202
333
commit to be valid, deletes against the basis MUST be recorded via
203
334
builder.record_delete().
205
raise NotImplementedError(self.will_record_deletes)
207
def record_iter_changes(self, tree, basis_revision_id, iter_changes):
336
self._recording_deletes = True
338
basis_id = self.parents[0]
340
basis_id = _mod_revision.NULL_REVISION
341
self.basis_delta_revision = basis_id
343
def record_entry_contents(self, ie, parent_invs, path, tree,
345
"""Record the content of ie from tree into the commit if needed.
347
Side effect: sets ie.revision when unchanged
349
:param ie: An inventory entry present in the commit.
350
:param parent_invs: The inventories of the parent revisions of the
352
:param path: The path the entry is at in the tree.
353
:param tree: The tree which contains this entry and should be used to
355
:param content_summary: Summary data from the tree about the paths
356
content - stat, length, exec, sha/link target. This is only
357
accessed when the entry has a revision of None - that is when it is
358
a candidate to commit.
359
:return: A tuple (change_delta, version_recorded, fs_hash).
360
change_delta is an inventory_delta change for this entry against
361
the basis tree of the commit, or None if no change occured against
363
version_recorded is True if a new version of the entry has been
364
recorded. For instance, committing a merge where a file was only
365
changed on the other side will return (delta, False).
366
fs_hash is either None, or the hash details for the path (currently
367
a tuple of the contents sha1 and the statvalue returned by
368
tree.get_file_with_stat()).
370
if self.new_inventory.root is None:
371
if ie.parent_id is not None:
372
raise errors.RootMissing()
373
self._check_root(ie, parent_invs, tree)
374
if ie.revision is None:
375
kind = content_summary[0]
377
# ie is carried over from a prior commit
379
# XXX: repository specific check for nested tree support goes here - if
380
# the repo doesn't want nested trees we skip it ?
381
if (kind == 'tree-reference' and
382
not self.repository._format.supports_tree_reference):
383
# mismatch between commit builder logic and repository:
384
# this needs the entry creation pushed down into the builder.
385
raise NotImplementedError('Missing repository subtree support.')
386
self.new_inventory.add(ie)
388
# TODO: slow, take it out of the inner loop.
390
basis_inv = parent_invs[0]
392
basis_inv = Inventory(root_id=None)
394
# ie.revision is always None if the InventoryEntry is considered
395
# for committing. We may record the previous parents revision if the
396
# content is actually unchanged against a sole head.
397
if ie.revision is not None:
398
if not self._versioned_root and path == '':
399
# repositories that do not version the root set the root's
400
# revision to the new commit even when no change occurs (more
401
# specifically, they do not record a revision on the root; and
402
# the rev id is assigned to the root during deserialisation -
403
# this masks when a change may have occurred against the basis.
404
# To match this we always issue a delta, because the revision
405
# of the root will always be changing.
406
if ie.file_id in basis_inv:
407
delta = (basis_inv.id2path(ie.file_id), path,
411
delta = (None, path, ie.file_id, ie)
412
self._basis_delta.append(delta)
413
return delta, False, None
415
# we don't need to commit this, because the caller already
416
# determined that an existing revision of this file is
417
# appropriate. If its not being considered for committing then
418
# it and all its parents to the root must be unaltered so
419
# no-change against the basis.
420
if ie.revision == self._new_revision_id:
421
raise AssertionError("Impossible situation, a skipped "
422
"inventory entry (%r) claims to be modified in this "
423
"commit (%r).", (ie, self._new_revision_id))
424
return None, False, None
425
# XXX: Friction: parent_candidates should return a list not a dict
426
# so that we don't have to walk the inventories again.
427
parent_candiate_entries = ie.parent_candidates(parent_invs)
428
head_set = self._heads(ie.file_id, parent_candiate_entries.keys())
430
for inv in parent_invs:
431
if ie.file_id in inv:
432
old_rev = inv[ie.file_id].revision
433
if old_rev in head_set:
434
heads.append(inv[ie.file_id].revision)
435
head_set.remove(inv[ie.file_id].revision)
438
# now we check to see if we need to write a new record to the
440
# We write a new entry unless there is one head to the ancestors, and
441
# the kind-derived content is unchanged.
443
# Cheapest check first: no ancestors, or more the one head in the
444
# ancestors, we write a new node.
448
# There is a single head, look it up for comparison
449
parent_entry = parent_candiate_entries[heads[0]]
450
# if the non-content specific data has changed, we'll be writing a
452
if (parent_entry.parent_id != ie.parent_id or
453
parent_entry.name != ie.name):
455
# now we need to do content specific checks:
457
# if the kind changed the content obviously has
458
if kind != parent_entry.kind:
460
# Stat cache fingerprint feedback for the caller - None as we usually
461
# don't generate one.
464
if content_summary[2] is None:
465
raise ValueError("Files must not have executable = None")
467
if (# if the file length changed we have to store:
468
parent_entry.text_size != content_summary[1] or
469
# if the exec bit has changed we have to store:
470
parent_entry.executable != content_summary[2]):
472
elif parent_entry.text_sha1 == content_summary[3]:
473
# all meta and content is unchanged (using a hash cache
474
# hit to check the sha)
475
ie.revision = parent_entry.revision
476
ie.text_size = parent_entry.text_size
477
ie.text_sha1 = parent_entry.text_sha1
478
ie.executable = parent_entry.executable
479
return self._get_delta(ie, basis_inv, path), False, None
481
# Either there is only a hash change(no hash cache entry,
482
# or same size content change), or there is no change on
484
# Provide the parent's hash to the store layer, so that the
485
# content is unchanged we will not store a new node.
486
nostore_sha = parent_entry.text_sha1
488
# We want to record a new node regardless of the presence or
489
# absence of a content change in the file.
491
ie.executable = content_summary[2]
492
file_obj, stat_value = tree.get_file_with_stat(ie.file_id, path)
494
text = file_obj.read()
498
ie.text_sha1, ie.text_size = self._add_text_to_weave(
499
ie.file_id, text, heads, nostore_sha)
500
# Let the caller know we generated a stat fingerprint.
501
fingerprint = (ie.text_sha1, stat_value)
502
except errors.ExistingContent:
503
# Turns out that the file content was unchanged, and we were
504
# only going to store a new node if it was changed. Carry over
506
ie.revision = parent_entry.revision
507
ie.text_size = parent_entry.text_size
508
ie.text_sha1 = parent_entry.text_sha1
509
ie.executable = parent_entry.executable
510
return self._get_delta(ie, basis_inv, path), False, None
511
elif kind == 'directory':
513
# all data is meta here, nothing specific to directory, so
515
ie.revision = parent_entry.revision
516
return self._get_delta(ie, basis_inv, path), False, None
517
self._add_text_to_weave(ie.file_id, '', heads, None)
518
elif kind == 'symlink':
519
current_link_target = content_summary[3]
521
# symlink target is not generic metadata, check if it has
523
if current_link_target != parent_entry.symlink_target:
526
# unchanged, carry over.
527
ie.revision = parent_entry.revision
528
ie.symlink_target = parent_entry.symlink_target
529
return self._get_delta(ie, basis_inv, path), False, None
530
ie.symlink_target = current_link_target
531
self._add_text_to_weave(ie.file_id, '', heads, None)
532
elif kind == 'tree-reference':
534
if content_summary[3] != parent_entry.reference_revision:
537
# unchanged, carry over.
538
ie.reference_revision = parent_entry.reference_revision
539
ie.revision = parent_entry.revision
540
return self._get_delta(ie, basis_inv, path), False, None
541
ie.reference_revision = content_summary[3]
542
self._add_text_to_weave(ie.file_id, '', heads, None)
544
raise NotImplementedError('unknown kind')
545
ie.revision = self._new_revision_id
546
self._any_changes = True
547
return self._get_delta(ie, basis_inv, path), True, fingerprint
549
def record_iter_changes(self, tree, basis_revision_id, iter_changes,
550
_entry_factory=entry_factory):
208
551
"""Record a new tree via iter_changes.
210
553
:param tree: The tree to obtain text contents from for changed objects.
215
558
to basis_revision_id. The iterator must not include any items with
216
559
a current kind of None - missing items must be either filtered out
217
560
or errored-on beefore record_iter_changes sees the item.
561
:param _entry_factory: Private method to bind entry_factory locally for
218
563
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
219
564
tree._observed_sha1.
221
raise NotImplementedError(self.record_iter_changes)
224
class RepositoryWriteLockResult(LogicalLockResult):
225
"""The result of write locking a repository.
227
:ivar repository_token: The token obtained from the underlying lock, or
229
:ivar unlock: A callable which will unlock the lock.
232
def __init__(self, unlock, repository_token):
233
LogicalLockResult.__init__(self, unlock)
234
self.repository_token = repository_token
237
return "RepositoryWriteLockResult(%s, %s)" % (self.repository_token,
566
# Create an inventory delta based on deltas between all the parents and
567
# deltas between all the parent inventories. We use inventory delta's
568
# between the inventory objects because iter_changes masks
569
# last-changed-field only changes.
571
# file_id -> change map, change is fileid, paths, changed, versioneds,
572
# parents, names, kinds, executables
574
# {file_id -> revision_id -> inventory entry, for entries in parent
575
# trees that are not parents[0]
579
revtrees = list(self.repository.revision_trees(self.parents))
580
except errors.NoSuchRevision:
581
# one or more ghosts, slow path.
583
for revision_id in self.parents:
585
revtrees.append(self.repository.revision_tree(revision_id))
586
except errors.NoSuchRevision:
588
basis_revision_id = _mod_revision.NULL_REVISION
590
revtrees.append(self.repository.revision_tree(
591
_mod_revision.NULL_REVISION))
592
# The basis inventory from a repository
594
basis_inv = revtrees[0].inventory
596
basis_inv = self.repository.revision_tree(
597
_mod_revision.NULL_REVISION).inventory
598
if len(self.parents) > 0:
599
if basis_revision_id != self.parents[0] and not ghost_basis:
601
"arbitrary basis parents not yet supported with merges")
602
for revtree in revtrees[1:]:
603
for change in revtree.inventory._make_delta(basis_inv):
604
if change[1] is None:
605
# Not present in this parent.
607
if change[2] not in merged_ids:
608
if change[0] is not None:
609
basis_entry = basis_inv[change[2]]
610
merged_ids[change[2]] = [
612
basis_entry.revision,
615
parent_entries[change[2]] = {
617
basis_entry.revision:basis_entry,
619
change[3].revision:change[3],
622
merged_ids[change[2]] = [change[3].revision]
623
parent_entries[change[2]] = {change[3].revision:change[3]}
625
merged_ids[change[2]].append(change[3].revision)
626
parent_entries[change[2]][change[3].revision] = change[3]
629
# Setup the changes from the tree:
630
# changes maps file_id -> (change, [parent revision_ids])
632
for change in iter_changes:
633
# This probably looks up in basis_inv way to much.
634
if change[1][0] is not None:
635
head_candidate = [basis_inv[change[0]].revision]
638
changes[change[0]] = change, merged_ids.get(change[0],
640
unchanged_merged = set(merged_ids) - set(changes)
641
# Extend the changes dict with synthetic changes to record merges of
643
for file_id in unchanged_merged:
644
# Record a merged version of these items that did not change vs the
645
# basis. This can be either identical parallel changes, or a revert
646
# of a specific file after a merge. The recorded content will be
647
# that of the current tree (which is the same as the basis), but
648
# the per-file graph will reflect a merge.
649
# NB:XXX: We are reconstructing path information we had, this
650
# should be preserved instead.
651
# inv delta change: (file_id, (path_in_source, path_in_target),
652
# changed_content, versioned, parent, name, kind,
655
basis_entry = basis_inv[file_id]
656
except errors.NoSuchId:
657
# a change from basis->some_parents but file_id isn't in basis
658
# so was new in the merge, which means it must have changed
659
# from basis -> current, and as it hasn't the add was reverted
660
# by the user. So we discard this change.
664
(basis_inv.id2path(file_id), tree.id2path(file_id)),
666
(basis_entry.parent_id, basis_entry.parent_id),
667
(basis_entry.name, basis_entry.name),
668
(basis_entry.kind, basis_entry.kind),
669
(basis_entry.executable, basis_entry.executable))
670
changes[file_id] = (change, merged_ids[file_id])
671
# changes contains tuples with the change and a set of inventory
672
# candidates for the file.
674
# old_path, new_path, file_id, new_inventory_entry
675
seen_root = False # Is the root in the basis delta?
