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)
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.
224
863
class RepositoryWriteLockResult(LogicalLockResult):
313
1018
return InterRepository._assert_same_model(self, repository)
1020
def add_inventory(self, revision_id, inv, parents):
1021
"""Add the inventory inv to the repository as revision_id.
1023
:param parents: The revision ids of the parents that revision_id
1024
is known to have and are in the repository already.
1026
:returns: The validator(which is a sha1 digest, though what is sha'd is
1027
repository format specific) of the serialized inventory.
1029
if not self.is_in_write_group():
1030
raise AssertionError("%r not in write group" % (self,))
1031
_mod_revision.check_not_reserved_id(revision_id)
1032
if not (inv.revision_id is None or inv.revision_id == revision_id):
1033
raise AssertionError(
1034
"Mismatch between inventory revision"
1035
" id and insertion revid (%r, %r)"
1036
% (inv.revision_id, revision_id))
1037
if inv.root is None:
1038
raise AssertionError()
1039
return self._add_inventory_checked(revision_id, inv, parents)
1041
def _add_inventory_checked(self, revision_id, inv, parents):
1042
"""Add inv to the repository after checking the inputs.
1044
This function can be overridden to allow different inventory styles.
1046
:seealso: add_inventory, for the contract.
1048
inv_lines = self._serializer.write_inventory_to_lines(inv)
1049
return self._inventory_add_lines(revision_id, parents,
1050
inv_lines, check_content=False)
1052
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1053
parents, basis_inv=None, propagate_caches=False):
1054
"""Add a new inventory expressed as a delta against another revision.
1056
See the inventory developers documentation for the theory behind
1059
:param basis_revision_id: The inventory id the delta was created
1060
against. (This does not have to be a direct parent.)
1061
:param delta: The inventory delta (see Inventory.apply_delta for
1063
:param new_revision_id: The revision id that the inventory is being
1065
:param parents: The revision ids of the parents that revision_id is
1066
known to have and are in the repository already. These are supplied
1067
for repositories that depend on the inventory graph for revision
1068
graph access, as well as for those that pun ancestry with delta
1070
:param basis_inv: The basis inventory if it is already known,
1072
:param propagate_caches: If True, the caches for this inventory are
1073
copied to and updated for the result if possible.
1075
:returns: (validator, new_inv)
1076
The validator(which is a sha1 digest, though what is sha'd is
1077
repository format specific) of the serialized inventory, and the
1078
resulting inventory.
1080
if not self.is_in_write_group():
1081
raise AssertionError("%r not in write group" % (self,))
1082
_mod_revision.check_not_reserved_id(new_revision_id)
1083
basis_tree = self.revision_tree(basis_revision_id)
1084
basis_tree.lock_read()
1086
# Note that this mutates the inventory of basis_tree, which not all
1087
# inventory implementations may support: A better idiom would be to
1088
# return a new inventory, but as there is no revision tree cache in
1089
# repository this is safe for now - RBC 20081013
1090
if basis_inv is None:
1091
basis_inv = basis_tree.inventory
1092
basis_inv.apply_delta(delta)
1093
basis_inv.revision_id = new_revision_id
1094
return (self.add_inventory(new_revision_id, basis_inv, parents),
1099
def _inventory_add_lines(self, revision_id, parents, lines,
1100
check_content=True):
1101
"""Store lines in inv_vf and return the sha1 of the inventory."""
1102
parents = [(parent,) for parent in parents]
1103
result = self.inventories.add_lines((revision_id,), parents, lines,
1104
check_content=check_content)[0]
1105
self.inventories._access.flush()
1108
def add_revision(self, revision_id, rev, inv=None, config=None):
1109
"""Add rev to the revision store as revision_id.
1111
:param revision_id: the revision id to use.
1112
:param rev: The revision object.
1113
:param inv: The inventory for the revision. if None, it will be looked
1114
up in the inventory storer
1115
:param config: If None no digital signature will be created.
1116
If supplied its signature_needed method will be used
1117
to determine if a signature should be made.
1119
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1121
_mod_revision.check_not_reserved_id(revision_id)
1122
if config is not None and config.signature_needed():
1124
inv = self.get_inventory(revision_id)
1125
plaintext = Testament(rev, inv).as_short_text()
1126
self.store_revision_signature(
1127
gpg.GPGStrategy(config), plaintext, revision_id)
1128
# check inventory present
1129
if not self.inventories.get_parent_map([(revision_id,)]):
1131
raise errors.WeaveRevisionNotPresent(revision_id,
1134
# yes, this is not suitable for adding with ghosts.
1135
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1138
key = (revision_id,)
1139
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1140
self._add_revision(rev)
1142
def _add_revision(self, revision):
1143
text = self._serializer.write_revision_to_string(revision)
1144
key = (revision.revision_id,)
1145
parents = tuple((parent,) for parent in revision.parent_ids)
1146
self.revisions.add_lines(key, parents, osutils.split_lines(text))
315
1148
def all_revision_ids(self):
316
1149
"""Returns a list of all the revision ids in the repository.
352
1185
# The old API returned a list, should this actually be a set?
353
1186
return parent_map.keys()
1188
def _check_inventories(self, checker):
1189
"""Check the inventories found from the revision scan.
1191
This is responsible for verifying the sha1 of inventories and
1192
creating a pending_keys set that covers data referenced by inventories.
1194
bar = ui.ui_factory.nested_progress_bar()
1196
self._do_check_inventories(checker, bar)
1200
def _do_check_inventories(self, checker, bar):
1201
"""Helper for _check_inventories."""
1203
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1204
kinds = ['chk_bytes', 'texts']
1205
count = len(checker.pending_keys)
1206
bar.update("inventories", 0, 2)
1207
current_keys = checker.pending_keys
1208
checker.pending_keys = {}
1209
# Accumulate current checks.
1210
for key in current_keys:
1211
if key[0] != 'inventories' and key[0] not in kinds:
1212
checker._report_items.append('unknown key type %r' % (key,))
1213
keys[key[0]].add(key[1:])
1214
if keys['inventories']:
1215
# NB: output order *should* be roughly sorted - topo or
1216
# inverse topo depending on repository - either way decent
1217
# to just delta against. However, pre-CHK formats didn't
1218
# try to optimise inventory layout on disk. As such the
1219
# pre-CHK code path does not use inventory deltas.
1221
for record in self.inventories.check(keys=keys['inventories']):
1222
if record.storage_kind == 'absent':
1223
checker._report_items.append(
1224
'Missing inventory {%s}' % (record.key,))
1226
last_object = self._check_record('inventories', record,
1227
checker, last_object,
1228
current_keys[('inventories',) + record.key])
1229
del keys['inventories']
1232
bar.update("texts", 1)
1233
while (checker.pending_keys or keys['chk_bytes']
1235
# Something to check.
1236
current_keys = checker.pending_keys
1237
checker.pending_keys = {}
1238
# Accumulate current checks.
1239
for key in current_keys:
1240
if key[0] not in kinds:
1241
checker._report_items.append('unknown key type %r' % (key,))
1242
keys[key[0]].add(key[1:])
1243
# Check the outermost kind only - inventories || chk_bytes || texts
1247
for record in getattr(self, kind).check(keys=keys[kind]):
1248
if record.storage_kind == 'absent':
1249
checker._report_items.append(
1250
'Missing %s {%s}' % (kind, record.key,))
1252
last_object = self._check_record(kind, record,
1253
checker, last_object, current_keys[(kind,) + record.key])
1257
def _check_record(self, kind, record, checker, last_object, item_data):
1258
"""Check a single text from this repository."""
1259
if kind == 'inventories':
1260
rev_id = record.key[0]
1261
inv = self._deserialise_inventory(rev_id,
1262
record.get_bytes_as('fulltext'))
1263
if last_object is not None:
1264
delta = inv._make_delta(last_object)
1265
for old_path, path, file_id, ie in delta:
1268
ie.check(checker, rev_id, inv)
1270
for path, ie in inv.iter_entries():
1271
ie.check(checker, rev_id, inv)
1272
if self._format.fast_deltas:
1274
elif kind == 'chk_bytes':
1275
# No code written to check chk_bytes for this repo format.
1276
checker._report_items.append(
1277
'unsupported key type chk_bytes for %s' % (record.key,))
1278
elif kind == 'texts':
1279
self._check_text(record, checker, item_data)
1281
checker._report_items.append(
1282
'unknown key type %s for %s' % (kind, record.key))
1284
def _check_text(self, record, checker, item_data):
1285
"""Check a single text."""
1286
# Check it is extractable.
1287
# TODO: check length.
1288
if record.storage_kind == 'chunked':
1289
chunks = record.get_bytes_as(record.storage_kind)
1290
sha1 = osutils.sha_strings(chunks)
1291
length = sum(map(len, chunks))
1293
content = record.get_bytes_as('fulltext')
1294
sha1 = osutils.sha_string(content)
1295
length = len(content)
1296
if item_data and sha1 != item_data[1]:
1297
checker._report_items.append(
1298
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1299
(record.key, sha1, item_data[1], item_data[2]))
356
1302
def create(a_bzrdir):
357
1303
"""Construct the current default format repository in a_bzrdir."""
