203
333
commit to be valid, deletes against the basis MUST be recorded via
204
334
builder.record_delete().
206
raise NotImplementedError(self.will_record_deletes)
208
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):
209
551
"""Record a new tree via iter_changes.
211
553
:param tree: The tree to obtain text contents from for changed objects.
216
558
to basis_revision_id. The iterator must not include any items with
217
559
a current kind of None - missing items must be either filtered out
218
560
or errored-on beefore record_iter_changes sees the item.
561
:param _entry_factory: Private method to bind entry_factory locally for
219
563
:return: A generator of (file_id, relpath, fs_hash) tuples for use with
220
564
tree._observed_sha1.
222
raise NotImplementedError(self.record_iter_changes)
225
class RepositoryWriteLockResult(LogicalLockResult):
226
"""The result of write locking a repository.
228
:ivar repository_token: The token obtained from the underlying lock, or
230
:ivar unlock: A callable which will unlock the lock.
233
def __init__(self, unlock, repository_token):
234
LogicalLockResult.__init__(self, unlock)
235
self.repository_token = repository_token
238
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.
242
848
######################################################################
246
class Repository(_RelockDebugMixin, controldir.ControlComponent):
852
class Repository(object):
247
853
"""Repository holding history for one or more branches.
249
855
The repository holds and retrieves historical information including
250
856
revisions and file history. It's normally accessed only by the Branch,
251
857
which views a particular line of development through that history.
253
See VersionedFileRepository in bzrlib.vf_repository for the
254
base class for most Bazaar repositories.
859
The Repository builds on top of some byte storage facilies (the revisions,
860
signatures, inventories, texts and chk_bytes attributes) and a Transport,
861
which respectively provide byte storage and a means to access the (possibly
864
The byte storage facilities are addressed via tuples, which we refer to
865
as 'keys' throughout the code base. Revision_keys, inventory_keys and
866
signature_keys are all 1-tuples: (revision_id,). text_keys are two-tuples:
867
(file_id, revision_id). chk_bytes uses CHK keys - a 1-tuple with a single
868
byte string made up of a hash identifier and a hash value.
869
We use this interface because it allows low friction with the underlying
870
code that implements disk indices, network encoding and other parts of
873
:ivar revisions: A bzrlib.versionedfile.VersionedFiles instance containing
874
the serialised revisions for the repository. This can be used to obtain
875
revision graph information or to access raw serialised revisions.
876
The result of trying to insert data into the repository via this store
877
is undefined: it should be considered read-only except for implementors
879
:ivar signatures: A bzrlib.versionedfile.VersionedFiles instance containing
880
the serialised signatures for the repository. This can be used to
881
obtain access to raw serialised signatures. The result of trying to
882
insert data into the repository via this store is undefined: it should
883
be considered read-only except for implementors of repositories.
884
:ivar inventories: A bzrlib.versionedfile.VersionedFiles instance containing
885
the serialised inventories for the repository. This can be used to
886
obtain unserialised inventories. The result of trying to insert data
887
into the repository via this store is undefined: it should be
888
considered read-only except for implementors of repositories.
889
:ivar texts: A bzrlib.versionedfile.VersionedFiles instance containing the
890
texts of files and directories for the repository. This can be used to
891
obtain file texts or file graphs. Note that Repository.iter_file_bytes
892
is usually a better interface for accessing file texts.
893
The result of trying to insert data into the repository via this store
894
is undefined: it should be considered read-only except for implementors
896
:ivar chk_bytes: A bzrlib.versionedfile.VersionedFiles instance containing
897
any data the repository chooses to store or have indexed by its hash.
898
The result of trying to insert data into the repository via this store
899
is undefined: it should be considered read-only except for implementors
901
:ivar _transport: Transport for file access to repository, typically
902
pointing to .bzr/repository.
905
# What class to use for a CommitBuilder. Often its simpler to change this
906
# in a Repository class subclass rather than to override
907
# get_commit_builder.
908
_commit_builder_class = CommitBuilder
909
# The search regex used by xml based repositories to determine what things
910
# where changed in a single commit.
911
_file_ids_altered_regex = lazy_regex.lazy_compile(
912
r'file_id="(?P<file_id>[^"]+)"'
913
r'.* revision="(?P<revision_id>[^"]+)"'
257
916
def abort_write_group(self, suppress_errors=False):
258
917
"""Commit the contents accrued within the current write group.
314
981
return InterRepository._assert_same_model(self, repository)
983
def add_inventory(self, revision_id, inv, parents):
984
"""Add the inventory inv to the repository as revision_id.
986
:param parents: The revision ids of the parents that revision_id
987
is known to have and are in the repository already.
989
:returns: The validator(which is a sha1 digest, though what is sha'd is
990
repository format specific) of the serialized inventory.
992
if not self.is_in_write_group():
993
raise AssertionError("%r not in write group" % (self,))
994
_mod_revision.check_not_reserved_id(revision_id)
995
if not (inv.revision_id is None or inv.revision_id == revision_id):
996
raise AssertionError(
997
"Mismatch between inventory revision"
998
" id and insertion revid (%r, %r)"
999
% (inv.revision_id, revision_id))
1000
if inv.root is None:
1001
raise AssertionError()
1002
return self._add_inventory_checked(revision_id, inv, parents)
1004
def _add_inventory_checked(self, revision_id, inv, parents):
1005
"""Add inv to the repository after checking the inputs.
1007
This function can be overridden to allow different inventory styles.
1009
:seealso: add_inventory, for the contract.
1011
inv_lines = self._serialise_inventory_to_lines(inv)
1012
return self._inventory_add_lines(revision_id, parents,
1013
inv_lines, check_content=False)
1015
def add_inventory_by_delta(self, basis_revision_id, delta, new_revision_id,
1016
parents, basis_inv=None, propagate_caches=False):
1017
"""Add a new inventory expressed as a delta against another revision.
1019
See the inventory developers documentation for the theory behind
1022
:param basis_revision_id: The inventory id the delta was created
1023
against. (This does not have to be a direct parent.)
1024
:param delta: The inventory delta (see Inventory.apply_delta for
1026
:param new_revision_id: The revision id that the inventory is being
1028
:param parents: The revision ids of the parents that revision_id is
1029
known to have and are in the repository already. These are supplied
1030
for repositories that depend on the inventory graph for revision
1031
graph access, as well as for those that pun ancestry with delta
1033
:param basis_inv: The basis inventory if it is already known,
1035
:param propagate_caches: If True, the caches for this inventory are
1036
copied to and updated for the result if possible.
1038
:returns: (validator, new_inv)
1039
The validator(which is a sha1 digest, though what is sha'd is
1040
repository format specific) of the serialized inventory, and the
1041
resulting inventory.
1043
if not self.is_in_write_group():
1044
raise AssertionError("%r not in write group" % (self,))
1045
_mod_revision.check_not_reserved_id(new_revision_id)
1046
basis_tree = self.revision_tree(basis_revision_id)
1047
basis_tree.lock_read()
1049
# Note that this mutates the inventory of basis_tree, which not all
1050
# inventory implementations may support: A better idiom would be to
1051
# return a new inventory, but as there is no revision tree cache in
1052
# repository this is safe for now - RBC 20081013
1053
if basis_inv is None:
1054
basis_inv = basis_tree.inventory
1055
basis_inv.apply_delta(delta)
1056
basis_inv.revision_id = new_revision_id
1057
return (self.add_inventory(new_revision_id, basis_inv, parents),
1062
def _inventory_add_lines(self, revision_id, parents, lines,
1063
check_content=True):
1064
"""Store lines in inv_vf and return the sha1 of the inventory."""
1065
parents = [(parent,) for parent in parents]
1066
return self.inventories.add_lines((revision_id,), parents, lines,
1067
check_content=check_content)[0]
1069
def add_revision(self, revision_id, rev, inv=None, config=None):
1070
"""Add rev to the revision store as revision_id.
1072
:param revision_id: the revision id to use.
1073
:param rev: The revision object.
1074
:param inv: The inventory for the revision. if None, it will be looked
1075
up in the inventory storer
1076
:param config: If None no digital signature will be created.
1077
If supplied its signature_needed method will be used
1078
to determine if a signature should be made.
1080
# TODO: jam 20070210 Shouldn't we check rev.revision_id and
1082
_mod_revision.check_not_reserved_id(revision_id)
1083
if config is not None and config.signature_needed():
1085
inv = self.get_inventory(revision_id)
1086
plaintext = Testament(rev, inv).as_short_text()
1087
self.store_revision_signature(
1088
gpg.GPGStrategy(config), plaintext, revision_id)
1089
# check inventory present
1090
if not self.inventories.get_parent_map([(revision_id,)]):
1092
raise errors.WeaveRevisionNotPresent(revision_id,
1095
# yes, this is not suitable for adding with ghosts.
1096
rev.inventory_sha1 = self.add_inventory(revision_id, inv,
1099
key = (revision_id,)
1100
rev.inventory_sha1 = self.inventories.get_sha1s([key])[key]
1101
self._add_revision(rev)
1103
def _add_revision(self, revision):
1104
text = self._serializer.write_revision_to_string(revision)
1105
key = (revision.revision_id,)
1106
parents = tuple((parent,) for parent in revision.parent_ids)
1107
self.revisions.add_lines(key, parents, osutils.split_lines(text))
316
1109
def all_revision_ids(self):
317
1110
"""Returns a list of all the revision ids in the repository.
909
1810
signature = gpg_strategy.sign(plaintext)
910
1811
self.add_signature_text(revision_id, signature)
912
1814
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)
1815
self.signatures.add_lines((revision_id,), (),
1816
osutils.split_lines(signature))
1818
def find_text_key_references(self):
1819
"""Find the text key references within the repository.
1821
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1822
to whether they were referred to by the inventory of the
1823
revision_id that they contain. The inventory texts from all present
1824
revision ids are assessed to generate this report.
