391
201
commit to be valid, deletes against the basis MUST be recorded via
392
202
builder.record_delete().
394
self._recording_deletes = True
396
basis_id = self.parents[0]
398
basis_id = _mod_revision.NULL_REVISION
399
self.basis_delta_revision = basis_id
401
def record_entry_contents(self, ie, parent_invs, path, tree,
403
"""Record the content of ie from tree into the commit if needed.
405
Side effect: sets ie.revision when unchanged
407
:param ie: An inventory entry present in the commit.
408
:param parent_invs: The inventories of the parent revisions of the
410
:param path: The path the entry is at in the tree.
411
:param tree: The tree which contains this entry and should be used to
413
:param content_summary: Summary data from the tree about the paths
414
content - stat, length, exec, sha/link target. This is only
415
accessed when the entry has a revision of None - that is when it is
416
a candidate to commit.
417
:return: A tuple (change_delta, version_recorded, fs_hash).
418
change_delta is an inventory_delta change for this entry against
419
the basis tree of the commit, or None if no change occured against
421
version_recorded is True if a new version of the entry has been
422
recorded. For instance, committing a merge where a file was only
423
changed on the other side will return (delta, False).
424
fs_hash is either None, or the hash details for the path (currently
425
a tuple of the contents sha1 and the statvalue returned by
426
tree.get_file_with_stat()).
428
if self.new_inventory.root is None:
429
if ie.parent_id is not None:
430
raise errors.RootMissing()
431
self._check_root(ie, parent_invs, tree)
432
if ie.revision is None:
433
kind = content_summary[0]
435
# ie is carried over from a prior commit
437
# XXX: repository specific check for nested tree support goes here - if
438
# the repo doesn't want nested trees we skip it ?
439
if (kind == 'tree-reference' and
440
not self.repository._format.supports_tree_reference):
441
# mismatch between commit builder logic and repository:
442
# this needs the entry creation pushed down into the builder.
443
raise NotImplementedError('Missing repository subtree support.')
444
self.new_inventory.add(ie)
446
# TODO: slow, take it out of the inner loop.
448
basis_inv = parent_invs[0]
450
basis_inv = Inventory(root_id=None)
452
# ie.revision is always None if the InventoryEntry is considered
453
# for committing. We may record the previous parents revision if the
454
# content is actually unchanged against a sole head.
455
if ie.revision is not None:
456
if not self._versioned_root and path == '':
457
# repositories that do not version the root set the root's
458
# revision to the new commit even when no change occurs (more
459
# specifically, they do not record a revision on the root; and
460
# the rev id is assigned to the root during deserialisation -
461
# this masks when a change may have occurred against the basis.
462
# To match this we always issue a delta, because the revision
463
# of the root will always be changing.
464
if ie.file_id in basis_inv:
465
delta = (basis_inv.id2path(ie.file_id), path,
469
delta = (None, path, ie.file_id, ie)
470
self._basis_delta.append(delta)
471
return delta, False, None
473
# we don't need to commit this, because the caller already
474
# determined that an existing revision of this file is
475
# appropriate. If it's not being considered for committing then
476
# it and all its parents to the root must be unaltered so
477
# no-change against the basis.
478
if ie.revision == self._new_revision_id:
479
raise AssertionError("Impossible situation, a skipped "
480
"inventory entry (%r) claims to be modified in this "
481
"commit (%r).", (ie, self._new_revision_id))
482
return None, False, None
483
# XXX: Friction: parent_candidates should return a list not a dict
484
# so that we don't have to walk the inventories again.
485
parent_candiate_entries = ie.parent_candidates(parent_invs)
486
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
488
for inv in parent_invs:
489
if ie.file_id in inv:
490
old_rev = inv[ie.file_id].revision
491
if old_rev in head_set:
492
heads.append(inv[ie.file_id].revision)
493
head_set.remove(inv[ie.file_id].revision)
496
# now we check to see if we need to write a new record to the
498
# We write a new entry unless there is one head to the ancestors, and
499
# the kind-derived content is unchanged.
501
# Cheapest check first: no ancestors, or more the one head in the
502
# ancestors, we write a new node.
506
# There is a single head, look it up for comparison
507
parent_entry = parent_candiate_entries[heads[0]]
508
# if the non-content specific data has changed, we'll be writing a
510
if (parent_entry.parent_id != ie.parent_id or
511
parent_entry.name != ie.name):
513
# now we need to do content specific checks:
515
# if the kind changed the content obviously has
516
if kind != parent_entry.kind:
518
# Stat cache fingerprint feedback for the caller - None as we usually
519
# don't generate one.
522
if content_summary[2] is None:
523
raise ValueError("Files must not have executable = None")
525
# We can't trust a check of the file length because of content
527
if (# if the exec bit has changed we have to store:
528
parent_entry.executable != content_summary[2]):
530
elif parent_entry.text_sha1 == content_summary[3]:
531
# all meta and content is unchanged (using a hash cache
532
# hit to check the sha)
533
ie.revision = parent_entry.revision
534
ie.text_size = parent_entry.text_size
535
ie.text_sha1 = parent_entry.text_sha1
536
ie.executable = parent_entry.executable
537
return self._get_delta(ie, basis_inv, path), False, None
539
# Either there is only a hash change(no hash cache entry,
540
# or same size content change), or there is no change on
542
# Provide the parent's hash to the store layer, so that the
543
# content is unchanged we will not store a new node.
544
nostore_sha = parent_entry.text_sha1
546
# We want to record a new node regardless of the presence or
547
# absence of a content change in the file.
549
ie.executable = content_summary[2]
550
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
552
text = file_obj.read()
556
ie.text_sha1, ie.text_size = self._add_text_to_weave(
557
ie.file_id, text, heads, nostore_sha)
558
# Let the caller know we generated a stat fingerprint.
559
fingerprint = (ie.text_sha1, stat_value)
560
except errors.ExistingContent:
561
# Turns out that the file content was unchanged, and we were
562
# only going to store a new node if it was changed. Carry over
564
ie.revision = parent_entry.revision
565
ie.text_size = parent_entry.text_size
566
ie.text_sha1 = parent_entry.text_sha1
567
ie.executable = parent_entry.executable
568
return self._get_delta(ie, basis_inv, path), False, None
569
elif kind == 'directory':
571
# all data is meta here, nothing specific to directory, so
573
ie.revision = parent_entry.revision
574
return self._get_delta(ie, basis_inv, path), False, None
575
self._add_text_to_weave(ie.file_id, '', heads, None)
576
elif kind == 'symlink':
577
current_link_target = content_summary[3]
579
# symlink target is not generic metadata, check if it has
581
if current_link_target != parent_entry.symlink_target:
584
# unchanged, carry over.
585
ie.revision = parent_entry.revision
586
ie.symlink_target = parent_entry.symlink_target
587
return self._get_delta(ie, basis_inv, path), False, None
588
ie.symlink_target = current_link_target
589
self._add_text_to_weave(ie.file_id, '', heads, None)
590
elif kind == 'tree-reference':
592
if content_summary[3] != parent_entry.reference_revision:
595
# unchanged, carry over.
596
ie.reference_revision = parent_entry.reference_revision
597
ie.revision = parent_entry.revision
598
return self._get_delta(ie, basis_inv, path), False, None
599
ie.reference_revision = content_summary[3]
600
if ie.reference_revision is None:
601
raise AssertionError("invalid content_summary for nested tree: %r"
602
% (content_summary,))
603
self._add_text_to_weave(ie.file_id, '', heads, None)
605
raise NotImplementedError('unknown kind')
606
ie.revision = self._new_revision_id
607
self._any_changes = True
608
return self._get_delta(ie, basis_inv, path), True, fingerprint
610
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
611
_entry_factory=entry_factory):
204
raise NotImplementedError(self.will_record_deletes)
206
def record_iter_changes(self, tree, basis_revision_id, iter_changes):
612
207
"""Record a new tree via iter_changes.
614
209
:param tree: The tree to obtain text contents from for changed objects.
619
214
to basis_revision_id. The iterator must not include any items with
620
215
a current kind of None - missing items must be either filtered out
621
216
or errored-on beefore record_iter_changes sees the item.
622
:param _entry_factory: Private method to bind entry_factory locally for
624
217
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
625
218
tree._observed_sha1.