921
1986
signature = gpg_strategy.sign(plaintext)
922
1987
self.add_signature_text(revision_id, signature)
924
1990
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)
1991
self.signatures.add_lines((revision_id,), (),
1992
osutils.split_lines(signature))
1994
def find_text_key_references(self):
1995
"""Find the text key references within the repository.
1997
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1998
to whether they were referred to by the inventory of the
1999
revision_id that they contain. The inventory texts from all present
2000
revision ids are assessed to generate this report.
2002
revision_keys = self.revisions.keys()
2003
w = self.inventories
2004
pb = ui.ui_factory.nested_progress_bar()
2006
return self._find_text_key_references_from_xml_inventory_lines(
2007
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
2011
def _find_text_key_references_from_xml_inventory_lines(self,
2013
"""Core routine for extracting references to texts from inventories.
2015
This performs the translation of xml lines to revision ids.
2017
:param line_iterator: An iterator of lines, origin_version_id
2018
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
2019
to whether they were referred to by the inventory of the
2020
revision_id that they contain. Note that if that revision_id was
2021
not part of the line_iterator's output then False will be given -
2022
even though it may actually refer to that key.
2024
if not self._serializer.support_altered_by_hack:
2025
raise AssertionError(
2026
"_find_text_key_references_from_xml_inventory_lines only "
2027
"supported for branches which store inventory as unnested xml"
2028
", not on %r" % self)
2031
# this code needs to read every new line in every inventory for the
2032
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
2033
# not present in one of those inventories is unnecessary but not
2034
# harmful because we are filtering by the revision id marker in the
2035
# inventory lines : we only select file ids altered in one of those
2036
# revisions. We don't need to see all lines in the inventory because
2037
# only those added in an inventory in rev X can contain a revision=X
2039
unescape_revid_cache = {}
2040
unescape_fileid_cache = {}
2042
# jam 20061218 In a big fetch, this handles hundreds of thousands
2043
# of lines, so it has had a lot of inlining and optimizing done.
2044
# Sorry that it is a little bit messy.
2045
# Move several functions to be local variables, since this is a long
2047
search = self._file_ids_altered_regex.search
2048
unescape = _unescape_xml
2049
setdefault = result.setdefault
2050
for line, line_key in line_iterator:
2051
match = search(line)
2054
# One call to match.group() returning multiple items is quite a
2055
# bit faster than 2 calls to match.group() each returning 1
2056
file_id, revision_id = match.group('file_id', 'revision_id')
2058
# Inlining the cache lookups helps a lot when you make 170,000
2059
# lines and 350k ids, versus 8.4 unique ids.
2060
# Using a cache helps in 2 ways:
2061
# 1) Avoids unnecessary decoding calls
2062
# 2) Re-uses cached strings, which helps in future set and
2064
# (2) is enough that removing encoding entirely along with
2065
# the cache (so we are using plain strings) results in no
2066
# performance improvement.
2068
revision_id = unescape_revid_cache[revision_id]
2070
unescaped = unescape(revision_id)
2071
unescape_revid_cache[revision_id] = unescaped
2072
revision_id = unescaped
2074
# Note that unconditionally unescaping means that we deserialise
2075
# every fileid, which for general 'pull' is not great, but we don't
2076
# really want to have some many fulltexts that this matters anyway.
2079
file_id = unescape_fileid_cache[file_id]
2081
unescaped = unescape(file_id)
2082
unescape_fileid_cache[file_id] = unescaped
2085
key = (file_id, revision_id)
2086
setdefault(key, False)
2087
if revision_id == line_key[-1]:
2091
def _inventory_xml_lines_for_keys(self, keys):
2092
"""Get a line iterator of the sort needed for findind references.
2094
Not relevant for non-xml inventory repositories.
2096
Ghosts in revision_keys are ignored.
2098
:param revision_keys: The revision keys for the inventories to inspect.
2099
:return: An iterator over (inventory line, revid) for the fulltexts of
2100
all of the xml inventories specified by revision_keys.
2102
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2103
for record in stream:
2104
if record.storage_kind != 'absent':
2105
chunks = record.get_bytes_as('chunked')
2106
revid = record.key[-1]
2107
lines = osutils.chunks_to_lines(chunks)
2111
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
2113
"""Helper routine for fileids_altered_by_revision_ids.
2115
This performs the translation of xml lines to revision ids.
2117
:param line_iterator: An iterator of lines, origin_version_id
2118
:param revision_keys: The revision ids to filter for. This should be a
2119
set or other type which supports efficient __contains__ lookups, as
2120
the revision key from each parsed line will be looked up in the
2121
revision_keys filter.
2122
:return: a dictionary mapping altered file-ids to an iterable of
2123
revision_ids. Each altered file-ids has the exact revision_ids that
2124
altered it listed explicitly.
2126
seen = set(self._find_text_key_references_from_xml_inventory_lines(
2127
line_iterator).iterkeys())
2128
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
2129
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
2130
self._inventory_xml_lines_for_keys(parent_keys)))
2131
new_keys = seen - parent_seen
2133
setdefault = result.setdefault
2134
for key in new_keys:
2135
setdefault(key[0], set()).add(key[-1])
932
2138
def _find_parent_ids_of_revisions(self, revision_ids):
933
2139
"""Find all parent ids that are mentioned in the revision graph.
963
2190
uniquely identify the file version in the caller's context. (Examples:
964
2191
an index number or a TreeTransform trans_id.)
2193
bytes_iterator is an iterable of bytestrings for the file. The
2194
kind of iterable and length of the bytestrings are unspecified, but for
2195
this implementation, it is a list of bytes produced by
2196
VersionedFile.get_record_stream().
966
2198
:param desired_files: a list of (file_id, revision_id, identifier)
969
raise NotImplementedError(self.iter_files_bytes)
2202
for file_id, revision_id, callable_data in desired_files:
2203
text_keys[(file_id, revision_id)] = callable_data
2204
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2205
if record.storage_kind == 'absent':
2206
raise errors.RevisionNotPresent(record.key, self)
2207
yield text_keys[record.key], record.get_bytes_as('chunked')
2209
def _generate_text_key_index(self, text_key_references=None,
2211
"""Generate a new text key index for the repository.
2213
This is an expensive function that will take considerable time to run.
2215
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2216
list of parents, also text keys. When a given key has no parents,
2217
the parents list will be [NULL_REVISION].
2219
# All revisions, to find inventory parents.
2220
if ancestors is None:
2221
graph = self.get_graph()
2222
ancestors = graph.get_parent_map(self.all_revision_ids())
2223
if text_key_references is None:
2224
text_key_references = self.find_text_key_references()
2225
pb = ui.ui_factory.nested_progress_bar()
2227
return self._do_generate_text_key_index(ancestors,
2228
text_key_references, pb)
2232
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2233
"""Helper for _generate_text_key_index to avoid deep nesting."""
2234
revision_order = tsort.topo_sort(ancestors)
2235
invalid_keys = set()
2237
for revision_id in revision_order:
2238
revision_keys[revision_id] = set()
2239
text_count = len(text_key_references)
2240
# a cache of the text keys to allow reuse; costs a dict of all the
2241
# keys, but saves a 2-tuple for every child of a given key.
2243
for text_key, valid in text_key_references.iteritems():
2245
invalid_keys.add(text_key)
2247
revision_keys[text_key[1]].add(text_key)
2248
text_key_cache[text_key] = text_key
2249
del text_key_references
2251
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2252
NULL_REVISION = _mod_revision.NULL_REVISION
2253
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2254
# too small for large or very branchy trees. However, for 55K path
2255
# trees, it would be easy to use too much memory trivially. Ideally we
2256
# could gauge this by looking at available real memory etc, but this is
2257
# always a tricky proposition.
2258
inventory_cache = lru_cache.LRUCache(10)
2259
batch_size = 10 # should be ~150MB on a 55K path tree
2260
batch_count = len(revision_order) / batch_size + 1
2262
pb.update("Calculating text parents", processed_texts, text_count)
2263
for offset in xrange(batch_count):
2264
to_query = revision_order[offset * batch_size:(offset + 1) *
2268
for revision_id in to_query:
2269
parent_ids = ancestors[revision_id]
2270
for text_key in revision_keys[revision_id]:
2271
pb.update("Calculating text parents", processed_texts)
2272
processed_texts += 1
2273
candidate_parents = []
2274
for parent_id in parent_ids:
2275
parent_text_key = (text_key[0], parent_id)
2277
check_parent = parent_text_key not in \
2278
revision_keys[parent_id]
2280
# the parent parent_id is a ghost:
2281
check_parent = False
2282
# truncate the derived graph against this ghost.
2283
parent_text_key = None
2285
# look at the parent commit details inventories to
2286
# determine possible candidates in the per file graph.
2289
inv = inventory_cache[parent_id]
2291
inv = self.revision_tree(parent_id).inventory
2292
inventory_cache[parent_id] = inv
2294
parent_entry = inv[text_key[0]]
2295
except (KeyError, errors.NoSuchId):
2297
if parent_entry is not None:
2299
text_key[0], parent_entry.revision)
2301
parent_text_key = None
2302
if parent_text_key is not None:
2303
candidate_parents.append(
2304
text_key_cache[parent_text_key])
2305
parent_heads = text_graph.heads(candidate_parents)
2306
new_parents = list(parent_heads)
2307
new_parents.sort(key=lambda x:candidate_parents.index(x))
2308
if new_parents == []:
2309
new_parents = [NULL_REVISION]
2310
text_index[text_key] = new_parents
2312
for text_key in invalid_keys:
2313
text_index[text_key] = [NULL_REVISION]
2316
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2317
"""Get an iterable listing the keys of all the data introduced by a set
2320
The keys will be ordered so that the corresponding items can be safely
2321
fetched and inserted in that order.