1826
revision_keys = self.revisions.keys()
1827
w = self.inventories
1828
pb = ui.ui_factory.nested_progress_bar()
1830
return self._find_text_key_references_from_xml_inventory_lines(
1831
w.iter_lines_added_or_present_in_keys(revision_keys, pb=pb))
1835
def _find_text_key_references_from_xml_inventory_lines(self,
1837
"""Core routine for extracting references to texts from inventories.
1839
This performs the translation of xml lines to revision ids.
1841
:param line_iterator: An iterator of lines, origin_version_id
1842
:return: A dictionary mapping text keys ((fileid, revision_id) tuples)
1843
to whether they were referred to by the inventory of the
1844
revision_id that they contain. Note that if that revision_id was
1845
not part of the line_iterator's output then False will be given -
1846
even though it may actually refer to that key.
1848
if not self._serializer.support_altered_by_hack:
1849
raise AssertionError(
1850
"_find_text_key_references_from_xml_inventory_lines only "
1851
"supported for branches which store inventory as unnested xml"
1852
", not on %r" % self)
1855
# this code needs to read every new line in every inventory for the
1856
# inventories [revision_ids]. Seeing a line twice is ok. Seeing a line
1857
# not present in one of those inventories is unnecessary but not
1858
# harmful because we are filtering by the revision id marker in the
1859
# inventory lines : we only select file ids altered in one of those
1860
# revisions. We don't need to see all lines in the inventory because
1861
# only those added in an inventory in rev X can contain a revision=X
1863
unescape_revid_cache = {}
1864
unescape_fileid_cache = {}
1866
# jam 20061218 In a big fetch, this handles hundreds of thousands
1867
# of lines, so it has had a lot of inlining and optimizing done.
1868
# Sorry that it is a little bit messy.
1869
# Move several functions to be local variables, since this is a long
1871
search = self._file_ids_altered_regex.search
1872
unescape = _unescape_xml
1873
setdefault = result.setdefault
1874
for line, line_key in line_iterator:
1875
match = search(line)
1878
# One call to match.group() returning multiple items is quite a
1879
# bit faster than 2 calls to match.group() each returning 1
1880
file_id, revision_id = match.group('file_id', 'revision_id')
1882
# Inlining the cache lookups helps a lot when you make 170,000
1883
# lines and 350k ids, versus 8.4 unique ids.
1884
# Using a cache helps in 2 ways:
1885
# 1) Avoids unnecessary decoding calls
1886
# 2) Re-uses cached strings, which helps in future set and
1888
# (2) is enough that removing encoding entirely along with
1889
# the cache (so we are using plain strings) results in no
1890
# performance improvement.
1892
revision_id = unescape_revid_cache[revision_id]
1894
unescaped = unescape(revision_id)
1895
unescape_revid_cache[revision_id] = unescaped
1896
revision_id = unescaped
1898
# Note that unconditionally unescaping means that we deserialise
1899
# every fileid, which for general 'pull' is not great, but we don't
1900
# really want to have some many fulltexts that this matters anyway.
1903
file_id = unescape_fileid_cache[file_id]
1905
unescaped = unescape(file_id)
1906
unescape_fileid_cache[file_id] = unescaped
1909
key = (file_id, revision_id)
1910
setdefault(key, False)
1911
if revision_id == line_key[-1]:
1915
def _inventory_xml_lines_for_keys(self, keys):
1916
"""Get a line iterator of the sort needed for findind references.
1918
Not relevant for non-xml inventory repositories.
1920
Ghosts in revision_keys are ignored.
1922
:param revision_keys: The revision keys for the inventories to inspect.
1923
:return: An iterator over (inventory line, revid) for the fulltexts of
1924
all of the xml inventories specified by revision_keys.
1926
stream = self.inventories.get_record_stream(keys, 'unordered', True)
1927
for record in stream:
1928
if record.storage_kind != 'absent':
1929
chunks = record.get_bytes_as('chunked')
1930
revid = record.key[-1]
1931
lines = osutils.chunks_to_lines(chunks)
1935
def _find_file_ids_from_xml_inventory_lines(self, line_iterator,
1937
"""Helper routine for fileids_altered_by_revision_ids.
1939
This performs the translation of xml lines to revision ids.
1941
:param line_iterator: An iterator of lines, origin_version_id
1942
:param revision_keys: The revision ids to filter for. This should be a
1943
set or other type which supports efficient __contains__ lookups, as
1944
the revision key from each parsed line will be looked up in the
1945
revision_keys filter.
1946
:return: a dictionary mapping altered file-ids to an iterable of
1947
revision_ids. Each altered file-ids has the exact revision_ids that
1948
altered it listed explicitly.
1950
seen = set(self._find_text_key_references_from_xml_inventory_lines(
1951
line_iterator).iterkeys())
1952
parent_keys = self._find_parent_keys_of_revisions(revision_keys)
1953
parent_seen = set(self._find_text_key_references_from_xml_inventory_lines(
1954
self._inventory_xml_lines_for_keys(parent_keys)))
1955
new_keys = seen - parent_seen
1957
setdefault = result.setdefault
1958
for key in new_keys:
1959
setdefault(key[0], set()).add(key[-1])
920
1962
def _find_parent_ids_of_revisions(self, revision_ids):
921
1963
"""Find all parent ids that are mentioned in the revision graph.
941
2018
uniquely identify the file version in the caller's context. (Examples:
942
2019
an index number or a TreeTransform trans_id.)
2021
bytes_iterator is an iterable of bytestrings for the file. The
2022
kind of iterable and length of the bytestrings are unspecified, but for
2023
this implementation, it is a list of bytes produced by
2024
VersionedFile.get_record_stream().
944
2026
:param desired_files: a list of (file_id, revision_id, identifier)
947
raise NotImplementedError(self.iter_files_bytes)
2030
for file_id, revision_id, callable_data in desired_files:
2031
text_keys[(file_id, revision_id)] = callable_data
2032
for record in self.texts.get_record_stream(text_keys, 'unordered', True):
2033
if record.storage_kind == 'absent':
2034
raise errors.RevisionNotPresent(record.key, self)
2035
yield text_keys[record.key], record.get_bytes_as('chunked')
2037
def _generate_text_key_index(self, text_key_references=None,
2039
"""Generate a new text key index for the repository.
2041
This is an expensive function that will take considerable time to run.
2043
:return: A dict mapping text keys ((file_id, revision_id) tuples) to a
2044
list of parents, also text keys. When a given key has no parents,
2045
the parents list will be [NULL_REVISION].
2047
# All revisions, to find inventory parents.
2048
if ancestors is None:
2049
graph = self.get_graph()
2050
ancestors = graph.get_parent_map(self.all_revision_ids())
2051
if text_key_references is None:
2052
text_key_references = self.find_text_key_references()
2053
pb = ui.ui_factory.nested_progress_bar()
2055
return self._do_generate_text_key_index(ancestors,
2056
text_key_references, pb)
2060
def _do_generate_text_key_index(self, ancestors, text_key_references, pb):
2061
"""Helper for _generate_text_key_index to avoid deep nesting."""
2062
revision_order = tsort.topo_sort(ancestors)
2063
invalid_keys = set()
2065
for revision_id in revision_order:
2066
revision_keys[revision_id] = set()
2067
text_count = len(text_key_references)
2068
# a cache of the text keys to allow reuse; costs a dict of all the
2069
# keys, but saves a 2-tuple for every child of a given key.
2071
for text_key, valid in text_key_references.iteritems():
2073
invalid_keys.add(text_key)
2075
revision_keys[text_key[1]].add(text_key)
2076
text_key_cache[text_key] = text_key
2077
del text_key_references
2079
text_graph = graph.Graph(graph.DictParentsProvider(text_index))
2080
NULL_REVISION = _mod_revision.NULL_REVISION
2081
# Set a cache with a size of 10 - this suffices for bzr.dev but may be
2082
# too small for large or very branchy trees. However, for 55K path
2083
# trees, it would be easy to use too much memory trivially. Ideally we
2084
# could gauge this by looking at available real memory etc, but this is
2085
# always a tricky proposition.
2086
inventory_cache = lru_cache.LRUCache(10)
2087
batch_size = 10 # should be ~150MB on a 55K path tree
2088
batch_count = len(revision_order) / batch_size + 1
2090
pb.update("Calculating text parents", processed_texts, text_count)
2091
for offset in xrange(batch_count):
2092
to_query = revision_order[offset * batch_size:(offset + 1) *
2096
for rev_tree in self.revision_trees(to_query):
2097
revision_id = rev_tree.get_revision_id()
2098
parent_ids = ancestors[revision_id]
2099
for text_key in revision_keys[revision_id]:
2100
pb.update("Calculating text parents", processed_texts)
2101
processed_texts += 1
2102
candidate_parents = []
2103
for parent_id in parent_ids:
2104
parent_text_key = (text_key[0], parent_id)
2106
check_parent = parent_text_key not in \
2107
revision_keys[parent_id]
2109
# the parent parent_id is a ghost:
2110
check_parent = False
2111
# truncate the derived graph against this ghost.
2112
parent_text_key = None
2114
# look at the parent commit details inventories to
2115
# determine possible candidates in the per file graph.
2118
inv = inventory_cache[parent_id]
2120
inv = self.revision_tree(parent_id).inventory
2121
inventory_cache[parent_id] = inv
2123
parent_entry = inv[text_key[0]]
2124
except (KeyError, errors.NoSuchId):
2126
if parent_entry is not None:
2128
text_key[0], parent_entry.revision)
2130
parent_text_key = None
2131
if parent_text_key is not None:
2132
candidate_parents.append(
2133
text_key_cache[parent_text_key])
2134
parent_heads = text_graph.heads(candidate_parents)
2135
new_parents = list(parent_heads)
2136
new_parents.sort(key=lambda x:candidate_parents.index(x))
2137
if new_parents == []:
2138
new_parents = [NULL_REVISION]
2139
text_index[text_key] = new_parents
2141
for text_key in invalid_keys:
2142
text_index[text_key] = [NULL_REVISION]
2145
def item_keys_introduced_by(self, revision_ids, _files_pb=None):
2146
"""Get an iterable listing the keys of all the data introduced by a set
2149
The keys will be ordered so that the corresponding items can be safely
2150
fetched and inserted in that order.