627
# Create an inventory delta based on deltas between all the parents and
628
# deltas between all the parent inventories. We use inventory delta's
629
# between the inventory objects because iter_changes masks
630
# last-changed-field only changes.
632
# file_id -> change map, change is fileid, paths, changed, versioneds,
633
# parents, names, kinds, executables
635
# {file_id -> revision_id -> inventory entry, for entries in parent
636
# trees that are not parents[0]
640
revtrees = list(self.repository.revision_trees(self.parents))
641
except errors.NoSuchRevision:
642
# one or more ghosts, slow path.
644
for revision_id in self.parents:
646
revtrees.append(self.repository.revision_tree(revision_id))
647
except errors.NoSuchRevision:
649
basis_revision_id = _mod_revision.NULL_REVISION
651
revtrees.append(self.repository.revision_tree(
652
_mod_revision.NULL_REVISION))
653
# The basis inventory from a repository
655
basis_inv = revtrees[0].inventory
657
basis_inv = self.repository.revision_tree(
658
_mod_revision.NULL_REVISION).inventory
659
if len(self.parents) > 0:
660
if basis_revision_id != self.parents[0] and not ghost_basis:
662
"arbitrary basis parents not yet supported with merges")
663
for revtree in revtrees[1:]:
664
for change in revtree.inventory._make_delta(basis_inv):
665
if change[1] is None:
666
# Not present in this parent.
668
if change[2] not in merged_ids:
669
if change[0] is not None:
670
basis_entry = basis_inv[change[2]]
671
merged_ids[change[2]] = [
673
basis_entry.revision,
676
parent_entries[change[2]] = {
678
basis_entry.revision:basis_entry,
680
change[3].revision:change[3],
683
merged_ids[change[2]] = [change[3].revision]
684
parent_entries[change[2]] = {change[3].revision:change[3]}
686
merged_ids[change[2]].append(change[3].revision)
687
parent_entries[change[2]][change[3].revision] = change[3]
690
# Setup the changes from the tree:
691
# changes maps file_id -> (change, [parent revision_ids])
693
for change in iter_changes:
694
# This probably looks up in basis_inv way to much.
695
if change[1][0] is not None:
696
head_candidate = [basis_inv[change[0]].revision]
699
changes[change[0]] = change, merged_ids.get(change[0],
701
unchanged_merged = set(merged_ids) - set(changes)
702
# Extend the changes dict with synthetic changes to record merges of
704
for file_id in unchanged_merged:
705
# Record a merged version of these items that did not change vs the
706
# basis. This can be either identical parallel changes, or a revert
707
# of a specific file after a merge. The recorded content will be
708
# that of the current tree (which is the same as the basis), but
709
# the per-file graph will reflect a merge.
710
# NB:XXX: We are reconstructing path information we had, this
711
# should be preserved instead.
712
# inv delta change: (file_id, (path_in_source, path_in_target),
713
# changed_content, versioned, parent, name, kind,
716
basis_entry = basis_inv[file_id]
717
except errors.NoSuchId:
718
# a change from basis->some_parents but file_id isn't in basis
719
# so was new in the merge, which means it must have changed
720
# from basis -> current, and as it hasn't the add was reverted
721
# by the user. So we discard this change.
725
(basis_inv.id2path(file_id), tree.id2path(file_id)),
727
(basis_entry.parent_id, basis_entry.parent_id),
728
(basis_entry.name, basis_entry.name),
729
(basis_entry.kind, basis_entry.kind),
730
(basis_entry.executable, basis_entry.executable))
731
changes[file_id] = (change, merged_ids[file_id])
732
# changes contains tuples with the change and a set of inventory
733
# candidates for the file.
735
# old_path, new_path, file_id, new_inventory_entry
736
seen_root = False # Is the root in the basis delta?
737
inv_delta = self._basis_delta
738
modified_rev = self._new_revision_id
739
for change, head_candidates in changes.values():
740
if change[3][1]: # versioned in target.
741
# Several things may be happening here:
742
# We may have a fork in the per-file graph
743
# - record a change with the content from tree
744
# We may have a change against < all trees
745
# - carry over the tree that hasn't changed
746
# We may have a change against all trees
747
# - record the change with the content from tree
750
entry = _entry_factory[kind](file_id, change[5][1],
752
head_set = self._heads(change[0], set(head_candidates))
755
for head_candidate in head_candidates:
756
if head_candidate in head_set:
757
heads.append(head_candidate)
758
head_set.remove(head_candidate)
761
# Could be a carry-over situation:
762
parent_entry_revs = parent_entries.get(file_id, None)
763
if parent_entry_revs:
764
parent_entry = parent_entry_revs.get(heads[0], None)
767
if parent_entry is None:
768
# The parent iter_changes was called against is the one
769
# that is the per-file head, so any change is relevant
770
# iter_changes is valid.
771
carry_over_possible = False
773
# could be a carry over situation
774
# A change against the basis may just indicate a merge,
775
# we need to check the content against the source of the
776
# merge to determine if it was changed after the merge
778
if (parent_entry.kind != entry.kind or
779
parent_entry.parent_id != entry.parent_id or
780
parent_entry.name != entry.name):
781
# Metadata common to all entries has changed
782
# against per-file parent
783
carry_over_possible = False
785
carry_over_possible = True
786
# per-type checks for changes against the parent_entry
789
# Cannot be a carry-over situation
790
carry_over_possible = False
791
# Populate the entry in the delta
793
# XXX: There is still a small race here: If someone reverts the content of a file
794
# after iter_changes examines and decides it has changed,
795
# we will unconditionally record a new version even if some
796
# other process reverts it while commit is running (with
797
# the revert happening after iter_changes did its
800
entry.executable = True
802
entry.executable = False
803
if (carry_over_possible and
804
parent_entry.executable == entry.executable):
805
# Check the file length, content hash after reading
807
nostore_sha = parent_entry.text_sha1
810
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
812
text = file_obj.read()
816
entry.text_sha1, entry.text_size = self._add_text_to_weave(
817
file_id, text, heads, nostore_sha)
818
yield file_id, change[1][1], (entry.text_sha1, stat_value)
819
except errors.ExistingContent:
820
# No content change against a carry_over parent
821
# Perhaps this should also yield a fs hash update?
823
entry.text_size = parent_entry.text_size
824
entry.text_sha1 = parent_entry.text_sha1
825
elif kind == 'symlink':
827
entry.symlink_target = tree.get_symlink_target(file_id)
828
if (carry_over_possible and
829
parent_entry.symlink_target == entry.symlink_target):
832
self._add_text_to_weave(change[0], '', heads, None)
833
elif kind == 'directory':
834
if carry_over_possible:
837
# Nothing to set on the entry.
838
# XXX: split into the Root and nonRoot versions.
839
if change[1][1] != '' or self.repository.supports_rich_root():
840
self._add_text_to_weave(change[0], '', heads, None)
841
elif kind == 'tree-reference':
842
if not self.repository._format.supports_tree_reference:
843
# This isn't quite sane as an error, but we shouldn't
844
# ever see this code path in practice: tree's don't
845
# permit references when the repo doesn't support tree
847
raise errors.UnsupportedOperation(tree.add_reference,
849
reference_revision = tree.get_reference_revision(change[0])
850
entry.reference_revision = reference_revision
851
if (carry_over_possible and
852
parent_entry.reference_revision == reference_revision):
855
self._add_text_to_weave(change[0], '', heads, None)
857
raise AssertionError('unknown kind %r' % kind)
859
entry.revision = modified_rev
861
entry.revision = parent_entry.revision
864
new_path = change[1][1]
865
inv_delta.append((change[1][0], new_path, change[0], entry))
868
self.new_inventory = None
870
# This should perhaps be guarded by a check that the basis we
871
# commit against is the basis for the commit and if not do a delta
873
self._any_changes = True
875
# housekeeping root entry changes do not affect no-change commits.
876
self._require_root_change(tree)
877
self.basis_delta_revision = basis_revision_id
879
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
880
parent_keys = tuple([(file_id, parent) for parent in parents])
881
return self.repository.texts._add_text(
882
(file_id, self._new_revision_id), parent_keys, new_text,
883
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
886
class RootCommitBuilder(CommitBuilder):
887
"""This commitbuilder actually records the root id"""
889
# the root entry gets versioned properly by this builder.
890
_versioned_root = True
892
def _check_root(self, ie, parent_invs, tree):
893
"""Helper for record_entry_contents.
895
:param ie: An entry being added.
896
:param parent_invs: The inventories of the parent revisions of the
898
:param tree: The tree that is being committed.
901
def _require_root_change(self, tree):
902
"""Enforce an appropriate root object change.
904
This is called once when record_iter_changes is called, if and only if
905
the root was not in the delta calculated by record_iter_changes.
907
:param tree: The tree which is being committed.
909
# versioned roots do not change unless the tree found a change.
220
raise NotImplementedError(self.record_iter_changes)
912
223
class RepositoryWriteLockResult(LogicalLockResult):
1061
312
return InterRepository._assert_same_model(self, repository)
1063
def add_inventory(self, revision_id, inv, parents):
1064
"""Add the inventory inv to the repository as revision_id.
1066
:param parents: The revision ids of the parents that revision_id
1067
is known to have and are in the repository already.
1069
:returns: The validator(which is a sha1 digest, though what is sha'd is
1070
repository format specific) of the serialized inventory.