2323
:returns: An iterable producing tuples of (knit-kind, file-id,
2324
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2325
'revisions'. file-id is None unless knit-kind is 'file'.
2327
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2330
for result in self._find_non_file_keys_to_fetch(revision_ids):
2333
def _find_file_keys_to_fetch(self, revision_ids, pb):
2334
# XXX: it's a bit weird to control the inventory weave caching in this
2335
# generator. Ideally the caching would be done in fetch.py I think. Or
2336
# maybe this generator should explicitly have the contract that it
2337
# should not be iterated until the previously yielded item has been
2339
inv_w = self.inventories
2341
# file ids that changed
2342
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2344
num_file_ids = len(file_ids)
2345
for file_id, altered_versions in file_ids.iteritems():
2347
pb.update("Fetch texts", count, num_file_ids)
2349
yield ("file", file_id, altered_versions)
2351
def _find_non_file_keys_to_fetch(self, revision_ids):
2353
yield ("inventory", None, revision_ids)
2356
# XXX: Note ATM no callers actually pay attention to this return
2357
# instead they just use the list of revision ids and ignore
2358
# missing sigs. Consider removing this work entirely
2359
revisions_with_signatures = set(self.signatures.get_parent_map(
2360
[(r,) for r in revision_ids]))
2361
revisions_with_signatures = set(
2362
[r for (r,) in revisions_with_signatures])
2363
revisions_with_signatures.intersection_update(revision_ids)
2364
yield ("signatures", None, revisions_with_signatures)
2367
yield ("revisions", None, revision_ids)
2370
def get_inventory(self, revision_id):
2371
"""Get Inventory object by revision id."""
2372
return self.iter_inventories([revision_id]).next()
2374
def iter_inventories(self, revision_ids, ordering=None):
2375
"""Get many inventories by revision_ids.
2377
This will buffer some or all of the texts used in constructing the
2378
inventories in memory, but will only parse a single inventory at a
2381
:param revision_ids: The expected revision ids of the inventories.
2382
:param ordering: optional ordering, e.g. 'topological'. If not
2383
specified, the order of revision_ids will be preserved (by
2384
buffering if necessary).
2385
:return: An iterator of inventories.
2387
if ((None in revision_ids)
2388
or (_mod_revision.NULL_REVISION in revision_ids)):
2389
raise ValueError('cannot get null revision inventory')
2390
return self._iter_inventories(revision_ids, ordering)
2392
def _iter_inventories(self, revision_ids, ordering):
2393
"""single-document based inventory iteration."""
2394
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2395
for text, revision_id in inv_xmls:
2396
yield self._deserialise_inventory(revision_id, text)
2398
def _iter_inventory_xmls(self, revision_ids, ordering):
2399
if ordering is None:
2400
order_as_requested = True
2401
ordering = 'unordered'
2403
order_as_requested = False
2404
keys = [(revision_id,) for revision_id in revision_ids]
2407
if order_as_requested:
2408
key_iter = iter(keys)
2409
next_key = key_iter.next()
2410
stream = self.inventories.get_record_stream(keys, ordering, True)
2412
for record in stream:
2413
if record.storage_kind != 'absent':
2414
chunks = record.get_bytes_as('chunked')
2415
if order_as_requested:
2416
text_chunks[record.key] = chunks
2418
yield ''.join(chunks), record.key[-1]
2420
raise errors.NoSuchRevision(self, record.key)
2421
if order_as_requested:
2422
# Yield as many results as we can while preserving order.
2423
while next_key in text_chunks:
2424
chunks = text_chunks.pop(next_key)
2425
yield ''.join(chunks), next_key[-1]
2427
next_key = key_iter.next()
2428
except StopIteration:
2429
# We still want to fully consume the get_record_stream,
2430
# just in case it is not actually finished at this point
2434
def _deserialise_inventory(self, revision_id, xml):
2435
"""Transform the xml into an inventory object.
2437
:param revision_id: The expected revision id of the inventory.
2438
:param xml: A serialised inventory.
2440
result = self._serializer.read_inventory_from_string(xml, revision_id,
2441
entry_cache=self._inventory_entry_cache,
2442
return_from_cache=self._safe_to_return_from_cache)
2443
if result.revision_id != revision_id:
2444
raise AssertionError('revision id mismatch %s != %s' % (
2445
result.revision_id, revision_id))
2448
def get_serializer_format(self):
2449
return self._serializer.format_num
2452
def _get_inventory_xml(self, revision_id):
2453
"""Get serialized inventory as a string."""
2454
texts = self._iter_inventory_xmls([revision_id], 'unordered')
2456
text, revision_id = texts.next()
2457
except StopIteration:
2458
raise errors.HistoryMissing(self, 'inventory', revision_id)
971
2461
def get_rev_id_for_revno(self, revno, known_pair):
972
2462
"""Return the revision id of a revno, given a later (revno, revid)
1289
2806
except UnicodeDecodeError:
1290
2807
raise errors.NonAsciiRevisionId(method, self)
2809
def revision_graph_can_have_wrong_parents(self):
2810
"""Is it possible for this repository to have a revision graph with
2813
If True, then this repository must also implement
2814
_find_inconsistent_revision_parents so that check and reconcile can
2815
check for inconsistencies before proceeding with other checks that may
2816
depend on the revision index being consistent.
2818
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2821
# remove these delegates a while after bzr 0.15
2822
def __make_delegated(name, from_module):
2823
def _deprecated_repository_forwarder():
2824
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2825
% (name, from_module),
2828
m = __import__(from_module, globals(), locals(), [name])
2830
return getattr(m, name)
2831
except AttributeError:
2832
raise AttributeError('module %s has no name %s'
2834
globals()[name] = _deprecated_repository_forwarder
2837
'AllInOneRepository',
2838
'WeaveMetaDirRepository',
2839
'PreSplitOutRepositoryFormat',
2840
'RepositoryFormat4',
2841
'RepositoryFormat5',
2842
'RepositoryFormat6',
2843
'RepositoryFormat7',
2845
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2849
'RepositoryFormatKnit',
2850
'RepositoryFormatKnit1',
2852
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2855
def install_revision(repository, rev, revision_tree):
2856
"""Install all revision data into a repository."""
2857
install_revisions(repository, [(rev, revision_tree, None)])
2860
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2861
"""Install all revision data into a repository.
2863
Accepts an iterable of revision, tree, signature tuples. The signature
2866
repository.start_write_group()
2868
inventory_cache = lru_cache.LRUCache(10)
2869
for n, (revision, revision_tree, signature) in enumerate(iterable):
2870
_install_revision(repository, revision, revision_tree, signature,
2873
pb.update('Transferring revisions', n + 1, num_revisions)
2875
repository.abort_write_group()
2878
repository.commit_write_group()
2881
def _install_revision(repository, rev, revision_tree, signature,
2883
"""Install all revision data into a repository."""
