta(*delta)
3926
if self.target._fallback_repositories:
3927
# Make sure this stacked repository has all the parent inventories
3928
# for the new revisions that we are about to insert. We do this
3929
# before adding the revisions so that no revision is added until
3930
# all the inventories it may depend on are added.
3931
# Note that this is overzealous, as we may have fetched these in an
3934
revision_ids = set()
3935
for revision in pending_revisions:
3936
revision_ids.add(revision.revision_id)
3937
parent_ids.update(revision.parent_ids)
3938
parent_ids.difference_update(revision_ids)
3939
parent_ids.discard(_mod_revision.NULL_REVISION)
3940
parent_map = self.source.get_parent_map(parent_ids)
3941
# we iterate over parent_map and not parent_ids because we don't
3942
# want to try copying any revision which is a ghost
3943
for parent_tree in self.source.revision_trees(parent_map):
3944
current_revision_id = parent_tree.get_revision_id()
3945
parents_parents = parent_map[current_revision_id]
3946
possible_trees = self._get_trees(parents_parents, cache)
3947
if len(possible_trees) == 0:
3948
# There either aren't any parents, or the parents are
3949
# ghosts, so just use the last converted tree.
3950
possible_trees.append((basis_id, cache[basis_id]))
3951
basis_id, delta = self._get_delta_for_revision(parent_tree,
3952
parents_parents, possible_trees)
3953
self.target.add_inventory_by_delta(
3954
basis_id, delta, current_revision_id, parents_parents)
3955
# insert signatures and revisions
3956
for revision in pending_revisions:
3958
signature = self.source.get_signature_text(
3959
revision.revision_id)
3960
self.target.add_signature_text(revision.revision_id,
3962
except errors.NoSuchRevision:
3964
self.target.add_revision(revision.revision_id, revision)
3967
def _fetch_all_revisions(self, revision_ids, pb):
3968
"""Fetch everything for the list of revisions.
3970
:param revision_ids: The list of revisions to fetch. Must be in
3972
:param pb: A ProgressTask
3975
basis_id, basis_tree = self._get_basis(revision_ids[0])
3977
cache = lru_cache.LRUCache(100)
3978
cache[basis_id] = basis_tree
3979
del basis_tree # We don't want to hang on to it here
3981
if self._converting_to_rich_root and len(revision_ids) > 100:
3982
a_graph = _mod_fetch._get_rich_root_heads_graph(self.source,
3987
for offset in range(0, len(revision_ids), batch_size):
3988
self.target.start_write_group()
3990
pb.update('Transferring revisions', offset,
3992
batch = revision_ids[offset:offset+batch_size]
3993
basis_id = self._fetch_batch(batch, basis_id, cache,
3996
self.source._safe_to_return_from_cache = False
3997
self.target.abort_write_group()
4000
hint = self.target.commit_write_group()
4003
if hints and self.target._format.pack_compresses:
4004
self.target.pack(hint=hints)
4005
pb.update('Transferring revisions', len(revision_ids),
4009
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
4011
"""See InterRepository.fetch()."""
4012
if fetch_spec is not None:
4013
raise AssertionError("Not implemented yet...")
4014
ui.ui_factory.warn_experimental_format_fetch(self)
4015
if (not self.source.supports_rich_root()
4016
and self.target.supports_rich_root()):
4017
self._converting_to_rich_root = True
4018
self._revision_id_to_root_id = {}
4020
self._converting_to_rich_root = False
4021
# See <https://launchpad.net/bugs/456077> asking for a warning here
4022
if self.source._format.network_name() != self.target._format.network_name():
4023
ui.ui_factory.show_user_warning('cross_format_fetch',
4024
from_format=self.source._format,
4025
to_format=self.target._format)
4026
revision_ids = self.target.search_missing_revision_ids(self.source,
4027
revision_id, find_ghosts=find_ghosts).get_keys()
4028
if not revision_ids:
4030
revision_ids = tsort.topo_sort(
4031
self.source.get_graph().get_parent_map(revision_ids))
4032
if not revision_ids:
4034
# Walk though all revisions; get inventory deltas, copy referenced
4035
# texts that delta references, insert the delta, revision and
4038
my_pb = ui.ui_factory.nested_progress_bar()
4041
symbol_versioning.warn(
4042
symbol_versioning.deprecated_in((1, 14, 0))
4043
% "pb parameter to fetch()")
4046
self._fetch_all_revisions(revision_ids, pb)
4048
if my_pb is not None:
4050
return len(revision_ids), 0
4052
def _get_basis(self, first_revision_id):
4053
"""Get a revision and tree which exists in the target.