1072
if not self.is_in_write_group():
1073
raise AssertionError("%r not in write group" % (self,))
1074
_mod_revision.check_not_reserved_id(revision_id)
1075
if not (inv.revision_id is None or inv.revision_id == revision_id):
1076
raise AssertionError(
1077
"Mismatch between inventory revision"
1078
" id and insertion revid (%r, %r)"
1079
% (inv.revision_id, revision_id))
1080
if inv.root is None:
1081
raise errors.RootMissing()
1082
return self._add_inventory_checked(revision_id, inv, parents)
1084
def _add_inventory_checked(self, revision_id, inv, parents):
1085
"""Add inv to the repository after checking the inputs.
1087
This function can be overridden to allow different inventory styles.
1089
:seealso: add_inventory, for the contract.
1091
inv_lines = self._serializer.write_inventory_to_lines(inv)
1092
return self._inventory_add_lines(revision_id, parents,
1093
inv_lines, check_content=False)
1095
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1096
parents, basis_inv=None, propagate_caches=False):
1097
"""Add a new inventory expressed as a delta against another revision.
1099
See the inventory developers documentation for the theory behind
1102
:param basis_revision_id: The inventory id the delta was created
1103
against. (This does not have to be a direct parent.)
1104
:param delta: The inventory delta (see Inventory.apply_delta for
1106
:param new_revision_id: The revision id that the inventory is being
1108
:param parents: The revision ids of the parents that revision_id is
1109
known to have and are in the repository already. These are supplied
1110
for repositories that depend on the inventory graph for revision
1111
graph access, as well as for those that pun ancestry with delta
1113
:param basis_inv: The basis inventory if it is already known,
1115
:param propagate_caches: If True, the caches for this inventory are
1116
copied to and updated for the result if possible.
1118
:returns: (validator, new_inv)
1119
The validator(which is a sha1 digest, though what is sha'd is
1120
repository format specific) of the serialized inventory, and the
1121
resulting inventory.
1123
if not self.is_in_write_group():
1124
raise AssertionError("%r not in write group" % (self,))
1125
_mod_revision.check_not_reserved_id(new_revision_id)
1126
basis_tree = self.revision_tree(basis_revision_id)
1127
basis_tree.lock_read()
1129
# Note that this mutates the inventory of basis_tree, which not all
1130
# inventory implementations may support: A better idiom would be to
1131
# return a new inventory, but as there is no revision tree cache in
1132
# repository this is safe for now - RBC 20081013
1133
if basis_inv is None:
1134
basis_inv = basis_tree.inventory
1135
basis_inv.apply_delta(delta)
1136
basis_inv.revision_id = new_revision_id
1137
return (self.add_inventory(new_revision_id, basis_inv, parents),
1142
def _inventory_add_lines(self, revision_id, parents, lines,
1143
check_content=True):
1144
"""Store lines in inv_vf and return the sha1 of the inventory."""
1145
parents = [(parent,) for parent in parents]
1146
result = self.inventories.add_lines((revision_id,), parents, lines,
1147
check_content=check_content)[0]
1148
self.inventories._access.flush()
1151
def add_revision(self, revision_id, rev, inv=None, config=None):
1152
"""Add rev to the revision store as revision_id.
1154
:param revision_id: the revision id to use.
1155
:param rev: The revision object.
1156
:param inv: The inventory for the revision. if None, it will be looked
1157
up in the inventory storer
1158
:param config: If None no digital signature will be created.
1159
If supplied its signature_needed method will be used
1160
to determine if a signature should be made.
1162
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1164
_mod_revision.check_not_reserved_id(revision_id)
1165
if config is not None and config.signature_needed():
1167
inv = self.get_inventory(revision_id)
1168
plaintext = Testament(rev, inv).as_short_text()
1169
self.store_revision_signature(
1170
gpg.GPGStrategy(config), plaintext, revision_id)
1171
# check inventory present
1172
if not self.inventories.get_parent_map([(revision_id,)]):
1174
raise errors.WeaveRevisionNotPresent(revision_id,
1177
# yes, this is not suitable for adding with ghosts.
1178
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1181
key = (revision_id,)
1182
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1183
self._add_revision(rev)
1185
def _add_revision(self, revision):
1186
text = self._serializer.write_revision_to_string(revision)
1187
key = (revision.revision_id,)
1188
parents = tuple((parent,) for parent in revision.parent_ids)
1189
self.revisions.add_lines(key, parents, osutils.split_lines(text))
1191
314
def all_revision_ids(self):
1192
315
"""Returns a list of all the revision ids in the repository.
1228
351
# The old API returned a list, should this actually be a set?
1229
352
return parent_map.keys()
1231
def _check_inventories(self, checker):
1232
"""Check the inventories found from the revision scan.
1234
This is responsible for verifying the sha1 of inventories and
1235
creating a pending_keys set that covers data referenced by inventories.
1237
bar = ui.ui_factory.nested_progress_bar()
1239
self._do_check_inventories(checker, bar)
1243
def _do_check_inventories(self, checker, bar):
1244
"""Helper for _check_inventories."""
1246
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1247
kinds = ['chk_bytes', 'texts']
1248
count = len(checker.pending_keys)
1249
bar.update("inventories", 0, 2)
1250
current_keys = checker.pending_keys
1251
checker.pending_keys = {}
1252
# Accumulate current checks.
1253
for key in current_keys:
1254
if key[0] != 'inventories' and key[0] not in kinds:
1255
checker._report_items.append('unknown key type %r' % (key,))
1256
keys[key[0]].add(key[1:])
1257
if keys['inventories']:
1258
# NB: output order *should* be roughly sorted - topo or
1259
# inverse topo depending on repository - either way decent
1260
# to just delta against. However, pre-CHK formats didn't
1261
# try to optimise inventory layout on disk. As such the
1262
# pre-CHK code path does not use inventory deltas.
1264
for record in self.inventories.check(keys=keys['inventories']):
1265
if record.storage_kind == 'absent':
1266
checker._report_items.append(
1267
'Missing inventory {%s}' % (record.key,))
1269
last_object = self._check_record('inventories', record,
1270
checker, last_object,
1271
current_keys[('inventories',) + record.key])
1272
del keys['inventories']
1275
bar.update("texts", 1)
1276
while (checker.pending_keys or keys['chk_bytes']
1278
# Something to check.
1279
current_keys = checker.pending_keys
1280
checker.pending_keys = {}
1281
# Accumulate current checks.
1282
for key in current_keys:
1283
if key[0] not in kinds:
1284
checker._report_items.append('unknown key type %r' % (key,))
1285
keys[key[0]].add(key[1:])
1286
# Check the outermost kind only - inventories || chk_bytes || texts
1290
for record in getattr(self, kind).check(keys=keys[kind]):
1291
if record.storage_kind == 'absent':
1292
checker._report_items.append(
1293
'Missing %s {%s}' % (kind, record.key,))
1295
last_object = self._check_record(kind, record,
1296
checker, last_object, current_keys[(kind,) + record.key])
1300
def _check_record(self, kind, record, checker, last_object, item_data):
1301
"""Check a single text from this repository."""
1302
if kind == 'inventories':
1303
rev_id = record.key[0]
1304
inv = self._deserialise_inventory(rev_id,
1305
record.get_bytes_as('fulltext'))
1306
if last_object is not None:
1307
delta = inv._make_delta(last_object)
1308
for old_path, path, file_id, ie in delta:
1311
ie.check(checker, rev_id, inv)
1313
for path, ie in inv.iter_entries():
1314
ie.check(checker, rev_id, inv)
1315
if self._format.fast_deltas:
1317
elif kind == 'chk_bytes':
1318
# No code written to check chk_bytes for this repo format.
1319
checker._report_items.append(
1320
'unsupported key type chk_bytes for %s' % (record.key,))
1321
elif kind == 'texts':
1322
self._check_text(record, checker, item_data)
1324
checker._report_items.append(
1325
'unknown key type %s for %s' % (kind, record.key))
1327
def _check_text(self, record, checker, item_data):
1328
"""Check a single text."""
1329
# Check it is extractable.
1330
# TODO: check length.
1331
if record.storage_kind == 'chunked':
1332
chunks = record.get_bytes_as(record.storage_kind)
1333
sha1 = osutils.sha_strings(chunks)
1334
length = sum(map(len, chunks))
1336
content = record.get_bytes_as('fulltext')
1337
sha1 = osutils.sha_string(content)
1338
length = len(content)
1339
if item_data and sha1 != item_data[1]:
1340
checker._report_items.append(
1341
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1342
(record.key, sha1, item_data[1], item_data[2]))
1345
355
def create(a_bzrdir):
1346
356
"""Construct the current default format repository in a_bzrdir."""