676
inv_delta = self._basis_delta
677
modified_rev = self._new_revision_id
678
for change, head_candidates in changes.values():
679
if change[3][1]: # versioned in target.
680
# Several things may be happening here:
681
# We may have a fork in the per-file graph
682
# - record a change with the content from tree
683
# We may have a change against < all trees
684
# - carry over the tree that hasn't changed
685
# We may have a change against all trees
686
# - record the change with the content from tree
689
entry = _entry_factory[kind](file_id, change[5][1],
691
head_set = self._heads(change[0], set(head_candidates))
694
for head_candidate in head_candidates:
695
if head_candidate in head_set:
696
heads.append(head_candidate)
697
head_set.remove(head_candidate)
700
# Could be a carry-over situation:
701
parent_entry_revs = parent_entries.get(file_id, None)
702
if parent_entry_revs:
703
parent_entry = parent_entry_revs.get(heads[0], None)
706
if parent_entry is None:
707
# The parent iter_changes was called against is the one
708
# that is the per-file head, so any change is relevant
709
# iter_changes is valid.
710
carry_over_possible = False
712
# could be a carry over situation
713
# A change against the basis may just indicate a merge,
714
# we need to check the content against the source of the
715
# merge to determine if it was changed after the merge
717
if (parent_entry.kind != entry.kind or
718
parent_entry.parent_id != entry.parent_id or
719
parent_entry.name != entry.name):
720
# Metadata common to all entries has changed
721
# against per-file parent
722
carry_over_possible = False
724
carry_over_possible = True
725
# per-type checks for changes against the parent_entry
728
# Cannot be a carry-over situation
729
carry_over_possible = False
730
# Populate the entry in the delta
732
# XXX: There is still a small race here: If someone reverts the content of a file
733
# after iter_changes examines and decides it has changed,
734
# we will unconditionally record a new version even if some
735
# other process reverts it while commit is running (with
736
# the revert happening after iter_changes did it's
739
entry.executable = True
741
entry.executable = False
742
if (carry_over_possible and
743
parent_entry.executable == entry.executable):
744
# Check the file length, content hash after reading
746
nostore_sha = parent_entry.text_sha1
749
file_obj, stat_value = tree.get_file_with_stat(file_id, change[1][1])
751
text = file_obj.read()
755
entry.text_sha1, entry.text_size = self._add_text_to_weave(
756
file_id, text, heads, nostore_sha)
757
yield file_id, change[1][1], (entry.text_sha1, stat_value)
758
except errors.ExistingContent:
759
# No content change against a carry_over parent
760
# Perhaps this should also yield a fs hash update?
762
entry.text_size = parent_entry.text_size
763
entry.text_sha1 = parent_entry.text_sha1
764
elif kind == 'symlink':
766
entry.symlink_target = tree.get_symlink_target(file_id)
767
if (carry_over_possible and
768
parent_entry.symlink_target == entry.symlink_target):
771
self._add_text_to_weave(change[0], '', heads, None)
772
elif kind == 'directory':
773
if carry_over_possible:
776
# Nothing to set on the entry.
777
# XXX: split into the Root and nonRoot versions.
778
if change[1][1] != '' or self.repository.supports_rich_root():
779
self._add_text_to_weave(change[0], '', heads, None)
780
elif kind == 'tree-reference':
781
if not self.repository._format.supports_tree_reference:
782
# This isn't quite sane as an error, but we shouldn't
783
# ever see this code path in practice: tree's don't
784
# permit references when the repo doesn't support tree
786
raise errors.UnsupportedOperation(tree.add_reference,
788
reference_revision = tree.get_reference_revision(change[0])
789
entry.reference_revision = reference_revision
790
if (carry_over_possible and
791
parent_entry.reference_revision == reference_revision):
794
self._add_text_to_weave(change[0], '', heads, None)
796
raise AssertionError('unknown kind %r' % kind)
798
entry.revision = modified_rev
800
entry.revision = parent_entry.revision
803
new_path = change[1][1]
804
inv_delta.append((change[1][0], new_path, change[0], entry))
807
self.new_inventory = None
809
self._any_changes = True
811
# housekeeping root entry changes do not affect no-change commits.
812
self._require_root_change(tree)
813
self.basis_delta_revision = basis_revision_id
815
def _add_text_to_weave(self, file_id, new_text, parents, nostore_sha):
816
parent_keys = tuple([(file_id, parent) for parent in parents])
817
return self.repository.texts._add_text(
818
(file_id, self._new_revision_id), parent_keys, new_text,
819
nostore_sha=nostore_sha, random_id=self.random_revid)[0:2]
822
class RootCommitBuilder(CommitBuilder):
823
"""This commitbuilder actually records the root id"""
825
# the root entry gets versioned properly by this builder.
826
_versioned_root = True
828
def _check_root(self, ie, parent_invs, tree):
829
"""Helper for record_entry_contents.
831
:param ie: An entry being added.
832
:param parent_invs: The inventories of the parent revisions of the
834
:param tree: The tree that is being committed.
837
def _require_root_change(self, tree):
838
"""Enforce an appropriate root object change.
840
This is called once when record_iter_changes is called, if and only if
841
the root was not in the delta calculated by record_iter_changes.
843
:param tree: The tree which is being committed.
845
# versioned roots do not change unless the tree found a change.
241
848
######################################################################
245
class Repository(_RelockDebugMixin, controldir.ControlComponent):
851
class Repository(object):
246
852
"""Repository holding history for one or more branches.
248
854
The repository holds and retrieves historical information including
249
855
revisions and file history. It's normally accessed only by the Branch,
250
856
which views a particular line of development through that history.
252
See VersionedFileRepository in bzrlib.vf_repository for the
253
base class for most Bazaar repositories.
858
The Repository builds on top of some byte storage facilies (the revisions,
859
signatures, inventories, texts and chk_bytes attributes) and a Transport,
860
which respectively provide byte storage and a means to access the (possibly
863
The byte storage facilities are addressed via tuples, which we refer to
864
as 'keys' throughout the code base. Revision_keys, inventory_keys and
865
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
866
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
867
byte string made up of a hash identifier and a hash value.
868
We use this interface because it allows low friction with the underlying
869
code that implements disk indices, network encoding and other parts of
872
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
873
the serialised revisions for the repository. This can be used to obtain
874
revision graph information or to access raw serialised revisions.
875
The result of trying to insert data into the repository via this store
876
is undefined: it should be considered read-only except for implementors
878
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
879
the serialised signatures for the repository. This can be used to
880
obtain access to raw serialised signatures. The result of trying to
881
insert data into the repository via this store is undefined: it should
882
be considered read-only except for implementors of repositories.
883
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
884
the serialised inventories for the repository. This can be used to
885
obtain unserialised inventories. The result of trying to insert data
886
into the repository via this store is undefined: it should be
887
considered read-only except for implementors of repositories.
888
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
889
texts of files and directories for the repository. This can be used to
890
obtain file texts or file graphs. Note that Repository.iter_file_bytes
891
is usually a better interface for accessing file texts.
892
The result of trying to insert data into the repository via this store
893
is undefined: it should be considered read-only except for implementors
895
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
896
any data the repository chooses to store or have indexed by its hash.
897
The result of trying to insert data into the repository via this store
898
is undefined: it should be considered read-only except for implementors
900
:ivar _transport: Transport for file access to repository, typically
901
pointing to .bzr/repository.
904
# What class to use for a CommitBuilder. Often its simpler to change this
905
# in a Repository class subclass rather than to override
906
# get_commit_builder.
907
_commit_builder_class = CommitBuilder
908
# The search regex used by xml based repositories to determine what things
909
# where changed in a single commit.
910
_file_ids_altered_regex = lazy_regex.lazy_compile(
911
r'file_id="(?P<file_id>[^"]+)"'
912
r'.* revision="(?P<revision_id>[^"]+)"'
256
915
def abort_write_group(self, suppress_errors=False):
257
916
"""Commit the contents accrued within the current write group.
313
980
return InterRepository._assert_same_model(self, repository)
982
def add_inventory(self, revision_id, inv, parents):
983
"""Add the inventory inv to the repository as revision_id.
985
:param parents: The revision ids of the parents that revision_id
986
is known to have and are in the repository already.
988
:returns: The validator(which is a sha1 digest, though what is sha'd is
989
repository format specific) of the serialized inventory.
991
if not self.is_in_write_group():
992
raise AssertionError("%r not in write group" % (self,))
993
_mod_revision.check_not_reserved_id(revision_id)
994
if not (inv.revision_id is None or inv.revision_id == revision_id):
995
raise AssertionError(
996
"Mismatch between inventory revision"
997
" id and insertion revid (%r, %r)"
998
% (inv.revision_id, revision_id))
1000
raise AssertionError()
1001
return self._add_inventory_checked(revision_id, inv, parents)
1003
def _add_inventory_checked(self, revision_id, inv, parents):
1004
"""Add inv to the repository after checking the inputs.
1006
This function can be overridden to allow different inventory styles.
1008
:seealso: add_inventory, for the contract.
1010
inv_lines = self._serialise_inventory_to_lines(inv)
1011
return self._inventory_add_lines(revision_id, parents,
1012
inv_lines, check_content=False)
1014
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1015
parents, basis_inv=None, propagate_caches=False):
1016
"""Add a new inventory expressed as a delta against another revision.
1018
:param basis_revision_id: The inventory id the delta was created
1019
against. (This does not have to be a direct parent.)
1020
:param delta: The inventory delta (see Inventory.apply_delta for
1022
:param new_revision_id: The revision id that the inventory is being
1024
:param parents: The revision ids of the parents that revision_id is
1025
known to have and are in the repository already. These are supplied
1026
for repositories that depend on the inventory graph for revision
1027
graph access, as well as for those that pun ancestry with delta
1029
:param basis_inv: The basis inventory if it is already known,
1031
:param propagate_caches: If True, the caches for this inventory are
1032
copied to and updated for the result if possible.
1034
:returns: (validator, new_inv)
1035
The validator(which is a sha1 digest, though what is sha'd is
1036
repository format specific) of the serialized inventory, and the
1037
resulting inventory.
1039
if not self.is_in_write_group():
1040
raise AssertionError("%r not in write group" % (self,))
1041
_mod_revision.check_not_reserved_id(new_revision_id)
1042
basis_tree = self.revision_tree(basis_revision_id)
1043
basis_tree.lock_read()
1045
# Note that this mutates the inventory of basis_tree, which not all
1046
# inventory implementations may support: A better idiom would be to
1047
# return a new inventory, but as there is no revision tree cache in
1048
# repository this is safe for now - RBC 20081013
1049
if basis_inv is None:
1050
basis_inv = basis_tree.inventory
1051
basis_inv.apply_delta(delta)
1052
basis_inv.revision_id = new_revision_id
1053
return (self.add_inventory(new_revision_id, basis_inv, parents),
1058
def _inventory_add_lines(self, revision_id, parents, lines,
1059
check_content=True):
1060
"""Store lines in inv_vf and return the sha1 of the inventory."""