2152
:returns: An iterable producing tuples of (knit-kind, file-id,
2153
versions). knit-kind is one of 'file', 'inventory', 'signatures',
2154
'revisions'. file-id is None unless knit-kind is 'file'.
2156
for result in self._find_file_keys_to_fetch(revision_ids, _files_pb):
2159
for result in self._find_non_file_keys_to_fetch(revision_ids):
2162
def _find_file_keys_to_fetch(self, revision_ids, pb):
2163
# XXX: it's a bit weird to control the inventory weave caching in this
2164
# generator. Ideally the caching would be done in fetch.py I think. Or
2165
# maybe this generator should explicitly have the contract that it
2166
# should not be iterated until the previously yielded item has been
2168
inv_w = self.inventories
2170
# file ids that changed
2171
file_ids = self.fileids_altered_by_revision_ids(revision_ids, inv_w)
2173
num_file_ids = len(file_ids)
2174
for file_id, altered_versions in file_ids.iteritems():
2176
pb.update("fetch texts", count, num_file_ids)
2178
yield ("file", file_id, altered_versions)
2180
def _find_non_file_keys_to_fetch(self, revision_ids):
2182
yield ("inventory", None, revision_ids)
2185
# XXX: Note ATM no callers actually pay attention to this return
2186
# instead they just use the list of revision ids and ignore
2187
# missing sigs. Consider removing this work entirely
2188
revisions_with_signatures = set(self.signatures.get_parent_map(
2189
[(r,) for r in revision_ids]))
2190
revisions_with_signatures = set(
2191
[r for (r,) in revisions_with_signatures])
2192
revisions_with_signatures.intersection_update(revision_ids)
2193
yield ("signatures", None, revisions_with_signatures)
2196
yield ("revisions", None, revision_ids)
2199
def get_inventory(self, revision_id):
2200
"""Get Inventory object by revision id."""
2201
return self.iter_inventories([revision_id]).next()
2203
def iter_inventories(self, revision_ids):
2204
"""Get many inventories by revision_ids.
2206
This will buffer some or all of the texts used in constructing the
2207
inventories in memory, but will only parse a single inventory at a
2210
:param revision_ids: The expected revision ids of the inventories.
2211
:return: An iterator of inventories.
2213
if ((None in revision_ids)
2214
or (_mod_revision.NULL_REVISION in revision_ids)):
2215
raise ValueError('cannot get null revision inventory')
2216
return self._iter_inventories(revision_ids)
2218
def _iter_inventories(self, revision_ids):
2219
"""single-document based inventory iteration."""
2220
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2221
yield self.deserialise_inventory(revision_id, text)
2223
def _iter_inventory_xmls(self, revision_ids):
2224
keys = [(revision_id,) for revision_id in revision_ids]
2225
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2227
for record in stream:
2228
if record.storage_kind != 'absent':
2229
text_chunks[record.key] = record.get_bytes_as('chunked')
2231
raise errors.NoSuchRevision(self, record.key)
2233
chunks = text_chunks.pop(key)
2234
yield ''.join(chunks), key[-1]
2236
def deserialise_inventory(self, revision_id, xml):
2237
"""Transform the xml into an inventory object.
2239
:param revision_id: The expected revision id of the inventory.
2240
:param xml: A serialised inventory.
2242
result = self._serializer.read_inventory_from_string(xml, revision_id,
2243
entry_cache=self._inventory_entry_cache)
2244
if result.revision_id != revision_id:
2245
raise AssertionError('revision id mismatch %s != %s' % (
2246
result.revision_id, revision_id))
2249
def serialise_inventory(self, inv):
2250
return self._serializer.write_inventory_to_string(inv)
2252
def _serialise_inventory_to_lines(self, inv):
2253
return self._serializer.write_inventory_to_lines(inv)
2255
def get_serializer_format(self):
2256
return self._serializer.format_num
2259
def get_inventory_xml(self, revision_id):
2260
"""Get inventory XML as a file object."""
2261
texts = self._iter_inventory_xmls([revision_id])
2263
text, revision_id = texts.next()
2264
except StopIteration:
2265
raise errors.HistoryMissing(self, 'inventory', revision_id)
2269
def get_inventory_sha1(self, revision_id):
2270
"""Return the sha1 hash of the inventory entry
2272
return self.get_revision(revision_id).inventory_sha1
949
2274
def get_rev_id_for_revno(self, revno, known_pair):
950
2275
"""Return the revision id of a revno, given a later (revno, revid)
1181
2540
@needs_write_lock
1182
2541
def sign_revision(self, revision_id, gpg_strategy):
1183
testament = _mod_testament.Testament.from_revision(self, revision_id)
1184
plaintext = testament.as_short_text()
2542
plaintext = Testament.from_revision(self, revision_id).as_short_text()
1185
2543
self.store_revision_signature(gpg_strategy, plaintext, revision_id)
1187
2545
@needs_read_lock
1188
def verify_revision(self, revision_id, gpg_strategy):
1189
"""Verify the signature on a revision.
1191
:param revision_id: the revision to verify
1192
:gpg_strategy: the GPGStrategy object to used
1194
:return: gpg.SIGNATURE_VALID or a failed SIGNATURE_ value
1196
if not self.has_signature_for_revision_id(revision_id):
1197
return gpg.SIGNATURE_NOT_SIGNED, None
1198
signature = self.get_signature_text(revision_id)
1200
testament = _mod_testament.Testament.from_revision(self, revision_id)
1201
plaintext = testament.as_short_text()
1203
return gpg_strategy.verify(signature, plaintext)
1205
2546
def has_signature_for_revision_id(self, revision_id):
1206
2547
"""Query for a revision signature for revision_id in the repository."""
1207
raise NotImplementedError(self.has_signature_for_revision_id)
2548
if not self.has_revision(revision_id):
2549
raise errors.NoSuchRevision(self, revision_id)
2550
sig_present = (1 == len(
2551
self.signatures.get_parent_map([(revision_id,)])))
1209
2555
def get_signature_text(self, revision_id):
1210
2556
"""Return the text for a signature."""
1211
raise NotImplementedError(self.get_signature_text)
2557
stream = self.signatures.get_record_stream([(revision_id,)],
2559
record = stream.next()
2560
if record.storage_kind == 'absent':
2561
raise errors.NoSuchRevision(self, revision_id)
2562
return record.get_bytes_as('fulltext')
1213
def check(self, revision_ids=None, callback_refs=None, check_repo=True):
2565
def check(self, revision_ids=None):
1214
2566
"""Check consistency of all history of given revision_ids.
1216
2568
Different repository implementations should override _check().
1218
2570
:param revision_ids: A non-empty list of revision_ids whose ancestry
1219
2571
will be checked. Typically the last revision_id of a branch.
1220
:param callback_refs: A dict of check-refs to resolve and callback
1221
the check/_check method on the items listed as wanting the ref.
1223
:param check_repo: If False do not check the repository contents, just
1224
calculate the data callback_refs requires and call them back.
1226
return self._check(revision_ids=revision_ids, callback_refs=callback_refs,
1227
check_repo=check_repo)
1229
def _check(self, revision_ids=None, callback_refs=None, check_repo=True):
1230
raise NotImplementedError(self.check)
1232
def _warn_if_deprecated(self, branch=None):
1233
if not self._format.is_deprecated():
2573
return self._check(revision_ids)
2575
def _check(self, revision_ids):
2576
result = check.Check(self)
2580
def _warn_if_deprecated(self):
1235
2581
global _deprecation_warning_done
1236
2582
if _deprecation_warning_done:
1240
conf = config.GlobalConfig()
1242
conf = branch.get_config()
1243
if conf.suppress_warning('format_deprecation'):
1245
warning("Format %s for %s is deprecated -"
1246
" please use 'bzr upgrade' to get better performance"
1247
% (self._format, self.bzrdir.transport.base))
1249
_deprecation_warning_done = True
2584
_deprecation_warning_done = True
2585
warning("Format %s for %s is deprecated - please use 'bzr upgrade' to get better performance"
2586
% (self._format, self.bzrdir.transport.base))
1251
2588
def supports_rich_root(self):
1252
2589
return self._format.rich_root_data
1267
2604
except UnicodeDecodeError:
1268
2605
raise errors.NonAsciiRevisionId(method, self)
2607
def revision_graph_can_have_wrong_parents(self):
2608
"""Is it possible for this repository to have a revision graph with
2611
If True, then this repository must also implement
2612
_find_inconsistent_revision_parents so that check and reconcile can
2613
check for inconsistencies before proceeding with other checks that may
2614
depend on the revision index being consistent.
2616
raise NotImplementedError(self.revision_graph_can_have_wrong_parents)
2619
# remove these delegates a while after bzr 0.15
2620
def __make_delegated(name, from_module):
2621
def _deprecated_repository_forwarder():
2622
symbol_versioning.warn('%s moved to %s in bzr 0.15'
2623
% (name, from_module),
2626
m = __import__(from_module, globals(), locals(), [name])
2628
return getattr(m, name)
2629
except AttributeError:
2630
raise AttributeError('module %s has no name %s'
2632
globals()[name] = _deprecated_repository_forwarder
2635
'AllInOneRepository',
2636
'WeaveMetaDirRepository',
2637
'PreSplitOutRepositoryFormat',
2638
'RepositoryFormat4',
2639
'RepositoryFormat5',
2640
'RepositoryFormat6',
2641
'RepositoryFormat7',
2643
__make_delegated(_name, 'bzrlib.repofmt.weaverepo')
2647
'RepositoryFormatKnit',
2648
'RepositoryFormatKnit1',
2650
__make_delegated(_name, 'bzrlib.repofmt.knitrepo')
2653
def install_revision(repository, rev, revision_tree):
2654
"""Install all revision data into a repository."""
2655
install_revisions(repository, [(rev, revision_tree, None)])
2658
def install_revisions(repository, iterable, num_revisions=None, pb=None):
2659
"""Install all revision data into a repository.