2163
970
uniquely identify the file version in the caller's context. (Examples:
2164
971
an index number or a TreeTransform trans_id.)
2166
bytes_iterator is an iterable of bytestrings for the file. The
2167
kind of iterable and length of the bytestrings are unspecified, but for
2168
this implementation, it is a list of bytes produced by
2169
VersionedFile.get_record_stream().
2171
973
:param desired_files: a list of (file_id, revision_id, identifier)
2175
for file_id, revision_id, callable_data in desired_files:
2176
text_keys[(file_id, revision_id)] = callable_data
2177
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2178
if record.storage_kind == 'absent':
2179
raise errors.RevisionNotPresent(record.key, self)
2180
yield text_keys[record.key], record.get_bytes_as('chunked')
2182
def _generate_text_key_index(self, text_key_references=None,
2184
"""Generate a new text key index for the repository.
2186
This is an expensive function that will take considerable time to run.
2188
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2189
list of parents, also text keys. When a given key has no parents,
2190
the parents list will be [NULL_REVISION].
2192
# All revisions, to find inventory parents.
2193
if ancestors is None:
2194
graph = self.get_graph()
2195
ancestors = graph.get_parent_map(self.all_revision_ids())
2196
if text_key_references is None:
2197
text_key_references = self.find_text_key_references()
2198
pb = ui.ui_factory.nested_progress_bar()
2200
return self._do_generate_text_key_index(ancestors,
2201
text_key_references, pb)
2205
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2206
"""Helper for _generate_text_key_index to avoid deep nesting."""
2207
revision_order = tsort.topo_sort(ancestors)
2208
invalid_keys = set()
2210
for revision_id in revision_order:
2211
revision_keys[revision_id] = set()
2212
text_count = len(text_key_references)
2213
# a cache of the text keys to allow reuse; costs a dict of all the
2214
# keys, but saves a 2-tuple for every child of a given key.
2216
for text_key, valid in text_key_references.iteritems():
2218
invalid_keys.add(text_key)
2220
revision_keys[text_key[1]].add(text_key)
2221
text_key_cache[text_key] = text_key
2222
del text_key_references
2224
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2225
NULL_REVISION = _mod_revision.NULL_REVISION
2226
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2227
# too small for large or very branchy trees. However, for 55K path
2228
# trees, it would be easy to use too much memory trivially. Ideally we
2229
# could gauge this by looking at available real memory etc, but this is
2230
# always a tricky proposition.
2231
inventory_cache = lru_cache.LRUCache(10)
2232
batch_size = 10 # should be ~150MB on a 55K path tree
2233
batch_count = len(revision_order) / batch_size + 1
2235
pb.update("Calculating text parents", processed_texts, text_count)
2236
for offset in xrange(batch_count):
2237
to_query = revision_order[offset * batch_size:(offset + 1) *
2241
for revision_id in to_query:
2242
parent_ids = ancestors[revision_id]
2243
for text_key in revision_keys[revision_id]:
2244
pb.update("Calculating text parents", processed_texts)
2245
processed_texts += 1
2246
candidate_parents = []
2247
for parent_id in parent_ids:
2248
parent_text_key = (text_key[0], parent_id)
2250
check_parent = parent_text_key not in \
2251
revision_keys[parent_id]
2253
# the parent parent_id is a ghost:
2254
check_parent = False
2255
# truncate the derived graph against this ghost.
2256
parent_text_key = None
2258
# look at the parent commit details inventories to
2259
# determine possible candidates in the per file graph.
2262
inv = inventory_cache[parent_id]
2264
inv = self.revision_tree(parent_id).inventory
2265
inventory_cache[parent_id] = inv
2267
parent_entry = inv[text_key[0]]
2268
except (KeyError, errors.NoSuchId):
2270
if parent_entry is not None:
2272
text_key[0], parent_entry.revision)
2274
parent_text_key = None
2275
if parent_text_key is not None:
2276
candidate_parents.append(
2277
text_key_cache[parent_text_key])
2278
parent_heads = text_graph.heads(candidate_parents)
2279
new_parents = list(parent_heads)
2280
new_parents.sort(key=lambda x:candidate_parents.index(x))
2281
if new_parents == []:
2282
new_parents = [NULL_REVISION]
2283
text_index[text_key] = new_parents
2285
for text_key in invalid_keys:
2286
text_index[text_key] = [NULL_REVISION]
2289
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2290
"""Get an iterable listing the keys of all the data introduced by a set
2293
The keys will be ordered so that the corresponding items can be safely
2294
fetched and inserted in that order.
2296
:returns: An iterable producing tuples of (knit-kind, file-id,
2297
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2298
'revisions'. file-id is None unless knit-kind is 'file'.
2300
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2303
for result in self._find_non_file_keys_to_fetch(revision_ids):
2306
def _find_file_keys_to_fetch(self, revision_ids, pb):
2307
# XXX: it's a bit weird to control the inventory weave caching in this
2308
# generator. Ideally the caching would be done in fetch.py I think. Or
2309
# maybe this generator should explicitly have the contract that it
2310
# should not be iterated until the previously yielded item has been
2312
inv_w = self.inventories
2314
# file ids that changed
2315
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2317
num_file_ids = len(file_ids)
2318
for file_id, altered_versions in file_ids.iteritems():
2320
pb.update("Fetch texts", count, num_file_ids)
2322
yield ("file", file_id, altered_versions)
2324
def _find_non_file_keys_to_fetch(self, revision_ids):
2326
yield ("inventory", None, revision_ids)
2329
# XXX: Note ATM no callers actually pay attention to this return
2330
# instead they just use the list of revision ids and ignore
2331
# missing sigs. Consider removing this work entirely
2332
revisions_with_signatures = set(self.signatures.get_parent_map(
2333
[(r,) for r in revision_ids]))
2334
revisions_with_signatures = set(
2335
[r for (r,) in revisions_with_signatures])
2336
revisions_with_signatures.intersection_update(revision_ids)
2337
yield ("signatures", None, revisions_with_signatures)
2340
yield ("revisions", None, revision_ids)
2343
def get_inventory(self, revision_id):
2344
"""Get Inventory object by revision id."""
2345
return self.iter_inventories([revision_id]).next()
2347
def iter_inventories(self, revision_ids, ordering=None):
2348
"""Get many inventories by revision_ids.
2350
This will buffer some or all of the texts used in constructing the
2351
inventories in memory, but will only parse a single inventory at a
2354
:param revision_ids: The expected revision ids of the inventories.
2355
:param ordering: optional ordering, e.g. 'topological'. If not
2356
specified, the order of revision_ids will be preserved (by
2357
buffering if necessary).
2358
:return: An iterator of inventories.
2360
if ((None in revision_ids)
2361
or (_mod_revision.NULL_REVISION in revision_ids)):
2362
raise ValueError('cannot get null revision inventory')
2363
return self._iter_inventories(revision_ids, ordering)
2365
def _iter_inventories(self, revision_ids, ordering):
2366
"""single-document based inventory iteration."""
2367
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2368
for text, revision_id in inv_xmls:
2369
yield self._deserialise_inventory(revision_id, text)
2371
def _iter_inventory_xmls(self, revision_ids, ordering):
2372
if ordering is None:
2373
order_as_requested = True
2374
ordering = 'unordered'
2376
order_as_requested = False
2377
keys = [(revision_id,) for revision_id in revision_ids]
2380
if order_as_requested:
2381
key_iter = iter(keys)
2382
next_key = key_iter.next()
2383
stream = self.inventories.get_record_stream(keys, ordering, True)
2385
for record in stream:
2386
if record.storage_kind != 'absent':
2387
chunks = record.get_bytes_as('chunked')
2388
if order_as_requested:
2389
text_chunks[record.key] = chunks
2391
yield ''.join(chunks), record.key[-1]
2393
raise errors.NoSuchRevision(self, record.key)
2394
if order_as_requested:
2395
# Yield as many results as we can while preserving order.
2396
while next_key in text_chunks:
2397
chunks = text_chunks.pop(next_key)
2398
yield ''.join(chunks), next_key[-1]
2400
next_key = key_iter.next()
2401
except StopIteration:
2402
# We still want to fully consume the get_record_stream,
2403
# just in case it is not actually finished at this point
2407
def _deserialise_inventory(self, revision_id, xml):
2408
"""Transform the xml into an inventory object.
2410
:param revision_id: The expected revision id of the inventory.
2411
:param xml: A serialised inventory.
2413
result = self._serializer.read_inventory_from_string(xml, revision_id,
2414
entry_cache=self._inventory_entry_cache,
2415
return_from_cache=self._safe_to_return_from_cache)
2416
if result.revision_id != revision_id:
2417
raise AssertionError('revision id mismatch %s != %s' % (
2418
result.revision_id, revision_id))
2421
def get_serializer_format(self):
2422
return self._serializer.format_num
2425
def _get_inventory_xml(self, revision_id):
2426
"""Get serialized inventory as a string."""