1061
parents = [(parent,) for parent in parents]
1062
return self.inventories.add_lines((revision_id,), parents, lines,
1063
check_content=check_content)[0]
1065
def add_revision(self, revision_id, rev, inv=None, config=None):
1066
"""Add rev to the revision store as revision_id.
1068
:param revision_id: the revision id to use.
1069
:param rev: The revision object.
1070
:param inv: The inventory for the revision. if None, it will be looked
1071
up in the inventory storer
1072
:param config: If None no digital signature will be created.
1073
If supplied its signature_needed method will be used
1074
to determine if a signature should be made.
1076
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1078
_mod_revision.check_not_reserved_id(revision_id)
1079
if config is not None and config.signature_needed():
1081
inv = self.get_inventory(revision_id)
1082
plaintext = Testament(rev, inv).as_short_text()
1083
self.store_revision_signature(
1084
gpg.GPGStrategy(config), plaintext, revision_id)
1085
# check inventory present
1086
if not self.inventories.get_parent_map([(revision_id,)]):
1088
raise errors.WeaveRevisionNotPresent(revision_id,
1091
# yes, this is not suitable for adding with ghosts.
1092
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1095
key = (revision_id,)
1096
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1097
self._add_revision(rev)
1099
def _add_revision(self, revision):
1100
text = self._serializer.write_revision_to_string(revision)
1101
key = (revision.revision_id,)
1102
parents = tuple((parent,) for parent in revision.parent_ids)
1103
self.revisions.add_lines(key, parents, osutils.split_lines(text))
315
1105
def all_revision_ids(self):
316
1106
"""Returns a list of all the revision ids in the repository.
909
1785
signature = gpg_strategy.sign(plaintext)
910
1786
self.add_signature_text(revision_id, signature)
912
1789
def add_signature_text(self, revision_id, signature):
913
"""Store a signature text for a revision.
915
:param revision_id: Revision id of the revision
916
:param signature: Signature text.
918
raise NotImplementedError(self.add_signature_text)
1790
self.signatures.add_lines((revision_id,), (),
1791
osutils.split_lines(signature))
1793
def find_text_key_references(self):
1794
"""Find the text key references within the repository.
1796
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1797
to whether they were referred to by the inventory of the
1798
revision_id that they contain. The inventory texts from all present
1799
revision ids are assessed to generate this report.
1801
revision_keys = self.revisions.keys()
1802
w = self.inventories
1803
pb = ui.ui_factory.nested_progress_bar()
1805
return self._find_text_key_references_from_xml_inventory_lines(
1806
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1810
def _find_text_key_references_from_xml_inventory_lines(self,
1812
"""Core routine for extracting references to texts from inventories.
1814
This performs the translation of xml lines to revision ids.
1816
:param line_iterator: An iterator of lines, origin_version_id
1817
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1818
to whether they were referred to by the inventory of the
1819
revision_id that they contain. Note that if that revision_id was
1820
not part of the line_iterator's output then False will be given -
1821
even though it may actually refer to that key.
1823
if not self._serializer.support_altered_by_hack:
1824
raise AssertionError(
1825
"_find_text_key_references_from_xml_inventory_lines only "
1826
"supported for branches which store inventory as unnested xml"
1827
", not on %r" % self)
1830
# this code needs to read every new line in every inventory for the
1831
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
1832
# not present in one of those inventories is unnecessary but not
1833
# harmful because we are filtering by the revision id marker in the
1834
# inventory lines : we only select file ids altered in one of those
1835
# revisions. We don't need to see all lines in the inventory because
1836
# only those added in an inventory in rev X can contain a revision=X
1838
unescape_revid_cache = {}
1839
unescape_fileid_cache = {}
1841
# jam 20061218 In a big fetch, this handles hundreds of thousands
1842
# of lines, so it has had a lot of inlining and optimizing done.
1843
# Sorry that it is a little bit messy.
1844
# Move several functions to be local variables, since this is a long
1846
search = self._file_ids_altered_regex.search
1847
unescape = _unescape_xml
1848
setdefault = result.setdefault
1849
for line, line_key in line_iterator:
1850
match = search(line)
1853
# One call to match.group() returning multiple items is quite a
1854
# bit faster than 2 calls to match.group() each returning 1
1855
file_id, revision_id = match.group('file_id', 'revision_id')
1857
# Inlining the cache lookups helps a lot when you make 170,000
1858
# lines and 350k ids, versus 8.4 unique ids.
1859
# Using a cache helps in 2 ways:
1860
# 1) Avoids unnecessary decoding calls
1861
# 2) Re-uses cached strings, which helps in future set and
1863
# (2) is enough that removing encoding entirely along with
1864
# the cache (so we are using plain strings) results in no
1865
# performance improvement.
1867
revision_id = unescape_revid_cache[revision_id]
1869
unescaped = unescape(revision_id)
1870
unescape_revid_cache[revision_id] = unescaped
1871
revision_id = unescaped
1873
# Note that unconditionally unescaping means that we deserialise
1874
# every fileid, which for general 'pull' is not great, but we don't
1875
# really want to have some many fulltexts that this matters anyway.
1878
file_id = unescape_fileid_cache[file_id]
1880
unescaped = unescape(file_id)
1881
unescape_fileid_cache[file_id] = unescaped
1884
key = (file_id, revision_id)
1885
setdefault(key, False)
1886
if revision_id == line_key[-1]:
1890
def _inventory_xml_lines_for_keys(self, keys):
1891
"""Get a line iterator of the sort needed for findind references.
1893
Not relevant for non-xml inventory repositories.
1895
Ghosts in revision_keys are ignored.
1897
:param revision_keys: The revision keys for the inventories to inspect.
1898
:return: An iterator over (inventory line, revid) for the fulltexts of
1899
all of the xml inventories specified by revision_keys.
1901
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1902
for record in stream:
1903
if record.storage_kind != 'absent':
1904
chunks = record.get_bytes_as('chunked')
1905
revid = record.key[-1]
1906
lines = osutils.chunks_to_lines(chunks)
1910
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1912
"""Helper routine for fileids_altered_by_revision_ids.
1914
This performs the translation of xml lines to revision ids.
1916
:param line_iterator: An iterator of lines, origin_version_id
1917
:param revision_keys: The revision ids to filter for. This should be a
1918
set or other type which supports efficient __contains__ lookups, as
1919
the revision key from each parsed line will be looked up in the
1920
revision_keys filter.
1921
:return: a dictionary mapping altered file-ids to an iterable of
1922
revision_ids. Each altered file-ids has the exact revision_ids that
1923
altered it listed explicitly.
1925
seen = set(self._find_text_key_references_from_xml_inventory_lines(
1926
line_iterator).iterkeys())
1927
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
1928
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
1929
self._inventory_xml_lines_for_keys(parent_keys)))
1930
new_keys = seen - parent_seen
1932
setdefault = result.setdefault
1933
for key in new_keys:
1934
setdefault(key[0], set()).add(key[-1])
920
1937
def _find_parent_ids_of_revisions(self, revision_ids):
921
1938
"""Find all parent ids that are mentioned in the revision graph.
951
1993
uniquely identify the file version in the caller's context. (Examples:
952
1994
an index number or a TreeTransform trans_id.)
1996
bytes_iterator is an iterable of bytestrings for the file. The
1997
kind of iterable and length of the bytestrings are unspecified, but for
1998
this implementation, it is a list of bytes produced by
1999
VersionedFile.get_record_stream().
954
2001
:param desired_files: a list of (file_id, revision_id, identifier)
957
raise NotImplementedError(self.iter_files_bytes)
2005
for file_id, revision_id, callable_data in desired_files:
2006
text_keys[(file_id, revision_id)] = callable_data
2007
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2008
if record.storage_kind == 'absent':
2009
raise errors.RevisionNotPresent(record.key, self)
2010
yield text_keys[record.key], record.get_bytes_as('chunked')
2012
def _generate_text_key_index(self, text_key_references=None,
2014
"""Generate a new text key index for the repository.
2016
This is an expensive function that will take considerable time to run.
2018
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2019
list of parents, also text keys. When a given key has no parents,
2020
the parents list will be [NULL_REVISION].
2022
# All revisions, to find inventory parents.
2023
if ancestors is None:
2024
graph = self.get_graph()
2025
ancestors = graph.get_parent_map(self.all_revision_ids())
2026
if text_key_references is None:
2027
text_key_references = self.find_text_key_references()
2028
pb = ui.ui_factory.nested_progress_bar()
2030
return self._do_generate_text_key_index(ancestors,
2031
text_key_references, pb)
2035
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2036
"""Helper for _generate_text_key_index to avoid deep nesting."""
2037
revision_order = tsort.topo_sort(ancestors)
2038
invalid_keys = set()
2040
for revision_id in revision_order:
2041
revision_keys[revision_id] = set()
2042
text_count = len(text_key_references)
2043
# a cache of the text keys to allow reuse; costs a dict of all the
2044
# keys, but saves a 2-tuple for every child of a given key.
2046
for text_key, valid in text_key_references.iteritems():
2048
invalid_keys.add(text_key)
2050
revision_keys[text_key[1]].add(text_key)
2051
text_key_cache[text_key] = text_key
2052
del text_key_references
2054
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2055
NULL_REVISION = _mod_revision.NULL_REVISION
2056
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2057
# too small for large or very branchy trees. However, for 55K path
2058
# trees, it would be easy to use too much memory trivially. Ideally we
2059
# could gauge this by looking at available real memory etc, but this is
2060
# always a tricky proposition.
2061
inventory_cache = lru_cache.LRUCache(10)
2062
batch_size = 10 # should be ~150MB on a 55K path tree
2063
batch_count = len(revision_order) / batch_size + 1
2065
pb.update("Calculating text parents", processed_texts, text_count)
2066
for offset in xrange(batch_count):
2067
to_query = revision_order[offset * batch_size:(offset + 1) *
2071
for rev_tree in self.revision_trees(to_query):
2072
revision_id = rev_tree.get_revision_id()
2073
parent_ids = ancestors[revision_id]
2074
for text_key in revision_keys[revision_id]:
2075
pb.update("Calculating text parents", processed_texts)
2076
processed_texts += 1
2077
candidate_parents = []
2078
for parent_id in parent_ids:
2079
parent_text_key = (text_key[0], parent_id)
2081
check_parent = parent_text_key not in \
2082
revision_keys[parent_id]
2084
# the parent parent_id is a ghost:
2085
check_parent = False
2086
# truncate the derived graph against this ghost.
2087
parent_text_key = None
2089
# look at the parent commit details inventories to
2090
# determine possible candidates in the per file graph.
2093
inv = inventory_cache[parent_id]
2095
inv = self.revision_tree(parent_id).inventory
2096
inventory_cache[parent_id] = inv
2098
parent_entry = inv[text_key[0]]
2099
except (KeyError, errors.NoSuchId):
2101
if parent_entry is not None:
2103
text_key[0], parent_entry.revision)
2105
parent_text_key = None
2106
if parent_text_key is not None:
2107
candidate_parents.append(
2108
text_key_cache[parent_text_key])
2109
parent_heads = text_graph.heads(candidate_parents)
2110
new_parents = list(parent_heads)
2111
new_parents.sort(key=lambda x:candidate_parents.index(x))
2112
if new_parents == []:
2113
new_parents = [NULL_REVISION]
2114
text_index[text_key] = new_parents
2116
for text_key in invalid_keys:
2117
text_index[text_key] = [NULL_REVISION]
2120
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2121
"""Get an iterable listing the keys of all the data introduced by a set
2124
The keys will be ordered so that the corresponding items can be safely
2125
fetched and inserted in that order.