4055
This assumes that first_revision_id is selected for transmission
4056
because all other ancestors are already present. If we can't find an
4057
ancestor we fall back to NULL_REVISION since we know that is safe.
4059
:return: (basis_id, basis_tree)
4061
first_rev = self.source.get_revision(first_revision_id)
4063
basis_id = first_rev.parent_ids[0]
4064
# only valid as a basis if the target has it
4065
self.target.get_revision(basis_id)
4066
# Try to get a basis tree - if its a ghost it will hit the
4067
# NoSuchRevision case.
4068
basis_tree = self.source.revision_tree(basis_id)
4069
except (IndexError, errors.NoSuchRevision):
4070
basis_id = _mod_revision.NULL_REVISION
4071
basis_tree = self.source.revision_tree(basis_id)
4072
return basis_id, basis_tree
4075
InterRepository.register_optimiser(InterDifferingSerializer)
4076
InterRepository.register_optimiser(InterSameDataRepository)
4077
InterRepository.register_optimiser(InterWeaveRepo)
4078
InterRepository.register_optimiser(InterKnitRepo)
1788
4081
class CopyConverter(object):
1789
4082
"""A repository conversion tool which just performs a copy of the content.
1844
4239
return revision_graph
4242
class StreamSink(object):
4243
"""An object that can insert a stream into a repository.
4245
This interface handles the complexity of reserialising inventories and
4246
revisions from different formats, and allows unidirectional insertion into
4247
stacked repositories without looking for the missing basis parents
4251
def __init__(self, target_repo):
4252
self.target_repo = target_repo
4254
def insert_stream(self, stream, src_format, resume_tokens):
4255
"""Insert a stream's content into the target repository.
4257
:param src_format: a bzr repository format.
4259
:return: a list of resume tokens and an iterable of keys additional
4260
items required before the insertion can be completed.
4262
self.target_repo.lock_write()
4265
self.target_repo.resume_write_group(resume_tokens)
4268
self.target_repo.start_write_group()
4271
# locked_insert_stream performs a commit|suspend.
4272
return self._locked_insert_stream(stream, src_format, is_resume)
4274
self.target_repo.abort_write_group(suppress_errors=True)
4277
self.target_repo.unlock()
4279
def _locked_insert_stream(self, stream, src_format, is_resume):
4280
to_serializer = self.target_repo._format._serializer
4281
src_serializer = src_format._serializer
4283
if to_serializer == src_serializer:
4284
# If serializers match and the target is a pack repository, set the
4285
# write cache size on the new pack. This avoids poor performance
4286
# on transports where append is unbuffered (such as
4287
# RemoteTransport). This is safe to do because nothing should read
4288
# back from the target repository while a stream with matching
4289
# serialization is being inserted.
4290
# The exception is that a delta record from the source that should
4291
# be a fulltext may need to be expanded by the target (see
4292
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4293
# explicitly flush any buffered writes first in that rare case.
4295
new_pack = self.target_repo._pack_collection._new_pack
4296
except AttributeError:
4297
# Not a pack repository
4300
new_pack.set_write_cache_size(1024*1024)
4301
for substream_type, substream in stream:
4302
if 'stream' in debug.debug_flags:
4303
mutter('inserting substream: %s', substream_type)
4304
if substream_type == 'texts':
4305
self.target_repo.texts.insert_record_stream(substream)
4306
elif substream_type == 'inventories':
4307
if src_serializer == to_serializer:
4308
self.target_repo.inventories.insert_record_stream(
4311
self._extract_and_insert_inventories(
4312
substream, src_serializer)
4313
elif substream_type == 'inventory-deltas':
4314
self._extract_and_insert_inventory_deltas(
4315
substream, src_serializer)
4316
elif substream_type == 'chk_bytes':
4317
# XXX: This doesn't support conversions, as it assumes the
4318
# conversion was done in the fetch code.
4319
self.target_repo.chk_bytes.insert_record_stream(substream)
4320
elif substream_type == 'revisions':
4321
# This may fallback to extract-and-insert more often than
4322
# required if the serializers are different only in terms of
4324
if src_serializer == to_serializer:
4325
self.target_repo.revisions.insert_record_stream(
4328
self._extract_and_insert_revisions(substream,
4330
elif substream_type == 'signatures':
4331
self.target_repo.signatures.insert_record_stream(substream)
4333
raise AssertionError('kaboom! %s' % (substream_type,))
4334
# Done inserting data, and the missing_keys calculations will try to
4335
# read back from the inserted data, so flush the writes to the new pack
4336
# (if this is pack format).
4337
if new_pack is not None:
4338
new_pack._write_data('', flush=True)
4339
# Find all the new revisions (including ones from resume_tokens)
4340
missing_keys = self.target_repo.get_missing_parent_inventories(
4341
check_for_missing_texts=is_resume)
4343
for prefix, versioned_file in (
4344
('texts', self.target_repo.texts),
4345
('inventories', self.target_repo.inventories),
4346
('revisions', self.target_repo.revisions),
4347
('signatures', self.target_repo.signatures),
4348
('chk_bytes', self.target_repo.chk_bytes),
4350
if versioned_file is None:
4352
# TODO: key is often going to be a StaticTuple object
4353
# I don't believe we can define a method by which
4354
# (prefix,) + StaticTuple will work, though we could
4355
# define a StaticTuple.sq_concat that would allow you to
4356
# pass in either a tuple or a StaticTuple as the second
4357
# object, so instead we could have:
4358
# StaticTuple(prefix) + key here...
4359
missing_keys.update((prefix,) + key for key in
4360
versioned_file.get_missing_compression_parent_keys())
4361
except NotImplementedError:
4362
# cannot even attempt suspending, and missing would have failed
4363
# during stream insertion.
4364
missing_keys = set()
4367
# suspend the write group and tell the caller what we is
4368
# missing. We know we can suspend or else we would not have
4369
# entered this code path. (All repositories that can handle
4370
# missing keys can handle suspending a write group).
4371
write_group_tokens = self.target_repo.suspend_write_group()
4372
return write_group_tokens, missing_keys
4373
hint = self.target_repo.commit_write_group()
4374
if (to_serializer != src_serializer and
4375
self.target_repo._format.pack_compresses):
4376
self.target_repo.pack(hint=hint)
4379
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4380
target_rich_root = self.target_repo._format.rich_root_data
4381
target_tree_refs = self.target_repo._format.supports_tree_reference
4382
for record in substream:
4383
# Insert the delta directly
4384
inventory_delta_bytes = record.get_bytes_as('fulltext')
4385
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4387
parse_result = deserialiser.parse_text_bytes(
4388
inventory_delta_bytes)
4389
except inventory_delta.IncompatibleInventoryDelta, err:
4390
trace.mutter("Incompatible delta: %s", err.msg)
4391
raise errors.IncompatibleRevision(self.target_repo._format)
4392
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4393
revision_id = new_id
4394
parents = [key[0] for key in record.parents]
4395
self.target_repo.add_inventory_by_delta(
4396
basis_id, inv_delta, revision_id, parents)
4398
def _extract_and_insert_inventories(self, substream, serializer,
4400
"""Generate a new inventory versionedfile in target, converting data.