2427
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2429
text, revision_id = texts.next()
2430
except StopIteration:
2431
raise errors.HistoryMissing(self, 'inventory', revision_id)
976
raise NotImplementedError(self.iter_files_bytes)
2434
978
def get_rev_id_for_revno(self, revno, known_pair):
2435
979
"""Return the revision id of a revno, given a later (revno, revid)
2770
1272
except UnicodeDecodeError:
2771
1273
raise errors.NonAsciiRevisionId(method, self)
2773
def revision_graph_can_have_wrong_parents(self):
2774
"""Is it possible for this repository to have a revision graph with
2777
If True, then this repository must also implement
2778
_find_inconsistent_revision_parents so that check and reconcile can
2779
check for inconsistencies before proceeding with other checks that may
2780
depend on the revision index being consistent.
2782
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2785
def install_revision(repository, rev, revision_tree):
2786
"""Install all revision data into a repository."""
2787
install_revisions(repository, [(rev, revision_tree, None)])
2790
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2791
"""Install all revision data into a repository.
2793
Accepts an iterable of revision, tree, signature tuples. The signature
2796
repository.start_write_group()
2798
inventory_cache = lru_cache.LRUCache(10)
2799
for n, (revision, revision_tree, signature) in enumerate(iterable):
2800
_install_revision(repository, revision, revision_tree, signature,
2803
pb.update('Transferring revisions', n + 1, num_revisions)
2805
repository.abort_write_group()
2808
repository.commit_write_group()
2811
def _install_revision(repository, rev, revision_tree, signature,
2813
"""Install all revision data into a repository."""
2814
present_parents = []
2816
for p_id in rev.parent_ids:
2817
if repository.has_revision(p_id):
2818
present_parents.append(p_id)
2819
parent_trees[p_id] = repository.revision_tree(p_id)
2821
parent_trees[p_id] = repository.revision_tree(
2822
_mod_revision.NULL_REVISION)
2824
inv = revision_tree.inventory
2825
entries = inv.iter_entries()
2826
# backwards compatibility hack: skip the root id.
2827
if not repository.supports_rich_root():
2828
path, root = entries.next()
2829
if root.revision != rev.revision_id:
2830
raise errors.IncompatibleRevision(repr(repository))
2832
for path, ie in entries:
2833
text_keys[(ie.file_id, ie.revision)] = ie
2834
text_parent_map = repository.texts.get_parent_map(text_keys)
2835
missing_texts = set(text_keys) - set(text_parent_map)
2836
# Add the texts that are not already present
2837
for text_key in missing_texts:
2838
ie = text_keys[text_key]
2840
# FIXME: TODO: The following loop overlaps/duplicates that done by
2841
# commit to determine parents. There is a latent/real bug here where
2842
# the parents inserted are not those commit would do - in particular
2843
# they are not filtered by heads(). RBC, AB
2844
for revision, tree in parent_trees.iteritems():
2845
if ie.file_id not in tree:
2847
parent_id = tree.inventory[ie.file_id].revision
2848
if parent_id in text_parents:
2850
text_parents.append((ie.file_id, parent_id))
2851
lines = revision_tree.get_file(ie.file_id).readlines()
2852
repository.texts.add_lines(text_key, text_parents, lines)
2854
# install the inventory
2855
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2856
# Cache this inventory
2857
inventory_cache[rev.revision_id] = inv
2859
basis_inv = inventory_cache[rev.parent_ids[0]]
2861
repository.add_inventory(rev.revision_id, inv, present_parents)
2863
delta = inv._make_delta(basis_inv)
2864
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2865
rev.revision_id, present_parents)
2867
repository.add_inventory(rev.revision_id, inv, present_parents)
2868
except errors.RevisionAlreadyPresent:
2870
if signature is not None:
2871
repository.add_signature_text(rev.revision_id, signature)
2872
repository.add_revision(rev.revision_id, rev, inv)
2875
1276
class MetaDirRepository(Repository):
2876
1277
"""Repositories in the new meta-dir layout.
3516
1883
raise errors.IncompatibleRepositories(source, target,
3517
1884
"different serializers")
3520
class InterSameDataRepository(InterRepository):
3521
"""Code for converting between repositories that represent the same data.
3523
Data format and model must match for this to work.
3527
def _get_repo_format_to_test(self):
3528
"""Repository format for testing with.
3530
InterSameData can pull from subtree to subtree and from non-subtree to
3531
non-subtree, so we test this with the richest repository format.
3533
from bzrlib.repofmt import knitrepo
3534
return knitrepo.RepositoryFormatKnit3()
3537
def is_compatible(source, target):
3538
return InterRepository._same_model(source, target)
3541
class InterDifferingSerializer(InterRepository):
3544
1887
def _get_repo_format_to_test(self):
3548
def is_compatible(source, target):
3549
"""Be compatible with Knit2 source and Knit3 target"""
3550
# This is redundant with format.check_conversion_target(), however that
3551
# raises an exception, and we just want to say "False" as in we won't
3552
# support converting between these formats.
3553
if 'IDS_never' in debug.debug_flags:
3555
if source.supports_rich_root() and not target.supports_rich_root():
3557
if (source._format.supports_tree_reference
3558
and not target._format.supports_tree_reference):
3560
if target._fallback_repositories and target._format.supports_chks:
3561
# IDS doesn't know how to copy CHKs for the parent inventories it
3562
# adds to stacked repos.
3564
if 'IDS_always' in debug.debug_flags:
3566
# Only use this code path for local source and target. IDS does far
3567
# too much IO (both bandwidth and roundtrips) over a network.
3568
if not source.bzrdir.transport.base.startswith('file:///'):
3570
if not target.bzrdir.transport.base.startswith('file:///'):
1891
def is_compatible(cls, source, target):
1892
# The default implementation is compatible with everything
3574
def _get_trees(self, revision_ids, cache):
3576
for rev_id in revision_ids:
3578
possible_trees.append((rev_id, cache[rev_id]))
3580
# Not cached, but inventory might be present anyway.
3582
tree = self.source.revision_tree(rev_id)
3583
except errors.NoSuchRevision:
3584
# Nope, parent is ghost.
3587
cache[rev_id] = tree
3588
possible_trees.append((rev_id, tree))
3589
return possible_trees
3591
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3592
"""Get the best delta and base for this revision.
3594
:return: (basis_id, delta)
3597
# Generate deltas against each tree, to find the shortest.
3598
texts_possibly_new_in_tree = set()
3599
for basis_id, basis_tree in possible_trees:
3600
delta = tree.inventory._make_delta(basis_tree.inventory)
3601
for old_path, new_path, file_id, new_entry in delta:
3602
if new_path is None:
3603
# This file_id isn't present in the new rev, so we don't
3607
# Rich roots are handled elsewhere...
3609
kind = new_entry.kind
3610
if kind != 'directory' and kind != 'file':
3611
# No text record associated with this inventory entry.
3613
# This is a directory or file that has changed somehow.
3614
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3615
deltas.append((len(delta), basis_id, delta))
3617
return deltas[0][1:]
3619
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3620
"""Find all parent revisions that are absent, but for which the
3621
inventory is present, and copy those inventories.
3623
This is necessary to preserve correctness when the source is stacked
3624
without fallbacks configured. (Note that in cases like upgrade the
3625
source may be not have _fallback_repositories even though it is
3629
for parents in parent_map.values():
3630
parent_revs.update(parents)
3631
present_parents = self.source.get_parent_map(parent_revs)
3632
absent_parents = set(parent_revs).difference(present_parents)
3633
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3634
(rev_id,) for rev_id in absent_parents)
3635
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3636
for parent_tree in self.source.revision_trees(parent_inv_ids):
3637
current_revision_id = parent_tree.get_revision_id()
3638
parents_parents_keys = parent_invs_keys_for_stacking[
3639
(current_revision_id,)]
3640
parents_parents = [key[-1] for key in parents_parents_keys]
3641
basis_id = _mod_revision.NULL_REVISION
3642
basis_tree = self.source.revision_tree(basis_id)
3643
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3644
self.target.add_inventory_by_delta(
3645
basis_id, delta, current_revision_id, parents_parents)
3646
cache[current_revision_id] = parent_tree
3648
def _fetch_batch(self, revision_ids, basis_id, cache):
3649
"""Fetch across a few revisions.
3651
:param revision_ids: The revisions to copy
3652
:param basis_id: The revision_id of a tree that must be in cache, used
3653
as a basis for delta when no other base is available
3654
:param cache: A cache of RevisionTrees that we can use.