2127
:returns: An iterable producing tuples of (knit-kind, file-id,
2128
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2129
'revisions'. file-id is None unless knit-kind is 'file'.
2131
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2134
for result in self._find_non_file_keys_to_fetch(revision_ids):
2137
def _find_file_keys_to_fetch(self, revision_ids, pb):
2138
# XXX: it's a bit weird to control the inventory weave caching in this
2139
# generator. Ideally the caching would be done in fetch.py I think. Or
2140
# maybe this generator should explicitly have the contract that it
2141
# should not be iterated until the previously yielded item has been
2143
inv_w = self.inventories
2145
# file ids that changed
2146
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2148
num_file_ids = len(file_ids)
2149
for file_id, altered_versions in file_ids.iteritems():
2151
pb.update("fetch texts", count, num_file_ids)
2153
yield ("file", file_id, altered_versions)
2155
def _find_non_file_keys_to_fetch(self, revision_ids):
2157
yield ("inventory", None, revision_ids)
2160
# XXX: Note ATM no callers actually pay attention to this return
2161
# instead they just use the list of revision ids and ignore
2162
# missing sigs. Consider removing this work entirely
2163
revisions_with_signatures = set(self.signatures.get_parent_map(
2164
[(r,) for r in revision_ids]))
2165
revisions_with_signatures = set(
2166
[r for (r,) in revisions_with_signatures])
2167
revisions_with_signatures.intersection_update(revision_ids)
2168
yield ("signatures", None, revisions_with_signatures)
2171
yield ("revisions", None, revision_ids)
2174
def get_inventory(self, revision_id):
2175
"""Get Inventory object by revision id."""
2176
return self.iter_inventories([revision_id]).next()
2178
def iter_inventories(self, revision_ids):
2179
"""Get many inventories by revision_ids.
2181
This will buffer some or all of the texts used in constructing the
2182
inventories in memory, but will only parse a single inventory at a
2185
:param revision_ids: The expected revision ids of the inventories.
2186
:return: An iterator of inventories.
2188
if ((None in revision_ids)
2189
or (_mod_revision.NULL_REVISION in revision_ids)):
2190
raise ValueError('cannot get null revision inventory')
2191
return self._iter_inventories(revision_ids)
2193
def _iter_inventories(self, revision_ids):
2194
"""single-document based inventory iteration."""
2195
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2196
yield self.deserialise_inventory(revision_id, text)
2198
def _iter_inventory_xmls(self, revision_ids):
2199
keys = [(revision_id,) for revision_id in revision_ids]
2200
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2202
for record in stream:
2203
if record.storage_kind != 'absent':
2204
text_chunks[record.key] = record.get_bytes_as('chunked')
2206
raise errors.NoSuchRevision(self, record.key)
2208
chunks = text_chunks.pop(key)
2209
yield ''.join(chunks), key[-1]
2211
def deserialise_inventory(self, revision_id, xml):
2212
"""Transform the xml into an inventory object.
2214
:param revision_id: The expected revision id of the inventory.
2215
:param xml: A serialised inventory.
2217
result = self._serializer.read_inventory_from_string(xml, revision_id,
2218
entry_cache=self._inventory_entry_cache)
2219
if result.revision_id != revision_id:
2220
raise AssertionError('revision id mismatch %s != %s' % (
2221
result.revision_id, revision_id))
2224
def serialise_inventory(self, inv):
2225
return self._serializer.write_inventory_to_string(inv)
2227
def _serialise_inventory_to_lines(self, inv):
2228
return self._serializer.write_inventory_to_lines(inv)
2230
def get_serializer_format(self):
2231
return self._serializer.format_num
2234
def get_inventory_xml(self, revision_id):
2235
"""Get inventory XML as a file object."""
2236
texts = self._iter_inventory_xmls([revision_id])
2238
text, revision_id = texts.next()
2239
except StopIteration:
2240
raise errors.HistoryMissing(self, 'inventory', revision_id)
2244
def get_inventory_sha1(self, revision_id):
2245
"""Return the sha1 hash of the inventory entry
2247
return self.get_revision(revision_id).inventory_sha1
959
2249
def get_rev_id_for_revno(self, revno, known_pair):
960
2250
"""Return the revision id of a revno, given a later (revno, revid)
1191
2515
@needs_write_lock
1192
2516
def sign_revision(self, revision_id, gpg_strategy):
1193
testament = _mod_testament.Testament.from_revision(self, revision_id)
1194
plaintext = testament.as_short_text()
2517
plaintext = Testament.from_revision(self, revision_id).as_short_text()
1195
2518
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
1197
2520
@needs_read_lock
1198
def verify_revision(self, revision_id, gpg_strategy):
1199
"""Verify the signature on a revision.
1201
:param revision_id: the revision to verify
1202
:gpg_strategy: the GPGStrategy object to used
1204
:return: gpg.SIGNATURE_VALID or a failed SIGNATURE_ value
1206
if not self.has_signature_for_revision_id(revision_id):
1207
return gpg.SIGNATURE_NOT_SIGNED, None
1208
signature = self.get_signature_text(revision_id)
1210
testament = _mod_testament.Testament.from_revision(self, revision_id)
1211
plaintext = testament.as_short_text()
1213
return gpg_strategy.verify(signature, plaintext)
1215
2521
def has_signature_for_revision_id(self, revision_id):
1216
2522
"""Query for a revision signature for revision_id in the repository."""
1217
raise NotImplementedError(self.has_signature_for_revision_id)
2523
if not self.has_revision(revision_id):
2524
raise errors.NoSuchRevision(self, revision_id)
2525
sig_present = (1 == len(
2526
self.signatures.get_parent_map([(revision_id,)])))
1219
2530
def get_signature_text(self, revision_id):
1220
2531
"""Return the text for a signature."""
1221
raise NotImplementedError(self.get_signature_text)
2532
stream = self.signatures.get_record_stream([(revision_id,)],
2534
record = stream.next()
2535
if record.storage_kind == 'absent':
2536
raise errors.NoSuchRevision(self, revision_id)
2537
return record.get_bytes_as('fulltext')
1223
def check(self, revision_ids=None, callback_refs=None, check_repo=True):
2540
def check(self, revision_ids=None):
1224
2541
"""Check consistency of all history of given revision_ids.
1226
2543
Different repository implementations should override _check().
1228
2545
:param revision_ids: A non-empty list of revision_ids whose ancestry
1229
2546
will be checked. Typically the last revision_id of a branch.
1230
:param callback_refs: A dict of check-refs to resolve and callback
1231
the check/_check method on the items listed as wanting the ref.
1233
:param check_repo: If False do not check the repository contents, just
1234
calculate the data callback_refs requires and call them back.
1236
return self._check(revision_ids=revision_ids, callback_refs=callback_refs,
1237
check_repo=check_repo)
1239
def _check(self, revision_ids=None, callback_refs=None, check_repo=True):
1240
raise NotImplementedError(self.check)
1242
def _warn_if_deprecated(self, branch=None):
1243
if not self._format.is_deprecated():
2548
return self._check(revision_ids)
2550
def _check(self, revision_ids):
2551
result = check.Check(self)
2555
def _warn_if_deprecated(self):
1245
2556
global _deprecation_warning_done
1246
2557
if _deprecation_warning_done:
1250
conf = config.GlobalConfig()
1252
conf = branch.get_config()
1253
if conf.suppress_warning('format_deprecation'):
1255
warning("Format %s for %s is deprecated -"
1256
" please use 'bzr upgrade' to get better performance"
1257
% (self._format, self.bzrdir.transport.base))
1259
_deprecation_warning_done = True
2559
_deprecation_warning_done = True
2560
warning("Format %s for %s is deprecated - please use 'bzr upgrade' to get better performance"
2561
% (self._format, self.bzrdir.transport.base))
1261
2563
def supports_rich_root(self):
1262
2564
return self._format.rich_root_data
1277
2579
except UnicodeDecodeError:
1278
2580
raise errors.NonAsciiRevisionId(method, self)
2582
def revision_graph_can_have_wrong_parents(self):
2583
"""Is it possible for this repository to have a revision graph with
2586
If True, then this repository must also implement
2587
_find_inconsistent_revision_parents so that check and reconcile can
2588
check for inconsistencies before proceeding with other checks that may
2589
depend on the revision index being consistent.
2591
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2594
# remove these delegates a while after bzr 0.15
2595
def __make_delegated(name, from_module):
2596
def _deprecated_repository_forwarder():
2597
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2598
% (name, from_module),
2601
m = __import__(from_module, globals(), locals(), [name])
2603
return getattr(m, name)
2604
except AttributeError:
2605
raise AttributeError('module %s has no name %s'
2607
globals()[name] = _deprecated_repository_forwarder
2610
'AllInOneRepository',
2611
'WeaveMetaDirRepository',
2612
'PreSplitOutRepositoryFormat',
2613
'RepositoryFormat4',
2614
'RepositoryFormat5',
2615
'RepositoryFormat6',
2616
'RepositoryFormat7',
2618
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2622
'RepositoryFormatKnit',
2623
'RepositoryFormatKnit1',
2625
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2628
def install_revision(repository, rev, revision_tree):
2629
"""Install all revision data into a repository."""
2630
install_revisions(repository, [(rev, revision_tree, None)])
2633
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2634
"""Install all revision data into a repository.
2636
Accepts an iterable of revision, tree, signature tuples. The signature
2639
repository.start_write_group()
2641
inventory_cache = lru_cache.LRUCache(10)
2642
for n, (revision, revision_tree, signature) in enumerate(iterable):
2643
_install_revision(repository, revision, revision_tree, signature,
2646
pb.update('Transferring revisions', n + 1, num_revisions)
2648
repository.abort_write_group()
2651
repository.commit_write_group()
2654
def _install_revision(repository, rev, revision_tree, signature,
2656
"""Install all revision data into a repository."""
2657
present_parents = []
2659
for p_id in rev.parent_ids:
2660
if repository.has_revision(p_id):
2661
present_parents.append(p_id)
2662
parent_trees[p_id] = repository.revision_tree(p_id)
2664
parent_trees[p_id] = repository.revision_tree(
2665
_mod_revision.NULL_REVISION)
2667
inv = revision_tree.inventory
2668
entries = inv.iter_entries()
2669
# backwards compatibility hack: skip the root id.
2670
if not repository.supports_rich_root():
2671
path, root = entries.next()
2672
if root.revision != rev.revision_id:
2673
raise errors.IncompatibleRevision(repr(repository))
2675
for path, ie in entries:
2676
text_keys[(ie.file_id, ie.revision)] = ie
2677
text_parent_map = repository.texts.get_parent_map(text_keys)
2678
missing_texts = set(text_keys) - set(text_parent_map)
2679
# Add the texts that are not already present
2680
for text_key in missing_texts:
2681
ie = text_keys[text_key]
2683
# FIXME: TODO: The following loop overlaps/duplicates that done by
2684
# commit to determine parents. There is a latent/real bug here where
2685
# the parents inserted are not those commit would do - in particular
2686
# they are not filtered by heads(). RBC, AB
2687
for revision, tree in parent_trees.iteritems():
2688
if ie.file_id not in tree:
2690
parent_id = tree.inventory[ie.file_id].revision
2691
if parent_id in text_parents:
2693
text_parents.append((ie.file_id, parent_id))
2694
lines = revision_tree.get_file(ie.file_id).readlines()
2695
repository.texts.add_lines(text_key, text_parents, lines)
2697
# install the inventory
2698
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2699
# Cache this inventory
2700
inventory_cache[rev.revision_id] = inv
2702
basis_inv = inventory_cache[rev.parent_ids[0]]
2704
repository.add_inventory(rev.revision_id, inv, present_parents)
2706
delta = inv._make_delta(basis_inv)
2707
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2708
rev.revision_id, present_parents)
2710
repository.add_inventory(rev.revision_id, inv, present_parents)
2711
except errors.RevisionAlreadyPresent:
2713
if signature is not None:
2714
repository.add_signature_text(rev.revision_id, signature)
2715
repository.add_revision(rev.revision_id, rev, inv)
1281
2718
class MetaDirRepository(Repository):
1282
2719
"""Repositories in the new meta-dir layout.