2661
Accepts an iterable of revision, tree, signature tuples. The signature
2664
repository.start_write_group()
2666
inventory_cache = lru_cache.LRUCache(10)
2667
for n, (revision, revision_tree, signature) in enumerate(iterable):
2668
_install_revision(repository, revision, revision_tree, signature,
2671
pb.update('Transferring revisions', n + 1, num_revisions)
2673
repository.abort_write_group()
2676
repository.commit_write_group()
2679
def _install_revision(repository, rev, revision_tree, signature,
2681
"""Install all revision data into a repository."""
2682
present_parents = []
2684
for p_id in rev.parent_ids:
2685
if repository.has_revision(p_id):
2686
present_parents.append(p_id)
2687
parent_trees[p_id] = repository.revision_tree(p_id)
2689
parent_trees[p_id] = repository.revision_tree(
2690
_mod_revision.NULL_REVISION)
2692
inv = revision_tree.inventory
2693
entries = inv.iter_entries()
2694
# backwards compatibility hack: skip the root id.
2695
if not repository.supports_rich_root():
2696
path, root = entries.next()
2697
if root.revision != rev.revision_id:
2698
raise errors.IncompatibleRevision(repr(repository))
2700
for path, ie in entries:
2701
text_keys[(ie.file_id, ie.revision)] = ie
2702
text_parent_map = repository.texts.get_parent_map(text_keys)
2703
missing_texts = set(text_keys) - set(text_parent_map)
2704
# Add the texts that are not already present
2705
for text_key in missing_texts:
2706
ie = text_keys[text_key]
2708
# FIXME: TODO: The following loop overlaps/duplicates that done by
2709
# commit to determine parents. There is a latent/real bug here where
2710
# the parents inserted are not those commit would do - in particular
2711
# they are not filtered by heads(). RBC, AB
2712
for revision, tree in parent_trees.iteritems():
2713
if ie.file_id not in tree:
2715
parent_id = tree.inventory[ie.file_id].revision
2716
if parent_id in text_parents:
2718
text_parents.append((ie.file_id, parent_id))
2719
lines = revision_tree.get_file(ie.file_id).readlines()
2720
repository.texts.add_lines(text_key, text_parents, lines)
2722
# install the inventory
2723
if repository._format._commit_inv_deltas and len(rev.parent_ids):
2724
# Cache this inventory
2725
inventory_cache[rev.revision_id] = inv
2727
basis_inv = inventory_cache[rev.parent_ids[0]]
2729
repository.add_inventory(rev.revision_id, inv, present_parents)
2731
delta = inv._make_delta(basis_inv)
2732
repository.add_inventory_by_delta(rev.parent_ids[0], delta,
2733
rev.revision_id, present_parents)
2735
repository.add_inventory(rev.revision_id, inv, present_parents)
2736
except errors.RevisionAlreadyPresent:
2738
if signature is not None:
2739
repository.add_signature_text(rev.revision_id, signature)
2740
repository.add_revision(rev.revision_id, rev, inv)
1271
2743
class MetaDirRepository(Repository):
1272
2744
"""Repositories in the new meta-dir layout.
1613
3088
# NOTE: These are experimental in 0.92. Stable in 1.0 and above
1614
3089
format_registry.register_lazy(
1615
3090
'Bazaar pack repository format 1 (needs bzr 0.92)\n',
1616
'bzrlib.repofmt.knitpack_repo',
3091
'bzrlib.repofmt.pack_repo',
1617
3092
'RepositoryFormatKnitPack1',
1619
3094
format_registry.register_lazy(
1620
3095
'Bazaar pack repository format 1 with subtree support (needs bzr 0.92)\n',
1621
'bzrlib.repofmt.knitpack_repo',
3096
'bzrlib.repofmt.pack_repo',
1622
3097
'RepositoryFormatKnitPack3',
1624
3099
format_registry.register_lazy(
1625
3100
'Bazaar pack repository format 1 with rich root (needs bzr 1.0)\n',
1626
'bzrlib.repofmt.knitpack_repo',
3101
'bzrlib.repofmt.pack_repo',
1627
3102
'RepositoryFormatKnitPack4',
1629
3104
format_registry.register_lazy(
1630
3105
'Bazaar RepositoryFormatKnitPack5 (bzr 1.6)\n',
1631
'bzrlib.repofmt.knitpack_repo',
3106
'bzrlib.repofmt.pack_repo',
1632
3107
'RepositoryFormatKnitPack5',
1634
3109
format_registry.register_lazy(
1635
3110
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6.1)\n',
1636
'bzrlib.repofmt.knitpack_repo',
3111
'bzrlib.repofmt.pack_repo',
1637
3112
'RepositoryFormatKnitPack5RichRoot',
1639
3114
format_registry.register_lazy(
1640
3115
'Bazaar RepositoryFormatKnitPack5RichRoot (bzr 1.6)\n',
1641
'bzrlib.repofmt.knitpack_repo',
3116
'bzrlib.repofmt.pack_repo',
1642
3117
'RepositoryFormatKnitPack5RichRootBroken',
1644
3119
format_registry.register_lazy(
1645
3120
'Bazaar RepositoryFormatKnitPack6 (bzr 1.9)\n',
1646
'bzrlib.repofmt.knitpack_repo',
3121
'bzrlib.repofmt.pack_repo',
1647
3122
'RepositoryFormatKnitPack6',
1649
3124
format_registry.register_lazy(
1650
3125
'Bazaar RepositoryFormatKnitPack6RichRoot (bzr 1.9)\n',
1651
'bzrlib.repofmt.knitpack_repo',
3126
'bzrlib.repofmt.pack_repo',
1652
3127
'RepositoryFormatKnitPack6RichRoot',
1654
format_registry.register_lazy(
1655
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
1656
'bzrlib.repofmt.groupcompress_repo',
1657
'RepositoryFormat2a',
1660
3130
# Development formats.
1661
# Check their docstrings to see if/when they are obsolete.
3131
# Obsolete but kept pending a CHK based subtree format.
1662
3132
format_registry.register_lazy(
1663
3133
("Bazaar development format 2 with subtree support "
1664
3134
"(needs bzr.dev from before 1.8)\n"),
1665
'bzrlib.repofmt.knitpack_repo',
3135
'bzrlib.repofmt.pack_repo',
1666
3136
'RepositoryFormatPackDevelopment2Subtree',
1668
format_registry.register_lazy(
1669
'Bazaar development format 8\n',
1670
'bzrlib.repofmt.groupcompress_repo',
1671
'RepositoryFormat2aSubtree',
3139
# 1.14->1.16 go below here
3140
format_registry.register_lazy(
3141
'Bazaar development format - group compression and chk inventory'
3142
' (needs bzr.dev from 1.14)\n',
3143
'bzrlib.repofmt.groupcompress_repo',
3144
'RepositoryFormatCHK1',
3147
format_registry.register_lazy(
3148
'Bazaar development format - chk repository with bencode revision '
3149
'serialization (needs bzr.dev from 1.16)\n',
3150
'bzrlib.repofmt.groupcompress_repo',
3151
'RepositoryFormatCHK2',
3153
format_registry.register_lazy(
3154
'Bazaar repository format 2a (needs bzr 1.16 or later)\n',
3155
'bzrlib.repofmt.groupcompress_repo',
3156
'RepositoryFormat2a',
1713
3199
:param revision_id: if None all content is copied, if NULL_REVISION no
1714
3200
content is copied.
3201
:param pb: optional progress bar to use for progress reports. If not
3202
provided a default one will be created.
1717
raise NotImplementedError(self.fetch)
3205
from bzrlib.fetch import RepoFetcher
3206
f = RepoFetcher(to_repository=self.target,
3207
from_repository=self.source,
3208
last_revision=revision_id,
3209
fetch_spec=fetch_spec,
3210
pb=pb, find_ghosts=find_ghosts)
3212
def _walk_to_common_revisions(self, revision_ids):
3213
"""Walk out from revision_ids in source to revisions target has.
3215
:param revision_ids: The start point for the search.
3216
:return: A set of revision ids.
3218
target_graph = self.target.get_graph()
3219
revision_ids = frozenset(revision_ids)
3220
missing_revs = set()
3221
source_graph = self.source.get_graph()
3222
# ensure we don't pay silly lookup costs.
3223
searcher = source_graph._make_breadth_first_searcher(revision_ids)
3224
null_set = frozenset([_mod_revision.NULL_REVISION])
3225
searcher_exhausted = False
3229
# Iterate the searcher until we have enough next_revs
3230
while len(next_revs) < self._walk_to_common_revisions_batch_size:
3232
next_revs_part, ghosts_part = searcher.next_with_ghosts()
3233
next_revs.update(next_revs_part)
3234
ghosts.update(ghosts_part)
3235
except StopIteration:
3236
searcher_exhausted = True
3238
# If there are ghosts in the source graph, and the caller asked for
3239
# them, make sure that they are present in the target.
3240
# We don't care about other ghosts as we can't fetch them and
3241
# haven't been asked to.
3242
ghosts_to_check = set(revision_ids.intersection(ghosts))
3243
revs_to_get = set(next_revs).union(ghosts_to_check)
3245
have_revs = set(target_graph.get_parent_map(revs_to_get))
3246
# we always have NULL_REVISION present.
3247
have_revs = have_revs.union(null_set)
3248
# Check if the target is missing any ghosts we need.
3249
ghosts_to_check.difference_update(have_revs)
3251
# One of the caller's revision_ids is a ghost in both the
3252
# source and the target.
3253
raise errors.NoSuchRevision(
3254
self.source, ghosts_to_check.pop())
3255
missing_revs.update(next_revs - have_revs)
3256
# Because we may have walked past the original stop point, make
3257
# sure everything is stopped
3258
stop_revs = searcher.find_seen_ancestors(have_revs)
3259
searcher.stop_searching_any(stop_revs)
3260
if searcher_exhausted:
3262
return searcher.get_result()
1719
3264
@needs_read_lock
1720
def search_missing_revision_ids(self,
1721
revision_id=symbol_versioning.DEPRECATED_PARAMETER,
1722
find_ghosts=True, revision_ids=None, if_present_ids=None,
3265
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
1724
3266
"""Return the revision ids that source has that target does not.