4402
The inventory is retrieved from the source, (deserializing it), and
4403
stored in the target (reserializing it in a different format).
4405
target_rich_root = self.target_repo._format.rich_root_data
4406
target_tree_refs = self.target_repo._format.supports_tree_reference
4407
for record in substream:
4408
# It's not a delta, so it must be a fulltext in the source
4409
# serializer's format.
4410
bytes = record.get_bytes_as('fulltext')
4411
revision_id = record.key[0]
4412
inv = serializer.read_inventory_from_string(bytes, revision_id)
4413
parents = [key[0] for key in record.parents]
4414
self.target_repo.add_inventory(revision_id, inv, parents)
4415
# No need to keep holding this full inv in memory when the rest of
4416
# the substream is likely to be all deltas.
4419
def _extract_and_insert_revisions(self, substream, serializer):
4420
for record in substream:
4421
bytes = record.get_bytes_as('fulltext')
4422
revision_id = record.key[0]
4423
rev = serializer.read_revision_from_string(bytes)
4424
if rev.revision_id != revision_id:
4425
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4426
self.target_repo.add_revision(revision_id, rev)
4429
if self.target_repo._format._fetch_reconcile:
4430
self.target_repo.reconcile()
4433
class StreamSource(object):
4434
"""A source of a stream for fetching between repositories."""
4436
def __init__(self, from_repository, to_format):
4437
"""Create a StreamSource streaming from from_repository."""
4438
self.from_repository = from_repository
4439
self.to_format = to_format
4441
def delta_on_metadata(self):
4442
"""Return True if delta's are permitted on metadata streams.
4444
That is on revisions and signatures.
4446
src_serializer = self.from_repository._format._serializer
4447
target_serializer = self.to_format._serializer
4448
return (self.to_format._fetch_uses_deltas and
4449
src_serializer == target_serializer)
4451
def _fetch_revision_texts(self, revs):
4452
# fetch signatures first and then the revision texts
4453
# may need to be a InterRevisionStore call here.
4454
from_sf = self.from_repository.signatures
4455
# A missing signature is just skipped.
4456
keys = [(rev_id,) for rev_id in revs]
4457
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4459
self.to_format._fetch_order,
4460
not self.to_format._fetch_uses_deltas))
4461
# If a revision has a delta, this is actually expanded inside the
4462
# insert_record_stream code now, which is an alternate fix for
4464
from_rf = self.from_repository.revisions
4465
revisions = from_rf.get_record_stream(
4467
self.to_format._fetch_order,
4468
not self.delta_on_metadata())
4469
return [('signatures', signatures), ('revisions', revisions)]
4471
def _generate_root_texts(self, revs):
4472
"""This will be called by get_stream between fetching weave texts and
4473
fetching the inventory weave.
4475
if self._rich_root_upgrade():
4476
return _mod_fetch.Inter1and2Helper(
4477
self.from_repository).generate_root_texts(revs)
4481
def get_stream(self, search):
4483
revs = search.get_keys()
4484
graph = self.from_repository.get_graph()
4485
revs = tsort.topo_sort(graph.get_parent_map(revs))
4486
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4488
for knit_kind, file_id, revisions in data_to_fetch:
4489
if knit_kind != phase:
4491
# Make a new progress bar for this phase
4492
if knit_kind == "file":
4493
# Accumulate file texts
4494
text_keys.extend([(file_id, revision) for revision in
4496
elif knit_kind == "inventory":