3655
:return: The revision_id of the last converted tree. The RevisionTree
3656
for it will be in cache
3658
# Walk though all revisions; get inventory deltas, copy referenced
3659
# texts that delta references, insert the delta, revision and
3661
root_keys_to_create = set()
3664
pending_revisions = []
3665
parent_map = self.source.get_parent_map(revision_ids)
3666
self._fetch_parent_invs_for_stacking(parent_map, cache)
3667
self.source._safe_to_return_from_cache = True
3668
for tree in self.source.revision_trees(revision_ids):
3669
# Find a inventory delta for this revision.
3670
# Find text entries that need to be copied, too.
3671
current_revision_id = tree.get_revision_id()
3672
parent_ids = parent_map.get(current_revision_id, ())
3673
parent_trees = self._get_trees(parent_ids, cache)
3674
possible_trees = list(parent_trees)
3675
if len(possible_trees) == 0:
3676
# There either aren't any parents, or the parents are ghosts,
3677
# so just use the last converted tree.
3678
possible_trees.append((basis_id, cache[basis_id]))
3679
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3681
revision = self.source.get_revision(current_revision_id)
3682
pending_deltas.append((basis_id, delta,
3683
current_revision_id, revision.parent_ids))
3684
if self._converting_to_rich_root:
3685
self._revision_id_to_root_id[current_revision_id] = \
3687
# Determine which texts are in present in this revision but not in
3688
# any of the available parents.
3689
texts_possibly_new_in_tree = set()
3690
for old_path, new_path, file_id, entry in delta:
3691
if new_path is None:
3692
# This file_id isn't present in the new rev
3696
if not self.target.supports_rich_root():
3697
# The target doesn't support rich root, so we don't
3700
if self._converting_to_rich_root:
3701
# This can't be copied normally, we have to insert
3703
root_keys_to_create.add((file_id, entry.revision))
3706
texts_possibly_new_in_tree.add((file_id, entry.revision))
3707
for basis_id, basis_tree in possible_trees:
3708
basis_inv = basis_tree.inventory
3709
for file_key in list(texts_possibly_new_in_tree):
3710
file_id, file_revision = file_key
3712
entry = basis_inv[file_id]
3713
except errors.NoSuchId:
3715
if entry.revision == file_revision:
3716
texts_possibly_new_in_tree.remove(file_key)
3717
text_keys.update(texts_possibly_new_in_tree)
3718
pending_revisions.append(revision)
3719
cache[current_revision_id] = tree
3720
basis_id = current_revision_id
3721
self.source._safe_to_return_from_cache = False
3723
from_texts = self.source.texts
3724
to_texts = self.target.texts
3725
if root_keys_to_create:
3726
root_stream = _mod_fetch._new_root_data_stream(
3727
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3729
to_texts.insert_record_stream(root_stream)
3730
to_texts.insert_record_stream(from_texts.get_record_stream(
3731
text_keys, self.target._format._fetch_order,
3732
not self.target._format._fetch_uses_deltas))
3733
# insert inventory deltas
3734
for delta in pending_deltas:
3735
self.target.add_inventory_by_delta(*delta)
3736
if self.target._fallback_repositories:
3737
# Make sure this stacked repository has all the parent inventories
3738
# for the new revisions that we are about to insert. We do this
3739
# before adding the revisions so that no revision is added until
3740
# all the inventories it may depend on are added.
3741
# Note that this is overzealous, as we may have fetched these in an
3744
revision_ids = set()
3745
for revision in pending_revisions:
3746
revision_ids.add(revision.revision_id)
3747
parent_ids.update(revision.parent_ids)
3748
parent_ids.difference_update(revision_ids)
3749
parent_ids.discard(_mod_revision.NULL_REVISION)
3750
parent_map = self.source.get_parent_map(parent_ids)
3751
# we iterate over parent_map and not parent_ids because we don't
3752
# want to try copying any revision which is a ghost
3753
for parent_tree in self.source.revision_trees(parent_map):
3754
current_revision_id = parent_tree.get_revision_id()
3755
parents_parents = parent_map[current_revision_id]
3756
possible_trees = self._get_trees(parents_parents, cache)
3757
if len(possible_trees) == 0:
3758
# There either aren't any parents, or the parents are
3759
# ghosts, so just use the last converted tree.
3760
possible_trees.append((basis_id, cache[basis_id]))
3761
basis_id, delta = self._get_delta_for_revision(parent_tree,
3762
parents_parents, possible_trees)
3763
self.target.add_inventory_by_delta(
3764
basis_id, delta, current_revision_id, parents_parents)
3765
# insert signatures and revisions
3766
for revision in pending_revisions:
3768
signature = self.source.get_signature_text(
3769
revision.revision_id)
3770
self.target.add_signature_text(revision.revision_id,
3772
except errors.NoSuchRevision:
3774
self.target.add_revision(revision.revision_id, revision)
3777
def _fetch_all_revisions(self, revision_ids, pb):
3778
"""Fetch everything for the list of revisions.
3780
:param revision_ids: The list of revisions to fetch. Must be in
3782
:param pb: A ProgressTask
3785
basis_id, basis_tree = self._get_basis(revision_ids[0])
3787
cache = lru_cache.LRUCache(100)
3788
cache[basis_id] = basis_tree
3789
del basis_tree # We don't want to hang on to it here
3793
for offset in range(0, len(revision_ids), batch_size):
3794
self.target.start_write_group()
3796
pb.update('Transferring revisions', offset,
3798
batch = revision_ids[offset:offset+batch_size]
3799
basis_id = self._fetch_batch(batch, basis_id, cache)
3801
self.source._safe_to_return_from_cache = False
3802
self.target.abort_write_group()
3805
hint = self.target.commit_write_group()
3808
if hints and self.target._format.pack_compresses:
3809
self.target.pack(hint=hints)
3810
pb.update('Transferring revisions', len(revision_ids),
3814
def fetch(self, revision_id=None, find_ghosts=False,
3816
"""See InterRepository.fetch()."""
3817
if fetch_spec is not None:
3818
revision_ids = fetch_spec.get_keys()
3821
ui.ui_factory.warn_experimental_format_fetch(self)
3822
if (not self.source.supports_rich_root()
3823
and self.target.supports_rich_root()):
3824
self._converting_to_rich_root = True
3825
self._revision_id_to_root_id = {}
3827
self._converting_to_rich_root = False
3828
# See <https://launchpad.net/bugs/456077> asking for a warning here
3829
if self.source._format.network_name() != self.target._format.network_name():
3830
ui.ui_factory.show_user_warning('cross_format_fetch',
3831
from_format=self.source._format,
3832
to_format=self.target._format)
3833
if revision_ids is None:
3835
search_revision_ids = [revision_id]
3837
search_revision_ids = None
3838
revision_ids = self.target.search_missing_revision_ids(self.source,
3839
revision_ids=search_revision_ids,
3840
find_ghosts=find_ghosts).get_keys()
3841
if not revision_ids:
3843
revision_ids = tsort.topo_sort(
3844
self.source.get_graph().get_parent_map(revision_ids))
3845
if not revision_ids:
3847
# Walk though all revisions; get inventory deltas, copy referenced
3848
# texts that delta references, insert the delta, revision and
3850
pb = ui.ui_factory.nested_progress_bar()
3852
self._fetch_all_revisions(revision_ids, pb)
3855
return len(revision_ids), 0
3857
def _get_basis(self, first_revision_id):
3858
"""Get a revision and tree which exists in the target.
3860
This assumes that first_revision_id is selected for transmission
3861
because all other ancestors are already present. If we can't find an
3862
ancestor we fall back to NULL_REVISION since we know that is safe.
3864
:return: (basis_id, basis_tree)
3866
first_rev = self.source.get_revision(first_revision_id)
3868
basis_id = first_rev.parent_ids[0]
3869
# only valid as a basis if the target has it
3870
self.target.get_revision(basis_id)
3871
# Try to get a basis tree - if it's a ghost it will hit the
3872
# NoSuchRevision case.
3873
basis_tree = self.source.revision_tree(basis_id)
3874
except (IndexError, errors.NoSuchRevision):
3875
basis_id = _mod_revision.NULL_REVISION
3876
basis_tree = self.source.revision_tree(basis_id)
3877
return basis_id, basis_tree
3880
InterRepository.register_optimiser(InterDifferingSerializer)
3881
InterRepository.register_optimiser(InterSameDataRepository)
1896
InterRepository.register_optimiser(InterRepository)
3884
1899
class CopyConverter(object):
4012
1955
return revision_graph
4015
class StreamSink(object):
4016
"""An object that can insert a stream into a repository.
4018
This interface handles the complexity of reserialising inventories and
4019
revisions from different formats, and allows unidirectional insertion into
4020
stacked repositories without looking for the missing basis parents
4024
def __init__(self, target_repo):
4025
self.target_repo = target_repo
4027
def insert_stream(self, stream, src_format, resume_tokens):
4028
"""Insert a stream's content into the target repository.