1726
3268
:param revision_id: only return revision ids included by this
1728
:param revision_ids: return revision ids included by these
1729
revision_ids. NoSuchRevision will be raised if any of these
1730
revisions are not present.
1731
:param if_present_ids: like revision_ids, but will not cause
1732
NoSuchRevision if any of these are absent, instead they will simply
1733
not be in the result. This is useful for e.g. finding revisions
1734
to fetch for tags, which may reference absent revisions.
1735
3270
:param find_ghosts: If True find missing revisions in deep history
1736
3271
rather than just finding the surface difference.
1737
:param limit: Maximum number of revisions to return, topologically
1739
3272
:return: A bzrlib.graph.SearchResult.
1741
raise NotImplementedError(self.search_missing_revision_ids)
3274
# stop searching at found target revisions.
3275
if not find_ghosts and revision_id is not None:
3276
return self._walk_to_common_revisions([revision_id])
3277
# generic, possibly worst case, slow code path.
3278
target_ids = set(self.target.all_revision_ids())
3279
if revision_id is not None:
3280
source_ids = self.source.get_ancestry(revision_id)
3281
if source_ids[0] is not None:
3282
raise AssertionError()
3285
source_ids = self.source.all_revision_ids()
3286
result_set = set(source_ids).difference(target_ids)
3287
return self.source.revision_ids_to_search_result(result_set)
1744
3290
def _same_model(source, target):
1765
3311
"different serializers")
3314
class InterSameDataRepository(InterRepository):
3315
"""Code for converting between repositories that represent the same data.
3317
Data format and model must match for this to work.
3321
def _get_repo_format_to_test(self):
3322
"""Repository format for testing with.
3324
InterSameData can pull from subtree to subtree and from non-subtree to
3325
non-subtree, so we test this with the richest repository format.
3327
from bzrlib.repofmt import knitrepo
3328
return knitrepo.RepositoryFormatKnit3()
3331
def is_compatible(source, target):
3332
return InterRepository._same_model(source, target)
3335
class InterWeaveRepo(InterSameDataRepository):
3336
"""Optimised code paths between Weave based repositories.
3338
This should be in bzrlib/repofmt/weaverepo.py but we have not yet
3339
implemented lazy inter-object optimisation.
3343
def _get_repo_format_to_test(self):
3344
from bzrlib.repofmt import weaverepo
3345
return weaverepo.RepositoryFormat7()
3348
def is_compatible(source, target):
3349
"""Be compatible with known Weave formats.
3351
We don't test for the stores being of specific types because that
3352
could lead to confusing results, and there is no need to be
3355
from bzrlib.repofmt.weaverepo import (
3361
return (isinstance(source._format, (RepositoryFormat5,
3363
RepositoryFormat7)) and
3364
isinstance(target._format, (RepositoryFormat5,
3366
RepositoryFormat7)))
3367
except AttributeError:
3371
def copy_content(self, revision_id=None):
3372
"""See InterRepository.copy_content()."""
3373
# weave specific optimised path:
3375
self.target.set_make_working_trees(self.source.make_working_trees())
3376
except (errors.RepositoryUpgradeRequired, NotImplemented):
3378
# FIXME do not peek!
3379
if self.source._transport.listable():
3380
pb = ui.ui_factory.nested_progress_bar()
3382
self.target.texts.insert_record_stream(
3383
self.source.texts.get_record_stream(
3384
self.source.texts.keys(), 'topological', False))
3385
pb.update('copying inventory', 0, 1)
3386
self.target.inventories.insert_record_stream(
3387
self.source.inventories.get_record_stream(
3388
self.source.inventories.keys(), 'topological', False))
3389
self.target.signatures.insert_record_stream(
3390
self.source.signatures.get_record_stream(
3391
self.source.signatures.keys(),
3393
self.target.revisions.insert_record_stream(
3394
self.source.revisions.get_record_stream(
3395
self.source.revisions.keys(),
3396
'topological', True))
3400
self.target.fetch(self.source, revision_id=revision_id)
3403
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3404
"""See InterRepository.missing_revision_ids()."""
3405
# we want all revisions to satisfy revision_id in source.
3406
# but we don't want to stat every file here and there.
3407
# we want then, all revisions other needs to satisfy revision_id
3408
# checked, but not those that we have locally.
3409
# so the first thing is to get a subset of the revisions to
3410
# satisfy revision_id in source, and then eliminate those that
3411
# we do already have.
3412
# this is slow on high latency connection to self, but as this
3413
# disk format scales terribly for push anyway due to rewriting
3414
# inventory.weave, this is considered acceptable.
3416
if revision_id is not None:
3417
source_ids = self.source.get_ancestry(revision_id)
3418
if source_ids[0] is not None:
3419
raise AssertionError()
3422
source_ids = self.source._all_possible_ids()
3423
source_ids_set = set(source_ids)
3424
# source_ids is the worst possible case we may need to pull.
3425
# now we want to filter source_ids against what we actually
3426
# have in target, but don't try to check for existence where we know
3427
# we do not have a revision as that would be pointless.
3428
target_ids = set(self.target._all_possible_ids())
3429
possibly_present_revisions = target_ids.intersection(source_ids_set)
3430
actually_present_revisions = set(
3431
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3432
required_revisions = source_ids_set.difference(actually_present_revisions)
3433
if revision_id is not None:
3434
# we used get_ancestry to determine source_ids then we are assured all
3435
# revisions referenced are present as they are installed in topological order.
3436
# and the tip revision was validated by get_ancestry.
3437
result_set = required_revisions
3439
# if we just grabbed the possibly available ids, then
3440
# we only have an estimate of whats available and need to validate
3441
# that against the revision records.
3443
self.source._eliminate_revisions_not_present(required_revisions))
3444
return self.source.revision_ids_to_search_result(result_set)
3447
class InterKnitRepo(InterSameDataRepository):
3448
"""Optimised code paths between Knit based repositories."""
3451
def _get_repo_format_to_test(self):
3452
from bzrlib.repofmt import knitrepo
3453
return knitrepo.RepositoryFormatKnit1()
3456
def is_compatible(source, target):
3457
"""Be compatible with known Knit formats.
3459
We don't test for the stores being of specific types because that
3460
could lead to confusing results, and there is no need to be
3463
from bzrlib.repofmt.knitrepo import RepositoryFormatKnit
3465
are_knits = (isinstance(source._format, RepositoryFormatKnit) and
3466
isinstance(target._format, RepositoryFormatKnit))
3467
except AttributeError:
3469
return are_knits and InterRepository._same_model(source, target)
3472
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3473
"""See InterRepository.missing_revision_ids()."""
3474
if revision_id is not None:
3475
source_ids = self.source.get_ancestry(revision_id)
3476
if source_ids[0] is not None:
3477
raise AssertionError()
3480
source_ids = self.source.all_revision_ids()
3481
source_ids_set = set(source_ids)
3482
# source_ids is the worst possible case we may need to pull.
3483
# now we want to filter source_ids against what we actually
3484
# have in target, but don't try to check for existence where we know
3485
# we do not have a revision as that would be pointless.
3486
target_ids = set(self.target.all_revision_ids())
3487
possibly_present_revisions = target_ids.intersection(source_ids_set)
3488
actually_present_revisions = set(
3489
self.target._eliminate_revisions_not_present(possibly_present_revisions))
3490
required_revisions = source_ids_set.difference(actually_present_revisions)
3491
if revision_id is not None:
3492
# we used get_ancestry to determine source_ids then we are assured all
3493
# revisions referenced are present as they are installed in topological order.
3494
# and the tip revision was validated by get_ancestry.
3495
result_set = required_revisions
3497
# if we just grabbed the possibly available ids, then
3498
# we only have an estimate of whats available and need to validate
3499
# that against the revision records.
3501
self.source._eliminate_revisions_not_present(required_revisions))
3502
return self.source.revision_ids_to_search_result(result_set)
3505
class InterDifferingSerializer(InterRepository):
3508
def _get_repo_format_to_test(self):
3512
def is_compatible(source, target):
3513
"""Be compatible with Knit2 source and Knit3 target"""
3514
# This is redundant with format.check_conversion_target(), however that
3515
# raises an exception, and we just want to say "False" as in we won't
3516
# support converting between these formats.
3517
if source.supports_rich_root() and not target.supports_rich_root():
3519
if (source._format.supports_tree_reference
3520
and not target._format.supports_tree_reference):
3524
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3525
"""Get the best delta and base for this revision.
3527
:return: (basis_id, delta)
3529
possible_trees = [(parent_id, cache[parent_id])
3530
for parent_id in parent_ids
3531
if parent_id in cache]
3532
if len(possible_trees) == 0:
3533
# There either aren't any parents, or the parents aren't in the
3534
# cache, so just use the last converted tree
3535
possible_trees.append((basis_id, cache[basis_id]))
3537
for basis_id, basis_tree in possible_trees:
3538
delta = tree.inventory._make_delta(basis_tree.inventory)
3539
deltas.append((len(delta), basis_id, delta))
3541
return deltas[0][1:]
3543
def _get_parent_keys(self, root_key, parent_map):
3544
"""Get the parent keys for a given root id."""