4497
# Now copy the file texts.
4498
from_texts = self.from_repository.texts
4499
yield ('texts', from_texts.get_record_stream(
4500
text_keys, self.to_format._fetch_order,
4501
not self.to_format._fetch_uses_deltas))
4502
# Cause an error if a text occurs after we have done the
4505
# Before we process the inventory we generate the root
4506
# texts (if necessary) so that the inventories references
4508
for _ in self._generate_root_texts(revs):
4510
# we fetch only the referenced inventories because we do not
4511
# know for unselected inventories whether all their required
4512
# texts are present in the other repository - it could be
4514
for info in self._get_inventory_stream(revs):
4516
elif knit_kind == "signatures":
4517
# Nothing to do here; this will be taken care of when
4518
# _fetch_revision_texts happens.
4520
elif knit_kind == "revisions":
4521
for record in self._fetch_revision_texts(revs):
4524
raise AssertionError("Unknown knit kind %r" % knit_kind)
4526
def get_stream_for_missing_keys(self, missing_keys):
4527
# missing keys can only occur when we are byte copying and not
4528
# translating (because translation means we don't send
4529
# unreconstructable deltas ever).
4531
keys['texts'] = set()
4532
keys['revisions'] = set()
4533
keys['inventories'] = set()
4534
keys['chk_bytes'] = set()
4535
keys['signatures'] = set()
4536
for key in missing_keys:
4537
keys[key[0]].add(key[1:])
4538
if len(keys['revisions']):
4539
# If we allowed copying revisions at this point, we could end up
4540
# copying a revision without copying its required texts: a
4541
# violation of the requirements for repository integrity.
4542
raise AssertionError(
4543
'cannot copy revisions to fill in missing deltas %s' % (
4544
keys['revisions'],))
4545
for substream_kind, keys in keys.iteritems():
4546
vf = getattr(self.from_repository, substream_kind)
4547
if vf is None and keys:
4548
raise AssertionError(
4549
"cannot fill in keys for a versioned file we don't"
4550
" have: %s needs %s" % (substream_kind, keys))
4552
# No need to stream something we don't have
4554
if substream_kind == 'inventories':
4555
# Some missing keys are genuinely ghosts, filter those out.
4556
present = self.from_repository.inventories.get_parent_map(keys)
4557
revs = [key[0] for key in present]
4558
# Get the inventory stream more-or-less as we do for the
4559
# original stream; there's no reason to assume that records
4560
# direct from the source will be suitable for the sink. (Think
4561
# e.g. 2a -> 1.9-rich-root).
4562
for info in self._get_inventory_stream(revs, missing=True):
4566
# Ask for full texts always so that we don't need more round trips
4567
# after this stream.
4568
# Some of the missing keys are genuinely ghosts, so filter absent
4569
# records. The Sink is responsible for doing another check to
4570
# ensure that ghosts don't introduce missing data for future
4572
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4573
self.to_format._fetch_order, True))
4574
yield substream_kind, stream
4576
def inventory_fetch_order(self):
4577
if self._rich_root_upgrade():
4578
return 'topological'
4580
return self.to_format._fetch_order
4582
def _rich_root_upgrade(self):
4583
return (not self.from_repository._format.rich_root_data and
4584
self.to_format.rich_root_data)
4586
def _get_inventory_stream(self, revision_ids, missing=False):
4587
from_format = self.from_repository._format
4588
if (from_format.supports_chks and self.to_format.supports_chks and
4589
from_format.network_name() == self.to_format.network_name()):
4590
raise AssertionError(
4591
"this case should be handled by GroupCHKStreamSource")
4592
elif 'forceinvdeltas' in debug.debug_flags:
4593
return self._get_convertable_inventory_stream(revision_ids,
4594
delta_versus_null=missing)
4595
elif from_format.network_name() == self.to_format.network_name():
4597
return self._get_simple_inventory_stream(revision_ids,
4599
elif (not from_format.supports_chks and not self.to_format.supports_chks
4600
and from_format._serializer == self.to_format._serializer):