1621
3063
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
1622
3064
format_registry.register_lazy(
1623
3065
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
1624
'bzrlib.repofmt.knitpack_repo',
3066
'bzrlib.repofmt.pack_repo',
1625
3067
'RepositoryFormatKnitPack1',
1627
3069
format_registry.register_lazy(
1628
3070
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
1629
'bzrlib.repofmt.knitpack_repo',
3071
'bzrlib.repofmt.pack_repo',
1630
3072
'RepositoryFormatKnitPack3',
1632
3074
format_registry.register_lazy(
1633
3075
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
1634
'bzrlib.repofmt.knitpack_repo',
3076
'bzrlib.repofmt.pack_repo',
1635
3077
'RepositoryFormatKnitPack4',
1637
3079
format_registry.register_lazy(
1638
3080
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
1639
'bzrlib.repofmt.knitpack_repo',
3081
'bzrlib.repofmt.pack_repo',
1640
3082
'RepositoryFormatKnitPack5',
1642
3084
format_registry.register_lazy(
1643
3085
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
1644
'bzrlib.repofmt.knitpack_repo',
3086
'bzrlib.repofmt.pack_repo',
1645
3087
'RepositoryFormatKnitPack5RichRoot',
1647
3089
format_registry.register_lazy(
1648
3090
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
1649
'bzrlib.repofmt.knitpack_repo',
3091
'bzrlib.repofmt.pack_repo',
1650
3092
'RepositoryFormatKnitPack5RichRootBroken',
1652
3094
format_registry.register_lazy(
1653
3095
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
1654
'bzrlib.repofmt.knitpack_repo',
3096
'bzrlib.repofmt.pack_repo',
1655
3097
'RepositoryFormatKnitPack6',
1657
3099
format_registry.register_lazy(
1658
3100
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
1659
'bzrlib.repofmt.knitpack_repo',
3101
'bzrlib.repofmt.pack_repo',
1660
3102
'RepositoryFormatKnitPack6RichRoot',
1662
format_registry.register_lazy(
1663
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
1664
'bzrlib.repofmt.groupcompress_repo',
1665
'RepositoryFormat2a',
1668
3105
# Development formats.
1669
# Check their docstrings to see if/when they are obsolete.
3106
# Obsolete but kept pending a CHK based subtree format.
1670
3107
format_registry.register_lazy(
1671
3108
("Bazaar development format 2 with subtree support "
1672
3109
"(needs bzr.dev from before 1.8)\n"),
1673
'bzrlib.repofmt.knitpack_repo',
3110
'bzrlib.repofmt.pack_repo',
1674
3111
'RepositoryFormatPackDevelopment2Subtree',
1676
format_registry.register_lazy(
1677
'Bazaar development format 8\n',
1678
'bzrlib.repofmt.groupcompress_repo',
1679
'RepositoryFormat2aSubtree',
3114
# 1.14->1.16 go below here
3115
format_registry.register_lazy(
3116
'Bazaar development format - group compression and chk inventory'
3117
' (needs bzr.dev from 1.14)\n',
3118
'bzrlib.repofmt.groupcompress_repo',
3119
'RepositoryFormatCHK1',
3122
format_registry.register_lazy(
3123
'Bazaar development format - chk repository with bencode revision '
3124
'serialization (needs bzr.dev from 1.16)\n',
3125
'bzrlib.repofmt.groupcompress_repo',
3126
'RepositoryFormatCHK2',
3128
format_registry.register_lazy(
3129
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3130
'bzrlib.repofmt.groupcompress_repo',
3131
'RepositoryFormat2a',
1721
3174
:param revision_id: if None all content is copied, if NULL_REVISION no
1722
3175
content is copied.
3176
:param pb: optional progress bar to use for progress reports. If not
3177
provided a default one will be created.
1725
raise NotImplementedError(self.fetch)
3180
from bzrlib.fetch import RepoFetcher
3181
f = RepoFetcher(to_repository=self.target,
3182
from_repository=self.source,
3183
last_revision=revision_id,
3184
fetch_spec=fetch_spec,
3185
pb=pb, find_ghosts=find_ghosts)
3187
def _walk_to_common_revisions(self, revision_ids):
3188
"""Walk out from revision_ids in source to revisions target has.
3190
:param revision_ids: The start point for the search.
3191
:return: A set of revision ids.
3193
target_graph = self.target.get_graph()
3194
revision_ids = frozenset(revision_ids)
3195
missing_revs = set()
3196
source_graph = self.source.get_graph()
3197
# ensure we don't pay silly lookup costs.
3198
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3199
null_set = frozenset([_mod_revision.NULL_REVISION])
3200
searcher_exhausted = False
3204
# Iterate the searcher until we have enough next_revs
3205
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3207
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3208
next_revs.update(next_revs_part)
3209
ghosts.update(ghosts_part)
3210
except StopIteration:
3211
searcher_exhausted = True
3213
# If there are ghosts in the source graph, and the caller asked for
3214
# them, make sure that they are present in the target.
3215
# We don't care about other ghosts as we can't fetch them and
3216
# haven't been asked to.
3217
ghosts_to_check = set(revision_ids.intersection(ghosts))
3218
revs_to_get = set(next_revs).union(ghosts_to_check)
3220
have_revs = set(target_graph.get_parent_map(revs_to_get))
3221
# we always have NULL_REVISION present.
3222
have_revs = have_revs.union(null_set)
3223
# Check if the target is missing any ghosts we need.
3224
ghosts_to_check.difference_update(have_revs)
3226
# One of the caller's revision_ids is a ghost in both the
3227
# source and the target.
3228
raise errors.NoSuchRevision(
3229
self.source, ghosts_to_check.pop())
3230
missing_revs.update(next_revs - have_revs)
3231
# Because we may have walked past the original stop point, make
3232
# sure everything is stopped
3233
stop_revs = searcher.find_seen_ancestors(have_revs)
3234
searcher.stop_searching_any(stop_revs)
3235
if searcher_exhausted:
3237
return searcher.get_result()
1727
3239
@needs_read_lock
1728
def search_missing_revision_ids(self,
1729
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
1730
find_ghosts=True, revision_ids=None, if_present_ids=None,
3240
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1732
3241
"""Return the revision ids that source has that target does not.
1734
3243
:param revision_id: only return revision ids included by this
1736
:param revision_ids: return revision ids included by these
1737
revision_ids. NoSuchRevision will be raised if any of these
1738
revisions are not present.
1739
:param if_present_ids: like revision_ids, but will not cause
1740
NoSuchRevision if any of these are absent, instead they will simply
1741
not be in the result. This is useful for e.g. finding revisions
1742
to fetch for tags, which may reference absent revisions.
1743
3245
:param find_ghosts: If True find missing revisions in deep history
1744
3246
rather than just finding the surface difference.
1745
:param limit: Maximum number of revisions to return, topologically
1747
3247
:return: A bzrlib.graph.SearchResult.
1749
raise NotImplementedError(self.search_missing_revision_ids)
3249
# stop searching at found target revisions.
3250
if not find_ghosts and revision_id is not None:
3251
return self._walk_to_common_revisions([revision_id])
3252
# generic, possibly worst case, slow code path.
3253
target_ids = set(self.target.all_revision_ids())
3254
if revision_id is not None:
3255
source_ids = self.source.get_ancestry(revision_id)
3256
if source_ids[0] is not None:
3257
raise AssertionError()
3260
source_ids = self.source.all_revision_ids()
3261
result_set = set(source_ids).difference(target_ids)
3262
return self.source.revision_ids_to_search_result(result_set)
1752
3265
def _same_model(source, target):
1773
3286
"different serializers")
3289
class InterSameDataRepository(InterRepository):
3290
"""Code for converting between repositories that represent the same data.
3292
Data format and model must match for this to work.
3296
def _get_repo_format_to_test(self):
3297
"""Repository format for testing with.
3299
InterSameData can pull from subtree to subtree and from non-subtree to
3300
non-subtree, so we test this with the richest repository format.
3302
from bzrlib.repofmt import knitrepo
3303
return knitrepo.RepositoryFormatKnit3()
3306
def is_compatible(source, target):
3307
return InterRepository._same_model(source, target)
3310
class InterWeaveRepo(InterSameDataRepository):
3311
"""Optimised code paths between Weave based repositories.
3313
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3314
implemented lazy inter-object optimisation.
3318
def _get_repo_format_to_test(self):
3319
from bzrlib.repofmt import weaverepo
3320
return weaverepo.RepositoryFormat7()
3323
def is_compatible(source, target):
3324
"""Be compatible with known Weave formats.
3326
We don't test for the stores being of specific types because that
3327
could lead to confusing results, and there is no need to be
3330
from bzrlib.repofmt.weaverepo import (
3336
return (isinstance(source._format, (RepositoryFormat5,
3338
RepositoryFormat7)) and
3339
isinstance(target._format, (RepositoryFormat5,
3341
RepositoryFormat7)))
3342
except AttributeError:
3346
def copy_content(self, revision_id=None):
3347
"""See InterRepository.copy_content()."""
3348
# weave specific optimised path:
3350
self.target.set_make_working_trees(self.source.make_working_trees())
3351
except (errors.RepositoryUpgradeRequired, NotImplemented):
3353
# FIXME do not peek!
3354
if self.source._transport.listable():
3355
pb = ui.ui_factory.nested_progress_bar()
3357
self.target.texts.insert_record_stream(
3358
self.source.texts.get_record_stream(
3359
self.source.texts.keys(), 'topological', False))
3360
pb.update('copying inventory', 0, 1)
3361
self.target.inventories.insert_record_stream(
3362
self.source.inventories.get_record_stream(
3363
self.source.inventories.keys(), 'topological', False))
3364
self.target.signatures.insert_record_stream(
3365
self.source.signatures.get_record_stream(
3366
self.source.signatures.keys(),
3368
self.target.revisions.insert_record_stream(
3369
self.source.revisions.get_record_stream(
3370
self.source.revisions.keys(),
3371
'topological', True))
3375
self.target.fetch(self.source, revision_id=revision_id)
3378
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3379
"""See InterRepository.missing_revision_ids()."""