3545
root_id, rev_id = root_key
3546
# Include direct parents of the revision, but only if they used
3547
# the same root_id and are heads.
3549
for parent_id in parent_map[rev_id]:
3550
if parent_id == _mod_revision.NULL_REVISION:
3552
if parent_id not in self._revision_id_to_root_id:
3553
# We probably didn't read this revision, go spend the
3554
# extra effort to actually check
3556
tree = self.source.revision_tree(parent_id)
3557
except errors.NoSuchRevision:
3558
# Ghost, fill out _revision_id_to_root_id in case we
3559
# encounter this again.
3560
# But set parent_root_id to None since we don't really know
3561
parent_root_id = None
3563
parent_root_id = tree.get_root_id()
3564
self._revision_id_to_root_id[parent_id] = None
3566
parent_root_id = self._revision_id_to_root_id[parent_id]
3567
if root_id == parent_root_id:
3568
# With stacking we _might_ want to refer to a non-local
3569
# revision, but this code path only applies when we have the
3570
# full content available, so ghosts really are ghosts, not just
3571
# the edge of local data.
3572
parent_keys.append((parent_id,))
3574
# root_id may be in the parent anyway.
3576
tree = self.source.revision_tree(parent_id)
3577
except errors.NoSuchRevision:
3578
# ghost, can't refer to it.
3582
parent_keys.append((tree.inventory[root_id].revision,))
3583
except errors.NoSuchId:
3586
g = graph.Graph(self.source.revisions)
3587
heads = g.heads(parent_keys)
3589
for key in parent_keys:
3590
if key in heads and key not in selected_keys:
3591
selected_keys.append(key)
3592
return tuple([(root_id,)+ key for key in selected_keys])
3594
def _new_root_data_stream(self, root_keys_to_create, parent_map):
3595
for root_key in root_keys_to_create:
3596
parent_keys = self._get_parent_keys(root_key, parent_map)
3597
yield versionedfile.FulltextContentFactory(root_key,
3598
parent_keys, None, '')
3600
def _fetch_batch(self, revision_ids, basis_id, cache):
3601
"""Fetch across a few revisions.
3603
:param revision_ids: The revisions to copy
3604
:param basis_id: The revision_id of a tree that must be in cache, used
3605
as a basis for delta when no other base is available
3606
:param cache: A cache of RevisionTrees that we can use.
3607
:return: The revision_id of the last converted tree. The RevisionTree
3608
for it will be in cache
3610
# Walk though all revisions; get inventory deltas, copy referenced
3611
# texts that delta references, insert the delta, revision and
3613
root_keys_to_create = set()
3616
pending_revisions = []
3617
parent_map = self.source.get_parent_map(revision_ids)
3618
for tree in self.source.revision_trees(revision_ids):
3619
current_revision_id = tree.get_revision_id()
3620
parent_ids = parent_map.get(current_revision_id, ())
3621
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3623
if self._converting_to_rich_root:
3624
self._revision_id_to_root_id[current_revision_id] = \
3626
# Find text entries that need to be copied
3627
for old_path, new_path, file_id, entry in delta:
3628
if new_path is not None:
3631
if not self.target.supports_rich_root():
3632
# The target doesn't support rich root, so we don't
3635
if self._converting_to_rich_root:
3636
# This can't be copied normally, we have to insert
3638
root_keys_to_create.add((file_id, entry.revision))
3640
text_keys.add((file_id, entry.revision))
3641
revision = self.source.get_revision(current_revision_id)
3642
pending_deltas.append((basis_id, delta,
3643
current_revision_id, revision.parent_ids))
3644
pending_revisions.append(revision)
3645
cache[current_revision_id] = tree
3646
basis_id = current_revision_id
3648
from_texts = self.source.texts
3649
to_texts = self.target.texts
3650
if root_keys_to_create:
3651
root_stream = self._new_root_data_stream(root_keys_to_create,
3653
to_texts.insert_record_stream(root_stream)
3654
to_texts.insert_record_stream(from_texts.get_record_stream(
3655
text_keys, self.target._format._fetch_order,
3656
not self.target._format._fetch_uses_deltas))
3657
# insert inventory deltas
3658
for delta in pending_deltas:
3659
self.target.add_inventory_by_delta(*delta)
3660
if self.target._fallback_repositories:
3661
# Make sure this stacked repository has all the parent inventories
3662
# for the new revisions that we are about to insert. We do this
3663
# before adding the revisions so that no revision is added until
3664
# all the inventories it may depend on are added.
3666
revision_ids = set()
3667
for revision in pending_revisions:
3668
revision_ids.add(revision.revision_id)
3669
parent_ids.update(revision.parent_ids)
3670
parent_ids.difference_update(revision_ids)
3671
parent_ids.discard(_mod_revision.NULL_REVISION)
3672
parent_map = self.source.get_parent_map(parent_ids)
3673
for parent_tree in self.source.revision_trees(parent_ids):
3674
basis_id, delta = self._get_delta_for_revision(tree, parent_ids, basis_id, cache)
3675
current_revision_id = parent_tree.get_revision_id()
3676
parents_parents = parent_map[current_revision_id]
3677
self.target.add_inventory_by_delta(
3678
basis_id, delta, current_revision_id, parents_parents)
3679
# insert signatures and revisions
3680
for revision in pending_revisions:
3682
signature = self.source.get_signature_text(
3683
revision.revision_id)
3684
self.target.add_signature_text(revision.revision_id,
3686
except errors.NoSuchRevision:
3688
self.target.add_revision(revision.revision_id, revision)
3691
def _fetch_all_revisions(self, revision_ids, pb):
3692
"""Fetch everything for the list of revisions.
3694
:param revision_ids: The list of revisions to fetch. Must be in
3696
:param pb: A ProgressBar
3699
basis_id, basis_tree = self._get_basis(revision_ids[0])
3701
cache = lru_cache.LRUCache(100)
3702
cache[basis_id] = basis_tree
3703
del basis_tree # We don't want to hang on to it here
3705
for offset in range(0, len(revision_ids), batch_size):
3706
self.target.start_write_group()
3708
pb.update('Transferring revisions', offset,
3710
batch = revision_ids[offset:offset+batch_size]
3711
basis_id = self._fetch_batch(batch, basis_id, cache)
3713
self.target.abort_write_group()
3716
hint = self.target.commit_write_group()
3719
if hints and self.target._format.pack_compresses:
3720
self.target.pack(hint=hints)
3721
pb.update('Transferring revisions', len(revision_ids),
3725
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3727
"""See InterRepository.fetch()."""
3728
if fetch_spec is not None:
3729
raise AssertionError("Not implemented yet...")
3730
if (not self.source.supports_rich_root()
3731
and self.target.supports_rich_root()):
3732
self._converting_to_rich_root = True
3733
self._revision_id_to_root_id = {}
3735
self._converting_to_rich_root = False
3736
revision_ids = self.target.search_missing_revision_ids(self.source,
3737
revision_id, find_ghosts=find_ghosts).get_keys()
3738
if not revision_ids:
3740
revision_ids = tsort.topo_sort(
3741
self.source.get_graph().get_parent_map(revision_ids))
3742
if not revision_ids:
3744
# Walk though all revisions; get inventory deltas, copy referenced
3745
# texts that delta references, insert the delta, revision and
3747
first_rev = self.source.get_revision(revision_ids[0])
3749
my_pb = ui.ui_factory.nested_progress_bar()
3752
symbol_versioning.warn(
3753
symbol_versioning.deprecated_in((1, 14, 0))
3754
% "pb parameter to fetch()")
3757
self._fetch_all_revisions(revision_ids, pb)
3759
if my_pb is not None:
3761
return len(revision_ids), 0
3763
def _get_basis(self, first_revision_id):
3764
"""Get a revision and tree which exists in the target.
3766
This assumes that first_revision_id is selected for transmission
3767
because all other ancestors are already present. If we can't find an
3768
ancestor we fall back to NULL_REVISION since we know that is safe.
3770
:return: (basis_id, basis_tree)
3772
first_rev = self.source.get_revision(first_revision_id)
3774
basis_id = first_rev.parent_ids[0]
3775
# only valid as a basis if the target has it
3776
self.target.get_revision(basis_id)
3777
# Try to get a basis tree - if its a ghost it will hit the
3778
# NoSuchRevision case.
3779
basis_tree = self.source.revision_tree(basis_id)
3780
except (IndexError, errors.NoSuchRevision):
3781
basis_id = _mod_revision.NULL_REVISION
3782
basis_tree = self.source.revision_tree(basis_id)
3783
return basis_id, basis_tree
3786
InterRepository.register_optimiser(InterDifferingSerializer)
3787
InterRepository.register_optimiser(InterSameDataRepository)
3788
InterRepository.register_optimiser(InterWeaveRepo)
3789
InterRepository.register_optimiser(InterKnitRepo)
1768
3792
class CopyConverter(object):
1769
3793
"""A repository conversion tool which just performs a copy of the content.
1784
3808
:param to_convert: The disk object to convert.
1785
3809
:param pb: a progress bar to use for progress information.
1787
pb = ui.ui_factory.nested_progress_bar()
1790
3814
# this is only useful with metadir layouts - separated repo content.
1791
3815
# trigger an assertion if not such
1792
3816
repo._format.get_format_string()
1793
3817
self.repo_dir = repo.bzrdir
1794
pb.update(gettext('Moving repository to repository.backup'))
3818
self.step('Moving repository to repository.backup')
1795
3819
self.repo_dir.transport.move('repository', 'repository.backup')
1796
3820
backup_transport = self.repo_dir.transport.clone('repository.backup')
1797
3821
repo._format.check_conversion_target(self.target_format)
1798
3822
self.source_repo = repo._format.open(self.repo_dir,
1800
3824
_override_transport=backup_transport)
1801
pb.update(gettext('Creating new repository'))
3825
self.step('Creating new repository')
1802
3826
converted = self.target_format.initialize(self.repo_dir,
1803
3827
self.source_repo.is_shared())
1804
3828
converted.lock_write()
1806
pb.update(gettext('Copying content'))
3830
self.step('Copying content into repository.')
1807
3831
self.source_repo.copy_content_into(converted)
1809
3833
converted.unlock()
1810
pb.update(gettext('Deleting old repository content'))
3834
self.step('Deleting old repository content.')
1811
3835
self.repo_dir.transport.delete_tree('repository.backup')
1812
ui.ui_factory.note(gettext('repository converted'))
3836
self.pb.note('repository converted')
3838
def step(self, message):
3839
"""Update the pb by a step."""