2884
present_parents = []
2886
for p_id in rev.parent_ids:
2887
if repository.has_revision(p_id):
2888
present_parents.append(p_id)
2889
parent_trees[p_id] = repository.revision_tree(p_id)
2891
parent_trees[p_id] = repository.revision_tree(
2892
_mod_revision.NULL_REVISION)
2894
inv = revision_tree.inventory
2895
entries = inv.iter_entries()
2896
# backwards compatibility hack: skip the root id.
2897
if not repository.supports_rich_root():
2898
path, root = entries.next()
2899
if root.revision != rev.revision_id:
2900
raise errors.IncompatibleRevision(repr(repository))
2902
for path, ie in entries:
2903
text_keys[(ie.file_id, ie.revision)] = ie
2904
text_parent_map = repository.texts.get_parent_map(text_keys)
2905
missing_texts = set(text_keys) - set(text_parent_map)
2906
# Add the texts that are not already present
2907
for text_key in missing_texts:
2908
ie = text_keys[text_key]
2910
# FIXME: TODO: The following loop overlaps/duplicates that done by
2911
# commit to determine parents. There is a latent/real bug here where
2912
# the parents inserted are not those commit would do - in particular
2913
# they are not filtered by heads(). RBC, AB
2914
for revision, tree in parent_trees.iteritems():
2915
if ie.file_id not in tree:
2917
parent_id = tree.inventory[ie.file_id].revision
2918
if parent_id in text_parents:
2920
text_parents.append((ie.file_id, parent_id))
2921
lines = revision_tree.get_file(ie.file_id).readlines()
2922
repository.texts.add_lines(text_key, text_parents, lines)
2924
# install the inventory
2925
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2926
# Cache this inventory
2927
inventory_cache[rev.revision_id] = inv
2929
basis_inv = inventory_cache[rev.parent_ids[0]]
2931
repository.add_inventory(rev.revision_id, inv, present_parents)
2933
delta = inv._make_delta(basis_inv)
2934
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2935
rev.revision_id, present_parents)
2937
repository.add_inventory(rev.revision_id, inv, present_parents)
2938
except errors.RevisionAlreadyPresent:
2940
if signature is not None:
2941
repository.add_signature_text(rev.revision_id, signature)
2942
repository.add_revision(rev.revision_id, rev, inv)
1293
2945
class MetaDirRepository(Repository):
1294
2946
"""Repositories in the new meta-dir layout.
1633
3311
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
1634
3312
format_registry.register_lazy(
1635
3313
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
1636
'bzrlib.repofmt.knitpack_repo',
3314
'bzrlib.repofmt.pack_repo',
1637
3315
'RepositoryFormatKnitPack1',
1639
3317
format_registry.register_lazy(
1640
3318
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
1641
'bzrlib.repofmt.knitpack_repo',
3319
'bzrlib.repofmt.pack_repo',
1642
3320
'RepositoryFormatKnitPack3',
1644
3322
format_registry.register_lazy(
1645
3323
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
1646
'bzrlib.repofmt.knitpack_repo',
3324
'bzrlib.repofmt.pack_repo',
1647
3325
'RepositoryFormatKnitPack4',
1649
3327
format_registry.register_lazy(
1650
3328
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
1651
'bzrlib.repofmt.knitpack_repo',
3329
'bzrlib.repofmt.pack_repo',
1652
3330
'RepositoryFormatKnitPack5',
1654
3332
format_registry.register_lazy(
1655
3333
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
1656
'bzrlib.repofmt.knitpack_repo',
3334
'bzrlib.repofmt.pack_repo',
1657
3335
'RepositoryFormatKnitPack5RichRoot',
1659
3337
format_registry.register_lazy(
1660
3338
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
1661
'bzrlib.repofmt.knitpack_repo',
3339
'bzrlib.repofmt.pack_repo',
1662
3340
'RepositoryFormatKnitPack5RichRootBroken',
1664
3342
format_registry.register_lazy(
1665
3343
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
1666
'bzrlib.repofmt.knitpack_repo',
3344
'bzrlib.repofmt.pack_repo',
1667
3345
'RepositoryFormatKnitPack6',
1669
3347
format_registry.register_lazy(
1670
3348
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
1671
'bzrlib.repofmt.knitpack_repo',
3349
'bzrlib.repofmt.pack_repo',
1672
3350
'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
3353
# Development formats.
1681
# Check their docstrings to see if/when they are obsolete.
3354
# Obsolete but kept pending a CHK based subtree format.
1682
3355
format_registry.register_lazy(
1683
3356
("Bazaar development format 2 with subtree support "
1684
3357
"(needs bzr.dev from before 1.8)\n"),
1685
'bzrlib.repofmt.knitpack_repo',
3358
'bzrlib.repofmt.pack_repo',
1686
3359
'RepositoryFormatPackDevelopment2Subtree',
1688
format_registry.register_lazy(
1689
'Bazaar development format 8\n',
1690
'bzrlib.repofmt.groupcompress_repo',
1691
'RepositoryFormat2aSubtree',
3362
# 1.14->1.16 go below here
3363
format_registry.register_lazy(
3364
'Bazaar development format - group compression and chk inventory'
3365
' (needs bzr.dev from 1.14)\n',
3366
'bzrlib.repofmt.groupcompress_repo',
3367
'RepositoryFormatCHK1',
3370
format_registry.register_lazy(
3371
'Bazaar development format - chk repository with bencode revision '
3372
'serialization (needs bzr.dev from 1.16)\n',
3373
'bzrlib.repofmt.groupcompress_repo',
3374
'RepositoryFormatCHK2',
3376
format_registry.register_lazy(
3377
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3378
'bzrlib.repofmt.groupcompress_repo',
3379
'RepositoryFormat2a',
1733
3422
:param revision_id: if None all content is copied, if NULL_REVISION no
1734
3423
content is copied.
1737
raise NotImplementedError(self.fetch)
3427
ui.ui_factory.warn_experimental_format_fetch(self)
3428
from bzrlib.fetch import RepoFetcher
3429
# See <https://launchpad.net/bugs/456077> asking for a warning here
3430
if self.source._format.network_name() != self.target._format.network_name():
3431
ui.ui_factory.show_user_warning('cross_format_fetch',
3432
from_format=self.source._format,
3433
to_format=self.target._format)
3434
f = RepoFetcher(to_repository=self.target,
3435
from_repository=self.source,
3436
last_revision=revision_id,
3437
fetch_spec=fetch_spec,
3438
find_ghosts=find_ghosts)
3440
def _walk_to_common_revisions(self, revision_ids):
3441
"""Walk out from revision_ids in source to revisions target has.
3443
:param revision_ids: The start point for the search.
3444
:return: A set of revision ids.
3446
target_graph = self.target.get_graph()
3447
revision_ids = frozenset(revision_ids)
3448
missing_revs = set()
3449
source_graph = self.source.get_graph()
3450
# ensure we don't pay silly lookup costs.
3451
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3452
null_set = frozenset([_mod_revision.NULL_REVISION])
3453
searcher_exhausted = False
3457
# Iterate the searcher until we have enough next_revs
3458
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3460
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3461
next_revs.update(next_revs_part)
3462
ghosts.update(ghosts_part)
3463
except StopIteration:
3464
searcher_exhausted = True
3466
# If there are ghosts in the source graph, and the caller asked for
3467
# them, make sure that they are present in the target.
3468
# We don't care about other ghosts as we can't fetch them and
3469
# haven't been asked to.
3470
ghosts_to_check = set(revision_ids.intersection(ghosts))
3471
revs_to_get = set(next_revs).union(ghosts_to_check)
3473
have_revs = set(target_graph.get_parent_map(revs_to_get))
3474
# we always have NULL_REVISION present.
3475
have_revs = have_revs.union(null_set)
3476
# Check if the target is missing any ghosts we need.
3477
ghosts_to_check.difference_update(have_revs)
3479
# One of the caller's revision_ids is a ghost in both the
3480
# source and the target.
3481
raise errors.NoSuchRevision(
3482
self.source, ghosts_to_check.pop())
3483
missing_revs.update(next_revs - have_revs)
3484
# Because we may have walked past the original stop point, make
3485
# sure everything is stopped
3486
stop_revs = searcher.find_seen_ancestors(have_revs)
3487
searcher.stop_searching_any(stop_revs)
3488
if searcher_exhausted:
3490
return searcher.get_result()
1739
3492
@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,
3493
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1744
3494
"""Return the revision ids that source has that target does not.
1746
3496
: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
3498
:param find_ghosts: If True find missing revisions in deep history
1756
3499
rather than just finding the surface difference.
1757
:param limit: Maximum number of revisions to return, topologically
1759
3500
:return: A bzrlib.graph.SearchResult.
1761
raise NotImplementedError(self.search_missing_revision_ids)
3502
# stop searching at found target revisions.
3503
if not find_ghosts and revision_id is not None:
3504
return self._walk_to_common_revisions([revision_id])
3505
# generic, possibly worst case, slow code path.
3506
target_ids = set(self.target.all_revision_ids())
3507
if revision_id is not None:
3508
source_ids = self.source.get_ancestry(revision_id)
3509
if source_ids[0] is not None:
3510
raise AssertionError()
3513
source_ids = self.source.all_revision_ids()
3514
result_set = set(source_ids).difference(target_ids)
3515
return self.source.revision_ids_to_search_result(result_set)
1764
3518
def _same_model(source, target):
1785
3539
"different serializers")
3542
class InterSameDataRepository(InterRepository):
3543
"""Code for converting between repositories that represent the same data.
3545
Data format and model must match for this to work.
3549
def _get_repo_format_to_test(self):
3550
"""Repository format for testing with.
3552
InterSameData can pull from subtree to subtree and from non-subtree to
3553
non-subtree, so we test this with the richest repository format.