3380
# we want all revisions to satisfy revision_id in source.
3381
# but we don't want to stat every file here and there.
3382
# we want then, all revisions other needs to satisfy revision_id
3383
# checked, but not those that we have locally.
3384
# so the first thing is to get a subset of the revisions to
3385
# satisfy revision_id in source, and then eliminate those that
3386
# we do already have.
3387
# this is slow on high latency connection to self, but as this
3388
# disk format scales terribly for push anyway due to rewriting
3389
# inventory.weave, this is considered acceptable.
3391
if revision_id is not None:
3392
source_ids = self.source.get_ancestry(revision_id)
3393
if source_ids[0] is not None:
3394
raise AssertionError()
3397
source_ids = self.source._all_possible_ids()
3398
source_ids_set = set(source_ids)
3399
# source_ids is the worst possible case we may need to pull.
3400
# now we want to filter source_ids against what we actually
3401
# have in target, but don't try to check for existence where we know
3402
# we do not have a revision as that would be pointless.
3403
target_ids = set(self.target._all_possible_ids())
3404
possibly_present_revisions = target_ids.intersection(source_ids_set)
3405
actually_present_revisions = set(
3406
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3407
required_revisions = source_ids_set.difference(actually_present_revisions)
3408
if revision_id is not None:
3409
# we used get_ancestry to determine source_ids then we are assured all
3410
# revisions referenced are present as they are installed in topological order.
3411
# and the tip revision was validated by get_ancestry.
3412
result_set = required_revisions
3414
# if we just grabbed the possibly available ids, then
3415
# we only have an estimate of whats available and need to validate
3416
# that against the revision records.
3418
self.source._eliminate_revisions_not_present(required_revisions))
3419
return self.source.revision_ids_to_search_result(result_set)
3422
class InterKnitRepo(InterSameDataRepository):
3423
"""Optimised code paths between Knit based repositories."""
3426
def _get_repo_format_to_test(self):
3427
from bzrlib.repofmt import knitrepo
3428
return knitrepo.RepositoryFormatKnit1()
3431
def is_compatible(source, target):
3432
"""Be compatible with known Knit formats.
3434
We don't test for the stores being of specific types because that
3435
could lead to confusing results, and there is no need to be
3438
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3440
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3441
isinstance(target._format, RepositoryFormatKnit))
3442
except AttributeError:
3444
return are_knits and InterRepository._same_model(source, target)
3447
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3448
"""See InterRepository.missing_revision_ids()."""
3449
if revision_id is not None:
3450
source_ids = self.source.get_ancestry(revision_id)
3451
if source_ids[0] is not None:
3452
raise AssertionError()
3455
source_ids = self.source.all_revision_ids()
3456
source_ids_set = set(source_ids)
3457
# source_ids is the worst possible case we may need to pull.
3458
# now we want to filter source_ids against what we actually
3459
# have in target, but don't try to check for existence where we know
3460
# we do not have a revision as that would be pointless.
3461
target_ids = set(self.target.all_revision_ids())
3462
possibly_present_revisions = target_ids.intersection(source_ids_set)
3463
actually_present_revisions = set(
3464
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3465
required_revisions = source_ids_set.difference(actually_present_revisions)
3466
if revision_id is not None:
3467
# we used get_ancestry to determine source_ids then we are assured all
3468
# revisions referenced are present as they are installed in topological order.
3469
# and the tip revision was validated by get_ancestry.
3470
result_set = required_revisions
3472
# if we just grabbed the possibly available ids, then
3473
# we only have an estimate of whats available and need to validate
3474
# that against the revision records.
3476
self.source._eliminate_revisions_not_present(required_revisions))
3477
return self.source.revision_ids_to_search_result(result_set)
3480
class InterDifferingSerializer(InterRepository):
3483
def _get_repo_format_to_test(self):
3487
def is_compatible(source, target):
3488
"""Be compatible with Knit2 source and Knit3 target"""
3489
# This is redundant with format.check_conversion_target(), however that
3490
# raises an exception, and we just want to say "False" as in we won't
3491
# support converting between these formats.
3492
if source.supports_rich_root() and not target.supports_rich_root():
3494
if (source._format.supports_tree_reference
3495
and not target._format.supports_tree_reference):
3499
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3500
"""Get the best delta and base for this revision.
3502
:return: (basis_id, delta)
3504
possible_trees = [(parent_id, cache[parent_id])
3505
for parent_id in parent_ids
3506
if parent_id in cache]
3507
if len(possible_trees) == 0:
3508
# There either aren't any parents, or the parents aren't in the
3509
# cache, so just use the last converted tree
3510
possible_trees.append((basis_id, cache[basis_id]))
3512
for basis_id, basis_tree in possible_trees:
3513
delta = tree.inventory._make_delta(basis_tree.inventory)
3514
deltas.append((len(delta), basis_id, delta))
3516
return deltas[0][1:]
3518
def _get_parent_keys(self, root_key, parent_map):
3519
"""Get the parent keys for a given root id."""
3520
root_id, rev_id = root_key
3521
# Include direct parents of the revision, but only if they used
3522
# the same root_id and are heads.
3524
for parent_id in parent_map[rev_id]:
3525
if parent_id == _mod_revision.NULL_REVISION:
3527
if parent_id not in self._revision_id_to_root_id:
3528
# We probably didn't read this revision, go spend the
3529
# extra effort to actually check
3531
tree = self.source.revision_tree(parent_id)
3532
except errors.NoSuchRevision:
3533
# Ghost, fill out _revision_id_to_root_id in case we
3534
# encounter this again.
3535
# But set parent_root_id to None since we don't really know
3536
parent_root_id = None
3538
parent_root_id = tree.get_root_id()
3539
self._revision_id_to_root_id[parent_id] = None
3541
parent_root_id = self._revision_id_to_root_id[parent_id]
3542
if root_id == parent_root_id:
3543
# With stacking we _might_ want to refer to a non-local
3544
# revision, but this code path only applies when we have the
3545
# full content available, so ghosts really are ghosts, not just
3546
# the edge of local data.
3547
parent_keys.append((parent_id,))
3549
# root_id may be in the parent anyway.
3551
tree = self.source.revision_tree(parent_id)
3552
except errors.NoSuchRevision:
3553
# ghost, can't refer to it.
3557
parent_keys.append((tree.inventory[root_id].revision,))
3558
except errors.NoSuchId:
3561
g = graph.Graph(self.source.revisions)
3562
heads = g.heads(parent_keys)
3564
for key in parent_keys:
3565
if key in heads and key not in selected_keys:
3566
selected_keys.append(key)
3567
return tuple([(root_id,)+ key for key in selected_keys])
3569
def _new_root_data_stream(self, root_keys_to_create, parent_map):
3570
for root_key in root_keys_to_create:
3571
parent_keys = self._get_parent_keys(root_key, parent_map)
3572
yield versionedfile.FulltextContentFactory(root_key,
3573
parent_keys, None, '')
3575
def _fetch_batch(self, revision_ids, basis_id, cache):
3576
"""Fetch across a few revisions.
3578
:param revision_ids: The revisions to copy
3579
:param basis_id: The revision_id of a tree that must be in cache, used
3580
as a basis for delta when no other base is available
3581
:param cache: A cache of RevisionTrees that we can use.
3582
:return: The revision_id of the last converted tree. The RevisionTree
3583
for it will be in cache
3585
# Walk though all revisions; get inventory deltas, copy referenced
3586
# texts that delta references, insert the delta, revision and
3588
root_keys_to_create = set()
3591
pending_revisions = []
3592
parent_map = self.source.get_parent_map(revision_ids)
3593
for tree in self.source.revision_trees(revision_ids):
3594
current_revision_id = tree.get_revision_id()
3595
parent_ids = parent_map.get(current_revision_id, ())
3596
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3598
if self._converting_to_rich_root:
3599
self._revision_id_to_root_id[current_revision_id] = \
3601
# Find text entries that need to be copied
3602
for old_path, new_path, file_id, entry in delta:
3603
if new_path is not None:
3606
if not self.target.supports_rich_root():
3607
# The target doesn't support rich root, so we don't
3610
if self._converting_to_rich_root:
3611
# This can't be copied normally, we have to insert
3613
root_keys_to_create.add((file_id, entry.revision))
3615
text_keys.add((file_id, entry.revision))
3616
revision = self.source.get_revision(current_revision_id)
3617
pending_deltas.append((basis_id, delta,
3618
current_revision_id, revision.parent_ids))
3619
pending_revisions.append(revision)
3620
cache[current_revision_id] = tree
3621
basis_id = current_revision_id
3623
from_texts = self.source.texts
3624
to_texts = self.target.texts
3625
if root_keys_to_create:
3626
root_stream = self._new_root_data_stream(root_keys_to_create,
3628
to_texts.insert_record_stream(root_stream)
3629
to_texts.insert_record_stream(from_texts.get_record_stream(
3630
text_keys, self.target._format._fetch_order,
3631
not self.target._format._fetch_uses_deltas))
3632
# insert inventory deltas
3633
for delta in pending_deltas:
3634
self.target.add_inventory_by_delta(*delta)
3635
if self.target._fallback_repositories:
3636
# Make sure this stacked repository has all the parent inventories
3637
# for the new revisions that we are about to insert. We do this
3638
# before adding the revisions so that no revision is added until
3639
# all the inventories it may depend on are added.
3641
revision_ids = set()
3642
for revision in pending_revisions:
3643
revision_ids.add(revision.revision_id)
3644
parent_ids.update(revision.parent_ids)
3645
parent_ids.difference_update(revision_ids)
3646
parent_ids.discard(_mod_revision.NULL_REVISION)
3647
parent_map = self.source.get_parent_map(parent_ids)
3648
for parent_tree in self.source.revision_trees(parent_ids):
3649
basis_id, delta = self._get_delta_for_revision(tree, parent_ids, basis_id, cache)
3650
current_revision_id = parent_tree.get_revision_id()
3651
parents_parents = parent_map[current_revision_id]
3652
self.target.add_inventory_by_delta(
3653
basis_id, delta, current_revision_id, parents_parents)
3654
# insert signatures and revisions
3655
for revision in pending_revisions:
3657
signature = self.source.get_signature_text(
3658
revision.revision_id)
3659
self.target.add_signature_text(revision.revision_id,
3661
except errors.NoSuchRevision:
3663
self.target.add_revision(revision.revision_id, revision)
3666
def _fetch_all_revisions(self, revision_ids, pb):
3667
"""Fetch everything for the list of revisions.
3669
:param revision_ids: The list of revisions to fetch. Must be in
3671
:param pb: A ProgressBar
3674
basis_id, basis_tree = self._get_basis(revision_ids[0])
3676
cache = lru_cache.LRUCache(100)
3677
cache[basis_id] = basis_tree
3678
del basis_tree # We don't want to hang on to it here
3680
for offset in range(0, len(revision_ids), batch_size):
3681
self.target.start_write_group()
3683
pb.update('Transferring revisions', offset,
3685
batch = revision_ids[offset:offset+batch_size]
3686
basis_id = self._fetch_batch(batch, basis_id, cache)
3688
self.target.abort_write_group()
3691
hint = self.target.commit_write_group()
3694
if hints and self.target._format.pack_compresses:
3695
self.target.pack(hint=hints)
3696
pb.update('Transferring revisions', len(revision_ids),
3700
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3702
"""See InterRepository.fetch()."""
3703
if fetch_spec is not None:
3704
raise AssertionError("Not implemented yet...")
3705
if (not self.source.supports_rich_root()
3706
and self.target.supports_rich_root()):
3707
self._converting_to_rich_root = True
3708
self._revision_id_to_root_id = {}
3710
self._converting_to_rich_root = False
3711
revision_ids = self.target.search_missing_revision_ids(self.source,
3712
revision_id, find_ghosts=find_ghosts).get_keys()
3713
if not revision_ids:
3715
revision_ids = tsort.topo_sort(
3716
self.source.get_graph().get_parent_map(revision_ids))
3717
if not revision_ids:
3719
# Walk though all revisions; get inventory deltas, copy referenced
3720
# texts that delta references, insert the delta, revision and
3722
first_rev = self.source.get_revision(revision_ids[0])
3724
my_pb = ui.ui_factory.nested_progress_bar()
3727
symbol_versioning.warn(
3728
symbol_versioning.deprecated_in((1, 14, 0))
3729
% "pb parameter to fetch()")
3732
self._fetch_all_revisions(revision_ids, pb)
3734
if my_pb is not None:
3736
return len(revision_ids), 0
3738
def _get_basis(self, first_revision_id):
3739
"""Get a revision and tree which exists in the target.