4601
# Essentially the same format.
4602
return self._get_simple_inventory_stream(revision_ids,
4605
# Any time we switch serializations, we want to use an
4606
# inventory-delta based approach.
4607
return self._get_convertable_inventory_stream(revision_ids,
4608
delta_versus_null=missing)
4610
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4611
# NB: This currently reopens the inventory weave in source;
4612
# using a single stream interface instead would avoid this.
4613
from_weave = self.from_repository.inventories
4615
delta_closure = True
4617
delta_closure = not self.delta_on_metadata()
4618
yield ('inventories', from_weave.get_record_stream(
4619
[(rev_id,) for rev_id in revision_ids],
4620
self.inventory_fetch_order(), delta_closure))
4622
def _get_convertable_inventory_stream(self, revision_ids,
4623
delta_versus_null=False):
4624
# The two formats are sufficiently different that there is no fast
4625
# path, so we need to send just inventorydeltas, which any
4626
# sufficiently modern client can insert into any repository.
4627
# The StreamSink code expects to be able to
4628
# convert on the target, so we need to put bytes-on-the-wire that can
4629
# be converted. That means inventory deltas (if the remote is <1.19,
4630
# RemoteStreamSink will fallback to VFS to insert the deltas).
4631
yield ('inventory-deltas',
4632
self._stream_invs_as_deltas(revision_ids,
4633
delta_versus_null=delta_versus_null))
4635
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4636
"""Return a stream of inventory-deltas for the given rev ids.
4638
:param revision_ids: The list of inventories to transmit
4639
:param delta_versus_null: Don't try to find a minimal delta for this
4640
entry, instead compute the delta versus the NULL_REVISION. This
4641
effectively streams a complete inventory. Used for stuff like
4642
filling in missing parents, etc.
4644
from_repo = self.from_repository
4645
revision_keys = [(rev_id,) for rev_id in revision_ids]
4646
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4647
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4649
inventories = self.from_repository.iter_inventories(
4650
revision_ids, 'topological')
4651
format = from_repo._format
4652
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4653
inventory_cache = lru_cache.LRUCache(50)
4654
null_inventory = from_repo.revision_tree(
4655
_mod_revision.NULL_REVISION).inventory
4656
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4657
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4658
# repo back into a non-rich-root repo ought to be allowed)
4659
serializer = inventory_delta.InventoryDeltaSerializer(
4660
versioned_root=format.rich_root_data,
4661
tree_references=format.supports_tree_reference)
4662
for inv in inventories:
4663
key = (inv.revision_id,)
4664
parent_keys = parent_map.get(key, ())
4666
if not delta_versus_null and parent_keys:
4667
# The caller did not ask for complete inventories and we have
4668
# some parents that we can delta against. Make a delta against
4669
# each parent so that we can find the smallest.
4670
parent_ids = [parent_key[0] for parent_key in parent_keys]
4671
for parent_id in parent_ids:
4672
if parent_id not in invs_sent_so_far:
4673
# We don't know that the remote side has this basis, so
4676
if parent_id == _mod_revision.NULL_REVISION:
4677
parent_inv = null_inventory
4679
parent_inv = inventory_cache.get(parent_id, None)
4680
if parent_inv is None:
4681
parent_inv = from_repo.get_inventory(parent_id)
4682
candidate_delta = inv._make_delta(parent_inv)
4683
if (delta is None or
4684
len(delta) > len(candidate_delta)):
4685
delta = candidate_delta
4686
basis_id = parent_id
4688
# Either none of the parents ended up being suitable, or we
4689
# were asked to delta against NULL
4690
basis_id = _mod_revision.NULL_REVISION
4691
delta = inv._make_delta(null_inventory)
4692
invs_sent_so_far.add(inv.revision_id)
4693
inventory_cache[inv.revision_id] = inv
4694
delta_serialized = ''.join(
4695
serializer.delta_to_lines(basis_id, key[-1], delta))
4696
yield versionedfile.FulltextContentFactory(
4697
key, parent_keys, None, delta_serialized)
1847
4700
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
1848
4701
stop_revision=None):
1849
4702
"""Extend the partial history to include a given index