3555
from bzrlib.repofmt import knitrepo
3556
return knitrepo.RepositoryFormatKnit3()
3559
def is_compatible(source, target):
3560
return InterRepository._same_model(source, target)
3563
class InterWeaveRepo(InterSameDataRepository):
3564
"""Optimised code paths between Weave based repositories.
3566
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3567
implemented lazy inter-object optimisation.
3571
def _get_repo_format_to_test(self):
3572
from bzrlib.repofmt import weaverepo
3573
return weaverepo.RepositoryFormat7()
3576
def is_compatible(source, target):
3577
"""Be compatible with known Weave formats.
3579
We don't test for the stores being of specific types because that
3580
could lead to confusing results, and there is no need to be
3583
from bzrlib.repofmt.weaverepo import (
3589
return (isinstance(source._format, (RepositoryFormat5,
3591
RepositoryFormat7)) and
3592
isinstance(target._format, (RepositoryFormat5,
3594
RepositoryFormat7)))
3595
except AttributeError:
3599
def copy_content(self, revision_id=None):
3600
"""See InterRepository.copy_content()."""
3601
# weave specific optimised path:
3603
self.target.set_make_working_trees(self.source.make_working_trees())
3604
except (errors.RepositoryUpgradeRequired, NotImplemented):
3606
# FIXME do not peek!
3607
if self.source._transport.listable():
3608
pb = ui.ui_factory.nested_progress_bar()
3610
self.target.texts.insert_record_stream(
3611
self.source.texts.get_record_stream(
3612
self.source.texts.keys(), 'topological', False))
3613
pb.update('Copying inventory', 0, 1)
3614
self.target.inventories.insert_record_stream(
3615
self.source.inventories.get_record_stream(
3616
self.source.inventories.keys(), 'topological', False))
3617
self.target.signatures.insert_record_stream(
3618
self.source.signatures.get_record_stream(
3619
self.source.signatures.keys(),
3621
self.target.revisions.insert_record_stream(
3622
self.source.revisions.get_record_stream(
3623
self.source.revisions.keys(),
3624
'topological', True))
3628
self.target.fetch(self.source, revision_id=revision_id)
3631
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3632
"""See InterRepository.missing_revision_ids()."""
3633
# we want all revisions to satisfy revision_id in source.
3634
# but we don't want to stat every file here and there.
3635
# we want then, all revisions other needs to satisfy revision_id
3636
# checked, but not those that we have locally.
3637
# so the first thing is to get a subset of the revisions to
3638
# satisfy revision_id in source, and then eliminate those that
3639
# we do already have.
3640
# this is slow on high latency connection to self, but as this
3641
# disk format scales terribly for push anyway due to rewriting
3642
# inventory.weave, this is considered acceptable.
3644
if revision_id is not None:
3645
source_ids = self.source.get_ancestry(revision_id)
3646
if source_ids[0] is not None:
3647
raise AssertionError()
3650
source_ids = self.source._all_possible_ids()
3651
source_ids_set = set(source_ids)
3652
# source_ids is the worst possible case we may need to pull.
3653
# now we want to filter source_ids against what we actually
3654
# have in target, but don't try to check for existence where we know
3655
# we do not have a revision as that would be pointless.
3656
target_ids = set(self.target._all_possible_ids())
3657
possibly_present_revisions = target_ids.intersection(source_ids_set)
3658
actually_present_revisions = set(
3659
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3660
required_revisions = source_ids_set.difference(actually_present_revisions)
3661
if revision_id is not None:
3662
# we used get_ancestry to determine source_ids then we are assured all
3663
# revisions referenced are present as they are installed in topological order.
3664
# and the tip revision was validated by get_ancestry.
3665
result_set = required_revisions
3667
# if we just grabbed the possibly available ids, then
3668
# we only have an estimate of whats available and need to validate
3669
# that against the revision records.
3671
self.source._eliminate_revisions_not_present(required_revisions))
3672
return self.source.revision_ids_to_search_result(result_set)
3675
class InterKnitRepo(InterSameDataRepository):
3676
"""Optimised code paths between Knit based repositories."""
3679
def _get_repo_format_to_test(self):
3680
from bzrlib.repofmt import knitrepo
3681
return knitrepo.RepositoryFormatKnit1()
3684
def is_compatible(source, target):
3685
"""Be compatible with known Knit formats.
3687
We don't test for the stores being of specific types because that
3688
could lead to confusing results, and there is no need to be
3691
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3693
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3694
isinstance(target._format, RepositoryFormatKnit))
3695
except AttributeError:
3697
return are_knits and InterRepository._same_model(source, target)
3700
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3701
"""See InterRepository.missing_revision_ids()."""
3702
if revision_id is not None:
3703
source_ids = self.source.get_ancestry(revision_id)
3704
if source_ids[0] is not None:
3705
raise AssertionError()
3708
source_ids = self.source.all_revision_ids()
3709
source_ids_set = set(source_ids)
3710
# source_ids is the worst possible case we may need to pull.
3711
# now we want to filter source_ids against what we actually
3712
# have in target, but don't try to check for existence where we know
3713
# we do not have a revision as that would be pointless.
3714
target_ids = set(self.target.all_revision_ids())
3715
possibly_present_revisions = target_ids.intersection(source_ids_set)
3716
actually_present_revisions = set(
3717
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3718
required_revisions = source_ids_set.difference(actually_present_revisions)
3719
if revision_id is not None:
3720
# we used get_ancestry to determine source_ids then we are assured all
3721
# revisions referenced are present as they are installed in topological order.
3722
# and the tip revision was validated by get_ancestry.
3723
result_set = required_revisions
3725
# if we just grabbed the possibly available ids, then
3726
# we only have an estimate of whats available and need to validate
3727
# that against the revision records.
3729
self.source._eliminate_revisions_not_present(required_revisions))
3730
return self.source.revision_ids_to_search_result(result_set)
3733
class InterDifferingSerializer(InterRepository):
3736
def _get_repo_format_to_test(self):
3740
def is_compatible(source, target):
3741
"""Be compatible with Knit2 source and Knit3 target"""
3742
# This is redundant with format.check_conversion_target(), however that
3743
# raises an exception, and we just want to say "False" as in we won't
3744
# support converting between these formats.
3745
if 'IDS_never' in debug.debug_flags:
3747
if source.supports_rich_root() and not target.supports_rich_root():
3749
if (source._format.supports_tree_reference
3750
and not target._format.supports_tree_reference):
3752
if target._fallback_repositories and target._format.supports_chks:
3753
# IDS doesn't know how to copy CHKs for the parent inventories it
3754
# adds to stacked repos.
3756
if 'IDS_always' in debug.debug_flags:
3758
# Only use this code path for local source and target. IDS does far
3759
# too much IO (both bandwidth and roundtrips) over a network.
3760
if not source.bzrdir.transport.base.startswith('file:///'):
3762
if not target.bzrdir.transport.base.startswith('file:///'):
3766
def _get_trees(self, revision_ids, cache):
3768
for rev_id in revision_ids:
3770
possible_trees.append((rev_id, cache[rev_id]))
3772
# Not cached, but inventory might be present anyway.
3774
tree = self.source.revision_tree(rev_id)
3775
except errors.NoSuchRevision:
3776
# Nope, parent is ghost.
3779
cache[rev_id] = tree
3780
possible_trees.append((rev_id, tree))
3781
return possible_trees
3783
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3784
"""Get the best delta and base for this revision.
3786
:return: (basis_id, delta)
3789
# Generate deltas against each tree, to find the shortest.
3790
texts_possibly_new_in_tree = set()
3791
for basis_id, basis_tree in possible_trees:
3792
delta = tree.inventory._make_delta(basis_tree.inventory)
3793
for old_path, new_path, file_id, new_entry in delta:
3794
if new_path is None:
3795
# This file_id isn't present in the new rev, so we don't
3799
# Rich roots are handled elsewhere...
3801
kind = new_entry.kind
3802
if kind != 'directory' and kind != 'file':
3803
# No text record associated with this inventory entry.
3805
# This is a directory or file that has changed somehow.
3806
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3807
deltas.append((len(delta), basis_id, delta))
3809
return deltas[0][1:]
3811
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3812
"""Find all parent revisions that are absent, but for which the
3813
inventory is present, and copy those inventories.
3815
This is necessary to preserve correctness when the source is stacked
3816
without fallbacks configured. (Note that in cases like upgrade the
3817
source may be not have _fallback_repositories even though it is
3821
for parents in parent_map.values():
3822
parent_revs.update(parents)
3823
present_parents = self.source.get_parent_map(parent_revs)
3824
absent_parents = set(parent_revs).difference(present_parents)
3825
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3826
(rev_id,) for rev_id in absent_parents)
3827
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3828
for parent_tree in self.source.revision_trees(parent_inv_ids):
3829
current_revision_id = parent_tree.get_revision_id()
3830
parents_parents_keys = parent_invs_keys_for_stacking[
3831
(current_revision_id,)]
3832
parents_parents = [key[-1] for key in parents_parents_keys]
3833
basis_id = _mod_revision.NULL_REVISION
3834
basis_tree = self.source.revision_tree(basis_id)
3835
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3836
self.target.add_inventory_by_delta(
3837
basis_id, delta, current_revision_id, parents_parents)
3838
cache[current_revision_id] = parent_tree
3840
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3841
"""Fetch across a few revisions.