3841
self.pb.update(message, self.count, self.total)
3853
def _unescaper(match, _map=_unescape_map):
3854
code = match.group(1)
3858
if not code.startswith('#'):
3860
return unichr(int(code[1:])).encode('utf8')
3866
def _unescape_xml(data):
3867
"""Unescape predefined XML entities in a string of data."""
3869
if _unescape_re is None:
3870
_unescape_re = re.compile('\&([^;]*);')
3871
return _unescape_re.sub(_unescaper, data)
3874
class _VersionedFileChecker(object):
3876
def __init__(self, repository, text_key_references=None):
3877
self.repository = repository
3878
self.text_index = self.repository._generate_text_key_index(
3879
text_key_references=text_key_references)
3881
def calculate_file_version_parents(self, text_key):
3882
"""Calculate the correct parents for a file version according to
3885
parent_keys = self.text_index[text_key]
3886
if parent_keys == [_mod_revision.NULL_REVISION]:
3888
return tuple(parent_keys)
3890
def check_file_version_parents(self, texts, progress_bar=None):
3891
"""Check the parents stored in a versioned file are correct.
3893
It also detects file versions that are not referenced by their
3894
corresponding revision's inventory.
3896
:returns: A tuple of (wrong_parents, dangling_file_versions).
3897
wrong_parents is a dict mapping {revision_id: (stored_parents,
3898
correct_parents)} for each revision_id where the stored parents
3899
are not correct. dangling_file_versions is a set of (file_id,
3900
revision_id) tuples for versions that are present in this versioned
3901
file, but not used by the corresponding inventory.
3904
self.file_ids = set([file_id for file_id, _ in
3905
self.text_index.iterkeys()])
3906
# text keys is now grouped by file_id
3907
n_weaves = len(self.file_ids)
3908
files_in_revisions = {}
3909
revisions_of_files = {}
3910
n_versions = len(self.text_index)
3911
progress_bar.update('loading text store', 0, n_versions)
3912
parent_map = self.repository.texts.get_parent_map(self.text_index)
3913
# On unlistable transports this could well be empty/error...
3914
text_keys = self.repository.texts.keys()
3915
unused_keys = frozenset(text_keys) - set(self.text_index)
3916
for num, key in enumerate(self.text_index.iterkeys()):
3917
if progress_bar is not None:
3918
progress_bar.update('checking text graph', num, n_versions)
3919
correct_parents = self.calculate_file_version_parents(key)
3921
knit_parents = parent_map[key]
3922
except errors.RevisionNotPresent:
3925
if correct_parents != knit_parents:
3926
wrong_parents[key] = (knit_parents, correct_parents)
3927
return wrong_parents, unused_keys
3930
def _old_get_graph(repository, revision_id):
3931
"""DO NOT USE. That is all. I'm serious."""
3932
graph = repository.get_graph()
3933
revision_graph = dict(((key, value) for key, value in
3934
graph.iter_ancestry([revision_id]) if value is not None))
3935
return _strip_NULL_ghosts(revision_graph)
1816
3938
def _strip_NULL_ghosts(revision_graph):
1824
3946
return revision_graph
3949
class StreamSink(object):
3950
"""An object that can insert a stream into a repository.
3952
This interface handles the complexity of reserialising inventories and
3953
revisions from different formats, and allows unidirectional insertion into
3954
stacked repositories without looking for the missing basis parents
3958
def __init__(self, target_repo):
3959
self.target_repo = target_repo
3961
def insert_stream(self, stream, src_format, resume_tokens):
3962
"""Insert a stream's content into the target repository.
3964
:param src_format: a bzr repository format.
3966
:return: a list of resume tokens and an iterable of keys additional
3967
items required before the insertion can be completed.
3969
self.target_repo.lock_write()
3972
self.target_repo.resume_write_group(resume_tokens)
3975
self.target_repo.start_write_group()
3978
# locked_insert_stream performs a commit|suspend.
3979
return self._locked_insert_stream(stream, src_format, is_resume)
3981
self.target_repo.abort_write_group(suppress_errors=True)
3984
self.target_repo.unlock()
3986
def _locked_insert_stream(self, stream, src_format, is_resume):
3987
to_serializer = self.target_repo._format._serializer
3988
src_serializer = src_format._serializer
3990
if to_serializer == src_serializer:
3991
# If serializers match and the target is a pack repository, set the
3992
# write cache size on the new pack. This avoids poor performance
3993
# on transports where append is unbuffered (such as
3994
# RemoteTransport). This is safe to do because nothing should read
3995
# back from the target repository while a stream with matching
3996
# serialization is being inserted.
3997
# The exception is that a delta record from the source that should
3998
# be a fulltext may need to be expanded by the target (see
3999
# test_fetch_revisions_with_deltas_into_pack); but we take care to
4000
# explicitly flush any buffered writes first in that rare case.
4002
new_pack = self.target_repo._pack_collection._new_pack
4003
except AttributeError:
4004
# Not a pack repository
4007
new_pack.set_write_cache_size(1024*1024)
4008
for substream_type, substream in stream:
4009
if substream_type == 'texts':
4010
self.target_repo.texts.insert_record_stream(substream)
4011
elif substream_type == 'inventories':
4012
if src_serializer == to_serializer:
4013
self.target_repo.inventories.insert_record_stream(
4016
self._extract_and_insert_inventories(
4017
substream, src_serializer)
4018
elif substream_type == 'chk_bytes':
4019
# XXX: This doesn't support conversions, as it assumes the
4020
# conversion was done in the fetch code.
4021
self.target_repo.chk_bytes.insert_record_stream(substream)
4022
elif substream_type == 'revisions':
4023
# This may fallback to extract-and-insert more often than
4024
# required if the serializers are different only in terms of
4026
if src_serializer == to_serializer:
4027
self.target_repo.revisions.insert_record_stream(
4030
self._extract_and_insert_revisions(substream,
4032
elif substream_type == 'signatures':
4033
self.target_repo.signatures.insert_record_stream(substream)
4035
raise AssertionError('kaboom! %s' % (substream_type,))
4036
# Done inserting data, and the missing_keys calculations will try to
4037
# read back from the inserted data, so flush the writes to the new pack
4038
# (if this is pack format).
4039
if new_pack is not None:
4040
new_pack._write_data('', flush=True)
4041
# Find all the new revisions (including ones from resume_tokens)
4042
missing_keys = self.target_repo.get_missing_parent_inventories(
4043
check_for_missing_texts=is_resume)
4045
for prefix, versioned_file in (
4046
('texts', self.target_repo.texts),
4047
('inventories', self.target_repo.inventories),
4048
('revisions', self.target_repo.revisions),
4049
('signatures', self.target_repo.signatures),
4050
('chk_bytes', self.target_repo.chk_bytes),
4052
if versioned_file is None:
4054
missing_keys.update((prefix,) + key for key in
4055
versioned_file.get_missing_compression_parent_keys())
4056
except NotImplementedError:
4057
# cannot even attempt suspending, and missing would have failed
4058
# during stream insertion.
4059
missing_keys = set()
4062
# suspend the write group and tell the caller what we is
4063
# missing. We know we can suspend or else we would not have
4064
# entered this code path. (All repositories that can handle
4065
# missing keys can handle suspending a write group).
4066
write_group_tokens = self.target_repo.suspend_write_group()
4067
return write_group_tokens, missing_keys
4068
hint = self.target_repo.commit_write_group()
4069
if (to_serializer != src_serializer and
4070
self.target_repo._format.pack_compresses):
4071
self.target_repo.pack(hint=hint)
4074
def _extract_and_insert_inventories(self, substream, serializer):
4075
"""Generate a new inventory versionedfile in target, converting data.
4077
The inventory is retrieved from the source, (deserializing it), and
4078
stored in the target (reserializing it in a different format).
4080
for record in substream:
4081
bytes = record.get_bytes_as('fulltext')
4082
revision_id = record.key[0]
4083
inv = serializer.read_inventory_from_string(bytes, revision_id)
4084
parents = [key[0] for key in record.parents]
4085
self.target_repo.add_inventory(revision_id, inv, parents)
4087
def _extract_and_insert_revisions(self, substream, serializer):
4088
for record in substream:
4089
bytes = record.get_bytes_as('fulltext')
4090
revision_id = record.key[0]
4091
rev = serializer.read_revision_from_string(bytes)
4092
if rev.revision_id != revision_id:
4093
raise AssertionError('wtf: %s != %s' % (rev, revision_id))
4094
self.target_repo.add_revision(revision_id, rev)
4097
if self.target_repo._format._fetch_reconcile:
4098
self.target_repo.reconcile()
4101
class StreamSource(object):
4102
"""A source of a stream for fetching between repositories."""
4104
def __init__(self, from_repository, to_format):
4105
"""Create a StreamSource streaming from from_repository."""
4106
self.from_repository = from_repository
4107
self.to_format = to_format
4109
def delta_on_metadata(self):
4110
"""Return True if delta's are permitted on metadata streams.
4112
That is on revisions and signatures.
4114
src_serializer = self.from_repository._format._serializer
4115
target_serializer = self.to_format._serializer
4116
return (self.to_format._fetch_uses_deltas and
4117
src_serializer == target_serializer)
4119
def _fetch_revision_texts(self, revs):
4120
# fetch signatures first and then the revision texts
4121
# may need to be a InterRevisionStore call here.
4122
from_sf = self.from_repository.signatures
4123
# A missing signature is just skipped.
4124
keys = [(rev_id,) for rev_id in revs]
4125
signatures = versionedfile.filter_absent(from_sf.get_record_stream(
4127
self.to_format._fetch_order,
4128
not self.to_format._fetch_uses_deltas))
4129
# If a revision has a delta, this is actually expanded inside the
4130
# insert_record_stream code now, which is an alternate fix for
4132
from_rf = self.from_repository.revisions
4133
revisions = from_rf.get_record_stream(
4135
self.to_format._fetch_order,
4136
not self.delta_on_metadata())
4137
return [('signatures', signatures), ('revisions', revisions)]
4139
def _generate_root_texts(self, revs):
4140
"""This will be called by __fetch between fetching weave texts and
4141
fetching the inventory weave.