4030
:param src_format: a bzr repository format.
4032
:return: a list of resume tokens and an iterable of keys additional
4033
items required before the insertion can be completed.
4035
self.target_repo.lock_write()
4038
self.target_repo.resume_write_group(resume_tokens)
4041
self.target_repo.start_write_group()
4044
# locked_insert_stream performs a commit|suspend.
4045
missing_keys = self.insert_stream_without_locking(stream,
4046
src_format, is_resume)
4048
# suspend the write group and tell the caller what we is
4049
# missing. We know we can suspend or else we would not have
4050
# entered this code path. (All repositories that can handle
4051
# missing keys can handle suspending a write group).
4052
write_group_tokens = self.target_repo.suspend_write_group()
4053
return write_group_tokens, missing_keys
4054
hint = self.target_repo.commit_write_group()
4055
to_serializer = self.target_repo._format._serializer
4056
src_serializer = src_format._serializer
4057
if (to_serializer != src_serializer and
4058
self.target_repo._format.pack_compresses):
4059
self.target_repo.pack(hint=hint)
4062
self.target_repo.abort_write_group(suppress_errors=True)
4065
self.target_repo.unlock()
4067
def insert_stream_without_locking(self, stream, src_format,
4069
"""Insert a stream's content into the target repository.
4071
This assumes that you already have a locked repository and an active
4074
:param src_format: a bzr repository format.
4075
:param is_resume: Passed down to get_missing_parent_inventories to
4076
indicate if we should be checking for missing texts at the same
4079
:return: A set of keys that are missing.
4081
if not self.target_repo.is_write_locked():
4082
raise errors.ObjectNotLocked(self)
4083
if not self.target_repo.is_in_write_group():
4084
raise errors.BzrError('you must already be in a write group')
4085
to_serializer = self.target_repo._format._serializer
4086
src_serializer = src_format._serializer
4088
if to_serializer == src_serializer:
4089
# If serializers match and the target is a pack repository, set the
4090
# write cache size on the new pack. This avoids poor performance
4091
# on transports where append is unbuffered (such as
4092
# RemoteTransport). This is safe to do because nothing should read
4093
# back from the target repository while a stream with matching
4094
# serialization is being inserted.
4095
# The exception is that a delta record from the source that should
4096
# be a fulltext may need to be expanded by the target (see
4097
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4098
# explicitly flush any buffered writes first in that rare case.
4100
new_pack = self.target_repo._pack_collection._new_pack
4101
except AttributeError:
4102
# Not a pack repository
4105
new_pack.set_write_cache_size(1024*1024)
4106
for substream_type, substream in stream:
4107
if 'stream' in debug.debug_flags:
4108
mutter('inserting substream: %s', substream_type)
4109
if substream_type == 'texts':
4110
self.target_repo.texts.insert_record_stream(substream)
4111
elif substream_type == 'inventories':
4112
if src_serializer == to_serializer:
4113
self.target_repo.inventories.insert_record_stream(
4116
self._extract_and_insert_inventories(
4117
substream, src_serializer)
4118
elif substream_type == 'inventory-deltas':
4119
self._extract_and_insert_inventory_deltas(
4120
substream, src_serializer)
4121
elif substream_type == 'chk_bytes':
4122
# XXX: This doesn't support conversions, as it assumes the
4123
# conversion was done in the fetch code.
4124
self.target_repo.chk_bytes.insert_record_stream(substream)
4125
elif substream_type == 'revisions':
4126
# This may fallback to extract-and-insert more often than
4127
# required if the serializers are different only in terms of
4129
if src_serializer == to_serializer:
4130
self.target_repo.revisions.insert_record_stream(substream)
4132
self._extract_and_insert_revisions(substream,
4134
elif substream_type == 'signatures':
4135
self.target_repo.signatures.insert_record_stream(substream)
4137
raise AssertionError('kaboom! %s' % (substream_type,))
4138
# Done inserting data, and the missing_keys calculations will try to
4139
# read back from the inserted data, so flush the writes to the new pack
4140
# (if this is pack format).
4141
if new_pack is not None:
4142
new_pack._write_data('', flush=True)
4143
# Find all the new revisions (including ones from resume_tokens)
4144
missing_keys = self.target_repo.get_missing_parent_inventories(
4145
check_for_missing_texts=is_resume)
4147
for prefix, versioned_file in (
4148
('texts', self.target_repo.texts),
4149
('inventories', self.target_repo.inventories),
4150
('revisions', self.target_repo.revisions),
4151
('signatures', self.target_repo.signatures),
4152
('chk_bytes', self.target_repo.chk_bytes),
4154
if versioned_file is None:
4156
# TODO: key is often going to be a StaticTuple object
4157
# I don't believe we can define a method by which
4158
# (prefix,) + StaticTuple will work, though we could
4159
# define a StaticTuple.sq_concat that would allow you to
4160
# pass in either a tuple or a StaticTuple as the second
4161
# object, so instead we could have:
4162
# StaticTuple(prefix) + key here...
4163
missing_keys.update((prefix,) + key for key in
4164
versioned_file.get_missing_compression_parent_keys())
4165
except NotImplementedError:
4166
# cannot even attempt suspending, and missing would have failed
4167
# during stream insertion.
4168
missing_keys = set()
4171
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4172
target_rich_root = self.target_repo._format.rich_root_data
4173
target_tree_refs = self.target_repo._format.supports_tree_reference
4174
for record in substream:
4175
# Insert the delta directly
4176
inventory_delta_bytes = record.get_bytes_as('fulltext')
4177
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4179
parse_result = deserialiser.parse_text_bytes(
4180
inventory_delta_bytes)
4181
except inventory_delta.IncompatibleInventoryDelta, err:
4182
trace.mutter("Incompatible delta: %s", err.msg)
4183
raise errors.IncompatibleRevision(self.target_repo._format)
4184
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4185
revision_id = new_id
4186
parents = [key[0] for key in record.parents]
4187
self.target_repo.add_inventory_by_delta(
4188
basis_id, inv_delta, revision_id, parents)
4190
def _extract_and_insert_inventories(self, substream, serializer,
4192
"""Generate a new inventory versionedfile in target, converting data.
4194
The inventory is retrieved from the source, (deserializing it), and
4195
stored in the target (reserializing it in a different format).
4197
target_rich_root = self.target_repo._format.rich_root_data
4198
target_tree_refs = self.target_repo._format.supports_tree_reference
4199
for record in substream:
4200
# It's not a delta, so it must be a fulltext in the source
4201
# serializer's format.
4202
bytes = record.get_bytes_as('fulltext')
4203
revision_id = record.key[0]
4204
inv = serializer.read_inventory_from_string(bytes, revision_id)
4205
parents = [key[0] for key in record.parents]
4206
self.target_repo.add_inventory(revision_id, inv, parents)
4207
# No need to keep holding this full inv in memory when the rest of
4208
# the substream is likely to be all deltas.
4211
def _extract_and_insert_revisions(self, substream, serializer):
4212
for record in substream:
4213
bytes = record.get_bytes_as('fulltext')
4214
revision_id = record.key[0]
4215
rev = serializer.read_revision_from_string(bytes)
4216
if rev.revision_id != revision_id:
4217
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4218
self.target_repo.add_revision(revision_id, rev)
4221
if self.target_repo._format._fetch_reconcile:
4222
self.target_repo.reconcile()
4225
class StreamSource(object):
4226
"""A source of a stream for fetching between repositories."""
4228
def __init__(self, from_repository, to_format):
4229
"""Create a StreamSource streaming from from_repository."""
4230
self.from_repository = from_repository
4231
self.to_format = to_format
4232
self._record_counter = RecordCounter()
4234
def delta_on_metadata(self):
4235
"""Return True if delta's are permitted on metadata streams.
4237
That is on revisions and signatures.
4239
src_serializer = self.from_repository._format._serializer
4240
target_serializer = self.to_format._serializer
4241
return (self.to_format._fetch_uses_deltas and
4242
src_serializer == target_serializer)
4244
def _fetch_revision_texts(self, revs):
4245
# fetch signatures first and then the revision texts
4246
# may need to be a InterRevisionStore call here.
4247
from_sf = self.from_repository.signatures
4248
# A missing signature is just skipped.
4249
keys = [(rev_id,) for rev_id in revs]
4250
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4252
self.to_format._fetch_order,
4253
not self.to_format._fetch_uses_deltas))
4254
# If a revision has a delta, this is actually expanded inside the
4255
# insert_record_stream code now, which is an alternate fix for
4257
from_rf = self.from_repository.revisions
4258
revisions = from_rf.get_record_stream(
4260
self.to_format._fetch_order,
4261
not self.delta_on_metadata())
4262
return [('signatures', signatures), ('revisions', revisions)]
4264
def _generate_root_texts(self, revs):
4265
"""This will be called by get_stream between fetching weave texts and
4266
fetching the inventory weave.