3843
:param revision_ids: The revisions to copy
3844
:param basis_id: The revision_id of a tree that must be in cache, used
3845
as a basis for delta when no other base is available
3846
:param cache: A cache of RevisionTrees that we can use.
3847
:param a_graph: A Graph object to determine the heads() of the
3848
rich-root data stream.
3849
:return: The revision_id of the last converted tree. The RevisionTree
3850
for it will be in cache
3852
# Walk though all revisions; get inventory deltas, copy referenced
3853
# texts that delta references, insert the delta, revision and
3855
root_keys_to_create = set()
3858
pending_revisions = []
3859
parent_map = self.source.get_parent_map(revision_ids)
3860
self._fetch_parent_invs_for_stacking(parent_map, cache)
3861
self.source._safe_to_return_from_cache = True
3862
for tree in self.source.revision_trees(revision_ids):
3863
# Find a inventory delta for this revision.
3864
# Find text entries that need to be copied, too.
3865
current_revision_id = tree.get_revision_id()
3866
parent_ids = parent_map.get(current_revision_id, ())
3867
parent_trees = self._get_trees(parent_ids, cache)
3868
possible_trees = list(parent_trees)
3869
if len(possible_trees) == 0:
3870
# There either aren't any parents, or the parents are ghosts,
3871
# so just use the last converted tree.
3872
possible_trees.append((basis_id, cache[basis_id]))
3873
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3875
revision = self.source.get_revision(current_revision_id)
3876
pending_deltas.append((basis_id, delta,
3877
current_revision_id, revision.parent_ids))
3878
if self._converting_to_rich_root:
3879
self._revision_id_to_root_id[current_revision_id] = \
3881
# Determine which texts are in present in this revision but not in
3882
# any of the available parents.
3883
texts_possibly_new_in_tree = set()
3884
for old_path, new_path, file_id, entry in delta:
3885
if new_path is None:
3886
# This file_id isn't present in the new rev
3890
if not self.target.supports_rich_root():
3891
# The target doesn't support rich root, so we don't
3894
if self._converting_to_rich_root:
3895
# This can't be copied normally, we have to insert
3897
root_keys_to_create.add((file_id, entry.revision))
3900
texts_possibly_new_in_tree.add((file_id, entry.revision))
3901
for basis_id, basis_tree in possible_trees:
3902
basis_inv = basis_tree.inventory
3903
for file_key in list(texts_possibly_new_in_tree):
3904
file_id, file_revision = file_key
3906
entry = basis_inv[file_id]
3907
except errors.NoSuchId:
3909
if entry.revision == file_revision:
3910
texts_possibly_new_in_tree.remove(file_key)
3911
text_keys.update(texts_possibly_new_in_tree)
3912
pending_revisions.append(revision)
3913
cache[current_revision_id] = tree
3914
basis_id = current_revision_id
3915
self.source._safe_to_return_from_cache = False
3917
from_texts = self.source.texts
3918
to_texts = self.target.texts
3919
if root_keys_to_create:
3920
root_stream = _mod_fetch._new_root_data_stream(
3921
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3922
self.source, graph=a_graph)
3923
to_texts.insert_record_stream(root_stream)
3924
to_texts.insert_record_stream(from_texts.get_record_stream(
3925
text_keys, self.target._format._fetch_order,
3926
not self.target._format._fetch_uses_deltas))
3927
# insert inventory deltas
3928
for delta in pending_deltas:
3929
self.target.add_inventory_by_delta(*delta)
3930
if self.target._fallback_repositories:
3931
# Make sure this stacked repository has all the parent inventories
3932
# for the new revisions that we are about to insert. We do this
3933
# before adding the revisions so that no revision is added until
3934
# all the inventories it may depend on are added.
3935
# Note that this is overzealous, as we may have fetched these in an
3938
revision_ids = set()
3939
for revision in pending_revisions:
3940
revision_ids.add(revision.revision_id)
3941
parent_ids.update(revision.parent_ids)
3942
parent_ids.difference_update(revision_ids)
3943
parent_ids.discard(_mod_revision.NULL_REVISION)
3944
parent_map = self.source.get_parent_map(parent_ids)
3945
# we iterate over parent_map and not parent_ids because we don't
3946
# want to try copying any revision which is a ghost
3947
for parent_tree in self.source.revision_trees(parent_map):
3948
current_revision_id = parent_tree.get_revision_id()
3949
parents_parents = parent_map[current_revision_id]
3950
possible_trees = self._get_trees(parents_parents, cache)
3951
if len(possible_trees) == 0:
3952
# There either aren't any parents, or the parents are
3953
# ghosts, so just use the last converted tree.
3954
possible_trees.append((basis_id, cache[basis_id]))
3955
basis_id, delta = self._get_delta_for_revision(parent_tree,
3956
parents_parents, possible_trees)
3957
self.target.add_inventory_by_delta(
3958
basis_id, delta, current_revision_id, parents_parents)
3959
# insert signatures and revisions
3960
for revision in pending_revisions:
3962
signature = self.source.get_signature_text(
3963
revision.revision_id)
3964
self.target.add_signature_text(revision.revision_id,
3966
except errors.NoSuchRevision:
3968
self.target.add_revision(revision.revision_id, revision)
3971
def _fetch_all_revisions(self, revision_ids, pb):
3972
"""Fetch everything for the list of revisions.
3974
:param revision_ids: The list of revisions to fetch. Must be in
3976
:param pb: A ProgressTask
3979
basis_id, basis_tree = self._get_basis(revision_ids[0])
3981
cache = lru_cache.LRUCache(100)
3982
cache[basis_id] = basis_tree
3983
del basis_tree # We don't want to hang on to it here
3985
if self._converting_to_rich_root and len(revision_ids) > 100:
3986
a_graph = _mod_fetch._get_rich_root_heads_graph(self.source,
3991
for offset in range(0, len(revision_ids), batch_size):
3992
self.target.start_write_group()
3994
pb.update('Transferring revisions', offset,
3996
batch = revision_ids[offset:offset+batch_size]
3997
basis_id = self._fetch_batch(batch, basis_id, cache,
4000
self.source._safe_to_return_from_cache = False
4001
self.target.abort_write_group()
4004
hint = self.target.commit_write_group()
4007
if hints and self.target._format.pack_compresses:
4008
self.target.pack(hint=hints)
4009
pb.update('Transferring revisions', len(revision_ids),
4013
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
4015
"""See InterRepository.fetch()."""
4016
if fetch_spec is not None:
4017
raise AssertionError("Not implemented yet...")
4018
ui.ui_factory.warn_experimental_format_fetch(self)
4019
if (not self.source.supports_rich_root()
4020
and self.target.supports_rich_root()):
4021
self._converting_to_rich_root = True
4022
self._revision_id_to_root_id = {}
4024
self._converting_to_rich_root = False
4025
# See <https://launchpad.net/bugs/456077> asking for a warning here
4026
if self.source._format.network_name() != self.target._format.network_name():
4027
ui.ui_factory.show_user_warning('cross_format_fetch',
4028
from_format=self.source._format,
4029
to_format=self.target._format)
4030
revision_ids = self.target.search_missing_revision_ids(self.source,
4031
revision_id, find_ghosts=find_ghosts).get_keys()
4032
if not revision_ids:
4034
revision_ids = tsort.topo_sort(
4035
self.source.get_graph().get_parent_map(revision_ids))
4036
if not revision_ids:
4038
# Walk though all revisions; get inventory deltas, copy referenced
4039
# texts that delta references, insert the delta, revision and
4042
my_pb = ui.ui_factory.nested_progress_bar()
4045
symbol_versioning.warn(
4046
symbol_versioning.deprecated_in((1, 14, 0))
4047
% "pb parameter to fetch()")
4050
self._fetch_all_revisions(revision_ids, pb)
4052
if my_pb is not None:
4054
return len(revision_ids), 0
4056
def _get_basis(self, first_revision_id):
4057
"""Get a revision and tree which exists in the target.
4059
This assumes that first_revision_id is selected for transmission
4060
because all other ancestors are already present. If we can't find an
4061
ancestor we fall back to NULL_REVISION since we know that is safe.