4143
Subclasses should override this if they need to generate root texts
4144
after fetching weave texts.
4146
if self._rich_root_upgrade():
4148
return bzrlib.fetch.Inter1and2Helper(
4149
self.from_repository).generate_root_texts(revs)
4153
def get_stream(self, search):
4155
revs = search.get_keys()
4156
graph = self.from_repository.get_graph()
4157
revs = list(graph.iter_topo_order(revs))
4158
data_to_fetch = self.from_repository.item_keys_introduced_by(revs)
4160
for knit_kind, file_id, revisions in data_to_fetch:
4161
if knit_kind != phase:
4163
# Make a new progress bar for this phase
4164
if knit_kind == "file":
4165
# Accumulate file texts
4166
text_keys.extend([(file_id, revision) for revision in
4168
elif knit_kind == "inventory":
4169
# Now copy the file texts.
4170
from_texts = self.from_repository.texts
4171
yield ('texts', from_texts.get_record_stream(
4172
text_keys, self.to_format._fetch_order,
4173
not self.to_format._fetch_uses_deltas))
4174
# Cause an error if a text occurs after we have done the
4177
# Before we process the inventory we generate the root
4178
# texts (if necessary) so that the inventories references
4180
for _ in self._generate_root_texts(revs):
4182
# NB: This currently reopens the inventory weave in source;
4183
# using a single stream interface instead would avoid this.
4184
from_weave = self.from_repository.inventories
4185
# we fetch only the referenced inventories because we do not
4186
# know for unselected inventories whether all their required
4187
# texts are present in the other repository - it could be
4189
for info in self._get_inventory_stream(revs):
4191
elif knit_kind == "signatures":
4192
# Nothing to do here; this will be taken care of when
4193
# _fetch_revision_texts happens.
4195
elif knit_kind == "revisions":
4196
for record in self._fetch_revision_texts(revs):
4199
raise AssertionError("Unknown knit kind %r" % knit_kind)
4201
def get_stream_for_missing_keys(self, missing_keys):
4202
# missing keys can only occur when we are byte copying and not
4203
# translating (because translation means we don't send
4204
# unreconstructable deltas ever).
4206
keys['texts'] = set()
4207
keys['revisions'] = set()
4208
keys['inventories'] = set()
4209
keys['chk_bytes'] = set()
4210
keys['signatures'] = set()
4211
for key in missing_keys:
4212
keys[key[0]].add(key[1:])
4213
if len(keys['revisions']):
4214
# If we allowed copying revisions at this point, we could end up
4215
# copying a revision without copying its required texts: a
4216
# violation of the requirements for repository integrity.
4217
raise AssertionError(
4218
'cannot copy revisions to fill in missing deltas %s' % (
4219
keys['revisions'],))
4220
for substream_kind, keys in keys.iteritems():
4221
vf = getattr(self.from_repository, substream_kind)
4222
if vf is None and keys:
4223
raise AssertionError(
4224
"cannot fill in keys for a versioned file we don't"
4225
" have: %s needs %s" % (substream_kind, keys))
4227
# No need to stream something we don't have
4229
# Ask for full texts always so that we don't need more round trips
4230
# after this stream.
4231
# Some of the missing keys are genuinely ghosts, so filter absent
4232
# records. The Sink is responsible for doing another check to
4233
# ensure that ghosts don't introduce missing data for future
4235
stream = versionedfile.filter_absent(vf.get_record_stream(keys,
4236
self.to_format._fetch_order, True))
4237
yield substream_kind, stream
4239
def inventory_fetch_order(self):
4240
if self._rich_root_upgrade():
4241
return 'topological'
4243
return self.to_format._fetch_order
4245
def _rich_root_upgrade(self):
4246
return (not self.from_repository._format.rich_root_data and
4247
self.to_format.rich_root_data)
4249
def _get_inventory_stream(self, revision_ids):
4250
from_format = self.from_repository._format
4251
if (from_format.supports_chks and self.to_format.supports_chks
4252
and (from_format._serializer == self.to_format._serializer)):
4253
# Both sides support chks, and they use the same serializer, so it
4254
# is safe to transmit the chk pages and inventory pages across
4256
return self._get_chk_inventory_stream(revision_ids)
4257
elif (not from_format.supports_chks):
4258
# Source repository doesn't support chks. So we can transmit the
4259
# inventories 'as-is' and either they are just accepted on the
4260
# target, or the Sink will properly convert it.
4261
return self._get_simple_inventory_stream(revision_ids)
4263
# XXX: Hack to make not-chk->chk fetch: copy the inventories as
4264
# inventories. Note that this should probably be done somehow
4265
# as part of bzrlib.repository.StreamSink. Except JAM couldn't
4266
# figure out how a non-chk repository could possibly handle
4267
# deserializing an inventory stream from a chk repo, as it
4268
# doesn't have a way to understand individual pages.
4269
return self._get_convertable_inventory_stream(revision_ids)
4271
def _get_simple_inventory_stream(self, revision_ids):
4272
from_weave = self.from_repository.inventories
4273
yield ('inventories', from_weave.get_record_stream(
4274
[(rev_id,) for rev_id in revision_ids],
4275
self.inventory_fetch_order(),
4276
not self.delta_on_metadata()))
4278
def _get_chk_inventory_stream(self, revision_ids):
4279
"""Fetch the inventory texts, along with the associated chk maps."""
4280
# We want an inventory outside of the search set, so that we can filter
4281
# out uninteresting chk pages. For now we use
4282
# _find_revision_outside_set, but if we had a Search with cut_revs, we
4283
# could use that instead.
4284
start_rev_id = self.from_repository._find_revision_outside_set(
4286
start_rev_key = (start_rev_id,)
4287
inv_keys_to_fetch = [(rev_id,) for rev_id in revision_ids]
4288
if start_rev_id != _mod_revision.NULL_REVISION:
4289
inv_keys_to_fetch.append((start_rev_id,))
4290
# Any repo that supports chk_bytes must also support out-of-order
4291
# insertion. At least, that is how we expect it to work
4292
# We use get_record_stream instead of iter_inventories because we want
4293
# to be able to insert the stream as well. We could instead fetch
4294
# allowing deltas, and then iter_inventories, but we don't know whether
4295
# source or target is more 'local' anway.
4296
inv_stream = self.from_repository.inventories.get_record_stream(
4297
inv_keys_to_fetch, 'unordered',
4298
True) # We need them as full-texts so we can find their references
4299
uninteresting_chk_roots = set()
4300
interesting_chk_roots = set()
4301
def filter_inv_stream(inv_stream):
4302
for idx, record in enumerate(inv_stream):
4303
### child_pb.update('fetch inv', idx, len(inv_keys_to_fetch))
4304
bytes = record.get_bytes_as('fulltext')
4305
chk_inv = inventory.CHKInventory.deserialise(
4306
self.from_repository.chk_bytes, bytes, record.key)
4307
if record.key == start_rev_key:
4308
uninteresting_chk_roots.add(chk_inv.id_to_entry.key())
4309
p_id_map = chk_inv.parent_id_basename_to_file_id
4310
if p_id_map is not None:
4311
uninteresting_chk_roots.add(p_id_map.key())
4314
interesting_chk_roots.add(chk_inv.id_to_entry.key())
4315
p_id_map = chk_inv.parent_id_basename_to_file_id
4316
if p_id_map is not None:
4317
interesting_chk_roots.add(p_id_map.key())
4318
### pb.update('fetch inventory', 0, 2)
4319
yield ('inventories', filter_inv_stream(inv_stream))
4320
# Now that we have worked out all of the interesting root nodes, grab
4321
# all of the interesting pages and insert them
4322
### pb.update('fetch inventory', 1, 2)
4323
interesting = chk_map.iter_interesting_nodes(
4324
self.from_repository.chk_bytes, interesting_chk_roots,
4325
uninteresting_chk_roots)
4326
def to_stream_adapter():
4327
"""Adapt the iter_interesting_nodes result to a single stream.
4329
iter_interesting_nodes returns records as it processes them, along
4330
with keys. However, we only want to return the records themselves.
4332
for record, items in interesting:
4333
if record is not None:
4335
# XXX: We could instead call get_record_stream(records.keys())
4336
# ATM, this will always insert the records as fulltexts, and
4337
# requires that you can hang on to records once you have gone
4338
# on to the next one. Further, it causes the target to
4339
# recompress the data. Testing shows it to be faster than
4340
# requesting the records again, though.
4341
yield ('chk_bytes', to_stream_adapter())
4342
### pb.update('fetch inventory', 2, 2)
4344
def _get_convertable_inventory_stream(self, revision_ids):
4345
# XXX: One of source or target is using chks, and they don't have
4346
# compatible serializations. The StreamSink code expects to be
4347
# able to convert on the target, so we need to put
4348
# bytes-on-the-wire that can be converted
4349
yield ('inventories', self._stream_invs_as_fulltexts(revision_ids))
4351
def _stream_invs_as_fulltexts(self, revision_ids):
4352
from_repo = self.from_repository
4353
from_serializer = from_repo._format._serializer
4354
revision_keys = [(rev_id,) for rev_id in revision_ids]
4355
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4356
for inv in self.from_repository.iter_inventories(revision_ids):
4357
# XXX: This is a bit hackish, but it works. Basically,
4358
# CHKSerializer 'accidentally' supports
4359
# read/write_inventory_to_string, even though that is never
4360
# the format that is stored on disk. It *does* give us a
4361
# single string representation for an inventory, so live with
4363
# This would be far better if we had a 'serialized inventory
4364
# delta' form. Then we could use 'inventory._make_delta', and
4365
# transmit that. This would both be faster to generate, and
4366
# result in fewer bytes-on-the-wire.
4367
as_bytes = from_serializer.write_inventory_to_string(inv)
4368
key = (inv.revision_id,)
4369
parent_keys = parent_map.get(key, ())
4370
yield versionedfile.FulltextContentFactory(
4371
key, parent_keys, None, as_bytes)
1827
4374
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
1828
4375
stop_revision=None):
1829
4376
"""Extend the partial history to include a given index