4268
if self._rich_root_upgrade():
4269
return _mod_fetch.Inter1and2Helper(
4270
self.from_repository).generate_root_texts(revs)
4274
def get_stream(self, search):
4276
revs = search.get_keys()
4277
graph = self.from_repository.get_graph()
4278
revs = tsort.topo_sort(graph.get_parent_map(revs))
4279
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4281
for knit_kind, file_id, revisions in data_to_fetch:
4282
if knit_kind != phase:
4284
# Make a new progress bar for this phase
4285
if knit_kind == "file":
4286
# Accumulate file texts
4287
text_keys.extend([(file_id, revision) for revision in
4289
elif knit_kind == "inventory":
4290
# Now copy the file texts.
4291
from_texts = self.from_repository.texts
4292
yield ('texts', from_texts.get_record_stream(
4293
text_keys, self.to_format._fetch_order,
4294
not self.to_format._fetch_uses_deltas))
4295
# Cause an error if a text occurs after we have done the
4298
# Before we process the inventory we generate the root
4299
# texts (if necessary) so that the inventories references
4301
for _ in self._generate_root_texts(revs):
4303
# we fetch only the referenced inventories because we do not
4304
# know for unselected inventories whether all their required
4305
# texts are present in the other repository - it could be
4307
for info in self._get_inventory_stream(revs):
4309
elif knit_kind == "signatures":
4310
# Nothing to do here; this will be taken care of when
4311
# _fetch_revision_texts happens.
4313
elif knit_kind == "revisions":
4314
for record in self._fetch_revision_texts(revs):
4317
raise AssertionError("Unknown knit kind %r" % knit_kind)
4319
def get_stream_for_missing_keys(self, missing_keys):
4320
# missing keys can only occur when we are byte copying and not
4321
# translating (because translation means we don't send
4322
# unreconstructable deltas ever).
4324
keys['texts'] = set()
4325
keys['revisions'] = set()
4326
keys['inventories'] = set()
4327
keys['chk_bytes'] = set()
4328
keys['signatures'] = set()
4329
for key in missing_keys:
4330
keys[key[0]].add(key[1:])
4331
if len(keys['revisions']):
4332
# If we allowed copying revisions at this point, we could end up
4333
# copying a revision without copying its required texts: a
4334
# violation of the requirements for repository integrity.
4335
raise AssertionError(
4336
'cannot copy revisions to fill in missing deltas %s' % (
4337
keys['revisions'],))
4338
for substream_kind, keys in keys.iteritems():
4339
vf = getattr(self.from_repository, substream_kind)
4340
if vf is None and keys:
4341
raise AssertionError(
4342
"cannot fill in keys for a versioned file we don't"
4343
" have: %s needs %s" % (substream_kind, keys))
4345
# No need to stream something we don't have
4347
if substream_kind == 'inventories':
4348
# Some missing keys are genuinely ghosts, filter those out.
4349
present = self.from_repository.inventories.get_parent_map(keys)
4350
revs = [key[0] for key in present]
4351
# Get the inventory stream more-or-less as we do for the
4352
# original stream; there's no reason to assume that records
4353
# direct from the source will be suitable for the sink. (Think
4354
# e.g. 2a -> 1.9-rich-root).
4355
for info in self._get_inventory_stream(revs, missing=True):
4359
# Ask for full texts always so that we don't need more round trips
4360
# after this stream.
4361
# Some of the missing keys are genuinely ghosts, so filter absent
4362
# records. The Sink is responsible for doing another check to
4363
# ensure that ghosts don't introduce missing data for future
4365
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4366
self.to_format._fetch_order, True))
4367
yield substream_kind, stream
4369
def inventory_fetch_order(self):
4370
if self._rich_root_upgrade():
4371
return 'topological'
4373
return self.to_format._fetch_order
4375
def _rich_root_upgrade(self):
4376
return (not self.from_repository._format.rich_root_data and
4377
self.to_format.rich_root_data)
4379
def _get_inventory_stream(self, revision_ids, missing=False):
4380
from_format = self.from_repository._format
4381
if (from_format.supports_chks and self.to_format.supports_chks and
4382
from_format.network_name() == self.to_format.network_name()):
4383
raise AssertionError(
4384
"this case should be handled by GroupCHKStreamSource")
4385
elif 'forceinvdeltas' in debug.debug_flags:
4386
return self._get_convertable_inventory_stream(revision_ids,
4387
delta_versus_null=missing)
4388
elif from_format.network_name() == self.to_format.network_name():
4390
return self._get_simple_inventory_stream(revision_ids,
4392
elif (not from_format.supports_chks and not self.to_format.supports_chks
4393
and from_format._serializer == self.to_format._serializer):
4394
# Essentially the same format.
4395
return self._get_simple_inventory_stream(revision_ids,
4398
# Any time we switch serializations, we want to use an
4399
# inventory-delta based approach.
4400
return self._get_convertable_inventory_stream(revision_ids,
4401
delta_versus_null=missing)
4403
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4404
# NB: This currently reopens the inventory weave in source;
4405
# using a single stream interface instead would avoid this.
4406
from_weave = self.from_repository.inventories
4408
delta_closure = True
4410
delta_closure = not self.delta_on_metadata()
4411
yield ('inventories', from_weave.get_record_stream(
4412
[(rev_id,) for rev_id in revision_ids],
4413
self.inventory_fetch_order(), delta_closure))
4415
def _get_convertable_inventory_stream(self, revision_ids,
4416
delta_versus_null=False):
4417
# The two formats are sufficiently different that there is no fast
4418
# path, so we need to send just inventorydeltas, which any
4419
# sufficiently modern client can insert into any repository.
4420
# The StreamSink code expects to be able to
4421
# convert on the target, so we need to put bytes-on-the-wire that can
4422
# be converted. That means inventory deltas (if the remote is <1.19,
4423
# RemoteStreamSink will fallback to VFS to insert the deltas).
4424
yield ('inventory-deltas',
4425
self._stream_invs_as_deltas(revision_ids,
4426
delta_versus_null=delta_versus_null))
4428
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4429
"""Return a stream of inventory-deltas for the given rev ids.
4431
:param revision_ids: The list of inventories to transmit
4432
:param delta_versus_null: Don't try to find a minimal delta for this
4433
entry, instead compute the delta versus the NULL_REVISION. This
4434
effectively streams a complete inventory. Used for stuff like
4435
filling in missing parents, etc.
4437
from_repo = self.from_repository
4438
revision_keys = [(rev_id,) for rev_id in revision_ids]
4439
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4440
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4442
inventories = self.from_repository.iter_inventories(
4443
revision_ids, 'topological')
4444
format = from_repo._format
4445
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4446
inventory_cache = lru_cache.LRUCache(50)
4447
null_inventory = from_repo.revision_tree(
4448
_mod_revision.NULL_REVISION).inventory
4449
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4450
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4451
# repo back into a non-rich-root repo ought to be allowed)
4452
serializer = inventory_delta.InventoryDeltaSerializer(
4453
versioned_root=format.rich_root_data,
4454
tree_references=format.supports_tree_reference)
4455
for inv in inventories:
4456
key = (inv.revision_id,)
4457
parent_keys = parent_map.get(key, ())
4459
if not delta_versus_null and parent_keys:
4460
# The caller did not ask for complete inventories and we have
4461
# some parents that we can delta against. Make a delta against
4462
# each parent so that we can find the smallest.
4463
parent_ids = [parent_key[0] for parent_key in parent_keys]
4464
for parent_id in parent_ids:
4465
if parent_id not in invs_sent_so_far:
4466
# We don't know that the remote side has this basis, so
4469
if parent_id == _mod_revision.NULL_REVISION:
4470
parent_inv = null_inventory
4472
parent_inv = inventory_cache.get(parent_id, None)
4473
if parent_inv is None:
4474
parent_inv = from_repo.get_inventory(parent_id)
4475
candidate_delta = inv._make_delta(parent_inv)
4476
if (delta is None or
4477
len(delta) > len(candidate_delta)):
4478
delta = candidate_delta
4479
basis_id = parent_id
4481
# Either none of the parents ended up being suitable, or we
4482
# were asked to delta against NULL
4483
basis_id = _mod_revision.NULL_REVISION
4484
delta = inv._make_delta(null_inventory)
4485
invs_sent_so_far.add(inv.revision_id)
4486
inventory_cache[inv.revision_id] = inv
4487
delta_serialized = ''.join(
4488
serializer.delta_to_lines(basis_id, key[-1], delta))
4489
yield versionedfile.FulltextContentFactory(
4490
key, parent_keys, None, delta_serialized)
4493
1958
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4494
1959
stop_revision=None):
4495
1960
"""Extend the partial history to include a given index