4063
:return: (basis_id, basis_tree)
4065
first_rev = self.source.get_revision(first_revision_id)
4067
basis_id = first_rev.parent_ids[0]
4068
# only valid as a basis if the target has it
4069
self.target.get_revision(basis_id)
4070
# Try to get a basis tree - if its a ghost it will hit the
4071
# NoSuchRevision case.
4072
basis_tree = self.source.revision_tree(basis_id)
4073
except (IndexError, errors.NoSuchRevision):
4074
basis_id = _mod_revision.NULL_REVISION
4075
basis_tree = self.source.revision_tree(basis_id)
4076
return basis_id, basis_tree
4079
InterRepository.register_optimiser(InterDifferingSerializer)
4080
InterRepository.register_optimiser(InterSameDataRepository)
4081
InterRepository.register_optimiser(InterWeaveRepo)
4082
InterRepository.register_optimiser(InterKnitRepo)
1788
4085
class CopyConverter(object):
1789
4086
"""A repository conversion tool which just performs a copy of the content.
1844
4243
return revision_graph
4246
class StreamSink(object):
4247
"""An object that can insert a stream into a repository.
4249
This interface handles the complexity of reserialising inventories and
4250
revisions from different formats, and allows unidirectional insertion into
4251
stacked repositories without looking for the missing basis parents
4255
def __init__(self, target_repo):
4256
self.target_repo = target_repo
4258
def insert_stream(self, stream, src_format, resume_tokens):
4259
"""Insert a stream's content into the target repository.
4261
:param src_format: a bzr repository format.
4263
:return: a list of resume tokens and an iterable of keys additional
4264
items required before the insertion can be completed.
4266
self.target_repo.lock_write()
4269
self.target_repo.resume_write_group(resume_tokens)
4272
self.target_repo.start_write_group()
4275
# locked_insert_stream performs a commit|suspend.
4276
return self._locked_insert_stream(stream, src_format, is_resume)
4278
self.target_repo.abort_write_group(suppress_errors=True)
4281
self.target_repo.unlock()
4283
def _locked_insert_stream(self, stream, src_format, is_resume):
4284
to_serializer = self.target_repo._format._serializer
4285
src_serializer = src_format._serializer
4287
if to_serializer == src_serializer:
4288
# If serializers match and the target is a pack repository, set the
4289
# write cache size on the new pack. This avoids poor performance
4290
# on transports where append is unbuffered (such as
4291
# RemoteTransport). This is safe to do because nothing should read
4292
# back from the target repository while a stream with matching
4293
# serialization is being inserted.
4294
# The exception is that a delta record from the source that should
4295
# be a fulltext may need to be expanded by the target (see
4296
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4297
# explicitly flush any buffered writes first in that rare case.
4299
new_pack = self.target_repo._pack_collection._new_pack
4300
except AttributeError:
4301
# Not a pack repository
4304
new_pack.set_write_cache_size(1024*1024)
4305
for substream_type, substream in stream:
4306
if 'stream' in debug.debug_flags:
4307
mutter('inserting substream: %s', substream_type)
4308
if substream_type == 'texts':
4309
self.target_repo.texts.insert_record_stream(substream)
4310
elif substream_type == 'inventories':
4311
if src_serializer == to_serializer:
4312
self.target_repo.inventories.insert_record_stream(
4315
self._extract_and_insert_inventories(
4316
substream, src_serializer)
4317
elif substream_type == 'inventory-deltas':
4318
self._extract_and_insert_inventory_deltas(
4319
substream, src_serializer)
4320
elif substream_type == 'chk_bytes':
4321
# XXX: This doesn't support conversions, as it assumes the
4322
# conversion was done in the fetch code.
4323
self.target_repo.chk_bytes.insert_record_stream(substream)
4324
elif substream_type == 'revisions':
4325
# This may fallback to extract-and-insert more often than
4326
# required if the serializers are different only in terms of
4328
if src_serializer == to_serializer:
4329
self.target_repo.revisions.insert_record_stream(
4332
self._extract_and_insert_revisions(substream,
4334
elif substream_type == 'signatures':
4335
self.target_repo.signatures.insert_record_stream(substream)
4337
raise AssertionError('kaboom! %s' % (substream_type,))
4338
# Done inserting data, and the missing_keys calculations will try to
4339
# read back from the inserted data, so flush the writes to the new pack
4340
# (if this is pack format).
4341
if new_pack is not None:
4342
new_pack._write_data('', flush=True)
4343
# Find all the new revisions (including ones from resume_tokens)
4344
missing_keys = self.target_repo.get_missing_parent_inventories(
4345
check_for_missing_texts=is_resume)
4347
for prefix, versioned_file in (
4348
('texts', self.target_repo.texts),
4349
('inventories', self.target_repo.inventories),
4350
('revisions', self.target_repo.revisions),
4351
('signatures', self.target_repo.signatures),
4352
('chk_bytes', self.target_repo.chk_bytes),
4354
if versioned_file is None:
4356
# TODO: key is often going to be a StaticTuple object
4357
# I don't believe we can define a method by which
4358
# (prefix,) + StaticTuple will work, though we could
4359
# define a StaticTuple.sq_concat that would allow you to
4360
# pass in either a tuple or a StaticTuple as the second
4361
# object, so instead we could have:
4362
# StaticTuple(prefix) + key here...
4363
missing_keys.update((prefix,) + key for key in
4364
versioned_file.get_missing_compression_parent_keys())
4365
except NotImplementedError:
4366
# cannot even attempt suspending, and missing would have failed
4367
# during stream insertion.
4368
missing_keys = set()
4371
# suspend the write group and tell the caller what we is
4372
# missing. We know we can suspend or else we would not have
4373
# entered this code path. (All repositories that can handle
4374
# missing keys can handle suspending a write group).
4375
write_group_tokens = self.target_repo.suspend_write_group()
4376
return write_group_tokens, missing_keys
4377
hint = self.target_repo.commit_write_group()
4378
if (to_serializer != src_serializer and
4379
self.target_repo._format.pack_compresses):
4380
self.target_repo.pack(hint=hint)
4383
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4384
target_rich_root = self.target_repo._format.rich_root_data
4385
target_tree_refs = self.target_repo._format.supports_tree_reference
4386
for record in substream:
4387
# Insert the delta directly
4388
inventory_delta_bytes = record.get_bytes_as('fulltext')
4389
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4391
parse_result = deserialiser.parse_text_bytes(
4392
inventory_delta_bytes)
4393
except inventory_delta.IncompatibleInventoryDelta, err:
4394
trace.mutter("Incompatible delta: %s", err.msg)
4395
raise errors.IncompatibleRevision(self.target_repo._format)
4396
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4397
revision_id = new_id
4398
parents = [key[0] for key in record.parents]
4399
self.target_repo.add_inventory_by_delta(
4400
basis_id, inv_delta, revision_id, parents)
4402
def _extract_and_insert_inventories(self, substream, serializer,
4404
"""Generate a new inventory versionedfile in target, converting data.
4406
The inventory is retrieved from the source, (deserializing it), and
4407
stored in the target (reserializing it in a different format).
4409
target_rich_root = self.target_repo._format.rich_root_data
4410
target_tree_refs = self.target_repo._format.supports_tree_reference
4411
for record in substream:
4412
# It's not a delta, so it must be a fulltext in the source
4413
# serializer's format.
4414
bytes = record.get_bytes_as('fulltext')
4415
revision_id = record.key[0]
4416
inv = serializer.read_inventory_from_string(bytes, revision_id)
4417
parents = [key[0] for key in record.parents]
4418
self.target_repo.add_inventory(revision_id, inv, parents)
4419
# No need to keep holding this full inv in memory when the rest of
4420
# the substream is likely to be all deltas.
4423
def _extract_and_insert_revisions(self, substream, serializer):
4424
for record in substream:
4425
bytes = record.get_bytes_as('fulltext')
4426
revision_id = record.key[0]
4427
rev = serializer.read_revision_from_string(bytes)
4428
if rev.revision_id != revision_id:
4429
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4430
self.target_repo.add_revision(revision_id, rev)
4433
if self.target_repo._format._fetch_reconcile:
4434
self.target_repo.reconcile()
4437
class StreamSource(object):
4438
"""A source of a stream for fetching between repositories."""
4440
def __init__(self, from_repository, to_format):
4441
"""Create a StreamSource streaming from from_repository."""
4442
self.from_repository = from_repository
4443
self.to_format = to_format
4445
def delta_on_metadata(self):
4446
"""Return True if delta's are permitted on metadata streams.
4448
That is on revisions and signatures.
4450
src_serializer = self.from_repository._format._serializer
4451
target_serializer = self.to_format._serializer
4452
return (self.to_format._fetch_uses_deltas and
4453
src_serializer == target_serializer)
4455
def _fetch_revision_texts(self, revs):
4456
# fetch signatures first and then the revision texts
4457
# may need to be a InterRevisionStore call here.
4458
from_sf = self.from_repository.signatures
4459
# A missing signature is just skipped.
4460
keys = [(rev_id,) for rev_id in revs]
4461
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4463
self.to_format._fetch_order,
4464
not self.to_format._fetch_uses_deltas))
4465
# If a revision has a delta, this is actually expanded inside the
4466
# insert_record_stream code now, which is an alternate fix for
4468
from_rf = self.from_repository.revisions
4469
revisions = from_rf.get_record_stream(
4471
self.to_format._fetch_order,
4472
not self.delta_on_metadata())
4473
return [('signatures', signatures), ('revisions', revisions)]
4475
def _generate_root_texts(self, revs):
4476
"""This will be called by get_stream between fetching weave texts and
4477
fetching the inventory weave.