3741
This assumes that first_revision_id is selected for transmission
3742
because all other ancestors are already present. If we can't find an
3743
ancestor we fall back to NULL_REVISION since we know that is safe.
3745
:return: (basis_id, basis_tree)
3747
first_rev = self.source.get_revision(first_revision_id)
3749
basis_id = first_rev.parent_ids[0]
3750
# only valid as a basis if the target has it
3751
self.target.get_revision(basis_id)
3752
# Try to get a basis tree - if its a ghost it will hit the
3753
# NoSuchRevision case.
3754
basis_tree = self.source.revision_tree(basis_id)
3755
except (IndexError, errors.NoSuchRevision):
3756
basis_id = _mod_revision.NULL_REVISION
3757
basis_tree = self.source.revision_tree(basis_id)
3758
return basis_id, basis_tree
3761
InterRepository.register_optimiser(InterDifferingSerializer)
3762
InterRepository.register_optimiser(InterSameDataRepository)
3763
InterRepository.register_optimiser(InterWeaveRepo)
3764
InterRepository.register_optimiser(InterKnitRepo)
1776
3767
class CopyConverter(object):
1777
3768
"""A repository conversion tool which just performs a copy of the content.
1792
3783
:param to_convert: The disk object to convert.
1793
3784
:param pb: a progress bar to use for progress information.
1795
pb = ui.ui_factory.nested_progress_bar()
1798
3789
# this is only useful with metadir layouts - separated repo content.
1799
3790
# trigger an assertion if not such
1800
3791
repo._format.get_format_string()
1801
3792
self.repo_dir = repo.bzrdir
1802
pb.update('Moving repository to repository.backup')
3793
self.step('Moving repository to repository.backup')
1803
3794
self.repo_dir.transport.move('repository', 'repository.backup')
1804
3795
backup_transport = self.repo_dir.transport.clone('repository.backup')
1805
3796
repo._format.check_conversion_target(self.target_format)
1806
3797
self.source_repo = repo._format.open(self.repo_dir,
1808
3799
_override_transport=backup_transport)
1809
pb.update('Creating new repository')
3800
self.step('Creating new repository')
1810
3801
converted = self.target_format.initialize(self.repo_dir,
1811
3802
self.source_repo.is_shared())
1812
3803
converted.lock_write()
1814
pb.update('Copying content')
3805
self.step('Copying content into repository.')
1815
3806
self.source_repo.copy_content_into(converted)
1817
3808
converted.unlock()
1818
pb.update('Deleting old repository content')
3809
self.step('Deleting old repository content.')
1819
3810
self.repo_dir.transport.delete_tree('repository.backup')
1820
ui.ui_factory.note('repository converted')
3811
self.pb.note('repository converted')
3813
def step(self, message):
3814
"""Update the pb by a step."""
3816
self.pb.update(message, self.count, self.total)
3828
def _unescaper(match, _map=_unescape_map):
3829
code = match.group(1)
3833
if not code.startswith('#'):
3835
return unichr(int(code[1:])).encode('utf8')
3841
def _unescape_xml(data):
3842
"""Unescape predefined XML entities in a string of data."""
3844
if _unescape_re is None:
3845
_unescape_re = re.compile('\&([^;]*);')
3846
return _unescape_re.sub(_unescaper, data)
3849
class _VersionedFileChecker(object):
3851
def __init__(self, repository, text_key_references=None):
3852
self.repository = repository
3853
self.text_index = self.repository._generate_text_key_index(
3854
text_key_references=text_key_references)
3856
def calculate_file_version_parents(self, text_key):
3857
"""Calculate the correct parents for a file version according to
3860
parent_keys = self.text_index[text_key]
3861
if parent_keys == [_mod_revision.NULL_REVISION]:
3863
return tuple(parent_keys)
3865
def check_file_version_parents(self, texts, progress_bar=None):
3866
"""Check the parents stored in a versioned file are correct.
3868
It also detects file versions that are not referenced by their
3869
corresponding revision's inventory.
3871
:returns: A tuple of (wrong_parents, dangling_file_versions).
3872
wrong_parents is a dict mapping {revision_id: (stored_parents,
3873
correct_parents)} for each revision_id where the stored parents
3874
are not correct. dangling_file_versions is a set of (file_id,
3875
revision_id) tuples for versions that are present in this versioned
3876
file, but not used by the corresponding inventory.
3879
self.file_ids = set([file_id for file_id, _ in
3880
self.text_index.iterkeys()])
3881
# text keys is now grouped by file_id
3882
n_weaves = len(self.file_ids)
3883
files_in_revisions = {}
3884
revisions_of_files = {}
3885
n_versions = len(self.text_index)
3886
progress_bar.update('loading text store', 0, n_versions)
3887
parent_map = self.repository.texts.get_parent_map(self.text_index)
3888
# On unlistable transports this could well be empty/error...
3889
text_keys = self.repository.texts.keys()
3890
unused_keys = frozenset(text_keys) - set(self.text_index)
3891
for num, key in enumerate(self.text_index.iterkeys()):
3892
if progress_bar is not None:
3893
progress_bar.update('checking text graph', num, n_versions)
3894
correct_parents = self.calculate_file_version_parents(key)
3896
knit_parents = parent_map[key]
3897
except errors.RevisionNotPresent:
3900
if correct_parents != knit_parents:
3901
wrong_parents[key] = (knit_parents, correct_parents)
3902
return wrong_parents, unused_keys
3905
def _old_get_graph(repository, revision_id):
3906
"""DO NOT USE. That is all. I'm serious."""
3907
graph = repository.get_graph()
3908
revision_graph = dict(((key, value) for key, value in
3909
graph.iter_ancestry([revision_id]) if value is not None))
3910
return _strip_NULL_ghosts(revision_graph)
1824
3913
def _strip_NULL_ghosts(revision_graph):
1832
3921
return revision_graph
3924
class StreamSink(object):
3925
"""An object that can insert a stream into a repository.
3927
This interface handles the complexity of reserialising inventories and
3928
revisions from different formats, and allows unidirectional insertion into
3929
stacked repositories without looking for the missing basis parents
3933
def __init__(self, target_repo):
3934
self.target_repo = target_repo
3936
def insert_stream(self, stream, src_format, resume_tokens):
3937
"""Insert a stream's content into the target repository.
3939
:param src_format: a bzr repository format.
3941
:return: a list of resume tokens and an iterable of keys additional
3942
items required before the insertion can be completed.
3944
self.target_repo.lock_write()
3947
self.target_repo.resume_write_group(resume_tokens)
3950
self.target_repo.start_write_group()
3953
# locked_insert_stream performs a commit|suspend.
3954
return self._locked_insert_stream(stream, src_format, is_resume)
3956
self.target_repo.abort_write_group(suppress_errors=True)
3959
self.target_repo.unlock()
3961
def _locked_insert_stream(self, stream, src_format, is_resume):
3962
to_serializer = self.target_repo._format._serializer
3963
src_serializer = src_format._serializer
3965
if to_serializer == src_serializer:
3966
# If serializers match and the target is a pack repository, set the
3967
# write cache size on the new pack. This avoids poor performance
3968
# on transports where append is unbuffered (such as
3969
# RemoteTransport). This is safe to do because nothing should read
3970
# back from the target repository while a stream with matching
3971
# serialization is being inserted.
3972
# The exception is that a delta record from the source that should
3973
# be a fulltext may need to be expanded by the target (see
3974
# test_fetch_revisions_with_deltas_into_pack); but we take care to
3975
# explicitly flush any buffered writes first in that rare case.
3977
new_pack = self.target_repo._pack_collection._new_pack
3978
except AttributeError:
3979
# Not a pack repository
3982
new_pack.set_write_cache_size(1024*1024)
3983
for substream_type, substream in stream:
3984
if substream_type == 'texts':
3985
self.target_repo.texts.insert_record_stream(substream)
3986
elif substream_type == 'inventories':
3987
if src_serializer == to_serializer:
3988
self.target_repo.inventories.insert_record_stream(
3991
self._extract_and_insert_inventories(
3992
substream, src_serializer)
3993
elif substream_type == 'chk_bytes':
3994
# XXX: This doesn't support conversions, as it assumes the
3995
# conversion was done in the fetch code.
3996
self.target_repo.chk_bytes.insert_record_stream(substream)
3997
elif substream_type == 'revisions':
3998
# This may fallback to extract-and-insert more often than
3999
# required if the serializers are different only in terms of
4001
if src_serializer == to_serializer:
4002
self.target_repo.revisions.insert_record_stream(
4005
self._extract_and_insert_revisions(substream,
4007
elif substream_type == 'signatures':
4008
self.target_repo.signatures.insert_record_stream(substream)
4010
raise AssertionError('kaboom! %s' % (substream_type,))
4011
# Done inserting data, and the missing_keys calculations will try to
4012
# read back from the inserted data, so flush the writes to the new pack
4013
# (if this is pack format).
4014
if new_pack is not None:
4015
new_pack._write_data('', flush=True)
4016
# Find all the new revisions (including ones from resume_tokens)
4017
missing_keys = self.target_repo.get_missing_parent_inventories(
4018
check_for_missing_texts=is_resume)
4020
for prefix, versioned_file in (
4021
('texts', self.target_repo.texts),
4022
('inventories', self.target_repo.inventories),
4023
('revisions', self.target_repo.revisions),
4024
('signatures', self.target_repo.signatures),
4025
('chk_bytes', self.target_repo.chk_bytes),
4027
if versioned_file is None:
4029
missing_keys.update((prefix,) + key for key in
4030
versioned_file.get_missing_compression_parent_keys())
4031
except NotImplementedError:
4032
# cannot even attempt suspending, and missing would have failed
4033
# during stream insertion.
4034
missing_keys = set()
4037
# suspend the write group and tell the caller what we is
4038
# missing. We know we can suspend or else we would not have
4039
# entered this code path. (All repositories that can handle
4040
# missing keys can handle suspending a write group).
4041
write_group_tokens = self.target_repo.suspend_write_group()
4042
return write_group_tokens, missing_keys
4043
hint = self.target_repo.commit_write_group()
4044
if (to_serializer != src_serializer and
4045
self.target_repo._format.pack_compresses):
4046
self.target_repo.pack(hint=hint)
4049
def _extract_and_insert_inventories(self, substream, serializer):
4050
"""Generate a new inventory versionedfile in target, converting data.
4052
The inventory is retrieved from the source, (deserializing it), and
4053
stored in the target (reserializing it in a different format).
4055
for record in substream:
4056
bytes = record.get_bytes_as('fulltext')
4057
revision_id = record.key[0]
4058
inv = serializer.read_inventory_from_string(bytes, revision_id)
4059
parents = [key[0] for key in record.parents]
4060
self.target_repo.add_inventory(revision_id, inv, parents)
4062
def _extract_and_insert_revisions(self, substream, serializer):
4063
for record in substream:
4064
bytes = record.get_bytes_as('fulltext')
4065
revision_id = record.key[0]
4066
rev = serializer.read_revision_from_string(bytes)
4067
if rev.revision_id != revision_id:
4068
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4069
self.target_repo.add_revision(revision_id, rev)
4072
if self.target_repo._format._fetch_reconcile:
4073
self.target_repo.reconcile()
4076
class StreamSource(object):
4077
"""A source of a stream for fetching between repositories."""