4479
if self._rich_root_upgrade():
4480
return _mod_fetch.Inter1and2Helper(
4481
self.from_repository).generate_root_texts(revs)
4485
def get_stream(self, search):
4487
revs = search.get_keys()
4488
graph = self.from_repository.get_graph()
4489
revs = tsort.topo_sort(graph.get_parent_map(revs))
4490
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4492
for knit_kind, file_id, revisions in data_to_fetch:
4493
if knit_kind != phase:
4495
# Make a new progress bar for this phase
4496
if knit_kind == "file":
4497
# Accumulate file texts
4498
text_keys.extend([(file_id, revision) for revision in
4500
elif knit_kind == "inventory":
4501
# Now copy the file texts.
4502
from_texts = self.from_repository.texts
4503
yield ('texts', from_texts.get_record_stream(
4504
text_keys, self.to_format._fetch_order,
4505
not self.to_format._fetch_uses_deltas))
4506
# Cause an error if a text occurs after we have done the
4509
# Before we process the inventory we generate the root
4510
# texts (if necessary) so that the inventories references
4512
for _ in self._generate_root_texts(revs):
4514
# we fetch only the referenced inventories because we do not
4515
# know for unselected inventories whether all their required
4516
# texts are present in the other repository - it could be
4518
for info in self._get_inventory_stream(revs):
4520
elif knit_kind == "signatures":
4521
# Nothing to do here; this will be taken care of when
4522
# _fetch_revision_texts happens.
4524
elif knit_kind == "revisions":
4525
for record in self._fetch_revision_texts(revs):
4528
raise AssertionError("Unknown knit kind %r" % knit_kind)
4530
def get_stream_for_missing_keys(self, missing_keys):
4531
# missing keys can only occur when we are byte copying and not
4532
# translating (because translation means we don't send
4533
# unreconstructable deltas ever).
4535
keys['texts'] = set()
4536
keys['revisions'] = set()
4537
keys['inventories'] = set()
4538
keys['chk_bytes'] = set()
4539
keys['signatures'] = set()
4540
for key in missing_keys:
4541
keys[key[0]].add(key[1:])
4542
if len(keys['revisions']):
4543
# If we allowed copying revisions at this point, we could end up
4544
# copying a revision without copying its required texts: a
4545
# violation of the requirements for repository integrity.
4546
raise AssertionError(
4547
'cannot copy revisions to fill in missing deltas %s' % (
4548
keys['revisions'],))
4549
for substream_kind, keys in keys.iteritems():
4550
vf = getattr(self.from_repository, substream_kind)
4551
if vf is None and keys:
4552
raise AssertionError(
4553
"cannot fill in keys for a versioned file we don't"
4554
" have: %s needs %s" % (substream_kind, keys))
4556
# No need to stream something we don't have
4558
if substream_kind == 'inventories':
4559
# Some missing keys are genuinely ghosts, filter those out.
4560
present = self.from_repository.inventories.get_parent_map(keys)
4561
revs = [key[0] for key in present]
4562
# Get the inventory stream more-or-less as we do for the
4563
# original stream; there's no reason to assume that records
4564
# direct from the source will be suitable for the sink. (Think
4565
# e.g. 2a -> 1.9-rich-root).
4566
for info in self._get_inventory_stream(revs, missing=True):
4570
# Ask for full texts always so that we don't need more round trips
4571
# after this stream.
4572
# Some of the missing keys are genuinely ghosts, so filter absent
4573
# records. The Sink is responsible for doing another check to
4574
# ensure that ghosts don't introduce missing data for future
4576
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4577
self.to_format._fetch_order, True))
4578
yield substream_kind, stream
4580
def inventory_fetch_order(self):
4581
if self._rich_root_upgrade():
4582
return 'topological'
4584
return self.to_format._fetch_order
4586
def _rich_root_upgrade(self):
4587
return (not self.from_repository._format.rich_root_data and
4588
self.to_format.rich_root_data)
4590
def _get_inventory_stream(self, revision_ids, missing=False):
4591
from_format = self.from_repository._format
4592
if (from_format.supports_chks and self.to_format.supports_chks and
4593
from_format.network_name() == self.to_format.network_name()):
4594
raise AssertionError(
4595
"this case should be handled by GroupCHKStreamSource")
4596
elif 'forceinvdeltas' in debug.debug_flags:
4597
return self._get_convertable_inventory_stream(revision_ids,
4598
delta_versus_null=missing)
4599
elif from_format.network_name() == self.to_format.network_name():
4601
return self._get_simple_inventory_stream(revision_ids,
4603
elif (not from_format.supports_chks and not self.to_format.supports_chks
4604
and from_format._serializer == self.to_format._serializer):
4605
# Essentially the same format.
4606
return self._get_simple_inventory_stream(revision_ids,
4609
# Any time we switch serializations, we want to use an
4610
# inventory-delta based approach.
4611
return self._get_convertable_inventory_stream(revision_ids,
4612
delta_versus_null=missing)
4614
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4615
# NB: This currently reopens the inventory weave in source;
4616
# using a single stream interface instead would avoid this.
4617
from_weave = self.from_repository.inventories
4619
delta_closure = True
4621
delta_closure = not self.delta_on_metadata()
4622
yield ('inventories', from_weave.get_record_stream(
4623
[(rev_id,) for rev_id in revision_ids],
4624
self.inventory_fetch_order(), delta_closure))
4626
def _get_convertable_inventory_stream(self, revision_ids,
4627
delta_versus_null=False):
4628
# The two formats are sufficiently different that there is no fast
4629
# path, so we need to send just inventorydeltas, which any
4630
# sufficiently modern client can insert into any repository.
4631
# The StreamSink code expects to be able to
4632
# convert on the target, so we need to put bytes-on-the-wire that can
4633
# be converted. That means inventory deltas (if the remote is <1.19,
4634
# RemoteStreamSink will fallback to VFS to insert the deltas).
4635
yield ('inventory-deltas',
4636
self._stream_invs_as_deltas(revision_ids,
4637
delta_versus_null=delta_versus_null))
4639
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4640
"""Return a stream of inventory-deltas for the given rev ids.
4642
:param revision_ids: The list of inventories to transmit
4643
:param delta_versus_null: Don't try to find a minimal delta for this
4644
entry, instead compute the delta versus the NULL_REVISION. This
4645
effectively streams a complete inventory. Used for stuff like
4646
filling in missing parents, etc.
4648
from_repo = self.from_repository
4649
revision_keys = [(rev_id,) for rev_id in revision_ids]
4650
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4651
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4653
inventories = self.from_repository.iter_inventories(
4654
revision_ids, 'topological')
4655
format = from_repo._format
4656
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4657
inventory_cache = lru_cache.LRUCache(50)
4658
null_inventory = from_repo.revision_tree(
4659
_mod_revision.NULL_REVISION).inventory
4660
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4661
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4662
# repo back into a non-rich-root repo ought to be allowed)
4663
serializer = inventory_delta.InventoryDeltaSerializer(
4664
versioned_root=format.rich_root_data,
4665
tree_references=format.supports_tree_reference)
4666
for inv in inventories:
4667
key = (inv.revision_id,)
4668
parent_keys = parent_map.get(key, ())
4670
if not delta_versus_null and parent_keys:
4671
# The caller did not ask for complete inventories and we have
4672
# some parents that we can delta against. Make a delta against
4673
# each parent so that we can find the smallest.
4674
parent_ids = [parent_key[0] for parent_key in parent_keys]
4675
for parent_id in parent_ids:
4676
if parent_id not in invs_sent_so_far:
4677
# We don't know that the remote side has this basis, so
4680
if parent_id == _mod_revision.NULL_REVISION:
4681
parent_inv = null_inventory
4683
parent_inv = inventory_cache.get(parent_id, None)
4684
if parent_inv is None:
4685
parent_inv = from_repo.get_inventory(parent_id)
4686
candidate_delta = inv._make_delta(parent_inv)
4687
if (delta is None or
4688
len(delta) > len(candidate_delta)):
4689
delta = candidate_delta
4690
basis_id = parent_id
4692
# Either none of the parents ended up being suitable, or we
4693
# were asked to delta against NULL
4694
basis_id = _mod_revision.NULL_REVISION
4695
delta = inv._make_delta(null_inventory)
4696
invs_sent_so_far.add(inv.revision_id)
4697
inventory_cache[inv.revision_id] = inv
4698
delta_serialized = ''.join(
4699
serializer.delta_to_lines(basis_id, key[-1], delta))
4700
yield versionedfile.FulltextContentFactory(
4701
key, parent_keys, None, delta_serialized)
1847
4704
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
1848
4705
stop_revision=None):
1849
4706
"""Extend the partial history to include a given index