4079
def __init__(self, from_repository, to_format):
4080
"""Create a StreamSource streaming from from_repository."""
4081
self.from_repository = from_repository
4082
self.to_format = to_format
4084
def delta_on_metadata(self):
4085
"""Return True if delta's are permitted on metadata streams.
4087
That is on revisions and signatures.
4089
src_serializer = self.from_repository._format._serializer
4090
target_serializer = self.to_format._serializer
4091
return (self.to_format._fetch_uses_deltas and
4092
src_serializer == target_serializer)
4094
def _fetch_revision_texts(self, revs):
4095
# fetch signatures first and then the revision texts
4096
# may need to be a InterRevisionStore call here.
4097
from_sf = self.from_repository.signatures
4098
# A missing signature is just skipped.
4099
keys = [(rev_id,) for rev_id in revs]
4100
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4102
self.to_format._fetch_order,
4103
not self.to_format._fetch_uses_deltas))
4104
# If a revision has a delta, this is actually expanded inside the
4105
# insert_record_stream code now, which is an alternate fix for
4107
from_rf = self.from_repository.revisions
4108
revisions = from_rf.get_record_stream(
4110
self.to_format._fetch_order,
4111
not self.delta_on_metadata())
4112
return [('signatures', signatures), ('revisions', revisions)]
4114
def _generate_root_texts(self, revs):
4115
"""This will be called by __fetch between fetching weave texts and
4116
fetching the inventory weave.
4118
Subclasses should override this if they need to generate root texts
4119
after fetching weave texts.
4121
if self._rich_root_upgrade():
4123
return bzrlib.fetch.Inter1and2Helper(
4124
self.from_repository).generate_root_texts(revs)
4128
def get_stream(self, search):
4130
revs = search.get_keys()
4131
graph = self.from_repository.get_graph()
4132
revs = list(graph.iter_topo_order(revs))
4133
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4135
for knit_kind, file_id, revisions in data_to_fetch:
4136
if knit_kind != phase:
4138
# Make a new progress bar for this phase
4139
if knit_kind == "file":
4140
# Accumulate file texts
4141
text_keys.extend([(file_id, revision) for revision in
4143
elif knit_kind == "inventory":
4144
# Now copy the file texts.
4145
from_texts = self.from_repository.texts
4146
yield ('texts', from_texts.get_record_stream(
4147
text_keys, self.to_format._fetch_order,
4148
not self.to_format._fetch_uses_deltas))
4149
# Cause an error if a text occurs after we have done the
4152
# Before we process the inventory we generate the root
4153
# texts (if necessary) so that the inventories references
4155
for _ in self._generate_root_texts(revs):
4157
# NB: This currently reopens the inventory weave in source;
4158
# using a single stream interface instead would avoid this.
4159
from_weave = self.from_repository.inventories
4160
# we fetch only the referenced inventories because we do not
4161
# know for unselected inventories whether all their required
4162
# texts are present in the other repository - it could be
4164
for info in self._get_inventory_stream(revs):
4166
elif knit_kind == "signatures":
4167
# Nothing to do here; this will be taken care of when
4168
# _fetch_revision_texts happens.
4170
elif knit_kind == "revisions":
4171
for record in self._fetch_revision_texts(revs):
4174
raise AssertionError("Unknown knit kind %r" % knit_kind)
4176
def get_stream_for_missing_keys(self, missing_keys):
4177
# missing keys can only occur when we are byte copying and not
4178
# translating (because translation means we don't send
4179
# unreconstructable deltas ever).
4181
keys['texts'] = set()
4182
keys['revisions'] = set()
4183
keys['inventories'] = set()
4184
keys['chk_bytes'] = set()
4185
keys['signatures'] = set()
4186
for key in missing_keys:
4187
keys[key[0]].add(key[1:])
4188
if len(keys['revisions']):
4189
# If we allowed copying revisions at this point, we could end up
4190
# copying a revision without copying its required texts: a
4191
# violation of the requirements for repository integrity.
4192
raise AssertionError(
4193
'cannot copy revisions to fill in missing deltas %s' % (
4194
keys['revisions'],))
4195
for substream_kind, keys in keys.iteritems():
4196
vf = getattr(self.from_repository, substream_kind)
4197
if vf is None and keys:
4198
raise AssertionError(
4199
"cannot fill in keys for a versioned file we don't"
4200
" have: %s needs %s" % (substream_kind, keys))
4202
# No need to stream something we don't have
4204
# Ask for full texts always so that we don't need more round trips
4205
# after this stream.
4206
# Some of the missing keys are genuinely ghosts, so filter absent
4207
# records. The Sink is responsible for doing another check to
4208
# ensure that ghosts don't introduce missing data for future
4210
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4211
self.to_format._fetch_order, True))
4212
yield substream_kind, stream
4214
def inventory_fetch_order(self):
4215
if self._rich_root_upgrade():
4216
return 'topological'
4218
return self.to_format._fetch_order
4220
def _rich_root_upgrade(self):
4221
return (not self.from_repository._format.rich_root_data and
4222
self.to_format.rich_root_data)
4224
def _get_inventory_stream(self, revision_ids):
4225
from_format = self.from_repository._format
4226
if (from_format.supports_chks and self.to_format.supports_chks
4227
and (from_format._serializer == self.to_format._serializer)):
4228
# Both sides support chks, and they use the same serializer, so it
4229
# is safe to transmit the chk pages and inventory pages across
4231
return self._get_chk_inventory_stream(revision_ids)
4232
elif (not from_format.supports_chks):
4233
# Source repository doesn't support chks. So we can transmit the
4234
# inventories 'as-is' and either they are just accepted on the
4235
# target, or the Sink will properly convert it.
4236
return self._get_simple_inventory_stream(revision_ids)
4238
# XXX: Hack to make not-chk->chk fetch: copy the inventories as
4239
# inventories. Note that this should probably be done somehow
4240
# as part of bzrlib.repository.StreamSink. Except JAM couldn't
4241
# figure out how a non-chk repository could possibly handle
4242
# deserializing an inventory stream from a chk repo, as it
4243
# doesn't have a way to understand individual pages.
4244
return self._get_convertable_inventory_stream(revision_ids)
4246
def _get_simple_inventory_stream(self, revision_ids):
4247
from_weave = self.from_repository.inventories
4248
yield ('inventories', from_weave.get_record_stream(
4249
[(rev_id,) for rev_id in revision_ids],
4250
self.inventory_fetch_order(),
4251
not self.delta_on_metadata()))
4253
def _get_chk_inventory_stream(self, revision_ids):
4254
"""Fetch the inventory texts, along with the associated chk maps."""
4255
# We want an inventory outside of the search set, so that we can filter
4256
# out uninteresting chk pages. For now we use
4257
# _find_revision_outside_set, but if we had a Search with cut_revs, we
4258
# could use that instead.
4259
start_rev_id = self.from_repository._find_revision_outside_set(
4261
start_rev_key = (start_rev_id,)
4262
inv_keys_to_fetch = [(rev_id,) for rev_id in revision_ids]
4263
if start_rev_id != _mod_revision.NULL_REVISION:
4264
inv_keys_to_fetch.append((start_rev_id,))
4265
# Any repo that supports chk_bytes must also support out-of-order
4266
# insertion. At least, that is how we expect it to work
4267
# We use get_record_stream instead of iter_inventories because we want
4268
# to be able to insert the stream as well. We could instead fetch
4269
# allowing deltas, and then iter_inventories, but we don't know whether
4270
# source or target is more 'local' anway.
4271
inv_stream = self.from_repository.inventories.get_record_stream(
4272
inv_keys_to_fetch, 'unordered',
4273
True) # We need them as full-texts so we can find their references
4274
uninteresting_chk_roots = set()
4275
interesting_chk_roots = set()
4276
def filter_inv_stream(inv_stream):
4277
for idx, record in enumerate(inv_stream):
4278
### child_pb.update('fetch inv', idx, len(inv_keys_to_fetch))
4279
bytes = record.get_bytes_as('fulltext')
4280
chk_inv = inventory.CHKInventory.deserialise(
4281
self.from_repository.chk_bytes, bytes, record.key)
4282
if record.key == start_rev_key:
4283
uninteresting_chk_roots.add(chk_inv.id_to_entry.key())
4284
p_id_map = chk_inv.parent_id_basename_to_file_id
4285
if p_id_map is not None:
4286
uninteresting_chk_roots.add(p_id_map.key())
4289
interesting_chk_roots.add(chk_inv.id_to_entry.key())
4290
p_id_map = chk_inv.parent_id_basename_to_file_id
4291
if p_id_map is not None:
4292
interesting_chk_roots.add(p_id_map.key())
4293
### pb.update('fetch inventory', 0, 2)
4294
yield ('inventories', filter_inv_stream(inv_stream))
4295
# Now that we have worked out all of the interesting root nodes, grab
4296
# all of the interesting pages and insert them
4297
### pb.update('fetch inventory', 1, 2)
4298
interesting = chk_map.iter_interesting_nodes(
4299
self.from_repository.chk_bytes, interesting_chk_roots,
4300
uninteresting_chk_roots)
4301
def to_stream_adapter():
4302
"""Adapt the iter_interesting_nodes result to a single stream.
4304
iter_interesting_nodes returns records as it processes them, along
4305
with keys. However, we only want to return the records themselves.
4307
for record, items in interesting:
4308
if record is not None:
4310
# XXX: We could instead call get_record_stream(records.keys())
4311
# ATM, this will always insert the records as fulltexts, and
4312
# requires that you can hang on to records once you have gone
4313
# on to the next one. Further, it causes the target to
4314
# recompress the data. Testing shows it to be faster than
4315
# requesting the records again, though.
4316
yield ('chk_bytes', to_stream_adapter())
4317
### pb.update('fetch inventory', 2, 2)
4319
def _get_convertable_inventory_stream(self, revision_ids):
4320
# XXX: One of source or target is using chks, and they don't have
4321
# compatible serializations. The StreamSink code expects to be
4322
# able to convert on the target, so we need to put
4323
# bytes-on-the-wire that can be converted
4324
yield ('inventories', self._stream_invs_as_fulltexts(revision_ids))
4326
def _stream_invs_as_fulltexts(self, revision_ids):
4327
from_repo = self.from_repository
4328
from_serializer = from_repo._format._serializer
4329
revision_keys = [(rev_id,) for rev_id in revision_ids]
4330
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4331
for inv in self.from_repository.iter_inventories(revision_ids):
4332
# XXX: This is a bit hackish, but it works. Basically,
4333
# CHKSerializer 'accidentally' supports
4334
# read/write_inventory_to_string, even though that is never
4335
# the format that is stored on disk. It *does* give us a
4336
# single string representation for an inventory, so live with
4338
# This would be far better if we had a 'serialized inventory
4339
# delta' form. Then we could use 'inventory._make_delta', and
4340
# transmit that. This would both be faster to generate, and
4341
# result in fewer bytes-on-the-wire.
4342
as_bytes = from_serializer.write_inventory_to_string(inv)
4343
key = (inv.revision_id,)
4344
parent_keys = parent_map.get(key, ())
4345
yield versionedfile.FulltextContentFactory(
4346
key, parent_keys, None, as_bytes)
1835
4349
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
1836
4350
stop_revision=None):
1837
4351
"""Extend the partial history to include a given index